]>
Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_receiver.c | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | drbd is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation; either version 2, or (at your option) | |
13 | any later version. | |
14 | ||
15 | drbd is distributed in the hope that it will be useful, | |
16 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | GNU General Public License for more details. | |
19 | ||
20 | You should have received a copy of the GNU General Public License | |
21 | along with drbd; see the file COPYING. If not, write to | |
22 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | */ | |
24 | ||
25 | ||
b411b363 PR |
26 | #include <linux/module.h> |
27 | ||
28 | #include <asm/uaccess.h> | |
29 | #include <net/sock.h> | |
30 | ||
b411b363 PR |
31 | #include <linux/drbd.h> |
32 | #include <linux/fs.h> | |
33 | #include <linux/file.h> | |
34 | #include <linux/in.h> | |
35 | #include <linux/mm.h> | |
36 | #include <linux/memcontrol.h> | |
37 | #include <linux/mm_inline.h> | |
38 | #include <linux/slab.h> | |
b411b363 PR |
39 | #include <linux/pkt_sched.h> |
40 | #define __KERNEL_SYSCALLS__ | |
41 | #include <linux/unistd.h> | |
42 | #include <linux/vmalloc.h> | |
43 | #include <linux/random.h> | |
b411b363 PR |
44 | #include <linux/string.h> |
45 | #include <linux/scatterlist.h> | |
46 | #include "drbd_int.h" | |
b411b363 PR |
47 | #include "drbd_req.h" |
48 | ||
49 | #include "drbd_vli.h" | |
50 | ||
b411b363 PR |
51 | enum finish_epoch { |
52 | FE_STILL_LIVE, | |
53 | FE_DESTROYED, | |
54 | FE_RECYCLED, | |
55 | }; | |
56 | ||
57 | static int drbd_do_handshake(struct drbd_conf *mdev); | |
58 | static int drbd_do_auth(struct drbd_conf *mdev); | |
59 | ||
60 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *, struct drbd_epoch *, enum epoch_event); | |
61 | static int e_end_block(struct drbd_conf *, struct drbd_work *, int); | |
62 | ||
b411b363 PR |
63 | |
64 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
65 | ||
45bb912b LE |
66 | /* |
67 | * some helper functions to deal with single linked page lists, | |
68 | * page->private being our "next" pointer. | |
69 | */ | |
70 | ||
71 | /* If at least n pages are linked at head, get n pages off. | |
72 | * Otherwise, don't modify head, and return NULL. | |
73 | * Locking is the responsibility of the caller. | |
74 | */ | |
75 | static struct page *page_chain_del(struct page **head, int n) | |
76 | { | |
77 | struct page *page; | |
78 | struct page *tmp; | |
79 | ||
80 | BUG_ON(!n); | |
81 | BUG_ON(!head); | |
82 | ||
83 | page = *head; | |
23ce4227 PR |
84 | |
85 | if (!page) | |
86 | return NULL; | |
87 | ||
45bb912b LE |
88 | while (page) { |
89 | tmp = page_chain_next(page); | |
90 | if (--n == 0) | |
91 | break; /* found sufficient pages */ | |
92 | if (tmp == NULL) | |
93 | /* insufficient pages, don't use any of them. */ | |
94 | return NULL; | |
95 | page = tmp; | |
96 | } | |
97 | ||
98 | /* add end of list marker for the returned list */ | |
99 | set_page_private(page, 0); | |
100 | /* actual return value, and adjustment of head */ | |
101 | page = *head; | |
102 | *head = tmp; | |
103 | return page; | |
104 | } | |
105 | ||
106 | /* may be used outside of locks to find the tail of a (usually short) | |
107 | * "private" page chain, before adding it back to a global chain head | |
108 | * with page_chain_add() under a spinlock. */ | |
109 | static struct page *page_chain_tail(struct page *page, int *len) | |
110 | { | |
111 | struct page *tmp; | |
112 | int i = 1; | |
113 | while ((tmp = page_chain_next(page))) | |
114 | ++i, page = tmp; | |
115 | if (len) | |
116 | *len = i; | |
117 | return page; | |
118 | } | |
119 | ||
120 | static int page_chain_free(struct page *page) | |
121 | { | |
122 | struct page *tmp; | |
123 | int i = 0; | |
124 | page_chain_for_each_safe(page, tmp) { | |
125 | put_page(page); | |
126 | ++i; | |
127 | } | |
128 | return i; | |
129 | } | |
130 | ||
131 | static void page_chain_add(struct page **head, | |
132 | struct page *chain_first, struct page *chain_last) | |
133 | { | |
134 | #if 1 | |
135 | struct page *tmp; | |
136 | tmp = page_chain_tail(chain_first, NULL); | |
137 | BUG_ON(tmp != chain_last); | |
138 | #endif | |
139 | ||
140 | /* add chain to head */ | |
141 | set_page_private(chain_last, (unsigned long)*head); | |
142 | *head = chain_first; | |
143 | } | |
144 | ||
145 | static struct page *drbd_pp_first_pages_or_try_alloc(struct drbd_conf *mdev, int number) | |
b411b363 PR |
146 | { |
147 | struct page *page = NULL; | |
45bb912b LE |
148 | struct page *tmp = NULL; |
149 | int i = 0; | |
b411b363 PR |
150 | |
151 | /* Yes, testing drbd_pp_vacant outside the lock is racy. | |
152 | * So what. It saves a spin_lock. */ | |
45bb912b | 153 | if (drbd_pp_vacant >= number) { |
b411b363 | 154 | spin_lock(&drbd_pp_lock); |
45bb912b LE |
155 | page = page_chain_del(&drbd_pp_pool, number); |
156 | if (page) | |
157 | drbd_pp_vacant -= number; | |
b411b363 | 158 | spin_unlock(&drbd_pp_lock); |
45bb912b LE |
159 | if (page) |
160 | return page; | |
b411b363 | 161 | } |
45bb912b | 162 | |
b411b363 PR |
163 | /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD |
164 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
165 | * which in turn might block on the other node at this very place. */ | |
45bb912b LE |
166 | for (i = 0; i < number; i++) { |
167 | tmp = alloc_page(GFP_TRY); | |
168 | if (!tmp) | |
169 | break; | |
170 | set_page_private(tmp, (unsigned long)page); | |
171 | page = tmp; | |
172 | } | |
173 | ||
174 | if (i == number) | |
175 | return page; | |
176 | ||
177 | /* Not enough pages immediately available this time. | |
178 | * No need to jump around here, drbd_pp_alloc will retry this | |
179 | * function "soon". */ | |
180 | if (page) { | |
181 | tmp = page_chain_tail(page, NULL); | |
182 | spin_lock(&drbd_pp_lock); | |
183 | page_chain_add(&drbd_pp_pool, page, tmp); | |
184 | drbd_pp_vacant += i; | |
185 | spin_unlock(&drbd_pp_lock); | |
186 | } | |
187 | return NULL; | |
b411b363 PR |
188 | } |
189 | ||
b411b363 PR |
190 | static void reclaim_net_ee(struct drbd_conf *mdev, struct list_head *to_be_freed) |
191 | { | |
192 | struct drbd_epoch_entry *e; | |
193 | struct list_head *le, *tle; | |
194 | ||
195 | /* The EEs are always appended to the end of the list. Since | |
196 | they are sent in order over the wire, they have to finish | |
197 | in order. As soon as we see the first not finished we can | |
198 | stop to examine the list... */ | |
199 | ||
200 | list_for_each_safe(le, tle, &mdev->net_ee) { | |
201 | e = list_entry(le, struct drbd_epoch_entry, w.list); | |
45bb912b | 202 | if (drbd_ee_has_active_page(e)) |
b411b363 PR |
203 | break; |
204 | list_move(le, to_be_freed); | |
205 | } | |
206 | } | |
207 | ||
208 | static void drbd_kick_lo_and_reclaim_net(struct drbd_conf *mdev) | |
209 | { | |
210 | LIST_HEAD(reclaimed); | |
211 | struct drbd_epoch_entry *e, *t; | |
212 | ||
b411b363 PR |
213 | spin_lock_irq(&mdev->req_lock); |
214 | reclaim_net_ee(mdev, &reclaimed); | |
215 | spin_unlock_irq(&mdev->req_lock); | |
216 | ||
217 | list_for_each_entry_safe(e, t, &reclaimed, w.list) | |
435f0740 | 218 | drbd_free_net_ee(mdev, e); |
b411b363 PR |
219 | } |
220 | ||
221 | /** | |
45bb912b | 222 | * drbd_pp_alloc() - Returns @number pages, retries forever (or until signalled) |
b411b363 | 223 | * @mdev: DRBD device. |
45bb912b LE |
224 | * @number: number of pages requested |
225 | * @retry: whether to retry, if not enough pages are available right now | |
226 | * | |
227 | * Tries to allocate number pages, first from our own page pool, then from | |
228 | * the kernel, unless this allocation would exceed the max_buffers setting. | |
229 | * Possibly retry until DRBD frees sufficient pages somewhere else. | |
b411b363 | 230 | * |
45bb912b | 231 | * Returns a page chain linked via page->private. |
b411b363 | 232 | */ |
45bb912b | 233 | static struct page *drbd_pp_alloc(struct drbd_conf *mdev, unsigned number, bool retry) |
b411b363 PR |
234 | { |
235 | struct page *page = NULL; | |
236 | DEFINE_WAIT(wait); | |
237 | ||
45bb912b LE |
238 | /* Yes, we may run up to @number over max_buffers. If we |
239 | * follow it strictly, the admin will get it wrong anyways. */ | |
240 | if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) | |
241 | page = drbd_pp_first_pages_or_try_alloc(mdev, number); | |
b411b363 | 242 | |
45bb912b | 243 | while (page == NULL) { |
b411b363 PR |
244 | prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE); |
245 | ||
246 | drbd_kick_lo_and_reclaim_net(mdev); | |
247 | ||
248 | if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) { | |
45bb912b | 249 | page = drbd_pp_first_pages_or_try_alloc(mdev, number); |
b411b363 PR |
250 | if (page) |
251 | break; | |
252 | } | |
253 | ||
254 | if (!retry) | |
255 | break; | |
256 | ||
257 | if (signal_pending(current)) { | |
258 | dev_warn(DEV, "drbd_pp_alloc interrupted!\n"); | |
259 | break; | |
260 | } | |
261 | ||
262 | schedule(); | |
263 | } | |
264 | finish_wait(&drbd_pp_wait, &wait); | |
265 | ||
45bb912b LE |
266 | if (page) |
267 | atomic_add(number, &mdev->pp_in_use); | |
b411b363 PR |
268 | return page; |
269 | } | |
270 | ||
271 | /* Must not be used from irq, as that may deadlock: see drbd_pp_alloc. | |
45bb912b LE |
272 | * Is also used from inside an other spin_lock_irq(&mdev->req_lock); |
273 | * Either links the page chain back to the global pool, | |
274 | * or returns all pages to the system. */ | |
435f0740 | 275 | static void drbd_pp_free(struct drbd_conf *mdev, struct page *page, int is_net) |
b411b363 | 276 | { |
435f0740 | 277 | atomic_t *a = is_net ? &mdev->pp_in_use_by_net : &mdev->pp_in_use; |
b411b363 | 278 | int i; |
435f0740 | 279 | |
1816a2b4 | 280 | if (drbd_pp_vacant > (DRBD_MAX_BIO_SIZE/PAGE_SIZE)*minor_count) |
45bb912b LE |
281 | i = page_chain_free(page); |
282 | else { | |
283 | struct page *tmp; | |
284 | tmp = page_chain_tail(page, &i); | |
285 | spin_lock(&drbd_pp_lock); | |
286 | page_chain_add(&drbd_pp_pool, page, tmp); | |
287 | drbd_pp_vacant += i; | |
288 | spin_unlock(&drbd_pp_lock); | |
b411b363 | 289 | } |
435f0740 | 290 | i = atomic_sub_return(i, a); |
45bb912b | 291 | if (i < 0) |
435f0740 LE |
292 | dev_warn(DEV, "ASSERTION FAILED: %s: %d < 0\n", |
293 | is_net ? "pp_in_use_by_net" : "pp_in_use", i); | |
b411b363 PR |
294 | wake_up(&drbd_pp_wait); |
295 | } | |
296 | ||
297 | /* | |
298 | You need to hold the req_lock: | |
299 | _drbd_wait_ee_list_empty() | |
300 | ||
301 | You must not have the req_lock: | |
302 | drbd_free_ee() | |
303 | drbd_alloc_ee() | |
304 | drbd_init_ee() | |
305 | drbd_release_ee() | |
306 | drbd_ee_fix_bhs() | |
307 | drbd_process_done_ee() | |
308 | drbd_clear_done_ee() | |
309 | drbd_wait_ee_list_empty() | |
310 | */ | |
311 | ||
312 | struct drbd_epoch_entry *drbd_alloc_ee(struct drbd_conf *mdev, | |
313 | u64 id, | |
314 | sector_t sector, | |
315 | unsigned int data_size, | |
316 | gfp_t gfp_mask) __must_hold(local) | |
317 | { | |
b411b363 PR |
318 | struct drbd_epoch_entry *e; |
319 | struct page *page; | |
45bb912b | 320 | unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b411b363 | 321 | |
0cf9d27e | 322 | if (drbd_insert_fault(mdev, DRBD_FAULT_AL_EE)) |
b411b363 PR |
323 | return NULL; |
324 | ||
325 | e = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM); | |
326 | if (!e) { | |
327 | if (!(gfp_mask & __GFP_NOWARN)) | |
328 | dev_err(DEV, "alloc_ee: Allocation of an EE failed\n"); | |
329 | return NULL; | |
330 | } | |
331 | ||
45bb912b LE |
332 | page = drbd_pp_alloc(mdev, nr_pages, (gfp_mask & __GFP_WAIT)); |
333 | if (!page) | |
334 | goto fail; | |
b411b363 | 335 | |
24c4830c | 336 | INIT_HLIST_NODE(&e->collision); |
b411b363 | 337 | e->epoch = NULL; |
45bb912b LE |
338 | e->mdev = mdev; |
339 | e->pages = page; | |
340 | atomic_set(&e->pending_bios, 0); | |
341 | e->size = data_size; | |
b411b363 | 342 | e->flags = 0; |
45bb912b | 343 | e->sector = sector; |
9a8e7753 AG |
344 | /* |
345 | * The block_id is opaque to the receiver. It is not endianness | |
346 | * converted, and sent back to the sender unchanged. | |
347 | */ | |
45bb912b | 348 | e->block_id = id; |
b411b363 | 349 | |
b411b363 PR |
350 | return e; |
351 | ||
45bb912b | 352 | fail: |
b411b363 | 353 | mempool_free(e, drbd_ee_mempool); |
b411b363 PR |
354 | return NULL; |
355 | } | |
356 | ||
435f0740 | 357 | void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, int is_net) |
b411b363 | 358 | { |
c36c3ced LE |
359 | if (e->flags & EE_HAS_DIGEST) |
360 | kfree(e->digest); | |
435f0740 | 361 | drbd_pp_free(mdev, e->pages, is_net); |
45bb912b | 362 | D_ASSERT(atomic_read(&e->pending_bios) == 0); |
24c4830c | 363 | D_ASSERT(hlist_unhashed(&e->collision)); |
b411b363 PR |
364 | mempool_free(e, drbd_ee_mempool); |
365 | } | |
366 | ||
367 | int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list) | |
368 | { | |
369 | LIST_HEAD(work_list); | |
370 | struct drbd_epoch_entry *e, *t; | |
371 | int count = 0; | |
435f0740 | 372 | int is_net = list == &mdev->net_ee; |
b411b363 PR |
373 | |
374 | spin_lock_irq(&mdev->req_lock); | |
375 | list_splice_init(list, &work_list); | |
376 | spin_unlock_irq(&mdev->req_lock); | |
377 | ||
378 | list_for_each_entry_safe(e, t, &work_list, w.list) { | |
435f0740 | 379 | drbd_free_some_ee(mdev, e, is_net); |
b411b363 PR |
380 | count++; |
381 | } | |
382 | return count; | |
383 | } | |
384 | ||
385 | ||
386 | /* | |
387 | * This function is called from _asender only_ | |
388 | * but see also comments in _req_mod(,barrier_acked) | |
389 | * and receive_Barrier. | |
390 | * | |
391 | * Move entries from net_ee to done_ee, if ready. | |
392 | * Grab done_ee, call all callbacks, free the entries. | |
393 | * The callbacks typically send out ACKs. | |
394 | */ | |
395 | static int drbd_process_done_ee(struct drbd_conf *mdev) | |
396 | { | |
397 | LIST_HEAD(work_list); | |
398 | LIST_HEAD(reclaimed); | |
399 | struct drbd_epoch_entry *e, *t; | |
400 | int ok = (mdev->state.conn >= C_WF_REPORT_PARAMS); | |
401 | ||
402 | spin_lock_irq(&mdev->req_lock); | |
403 | reclaim_net_ee(mdev, &reclaimed); | |
404 | list_splice_init(&mdev->done_ee, &work_list); | |
405 | spin_unlock_irq(&mdev->req_lock); | |
406 | ||
407 | list_for_each_entry_safe(e, t, &reclaimed, w.list) | |
435f0740 | 408 | drbd_free_net_ee(mdev, e); |
b411b363 PR |
409 | |
410 | /* possible callbacks here: | |
411 | * e_end_block, and e_end_resync_block, e_send_discard_ack. | |
412 | * all ignore the last argument. | |
413 | */ | |
414 | list_for_each_entry_safe(e, t, &work_list, w.list) { | |
b411b363 PR |
415 | /* list_del not necessary, next/prev members not touched */ |
416 | ok = e->w.cb(mdev, &e->w, !ok) && ok; | |
417 | drbd_free_ee(mdev, e); | |
418 | } | |
419 | wake_up(&mdev->ee_wait); | |
420 | ||
421 | return ok; | |
422 | } | |
423 | ||
424 | void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) | |
425 | { | |
426 | DEFINE_WAIT(wait); | |
427 | ||
428 | /* avoids spin_lock/unlock | |
429 | * and calling prepare_to_wait in the fast path */ | |
430 | while (!list_empty(head)) { | |
431 | prepare_to_wait(&mdev->ee_wait, &wait, TASK_UNINTERRUPTIBLE); | |
432 | spin_unlock_irq(&mdev->req_lock); | |
7eaceacc | 433 | io_schedule(); |
b411b363 PR |
434 | finish_wait(&mdev->ee_wait, &wait); |
435 | spin_lock_irq(&mdev->req_lock); | |
436 | } | |
437 | } | |
438 | ||
439 | void drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) | |
440 | { | |
441 | spin_lock_irq(&mdev->req_lock); | |
442 | _drbd_wait_ee_list_empty(mdev, head); | |
443 | spin_unlock_irq(&mdev->req_lock); | |
444 | } | |
445 | ||
446 | /* see also kernel_accept; which is only present since 2.6.18. | |
447 | * also we want to log which part of it failed, exactly */ | |
448 | static int drbd_accept(struct drbd_conf *mdev, const char **what, | |
449 | struct socket *sock, struct socket **newsock) | |
450 | { | |
451 | struct sock *sk = sock->sk; | |
452 | int err = 0; | |
453 | ||
454 | *what = "listen"; | |
455 | err = sock->ops->listen(sock, 5); | |
456 | if (err < 0) | |
457 | goto out; | |
458 | ||
459 | *what = "sock_create_lite"; | |
460 | err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol, | |
461 | newsock); | |
462 | if (err < 0) | |
463 | goto out; | |
464 | ||
465 | *what = "accept"; | |
466 | err = sock->ops->accept(sock, *newsock, 0); | |
467 | if (err < 0) { | |
468 | sock_release(*newsock); | |
469 | *newsock = NULL; | |
470 | goto out; | |
471 | } | |
472 | (*newsock)->ops = sock->ops; | |
473 | ||
474 | out: | |
475 | return err; | |
476 | } | |
477 | ||
478 | static int drbd_recv_short(struct drbd_conf *mdev, struct socket *sock, | |
479 | void *buf, size_t size, int flags) | |
480 | { | |
481 | mm_segment_t oldfs; | |
482 | struct kvec iov = { | |
483 | .iov_base = buf, | |
484 | .iov_len = size, | |
485 | }; | |
486 | struct msghdr msg = { | |
487 | .msg_iovlen = 1, | |
488 | .msg_iov = (struct iovec *)&iov, | |
489 | .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) | |
490 | }; | |
491 | int rv; | |
492 | ||
493 | oldfs = get_fs(); | |
494 | set_fs(KERNEL_DS); | |
495 | rv = sock_recvmsg(sock, &msg, size, msg.msg_flags); | |
496 | set_fs(oldfs); | |
497 | ||
498 | return rv; | |
499 | } | |
500 | ||
501 | static int drbd_recv(struct drbd_conf *mdev, void *buf, size_t size) | |
502 | { | |
503 | mm_segment_t oldfs; | |
504 | struct kvec iov = { | |
505 | .iov_base = buf, | |
506 | .iov_len = size, | |
507 | }; | |
508 | struct msghdr msg = { | |
509 | .msg_iovlen = 1, | |
510 | .msg_iov = (struct iovec *)&iov, | |
511 | .msg_flags = MSG_WAITALL | MSG_NOSIGNAL | |
512 | }; | |
513 | int rv; | |
514 | ||
515 | oldfs = get_fs(); | |
516 | set_fs(KERNEL_DS); | |
517 | ||
518 | for (;;) { | |
519 | rv = sock_recvmsg(mdev->data.socket, &msg, size, msg.msg_flags); | |
520 | if (rv == size) | |
521 | break; | |
522 | ||
523 | /* Note: | |
524 | * ECONNRESET other side closed the connection | |
525 | * ERESTARTSYS (on sock) we got a signal | |
526 | */ | |
527 | ||
528 | if (rv < 0) { | |
529 | if (rv == -ECONNRESET) | |
530 | dev_info(DEV, "sock was reset by peer\n"); | |
531 | else if (rv != -ERESTARTSYS) | |
532 | dev_err(DEV, "sock_recvmsg returned %d\n", rv); | |
533 | break; | |
534 | } else if (rv == 0) { | |
535 | dev_info(DEV, "sock was shut down by peer\n"); | |
536 | break; | |
537 | } else { | |
538 | /* signal came in, or peer/link went down, | |
539 | * after we read a partial message | |
540 | */ | |
541 | /* D_ASSERT(signal_pending(current)); */ | |
542 | break; | |
543 | } | |
544 | }; | |
545 | ||
546 | set_fs(oldfs); | |
547 | ||
548 | if (rv != size) | |
549 | drbd_force_state(mdev, NS(conn, C_BROKEN_PIPE)); | |
550 | ||
551 | return rv; | |
552 | } | |
553 | ||
5dbf1673 LE |
554 | /* quoting tcp(7): |
555 | * On individual connections, the socket buffer size must be set prior to the | |
556 | * listen(2) or connect(2) calls in order to have it take effect. | |
557 | * This is our wrapper to do so. | |
558 | */ | |
559 | static void drbd_setbufsize(struct socket *sock, unsigned int snd, | |
560 | unsigned int rcv) | |
561 | { | |
562 | /* open coded SO_SNDBUF, SO_RCVBUF */ | |
563 | if (snd) { | |
564 | sock->sk->sk_sndbuf = snd; | |
565 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
566 | } | |
567 | if (rcv) { | |
568 | sock->sk->sk_rcvbuf = rcv; | |
569 | sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
570 | } | |
571 | } | |
572 | ||
b411b363 PR |
573 | static struct socket *drbd_try_connect(struct drbd_conf *mdev) |
574 | { | |
575 | const char *what; | |
576 | struct socket *sock; | |
577 | struct sockaddr_in6 src_in6; | |
578 | int err; | |
579 | int disconnect_on_error = 1; | |
580 | ||
581 | if (!get_net_conf(mdev)) | |
582 | return NULL; | |
583 | ||
584 | what = "sock_create_kern"; | |
585 | err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family, | |
586 | SOCK_STREAM, IPPROTO_TCP, &sock); | |
587 | if (err < 0) { | |
588 | sock = NULL; | |
589 | goto out; | |
590 | } | |
591 | ||
592 | sock->sk->sk_rcvtimeo = | |
593 | sock->sk->sk_sndtimeo = mdev->net_conf->try_connect_int*HZ; | |
5dbf1673 LE |
594 | drbd_setbufsize(sock, mdev->net_conf->sndbuf_size, |
595 | mdev->net_conf->rcvbuf_size); | |
b411b363 PR |
596 | |
597 | /* explicitly bind to the configured IP as source IP | |
598 | * for the outgoing connections. | |
599 | * This is needed for multihomed hosts and to be | |
600 | * able to use lo: interfaces for drbd. | |
601 | * Make sure to use 0 as port number, so linux selects | |
602 | * a free one dynamically. | |
603 | */ | |
604 | memcpy(&src_in6, mdev->net_conf->my_addr, | |
605 | min_t(int, mdev->net_conf->my_addr_len, sizeof(src_in6))); | |
606 | if (((struct sockaddr *)mdev->net_conf->my_addr)->sa_family == AF_INET6) | |
607 | src_in6.sin6_port = 0; | |
608 | else | |
609 | ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ | |
610 | ||
611 | what = "bind before connect"; | |
612 | err = sock->ops->bind(sock, | |
613 | (struct sockaddr *) &src_in6, | |
614 | mdev->net_conf->my_addr_len); | |
615 | if (err < 0) | |
616 | goto out; | |
617 | ||
618 | /* connect may fail, peer not yet available. | |
619 | * stay C_WF_CONNECTION, don't go Disconnecting! */ | |
620 | disconnect_on_error = 0; | |
621 | what = "connect"; | |
622 | err = sock->ops->connect(sock, | |
623 | (struct sockaddr *)mdev->net_conf->peer_addr, | |
624 | mdev->net_conf->peer_addr_len, 0); | |
625 | ||
626 | out: | |
627 | if (err < 0) { | |
628 | if (sock) { | |
629 | sock_release(sock); | |
630 | sock = NULL; | |
631 | } | |
632 | switch (-err) { | |
633 | /* timeout, busy, signal pending */ | |
634 | case ETIMEDOUT: case EAGAIN: case EINPROGRESS: | |
635 | case EINTR: case ERESTARTSYS: | |
636 | /* peer not (yet) available, network problem */ | |
637 | case ECONNREFUSED: case ENETUNREACH: | |
638 | case EHOSTDOWN: case EHOSTUNREACH: | |
639 | disconnect_on_error = 0; | |
640 | break; | |
641 | default: | |
642 | dev_err(DEV, "%s failed, err = %d\n", what, err); | |
643 | } | |
644 | if (disconnect_on_error) | |
645 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
646 | } | |
647 | put_net_conf(mdev); | |
648 | return sock; | |
649 | } | |
650 | ||
651 | static struct socket *drbd_wait_for_connect(struct drbd_conf *mdev) | |
652 | { | |
653 | int timeo, err; | |
654 | struct socket *s_estab = NULL, *s_listen; | |
655 | const char *what; | |
656 | ||
657 | if (!get_net_conf(mdev)) | |
658 | return NULL; | |
659 | ||
660 | what = "sock_create_kern"; | |
661 | err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family, | |
662 | SOCK_STREAM, IPPROTO_TCP, &s_listen); | |
663 | if (err) { | |
664 | s_listen = NULL; | |
665 | goto out; | |
666 | } | |
667 | ||
668 | timeo = mdev->net_conf->try_connect_int * HZ; | |
669 | timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */ | |
670 | ||
671 | s_listen->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
672 | s_listen->sk->sk_rcvtimeo = timeo; | |
673 | s_listen->sk->sk_sndtimeo = timeo; | |
5dbf1673 LE |
674 | drbd_setbufsize(s_listen, mdev->net_conf->sndbuf_size, |
675 | mdev->net_conf->rcvbuf_size); | |
b411b363 PR |
676 | |
677 | what = "bind before listen"; | |
678 | err = s_listen->ops->bind(s_listen, | |
679 | (struct sockaddr *) mdev->net_conf->my_addr, | |
680 | mdev->net_conf->my_addr_len); | |
681 | if (err < 0) | |
682 | goto out; | |
683 | ||
684 | err = drbd_accept(mdev, &what, s_listen, &s_estab); | |
685 | ||
686 | out: | |
687 | if (s_listen) | |
688 | sock_release(s_listen); | |
689 | if (err < 0) { | |
690 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
691 | dev_err(DEV, "%s failed, err = %d\n", what, err); | |
692 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
693 | } | |
694 | } | |
695 | put_net_conf(mdev); | |
696 | ||
697 | return s_estab; | |
698 | } | |
699 | ||
700 | static int drbd_send_fp(struct drbd_conf *mdev, | |
701 | struct socket *sock, enum drbd_packets cmd) | |
702 | { | |
02918be2 | 703 | struct p_header80 *h = &mdev->data.sbuf.header.h80; |
b411b363 PR |
704 | |
705 | return _drbd_send_cmd(mdev, sock, cmd, h, sizeof(*h), 0); | |
706 | } | |
707 | ||
708 | static enum drbd_packets drbd_recv_fp(struct drbd_conf *mdev, struct socket *sock) | |
709 | { | |
02918be2 | 710 | struct p_header80 *h = &mdev->data.rbuf.header.h80; |
b411b363 PR |
711 | int rr; |
712 | ||
713 | rr = drbd_recv_short(mdev, sock, h, sizeof(*h), 0); | |
714 | ||
ca9bc12b | 715 | if (rr == sizeof(*h) && h->magic == cpu_to_be32(DRBD_MAGIC)) |
b411b363 PR |
716 | return be16_to_cpu(h->command); |
717 | ||
718 | return 0xffff; | |
719 | } | |
720 | ||
721 | /** | |
722 | * drbd_socket_okay() - Free the socket if its connection is not okay | |
723 | * @mdev: DRBD device. | |
724 | * @sock: pointer to the pointer to the socket. | |
725 | */ | |
726 | static int drbd_socket_okay(struct drbd_conf *mdev, struct socket **sock) | |
727 | { | |
728 | int rr; | |
729 | char tb[4]; | |
730 | ||
731 | if (!*sock) | |
81e84650 | 732 | return false; |
b411b363 PR |
733 | |
734 | rr = drbd_recv_short(mdev, *sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); | |
735 | ||
736 | if (rr > 0 || rr == -EAGAIN) { | |
81e84650 | 737 | return true; |
b411b363 PR |
738 | } else { |
739 | sock_release(*sock); | |
740 | *sock = NULL; | |
81e84650 | 741 | return false; |
b411b363 PR |
742 | } |
743 | } | |
744 | ||
745 | /* | |
746 | * return values: | |
747 | * 1 yes, we have a valid connection | |
748 | * 0 oops, did not work out, please try again | |
749 | * -1 peer talks different language, | |
750 | * no point in trying again, please go standalone. | |
751 | * -2 We do not have a network config... | |
752 | */ | |
753 | static int drbd_connect(struct drbd_conf *mdev) | |
754 | { | |
755 | struct socket *s, *sock, *msock; | |
756 | int try, h, ok; | |
757 | ||
758 | D_ASSERT(!mdev->data.socket); | |
759 | ||
b411b363 PR |
760 | if (drbd_request_state(mdev, NS(conn, C_WF_CONNECTION)) < SS_SUCCESS) |
761 | return -2; | |
762 | ||
763 | clear_bit(DISCARD_CONCURRENT, &mdev->flags); | |
764 | ||
765 | sock = NULL; | |
766 | msock = NULL; | |
767 | ||
768 | do { | |
769 | for (try = 0;;) { | |
770 | /* 3 tries, this should take less than a second! */ | |
771 | s = drbd_try_connect(mdev); | |
772 | if (s || ++try >= 3) | |
773 | break; | |
774 | /* give the other side time to call bind() & listen() */ | |
20ee6390 | 775 | schedule_timeout_interruptible(HZ / 10); |
b411b363 PR |
776 | } |
777 | ||
778 | if (s) { | |
779 | if (!sock) { | |
780 | drbd_send_fp(mdev, s, P_HAND_SHAKE_S); | |
781 | sock = s; | |
782 | s = NULL; | |
783 | } else if (!msock) { | |
784 | drbd_send_fp(mdev, s, P_HAND_SHAKE_M); | |
785 | msock = s; | |
786 | s = NULL; | |
787 | } else { | |
788 | dev_err(DEV, "Logic error in drbd_connect()\n"); | |
789 | goto out_release_sockets; | |
790 | } | |
791 | } | |
792 | ||
793 | if (sock && msock) { | |
a8e40792 | 794 | schedule_timeout_interruptible(mdev->net_conf->ping_timeo*HZ/10); |
b411b363 PR |
795 | ok = drbd_socket_okay(mdev, &sock); |
796 | ok = drbd_socket_okay(mdev, &msock) && ok; | |
797 | if (ok) | |
798 | break; | |
799 | } | |
800 | ||
801 | retry: | |
802 | s = drbd_wait_for_connect(mdev); | |
803 | if (s) { | |
804 | try = drbd_recv_fp(mdev, s); | |
805 | drbd_socket_okay(mdev, &sock); | |
806 | drbd_socket_okay(mdev, &msock); | |
807 | switch (try) { | |
808 | case P_HAND_SHAKE_S: | |
809 | if (sock) { | |
810 | dev_warn(DEV, "initial packet S crossed\n"); | |
811 | sock_release(sock); | |
812 | } | |
813 | sock = s; | |
814 | break; | |
815 | case P_HAND_SHAKE_M: | |
816 | if (msock) { | |
817 | dev_warn(DEV, "initial packet M crossed\n"); | |
818 | sock_release(msock); | |
819 | } | |
820 | msock = s; | |
821 | set_bit(DISCARD_CONCURRENT, &mdev->flags); | |
822 | break; | |
823 | default: | |
824 | dev_warn(DEV, "Error receiving initial packet\n"); | |
825 | sock_release(s); | |
826 | if (random32() & 1) | |
827 | goto retry; | |
828 | } | |
829 | } | |
830 | ||
831 | if (mdev->state.conn <= C_DISCONNECTING) | |
832 | goto out_release_sockets; | |
833 | if (signal_pending(current)) { | |
834 | flush_signals(current); | |
835 | smp_rmb(); | |
836 | if (get_t_state(&mdev->receiver) == Exiting) | |
837 | goto out_release_sockets; | |
838 | } | |
839 | ||
840 | if (sock && msock) { | |
841 | ok = drbd_socket_okay(mdev, &sock); | |
842 | ok = drbd_socket_okay(mdev, &msock) && ok; | |
843 | if (ok) | |
844 | break; | |
845 | } | |
846 | } while (1); | |
847 | ||
848 | msock->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
849 | sock->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
850 | ||
851 | sock->sk->sk_allocation = GFP_NOIO; | |
852 | msock->sk->sk_allocation = GFP_NOIO; | |
853 | ||
854 | sock->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; | |
855 | msock->sk->sk_priority = TC_PRIO_INTERACTIVE; | |
856 | ||
b411b363 PR |
857 | /* NOT YET ... |
858 | * sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; | |
859 | * sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; | |
860 | * first set it to the P_HAND_SHAKE timeout, | |
861 | * which we set to 4x the configured ping_timeout. */ | |
862 | sock->sk->sk_sndtimeo = | |
863 | sock->sk->sk_rcvtimeo = mdev->net_conf->ping_timeo*4*HZ/10; | |
864 | ||
865 | msock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; | |
866 | msock->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ; | |
867 | ||
868 | /* we don't want delays. | |
25985edc | 869 | * we use TCP_CORK where appropriate, though */ |
b411b363 PR |
870 | drbd_tcp_nodelay(sock); |
871 | drbd_tcp_nodelay(msock); | |
872 | ||
873 | mdev->data.socket = sock; | |
874 | mdev->meta.socket = msock; | |
875 | mdev->last_received = jiffies; | |
876 | ||
877 | D_ASSERT(mdev->asender.task == NULL); | |
878 | ||
879 | h = drbd_do_handshake(mdev); | |
880 | if (h <= 0) | |
881 | return h; | |
882 | ||
883 | if (mdev->cram_hmac_tfm) { | |
884 | /* drbd_request_state(mdev, NS(conn, WFAuth)); */ | |
b10d96cb JT |
885 | switch (drbd_do_auth(mdev)) { |
886 | case -1: | |
b411b363 PR |
887 | dev_err(DEV, "Authentication of peer failed\n"); |
888 | return -1; | |
b10d96cb JT |
889 | case 0: |
890 | dev_err(DEV, "Authentication of peer failed, trying again.\n"); | |
891 | return 0; | |
b411b363 PR |
892 | } |
893 | } | |
894 | ||
895 | if (drbd_request_state(mdev, NS(conn, C_WF_REPORT_PARAMS)) < SS_SUCCESS) | |
896 | return 0; | |
897 | ||
898 | sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; | |
899 | sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; | |
900 | ||
901 | atomic_set(&mdev->packet_seq, 0); | |
902 | mdev->peer_seq = 0; | |
903 | ||
904 | drbd_thread_start(&mdev->asender); | |
905 | ||
148efa16 | 906 | if (drbd_send_protocol(mdev) == -1) |
7e2455c1 | 907 | return -1; |
b411b363 | 908 | drbd_send_sync_param(mdev, &mdev->sync_conf); |
e89b591c | 909 | drbd_send_sizes(mdev, 0, 0); |
b411b363 PR |
910 | drbd_send_uuids(mdev); |
911 | drbd_send_state(mdev); | |
912 | clear_bit(USE_DEGR_WFC_T, &mdev->flags); | |
913 | clear_bit(RESIZE_PENDING, &mdev->flags); | |
7fde2be9 | 914 | mod_timer(&mdev->request_timer, jiffies + HZ); /* just start it here. */ |
b411b363 PR |
915 | |
916 | return 1; | |
917 | ||
918 | out_release_sockets: | |
919 | if (sock) | |
920 | sock_release(sock); | |
921 | if (msock) | |
922 | sock_release(msock); | |
923 | return -1; | |
924 | } | |
925 | ||
02918be2 | 926 | static int drbd_recv_header(struct drbd_conf *mdev, enum drbd_packets *cmd, unsigned int *packet_size) |
b411b363 | 927 | { |
02918be2 | 928 | union p_header *h = &mdev->data.rbuf.header; |
b411b363 PR |
929 | int r; |
930 | ||
931 | r = drbd_recv(mdev, h, sizeof(*h)); | |
b411b363 | 932 | if (unlikely(r != sizeof(*h))) { |
0ddc5549 LE |
933 | if (!signal_pending(current)) |
934 | dev_warn(DEV, "short read expecting header on sock: r=%d\n", r); | |
81e84650 | 935 | return false; |
02918be2 PR |
936 | } |
937 | ||
ca9bc12b | 938 | if (likely(h->h80.magic == cpu_to_be32(DRBD_MAGIC))) { |
02918be2 PR |
939 | *cmd = be16_to_cpu(h->h80.command); |
940 | *packet_size = be16_to_cpu(h->h80.length); | |
ca9bc12b | 941 | } else if (h->h95.magic == cpu_to_be16(DRBD_MAGIC_BIG)) { |
02918be2 PR |
942 | *cmd = be16_to_cpu(h->h95.command); |
943 | *packet_size = be32_to_cpu(h->h95.length); | |
944 | } else { | |
004352fa LE |
945 | dev_err(DEV, "magic?? on data m: 0x%08x c: %d l: %d\n", |
946 | be32_to_cpu(h->h80.magic), | |
947 | be16_to_cpu(h->h80.command), | |
948 | be16_to_cpu(h->h80.length)); | |
81e84650 | 949 | return false; |
b411b363 PR |
950 | } |
951 | mdev->last_received = jiffies; | |
952 | ||
81e84650 | 953 | return true; |
b411b363 PR |
954 | } |
955 | ||
2451fc3b | 956 | static void drbd_flush(struct drbd_conf *mdev) |
b411b363 PR |
957 | { |
958 | int rv; | |
959 | ||
960 | if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) { | |
fbd9b09a | 961 | rv = blkdev_issue_flush(mdev->ldev->backing_bdev, GFP_KERNEL, |
dd3932ed | 962 | NULL); |
b411b363 PR |
963 | if (rv) { |
964 | dev_err(DEV, "local disk flush failed with status %d\n", rv); | |
965 | /* would rather check on EOPNOTSUPP, but that is not reliable. | |
966 | * don't try again for ANY return value != 0 | |
967 | * if (rv == -EOPNOTSUPP) */ | |
968 | drbd_bump_write_ordering(mdev, WO_drain_io); | |
969 | } | |
970 | put_ldev(mdev); | |
971 | } | |
b411b363 PR |
972 | } |
973 | ||
974 | /** | |
975 | * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. | |
976 | * @mdev: DRBD device. | |
977 | * @epoch: Epoch object. | |
978 | * @ev: Epoch event. | |
979 | */ | |
980 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev, | |
981 | struct drbd_epoch *epoch, | |
982 | enum epoch_event ev) | |
983 | { | |
2451fc3b | 984 | int epoch_size; |
b411b363 | 985 | struct drbd_epoch *next_epoch; |
b411b363 PR |
986 | enum finish_epoch rv = FE_STILL_LIVE; |
987 | ||
988 | spin_lock(&mdev->epoch_lock); | |
989 | do { | |
990 | next_epoch = NULL; | |
b411b363 PR |
991 | |
992 | epoch_size = atomic_read(&epoch->epoch_size); | |
993 | ||
994 | switch (ev & ~EV_CLEANUP) { | |
995 | case EV_PUT: | |
996 | atomic_dec(&epoch->active); | |
997 | break; | |
998 | case EV_GOT_BARRIER_NR: | |
999 | set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); | |
b411b363 PR |
1000 | break; |
1001 | case EV_BECAME_LAST: | |
1002 | /* nothing to do*/ | |
1003 | break; | |
1004 | } | |
1005 | ||
b411b363 PR |
1006 | if (epoch_size != 0 && |
1007 | atomic_read(&epoch->active) == 0 && | |
2451fc3b | 1008 | test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) { |
b411b363 PR |
1009 | if (!(ev & EV_CLEANUP)) { |
1010 | spin_unlock(&mdev->epoch_lock); | |
1011 | drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size); | |
1012 | spin_lock(&mdev->epoch_lock); | |
1013 | } | |
1014 | dec_unacked(mdev); | |
1015 | ||
1016 | if (mdev->current_epoch != epoch) { | |
1017 | next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); | |
1018 | list_del(&epoch->list); | |
1019 | ev = EV_BECAME_LAST | (ev & EV_CLEANUP); | |
1020 | mdev->epochs--; | |
b411b363 PR |
1021 | kfree(epoch); |
1022 | ||
1023 | if (rv == FE_STILL_LIVE) | |
1024 | rv = FE_DESTROYED; | |
1025 | } else { | |
1026 | epoch->flags = 0; | |
1027 | atomic_set(&epoch->epoch_size, 0); | |
698f9315 | 1028 | /* atomic_set(&epoch->active, 0); is already zero */ |
b411b363 PR |
1029 | if (rv == FE_STILL_LIVE) |
1030 | rv = FE_RECYCLED; | |
2451fc3b | 1031 | wake_up(&mdev->ee_wait); |
b411b363 PR |
1032 | } |
1033 | } | |
1034 | ||
1035 | if (!next_epoch) | |
1036 | break; | |
1037 | ||
1038 | epoch = next_epoch; | |
1039 | } while (1); | |
1040 | ||
1041 | spin_unlock(&mdev->epoch_lock); | |
1042 | ||
b411b363 PR |
1043 | return rv; |
1044 | } | |
1045 | ||
1046 | /** | |
1047 | * drbd_bump_write_ordering() - Fall back to an other write ordering method | |
1048 | * @mdev: DRBD device. | |
1049 | * @wo: Write ordering method to try. | |
1050 | */ | |
1051 | void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local) | |
1052 | { | |
1053 | enum write_ordering_e pwo; | |
1054 | static char *write_ordering_str[] = { | |
1055 | [WO_none] = "none", | |
1056 | [WO_drain_io] = "drain", | |
1057 | [WO_bdev_flush] = "flush", | |
b411b363 PR |
1058 | }; |
1059 | ||
1060 | pwo = mdev->write_ordering; | |
1061 | wo = min(pwo, wo); | |
b411b363 PR |
1062 | if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush) |
1063 | wo = WO_drain_io; | |
1064 | if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain) | |
1065 | wo = WO_none; | |
1066 | mdev->write_ordering = wo; | |
2451fc3b | 1067 | if (pwo != mdev->write_ordering || wo == WO_bdev_flush) |
b411b363 PR |
1068 | dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]); |
1069 | } | |
1070 | ||
45bb912b LE |
1071 | /** |
1072 | * drbd_submit_ee() | |
1073 | * @mdev: DRBD device. | |
1074 | * @e: epoch entry | |
1075 | * @rw: flag field, see bio->bi_rw | |
10f6d992 LE |
1076 | * |
1077 | * May spread the pages to multiple bios, | |
1078 | * depending on bio_add_page restrictions. | |
1079 | * | |
1080 | * Returns 0 if all bios have been submitted, | |
1081 | * -ENOMEM if we could not allocate enough bios, | |
1082 | * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a | |
1083 | * single page to an empty bio (which should never happen and likely indicates | |
1084 | * that the lower level IO stack is in some way broken). This has been observed | |
1085 | * on certain Xen deployments. | |
45bb912b LE |
1086 | */ |
1087 | /* TODO allocate from our own bio_set. */ | |
1088 | int drbd_submit_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, | |
1089 | const unsigned rw, const int fault_type) | |
1090 | { | |
1091 | struct bio *bios = NULL; | |
1092 | struct bio *bio; | |
1093 | struct page *page = e->pages; | |
1094 | sector_t sector = e->sector; | |
1095 | unsigned ds = e->size; | |
1096 | unsigned n_bios = 0; | |
1097 | unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT; | |
10f6d992 | 1098 | int err = -ENOMEM; |
45bb912b LE |
1099 | |
1100 | /* In most cases, we will only need one bio. But in case the lower | |
1101 | * level restrictions happen to be different at this offset on this | |
1102 | * side than those of the sending peer, we may need to submit the | |
1103 | * request in more than one bio. */ | |
1104 | next_bio: | |
1105 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
1106 | if (!bio) { | |
1107 | dev_err(DEV, "submit_ee: Allocation of a bio failed\n"); | |
1108 | goto fail; | |
1109 | } | |
1110 | /* > e->sector, unless this is the first bio */ | |
1111 | bio->bi_sector = sector; | |
1112 | bio->bi_bdev = mdev->ldev->backing_bdev; | |
45bb912b LE |
1113 | bio->bi_rw = rw; |
1114 | bio->bi_private = e; | |
1115 | bio->bi_end_io = drbd_endio_sec; | |
1116 | ||
1117 | bio->bi_next = bios; | |
1118 | bios = bio; | |
1119 | ++n_bios; | |
1120 | ||
1121 | page_chain_for_each(page) { | |
1122 | unsigned len = min_t(unsigned, ds, PAGE_SIZE); | |
1123 | if (!bio_add_page(bio, page, len, 0)) { | |
10f6d992 LE |
1124 | /* A single page must always be possible! |
1125 | * But in case it fails anyways, | |
1126 | * we deal with it, and complain (below). */ | |
1127 | if (bio->bi_vcnt == 0) { | |
1128 | dev_err(DEV, | |
1129 | "bio_add_page failed for len=%u, " | |
1130 | "bi_vcnt=0 (bi_sector=%llu)\n", | |
1131 | len, (unsigned long long)bio->bi_sector); | |
1132 | err = -ENOSPC; | |
1133 | goto fail; | |
1134 | } | |
45bb912b LE |
1135 | goto next_bio; |
1136 | } | |
1137 | ds -= len; | |
1138 | sector += len >> 9; | |
1139 | --nr_pages; | |
1140 | } | |
1141 | D_ASSERT(page == NULL); | |
1142 | D_ASSERT(ds == 0); | |
1143 | ||
1144 | atomic_set(&e->pending_bios, n_bios); | |
1145 | do { | |
1146 | bio = bios; | |
1147 | bios = bios->bi_next; | |
1148 | bio->bi_next = NULL; | |
1149 | ||
45bb912b | 1150 | drbd_generic_make_request(mdev, fault_type, bio); |
45bb912b | 1151 | } while (bios); |
45bb912b LE |
1152 | return 0; |
1153 | ||
1154 | fail: | |
1155 | while (bios) { | |
1156 | bio = bios; | |
1157 | bios = bios->bi_next; | |
1158 | bio_put(bio); | |
1159 | } | |
10f6d992 | 1160 | return err; |
45bb912b LE |
1161 | } |
1162 | ||
02918be2 | 1163 | static int receive_Barrier(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 1164 | { |
2451fc3b | 1165 | int rv; |
02918be2 | 1166 | struct p_barrier *p = &mdev->data.rbuf.barrier; |
b411b363 PR |
1167 | struct drbd_epoch *epoch; |
1168 | ||
b411b363 PR |
1169 | inc_unacked(mdev); |
1170 | ||
b411b363 PR |
1171 | mdev->current_epoch->barrier_nr = p->barrier; |
1172 | rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR); | |
1173 | ||
1174 | /* P_BARRIER_ACK may imply that the corresponding extent is dropped from | |
1175 | * the activity log, which means it would not be resynced in case the | |
1176 | * R_PRIMARY crashes now. | |
1177 | * Therefore we must send the barrier_ack after the barrier request was | |
1178 | * completed. */ | |
1179 | switch (mdev->write_ordering) { | |
b411b363 PR |
1180 | case WO_none: |
1181 | if (rv == FE_RECYCLED) | |
81e84650 | 1182 | return true; |
2451fc3b PR |
1183 | |
1184 | /* receiver context, in the writeout path of the other node. | |
1185 | * avoid potential distributed deadlock */ | |
1186 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1187 | if (epoch) | |
1188 | break; | |
1189 | else | |
1190 | dev_warn(DEV, "Allocation of an epoch failed, slowing down\n"); | |
1191 | /* Fall through */ | |
b411b363 PR |
1192 | |
1193 | case WO_bdev_flush: | |
1194 | case WO_drain_io: | |
b411b363 | 1195 | drbd_wait_ee_list_empty(mdev, &mdev->active_ee); |
2451fc3b PR |
1196 | drbd_flush(mdev); |
1197 | ||
1198 | if (atomic_read(&mdev->current_epoch->epoch_size)) { | |
1199 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1200 | if (epoch) | |
1201 | break; | |
b411b363 PR |
1202 | } |
1203 | ||
2451fc3b PR |
1204 | epoch = mdev->current_epoch; |
1205 | wait_event(mdev->ee_wait, atomic_read(&epoch->epoch_size) == 0); | |
1206 | ||
1207 | D_ASSERT(atomic_read(&epoch->active) == 0); | |
1208 | D_ASSERT(epoch->flags == 0); | |
b411b363 | 1209 | |
81e84650 | 1210 | return true; |
2451fc3b PR |
1211 | default: |
1212 | dev_err(DEV, "Strangeness in mdev->write_ordering %d\n", mdev->write_ordering); | |
81e84650 | 1213 | return false; |
b411b363 PR |
1214 | } |
1215 | ||
1216 | epoch->flags = 0; | |
1217 | atomic_set(&epoch->epoch_size, 0); | |
1218 | atomic_set(&epoch->active, 0); | |
1219 | ||
1220 | spin_lock(&mdev->epoch_lock); | |
1221 | if (atomic_read(&mdev->current_epoch->epoch_size)) { | |
1222 | list_add(&epoch->list, &mdev->current_epoch->list); | |
1223 | mdev->current_epoch = epoch; | |
1224 | mdev->epochs++; | |
b411b363 PR |
1225 | } else { |
1226 | /* The current_epoch got recycled while we allocated this one... */ | |
1227 | kfree(epoch); | |
1228 | } | |
1229 | spin_unlock(&mdev->epoch_lock); | |
1230 | ||
81e84650 | 1231 | return true; |
b411b363 PR |
1232 | } |
1233 | ||
1234 | /* used from receive_RSDataReply (recv_resync_read) | |
1235 | * and from receive_Data */ | |
1236 | static struct drbd_epoch_entry * | |
1237 | read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, int data_size) __must_hold(local) | |
1238 | { | |
6666032a | 1239 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); |
b411b363 | 1240 | struct drbd_epoch_entry *e; |
b411b363 | 1241 | struct page *page; |
45bb912b | 1242 | int dgs, ds, rr; |
b411b363 PR |
1243 | void *dig_in = mdev->int_dig_in; |
1244 | void *dig_vv = mdev->int_dig_vv; | |
6b4388ac | 1245 | unsigned long *data; |
b411b363 PR |
1246 | |
1247 | dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ? | |
1248 | crypto_hash_digestsize(mdev->integrity_r_tfm) : 0; | |
1249 | ||
1250 | if (dgs) { | |
1251 | rr = drbd_recv(mdev, dig_in, dgs); | |
1252 | if (rr != dgs) { | |
0ddc5549 LE |
1253 | if (!signal_pending(current)) |
1254 | dev_warn(DEV, | |
1255 | "short read receiving data digest: read %d expected %d\n", | |
1256 | rr, dgs); | |
b411b363 PR |
1257 | return NULL; |
1258 | } | |
1259 | } | |
1260 | ||
1261 | data_size -= dgs; | |
1262 | ||
d07c9c10 | 1263 | ERR_IF(data_size == 0) return NULL; |
b411b363 | 1264 | ERR_IF(data_size & 0x1ff) return NULL; |
1816a2b4 | 1265 | ERR_IF(data_size > DRBD_MAX_BIO_SIZE) return NULL; |
b411b363 | 1266 | |
6666032a LE |
1267 | /* even though we trust out peer, |
1268 | * we sometimes have to double check. */ | |
1269 | if (sector + (data_size>>9) > capacity) { | |
fdda6544 LE |
1270 | dev_err(DEV, "request from peer beyond end of local disk: " |
1271 | "capacity: %llus < sector: %llus + size: %u\n", | |
6666032a LE |
1272 | (unsigned long long)capacity, |
1273 | (unsigned long long)sector, data_size); | |
1274 | return NULL; | |
1275 | } | |
1276 | ||
b411b363 PR |
1277 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD |
1278 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
1279 | * which in turn might block on the other node at this very place. */ | |
1280 | e = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO); | |
1281 | if (!e) | |
1282 | return NULL; | |
45bb912b | 1283 | |
b411b363 | 1284 | ds = data_size; |
45bb912b LE |
1285 | page = e->pages; |
1286 | page_chain_for_each(page) { | |
1287 | unsigned len = min_t(int, ds, PAGE_SIZE); | |
6b4388ac | 1288 | data = kmap(page); |
45bb912b | 1289 | rr = drbd_recv(mdev, data, len); |
0cf9d27e | 1290 | if (drbd_insert_fault(mdev, DRBD_FAULT_RECEIVE)) { |
6b4388ac PR |
1291 | dev_err(DEV, "Fault injection: Corrupting data on receive\n"); |
1292 | data[0] = data[0] ^ (unsigned long)-1; | |
1293 | } | |
b411b363 | 1294 | kunmap(page); |
45bb912b | 1295 | if (rr != len) { |
b411b363 | 1296 | drbd_free_ee(mdev, e); |
0ddc5549 LE |
1297 | if (!signal_pending(current)) |
1298 | dev_warn(DEV, "short read receiving data: read %d expected %d\n", | |
1299 | rr, len); | |
b411b363 PR |
1300 | return NULL; |
1301 | } | |
1302 | ds -= rr; | |
1303 | } | |
1304 | ||
1305 | if (dgs) { | |
45bb912b | 1306 | drbd_csum_ee(mdev, mdev->integrity_r_tfm, e, dig_vv); |
b411b363 | 1307 | if (memcmp(dig_in, dig_vv, dgs)) { |
470be44a LE |
1308 | dev_err(DEV, "Digest integrity check FAILED: %llus +%u\n", |
1309 | (unsigned long long)sector, data_size); | |
b411b363 PR |
1310 | drbd_bcast_ee(mdev, "digest failed", |
1311 | dgs, dig_in, dig_vv, e); | |
1312 | drbd_free_ee(mdev, e); | |
1313 | return NULL; | |
1314 | } | |
1315 | } | |
1316 | mdev->recv_cnt += data_size>>9; | |
1317 | return e; | |
1318 | } | |
1319 | ||
1320 | /* drbd_drain_block() just takes a data block | |
1321 | * out of the socket input buffer, and discards it. | |
1322 | */ | |
1323 | static int drbd_drain_block(struct drbd_conf *mdev, int data_size) | |
1324 | { | |
1325 | struct page *page; | |
1326 | int rr, rv = 1; | |
1327 | void *data; | |
1328 | ||
c3470cde | 1329 | if (!data_size) |
81e84650 | 1330 | return true; |
c3470cde | 1331 | |
45bb912b | 1332 | page = drbd_pp_alloc(mdev, 1, 1); |
b411b363 PR |
1333 | |
1334 | data = kmap(page); | |
1335 | while (data_size) { | |
1336 | rr = drbd_recv(mdev, data, min_t(int, data_size, PAGE_SIZE)); | |
1337 | if (rr != min_t(int, data_size, PAGE_SIZE)) { | |
1338 | rv = 0; | |
0ddc5549 LE |
1339 | if (!signal_pending(current)) |
1340 | dev_warn(DEV, | |
1341 | "short read receiving data: read %d expected %d\n", | |
1342 | rr, min_t(int, data_size, PAGE_SIZE)); | |
b411b363 PR |
1343 | break; |
1344 | } | |
1345 | data_size -= rr; | |
1346 | } | |
1347 | kunmap(page); | |
435f0740 | 1348 | drbd_pp_free(mdev, page, 0); |
b411b363 PR |
1349 | return rv; |
1350 | } | |
1351 | ||
1352 | static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req, | |
1353 | sector_t sector, int data_size) | |
1354 | { | |
1355 | struct bio_vec *bvec; | |
1356 | struct bio *bio; | |
1357 | int dgs, rr, i, expect; | |
1358 | void *dig_in = mdev->int_dig_in; | |
1359 | void *dig_vv = mdev->int_dig_vv; | |
1360 | ||
1361 | dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ? | |
1362 | crypto_hash_digestsize(mdev->integrity_r_tfm) : 0; | |
1363 | ||
1364 | if (dgs) { | |
1365 | rr = drbd_recv(mdev, dig_in, dgs); | |
1366 | if (rr != dgs) { | |
0ddc5549 LE |
1367 | if (!signal_pending(current)) |
1368 | dev_warn(DEV, | |
1369 | "short read receiving data reply digest: read %d expected %d\n", | |
1370 | rr, dgs); | |
b411b363 PR |
1371 | return 0; |
1372 | } | |
1373 | } | |
1374 | ||
1375 | data_size -= dgs; | |
1376 | ||
1377 | /* optimistically update recv_cnt. if receiving fails below, | |
1378 | * we disconnect anyways, and counters will be reset. */ | |
1379 | mdev->recv_cnt += data_size>>9; | |
1380 | ||
1381 | bio = req->master_bio; | |
1382 | D_ASSERT(sector == bio->bi_sector); | |
1383 | ||
1384 | bio_for_each_segment(bvec, bio, i) { | |
1385 | expect = min_t(int, data_size, bvec->bv_len); | |
1386 | rr = drbd_recv(mdev, | |
1387 | kmap(bvec->bv_page)+bvec->bv_offset, | |
1388 | expect); | |
1389 | kunmap(bvec->bv_page); | |
1390 | if (rr != expect) { | |
0ddc5549 LE |
1391 | if (!signal_pending(current)) |
1392 | dev_warn(DEV, "short read receiving data reply: " | |
1393 | "read %d expected %d\n", | |
1394 | rr, expect); | |
b411b363 PR |
1395 | return 0; |
1396 | } | |
1397 | data_size -= rr; | |
1398 | } | |
1399 | ||
1400 | if (dgs) { | |
45bb912b | 1401 | drbd_csum_bio(mdev, mdev->integrity_r_tfm, bio, dig_vv); |
b411b363 PR |
1402 | if (memcmp(dig_in, dig_vv, dgs)) { |
1403 | dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n"); | |
1404 | return 0; | |
1405 | } | |
1406 | } | |
1407 | ||
1408 | D_ASSERT(data_size == 0); | |
1409 | return 1; | |
1410 | } | |
1411 | ||
1412 | /* e_end_resync_block() is called via | |
1413 | * drbd_process_done_ee() by asender only */ | |
1414 | static int e_end_resync_block(struct drbd_conf *mdev, struct drbd_work *w, int unused) | |
1415 | { | |
1416 | struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; | |
1417 | sector_t sector = e->sector; | |
1418 | int ok; | |
1419 | ||
24c4830c | 1420 | D_ASSERT(hlist_unhashed(&e->collision)); |
b411b363 | 1421 | |
45bb912b | 1422 | if (likely((e->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1423 | drbd_set_in_sync(mdev, sector, e->size); |
1424 | ok = drbd_send_ack(mdev, P_RS_WRITE_ACK, e); | |
1425 | } else { | |
1426 | /* Record failure to sync */ | |
1427 | drbd_rs_failed_io(mdev, sector, e->size); | |
1428 | ||
1429 | ok = drbd_send_ack(mdev, P_NEG_ACK, e); | |
1430 | } | |
1431 | dec_unacked(mdev); | |
1432 | ||
1433 | return ok; | |
1434 | } | |
1435 | ||
1436 | static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local) | |
1437 | { | |
1438 | struct drbd_epoch_entry *e; | |
1439 | ||
1440 | e = read_in_block(mdev, ID_SYNCER, sector, data_size); | |
45bb912b LE |
1441 | if (!e) |
1442 | goto fail; | |
b411b363 PR |
1443 | |
1444 | dec_rs_pending(mdev); | |
1445 | ||
b411b363 PR |
1446 | inc_unacked(mdev); |
1447 | /* corresponding dec_unacked() in e_end_resync_block() | |
1448 | * respective _drbd_clear_done_ee */ | |
1449 | ||
45bb912b LE |
1450 | e->w.cb = e_end_resync_block; |
1451 | ||
b411b363 PR |
1452 | spin_lock_irq(&mdev->req_lock); |
1453 | list_add(&e->w.list, &mdev->sync_ee); | |
1454 | spin_unlock_irq(&mdev->req_lock); | |
1455 | ||
0f0601f4 | 1456 | atomic_add(data_size >> 9, &mdev->rs_sect_ev); |
45bb912b | 1457 | if (drbd_submit_ee(mdev, e, WRITE, DRBD_FAULT_RS_WR) == 0) |
81e84650 | 1458 | return true; |
b411b363 | 1459 | |
10f6d992 LE |
1460 | /* don't care for the reason here */ |
1461 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
22cc37a9 LE |
1462 | spin_lock_irq(&mdev->req_lock); |
1463 | list_del(&e->w.list); | |
1464 | spin_unlock_irq(&mdev->req_lock); | |
1465 | ||
45bb912b LE |
1466 | drbd_free_ee(mdev, e); |
1467 | fail: | |
1468 | put_ldev(mdev); | |
81e84650 | 1469 | return false; |
b411b363 PR |
1470 | } |
1471 | ||
668eebc6 AG |
1472 | static struct drbd_request * |
1473 | find_request(struct drbd_conf *mdev, | |
1474 | struct hlist_head *(*hash_slot)(struct drbd_conf *, sector_t), | |
1475 | u64 id, sector_t sector, const char *func) | |
51624585 | 1476 | { |
668eebc6 | 1477 | struct hlist_head *slot = hash_slot(mdev, sector); |
51624585 AG |
1478 | struct hlist_node *n; |
1479 | struct drbd_request *req; | |
1480 | ||
1481 | hlist_for_each_entry(req, n, slot, collision) { | |
668eebc6 AG |
1482 | if ((unsigned long)req != (unsigned long)id) |
1483 | continue; | |
1484 | if (req->sector != sector) { | |
1485 | dev_err(DEV, "%s: found request %lu but it has " | |
1486 | "wrong sector (%llus versus %llus)\n", | |
1487 | func, (unsigned long)req, | |
1488 | (unsigned long long)req->sector, | |
1489 | (unsigned long long)sector); | |
1490 | break; | |
51624585 | 1491 | } |
668eebc6 | 1492 | return req; |
51624585 AG |
1493 | } |
1494 | return NULL; | |
1495 | } | |
1496 | ||
02918be2 | 1497 | static int receive_DataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
1498 | { |
1499 | struct drbd_request *req; | |
1500 | sector_t sector; | |
b411b363 | 1501 | int ok; |
02918be2 | 1502 | struct p_data *p = &mdev->data.rbuf.data; |
b411b363 PR |
1503 | |
1504 | sector = be64_to_cpu(p->sector); | |
1505 | ||
1506 | spin_lock_irq(&mdev->req_lock); | |
ae3388da | 1507 | req = find_request(mdev, ar_hash_slot, p->block_id, sector, __func__); |
b411b363 PR |
1508 | spin_unlock_irq(&mdev->req_lock); |
1509 | if (unlikely(!req)) { | |
1510 | dev_err(DEV, "Got a corrupt block_id/sector pair(1).\n"); | |
81e84650 | 1511 | return false; |
b411b363 PR |
1512 | } |
1513 | ||
24c4830c | 1514 | /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid |
b411b363 PR |
1515 | * special casing it there for the various failure cases. |
1516 | * still no race with drbd_fail_pending_reads */ | |
1517 | ok = recv_dless_read(mdev, req, sector, data_size); | |
1518 | ||
1519 | if (ok) | |
1520 | req_mod(req, data_received); | |
1521 | /* else: nothing. handled from drbd_disconnect... | |
1522 | * I don't think we may complete this just yet | |
1523 | * in case we are "on-disconnect: freeze" */ | |
1524 | ||
1525 | return ok; | |
1526 | } | |
1527 | ||
02918be2 | 1528 | static int receive_RSDataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
1529 | { |
1530 | sector_t sector; | |
b411b363 | 1531 | int ok; |
02918be2 | 1532 | struct p_data *p = &mdev->data.rbuf.data; |
b411b363 PR |
1533 | |
1534 | sector = be64_to_cpu(p->sector); | |
1535 | D_ASSERT(p->block_id == ID_SYNCER); | |
1536 | ||
1537 | if (get_ldev(mdev)) { | |
1538 | /* data is submitted to disk within recv_resync_read. | |
1539 | * corresponding put_ldev done below on error, | |
9c50842a | 1540 | * or in drbd_endio_sec. */ |
b411b363 PR |
1541 | ok = recv_resync_read(mdev, sector, data_size); |
1542 | } else { | |
1543 | if (__ratelimit(&drbd_ratelimit_state)) | |
1544 | dev_err(DEV, "Can not write resync data to local disk.\n"); | |
1545 | ||
1546 | ok = drbd_drain_block(mdev, data_size); | |
1547 | ||
2b2bf214 | 1548 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size); |
b411b363 PR |
1549 | } |
1550 | ||
778f271d PR |
1551 | atomic_add(data_size >> 9, &mdev->rs_sect_in); |
1552 | ||
b411b363 PR |
1553 | return ok; |
1554 | } | |
1555 | ||
1556 | /* e_end_block() is called via drbd_process_done_ee(). | |
1557 | * this means this function only runs in the asender thread | |
1558 | */ | |
1559 | static int e_end_block(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
1560 | { | |
1561 | struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; | |
1562 | sector_t sector = e->sector; | |
b411b363 PR |
1563 | int ok = 1, pcmd; |
1564 | ||
b411b363 | 1565 | if (mdev->net_conf->wire_protocol == DRBD_PROT_C) { |
45bb912b | 1566 | if (likely((e->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1567 | pcmd = (mdev->state.conn >= C_SYNC_SOURCE && |
1568 | mdev->state.conn <= C_PAUSED_SYNC_T && | |
1569 | e->flags & EE_MAY_SET_IN_SYNC) ? | |
1570 | P_RS_WRITE_ACK : P_WRITE_ACK; | |
1571 | ok &= drbd_send_ack(mdev, pcmd, e); | |
1572 | if (pcmd == P_RS_WRITE_ACK) | |
1573 | drbd_set_in_sync(mdev, sector, e->size); | |
1574 | } else { | |
1575 | ok = drbd_send_ack(mdev, P_NEG_ACK, e); | |
1576 | /* we expect it to be marked out of sync anyways... | |
1577 | * maybe assert this? */ | |
1578 | } | |
1579 | dec_unacked(mdev); | |
1580 | } | |
1581 | /* we delete from the conflict detection hash _after_ we sent out the | |
1582 | * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ | |
1583 | if (mdev->net_conf->two_primaries) { | |
1584 | spin_lock_irq(&mdev->req_lock); | |
24c4830c BVA |
1585 | D_ASSERT(!hlist_unhashed(&e->collision)); |
1586 | hlist_del_init(&e->collision); | |
b411b363 PR |
1587 | spin_unlock_irq(&mdev->req_lock); |
1588 | } else { | |
24c4830c | 1589 | D_ASSERT(hlist_unhashed(&e->collision)); |
b411b363 PR |
1590 | } |
1591 | ||
1592 | drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); | |
1593 | ||
1594 | return ok; | |
1595 | } | |
1596 | ||
1597 | static int e_send_discard_ack(struct drbd_conf *mdev, struct drbd_work *w, int unused) | |
1598 | { | |
1599 | struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; | |
1600 | int ok = 1; | |
1601 | ||
1602 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
1603 | ok = drbd_send_ack(mdev, P_DISCARD_ACK, e); | |
1604 | ||
1605 | spin_lock_irq(&mdev->req_lock); | |
24c4830c BVA |
1606 | D_ASSERT(!hlist_unhashed(&e->collision)); |
1607 | hlist_del_init(&e->collision); | |
b411b363 PR |
1608 | spin_unlock_irq(&mdev->req_lock); |
1609 | ||
1610 | dec_unacked(mdev); | |
1611 | ||
1612 | return ok; | |
1613 | } | |
1614 | ||
1615 | /* Called from receive_Data. | |
1616 | * Synchronize packets on sock with packets on msock. | |
1617 | * | |
1618 | * This is here so even when a P_DATA packet traveling via sock overtook an Ack | |
1619 | * packet traveling on msock, they are still processed in the order they have | |
1620 | * been sent. | |
1621 | * | |
1622 | * Note: we don't care for Ack packets overtaking P_DATA packets. | |
1623 | * | |
1624 | * In case packet_seq is larger than mdev->peer_seq number, there are | |
1625 | * outstanding packets on the msock. We wait for them to arrive. | |
1626 | * In case we are the logically next packet, we update mdev->peer_seq | |
1627 | * ourselves. Correctly handles 32bit wrap around. | |
1628 | * | |
1629 | * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, | |
1630 | * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds | |
1631 | * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have | |
1632 | * 1<<9 == 512 seconds aka ages for the 32bit wrap around... | |
1633 | * | |
1634 | * returns 0 if we may process the packet, | |
1635 | * -ERESTARTSYS if we were interrupted (by disconnect signal). */ | |
1636 | static int drbd_wait_peer_seq(struct drbd_conf *mdev, const u32 packet_seq) | |
1637 | { | |
1638 | DEFINE_WAIT(wait); | |
1639 | unsigned int p_seq; | |
1640 | long timeout; | |
1641 | int ret = 0; | |
1642 | spin_lock(&mdev->peer_seq_lock); | |
1643 | for (;;) { | |
1644 | prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE); | |
1645 | if (seq_le(packet_seq, mdev->peer_seq+1)) | |
1646 | break; | |
1647 | if (signal_pending(current)) { | |
1648 | ret = -ERESTARTSYS; | |
1649 | break; | |
1650 | } | |
1651 | p_seq = mdev->peer_seq; | |
1652 | spin_unlock(&mdev->peer_seq_lock); | |
1653 | timeout = schedule_timeout(30*HZ); | |
1654 | spin_lock(&mdev->peer_seq_lock); | |
1655 | if (timeout == 0 && p_seq == mdev->peer_seq) { | |
1656 | ret = -ETIMEDOUT; | |
1657 | dev_err(DEV, "ASSERT FAILED waited 30 seconds for sequence update, forcing reconnect\n"); | |
1658 | break; | |
1659 | } | |
1660 | } | |
1661 | finish_wait(&mdev->seq_wait, &wait); | |
1662 | if (mdev->peer_seq+1 == packet_seq) | |
1663 | mdev->peer_seq++; | |
1664 | spin_unlock(&mdev->peer_seq_lock); | |
1665 | return ret; | |
1666 | } | |
1667 | ||
688593c5 LE |
1668 | /* see also bio_flags_to_wire() |
1669 | * DRBD_REQ_*, because we need to semantically map the flags to data packet | |
1670 | * flags and back. We may replicate to other kernel versions. */ | |
1671 | static unsigned long wire_flags_to_bio(struct drbd_conf *mdev, u32 dpf) | |
76d2e7ec | 1672 | { |
688593c5 LE |
1673 | return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | |
1674 | (dpf & DP_FUA ? REQ_FUA : 0) | | |
1675 | (dpf & DP_FLUSH ? REQ_FLUSH : 0) | | |
1676 | (dpf & DP_DISCARD ? REQ_DISCARD : 0); | |
76d2e7ec PR |
1677 | } |
1678 | ||
b411b363 | 1679 | /* mirrored write */ |
02918be2 | 1680 | static int receive_Data(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
1681 | { |
1682 | sector_t sector; | |
1683 | struct drbd_epoch_entry *e; | |
02918be2 | 1684 | struct p_data *p = &mdev->data.rbuf.data; |
b411b363 PR |
1685 | int rw = WRITE; |
1686 | u32 dp_flags; | |
1687 | ||
b411b363 | 1688 | if (!get_ldev(mdev)) { |
b411b363 PR |
1689 | spin_lock(&mdev->peer_seq_lock); |
1690 | if (mdev->peer_seq+1 == be32_to_cpu(p->seq_num)) | |
1691 | mdev->peer_seq++; | |
1692 | spin_unlock(&mdev->peer_seq_lock); | |
1693 | ||
2b2bf214 | 1694 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size); |
b411b363 PR |
1695 | atomic_inc(&mdev->current_epoch->epoch_size); |
1696 | return drbd_drain_block(mdev, data_size); | |
1697 | } | |
1698 | ||
1699 | /* get_ldev(mdev) successful. | |
1700 | * Corresponding put_ldev done either below (on various errors), | |
9c50842a | 1701 | * or in drbd_endio_sec, if we successfully submit the data at |
b411b363 PR |
1702 | * the end of this function. */ |
1703 | ||
1704 | sector = be64_to_cpu(p->sector); | |
1705 | e = read_in_block(mdev, p->block_id, sector, data_size); | |
1706 | if (!e) { | |
1707 | put_ldev(mdev); | |
81e84650 | 1708 | return false; |
b411b363 PR |
1709 | } |
1710 | ||
b411b363 PR |
1711 | e->w.cb = e_end_block; |
1712 | ||
688593c5 LE |
1713 | dp_flags = be32_to_cpu(p->dp_flags); |
1714 | rw |= wire_flags_to_bio(mdev, dp_flags); | |
1715 | ||
1716 | if (dp_flags & DP_MAY_SET_IN_SYNC) | |
1717 | e->flags |= EE_MAY_SET_IN_SYNC; | |
1718 | ||
b411b363 PR |
1719 | spin_lock(&mdev->epoch_lock); |
1720 | e->epoch = mdev->current_epoch; | |
1721 | atomic_inc(&e->epoch->epoch_size); | |
1722 | atomic_inc(&e->epoch->active); | |
b411b363 PR |
1723 | spin_unlock(&mdev->epoch_lock); |
1724 | ||
b411b363 PR |
1725 | /* I'm the receiver, I do hold a net_cnt reference. */ |
1726 | if (!mdev->net_conf->two_primaries) { | |
1727 | spin_lock_irq(&mdev->req_lock); | |
1728 | } else { | |
1729 | /* don't get the req_lock yet, | |
1730 | * we may sleep in drbd_wait_peer_seq */ | |
1731 | const int size = e->size; | |
1732 | const int discard = test_bit(DISCARD_CONCURRENT, &mdev->flags); | |
1733 | DEFINE_WAIT(wait); | |
1734 | struct drbd_request *i; | |
1735 | struct hlist_node *n; | |
1736 | struct hlist_head *slot; | |
1737 | int first; | |
1738 | ||
1739 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
1740 | BUG_ON(mdev->ee_hash == NULL); | |
1741 | BUG_ON(mdev->tl_hash == NULL); | |
1742 | ||
1743 | /* conflict detection and handling: | |
1744 | * 1. wait on the sequence number, | |
1745 | * in case this data packet overtook ACK packets. | |
1746 | * 2. check our hash tables for conflicting requests. | |
1747 | * we only need to walk the tl_hash, since an ee can not | |
1748 | * have a conflict with an other ee: on the submitting | |
1749 | * node, the corresponding req had already been conflicting, | |
1750 | * and a conflicting req is never sent. | |
1751 | * | |
1752 | * Note: for two_primaries, we are protocol C, | |
1753 | * so there cannot be any request that is DONE | |
1754 | * but still on the transfer log. | |
1755 | * | |
1756 | * unconditionally add to the ee_hash. | |
1757 | * | |
1758 | * if no conflicting request is found: | |
1759 | * submit. | |
1760 | * | |
1761 | * if any conflicting request is found | |
1762 | * that has not yet been acked, | |
1763 | * AND I have the "discard concurrent writes" flag: | |
1764 | * queue (via done_ee) the P_DISCARD_ACK; OUT. | |
1765 | * | |
1766 | * if any conflicting request is found: | |
1767 | * block the receiver, waiting on misc_wait | |
1768 | * until no more conflicting requests are there, | |
1769 | * or we get interrupted (disconnect). | |
1770 | * | |
1771 | * we do not just write after local io completion of those | |
1772 | * requests, but only after req is done completely, i.e. | |
1773 | * we wait for the P_DISCARD_ACK to arrive! | |
1774 | * | |
1775 | * then proceed normally, i.e. submit. | |
1776 | */ | |
1777 | if (drbd_wait_peer_seq(mdev, be32_to_cpu(p->seq_num))) | |
1778 | goto out_interrupted; | |
1779 | ||
1780 | spin_lock_irq(&mdev->req_lock); | |
1781 | ||
24c4830c | 1782 | hlist_add_head(&e->collision, ee_hash_slot(mdev, sector)); |
b411b363 PR |
1783 | |
1784 | #define OVERLAPS overlaps(i->sector, i->size, sector, size) | |
1785 | slot = tl_hash_slot(mdev, sector); | |
1786 | first = 1; | |
1787 | for (;;) { | |
1788 | int have_unacked = 0; | |
1789 | int have_conflict = 0; | |
1790 | prepare_to_wait(&mdev->misc_wait, &wait, | |
1791 | TASK_INTERRUPTIBLE); | |
24c4830c | 1792 | hlist_for_each_entry(i, n, slot, collision) { |
b411b363 PR |
1793 | if (OVERLAPS) { |
1794 | /* only ALERT on first iteration, | |
1795 | * we may be woken up early... */ | |
1796 | if (first) | |
1797 | dev_alert(DEV, "%s[%u] Concurrent local write detected!" | |
1798 | " new: %llus +%u; pending: %llus +%u\n", | |
1799 | current->comm, current->pid, | |
1800 | (unsigned long long)sector, size, | |
1801 | (unsigned long long)i->sector, i->size); | |
1802 | if (i->rq_state & RQ_NET_PENDING) | |
1803 | ++have_unacked; | |
1804 | ++have_conflict; | |
1805 | } | |
1806 | } | |
1807 | #undef OVERLAPS | |
1808 | if (!have_conflict) | |
1809 | break; | |
1810 | ||
1811 | /* Discard Ack only for the _first_ iteration */ | |
1812 | if (first && discard && have_unacked) { | |
1813 | dev_alert(DEV, "Concurrent write! [DISCARD BY FLAG] sec=%llus\n", | |
1814 | (unsigned long long)sector); | |
1815 | inc_unacked(mdev); | |
1816 | e->w.cb = e_send_discard_ack; | |
1817 | list_add_tail(&e->w.list, &mdev->done_ee); | |
1818 | ||
1819 | spin_unlock_irq(&mdev->req_lock); | |
1820 | ||
1821 | /* we could probably send that P_DISCARD_ACK ourselves, | |
1822 | * but I don't like the receiver using the msock */ | |
1823 | ||
1824 | put_ldev(mdev); | |
1825 | wake_asender(mdev); | |
1826 | finish_wait(&mdev->misc_wait, &wait); | |
81e84650 | 1827 | return true; |
b411b363 PR |
1828 | } |
1829 | ||
1830 | if (signal_pending(current)) { | |
24c4830c | 1831 | hlist_del_init(&e->collision); |
b411b363 PR |
1832 | |
1833 | spin_unlock_irq(&mdev->req_lock); | |
1834 | ||
1835 | finish_wait(&mdev->misc_wait, &wait); | |
1836 | goto out_interrupted; | |
1837 | } | |
1838 | ||
1839 | spin_unlock_irq(&mdev->req_lock); | |
1840 | if (first) { | |
1841 | first = 0; | |
1842 | dev_alert(DEV, "Concurrent write! [W AFTERWARDS] " | |
1843 | "sec=%llus\n", (unsigned long long)sector); | |
1844 | } else if (discard) { | |
1845 | /* we had none on the first iteration. | |
1846 | * there must be none now. */ | |
1847 | D_ASSERT(have_unacked == 0); | |
1848 | } | |
1849 | schedule(); | |
1850 | spin_lock_irq(&mdev->req_lock); | |
1851 | } | |
1852 | finish_wait(&mdev->misc_wait, &wait); | |
1853 | } | |
1854 | ||
1855 | list_add(&e->w.list, &mdev->active_ee); | |
1856 | spin_unlock_irq(&mdev->req_lock); | |
1857 | ||
1858 | switch (mdev->net_conf->wire_protocol) { | |
1859 | case DRBD_PROT_C: | |
1860 | inc_unacked(mdev); | |
1861 | /* corresponding dec_unacked() in e_end_block() | |
1862 | * respective _drbd_clear_done_ee */ | |
1863 | break; | |
1864 | case DRBD_PROT_B: | |
1865 | /* I really don't like it that the receiver thread | |
1866 | * sends on the msock, but anyways */ | |
1867 | drbd_send_ack(mdev, P_RECV_ACK, e); | |
1868 | break; | |
1869 | case DRBD_PROT_A: | |
1870 | /* nothing to do */ | |
1871 | break; | |
1872 | } | |
1873 | ||
6719fb03 | 1874 | if (mdev->state.pdsk < D_INCONSISTENT) { |
b411b363 PR |
1875 | /* In case we have the only disk of the cluster, */ |
1876 | drbd_set_out_of_sync(mdev, e->sector, e->size); | |
1877 | e->flags |= EE_CALL_AL_COMPLETE_IO; | |
6719fb03 | 1878 | e->flags &= ~EE_MAY_SET_IN_SYNC; |
b411b363 PR |
1879 | drbd_al_begin_io(mdev, e->sector); |
1880 | } | |
1881 | ||
45bb912b | 1882 | if (drbd_submit_ee(mdev, e, rw, DRBD_FAULT_DT_WR) == 0) |
81e84650 | 1883 | return true; |
b411b363 | 1884 | |
10f6d992 LE |
1885 | /* don't care for the reason here */ |
1886 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
22cc37a9 LE |
1887 | spin_lock_irq(&mdev->req_lock); |
1888 | list_del(&e->w.list); | |
24c4830c | 1889 | hlist_del_init(&e->collision); |
22cc37a9 LE |
1890 | spin_unlock_irq(&mdev->req_lock); |
1891 | if (e->flags & EE_CALL_AL_COMPLETE_IO) | |
1892 | drbd_al_complete_io(mdev, e->sector); | |
1893 | ||
b411b363 | 1894 | out_interrupted: |
10f6d992 | 1895 | drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + EV_CLEANUP); |
b411b363 PR |
1896 | put_ldev(mdev); |
1897 | drbd_free_ee(mdev, e); | |
81e84650 | 1898 | return false; |
b411b363 PR |
1899 | } |
1900 | ||
0f0601f4 LE |
1901 | /* We may throttle resync, if the lower device seems to be busy, |
1902 | * and current sync rate is above c_min_rate. | |
1903 | * | |
1904 | * To decide whether or not the lower device is busy, we use a scheme similar | |
1905 | * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" | |
1906 | * (more than 64 sectors) of activity we cannot account for with our own resync | |
1907 | * activity, it obviously is "busy". | |
1908 | * | |
1909 | * The current sync rate used here uses only the most recent two step marks, | |
1910 | * to have a short time average so we can react faster. | |
1911 | */ | |
e3555d85 | 1912 | int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector) |
0f0601f4 LE |
1913 | { |
1914 | struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk; | |
1915 | unsigned long db, dt, dbdt; | |
e3555d85 | 1916 | struct lc_element *tmp; |
0f0601f4 LE |
1917 | int curr_events; |
1918 | int throttle = 0; | |
1919 | ||
1920 | /* feature disabled? */ | |
1921 | if (mdev->sync_conf.c_min_rate == 0) | |
1922 | return 0; | |
1923 | ||
e3555d85 PR |
1924 | spin_lock_irq(&mdev->al_lock); |
1925 | tmp = lc_find(mdev->resync, BM_SECT_TO_EXT(sector)); | |
1926 | if (tmp) { | |
1927 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
1928 | if (test_bit(BME_PRIORITY, &bm_ext->flags)) { | |
1929 | spin_unlock_irq(&mdev->al_lock); | |
1930 | return 0; | |
1931 | } | |
1932 | /* Do not slow down if app IO is already waiting for this extent */ | |
1933 | } | |
1934 | spin_unlock_irq(&mdev->al_lock); | |
1935 | ||
0f0601f4 LE |
1936 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
1937 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
1938 | atomic_read(&mdev->rs_sect_ev); | |
e3555d85 | 1939 | |
0f0601f4 LE |
1940 | if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) { |
1941 | unsigned long rs_left; | |
1942 | int i; | |
1943 | ||
1944 | mdev->rs_last_events = curr_events; | |
1945 | ||
1946 | /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, | |
1947 | * approx. */ | |
2649f080 LE |
1948 | i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS; |
1949 | ||
1950 | if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T) | |
1951 | rs_left = mdev->ov_left; | |
1952 | else | |
1953 | rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; | |
0f0601f4 LE |
1954 | |
1955 | dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ; | |
1956 | if (!dt) | |
1957 | dt++; | |
1958 | db = mdev->rs_mark_left[i] - rs_left; | |
1959 | dbdt = Bit2KB(db/dt); | |
1960 | ||
1961 | if (dbdt > mdev->sync_conf.c_min_rate) | |
1962 | throttle = 1; | |
1963 | } | |
1964 | return throttle; | |
1965 | } | |
1966 | ||
1967 | ||
02918be2 | 1968 | static int receive_DataRequest(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int digest_size) |
b411b363 PR |
1969 | { |
1970 | sector_t sector; | |
1971 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); | |
1972 | struct drbd_epoch_entry *e; | |
1973 | struct digest_info *di = NULL; | |
b18b37be | 1974 | int size, verb; |
b411b363 | 1975 | unsigned int fault_type; |
02918be2 | 1976 | struct p_block_req *p = &mdev->data.rbuf.block_req; |
b411b363 PR |
1977 | |
1978 | sector = be64_to_cpu(p->sector); | |
1979 | size = be32_to_cpu(p->blksize); | |
1980 | ||
1816a2b4 | 1981 | if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_BIO_SIZE) { |
b411b363 PR |
1982 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, |
1983 | (unsigned long long)sector, size); | |
81e84650 | 1984 | return false; |
b411b363 PR |
1985 | } |
1986 | if (sector + (size>>9) > capacity) { | |
1987 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, | |
1988 | (unsigned long long)sector, size); | |
81e84650 | 1989 | return false; |
b411b363 PR |
1990 | } |
1991 | ||
1992 | if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) { | |
b18b37be PR |
1993 | verb = 1; |
1994 | switch (cmd) { | |
1995 | case P_DATA_REQUEST: | |
1996 | drbd_send_ack_rp(mdev, P_NEG_DREPLY, p); | |
1997 | break; | |
1998 | case P_RS_DATA_REQUEST: | |
1999 | case P_CSUM_RS_REQUEST: | |
2000 | case P_OV_REQUEST: | |
2001 | drbd_send_ack_rp(mdev, P_NEG_RS_DREPLY , p); | |
2002 | break; | |
2003 | case P_OV_REPLY: | |
2004 | verb = 0; | |
2005 | dec_rs_pending(mdev); | |
2006 | drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, ID_IN_SYNC); | |
2007 | break; | |
2008 | default: | |
2009 | dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", | |
2010 | cmdname(cmd)); | |
2011 | } | |
2012 | if (verb && __ratelimit(&drbd_ratelimit_state)) | |
b411b363 PR |
2013 | dev_err(DEV, "Can not satisfy peer's read request, " |
2014 | "no local data.\n"); | |
b18b37be | 2015 | |
a821cc4a LE |
2016 | /* drain possibly payload */ |
2017 | return drbd_drain_block(mdev, digest_size); | |
b411b363 PR |
2018 | } |
2019 | ||
2020 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD | |
2021 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
2022 | * which in turn might block on the other node at this very place. */ | |
2023 | e = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO); | |
2024 | if (!e) { | |
2025 | put_ldev(mdev); | |
81e84650 | 2026 | return false; |
b411b363 PR |
2027 | } |
2028 | ||
02918be2 | 2029 | switch (cmd) { |
b411b363 PR |
2030 | case P_DATA_REQUEST: |
2031 | e->w.cb = w_e_end_data_req; | |
2032 | fault_type = DRBD_FAULT_DT_RD; | |
80a40e43 LE |
2033 | /* application IO, don't drbd_rs_begin_io */ |
2034 | goto submit; | |
2035 | ||
b411b363 PR |
2036 | case P_RS_DATA_REQUEST: |
2037 | e->w.cb = w_e_end_rsdata_req; | |
2038 | fault_type = DRBD_FAULT_RS_RD; | |
5f9915bb LE |
2039 | /* used in the sector offset progress display */ |
2040 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
b411b363 PR |
2041 | break; |
2042 | ||
2043 | case P_OV_REPLY: | |
2044 | case P_CSUM_RS_REQUEST: | |
2045 | fault_type = DRBD_FAULT_RS_RD; | |
b411b363 PR |
2046 | di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO); |
2047 | if (!di) | |
2048 | goto out_free_e; | |
2049 | ||
2050 | di->digest_size = digest_size; | |
2051 | di->digest = (((char *)di)+sizeof(struct digest_info)); | |
2052 | ||
c36c3ced LE |
2053 | e->digest = di; |
2054 | e->flags |= EE_HAS_DIGEST; | |
2055 | ||
b411b363 PR |
2056 | if (drbd_recv(mdev, di->digest, digest_size) != digest_size) |
2057 | goto out_free_e; | |
2058 | ||
02918be2 | 2059 | if (cmd == P_CSUM_RS_REQUEST) { |
b411b363 PR |
2060 | D_ASSERT(mdev->agreed_pro_version >= 89); |
2061 | e->w.cb = w_e_end_csum_rs_req; | |
5f9915bb LE |
2062 | /* used in the sector offset progress display */ |
2063 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
02918be2 | 2064 | } else if (cmd == P_OV_REPLY) { |
2649f080 LE |
2065 | /* track progress, we may need to throttle */ |
2066 | atomic_add(size >> 9, &mdev->rs_sect_in); | |
b411b363 PR |
2067 | e->w.cb = w_e_end_ov_reply; |
2068 | dec_rs_pending(mdev); | |
0f0601f4 LE |
2069 | /* drbd_rs_begin_io done when we sent this request, |
2070 | * but accounting still needs to be done. */ | |
2071 | goto submit_for_resync; | |
b411b363 PR |
2072 | } |
2073 | break; | |
2074 | ||
2075 | case P_OV_REQUEST: | |
b411b363 PR |
2076 | if (mdev->ov_start_sector == ~(sector_t)0 && |
2077 | mdev->agreed_pro_version >= 90) { | |
de228bba LE |
2078 | unsigned long now = jiffies; |
2079 | int i; | |
b411b363 PR |
2080 | mdev->ov_start_sector = sector; |
2081 | mdev->ov_position = sector; | |
30b743a2 LE |
2082 | mdev->ov_left = drbd_bm_bits(mdev) - BM_SECT_TO_BIT(sector); |
2083 | mdev->rs_total = mdev->ov_left; | |
de228bba LE |
2084 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
2085 | mdev->rs_mark_left[i] = mdev->ov_left; | |
2086 | mdev->rs_mark_time[i] = now; | |
2087 | } | |
b411b363 PR |
2088 | dev_info(DEV, "Online Verify start sector: %llu\n", |
2089 | (unsigned long long)sector); | |
2090 | } | |
2091 | e->w.cb = w_e_end_ov_req; | |
2092 | fault_type = DRBD_FAULT_RS_RD; | |
b411b363 PR |
2093 | break; |
2094 | ||
b411b363 PR |
2095 | default: |
2096 | dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", | |
02918be2 | 2097 | cmdname(cmd)); |
b411b363 | 2098 | fault_type = DRBD_FAULT_MAX; |
80a40e43 | 2099 | goto out_free_e; |
b411b363 PR |
2100 | } |
2101 | ||
0f0601f4 LE |
2102 | /* Throttle, drbd_rs_begin_io and submit should become asynchronous |
2103 | * wrt the receiver, but it is not as straightforward as it may seem. | |
2104 | * Various places in the resync start and stop logic assume resync | |
2105 | * requests are processed in order, requeuing this on the worker thread | |
2106 | * introduces a bunch of new code for synchronization between threads. | |
2107 | * | |
2108 | * Unlimited throttling before drbd_rs_begin_io may stall the resync | |
2109 | * "forever", throttling after drbd_rs_begin_io will lock that extent | |
2110 | * for application writes for the same time. For now, just throttle | |
2111 | * here, where the rest of the code expects the receiver to sleep for | |
2112 | * a while, anyways. | |
2113 | */ | |
2114 | ||
2115 | /* Throttle before drbd_rs_begin_io, as that locks out application IO; | |
2116 | * this defers syncer requests for some time, before letting at least | |
2117 | * on request through. The resync controller on the receiving side | |
2118 | * will adapt to the incoming rate accordingly. | |
2119 | * | |
2120 | * We cannot throttle here if remote is Primary/SyncTarget: | |
2121 | * we would also throttle its application reads. | |
2122 | * In that case, throttling is done on the SyncTarget only. | |
2123 | */ | |
e3555d85 PR |
2124 | if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev, sector)) |
2125 | schedule_timeout_uninterruptible(HZ/10); | |
2126 | if (drbd_rs_begin_io(mdev, sector)) | |
80a40e43 | 2127 | goto out_free_e; |
b411b363 | 2128 | |
0f0601f4 LE |
2129 | submit_for_resync: |
2130 | atomic_add(size >> 9, &mdev->rs_sect_ev); | |
2131 | ||
80a40e43 | 2132 | submit: |
b411b363 | 2133 | inc_unacked(mdev); |
80a40e43 LE |
2134 | spin_lock_irq(&mdev->req_lock); |
2135 | list_add_tail(&e->w.list, &mdev->read_ee); | |
2136 | spin_unlock_irq(&mdev->req_lock); | |
b411b363 | 2137 | |
45bb912b | 2138 | if (drbd_submit_ee(mdev, e, READ, fault_type) == 0) |
81e84650 | 2139 | return true; |
b411b363 | 2140 | |
10f6d992 LE |
2141 | /* don't care for the reason here */ |
2142 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
22cc37a9 LE |
2143 | spin_lock_irq(&mdev->req_lock); |
2144 | list_del(&e->w.list); | |
2145 | spin_unlock_irq(&mdev->req_lock); | |
2146 | /* no drbd_rs_complete_io(), we are dropping the connection anyways */ | |
2147 | ||
b411b363 | 2148 | out_free_e: |
b411b363 PR |
2149 | put_ldev(mdev); |
2150 | drbd_free_ee(mdev, e); | |
81e84650 | 2151 | return false; |
b411b363 PR |
2152 | } |
2153 | ||
2154 | static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local) | |
2155 | { | |
2156 | int self, peer, rv = -100; | |
2157 | unsigned long ch_self, ch_peer; | |
2158 | ||
2159 | self = mdev->ldev->md.uuid[UI_BITMAP] & 1; | |
2160 | peer = mdev->p_uuid[UI_BITMAP] & 1; | |
2161 | ||
2162 | ch_peer = mdev->p_uuid[UI_SIZE]; | |
2163 | ch_self = mdev->comm_bm_set; | |
2164 | ||
2165 | switch (mdev->net_conf->after_sb_0p) { | |
2166 | case ASB_CONSENSUS: | |
2167 | case ASB_DISCARD_SECONDARY: | |
2168 | case ASB_CALL_HELPER: | |
2169 | dev_err(DEV, "Configuration error.\n"); | |
2170 | break; | |
2171 | case ASB_DISCONNECT: | |
2172 | break; | |
2173 | case ASB_DISCARD_YOUNGER_PRI: | |
2174 | if (self == 0 && peer == 1) { | |
2175 | rv = -1; | |
2176 | break; | |
2177 | } | |
2178 | if (self == 1 && peer == 0) { | |
2179 | rv = 1; | |
2180 | break; | |
2181 | } | |
2182 | /* Else fall through to one of the other strategies... */ | |
2183 | case ASB_DISCARD_OLDER_PRI: | |
2184 | if (self == 0 && peer == 1) { | |
2185 | rv = 1; | |
2186 | break; | |
2187 | } | |
2188 | if (self == 1 && peer == 0) { | |
2189 | rv = -1; | |
2190 | break; | |
2191 | } | |
2192 | /* Else fall through to one of the other strategies... */ | |
ad19bf6e | 2193 | dev_warn(DEV, "Discard younger/older primary did not find a decision\n" |
b411b363 PR |
2194 | "Using discard-least-changes instead\n"); |
2195 | case ASB_DISCARD_ZERO_CHG: | |
2196 | if (ch_peer == 0 && ch_self == 0) { | |
2197 | rv = test_bit(DISCARD_CONCURRENT, &mdev->flags) | |
2198 | ? -1 : 1; | |
2199 | break; | |
2200 | } else { | |
2201 | if (ch_peer == 0) { rv = 1; break; } | |
2202 | if (ch_self == 0) { rv = -1; break; } | |
2203 | } | |
2204 | if (mdev->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG) | |
2205 | break; | |
2206 | case ASB_DISCARD_LEAST_CHG: | |
2207 | if (ch_self < ch_peer) | |
2208 | rv = -1; | |
2209 | else if (ch_self > ch_peer) | |
2210 | rv = 1; | |
2211 | else /* ( ch_self == ch_peer ) */ | |
2212 | /* Well, then use something else. */ | |
2213 | rv = test_bit(DISCARD_CONCURRENT, &mdev->flags) | |
2214 | ? -1 : 1; | |
2215 | break; | |
2216 | case ASB_DISCARD_LOCAL: | |
2217 | rv = -1; | |
2218 | break; | |
2219 | case ASB_DISCARD_REMOTE: | |
2220 | rv = 1; | |
2221 | } | |
2222 | ||
2223 | return rv; | |
2224 | } | |
2225 | ||
2226 | static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local) | |
2227 | { | |
6184ea21 | 2228 | int hg, rv = -100; |
b411b363 PR |
2229 | |
2230 | switch (mdev->net_conf->after_sb_1p) { | |
2231 | case ASB_DISCARD_YOUNGER_PRI: | |
2232 | case ASB_DISCARD_OLDER_PRI: | |
2233 | case ASB_DISCARD_LEAST_CHG: | |
2234 | case ASB_DISCARD_LOCAL: | |
2235 | case ASB_DISCARD_REMOTE: | |
2236 | dev_err(DEV, "Configuration error.\n"); | |
2237 | break; | |
2238 | case ASB_DISCONNECT: | |
2239 | break; | |
2240 | case ASB_CONSENSUS: | |
2241 | hg = drbd_asb_recover_0p(mdev); | |
2242 | if (hg == -1 && mdev->state.role == R_SECONDARY) | |
2243 | rv = hg; | |
2244 | if (hg == 1 && mdev->state.role == R_PRIMARY) | |
2245 | rv = hg; | |
2246 | break; | |
2247 | case ASB_VIOLENTLY: | |
2248 | rv = drbd_asb_recover_0p(mdev); | |
2249 | break; | |
2250 | case ASB_DISCARD_SECONDARY: | |
2251 | return mdev->state.role == R_PRIMARY ? 1 : -1; | |
2252 | case ASB_CALL_HELPER: | |
2253 | hg = drbd_asb_recover_0p(mdev); | |
2254 | if (hg == -1 && mdev->state.role == R_PRIMARY) { | |
bb437946 AG |
2255 | enum drbd_state_rv rv2; |
2256 | ||
2257 | drbd_set_role(mdev, R_SECONDARY, 0); | |
b411b363 PR |
2258 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2259 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2260 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2261 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2262 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2263 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2264 | } else { | |
2265 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2266 | rv = hg; | |
2267 | } | |
2268 | } else | |
2269 | rv = hg; | |
2270 | } | |
2271 | ||
2272 | return rv; | |
2273 | } | |
2274 | ||
2275 | static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local) | |
2276 | { | |
6184ea21 | 2277 | int hg, rv = -100; |
b411b363 PR |
2278 | |
2279 | switch (mdev->net_conf->after_sb_2p) { | |
2280 | case ASB_DISCARD_YOUNGER_PRI: | |
2281 | case ASB_DISCARD_OLDER_PRI: | |
2282 | case ASB_DISCARD_LEAST_CHG: | |
2283 | case ASB_DISCARD_LOCAL: | |
2284 | case ASB_DISCARD_REMOTE: | |
2285 | case ASB_CONSENSUS: | |
2286 | case ASB_DISCARD_SECONDARY: | |
2287 | dev_err(DEV, "Configuration error.\n"); | |
2288 | break; | |
2289 | case ASB_VIOLENTLY: | |
2290 | rv = drbd_asb_recover_0p(mdev); | |
2291 | break; | |
2292 | case ASB_DISCONNECT: | |
2293 | break; | |
2294 | case ASB_CALL_HELPER: | |
2295 | hg = drbd_asb_recover_0p(mdev); | |
2296 | if (hg == -1) { | |
bb437946 AG |
2297 | enum drbd_state_rv rv2; |
2298 | ||
b411b363 PR |
2299 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2300 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2301 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2302 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2303 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2304 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2305 | } else { | |
2306 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2307 | rv = hg; | |
2308 | } | |
2309 | } else | |
2310 | rv = hg; | |
2311 | } | |
2312 | ||
2313 | return rv; | |
2314 | } | |
2315 | ||
2316 | static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid, | |
2317 | u64 bits, u64 flags) | |
2318 | { | |
2319 | if (!uuid) { | |
2320 | dev_info(DEV, "%s uuid info vanished while I was looking!\n", text); | |
2321 | return; | |
2322 | } | |
2323 | dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", | |
2324 | text, | |
2325 | (unsigned long long)uuid[UI_CURRENT], | |
2326 | (unsigned long long)uuid[UI_BITMAP], | |
2327 | (unsigned long long)uuid[UI_HISTORY_START], | |
2328 | (unsigned long long)uuid[UI_HISTORY_END], | |
2329 | (unsigned long long)bits, | |
2330 | (unsigned long long)flags); | |
2331 | } | |
2332 | ||
2333 | /* | |
2334 | 100 after split brain try auto recover | |
2335 | 2 C_SYNC_SOURCE set BitMap | |
2336 | 1 C_SYNC_SOURCE use BitMap | |
2337 | 0 no Sync | |
2338 | -1 C_SYNC_TARGET use BitMap | |
2339 | -2 C_SYNC_TARGET set BitMap | |
2340 | -100 after split brain, disconnect | |
2341 | -1000 unrelated data | |
4a23f264 PR |
2342 | -1091 requires proto 91 |
2343 | -1096 requires proto 96 | |
b411b363 PR |
2344 | */ |
2345 | static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local) | |
2346 | { | |
2347 | u64 self, peer; | |
2348 | int i, j; | |
2349 | ||
2350 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2351 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2352 | ||
2353 | *rule_nr = 10; | |
2354 | if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) | |
2355 | return 0; | |
2356 | ||
2357 | *rule_nr = 20; | |
2358 | if ((self == UUID_JUST_CREATED || self == (u64)0) && | |
2359 | peer != UUID_JUST_CREATED) | |
2360 | return -2; | |
2361 | ||
2362 | *rule_nr = 30; | |
2363 | if (self != UUID_JUST_CREATED && | |
2364 | (peer == UUID_JUST_CREATED || peer == (u64)0)) | |
2365 | return 2; | |
2366 | ||
2367 | if (self == peer) { | |
2368 | int rct, dc; /* roles at crash time */ | |
2369 | ||
2370 | if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) { | |
2371 | ||
2372 | if (mdev->agreed_pro_version < 91) | |
4a23f264 | 2373 | return -1091; |
b411b363 PR |
2374 | |
2375 | if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) && | |
2376 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { | |
2377 | dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n"); | |
2378 | drbd_uuid_set_bm(mdev, 0UL); | |
2379 | ||
2380 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, | |
2381 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2382 | *rule_nr = 34; | |
2383 | } else { | |
2384 | dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n"); | |
2385 | *rule_nr = 36; | |
2386 | } | |
2387 | ||
2388 | return 1; | |
2389 | } | |
2390 | ||
2391 | if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) { | |
2392 | ||
2393 | if (mdev->agreed_pro_version < 91) | |
4a23f264 | 2394 | return -1091; |
b411b363 PR |
2395 | |
2396 | if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) && | |
2397 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) { | |
2398 | dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); | |
2399 | ||
2400 | mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START]; | |
2401 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP]; | |
2402 | mdev->p_uuid[UI_BITMAP] = 0UL; | |
2403 | ||
2404 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2405 | *rule_nr = 35; | |
2406 | } else { | |
2407 | dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n"); | |
2408 | *rule_nr = 37; | |
2409 | } | |
2410 | ||
2411 | return -1; | |
2412 | } | |
2413 | ||
2414 | /* Common power [off|failure] */ | |
2415 | rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) + | |
2416 | (mdev->p_uuid[UI_FLAGS] & 2); | |
2417 | /* lowest bit is set when we were primary, | |
2418 | * next bit (weight 2) is set when peer was primary */ | |
2419 | *rule_nr = 40; | |
2420 | ||
2421 | switch (rct) { | |
2422 | case 0: /* !self_pri && !peer_pri */ return 0; | |
2423 | case 1: /* self_pri && !peer_pri */ return 1; | |
2424 | case 2: /* !self_pri && peer_pri */ return -1; | |
2425 | case 3: /* self_pri && peer_pri */ | |
2426 | dc = test_bit(DISCARD_CONCURRENT, &mdev->flags); | |
2427 | return dc ? -1 : 1; | |
2428 | } | |
2429 | } | |
2430 | ||
2431 | *rule_nr = 50; | |
2432 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2433 | if (self == peer) | |
2434 | return -1; | |
2435 | ||
2436 | *rule_nr = 51; | |
2437 | peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); | |
2438 | if (self == peer) { | |
4a23f264 PR |
2439 | if (mdev->agreed_pro_version < 96 ? |
2440 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == | |
2441 | (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) : | |
2442 | peer + UUID_NEW_BM_OFFSET == (mdev->p_uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2443 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2444 | resync as sync source modifications of the peer's UUIDs. */ | |
2445 | ||
2446 | if (mdev->agreed_pro_version < 91) | |
4a23f264 | 2447 | return -1091; |
b411b363 PR |
2448 | |
2449 | mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START]; | |
2450 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1]; | |
4a23f264 PR |
2451 | |
2452 | dev_info(DEV, "Did not got last syncUUID packet, corrected:\n"); | |
2453 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2454 | ||
b411b363 PR |
2455 | return -1; |
2456 | } | |
2457 | } | |
2458 | ||
2459 | *rule_nr = 60; | |
2460 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2461 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2462 | peer = mdev->p_uuid[i] & ~((u64)1); | |
2463 | if (self == peer) | |
2464 | return -2; | |
2465 | } | |
2466 | ||
2467 | *rule_nr = 70; | |
2468 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2469 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2470 | if (self == peer) | |
2471 | return 1; | |
2472 | ||
2473 | *rule_nr = 71; | |
2474 | self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); | |
2475 | if (self == peer) { | |
4a23f264 PR |
2476 | if (mdev->agreed_pro_version < 96 ? |
2477 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == | |
2478 | (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) : | |
2479 | self + UUID_NEW_BM_OFFSET == (mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2480 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2481 | resync as sync source modifications of our UUIDs. */ | |
2482 | ||
2483 | if (mdev->agreed_pro_version < 91) | |
4a23f264 | 2484 | return -1091; |
b411b363 PR |
2485 | |
2486 | _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]); | |
2487 | _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]); | |
2488 | ||
4a23f264 | 2489 | dev_info(DEV, "Last syncUUID did not get through, corrected:\n"); |
b411b363 PR |
2490 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, |
2491 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2492 | ||
2493 | return 1; | |
2494 | } | |
2495 | } | |
2496 | ||
2497 | ||
2498 | *rule_nr = 80; | |
d8c2a36b | 2499 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); |
b411b363 PR |
2500 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
2501 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2502 | if (self == peer) | |
2503 | return 2; | |
2504 | } | |
2505 | ||
2506 | *rule_nr = 90; | |
2507 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2508 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2509 | if (self == peer && self != ((u64)0)) | |
2510 | return 100; | |
2511 | ||
2512 | *rule_nr = 100; | |
2513 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2514 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2515 | for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { | |
2516 | peer = mdev->p_uuid[j] & ~((u64)1); | |
2517 | if (self == peer) | |
2518 | return -100; | |
2519 | } | |
2520 | } | |
2521 | ||
2522 | return -1000; | |
2523 | } | |
2524 | ||
2525 | /* drbd_sync_handshake() returns the new conn state on success, or | |
2526 | CONN_MASK (-1) on failure. | |
2527 | */ | |
2528 | static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role, | |
2529 | enum drbd_disk_state peer_disk) __must_hold(local) | |
2530 | { | |
2531 | int hg, rule_nr; | |
2532 | enum drbd_conns rv = C_MASK; | |
2533 | enum drbd_disk_state mydisk; | |
2534 | ||
2535 | mydisk = mdev->state.disk; | |
2536 | if (mydisk == D_NEGOTIATING) | |
2537 | mydisk = mdev->new_state_tmp.disk; | |
2538 | ||
2539 | dev_info(DEV, "drbd_sync_handshake:\n"); | |
2540 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0); | |
2541 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, | |
2542 | mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2543 | ||
2544 | hg = drbd_uuid_compare(mdev, &rule_nr); | |
2545 | ||
2546 | dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr); | |
2547 | ||
2548 | if (hg == -1000) { | |
2549 | dev_alert(DEV, "Unrelated data, aborting!\n"); | |
2550 | return C_MASK; | |
2551 | } | |
4a23f264 PR |
2552 | if (hg < -1000) { |
2553 | dev_alert(DEV, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000); | |
b411b363 PR |
2554 | return C_MASK; |
2555 | } | |
2556 | ||
2557 | if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || | |
2558 | (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { | |
2559 | int f = (hg == -100) || abs(hg) == 2; | |
2560 | hg = mydisk > D_INCONSISTENT ? 1 : -1; | |
2561 | if (f) | |
2562 | hg = hg*2; | |
2563 | dev_info(DEV, "Becoming sync %s due to disk states.\n", | |
2564 | hg > 0 ? "source" : "target"); | |
2565 | } | |
2566 | ||
3a11a487 AG |
2567 | if (abs(hg) == 100) |
2568 | drbd_khelper(mdev, "initial-split-brain"); | |
2569 | ||
b411b363 PR |
2570 | if (hg == 100 || (hg == -100 && mdev->net_conf->always_asbp)) { |
2571 | int pcount = (mdev->state.role == R_PRIMARY) | |
2572 | + (peer_role == R_PRIMARY); | |
2573 | int forced = (hg == -100); | |
2574 | ||
2575 | switch (pcount) { | |
2576 | case 0: | |
2577 | hg = drbd_asb_recover_0p(mdev); | |
2578 | break; | |
2579 | case 1: | |
2580 | hg = drbd_asb_recover_1p(mdev); | |
2581 | break; | |
2582 | case 2: | |
2583 | hg = drbd_asb_recover_2p(mdev); | |
2584 | break; | |
2585 | } | |
2586 | if (abs(hg) < 100) { | |
2587 | dev_warn(DEV, "Split-Brain detected, %d primaries, " | |
2588 | "automatically solved. Sync from %s node\n", | |
2589 | pcount, (hg < 0) ? "peer" : "this"); | |
2590 | if (forced) { | |
2591 | dev_warn(DEV, "Doing a full sync, since" | |
2592 | " UUIDs where ambiguous.\n"); | |
2593 | hg = hg*2; | |
2594 | } | |
2595 | } | |
2596 | } | |
2597 | ||
2598 | if (hg == -100) { | |
2599 | if (mdev->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1)) | |
2600 | hg = -1; | |
2601 | if (!mdev->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1)) | |
2602 | hg = 1; | |
2603 | ||
2604 | if (abs(hg) < 100) | |
2605 | dev_warn(DEV, "Split-Brain detected, manually solved. " | |
2606 | "Sync from %s node\n", | |
2607 | (hg < 0) ? "peer" : "this"); | |
2608 | } | |
2609 | ||
2610 | if (hg == -100) { | |
580b9767 LE |
2611 | /* FIXME this log message is not correct if we end up here |
2612 | * after an attempted attach on a diskless node. | |
2613 | * We just refuse to attach -- well, we drop the "connection" | |
2614 | * to that disk, in a way... */ | |
3a11a487 | 2615 | dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n"); |
b411b363 PR |
2616 | drbd_khelper(mdev, "split-brain"); |
2617 | return C_MASK; | |
2618 | } | |
2619 | ||
2620 | if (hg > 0 && mydisk <= D_INCONSISTENT) { | |
2621 | dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n"); | |
2622 | return C_MASK; | |
2623 | } | |
2624 | ||
2625 | if (hg < 0 && /* by intention we do not use mydisk here. */ | |
2626 | mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) { | |
2627 | switch (mdev->net_conf->rr_conflict) { | |
2628 | case ASB_CALL_HELPER: | |
2629 | drbd_khelper(mdev, "pri-lost"); | |
2630 | /* fall through */ | |
2631 | case ASB_DISCONNECT: | |
2632 | dev_err(DEV, "I shall become SyncTarget, but I am primary!\n"); | |
2633 | return C_MASK; | |
2634 | case ASB_VIOLENTLY: | |
2635 | dev_warn(DEV, "Becoming SyncTarget, violating the stable-data" | |
2636 | "assumption\n"); | |
2637 | } | |
2638 | } | |
2639 | ||
cf14c2e9 PR |
2640 | if (mdev->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->flags)) { |
2641 | if (hg == 0) | |
2642 | dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n"); | |
2643 | else | |
2644 | dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.", | |
2645 | drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), | |
2646 | abs(hg) >= 2 ? "full" : "bit-map based"); | |
2647 | return C_MASK; | |
2648 | } | |
2649 | ||
b411b363 PR |
2650 | if (abs(hg) >= 2) { |
2651 | dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); | |
20ceb2b2 LE |
2652 | if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake", |
2653 | BM_LOCKED_SET_ALLOWED)) | |
b411b363 PR |
2654 | return C_MASK; |
2655 | } | |
2656 | ||
2657 | if (hg > 0) { /* become sync source. */ | |
2658 | rv = C_WF_BITMAP_S; | |
2659 | } else if (hg < 0) { /* become sync target */ | |
2660 | rv = C_WF_BITMAP_T; | |
2661 | } else { | |
2662 | rv = C_CONNECTED; | |
2663 | if (drbd_bm_total_weight(mdev)) { | |
2664 | dev_info(DEV, "No resync, but %lu bits in bitmap!\n", | |
2665 | drbd_bm_total_weight(mdev)); | |
2666 | } | |
2667 | } | |
2668 | ||
2669 | return rv; | |
2670 | } | |
2671 | ||
2672 | /* returns 1 if invalid */ | |
2673 | static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self) | |
2674 | { | |
2675 | /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ | |
2676 | if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) || | |
2677 | (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL)) | |
2678 | return 0; | |
2679 | ||
2680 | /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ | |
2681 | if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL || | |
2682 | self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL) | |
2683 | return 1; | |
2684 | ||
2685 | /* everything else is valid if they are equal on both sides. */ | |
2686 | if (peer == self) | |
2687 | return 0; | |
2688 | ||
2689 | /* everything es is invalid. */ | |
2690 | return 1; | |
2691 | } | |
2692 | ||
02918be2 | 2693 | static int receive_protocol(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 2694 | { |
02918be2 | 2695 | struct p_protocol *p = &mdev->data.rbuf.protocol; |
b411b363 | 2696 | int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; |
cf14c2e9 | 2697 | int p_want_lose, p_two_primaries, cf; |
b411b363 PR |
2698 | char p_integrity_alg[SHARED_SECRET_MAX] = ""; |
2699 | ||
b411b363 PR |
2700 | p_proto = be32_to_cpu(p->protocol); |
2701 | p_after_sb_0p = be32_to_cpu(p->after_sb_0p); | |
2702 | p_after_sb_1p = be32_to_cpu(p->after_sb_1p); | |
2703 | p_after_sb_2p = be32_to_cpu(p->after_sb_2p); | |
b411b363 | 2704 | p_two_primaries = be32_to_cpu(p->two_primaries); |
cf14c2e9 PR |
2705 | cf = be32_to_cpu(p->conn_flags); |
2706 | p_want_lose = cf & CF_WANT_LOSE; | |
2707 | ||
2708 | clear_bit(CONN_DRY_RUN, &mdev->flags); | |
2709 | ||
2710 | if (cf & CF_DRY_RUN) | |
2711 | set_bit(CONN_DRY_RUN, &mdev->flags); | |
b411b363 PR |
2712 | |
2713 | if (p_proto != mdev->net_conf->wire_protocol) { | |
2714 | dev_err(DEV, "incompatible communication protocols\n"); | |
2715 | goto disconnect; | |
2716 | } | |
2717 | ||
2718 | if (cmp_after_sb(p_after_sb_0p, mdev->net_conf->after_sb_0p)) { | |
2719 | dev_err(DEV, "incompatible after-sb-0pri settings\n"); | |
2720 | goto disconnect; | |
2721 | } | |
2722 | ||
2723 | if (cmp_after_sb(p_after_sb_1p, mdev->net_conf->after_sb_1p)) { | |
2724 | dev_err(DEV, "incompatible after-sb-1pri settings\n"); | |
2725 | goto disconnect; | |
2726 | } | |
2727 | ||
2728 | if (cmp_after_sb(p_after_sb_2p, mdev->net_conf->after_sb_2p)) { | |
2729 | dev_err(DEV, "incompatible after-sb-2pri settings\n"); | |
2730 | goto disconnect; | |
2731 | } | |
2732 | ||
2733 | if (p_want_lose && mdev->net_conf->want_lose) { | |
2734 | dev_err(DEV, "both sides have the 'want_lose' flag set\n"); | |
2735 | goto disconnect; | |
2736 | } | |
2737 | ||
2738 | if (p_two_primaries != mdev->net_conf->two_primaries) { | |
2739 | dev_err(DEV, "incompatible setting of the two-primaries options\n"); | |
2740 | goto disconnect; | |
2741 | } | |
2742 | ||
2743 | if (mdev->agreed_pro_version >= 87) { | |
2744 | unsigned char *my_alg = mdev->net_conf->integrity_alg; | |
2745 | ||
2746 | if (drbd_recv(mdev, p_integrity_alg, data_size) != data_size) | |
81e84650 | 2747 | return false; |
b411b363 PR |
2748 | |
2749 | p_integrity_alg[SHARED_SECRET_MAX-1] = 0; | |
2750 | if (strcmp(p_integrity_alg, my_alg)) { | |
2751 | dev_err(DEV, "incompatible setting of the data-integrity-alg\n"); | |
2752 | goto disconnect; | |
2753 | } | |
2754 | dev_info(DEV, "data-integrity-alg: %s\n", | |
2755 | my_alg[0] ? my_alg : (unsigned char *)"<not-used>"); | |
2756 | } | |
2757 | ||
81e84650 | 2758 | return true; |
b411b363 PR |
2759 | |
2760 | disconnect: | |
2761 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 2762 | return false; |
b411b363 PR |
2763 | } |
2764 | ||
2765 | /* helper function | |
2766 | * input: alg name, feature name | |
2767 | * return: NULL (alg name was "") | |
2768 | * ERR_PTR(error) if something goes wrong | |
2769 | * or the crypto hash ptr, if it worked out ok. */ | |
2770 | struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev, | |
2771 | const char *alg, const char *name) | |
2772 | { | |
2773 | struct crypto_hash *tfm; | |
2774 | ||
2775 | if (!alg[0]) | |
2776 | return NULL; | |
2777 | ||
2778 | tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); | |
2779 | if (IS_ERR(tfm)) { | |
2780 | dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n", | |
2781 | alg, name, PTR_ERR(tfm)); | |
2782 | return tfm; | |
2783 | } | |
2784 | if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) { | |
2785 | crypto_free_hash(tfm); | |
2786 | dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name); | |
2787 | return ERR_PTR(-EINVAL); | |
2788 | } | |
2789 | return tfm; | |
2790 | } | |
2791 | ||
02918be2 | 2792 | static int receive_SyncParam(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int packet_size) |
b411b363 | 2793 | { |
81e84650 | 2794 | int ok = true; |
02918be2 | 2795 | struct p_rs_param_95 *p = &mdev->data.rbuf.rs_param_95; |
b411b363 PR |
2796 | unsigned int header_size, data_size, exp_max_sz; |
2797 | struct crypto_hash *verify_tfm = NULL; | |
2798 | struct crypto_hash *csums_tfm = NULL; | |
2799 | const int apv = mdev->agreed_pro_version; | |
778f271d PR |
2800 | int *rs_plan_s = NULL; |
2801 | int fifo_size = 0; | |
b411b363 PR |
2802 | |
2803 | exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) | |
2804 | : apv == 88 ? sizeof(struct p_rs_param) | |
2805 | + SHARED_SECRET_MAX | |
8e26f9cc PR |
2806 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
2807 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 2808 | |
02918be2 | 2809 | if (packet_size > exp_max_sz) { |
b411b363 | 2810 | dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n", |
02918be2 | 2811 | packet_size, exp_max_sz); |
81e84650 | 2812 | return false; |
b411b363 PR |
2813 | } |
2814 | ||
2815 | if (apv <= 88) { | |
02918be2 PR |
2816 | header_size = sizeof(struct p_rs_param) - sizeof(struct p_header80); |
2817 | data_size = packet_size - header_size; | |
8e26f9cc | 2818 | } else if (apv <= 94) { |
02918be2 PR |
2819 | header_size = sizeof(struct p_rs_param_89) - sizeof(struct p_header80); |
2820 | data_size = packet_size - header_size; | |
b411b363 | 2821 | D_ASSERT(data_size == 0); |
8e26f9cc | 2822 | } else { |
02918be2 PR |
2823 | header_size = sizeof(struct p_rs_param_95) - sizeof(struct p_header80); |
2824 | data_size = packet_size - header_size; | |
b411b363 PR |
2825 | D_ASSERT(data_size == 0); |
2826 | } | |
2827 | ||
2828 | /* initialize verify_alg and csums_alg */ | |
2829 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); | |
2830 | ||
02918be2 | 2831 | if (drbd_recv(mdev, &p->head.payload, header_size) != header_size) |
81e84650 | 2832 | return false; |
b411b363 PR |
2833 | |
2834 | mdev->sync_conf.rate = be32_to_cpu(p->rate); | |
2835 | ||
2836 | if (apv >= 88) { | |
2837 | if (apv == 88) { | |
2838 | if (data_size > SHARED_SECRET_MAX) { | |
2839 | dev_err(DEV, "verify-alg too long, " | |
2840 | "peer wants %u, accepting only %u byte\n", | |
2841 | data_size, SHARED_SECRET_MAX); | |
81e84650 | 2842 | return false; |
b411b363 PR |
2843 | } |
2844 | ||
2845 | if (drbd_recv(mdev, p->verify_alg, data_size) != data_size) | |
81e84650 | 2846 | return false; |
b411b363 PR |
2847 | |
2848 | /* we expect NUL terminated string */ | |
2849 | /* but just in case someone tries to be evil */ | |
2850 | D_ASSERT(p->verify_alg[data_size-1] == 0); | |
2851 | p->verify_alg[data_size-1] = 0; | |
2852 | ||
2853 | } else /* apv >= 89 */ { | |
2854 | /* we still expect NUL terminated strings */ | |
2855 | /* but just in case someone tries to be evil */ | |
2856 | D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0); | |
2857 | D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0); | |
2858 | p->verify_alg[SHARED_SECRET_MAX-1] = 0; | |
2859 | p->csums_alg[SHARED_SECRET_MAX-1] = 0; | |
2860 | } | |
2861 | ||
2862 | if (strcmp(mdev->sync_conf.verify_alg, p->verify_alg)) { | |
2863 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { | |
2864 | dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", | |
2865 | mdev->sync_conf.verify_alg, p->verify_alg); | |
2866 | goto disconnect; | |
2867 | } | |
2868 | verify_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
2869 | p->verify_alg, "verify-alg"); | |
2870 | if (IS_ERR(verify_tfm)) { | |
2871 | verify_tfm = NULL; | |
2872 | goto disconnect; | |
2873 | } | |
2874 | } | |
2875 | ||
2876 | if (apv >= 89 && strcmp(mdev->sync_conf.csums_alg, p->csums_alg)) { | |
2877 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { | |
2878 | dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", | |
2879 | mdev->sync_conf.csums_alg, p->csums_alg); | |
2880 | goto disconnect; | |
2881 | } | |
2882 | csums_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
2883 | p->csums_alg, "csums-alg"); | |
2884 | if (IS_ERR(csums_tfm)) { | |
2885 | csums_tfm = NULL; | |
2886 | goto disconnect; | |
2887 | } | |
2888 | } | |
2889 | ||
8e26f9cc PR |
2890 | if (apv > 94) { |
2891 | mdev->sync_conf.rate = be32_to_cpu(p->rate); | |
2892 | mdev->sync_conf.c_plan_ahead = be32_to_cpu(p->c_plan_ahead); | |
2893 | mdev->sync_conf.c_delay_target = be32_to_cpu(p->c_delay_target); | |
2894 | mdev->sync_conf.c_fill_target = be32_to_cpu(p->c_fill_target); | |
2895 | mdev->sync_conf.c_max_rate = be32_to_cpu(p->c_max_rate); | |
778f271d PR |
2896 | |
2897 | fifo_size = (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ; | |
2898 | if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) { | |
2899 | rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL); | |
2900 | if (!rs_plan_s) { | |
2901 | dev_err(DEV, "kmalloc of fifo_buffer failed"); | |
2902 | goto disconnect; | |
2903 | } | |
2904 | } | |
8e26f9cc | 2905 | } |
b411b363 PR |
2906 | |
2907 | spin_lock(&mdev->peer_seq_lock); | |
2908 | /* lock against drbd_nl_syncer_conf() */ | |
2909 | if (verify_tfm) { | |
2910 | strcpy(mdev->sync_conf.verify_alg, p->verify_alg); | |
2911 | mdev->sync_conf.verify_alg_len = strlen(p->verify_alg) + 1; | |
2912 | crypto_free_hash(mdev->verify_tfm); | |
2913 | mdev->verify_tfm = verify_tfm; | |
2914 | dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg); | |
2915 | } | |
2916 | if (csums_tfm) { | |
2917 | strcpy(mdev->sync_conf.csums_alg, p->csums_alg); | |
2918 | mdev->sync_conf.csums_alg_len = strlen(p->csums_alg) + 1; | |
2919 | crypto_free_hash(mdev->csums_tfm); | |
2920 | mdev->csums_tfm = csums_tfm; | |
2921 | dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg); | |
2922 | } | |
778f271d PR |
2923 | if (fifo_size != mdev->rs_plan_s.size) { |
2924 | kfree(mdev->rs_plan_s.values); | |
2925 | mdev->rs_plan_s.values = rs_plan_s; | |
2926 | mdev->rs_plan_s.size = fifo_size; | |
2927 | mdev->rs_planed = 0; | |
2928 | } | |
b411b363 PR |
2929 | spin_unlock(&mdev->peer_seq_lock); |
2930 | } | |
2931 | ||
2932 | return ok; | |
2933 | disconnect: | |
2934 | /* just for completeness: actually not needed, | |
2935 | * as this is not reached if csums_tfm was ok. */ | |
2936 | crypto_free_hash(csums_tfm); | |
2937 | /* but free the verify_tfm again, if csums_tfm did not work out */ | |
2938 | crypto_free_hash(verify_tfm); | |
2939 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 2940 | return false; |
b411b363 PR |
2941 | } |
2942 | ||
b411b363 PR |
2943 | /* warn if the arguments differ by more than 12.5% */ |
2944 | static void warn_if_differ_considerably(struct drbd_conf *mdev, | |
2945 | const char *s, sector_t a, sector_t b) | |
2946 | { | |
2947 | sector_t d; | |
2948 | if (a == 0 || b == 0) | |
2949 | return; | |
2950 | d = (a > b) ? (a - b) : (b - a); | |
2951 | if (d > (a>>3) || d > (b>>3)) | |
2952 | dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s, | |
2953 | (unsigned long long)a, (unsigned long long)b); | |
2954 | } | |
2955 | ||
02918be2 | 2956 | static int receive_sizes(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 2957 | { |
02918be2 | 2958 | struct p_sizes *p = &mdev->data.rbuf.sizes; |
b411b363 | 2959 | enum determine_dev_size dd = unchanged; |
b411b363 PR |
2960 | sector_t p_size, p_usize, my_usize; |
2961 | int ldsc = 0; /* local disk size changed */ | |
e89b591c | 2962 | enum dds_flags ddsf; |
b411b363 | 2963 | |
b411b363 PR |
2964 | p_size = be64_to_cpu(p->d_size); |
2965 | p_usize = be64_to_cpu(p->u_size); | |
2966 | ||
2967 | if (p_size == 0 && mdev->state.disk == D_DISKLESS) { | |
2968 | dev_err(DEV, "some backing storage is needed\n"); | |
2969 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 2970 | return false; |
b411b363 PR |
2971 | } |
2972 | ||
2973 | /* just store the peer's disk size for now. | |
2974 | * we still need to figure out whether we accept that. */ | |
2975 | mdev->p_size = p_size; | |
2976 | ||
b411b363 PR |
2977 | if (get_ldev(mdev)) { |
2978 | warn_if_differ_considerably(mdev, "lower level device sizes", | |
2979 | p_size, drbd_get_max_capacity(mdev->ldev)); | |
2980 | warn_if_differ_considerably(mdev, "user requested size", | |
2981 | p_usize, mdev->ldev->dc.disk_size); | |
2982 | ||
2983 | /* if this is the first connect, or an otherwise expected | |
2984 | * param exchange, choose the minimum */ | |
2985 | if (mdev->state.conn == C_WF_REPORT_PARAMS) | |
2986 | p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size, | |
2987 | p_usize); | |
2988 | ||
2989 | my_usize = mdev->ldev->dc.disk_size; | |
2990 | ||
2991 | if (mdev->ldev->dc.disk_size != p_usize) { | |
2992 | mdev->ldev->dc.disk_size = p_usize; | |
2993 | dev_info(DEV, "Peer sets u_size to %lu sectors\n", | |
2994 | (unsigned long)mdev->ldev->dc.disk_size); | |
2995 | } | |
2996 | ||
2997 | /* Never shrink a device with usable data during connect. | |
2998 | But allow online shrinking if we are connected. */ | |
a393db6f | 2999 | if (drbd_new_dev_size(mdev, mdev->ldev, 0) < |
b411b363 PR |
3000 | drbd_get_capacity(mdev->this_bdev) && |
3001 | mdev->state.disk >= D_OUTDATED && | |
3002 | mdev->state.conn < C_CONNECTED) { | |
3003 | dev_err(DEV, "The peer's disk size is too small!\n"); | |
3004 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
3005 | mdev->ldev->dc.disk_size = my_usize; | |
3006 | put_ldev(mdev); | |
81e84650 | 3007 | return false; |
b411b363 PR |
3008 | } |
3009 | put_ldev(mdev); | |
3010 | } | |
b411b363 | 3011 | |
e89b591c | 3012 | ddsf = be16_to_cpu(p->dds_flags); |
b411b363 | 3013 | if (get_ldev(mdev)) { |
24c4830c | 3014 | dd = drbd_determine_dev_size(mdev, ddsf); |
b411b363 PR |
3015 | put_ldev(mdev); |
3016 | if (dd == dev_size_error) | |
81e84650 | 3017 | return false; |
b411b363 PR |
3018 | drbd_md_sync(mdev); |
3019 | } else { | |
3020 | /* I am diskless, need to accept the peer's size. */ | |
3021 | drbd_set_my_capacity(mdev, p_size); | |
3022 | } | |
3023 | ||
99432fcc PR |
3024 | mdev->peer_max_bio_size = be32_to_cpu(p->max_bio_size); |
3025 | drbd_reconsider_max_bio_size(mdev); | |
3026 | ||
b411b363 PR |
3027 | if (get_ldev(mdev)) { |
3028 | if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) { | |
3029 | mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); | |
3030 | ldsc = 1; | |
3031 | } | |
3032 | ||
b411b363 PR |
3033 | put_ldev(mdev); |
3034 | } | |
3035 | ||
3036 | if (mdev->state.conn > C_WF_REPORT_PARAMS) { | |
3037 | if (be64_to_cpu(p->c_size) != | |
3038 | drbd_get_capacity(mdev->this_bdev) || ldsc) { | |
3039 | /* we have different sizes, probably peer | |
3040 | * needs to know my new size... */ | |
e89b591c | 3041 | drbd_send_sizes(mdev, 0, ddsf); |
b411b363 PR |
3042 | } |
3043 | if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) || | |
3044 | (dd == grew && mdev->state.conn == C_CONNECTED)) { | |
3045 | if (mdev->state.pdsk >= D_INCONSISTENT && | |
e89b591c PR |
3046 | mdev->state.disk >= D_INCONSISTENT) { |
3047 | if (ddsf & DDSF_NO_RESYNC) | |
3048 | dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n"); | |
3049 | else | |
3050 | resync_after_online_grow(mdev); | |
3051 | } else | |
b411b363 PR |
3052 | set_bit(RESYNC_AFTER_NEG, &mdev->flags); |
3053 | } | |
3054 | } | |
3055 | ||
81e84650 | 3056 | return true; |
b411b363 PR |
3057 | } |
3058 | ||
02918be2 | 3059 | static int receive_uuids(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 3060 | { |
02918be2 | 3061 | struct p_uuids *p = &mdev->data.rbuf.uuids; |
b411b363 | 3062 | u64 *p_uuid; |
62b0da3a | 3063 | int i, updated_uuids = 0; |
b411b363 | 3064 | |
b411b363 PR |
3065 | p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); |
3066 | ||
3067 | for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) | |
3068 | p_uuid[i] = be64_to_cpu(p->uuid[i]); | |
3069 | ||
3070 | kfree(mdev->p_uuid); | |
3071 | mdev->p_uuid = p_uuid; | |
3072 | ||
3073 | if (mdev->state.conn < C_CONNECTED && | |
3074 | mdev->state.disk < D_INCONSISTENT && | |
3075 | mdev->state.role == R_PRIMARY && | |
3076 | (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { | |
3077 | dev_err(DEV, "Can only connect to data with current UUID=%016llX\n", | |
3078 | (unsigned long long)mdev->ed_uuid); | |
3079 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 3080 | return false; |
b411b363 PR |
3081 | } |
3082 | ||
3083 | if (get_ldev(mdev)) { | |
3084 | int skip_initial_sync = | |
3085 | mdev->state.conn == C_CONNECTED && | |
3086 | mdev->agreed_pro_version >= 90 && | |
3087 | mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && | |
3088 | (p_uuid[UI_FLAGS] & 8); | |
3089 | if (skip_initial_sync) { | |
3090 | dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n"); | |
3091 | drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, | |
20ceb2b2 LE |
3092 | "clear_n_write from receive_uuids", |
3093 | BM_LOCKED_TEST_ALLOWED); | |
b411b363 PR |
3094 | _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]); |
3095 | _drbd_uuid_set(mdev, UI_BITMAP, 0); | |
3096 | _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), | |
3097 | CS_VERBOSE, NULL); | |
3098 | drbd_md_sync(mdev); | |
62b0da3a | 3099 | updated_uuids = 1; |
b411b363 PR |
3100 | } |
3101 | put_ldev(mdev); | |
18a50fa2 PR |
3102 | } else if (mdev->state.disk < D_INCONSISTENT && |
3103 | mdev->state.role == R_PRIMARY) { | |
3104 | /* I am a diskless primary, the peer just created a new current UUID | |
3105 | for me. */ | |
62b0da3a | 3106 | updated_uuids = drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
b411b363 PR |
3107 | } |
3108 | ||
3109 | /* Before we test for the disk state, we should wait until an eventually | |
3110 | ongoing cluster wide state change is finished. That is important if | |
3111 | we are primary and are detaching from our disk. We need to see the | |
3112 | new disk state... */ | |
3113 | wait_event(mdev->misc_wait, !test_bit(CLUSTER_ST_CHANGE, &mdev->flags)); | |
3114 | if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT) | |
62b0da3a LE |
3115 | updated_uuids |= drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
3116 | ||
3117 | if (updated_uuids) | |
3118 | drbd_print_uuids(mdev, "receiver updated UUIDs to"); | |
b411b363 | 3119 | |
81e84650 | 3120 | return true; |
b411b363 PR |
3121 | } |
3122 | ||
3123 | /** | |
3124 | * convert_state() - Converts the peer's view of the cluster state to our point of view | |
3125 | * @ps: The state as seen by the peer. | |
3126 | */ | |
3127 | static union drbd_state convert_state(union drbd_state ps) | |
3128 | { | |
3129 | union drbd_state ms; | |
3130 | ||
3131 | static enum drbd_conns c_tab[] = { | |
3132 | [C_CONNECTED] = C_CONNECTED, | |
3133 | ||
3134 | [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, | |
3135 | [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, | |
3136 | [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ | |
3137 | [C_VERIFY_S] = C_VERIFY_T, | |
3138 | [C_MASK] = C_MASK, | |
3139 | }; | |
3140 | ||
3141 | ms.i = ps.i; | |
3142 | ||
3143 | ms.conn = c_tab[ps.conn]; | |
3144 | ms.peer = ps.role; | |
3145 | ms.role = ps.peer; | |
3146 | ms.pdsk = ps.disk; | |
3147 | ms.disk = ps.pdsk; | |
3148 | ms.peer_isp = (ps.aftr_isp | ps.user_isp); | |
3149 | ||
3150 | return ms; | |
3151 | } | |
3152 | ||
02918be2 | 3153 | static int receive_req_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 3154 | { |
02918be2 | 3155 | struct p_req_state *p = &mdev->data.rbuf.req_state; |
b411b363 | 3156 | union drbd_state mask, val; |
bf885f8a | 3157 | enum drbd_state_rv rv; |
b411b363 | 3158 | |
b411b363 PR |
3159 | mask.i = be32_to_cpu(p->mask); |
3160 | val.i = be32_to_cpu(p->val); | |
3161 | ||
3162 | if (test_bit(DISCARD_CONCURRENT, &mdev->flags) && | |
3163 | test_bit(CLUSTER_ST_CHANGE, &mdev->flags)) { | |
3164 | drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG); | |
81e84650 | 3165 | return true; |
b411b363 PR |
3166 | } |
3167 | ||
3168 | mask = convert_state(mask); | |
3169 | val = convert_state(val); | |
3170 | ||
3171 | rv = drbd_change_state(mdev, CS_VERBOSE, mask, val); | |
3172 | ||
3173 | drbd_send_sr_reply(mdev, rv); | |
3174 | drbd_md_sync(mdev); | |
3175 | ||
81e84650 | 3176 | return true; |
b411b363 PR |
3177 | } |
3178 | ||
02918be2 | 3179 | static int receive_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 3180 | { |
02918be2 | 3181 | struct p_state *p = &mdev->data.rbuf.state; |
4ac4aada | 3182 | union drbd_state os, ns, peer_state; |
b411b363 | 3183 | enum drbd_disk_state real_peer_disk; |
65d922c3 | 3184 | enum chg_state_flags cs_flags; |
b411b363 PR |
3185 | int rv; |
3186 | ||
b411b363 PR |
3187 | peer_state.i = be32_to_cpu(p->state); |
3188 | ||
3189 | real_peer_disk = peer_state.disk; | |
3190 | if (peer_state.disk == D_NEGOTIATING) { | |
3191 | real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; | |
3192 | dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); | |
3193 | } | |
3194 | ||
3195 | spin_lock_irq(&mdev->req_lock); | |
3196 | retry: | |
4ac4aada | 3197 | os = ns = mdev->state; |
b411b363 PR |
3198 | spin_unlock_irq(&mdev->req_lock); |
3199 | ||
e9ef7bb6 LE |
3200 | /* peer says his disk is uptodate, while we think it is inconsistent, |
3201 | * and this happens while we think we have a sync going on. */ | |
3202 | if (os.pdsk == D_INCONSISTENT && real_peer_disk == D_UP_TO_DATE && | |
3203 | os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { | |
3204 | /* If we are (becoming) SyncSource, but peer is still in sync | |
3205 | * preparation, ignore its uptodate-ness to avoid flapping, it | |
3206 | * will change to inconsistent once the peer reaches active | |
3207 | * syncing states. | |
3208 | * It may have changed syncer-paused flags, however, so we | |
3209 | * cannot ignore this completely. */ | |
3210 | if (peer_state.conn > C_CONNECTED && | |
3211 | peer_state.conn < C_SYNC_SOURCE) | |
3212 | real_peer_disk = D_INCONSISTENT; | |
3213 | ||
3214 | /* if peer_state changes to connected at the same time, | |
3215 | * it explicitly notifies us that it finished resync. | |
3216 | * Maybe we should finish it up, too? */ | |
3217 | else if (os.conn >= C_SYNC_SOURCE && | |
3218 | peer_state.conn == C_CONNECTED) { | |
3219 | if (drbd_bm_total_weight(mdev) <= mdev->rs_failed) | |
3220 | drbd_resync_finished(mdev); | |
81e84650 | 3221 | return true; |
e9ef7bb6 LE |
3222 | } |
3223 | } | |
3224 | ||
3225 | /* peer says his disk is inconsistent, while we think it is uptodate, | |
3226 | * and this happens while the peer still thinks we have a sync going on, | |
3227 | * but we think we are already done with the sync. | |
3228 | * We ignore this to avoid flapping pdsk. | |
3229 | * This should not happen, if the peer is a recent version of drbd. */ | |
3230 | if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && | |
3231 | os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) | |
3232 | real_peer_disk = D_UP_TO_DATE; | |
3233 | ||
4ac4aada LE |
3234 | if (ns.conn == C_WF_REPORT_PARAMS) |
3235 | ns.conn = C_CONNECTED; | |
b411b363 | 3236 | |
67531718 PR |
3237 | if (peer_state.conn == C_AHEAD) |
3238 | ns.conn = C_BEHIND; | |
3239 | ||
b411b363 PR |
3240 | if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING && |
3241 | get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3242 | int cr; /* consider resync */ | |
3243 | ||
3244 | /* if we established a new connection */ | |
4ac4aada | 3245 | cr = (os.conn < C_CONNECTED); |
b411b363 PR |
3246 | /* if we had an established connection |
3247 | * and one of the nodes newly attaches a disk */ | |
4ac4aada | 3248 | cr |= (os.conn == C_CONNECTED && |
b411b363 | 3249 | (peer_state.disk == D_NEGOTIATING || |
4ac4aada | 3250 | os.disk == D_NEGOTIATING)); |
b411b363 PR |
3251 | /* if we have both been inconsistent, and the peer has been |
3252 | * forced to be UpToDate with --overwrite-data */ | |
3253 | cr |= test_bit(CONSIDER_RESYNC, &mdev->flags); | |
3254 | /* if we had been plain connected, and the admin requested to | |
3255 | * start a sync by "invalidate" or "invalidate-remote" */ | |
4ac4aada | 3256 | cr |= (os.conn == C_CONNECTED && |
b411b363 PR |
3257 | (peer_state.conn >= C_STARTING_SYNC_S && |
3258 | peer_state.conn <= C_WF_BITMAP_T)); | |
3259 | ||
3260 | if (cr) | |
4ac4aada | 3261 | ns.conn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk); |
b411b363 PR |
3262 | |
3263 | put_ldev(mdev); | |
4ac4aada LE |
3264 | if (ns.conn == C_MASK) { |
3265 | ns.conn = C_CONNECTED; | |
b411b363 | 3266 | if (mdev->state.disk == D_NEGOTIATING) { |
82f59cc6 | 3267 | drbd_force_state(mdev, NS(disk, D_FAILED)); |
b411b363 PR |
3268 | } else if (peer_state.disk == D_NEGOTIATING) { |
3269 | dev_err(DEV, "Disk attach process on the peer node was aborted.\n"); | |
3270 | peer_state.disk = D_DISKLESS; | |
580b9767 | 3271 | real_peer_disk = D_DISKLESS; |
b411b363 | 3272 | } else { |
cf14c2e9 | 3273 | if (test_and_clear_bit(CONN_DRY_RUN, &mdev->flags)) |
81e84650 | 3274 | return false; |
4ac4aada | 3275 | D_ASSERT(os.conn == C_WF_REPORT_PARAMS); |
b411b363 | 3276 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); |
81e84650 | 3277 | return false; |
b411b363 PR |
3278 | } |
3279 | } | |
3280 | } | |
3281 | ||
3282 | spin_lock_irq(&mdev->req_lock); | |
4ac4aada | 3283 | if (mdev->state.i != os.i) |
b411b363 PR |
3284 | goto retry; |
3285 | clear_bit(CONSIDER_RESYNC, &mdev->flags); | |
b411b363 PR |
3286 | ns.peer = peer_state.role; |
3287 | ns.pdsk = real_peer_disk; | |
3288 | ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); | |
4ac4aada | 3289 | if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) |
b411b363 | 3290 | ns.disk = mdev->new_state_tmp.disk; |
4ac4aada LE |
3291 | cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); |
3292 | if (ns.pdsk == D_CONSISTENT && is_susp(ns) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && | |
481c6f50 PR |
3293 | test_bit(NEW_CUR_UUID, &mdev->flags)) { |
3294 | /* Do not allow tl_restart(resend) for a rebooted peer. We can only allow this | |
3295 | for temporal network outages! */ | |
3296 | spin_unlock_irq(&mdev->req_lock); | |
3297 | dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); | |
3298 | tl_clear(mdev); | |
3299 | drbd_uuid_new_current(mdev); | |
3300 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
3301 | drbd_force_state(mdev, NS2(conn, C_PROTOCOL_ERROR, susp, 0)); | |
81e84650 | 3302 | return false; |
481c6f50 | 3303 | } |
65d922c3 | 3304 | rv = _drbd_set_state(mdev, ns, cs_flags, NULL); |
b411b363 PR |
3305 | ns = mdev->state; |
3306 | spin_unlock_irq(&mdev->req_lock); | |
3307 | ||
3308 | if (rv < SS_SUCCESS) { | |
3309 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 3310 | return false; |
b411b363 PR |
3311 | } |
3312 | ||
4ac4aada LE |
3313 | if (os.conn > C_WF_REPORT_PARAMS) { |
3314 | if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && | |
b411b363 PR |
3315 | peer_state.disk != D_NEGOTIATING ) { |
3316 | /* we want resync, peer has not yet decided to sync... */ | |
3317 | /* Nowadays only used when forcing a node into primary role and | |
3318 | setting its disk to UpToDate with that */ | |
3319 | drbd_send_uuids(mdev); | |
3320 | drbd_send_state(mdev); | |
3321 | } | |
3322 | } | |
3323 | ||
3324 | mdev->net_conf->want_lose = 0; | |
3325 | ||
3326 | drbd_md_sync(mdev); /* update connected indicator, la_size, ... */ | |
3327 | ||
81e84650 | 3328 | return true; |
b411b363 PR |
3329 | } |
3330 | ||
02918be2 | 3331 | static int receive_sync_uuid(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 3332 | { |
02918be2 | 3333 | struct p_rs_uuid *p = &mdev->data.rbuf.rs_uuid; |
b411b363 PR |
3334 | |
3335 | wait_event(mdev->misc_wait, | |
3336 | mdev->state.conn == C_WF_SYNC_UUID || | |
c4752ef1 | 3337 | mdev->state.conn == C_BEHIND || |
b411b363 PR |
3338 | mdev->state.conn < C_CONNECTED || |
3339 | mdev->state.disk < D_NEGOTIATING); | |
3340 | ||
3341 | /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */ | |
3342 | ||
b411b363 PR |
3343 | /* Here the _drbd_uuid_ functions are right, current should |
3344 | _not_ be rotated into the history */ | |
3345 | if (get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3346 | _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid)); | |
3347 | _drbd_uuid_set(mdev, UI_BITMAP, 0UL); | |
3348 | ||
62b0da3a | 3349 | drbd_print_uuids(mdev, "updated sync uuid"); |
b411b363 PR |
3350 | drbd_start_resync(mdev, C_SYNC_TARGET); |
3351 | ||
3352 | put_ldev(mdev); | |
3353 | } else | |
3354 | dev_err(DEV, "Ignoring SyncUUID packet!\n"); | |
3355 | ||
81e84650 | 3356 | return true; |
b411b363 PR |
3357 | } |
3358 | ||
2c46407d AG |
3359 | /** |
3360 | * receive_bitmap_plain | |
3361 | * | |
3362 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3363 | * code upon failure. | |
3364 | */ | |
3365 | static int | |
02918be2 PR |
3366 | receive_bitmap_plain(struct drbd_conf *mdev, unsigned int data_size, |
3367 | unsigned long *buffer, struct bm_xfer_ctx *c) | |
b411b363 PR |
3368 | { |
3369 | unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset); | |
3370 | unsigned want = num_words * sizeof(long); | |
2c46407d | 3371 | int err; |
b411b363 | 3372 | |
02918be2 PR |
3373 | if (want != data_size) { |
3374 | dev_err(DEV, "%s:want (%u) != data_size (%u)\n", __func__, want, data_size); | |
2c46407d | 3375 | return -EIO; |
b411b363 PR |
3376 | } |
3377 | if (want == 0) | |
2c46407d AG |
3378 | return 0; |
3379 | err = drbd_recv(mdev, buffer, want); | |
3380 | if (err != want) { | |
3381 | if (err >= 0) | |
3382 | err = -EIO; | |
3383 | return err; | |
3384 | } | |
b411b363 PR |
3385 | |
3386 | drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer); | |
3387 | ||
3388 | c->word_offset += num_words; | |
3389 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
3390 | if (c->bit_offset > c->bm_bits) | |
3391 | c->bit_offset = c->bm_bits; | |
3392 | ||
2c46407d | 3393 | return 1; |
b411b363 PR |
3394 | } |
3395 | ||
2c46407d AG |
3396 | /** |
3397 | * recv_bm_rle_bits | |
3398 | * | |
3399 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3400 | * code upon failure. | |
3401 | */ | |
3402 | static int | |
b411b363 PR |
3403 | recv_bm_rle_bits(struct drbd_conf *mdev, |
3404 | struct p_compressed_bm *p, | |
3405 | struct bm_xfer_ctx *c) | |
3406 | { | |
3407 | struct bitstream bs; | |
3408 | u64 look_ahead; | |
3409 | u64 rl; | |
3410 | u64 tmp; | |
3411 | unsigned long s = c->bit_offset; | |
3412 | unsigned long e; | |
004352fa | 3413 | int len = be16_to_cpu(p->head.length) - (sizeof(*p) - sizeof(p->head)); |
b411b363 PR |
3414 | int toggle = DCBP_get_start(p); |
3415 | int have; | |
3416 | int bits; | |
3417 | ||
3418 | bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p)); | |
3419 | ||
3420 | bits = bitstream_get_bits(&bs, &look_ahead, 64); | |
3421 | if (bits < 0) | |
2c46407d | 3422 | return -EIO; |
b411b363 PR |
3423 | |
3424 | for (have = bits; have > 0; s += rl, toggle = !toggle) { | |
3425 | bits = vli_decode_bits(&rl, look_ahead); | |
3426 | if (bits <= 0) | |
2c46407d | 3427 | return -EIO; |
b411b363 PR |
3428 | |
3429 | if (toggle) { | |
3430 | e = s + rl -1; | |
3431 | if (e >= c->bm_bits) { | |
3432 | dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); | |
2c46407d | 3433 | return -EIO; |
b411b363 PR |
3434 | } |
3435 | _drbd_bm_set_bits(mdev, s, e); | |
3436 | } | |
3437 | ||
3438 | if (have < bits) { | |
3439 | dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", | |
3440 | have, bits, look_ahead, | |
3441 | (unsigned int)(bs.cur.b - p->code), | |
3442 | (unsigned int)bs.buf_len); | |
2c46407d | 3443 | return -EIO; |
b411b363 PR |
3444 | } |
3445 | look_ahead >>= bits; | |
3446 | have -= bits; | |
3447 | ||
3448 | bits = bitstream_get_bits(&bs, &tmp, 64 - have); | |
3449 | if (bits < 0) | |
2c46407d | 3450 | return -EIO; |
b411b363 PR |
3451 | look_ahead |= tmp << have; |
3452 | have += bits; | |
3453 | } | |
3454 | ||
3455 | c->bit_offset = s; | |
3456 | bm_xfer_ctx_bit_to_word_offset(c); | |
3457 | ||
2c46407d | 3458 | return (s != c->bm_bits); |
b411b363 PR |
3459 | } |
3460 | ||
2c46407d AG |
3461 | /** |
3462 | * decode_bitmap_c | |
3463 | * | |
3464 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3465 | * code upon failure. | |
3466 | */ | |
3467 | static int | |
b411b363 PR |
3468 | decode_bitmap_c(struct drbd_conf *mdev, |
3469 | struct p_compressed_bm *p, | |
3470 | struct bm_xfer_ctx *c) | |
3471 | { | |
3472 | if (DCBP_get_code(p) == RLE_VLI_Bits) | |
3473 | return recv_bm_rle_bits(mdev, p, c); | |
3474 | ||
3475 | /* other variants had been implemented for evaluation, | |
3476 | * but have been dropped as this one turned out to be "best" | |
3477 | * during all our tests. */ | |
3478 | ||
3479 | dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding); | |
3480 | drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); | |
2c46407d | 3481 | return -EIO; |
b411b363 PR |
3482 | } |
3483 | ||
3484 | void INFO_bm_xfer_stats(struct drbd_conf *mdev, | |
3485 | const char *direction, struct bm_xfer_ctx *c) | |
3486 | { | |
3487 | /* what would it take to transfer it "plaintext" */ | |
0b70a13d | 3488 | unsigned plain = sizeof(struct p_header80) * |
b411b363 PR |
3489 | ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1) |
3490 | + c->bm_words * sizeof(long); | |
3491 | unsigned total = c->bytes[0] + c->bytes[1]; | |
3492 | unsigned r; | |
3493 | ||
3494 | /* total can not be zero. but just in case: */ | |
3495 | if (total == 0) | |
3496 | return; | |
3497 | ||
3498 | /* don't report if not compressed */ | |
3499 | if (total >= plain) | |
3500 | return; | |
3501 | ||
3502 | /* total < plain. check for overflow, still */ | |
3503 | r = (total > UINT_MAX/1000) ? (total / (plain/1000)) | |
3504 | : (1000 * total / plain); | |
3505 | ||
3506 | if (r > 1000) | |
3507 | r = 1000; | |
3508 | ||
3509 | r = 1000 - r; | |
3510 | dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " | |
3511 | "total %u; compression: %u.%u%%\n", | |
3512 | direction, | |
3513 | c->bytes[1], c->packets[1], | |
3514 | c->bytes[0], c->packets[0], | |
3515 | total, r/10, r % 10); | |
3516 | } | |
3517 | ||
3518 | /* Since we are processing the bitfield from lower addresses to higher, | |
3519 | it does not matter if the process it in 32 bit chunks or 64 bit | |
3520 | chunks as long as it is little endian. (Understand it as byte stream, | |
3521 | beginning with the lowest byte...) If we would use big endian | |
3522 | we would need to process it from the highest address to the lowest, | |
3523 | in order to be agnostic to the 32 vs 64 bits issue. | |
3524 | ||
3525 | returns 0 on failure, 1 if we successfully received it. */ | |
02918be2 | 3526 | static int receive_bitmap(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
3527 | { |
3528 | struct bm_xfer_ctx c; | |
3529 | void *buffer; | |
2c46407d | 3530 | int err; |
81e84650 | 3531 | int ok = false; |
02918be2 | 3532 | struct p_header80 *h = &mdev->data.rbuf.header.h80; |
b411b363 | 3533 | |
20ceb2b2 LE |
3534 | drbd_bm_lock(mdev, "receive bitmap", BM_LOCKED_SET_ALLOWED); |
3535 | /* you are supposed to send additional out-of-sync information | |
3536 | * if you actually set bits during this phase */ | |
b411b363 PR |
3537 | |
3538 | /* maybe we should use some per thread scratch page, | |
3539 | * and allocate that during initial device creation? */ | |
3540 | buffer = (unsigned long *) __get_free_page(GFP_NOIO); | |
3541 | if (!buffer) { | |
3542 | dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__); | |
3543 | goto out; | |
3544 | } | |
3545 | ||
3546 | c = (struct bm_xfer_ctx) { | |
3547 | .bm_bits = drbd_bm_bits(mdev), | |
3548 | .bm_words = drbd_bm_words(mdev), | |
3549 | }; | |
3550 | ||
2c46407d | 3551 | for(;;) { |
02918be2 | 3552 | if (cmd == P_BITMAP) { |
2c46407d | 3553 | err = receive_bitmap_plain(mdev, data_size, buffer, &c); |
02918be2 | 3554 | } else if (cmd == P_COMPRESSED_BITMAP) { |
b411b363 PR |
3555 | /* MAYBE: sanity check that we speak proto >= 90, |
3556 | * and the feature is enabled! */ | |
3557 | struct p_compressed_bm *p; | |
3558 | ||
02918be2 | 3559 | if (data_size > BM_PACKET_PAYLOAD_BYTES) { |
b411b363 PR |
3560 | dev_err(DEV, "ReportCBitmap packet too large\n"); |
3561 | goto out; | |
3562 | } | |
3563 | /* use the page buff */ | |
3564 | p = buffer; | |
3565 | memcpy(p, h, sizeof(*h)); | |
02918be2 | 3566 | if (drbd_recv(mdev, p->head.payload, data_size) != data_size) |
b411b363 | 3567 | goto out; |
004352fa LE |
3568 | if (data_size <= (sizeof(*p) - sizeof(p->head))) { |
3569 | dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", data_size); | |
78fcbdae | 3570 | goto out; |
b411b363 | 3571 | } |
2c46407d | 3572 | err = decode_bitmap_c(mdev, p, &c); |
b411b363 | 3573 | } else { |
02918be2 | 3574 | dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", cmd); |
b411b363 PR |
3575 | goto out; |
3576 | } | |
3577 | ||
02918be2 PR |
3578 | c.packets[cmd == P_BITMAP]++; |
3579 | c.bytes[cmd == P_BITMAP] += sizeof(struct p_header80) + data_size; | |
b411b363 | 3580 | |
2c46407d AG |
3581 | if (err <= 0) { |
3582 | if (err < 0) | |
3583 | goto out; | |
b411b363 | 3584 | break; |
2c46407d | 3585 | } |
02918be2 | 3586 | if (!drbd_recv_header(mdev, &cmd, &data_size)) |
b411b363 | 3587 | goto out; |
2c46407d | 3588 | } |
b411b363 PR |
3589 | |
3590 | INFO_bm_xfer_stats(mdev, "receive", &c); | |
3591 | ||
3592 | if (mdev->state.conn == C_WF_BITMAP_T) { | |
de1f8e4a AG |
3593 | enum drbd_state_rv rv; |
3594 | ||
b411b363 PR |
3595 | ok = !drbd_send_bitmap(mdev); |
3596 | if (!ok) | |
3597 | goto out; | |
3598 | /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ | |
de1f8e4a AG |
3599 | rv = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); |
3600 | D_ASSERT(rv == SS_SUCCESS); | |
b411b363 PR |
3601 | } else if (mdev->state.conn != C_WF_BITMAP_S) { |
3602 | /* admin may have requested C_DISCONNECTING, | |
3603 | * other threads may have noticed network errors */ | |
3604 | dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n", | |
3605 | drbd_conn_str(mdev->state.conn)); | |
3606 | } | |
3607 | ||
81e84650 | 3608 | ok = true; |
b411b363 | 3609 | out: |
20ceb2b2 | 3610 | drbd_bm_unlock(mdev); |
b411b363 PR |
3611 | if (ok && mdev->state.conn == C_WF_BITMAP_S) |
3612 | drbd_start_resync(mdev, C_SYNC_SOURCE); | |
3613 | free_page((unsigned long) buffer); | |
3614 | return ok; | |
3615 | } | |
3616 | ||
02918be2 | 3617 | static int receive_skip(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
3618 | { |
3619 | /* TODO zero copy sink :) */ | |
3620 | static char sink[128]; | |
3621 | int size, want, r; | |
3622 | ||
02918be2 PR |
3623 | dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n", |
3624 | cmd, data_size); | |
b411b363 | 3625 | |
02918be2 | 3626 | size = data_size; |
b411b363 PR |
3627 | while (size > 0) { |
3628 | want = min_t(int, size, sizeof(sink)); | |
3629 | r = drbd_recv(mdev, sink, want); | |
3630 | ERR_IF(r <= 0) break; | |
3631 | size -= r; | |
3632 | } | |
3633 | return size == 0; | |
3634 | } | |
3635 | ||
02918be2 | 3636 | static int receive_UnplugRemote(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
0ced55a3 | 3637 | { |
e7f52dfb LE |
3638 | /* Make sure we've acked all the TCP data associated |
3639 | * with the data requests being unplugged */ | |
3640 | drbd_tcp_quickack(mdev->data.socket); | |
0ced55a3 | 3641 | |
81e84650 | 3642 | return true; |
0ced55a3 PR |
3643 | } |
3644 | ||
73a01a18 PR |
3645 | static int receive_out_of_sync(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
3646 | { | |
3647 | struct p_block_desc *p = &mdev->data.rbuf.block_desc; | |
3648 | ||
f735e363 LE |
3649 | switch (mdev->state.conn) { |
3650 | case C_WF_SYNC_UUID: | |
3651 | case C_WF_BITMAP_T: | |
3652 | case C_BEHIND: | |
3653 | break; | |
3654 | default: | |
3655 | dev_err(DEV, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n", | |
3656 | drbd_conn_str(mdev->state.conn)); | |
3657 | } | |
3658 | ||
73a01a18 PR |
3659 | drbd_set_out_of_sync(mdev, be64_to_cpu(p->sector), be32_to_cpu(p->blksize)); |
3660 | ||
81e84650 | 3661 | return true; |
73a01a18 PR |
3662 | } |
3663 | ||
02918be2 PR |
3664 | typedef int (*drbd_cmd_handler_f)(struct drbd_conf *, enum drbd_packets cmd, unsigned int to_receive); |
3665 | ||
3666 | struct data_cmd { | |
3667 | int expect_payload; | |
3668 | size_t pkt_size; | |
3669 | drbd_cmd_handler_f function; | |
3670 | }; | |
3671 | ||
3672 | static struct data_cmd drbd_cmd_handler[] = { | |
3673 | [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, | |
3674 | [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, | |
3675 | [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , | |
3676 | [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , | |
3677 | [P_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } , | |
3678 | [P_COMPRESSED_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } , | |
3679 | [P_UNPLUG_REMOTE] = { 0, sizeof(struct p_header80), receive_UnplugRemote }, | |
3680 | [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
3681 | [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
3682 | [P_SYNC_PARAM] = { 1, sizeof(struct p_header80), receive_SyncParam }, | |
3683 | [P_SYNC_PARAM89] = { 1, sizeof(struct p_header80), receive_SyncParam }, | |
3684 | [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, | |
3685 | [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, | |
3686 | [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, | |
3687 | [P_STATE] = { 0, sizeof(struct p_state), receive_state }, | |
3688 | [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, | |
3689 | [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, | |
3690 | [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
3691 | [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
3692 | [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
3693 | [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, | |
73a01a18 | 3694 | [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync }, |
b411b363 PR |
3695 | /* anything missing from this table is in |
3696 | * the asender_tbl, see get_asender_cmd */ | |
02918be2 | 3697 | [P_MAX_CMD] = { 0, 0, NULL }, |
b411b363 PR |
3698 | }; |
3699 | ||
02918be2 PR |
3700 | /* All handler functions that expect a sub-header get that sub-heder in |
3701 | mdev->data.rbuf.header.head.payload. | |
3702 | ||
3703 | Usually in mdev->data.rbuf.header.head the callback can find the usual | |
3704 | p_header, but they may not rely on that. Since there is also p_header95 ! | |
3705 | */ | |
b411b363 PR |
3706 | |
3707 | static void drbdd(struct drbd_conf *mdev) | |
3708 | { | |
02918be2 PR |
3709 | union p_header *header = &mdev->data.rbuf.header; |
3710 | unsigned int packet_size; | |
3711 | enum drbd_packets cmd; | |
3712 | size_t shs; /* sub header size */ | |
3713 | int rv; | |
b411b363 PR |
3714 | |
3715 | while (get_t_state(&mdev->receiver) == Running) { | |
3716 | drbd_thread_current_set_cpu(mdev); | |
02918be2 PR |
3717 | if (!drbd_recv_header(mdev, &cmd, &packet_size)) |
3718 | goto err_out; | |
b411b363 | 3719 | |
02918be2 PR |
3720 | if (unlikely(cmd >= P_MAX_CMD || !drbd_cmd_handler[cmd].function)) { |
3721 | dev_err(DEV, "unknown packet type %d, l: %d!\n", cmd, packet_size); | |
3722 | goto err_out; | |
0b33a916 | 3723 | } |
b411b363 | 3724 | |
02918be2 | 3725 | shs = drbd_cmd_handler[cmd].pkt_size - sizeof(union p_header); |
02918be2 PR |
3726 | if (packet_size - shs > 0 && !drbd_cmd_handler[cmd].expect_payload) { |
3727 | dev_err(DEV, "No payload expected %s l:%d\n", cmdname(cmd), packet_size); | |
3728 | goto err_out; | |
b411b363 | 3729 | } |
b411b363 | 3730 | |
c13f7e1a LE |
3731 | if (shs) { |
3732 | rv = drbd_recv(mdev, &header->h80.payload, shs); | |
3733 | if (unlikely(rv != shs)) { | |
0ddc5549 LE |
3734 | if (!signal_pending(current)) |
3735 | dev_warn(DEV, "short read while reading sub header: rv=%d\n", rv); | |
c13f7e1a LE |
3736 | goto err_out; |
3737 | } | |
3738 | } | |
3739 | ||
02918be2 | 3740 | rv = drbd_cmd_handler[cmd].function(mdev, cmd, packet_size - shs); |
b411b363 | 3741 | |
02918be2 | 3742 | if (unlikely(!rv)) { |
b411b363 | 3743 | dev_err(DEV, "error receiving %s, l: %d!\n", |
02918be2 PR |
3744 | cmdname(cmd), packet_size); |
3745 | goto err_out; | |
b411b363 PR |
3746 | } |
3747 | } | |
b411b363 | 3748 | |
02918be2 PR |
3749 | if (0) { |
3750 | err_out: | |
3751 | drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); | |
3752 | } | |
856c50c7 LE |
3753 | /* If we leave here, we probably want to update at least the |
3754 | * "Connected" indicator on stable storage. Do so explicitly here. */ | |
3755 | drbd_md_sync(mdev); | |
b411b363 PR |
3756 | } |
3757 | ||
3758 | void drbd_flush_workqueue(struct drbd_conf *mdev) | |
3759 | { | |
3760 | struct drbd_wq_barrier barr; | |
3761 | ||
3762 | barr.w.cb = w_prev_work_done; | |
3763 | init_completion(&barr.done); | |
3764 | drbd_queue_work(&mdev->data.work, &barr.w); | |
3765 | wait_for_completion(&barr.done); | |
3766 | } | |
3767 | ||
3768 | static void drbd_disconnect(struct drbd_conf *mdev) | |
3769 | { | |
3770 | enum drbd_fencing_p fp; | |
3771 | union drbd_state os, ns; | |
3772 | int rv = SS_UNKNOWN_ERROR; | |
3773 | unsigned int i; | |
3774 | ||
3775 | if (mdev->state.conn == C_STANDALONE) | |
3776 | return; | |
b411b363 PR |
3777 | |
3778 | /* asender does not clean up anything. it must not interfere, either */ | |
3779 | drbd_thread_stop(&mdev->asender); | |
b411b363 | 3780 | drbd_free_sock(mdev); |
b411b363 | 3781 | |
85719573 | 3782 | /* wait for current activity to cease. */ |
b411b363 PR |
3783 | spin_lock_irq(&mdev->req_lock); |
3784 | _drbd_wait_ee_list_empty(mdev, &mdev->active_ee); | |
3785 | _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee); | |
3786 | _drbd_wait_ee_list_empty(mdev, &mdev->read_ee); | |
3787 | spin_unlock_irq(&mdev->req_lock); | |
3788 | ||
3789 | /* We do not have data structures that would allow us to | |
3790 | * get the rs_pending_cnt down to 0 again. | |
3791 | * * On C_SYNC_TARGET we do not have any data structures describing | |
3792 | * the pending RSDataRequest's we have sent. | |
3793 | * * On C_SYNC_SOURCE there is no data structure that tracks | |
3794 | * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. | |
3795 | * And no, it is not the sum of the reference counts in the | |
3796 | * resync_LRU. The resync_LRU tracks the whole operation including | |
3797 | * the disk-IO, while the rs_pending_cnt only tracks the blocks | |
3798 | * on the fly. */ | |
3799 | drbd_rs_cancel_all(mdev); | |
3800 | mdev->rs_total = 0; | |
3801 | mdev->rs_failed = 0; | |
3802 | atomic_set(&mdev->rs_pending_cnt, 0); | |
3803 | wake_up(&mdev->misc_wait); | |
3804 | ||
7fde2be9 PR |
3805 | del_timer(&mdev->request_timer); |
3806 | ||
b411b363 PR |
3807 | /* make sure syncer is stopped and w_resume_next_sg queued */ |
3808 | del_timer_sync(&mdev->resync_timer); | |
b411b363 PR |
3809 | resync_timer_fn((unsigned long)mdev); |
3810 | ||
b411b363 PR |
3811 | /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, |
3812 | * w_make_resync_request etc. which may still be on the worker queue | |
3813 | * to be "canceled" */ | |
3814 | drbd_flush_workqueue(mdev); | |
3815 | ||
3816 | /* This also does reclaim_net_ee(). If we do this too early, we might | |
3817 | * miss some resync ee and pages.*/ | |
3818 | drbd_process_done_ee(mdev); | |
3819 | ||
3820 | kfree(mdev->p_uuid); | |
3821 | mdev->p_uuid = NULL; | |
3822 | ||
fb22c402 | 3823 | if (!is_susp(mdev->state)) |
b411b363 PR |
3824 | tl_clear(mdev); |
3825 | ||
b411b363 PR |
3826 | dev_info(DEV, "Connection closed\n"); |
3827 | ||
3828 | drbd_md_sync(mdev); | |
3829 | ||
3830 | fp = FP_DONT_CARE; | |
3831 | if (get_ldev(mdev)) { | |
3832 | fp = mdev->ldev->dc.fencing; | |
3833 | put_ldev(mdev); | |
3834 | } | |
3835 | ||
87f7be4c PR |
3836 | if (mdev->state.role == R_PRIMARY && fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN) |
3837 | drbd_try_outdate_peer_async(mdev); | |
b411b363 PR |
3838 | |
3839 | spin_lock_irq(&mdev->req_lock); | |
3840 | os = mdev->state; | |
3841 | if (os.conn >= C_UNCONNECTED) { | |
3842 | /* Do not restart in case we are C_DISCONNECTING */ | |
3843 | ns = os; | |
3844 | ns.conn = C_UNCONNECTED; | |
3845 | rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); | |
3846 | } | |
3847 | spin_unlock_irq(&mdev->req_lock); | |
3848 | ||
3849 | if (os.conn == C_DISCONNECTING) { | |
84dfb9f5 | 3850 | wait_event(mdev->net_cnt_wait, atomic_read(&mdev->net_cnt) == 0); |
b411b363 | 3851 | |
b411b363 PR |
3852 | crypto_free_hash(mdev->cram_hmac_tfm); |
3853 | mdev->cram_hmac_tfm = NULL; | |
3854 | ||
3855 | kfree(mdev->net_conf); | |
3856 | mdev->net_conf = NULL; | |
3857 | drbd_request_state(mdev, NS(conn, C_STANDALONE)); | |
3858 | } | |
3859 | ||
20ceb2b2 LE |
3860 | /* serialize with bitmap writeout triggered by the state change, |
3861 | * if any. */ | |
3862 | wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags)); | |
3863 | ||
b411b363 PR |
3864 | /* tcp_close and release of sendpage pages can be deferred. I don't |
3865 | * want to use SO_LINGER, because apparently it can be deferred for | |
3866 | * more than 20 seconds (longest time I checked). | |
3867 | * | |
3868 | * Actually we don't care for exactly when the network stack does its | |
3869 | * put_page(), but release our reference on these pages right here. | |
3870 | */ | |
3871 | i = drbd_release_ee(mdev, &mdev->net_ee); | |
3872 | if (i) | |
3873 | dev_info(DEV, "net_ee not empty, killed %u entries\n", i); | |
435f0740 LE |
3874 | i = atomic_read(&mdev->pp_in_use_by_net); |
3875 | if (i) | |
3876 | dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i); | |
b411b363 PR |
3877 | i = atomic_read(&mdev->pp_in_use); |
3878 | if (i) | |
45bb912b | 3879 | dev_info(DEV, "pp_in_use = %d, expected 0\n", i); |
b411b363 PR |
3880 | |
3881 | D_ASSERT(list_empty(&mdev->read_ee)); | |
3882 | D_ASSERT(list_empty(&mdev->active_ee)); | |
3883 | D_ASSERT(list_empty(&mdev->sync_ee)); | |
3884 | D_ASSERT(list_empty(&mdev->done_ee)); | |
3885 | ||
3886 | /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ | |
3887 | atomic_set(&mdev->current_epoch->epoch_size, 0); | |
3888 | D_ASSERT(list_empty(&mdev->current_epoch->list)); | |
3889 | } | |
3890 | ||
3891 | /* | |
3892 | * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version | |
3893 | * we can agree on is stored in agreed_pro_version. | |
3894 | * | |
3895 | * feature flags and the reserved array should be enough room for future | |
3896 | * enhancements of the handshake protocol, and possible plugins... | |
3897 | * | |
3898 | * for now, they are expected to be zero, but ignored. | |
3899 | */ | |
3900 | static int drbd_send_handshake(struct drbd_conf *mdev) | |
3901 | { | |
3902 | /* ASSERT current == mdev->receiver ... */ | |
3903 | struct p_handshake *p = &mdev->data.sbuf.handshake; | |
3904 | int ok; | |
3905 | ||
3906 | if (mutex_lock_interruptible(&mdev->data.mutex)) { | |
3907 | dev_err(DEV, "interrupted during initial handshake\n"); | |
3908 | return 0; /* interrupted. not ok. */ | |
3909 | } | |
3910 | ||
3911 | if (mdev->data.socket == NULL) { | |
3912 | mutex_unlock(&mdev->data.mutex); | |
3913 | return 0; | |
3914 | } | |
3915 | ||
3916 | memset(p, 0, sizeof(*p)); | |
3917 | p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); | |
3918 | p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); | |
3919 | ok = _drbd_send_cmd( mdev, mdev->data.socket, P_HAND_SHAKE, | |
0b70a13d | 3920 | (struct p_header80 *)p, sizeof(*p), 0 ); |
b411b363 PR |
3921 | mutex_unlock(&mdev->data.mutex); |
3922 | return ok; | |
3923 | } | |
3924 | ||
3925 | /* | |
3926 | * return values: | |
3927 | * 1 yes, we have a valid connection | |
3928 | * 0 oops, did not work out, please try again | |
3929 | * -1 peer talks different language, | |
3930 | * no point in trying again, please go standalone. | |
3931 | */ | |
3932 | static int drbd_do_handshake(struct drbd_conf *mdev) | |
3933 | { | |
3934 | /* ASSERT current == mdev->receiver ... */ | |
3935 | struct p_handshake *p = &mdev->data.rbuf.handshake; | |
02918be2 PR |
3936 | const int expect = sizeof(struct p_handshake) - sizeof(struct p_header80); |
3937 | unsigned int length; | |
3938 | enum drbd_packets cmd; | |
b411b363 PR |
3939 | int rv; |
3940 | ||
3941 | rv = drbd_send_handshake(mdev); | |
3942 | if (!rv) | |
3943 | return 0; | |
3944 | ||
02918be2 | 3945 | rv = drbd_recv_header(mdev, &cmd, &length); |
b411b363 PR |
3946 | if (!rv) |
3947 | return 0; | |
3948 | ||
02918be2 | 3949 | if (cmd != P_HAND_SHAKE) { |
b411b363 | 3950 | dev_err(DEV, "expected HandShake packet, received: %s (0x%04x)\n", |
02918be2 | 3951 | cmdname(cmd), cmd); |
b411b363 PR |
3952 | return -1; |
3953 | } | |
3954 | ||
02918be2 | 3955 | if (length != expect) { |
b411b363 | 3956 | dev_err(DEV, "expected HandShake length: %u, received: %u\n", |
02918be2 | 3957 | expect, length); |
b411b363 PR |
3958 | return -1; |
3959 | } | |
3960 | ||
3961 | rv = drbd_recv(mdev, &p->head.payload, expect); | |
3962 | ||
3963 | if (rv != expect) { | |
0ddc5549 LE |
3964 | if (!signal_pending(current)) |
3965 | dev_warn(DEV, "short read receiving handshake packet: l=%u\n", rv); | |
b411b363 PR |
3966 | return 0; |
3967 | } | |
3968 | ||
b411b363 PR |
3969 | p->protocol_min = be32_to_cpu(p->protocol_min); |
3970 | p->protocol_max = be32_to_cpu(p->protocol_max); | |
3971 | if (p->protocol_max == 0) | |
3972 | p->protocol_max = p->protocol_min; | |
3973 | ||
3974 | if (PRO_VERSION_MAX < p->protocol_min || | |
3975 | PRO_VERSION_MIN > p->protocol_max) | |
3976 | goto incompat; | |
3977 | ||
3978 | mdev->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); | |
3979 | ||
3980 | dev_info(DEV, "Handshake successful: " | |
3981 | "Agreed network protocol version %d\n", mdev->agreed_pro_version); | |
3982 | ||
3983 | return 1; | |
3984 | ||
3985 | incompat: | |
3986 | dev_err(DEV, "incompatible DRBD dialects: " | |
3987 | "I support %d-%d, peer supports %d-%d\n", | |
3988 | PRO_VERSION_MIN, PRO_VERSION_MAX, | |
3989 | p->protocol_min, p->protocol_max); | |
3990 | return -1; | |
3991 | } | |
3992 | ||
3993 | #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) | |
3994 | static int drbd_do_auth(struct drbd_conf *mdev) | |
3995 | { | |
3996 | dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); | |
3997 | dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); | |
b10d96cb | 3998 | return -1; |
b411b363 PR |
3999 | } |
4000 | #else | |
4001 | #define CHALLENGE_LEN 64 | |
b10d96cb JT |
4002 | |
4003 | /* Return value: | |
4004 | 1 - auth succeeded, | |
4005 | 0 - failed, try again (network error), | |
4006 | -1 - auth failed, don't try again. | |
4007 | */ | |
4008 | ||
b411b363 PR |
4009 | static int drbd_do_auth(struct drbd_conf *mdev) |
4010 | { | |
4011 | char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ | |
4012 | struct scatterlist sg; | |
4013 | char *response = NULL; | |
4014 | char *right_response = NULL; | |
4015 | char *peers_ch = NULL; | |
b411b363 PR |
4016 | unsigned int key_len = strlen(mdev->net_conf->shared_secret); |
4017 | unsigned int resp_size; | |
4018 | struct hash_desc desc; | |
02918be2 PR |
4019 | enum drbd_packets cmd; |
4020 | unsigned int length; | |
b411b363 PR |
4021 | int rv; |
4022 | ||
4023 | desc.tfm = mdev->cram_hmac_tfm; | |
4024 | desc.flags = 0; | |
4025 | ||
4026 | rv = crypto_hash_setkey(mdev->cram_hmac_tfm, | |
4027 | (u8 *)mdev->net_conf->shared_secret, key_len); | |
4028 | if (rv) { | |
4029 | dev_err(DEV, "crypto_hash_setkey() failed with %d\n", rv); | |
b10d96cb | 4030 | rv = -1; |
b411b363 PR |
4031 | goto fail; |
4032 | } | |
4033 | ||
4034 | get_random_bytes(my_challenge, CHALLENGE_LEN); | |
4035 | ||
4036 | rv = drbd_send_cmd2(mdev, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN); | |
4037 | if (!rv) | |
4038 | goto fail; | |
4039 | ||
02918be2 | 4040 | rv = drbd_recv_header(mdev, &cmd, &length); |
b411b363 PR |
4041 | if (!rv) |
4042 | goto fail; | |
4043 | ||
02918be2 | 4044 | if (cmd != P_AUTH_CHALLENGE) { |
b411b363 | 4045 | dev_err(DEV, "expected AuthChallenge packet, received: %s (0x%04x)\n", |
02918be2 | 4046 | cmdname(cmd), cmd); |
b411b363 PR |
4047 | rv = 0; |
4048 | goto fail; | |
4049 | } | |
4050 | ||
02918be2 | 4051 | if (length > CHALLENGE_LEN * 2) { |
b411b363 | 4052 | dev_err(DEV, "expected AuthChallenge payload too big.\n"); |
b10d96cb | 4053 | rv = -1; |
b411b363 PR |
4054 | goto fail; |
4055 | } | |
4056 | ||
02918be2 | 4057 | peers_ch = kmalloc(length, GFP_NOIO); |
b411b363 PR |
4058 | if (peers_ch == NULL) { |
4059 | dev_err(DEV, "kmalloc of peers_ch failed\n"); | |
b10d96cb | 4060 | rv = -1; |
b411b363 PR |
4061 | goto fail; |
4062 | } | |
4063 | ||
02918be2 | 4064 | rv = drbd_recv(mdev, peers_ch, length); |
b411b363 | 4065 | |
02918be2 | 4066 | if (rv != length) { |
0ddc5549 LE |
4067 | if (!signal_pending(current)) |
4068 | dev_warn(DEV, "short read AuthChallenge: l=%u\n", rv); | |
b411b363 PR |
4069 | rv = 0; |
4070 | goto fail; | |
4071 | } | |
4072 | ||
4073 | resp_size = crypto_hash_digestsize(mdev->cram_hmac_tfm); | |
4074 | response = kmalloc(resp_size, GFP_NOIO); | |
4075 | if (response == NULL) { | |
4076 | dev_err(DEV, "kmalloc of response failed\n"); | |
b10d96cb | 4077 | rv = -1; |
b411b363 PR |
4078 | goto fail; |
4079 | } | |
4080 | ||
4081 | sg_init_table(&sg, 1); | |
02918be2 | 4082 | sg_set_buf(&sg, peers_ch, length); |
b411b363 PR |
4083 | |
4084 | rv = crypto_hash_digest(&desc, &sg, sg.length, response); | |
4085 | if (rv) { | |
4086 | dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv); | |
b10d96cb | 4087 | rv = -1; |
b411b363 PR |
4088 | goto fail; |
4089 | } | |
4090 | ||
4091 | rv = drbd_send_cmd2(mdev, P_AUTH_RESPONSE, response, resp_size); | |
4092 | if (!rv) | |
4093 | goto fail; | |
4094 | ||
02918be2 | 4095 | rv = drbd_recv_header(mdev, &cmd, &length); |
b411b363 PR |
4096 | if (!rv) |
4097 | goto fail; | |
4098 | ||
02918be2 | 4099 | if (cmd != P_AUTH_RESPONSE) { |
b411b363 | 4100 | dev_err(DEV, "expected AuthResponse packet, received: %s (0x%04x)\n", |
02918be2 | 4101 | cmdname(cmd), cmd); |
b411b363 PR |
4102 | rv = 0; |
4103 | goto fail; | |
4104 | } | |
4105 | ||
02918be2 | 4106 | if (length != resp_size) { |
b411b363 PR |
4107 | dev_err(DEV, "expected AuthResponse payload of wrong size\n"); |
4108 | rv = 0; | |
4109 | goto fail; | |
4110 | } | |
4111 | ||
4112 | rv = drbd_recv(mdev, response , resp_size); | |
4113 | ||
4114 | if (rv != resp_size) { | |
0ddc5549 LE |
4115 | if (!signal_pending(current)) |
4116 | dev_warn(DEV, "short read receiving AuthResponse: l=%u\n", rv); | |
b411b363 PR |
4117 | rv = 0; |
4118 | goto fail; | |
4119 | } | |
4120 | ||
4121 | right_response = kmalloc(resp_size, GFP_NOIO); | |
2d1ee87d | 4122 | if (right_response == NULL) { |
b411b363 | 4123 | dev_err(DEV, "kmalloc of right_response failed\n"); |
b10d96cb | 4124 | rv = -1; |
b411b363 PR |
4125 | goto fail; |
4126 | } | |
4127 | ||
4128 | sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); | |
4129 | ||
4130 | rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); | |
4131 | if (rv) { | |
4132 | dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv); | |
b10d96cb | 4133 | rv = -1; |
b411b363 PR |
4134 | goto fail; |
4135 | } | |
4136 | ||
4137 | rv = !memcmp(response, right_response, resp_size); | |
4138 | ||
4139 | if (rv) | |
4140 | dev_info(DEV, "Peer authenticated using %d bytes of '%s' HMAC\n", | |
4141 | resp_size, mdev->net_conf->cram_hmac_alg); | |
b10d96cb JT |
4142 | else |
4143 | rv = -1; | |
b411b363 PR |
4144 | |
4145 | fail: | |
4146 | kfree(peers_ch); | |
4147 | kfree(response); | |
4148 | kfree(right_response); | |
4149 | ||
4150 | return rv; | |
4151 | } | |
4152 | #endif | |
4153 | ||
4154 | int drbdd_init(struct drbd_thread *thi) | |
4155 | { | |
4156 | struct drbd_conf *mdev = thi->mdev; | |
4157 | unsigned int minor = mdev_to_minor(mdev); | |
4158 | int h; | |
4159 | ||
4160 | sprintf(current->comm, "drbd%d_receiver", minor); | |
4161 | ||
4162 | dev_info(DEV, "receiver (re)started\n"); | |
4163 | ||
4164 | do { | |
4165 | h = drbd_connect(mdev); | |
4166 | if (h == 0) { | |
4167 | drbd_disconnect(mdev); | |
20ee6390 | 4168 | schedule_timeout_interruptible(HZ); |
b411b363 PR |
4169 | } |
4170 | if (h == -1) { | |
4171 | dev_warn(DEV, "Discarding network configuration.\n"); | |
4172 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
4173 | } | |
4174 | } while (h == 0); | |
4175 | ||
4176 | if (h > 0) { | |
4177 | if (get_net_conf(mdev)) { | |
4178 | drbdd(mdev); | |
4179 | put_net_conf(mdev); | |
4180 | } | |
4181 | } | |
4182 | ||
4183 | drbd_disconnect(mdev); | |
4184 | ||
4185 | dev_info(DEV, "receiver terminated\n"); | |
4186 | return 0; | |
4187 | } | |
4188 | ||
4189 | /* ********* acknowledge sender ******** */ | |
4190 | ||
0b70a13d | 4191 | static int got_RqSReply(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4192 | { |
4193 | struct p_req_state_reply *p = (struct p_req_state_reply *)h; | |
4194 | ||
4195 | int retcode = be32_to_cpu(p->retcode); | |
4196 | ||
4197 | if (retcode >= SS_SUCCESS) { | |
4198 | set_bit(CL_ST_CHG_SUCCESS, &mdev->flags); | |
4199 | } else { | |
4200 | set_bit(CL_ST_CHG_FAIL, &mdev->flags); | |
4201 | dev_err(DEV, "Requested state change failed by peer: %s (%d)\n", | |
4202 | drbd_set_st_err_str(retcode), retcode); | |
4203 | } | |
4204 | wake_up(&mdev->state_wait); | |
4205 | ||
81e84650 | 4206 | return true; |
b411b363 PR |
4207 | } |
4208 | ||
0b70a13d | 4209 | static int got_Ping(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4210 | { |
4211 | return drbd_send_ping_ack(mdev); | |
4212 | ||
4213 | } | |
4214 | ||
0b70a13d | 4215 | static int got_PingAck(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4216 | { |
4217 | /* restore idle timeout */ | |
4218 | mdev->meta.socket->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ; | |
309d1608 PR |
4219 | if (!test_and_set_bit(GOT_PING_ACK, &mdev->flags)) |
4220 | wake_up(&mdev->misc_wait); | |
b411b363 | 4221 | |
81e84650 | 4222 | return true; |
b411b363 PR |
4223 | } |
4224 | ||
0b70a13d | 4225 | static int got_IsInSync(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4226 | { |
4227 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4228 | sector_t sector = be64_to_cpu(p->sector); | |
4229 | int blksize = be32_to_cpu(p->blksize); | |
4230 | ||
4231 | D_ASSERT(mdev->agreed_pro_version >= 89); | |
4232 | ||
4233 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4234 | ||
1d53f09e LE |
4235 | if (get_ldev(mdev)) { |
4236 | drbd_rs_complete_io(mdev, sector); | |
4237 | drbd_set_in_sync(mdev, sector, blksize); | |
4238 | /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ | |
4239 | mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); | |
4240 | put_ldev(mdev); | |
4241 | } | |
b411b363 | 4242 | dec_rs_pending(mdev); |
778f271d | 4243 | atomic_add(blksize >> 9, &mdev->rs_sect_in); |
b411b363 | 4244 | |
81e84650 | 4245 | return true; |
b411b363 PR |
4246 | } |
4247 | ||
b411b363 | 4248 | static int validate_req_change_req_state(struct drbd_conf *mdev, |
28c455ce | 4249 | u64 id, sector_t sector, |
ae3388da | 4250 | struct hlist_head *(*hash_slot)(struct drbd_conf *, sector_t), |
b411b363 PR |
4251 | const char *func, enum drbd_req_event what) |
4252 | { | |
4253 | struct drbd_request *req; | |
4254 | struct bio_and_error m; | |
4255 | ||
4256 | spin_lock_irq(&mdev->req_lock); | |
ae3388da | 4257 | req = find_request(mdev, hash_slot, id, sector, func); |
b411b363 PR |
4258 | if (unlikely(!req)) { |
4259 | spin_unlock_irq(&mdev->req_lock); | |
2deb8336 PR |
4260 | |
4261 | dev_err(DEV, "%s: failed to find req %p, sector %llus\n", func, | |
4262 | (void *)(unsigned long)id, (unsigned long long)sector); | |
81e84650 | 4263 | return false; |
b411b363 PR |
4264 | } |
4265 | __req_mod(req, what, &m); | |
4266 | spin_unlock_irq(&mdev->req_lock); | |
4267 | ||
4268 | if (m.bio) | |
4269 | complete_master_bio(mdev, &m); | |
81e84650 | 4270 | return true; |
b411b363 PR |
4271 | } |
4272 | ||
0b70a13d | 4273 | static int got_BlockAck(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4274 | { |
4275 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4276 | sector_t sector = be64_to_cpu(p->sector); | |
4277 | int blksize = be32_to_cpu(p->blksize); | |
4278 | enum drbd_req_event what; | |
4279 | ||
4280 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4281 | ||
579b57ed | 4282 | if (p->block_id == ID_SYNCER) { |
b411b363 PR |
4283 | drbd_set_in_sync(mdev, sector, blksize); |
4284 | dec_rs_pending(mdev); | |
81e84650 | 4285 | return true; |
b411b363 PR |
4286 | } |
4287 | switch (be16_to_cpu(h->command)) { | |
4288 | case P_RS_WRITE_ACK: | |
4289 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
4290 | what = write_acked_by_peer_and_sis; | |
4291 | break; | |
4292 | case P_WRITE_ACK: | |
4293 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
4294 | what = write_acked_by_peer; | |
4295 | break; | |
4296 | case P_RECV_ACK: | |
4297 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_B); | |
4298 | what = recv_acked_by_peer; | |
4299 | break; | |
4300 | case P_DISCARD_ACK: | |
4301 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
4302 | what = conflict_discarded_by_peer; | |
4303 | break; | |
4304 | default: | |
4305 | D_ASSERT(0); | |
81e84650 | 4306 | return false; |
b411b363 PR |
4307 | } |
4308 | ||
4309 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
ae3388da | 4310 | tl_hash_slot, __func__, what); |
b411b363 PR |
4311 | } |
4312 | ||
0b70a13d | 4313 | static int got_NegAck(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4314 | { |
4315 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4316 | sector_t sector = be64_to_cpu(p->sector); | |
2deb8336 PR |
4317 | int size = be32_to_cpu(p->blksize); |
4318 | struct drbd_request *req; | |
4319 | struct bio_and_error m; | |
b411b363 PR |
4320 | |
4321 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4322 | ||
579b57ed | 4323 | if (p->block_id == ID_SYNCER) { |
b411b363 PR |
4324 | dec_rs_pending(mdev); |
4325 | drbd_rs_failed_io(mdev, sector, size); | |
81e84650 | 4326 | return true; |
b411b363 | 4327 | } |
2deb8336 PR |
4328 | |
4329 | spin_lock_irq(&mdev->req_lock); | |
ae3388da | 4330 | req = find_request(mdev, tl_hash_slot, p->block_id, sector, __func__); |
2deb8336 PR |
4331 | if (!req) { |
4332 | spin_unlock_irq(&mdev->req_lock); | |
4333 | if (mdev->net_conf->wire_protocol == DRBD_PROT_A || | |
4334 | mdev->net_conf->wire_protocol == DRBD_PROT_B) { | |
4335 | /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs. | |
4336 | The master bio might already be completed, therefore the | |
4337 | request is no longer in the collision hash. | |
4338 | => Do not try to validate block_id as request. */ | |
4339 | /* In Protocol B we might already have got a P_RECV_ACK | |
4340 | but then get a P_NEG_ACK after wards. */ | |
4341 | drbd_set_out_of_sync(mdev, sector, size); | |
4342 | return true; | |
4343 | } else { | |
4344 | dev_err(DEV, "%s: failed to find req %p, sector %llus\n", __func__, | |
4345 | (void *)(unsigned long)p->block_id, (unsigned long long)sector); | |
4346 | return false; | |
4347 | } | |
4348 | } | |
4349 | __req_mod(req, neg_acked, &m); | |
4350 | spin_unlock_irq(&mdev->req_lock); | |
4351 | ||
4352 | if (m.bio) | |
4353 | complete_master_bio(mdev, &m); | |
4354 | return true; | |
b411b363 PR |
4355 | } |
4356 | ||
0b70a13d | 4357 | static int got_NegDReply(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4358 | { |
4359 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4360 | sector_t sector = be64_to_cpu(p->sector); | |
4361 | ||
4362 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4363 | dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n", | |
4364 | (unsigned long long)sector, be32_to_cpu(p->blksize)); | |
4365 | ||
4366 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
ae3388da | 4367 | ar_hash_slot, __func__, neg_acked); |
b411b363 PR |
4368 | } |
4369 | ||
0b70a13d | 4370 | static int got_NegRSDReply(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4371 | { |
4372 | sector_t sector; | |
4373 | int size; | |
4374 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4375 | ||
4376 | sector = be64_to_cpu(p->sector); | |
4377 | size = be32_to_cpu(p->blksize); | |
b411b363 PR |
4378 | |
4379 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4380 | ||
4381 | dec_rs_pending(mdev); | |
4382 | ||
4383 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
4384 | drbd_rs_complete_io(mdev, sector); | |
d612d309 PR |
4385 | switch (be16_to_cpu(h->command)) { |
4386 | case P_NEG_RS_DREPLY: | |
4387 | drbd_rs_failed_io(mdev, sector, size); | |
4388 | case P_RS_CANCEL: | |
4389 | break; | |
4390 | default: | |
4391 | D_ASSERT(0); | |
4392 | put_ldev(mdev); | |
4393 | return false; | |
4394 | } | |
b411b363 PR |
4395 | put_ldev(mdev); |
4396 | } | |
4397 | ||
81e84650 | 4398 | return true; |
b411b363 PR |
4399 | } |
4400 | ||
0b70a13d | 4401 | static int got_BarrierAck(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4402 | { |
4403 | struct p_barrier_ack *p = (struct p_barrier_ack *)h; | |
4404 | ||
4405 | tl_release(mdev, p->barrier, be32_to_cpu(p->set_size)); | |
4406 | ||
c4752ef1 PR |
4407 | if (mdev->state.conn == C_AHEAD && |
4408 | atomic_read(&mdev->ap_in_flight) == 0 && | |
370a43e7 PR |
4409 | !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &mdev->current_epoch->flags)) { |
4410 | mdev->start_resync_timer.expires = jiffies + HZ; | |
4411 | add_timer(&mdev->start_resync_timer); | |
c4752ef1 PR |
4412 | } |
4413 | ||
81e84650 | 4414 | return true; |
b411b363 PR |
4415 | } |
4416 | ||
0b70a13d | 4417 | static int got_OVResult(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4418 | { |
4419 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4420 | struct drbd_work *w; | |
4421 | sector_t sector; | |
4422 | int size; | |
4423 | ||
4424 | sector = be64_to_cpu(p->sector); | |
4425 | size = be32_to_cpu(p->blksize); | |
4426 | ||
4427 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4428 | ||
4429 | if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) | |
4430 | drbd_ov_oos_found(mdev, sector, size); | |
4431 | else | |
4432 | ov_oos_print(mdev); | |
4433 | ||
1d53f09e | 4434 | if (!get_ldev(mdev)) |
81e84650 | 4435 | return true; |
1d53f09e | 4436 | |
b411b363 PR |
4437 | drbd_rs_complete_io(mdev, sector); |
4438 | dec_rs_pending(mdev); | |
4439 | ||
ea5442af LE |
4440 | --mdev->ov_left; |
4441 | ||
4442 | /* let's advance progress step marks only for every other megabyte */ | |
4443 | if ((mdev->ov_left & 0x200) == 0x200) | |
4444 | drbd_advance_rs_marks(mdev, mdev->ov_left); | |
4445 | ||
4446 | if (mdev->ov_left == 0) { | |
b411b363 PR |
4447 | w = kmalloc(sizeof(*w), GFP_NOIO); |
4448 | if (w) { | |
4449 | w->cb = w_ov_finished; | |
4450 | drbd_queue_work_front(&mdev->data.work, w); | |
4451 | } else { | |
4452 | dev_err(DEV, "kmalloc(w) failed."); | |
4453 | ov_oos_print(mdev); | |
4454 | drbd_resync_finished(mdev); | |
4455 | } | |
4456 | } | |
1d53f09e | 4457 | put_ldev(mdev); |
81e84650 | 4458 | return true; |
b411b363 PR |
4459 | } |
4460 | ||
02918be2 | 4461 | static int got_skip(struct drbd_conf *mdev, struct p_header80 *h) |
0ced55a3 | 4462 | { |
81e84650 | 4463 | return true; |
0ced55a3 PR |
4464 | } |
4465 | ||
b411b363 PR |
4466 | struct asender_cmd { |
4467 | size_t pkt_size; | |
0b70a13d | 4468 | int (*process)(struct drbd_conf *mdev, struct p_header80 *h); |
b411b363 PR |
4469 | }; |
4470 | ||
4471 | static struct asender_cmd *get_asender_cmd(int cmd) | |
4472 | { | |
4473 | static struct asender_cmd asender_tbl[] = { | |
4474 | /* anything missing from this table is in | |
4475 | * the drbd_cmd_handler (drbd_default_handler) table, | |
4476 | * see the beginning of drbdd() */ | |
0b70a13d PR |
4477 | [P_PING] = { sizeof(struct p_header80), got_Ping }, |
4478 | [P_PING_ACK] = { sizeof(struct p_header80), got_PingAck }, | |
b411b363 PR |
4479 | [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, |
4480 | [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4481 | [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4482 | [P_DISCARD_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4483 | [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, | |
4484 | [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, | |
4485 | [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply}, | |
4486 | [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, | |
4487 | [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, | |
4488 | [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, | |
4489 | [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, | |
02918be2 | 4490 | [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, |
d612d309 | 4491 | [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply}, |
b411b363 PR |
4492 | [P_MAX_CMD] = { 0, NULL }, |
4493 | }; | |
4494 | if (cmd > P_MAX_CMD || asender_tbl[cmd].process == NULL) | |
4495 | return NULL; | |
4496 | return &asender_tbl[cmd]; | |
4497 | } | |
4498 | ||
4499 | int drbd_asender(struct drbd_thread *thi) | |
4500 | { | |
4501 | struct drbd_conf *mdev = thi->mdev; | |
02918be2 | 4502 | struct p_header80 *h = &mdev->meta.rbuf.header.h80; |
b411b363 PR |
4503 | struct asender_cmd *cmd = NULL; |
4504 | ||
4505 | int rv, len; | |
4506 | void *buf = h; | |
4507 | int received = 0; | |
0b70a13d | 4508 | int expect = sizeof(struct p_header80); |
b411b363 | 4509 | int empty; |
f36af18c | 4510 | int ping_timeout_active = 0; |
b411b363 PR |
4511 | |
4512 | sprintf(current->comm, "drbd%d_asender", mdev_to_minor(mdev)); | |
4513 | ||
4514 | current->policy = SCHED_RR; /* Make this a realtime task! */ | |
4515 | current->rt_priority = 2; /* more important than all other tasks */ | |
4516 | ||
4517 | while (get_t_state(thi) == Running) { | |
4518 | drbd_thread_current_set_cpu(mdev); | |
4519 | if (test_and_clear_bit(SEND_PING, &mdev->flags)) { | |
4520 | ERR_IF(!drbd_send_ping(mdev)) goto reconnect; | |
4521 | mdev->meta.socket->sk->sk_rcvtimeo = | |
4522 | mdev->net_conf->ping_timeo*HZ/10; | |
f36af18c | 4523 | ping_timeout_active = 1; |
b411b363 PR |
4524 | } |
4525 | ||
4526 | /* conditionally cork; | |
4527 | * it may hurt latency if we cork without much to send */ | |
4528 | if (!mdev->net_conf->no_cork && | |
4529 | 3 < atomic_read(&mdev->unacked_cnt)) | |
4530 | drbd_tcp_cork(mdev->meta.socket); | |
4531 | while (1) { | |
4532 | clear_bit(SIGNAL_ASENDER, &mdev->flags); | |
4533 | flush_signals(current); | |
0f8488e1 | 4534 | if (!drbd_process_done_ee(mdev)) |
b411b363 | 4535 | goto reconnect; |
b411b363 PR |
4536 | /* to avoid race with newly queued ACKs */ |
4537 | set_bit(SIGNAL_ASENDER, &mdev->flags); | |
4538 | spin_lock_irq(&mdev->req_lock); | |
4539 | empty = list_empty(&mdev->done_ee); | |
4540 | spin_unlock_irq(&mdev->req_lock); | |
4541 | /* new ack may have been queued right here, | |
4542 | * but then there is also a signal pending, | |
4543 | * and we start over... */ | |
4544 | if (empty) | |
4545 | break; | |
4546 | } | |
4547 | /* but unconditionally uncork unless disabled */ | |
4548 | if (!mdev->net_conf->no_cork) | |
4549 | drbd_tcp_uncork(mdev->meta.socket); | |
4550 | ||
4551 | /* short circuit, recv_msg would return EINTR anyways. */ | |
4552 | if (signal_pending(current)) | |
4553 | continue; | |
4554 | ||
4555 | rv = drbd_recv_short(mdev, mdev->meta.socket, | |
4556 | buf, expect-received, 0); | |
4557 | clear_bit(SIGNAL_ASENDER, &mdev->flags); | |
4558 | ||
4559 | flush_signals(current); | |
4560 | ||
4561 | /* Note: | |
4562 | * -EINTR (on meta) we got a signal | |
4563 | * -EAGAIN (on meta) rcvtimeo expired | |
4564 | * -ECONNRESET other side closed the connection | |
4565 | * -ERESTARTSYS (on data) we got a signal | |
4566 | * rv < 0 other than above: unexpected error! | |
4567 | * rv == expected: full header or command | |
4568 | * rv < expected: "woken" by signal during receive | |
4569 | * rv == 0 : "connection shut down by peer" | |
4570 | */ | |
4571 | if (likely(rv > 0)) { | |
4572 | received += rv; | |
4573 | buf += rv; | |
4574 | } else if (rv == 0) { | |
4575 | dev_err(DEV, "meta connection shut down by peer.\n"); | |
4576 | goto reconnect; | |
4577 | } else if (rv == -EAGAIN) { | |
cb6518cb LE |
4578 | /* If the data socket received something meanwhile, |
4579 | * that is good enough: peer is still alive. */ | |
4580 | if (time_after(mdev->last_received, | |
4581 | jiffies - mdev->meta.socket->sk->sk_rcvtimeo)) | |
4582 | continue; | |
f36af18c | 4583 | if (ping_timeout_active) { |
b411b363 PR |
4584 | dev_err(DEV, "PingAck did not arrive in time.\n"); |
4585 | goto reconnect; | |
4586 | } | |
4587 | set_bit(SEND_PING, &mdev->flags); | |
4588 | continue; | |
4589 | } else if (rv == -EINTR) { | |
4590 | continue; | |
4591 | } else { | |
4592 | dev_err(DEV, "sock_recvmsg returned %d\n", rv); | |
4593 | goto reconnect; | |
4594 | } | |
4595 | ||
4596 | if (received == expect && cmd == NULL) { | |
ca9bc12b | 4597 | if (unlikely(h->magic != cpu_to_be32(DRBD_MAGIC))) { |
004352fa LE |
4598 | dev_err(DEV, "magic?? on meta m: 0x%08x c: %d l: %d\n", |
4599 | be32_to_cpu(h->magic), | |
4600 | be16_to_cpu(h->command), | |
4601 | be16_to_cpu(h->length)); | |
b411b363 PR |
4602 | goto reconnect; |
4603 | } | |
4604 | cmd = get_asender_cmd(be16_to_cpu(h->command)); | |
4605 | len = be16_to_cpu(h->length); | |
4606 | if (unlikely(cmd == NULL)) { | |
004352fa LE |
4607 | dev_err(DEV, "unknown command?? on meta m: 0x%08x c: %d l: %d\n", |
4608 | be32_to_cpu(h->magic), | |
4609 | be16_to_cpu(h->command), | |
4610 | be16_to_cpu(h->length)); | |
b411b363 PR |
4611 | goto disconnect; |
4612 | } | |
4613 | expect = cmd->pkt_size; | |
0b70a13d | 4614 | ERR_IF(len != expect-sizeof(struct p_header80)) |
b411b363 | 4615 | goto reconnect; |
b411b363 PR |
4616 | } |
4617 | if (received == expect) { | |
cb6518cb | 4618 | mdev->last_received = jiffies; |
b411b363 | 4619 | D_ASSERT(cmd != NULL); |
b411b363 PR |
4620 | if (!cmd->process(mdev, h)) |
4621 | goto reconnect; | |
4622 | ||
f36af18c LE |
4623 | /* the idle_timeout (ping-int) |
4624 | * has been restored in got_PingAck() */ | |
4625 | if (cmd == get_asender_cmd(P_PING_ACK)) | |
4626 | ping_timeout_active = 0; | |
4627 | ||
b411b363 PR |
4628 | buf = h; |
4629 | received = 0; | |
0b70a13d | 4630 | expect = sizeof(struct p_header80); |
b411b363 PR |
4631 | cmd = NULL; |
4632 | } | |
4633 | } | |
4634 | ||
4635 | if (0) { | |
4636 | reconnect: | |
4637 | drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE)); | |
856c50c7 | 4638 | drbd_md_sync(mdev); |
b411b363 PR |
4639 | } |
4640 | if (0) { | |
4641 | disconnect: | |
4642 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
856c50c7 | 4643 | drbd_md_sync(mdev); |
b411b363 PR |
4644 | } |
4645 | clear_bit(SIGNAL_ASENDER, &mdev->flags); | |
4646 | ||
4647 | D_ASSERT(mdev->state.conn < C_CONNECTED); | |
4648 | dev_info(DEV, "asender terminated\n"); | |
4649 | ||
4650 | return 0; | |
4651 | } |