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; |
45bb912b | 344 | e->block_id = id; |
b411b363 | 345 | |
b411b363 PR |
346 | return e; |
347 | ||
45bb912b | 348 | fail: |
b411b363 | 349 | mempool_free(e, drbd_ee_mempool); |
b411b363 PR |
350 | return NULL; |
351 | } | |
352 | ||
435f0740 | 353 | void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, int is_net) |
b411b363 | 354 | { |
c36c3ced LE |
355 | if (e->flags & EE_HAS_DIGEST) |
356 | kfree(e->digest); | |
435f0740 | 357 | drbd_pp_free(mdev, e->pages, is_net); |
45bb912b | 358 | D_ASSERT(atomic_read(&e->pending_bios) == 0); |
24c4830c | 359 | D_ASSERT(hlist_unhashed(&e->collision)); |
b411b363 PR |
360 | mempool_free(e, drbd_ee_mempool); |
361 | } | |
362 | ||
363 | int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list) | |
364 | { | |
365 | LIST_HEAD(work_list); | |
366 | struct drbd_epoch_entry *e, *t; | |
367 | int count = 0; | |
435f0740 | 368 | int is_net = list == &mdev->net_ee; |
b411b363 PR |
369 | |
370 | spin_lock_irq(&mdev->req_lock); | |
371 | list_splice_init(list, &work_list); | |
372 | spin_unlock_irq(&mdev->req_lock); | |
373 | ||
374 | list_for_each_entry_safe(e, t, &work_list, w.list) { | |
435f0740 | 375 | drbd_free_some_ee(mdev, e, is_net); |
b411b363 PR |
376 | count++; |
377 | } | |
378 | return count; | |
379 | } | |
380 | ||
381 | ||
382 | /* | |
383 | * This function is called from _asender only_ | |
384 | * but see also comments in _req_mod(,barrier_acked) | |
385 | * and receive_Barrier. | |
386 | * | |
387 | * Move entries from net_ee to done_ee, if ready. | |
388 | * Grab done_ee, call all callbacks, free the entries. | |
389 | * The callbacks typically send out ACKs. | |
390 | */ | |
391 | static int drbd_process_done_ee(struct drbd_conf *mdev) | |
392 | { | |
393 | LIST_HEAD(work_list); | |
394 | LIST_HEAD(reclaimed); | |
395 | struct drbd_epoch_entry *e, *t; | |
396 | int ok = (mdev->state.conn >= C_WF_REPORT_PARAMS); | |
397 | ||
398 | spin_lock_irq(&mdev->req_lock); | |
399 | reclaim_net_ee(mdev, &reclaimed); | |
400 | list_splice_init(&mdev->done_ee, &work_list); | |
401 | spin_unlock_irq(&mdev->req_lock); | |
402 | ||
403 | list_for_each_entry_safe(e, t, &reclaimed, w.list) | |
435f0740 | 404 | drbd_free_net_ee(mdev, e); |
b411b363 PR |
405 | |
406 | /* possible callbacks here: | |
407 | * e_end_block, and e_end_resync_block, e_send_discard_ack. | |
408 | * all ignore the last argument. | |
409 | */ | |
410 | list_for_each_entry_safe(e, t, &work_list, w.list) { | |
b411b363 PR |
411 | /* list_del not necessary, next/prev members not touched */ |
412 | ok = e->w.cb(mdev, &e->w, !ok) && ok; | |
413 | drbd_free_ee(mdev, e); | |
414 | } | |
415 | wake_up(&mdev->ee_wait); | |
416 | ||
417 | return ok; | |
418 | } | |
419 | ||
420 | void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) | |
421 | { | |
422 | DEFINE_WAIT(wait); | |
423 | ||
424 | /* avoids spin_lock/unlock | |
425 | * and calling prepare_to_wait in the fast path */ | |
426 | while (!list_empty(head)) { | |
427 | prepare_to_wait(&mdev->ee_wait, &wait, TASK_UNINTERRUPTIBLE); | |
428 | spin_unlock_irq(&mdev->req_lock); | |
7eaceacc | 429 | io_schedule(); |
b411b363 PR |
430 | finish_wait(&mdev->ee_wait, &wait); |
431 | spin_lock_irq(&mdev->req_lock); | |
432 | } | |
433 | } | |
434 | ||
435 | void drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) | |
436 | { | |
437 | spin_lock_irq(&mdev->req_lock); | |
438 | _drbd_wait_ee_list_empty(mdev, head); | |
439 | spin_unlock_irq(&mdev->req_lock); | |
440 | } | |
441 | ||
442 | /* see also kernel_accept; which is only present since 2.6.18. | |
443 | * also we want to log which part of it failed, exactly */ | |
444 | static int drbd_accept(struct drbd_conf *mdev, const char **what, | |
445 | struct socket *sock, struct socket **newsock) | |
446 | { | |
447 | struct sock *sk = sock->sk; | |
448 | int err = 0; | |
449 | ||
450 | *what = "listen"; | |
451 | err = sock->ops->listen(sock, 5); | |
452 | if (err < 0) | |
453 | goto out; | |
454 | ||
455 | *what = "sock_create_lite"; | |
456 | err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol, | |
457 | newsock); | |
458 | if (err < 0) | |
459 | goto out; | |
460 | ||
461 | *what = "accept"; | |
462 | err = sock->ops->accept(sock, *newsock, 0); | |
463 | if (err < 0) { | |
464 | sock_release(*newsock); | |
465 | *newsock = NULL; | |
466 | goto out; | |
467 | } | |
468 | (*newsock)->ops = sock->ops; | |
469 | ||
470 | out: | |
471 | return err; | |
472 | } | |
473 | ||
474 | static int drbd_recv_short(struct drbd_conf *mdev, struct socket *sock, | |
475 | void *buf, size_t size, int flags) | |
476 | { | |
477 | mm_segment_t oldfs; | |
478 | struct kvec iov = { | |
479 | .iov_base = buf, | |
480 | .iov_len = size, | |
481 | }; | |
482 | struct msghdr msg = { | |
483 | .msg_iovlen = 1, | |
484 | .msg_iov = (struct iovec *)&iov, | |
485 | .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) | |
486 | }; | |
487 | int rv; | |
488 | ||
489 | oldfs = get_fs(); | |
490 | set_fs(KERNEL_DS); | |
491 | rv = sock_recvmsg(sock, &msg, size, msg.msg_flags); | |
492 | set_fs(oldfs); | |
493 | ||
494 | return rv; | |
495 | } | |
496 | ||
497 | static int drbd_recv(struct drbd_conf *mdev, void *buf, size_t size) | |
498 | { | |
499 | mm_segment_t oldfs; | |
500 | struct kvec iov = { | |
501 | .iov_base = buf, | |
502 | .iov_len = size, | |
503 | }; | |
504 | struct msghdr msg = { | |
505 | .msg_iovlen = 1, | |
506 | .msg_iov = (struct iovec *)&iov, | |
507 | .msg_flags = MSG_WAITALL | MSG_NOSIGNAL | |
508 | }; | |
509 | int rv; | |
510 | ||
511 | oldfs = get_fs(); | |
512 | set_fs(KERNEL_DS); | |
513 | ||
514 | for (;;) { | |
515 | rv = sock_recvmsg(mdev->data.socket, &msg, size, msg.msg_flags); | |
516 | if (rv == size) | |
517 | break; | |
518 | ||
519 | /* Note: | |
520 | * ECONNRESET other side closed the connection | |
521 | * ERESTARTSYS (on sock) we got a signal | |
522 | */ | |
523 | ||
524 | if (rv < 0) { | |
525 | if (rv == -ECONNRESET) | |
526 | dev_info(DEV, "sock was reset by peer\n"); | |
527 | else if (rv != -ERESTARTSYS) | |
528 | dev_err(DEV, "sock_recvmsg returned %d\n", rv); | |
529 | break; | |
530 | } else if (rv == 0) { | |
531 | dev_info(DEV, "sock was shut down by peer\n"); | |
532 | break; | |
533 | } else { | |
534 | /* signal came in, or peer/link went down, | |
535 | * after we read a partial message | |
536 | */ | |
537 | /* D_ASSERT(signal_pending(current)); */ | |
538 | break; | |
539 | } | |
540 | }; | |
541 | ||
542 | set_fs(oldfs); | |
543 | ||
544 | if (rv != size) | |
545 | drbd_force_state(mdev, NS(conn, C_BROKEN_PIPE)); | |
546 | ||
547 | return rv; | |
548 | } | |
549 | ||
5dbf1673 LE |
550 | /* quoting tcp(7): |
551 | * On individual connections, the socket buffer size must be set prior to the | |
552 | * listen(2) or connect(2) calls in order to have it take effect. | |
553 | * This is our wrapper to do so. | |
554 | */ | |
555 | static void drbd_setbufsize(struct socket *sock, unsigned int snd, | |
556 | unsigned int rcv) | |
557 | { | |
558 | /* open coded SO_SNDBUF, SO_RCVBUF */ | |
559 | if (snd) { | |
560 | sock->sk->sk_sndbuf = snd; | |
561 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
562 | } | |
563 | if (rcv) { | |
564 | sock->sk->sk_rcvbuf = rcv; | |
565 | sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
566 | } | |
567 | } | |
568 | ||
b411b363 PR |
569 | static struct socket *drbd_try_connect(struct drbd_conf *mdev) |
570 | { | |
571 | const char *what; | |
572 | struct socket *sock; | |
573 | struct sockaddr_in6 src_in6; | |
574 | int err; | |
575 | int disconnect_on_error = 1; | |
576 | ||
577 | if (!get_net_conf(mdev)) | |
578 | return NULL; | |
579 | ||
580 | what = "sock_create_kern"; | |
581 | err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family, | |
582 | SOCK_STREAM, IPPROTO_TCP, &sock); | |
583 | if (err < 0) { | |
584 | sock = NULL; | |
585 | goto out; | |
586 | } | |
587 | ||
588 | sock->sk->sk_rcvtimeo = | |
589 | sock->sk->sk_sndtimeo = mdev->net_conf->try_connect_int*HZ; | |
5dbf1673 LE |
590 | drbd_setbufsize(sock, mdev->net_conf->sndbuf_size, |
591 | mdev->net_conf->rcvbuf_size); | |
b411b363 PR |
592 | |
593 | /* explicitly bind to the configured IP as source IP | |
594 | * for the outgoing connections. | |
595 | * This is needed for multihomed hosts and to be | |
596 | * able to use lo: interfaces for drbd. | |
597 | * Make sure to use 0 as port number, so linux selects | |
598 | * a free one dynamically. | |
599 | */ | |
600 | memcpy(&src_in6, mdev->net_conf->my_addr, | |
601 | min_t(int, mdev->net_conf->my_addr_len, sizeof(src_in6))); | |
602 | if (((struct sockaddr *)mdev->net_conf->my_addr)->sa_family == AF_INET6) | |
603 | src_in6.sin6_port = 0; | |
604 | else | |
605 | ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ | |
606 | ||
607 | what = "bind before connect"; | |
608 | err = sock->ops->bind(sock, | |
609 | (struct sockaddr *) &src_in6, | |
610 | mdev->net_conf->my_addr_len); | |
611 | if (err < 0) | |
612 | goto out; | |
613 | ||
614 | /* connect may fail, peer not yet available. | |
615 | * stay C_WF_CONNECTION, don't go Disconnecting! */ | |
616 | disconnect_on_error = 0; | |
617 | what = "connect"; | |
618 | err = sock->ops->connect(sock, | |
619 | (struct sockaddr *)mdev->net_conf->peer_addr, | |
620 | mdev->net_conf->peer_addr_len, 0); | |
621 | ||
622 | out: | |
623 | if (err < 0) { | |
624 | if (sock) { | |
625 | sock_release(sock); | |
626 | sock = NULL; | |
627 | } | |
628 | switch (-err) { | |
629 | /* timeout, busy, signal pending */ | |
630 | case ETIMEDOUT: case EAGAIN: case EINPROGRESS: | |
631 | case EINTR: case ERESTARTSYS: | |
632 | /* peer not (yet) available, network problem */ | |
633 | case ECONNREFUSED: case ENETUNREACH: | |
634 | case EHOSTDOWN: case EHOSTUNREACH: | |
635 | disconnect_on_error = 0; | |
636 | break; | |
637 | default: | |
638 | dev_err(DEV, "%s failed, err = %d\n", what, err); | |
639 | } | |
640 | if (disconnect_on_error) | |
641 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
642 | } | |
643 | put_net_conf(mdev); | |
644 | return sock; | |
645 | } | |
646 | ||
647 | static struct socket *drbd_wait_for_connect(struct drbd_conf *mdev) | |
648 | { | |
649 | int timeo, err; | |
650 | struct socket *s_estab = NULL, *s_listen; | |
651 | const char *what; | |
652 | ||
653 | if (!get_net_conf(mdev)) | |
654 | return NULL; | |
655 | ||
656 | what = "sock_create_kern"; | |
657 | err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family, | |
658 | SOCK_STREAM, IPPROTO_TCP, &s_listen); | |
659 | if (err) { | |
660 | s_listen = NULL; | |
661 | goto out; | |
662 | } | |
663 | ||
664 | timeo = mdev->net_conf->try_connect_int * HZ; | |
665 | timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */ | |
666 | ||
667 | s_listen->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
668 | s_listen->sk->sk_rcvtimeo = timeo; | |
669 | s_listen->sk->sk_sndtimeo = timeo; | |
5dbf1673 LE |
670 | drbd_setbufsize(s_listen, mdev->net_conf->sndbuf_size, |
671 | mdev->net_conf->rcvbuf_size); | |
b411b363 PR |
672 | |
673 | what = "bind before listen"; | |
674 | err = s_listen->ops->bind(s_listen, | |
675 | (struct sockaddr *) mdev->net_conf->my_addr, | |
676 | mdev->net_conf->my_addr_len); | |
677 | if (err < 0) | |
678 | goto out; | |
679 | ||
680 | err = drbd_accept(mdev, &what, s_listen, &s_estab); | |
681 | ||
682 | out: | |
683 | if (s_listen) | |
684 | sock_release(s_listen); | |
685 | if (err < 0) { | |
686 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
687 | dev_err(DEV, "%s failed, err = %d\n", what, err); | |
688 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
689 | } | |
690 | } | |
691 | put_net_conf(mdev); | |
692 | ||
693 | return s_estab; | |
694 | } | |
695 | ||
696 | static int drbd_send_fp(struct drbd_conf *mdev, | |
697 | struct socket *sock, enum drbd_packets cmd) | |
698 | { | |
02918be2 | 699 | struct p_header80 *h = &mdev->data.sbuf.header.h80; |
b411b363 PR |
700 | |
701 | return _drbd_send_cmd(mdev, sock, cmd, h, sizeof(*h), 0); | |
702 | } | |
703 | ||
704 | static enum drbd_packets drbd_recv_fp(struct drbd_conf *mdev, struct socket *sock) | |
705 | { | |
02918be2 | 706 | struct p_header80 *h = &mdev->data.rbuf.header.h80; |
b411b363 PR |
707 | int rr; |
708 | ||
709 | rr = drbd_recv_short(mdev, sock, h, sizeof(*h), 0); | |
710 | ||
711 | if (rr == sizeof(*h) && h->magic == BE_DRBD_MAGIC) | |
712 | return be16_to_cpu(h->command); | |
713 | ||
714 | return 0xffff; | |
715 | } | |
716 | ||
717 | /** | |
718 | * drbd_socket_okay() - Free the socket if its connection is not okay | |
719 | * @mdev: DRBD device. | |
720 | * @sock: pointer to the pointer to the socket. | |
721 | */ | |
722 | static int drbd_socket_okay(struct drbd_conf *mdev, struct socket **sock) | |
723 | { | |
724 | int rr; | |
725 | char tb[4]; | |
726 | ||
727 | if (!*sock) | |
81e84650 | 728 | return false; |
b411b363 PR |
729 | |
730 | rr = drbd_recv_short(mdev, *sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); | |
731 | ||
732 | if (rr > 0 || rr == -EAGAIN) { | |
81e84650 | 733 | return true; |
b411b363 PR |
734 | } else { |
735 | sock_release(*sock); | |
736 | *sock = NULL; | |
81e84650 | 737 | return false; |
b411b363 PR |
738 | } |
739 | } | |
740 | ||
741 | /* | |
742 | * return values: | |
743 | * 1 yes, we have a valid connection | |
744 | * 0 oops, did not work out, please try again | |
745 | * -1 peer talks different language, | |
746 | * no point in trying again, please go standalone. | |
747 | * -2 We do not have a network config... | |
748 | */ | |
749 | static int drbd_connect(struct drbd_conf *mdev) | |
750 | { | |
751 | struct socket *s, *sock, *msock; | |
752 | int try, h, ok; | |
753 | ||
754 | D_ASSERT(!mdev->data.socket); | |
755 | ||
b411b363 PR |
756 | if (drbd_request_state(mdev, NS(conn, C_WF_CONNECTION)) < SS_SUCCESS) |
757 | return -2; | |
758 | ||
759 | clear_bit(DISCARD_CONCURRENT, &mdev->flags); | |
760 | ||
761 | sock = NULL; | |
762 | msock = NULL; | |
763 | ||
764 | do { | |
765 | for (try = 0;;) { | |
766 | /* 3 tries, this should take less than a second! */ | |
767 | s = drbd_try_connect(mdev); | |
768 | if (s || ++try >= 3) | |
769 | break; | |
770 | /* give the other side time to call bind() & listen() */ | |
20ee6390 | 771 | schedule_timeout_interruptible(HZ / 10); |
b411b363 PR |
772 | } |
773 | ||
774 | if (s) { | |
775 | if (!sock) { | |
776 | drbd_send_fp(mdev, s, P_HAND_SHAKE_S); | |
777 | sock = s; | |
778 | s = NULL; | |
779 | } else if (!msock) { | |
780 | drbd_send_fp(mdev, s, P_HAND_SHAKE_M); | |
781 | msock = s; | |
782 | s = NULL; | |
783 | } else { | |
784 | dev_err(DEV, "Logic error in drbd_connect()\n"); | |
785 | goto out_release_sockets; | |
786 | } | |
787 | } | |
788 | ||
789 | if (sock && msock) { | |
a8e40792 | 790 | schedule_timeout_interruptible(mdev->net_conf->ping_timeo*HZ/10); |
b411b363 PR |
791 | ok = drbd_socket_okay(mdev, &sock); |
792 | ok = drbd_socket_okay(mdev, &msock) && ok; | |
793 | if (ok) | |
794 | break; | |
795 | } | |
796 | ||
797 | retry: | |
798 | s = drbd_wait_for_connect(mdev); | |
799 | if (s) { | |
800 | try = drbd_recv_fp(mdev, s); | |
801 | drbd_socket_okay(mdev, &sock); | |
802 | drbd_socket_okay(mdev, &msock); | |
803 | switch (try) { | |
804 | case P_HAND_SHAKE_S: | |
805 | if (sock) { | |
806 | dev_warn(DEV, "initial packet S crossed\n"); | |
807 | sock_release(sock); | |
808 | } | |
809 | sock = s; | |
810 | break; | |
811 | case P_HAND_SHAKE_M: | |
812 | if (msock) { | |
813 | dev_warn(DEV, "initial packet M crossed\n"); | |
814 | sock_release(msock); | |
815 | } | |
816 | msock = s; | |
817 | set_bit(DISCARD_CONCURRENT, &mdev->flags); | |
818 | break; | |
819 | default: | |
820 | dev_warn(DEV, "Error receiving initial packet\n"); | |
821 | sock_release(s); | |
822 | if (random32() & 1) | |
823 | goto retry; | |
824 | } | |
825 | } | |
826 | ||
827 | if (mdev->state.conn <= C_DISCONNECTING) | |
828 | goto out_release_sockets; | |
829 | if (signal_pending(current)) { | |
830 | flush_signals(current); | |
831 | smp_rmb(); | |
832 | if (get_t_state(&mdev->receiver) == Exiting) | |
833 | goto out_release_sockets; | |
834 | } | |
835 | ||
836 | if (sock && msock) { | |
837 | ok = drbd_socket_okay(mdev, &sock); | |
838 | ok = drbd_socket_okay(mdev, &msock) && ok; | |
839 | if (ok) | |
840 | break; | |
841 | } | |
842 | } while (1); | |
843 | ||
844 | msock->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
845 | sock->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
846 | ||
847 | sock->sk->sk_allocation = GFP_NOIO; | |
848 | msock->sk->sk_allocation = GFP_NOIO; | |
849 | ||
850 | sock->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; | |
851 | msock->sk->sk_priority = TC_PRIO_INTERACTIVE; | |
852 | ||
b411b363 PR |
853 | /* NOT YET ... |
854 | * sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; | |
855 | * sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; | |
856 | * first set it to the P_HAND_SHAKE timeout, | |
857 | * which we set to 4x the configured ping_timeout. */ | |
858 | sock->sk->sk_sndtimeo = | |
859 | sock->sk->sk_rcvtimeo = mdev->net_conf->ping_timeo*4*HZ/10; | |
860 | ||
861 | msock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; | |
862 | msock->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ; | |
863 | ||
864 | /* we don't want delays. | |
25985edc | 865 | * we use TCP_CORK where appropriate, though */ |
b411b363 PR |
866 | drbd_tcp_nodelay(sock); |
867 | drbd_tcp_nodelay(msock); | |
868 | ||
869 | mdev->data.socket = sock; | |
870 | mdev->meta.socket = msock; | |
871 | mdev->last_received = jiffies; | |
872 | ||
873 | D_ASSERT(mdev->asender.task == NULL); | |
874 | ||
875 | h = drbd_do_handshake(mdev); | |
876 | if (h <= 0) | |
877 | return h; | |
878 | ||
879 | if (mdev->cram_hmac_tfm) { | |
880 | /* drbd_request_state(mdev, NS(conn, WFAuth)); */ | |
b10d96cb JT |
881 | switch (drbd_do_auth(mdev)) { |
882 | case -1: | |
b411b363 PR |
883 | dev_err(DEV, "Authentication of peer failed\n"); |
884 | return -1; | |
b10d96cb JT |
885 | case 0: |
886 | dev_err(DEV, "Authentication of peer failed, trying again.\n"); | |
887 | return 0; | |
b411b363 PR |
888 | } |
889 | } | |
890 | ||
b411b363 PR |
891 | sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; |
892 | sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; | |
893 | ||
894 | atomic_set(&mdev->packet_seq, 0); | |
895 | mdev->peer_seq = 0; | |
896 | ||
148efa16 | 897 | if (drbd_send_protocol(mdev) == -1) |
7e2455c1 | 898 | return -1; |
b411b363 | 899 | drbd_send_sync_param(mdev, &mdev->sync_conf); |
e89b591c | 900 | drbd_send_sizes(mdev, 0, 0); |
b411b363 PR |
901 | drbd_send_uuids(mdev); |
902 | drbd_send_state(mdev); | |
903 | clear_bit(USE_DEGR_WFC_T, &mdev->flags); | |
904 | clear_bit(RESIZE_PENDING, &mdev->flags); | |
1e86ac48 PR |
905 | |
906 | if (drbd_request_state(mdev, NS(conn, C_WF_REPORT_PARAMS)) < SS_SUCCESS) | |
907 | return 0; | |
908 | ||
909 | drbd_thread_start(&mdev->asender); | |
7fde2be9 | 910 | mod_timer(&mdev->request_timer, jiffies + HZ); /* just start it here. */ |
b411b363 PR |
911 | |
912 | return 1; | |
913 | ||
914 | out_release_sockets: | |
915 | if (sock) | |
916 | sock_release(sock); | |
917 | if (msock) | |
918 | sock_release(msock); | |
919 | return -1; | |
920 | } | |
921 | ||
02918be2 | 922 | static int drbd_recv_header(struct drbd_conf *mdev, enum drbd_packets *cmd, unsigned int *packet_size) |
b411b363 | 923 | { |
02918be2 | 924 | union p_header *h = &mdev->data.rbuf.header; |
b411b363 PR |
925 | int r; |
926 | ||
927 | r = drbd_recv(mdev, h, sizeof(*h)); | |
b411b363 | 928 | if (unlikely(r != sizeof(*h))) { |
0ddc5549 LE |
929 | if (!signal_pending(current)) |
930 | dev_warn(DEV, "short read expecting header on sock: r=%d\n", r); | |
81e84650 | 931 | return false; |
02918be2 PR |
932 | } |
933 | ||
934 | if (likely(h->h80.magic == BE_DRBD_MAGIC)) { | |
935 | *cmd = be16_to_cpu(h->h80.command); | |
936 | *packet_size = be16_to_cpu(h->h80.length); | |
937 | } else if (h->h95.magic == BE_DRBD_MAGIC_BIG) { | |
938 | *cmd = be16_to_cpu(h->h95.command); | |
939 | *packet_size = be32_to_cpu(h->h95.length); | |
940 | } else { | |
004352fa LE |
941 | dev_err(DEV, "magic?? on data m: 0x%08x c: %d l: %d\n", |
942 | be32_to_cpu(h->h80.magic), | |
943 | be16_to_cpu(h->h80.command), | |
944 | be16_to_cpu(h->h80.length)); | |
81e84650 | 945 | return false; |
b411b363 PR |
946 | } |
947 | mdev->last_received = jiffies; | |
948 | ||
81e84650 | 949 | return true; |
b411b363 PR |
950 | } |
951 | ||
2451fc3b | 952 | static void drbd_flush(struct drbd_conf *mdev) |
b411b363 PR |
953 | { |
954 | int rv; | |
955 | ||
956 | if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) { | |
fbd9b09a | 957 | rv = blkdev_issue_flush(mdev->ldev->backing_bdev, GFP_KERNEL, |
dd3932ed | 958 | NULL); |
b411b363 | 959 | if (rv) { |
ebd2b0cd | 960 | dev_info(DEV, "local disk flush failed with status %d\n", rv); |
b411b363 PR |
961 | /* would rather check on EOPNOTSUPP, but that is not reliable. |
962 | * don't try again for ANY return value != 0 | |
963 | * if (rv == -EOPNOTSUPP) */ | |
964 | drbd_bump_write_ordering(mdev, WO_drain_io); | |
965 | } | |
966 | put_ldev(mdev); | |
967 | } | |
b411b363 PR |
968 | } |
969 | ||
970 | /** | |
971 | * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. | |
972 | * @mdev: DRBD device. | |
973 | * @epoch: Epoch object. | |
974 | * @ev: Epoch event. | |
975 | */ | |
976 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev, | |
977 | struct drbd_epoch *epoch, | |
978 | enum epoch_event ev) | |
979 | { | |
2451fc3b | 980 | int epoch_size; |
b411b363 | 981 | struct drbd_epoch *next_epoch; |
b411b363 PR |
982 | enum finish_epoch rv = FE_STILL_LIVE; |
983 | ||
984 | spin_lock(&mdev->epoch_lock); | |
985 | do { | |
986 | next_epoch = NULL; | |
b411b363 PR |
987 | |
988 | epoch_size = atomic_read(&epoch->epoch_size); | |
989 | ||
990 | switch (ev & ~EV_CLEANUP) { | |
991 | case EV_PUT: | |
992 | atomic_dec(&epoch->active); | |
993 | break; | |
994 | case EV_GOT_BARRIER_NR: | |
995 | set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); | |
b411b363 PR |
996 | break; |
997 | case EV_BECAME_LAST: | |
998 | /* nothing to do*/ | |
999 | break; | |
1000 | } | |
1001 | ||
b411b363 PR |
1002 | if (epoch_size != 0 && |
1003 | atomic_read(&epoch->active) == 0 && | |
80f9fd55 | 1004 | (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) || ev & EV_CLEANUP)) { |
b411b363 PR |
1005 | if (!(ev & EV_CLEANUP)) { |
1006 | spin_unlock(&mdev->epoch_lock); | |
1007 | drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size); | |
1008 | spin_lock(&mdev->epoch_lock); | |
1009 | } | |
80f9fd55 PR |
1010 | if (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) |
1011 | dec_unacked(mdev); | |
b411b363 PR |
1012 | |
1013 | if (mdev->current_epoch != epoch) { | |
1014 | next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); | |
1015 | list_del(&epoch->list); | |
1016 | ev = EV_BECAME_LAST | (ev & EV_CLEANUP); | |
1017 | mdev->epochs--; | |
b411b363 PR |
1018 | kfree(epoch); |
1019 | ||
1020 | if (rv == FE_STILL_LIVE) | |
1021 | rv = FE_DESTROYED; | |
1022 | } else { | |
1023 | epoch->flags = 0; | |
1024 | atomic_set(&epoch->epoch_size, 0); | |
698f9315 | 1025 | /* atomic_set(&epoch->active, 0); is already zero */ |
b411b363 PR |
1026 | if (rv == FE_STILL_LIVE) |
1027 | rv = FE_RECYCLED; | |
2451fc3b | 1028 | wake_up(&mdev->ee_wait); |
b411b363 PR |
1029 | } |
1030 | } | |
1031 | ||
1032 | if (!next_epoch) | |
1033 | break; | |
1034 | ||
1035 | epoch = next_epoch; | |
1036 | } while (1); | |
1037 | ||
1038 | spin_unlock(&mdev->epoch_lock); | |
1039 | ||
b411b363 PR |
1040 | return rv; |
1041 | } | |
1042 | ||
1043 | /** | |
1044 | * drbd_bump_write_ordering() - Fall back to an other write ordering method | |
1045 | * @mdev: DRBD device. | |
1046 | * @wo: Write ordering method to try. | |
1047 | */ | |
1048 | void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local) | |
1049 | { | |
1050 | enum write_ordering_e pwo; | |
1051 | static char *write_ordering_str[] = { | |
1052 | [WO_none] = "none", | |
1053 | [WO_drain_io] = "drain", | |
1054 | [WO_bdev_flush] = "flush", | |
b411b363 PR |
1055 | }; |
1056 | ||
1057 | pwo = mdev->write_ordering; | |
1058 | wo = min(pwo, wo); | |
b411b363 PR |
1059 | if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush) |
1060 | wo = WO_drain_io; | |
1061 | if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain) | |
1062 | wo = WO_none; | |
1063 | mdev->write_ordering = wo; | |
2451fc3b | 1064 | if (pwo != mdev->write_ordering || wo == WO_bdev_flush) |
b411b363 PR |
1065 | dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]); |
1066 | } | |
1067 | ||
45bb912b LE |
1068 | /** |
1069 | * drbd_submit_ee() | |
1070 | * @mdev: DRBD device. | |
1071 | * @e: epoch entry | |
1072 | * @rw: flag field, see bio->bi_rw | |
10f6d992 LE |
1073 | * |
1074 | * May spread the pages to multiple bios, | |
1075 | * depending on bio_add_page restrictions. | |
1076 | * | |
1077 | * Returns 0 if all bios have been submitted, | |
1078 | * -ENOMEM if we could not allocate enough bios, | |
1079 | * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a | |
1080 | * single page to an empty bio (which should never happen and likely indicates | |
1081 | * that the lower level IO stack is in some way broken). This has been observed | |
1082 | * on certain Xen deployments. | |
45bb912b LE |
1083 | */ |
1084 | /* TODO allocate from our own bio_set. */ | |
1085 | int drbd_submit_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, | |
1086 | const unsigned rw, const int fault_type) | |
1087 | { | |
1088 | struct bio *bios = NULL; | |
1089 | struct bio *bio; | |
1090 | struct page *page = e->pages; | |
1091 | sector_t sector = e->sector; | |
1092 | unsigned ds = e->size; | |
1093 | unsigned n_bios = 0; | |
1094 | unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT; | |
10f6d992 | 1095 | int err = -ENOMEM; |
45bb912b LE |
1096 | |
1097 | /* In most cases, we will only need one bio. But in case the lower | |
1098 | * level restrictions happen to be different at this offset on this | |
1099 | * side than those of the sending peer, we may need to submit the | |
1100 | * request in more than one bio. */ | |
1101 | next_bio: | |
1102 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
1103 | if (!bio) { | |
1104 | dev_err(DEV, "submit_ee: Allocation of a bio failed\n"); | |
1105 | goto fail; | |
1106 | } | |
1107 | /* > e->sector, unless this is the first bio */ | |
1108 | bio->bi_sector = sector; | |
1109 | bio->bi_bdev = mdev->ldev->backing_bdev; | |
45bb912b LE |
1110 | bio->bi_rw = rw; |
1111 | bio->bi_private = e; | |
1112 | bio->bi_end_io = drbd_endio_sec; | |
1113 | ||
1114 | bio->bi_next = bios; | |
1115 | bios = bio; | |
1116 | ++n_bios; | |
1117 | ||
1118 | page_chain_for_each(page) { | |
1119 | unsigned len = min_t(unsigned, ds, PAGE_SIZE); | |
1120 | if (!bio_add_page(bio, page, len, 0)) { | |
10f6d992 LE |
1121 | /* A single page must always be possible! |
1122 | * But in case it fails anyways, | |
1123 | * we deal with it, and complain (below). */ | |
1124 | if (bio->bi_vcnt == 0) { | |
1125 | dev_err(DEV, | |
1126 | "bio_add_page failed for len=%u, " | |
1127 | "bi_vcnt=0 (bi_sector=%llu)\n", | |
1128 | len, (unsigned long long)bio->bi_sector); | |
1129 | err = -ENOSPC; | |
1130 | goto fail; | |
1131 | } | |
45bb912b LE |
1132 | goto next_bio; |
1133 | } | |
1134 | ds -= len; | |
1135 | sector += len >> 9; | |
1136 | --nr_pages; | |
1137 | } | |
1138 | D_ASSERT(page == NULL); | |
1139 | D_ASSERT(ds == 0); | |
1140 | ||
1141 | atomic_set(&e->pending_bios, n_bios); | |
1142 | do { | |
1143 | bio = bios; | |
1144 | bios = bios->bi_next; | |
1145 | bio->bi_next = NULL; | |
1146 | ||
45bb912b | 1147 | drbd_generic_make_request(mdev, fault_type, bio); |
45bb912b | 1148 | } while (bios); |
45bb912b LE |
1149 | return 0; |
1150 | ||
1151 | fail: | |
1152 | while (bios) { | |
1153 | bio = bios; | |
1154 | bios = bios->bi_next; | |
1155 | bio_put(bio); | |
1156 | } | |
10f6d992 | 1157 | return err; |
45bb912b LE |
1158 | } |
1159 | ||
02918be2 | 1160 | static int receive_Barrier(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 1161 | { |
2451fc3b | 1162 | int rv; |
02918be2 | 1163 | struct p_barrier *p = &mdev->data.rbuf.barrier; |
b411b363 PR |
1164 | struct drbd_epoch *epoch; |
1165 | ||
b411b363 PR |
1166 | inc_unacked(mdev); |
1167 | ||
b411b363 PR |
1168 | mdev->current_epoch->barrier_nr = p->barrier; |
1169 | rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR); | |
1170 | ||
1171 | /* P_BARRIER_ACK may imply that the corresponding extent is dropped from | |
1172 | * the activity log, which means it would not be resynced in case the | |
1173 | * R_PRIMARY crashes now. | |
1174 | * Therefore we must send the barrier_ack after the barrier request was | |
1175 | * completed. */ | |
1176 | switch (mdev->write_ordering) { | |
b411b363 PR |
1177 | case WO_none: |
1178 | if (rv == FE_RECYCLED) | |
81e84650 | 1179 | return true; |
2451fc3b PR |
1180 | |
1181 | /* receiver context, in the writeout path of the other node. | |
1182 | * avoid potential distributed deadlock */ | |
1183 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1184 | if (epoch) | |
1185 | break; | |
1186 | else | |
1187 | dev_warn(DEV, "Allocation of an epoch failed, slowing down\n"); | |
1188 | /* Fall through */ | |
b411b363 PR |
1189 | |
1190 | case WO_bdev_flush: | |
1191 | case WO_drain_io: | |
b411b363 | 1192 | drbd_wait_ee_list_empty(mdev, &mdev->active_ee); |
2451fc3b PR |
1193 | drbd_flush(mdev); |
1194 | ||
1195 | if (atomic_read(&mdev->current_epoch->epoch_size)) { | |
1196 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1197 | if (epoch) | |
1198 | break; | |
b411b363 PR |
1199 | } |
1200 | ||
2451fc3b PR |
1201 | epoch = mdev->current_epoch; |
1202 | wait_event(mdev->ee_wait, atomic_read(&epoch->epoch_size) == 0); | |
1203 | ||
1204 | D_ASSERT(atomic_read(&epoch->active) == 0); | |
1205 | D_ASSERT(epoch->flags == 0); | |
b411b363 | 1206 | |
81e84650 | 1207 | return true; |
2451fc3b PR |
1208 | default: |
1209 | dev_err(DEV, "Strangeness in mdev->write_ordering %d\n", mdev->write_ordering); | |
81e84650 | 1210 | return false; |
b411b363 PR |
1211 | } |
1212 | ||
1213 | epoch->flags = 0; | |
1214 | atomic_set(&epoch->epoch_size, 0); | |
1215 | atomic_set(&epoch->active, 0); | |
1216 | ||
1217 | spin_lock(&mdev->epoch_lock); | |
1218 | if (atomic_read(&mdev->current_epoch->epoch_size)) { | |
1219 | list_add(&epoch->list, &mdev->current_epoch->list); | |
1220 | mdev->current_epoch = epoch; | |
1221 | mdev->epochs++; | |
b411b363 PR |
1222 | } else { |
1223 | /* The current_epoch got recycled while we allocated this one... */ | |
1224 | kfree(epoch); | |
1225 | } | |
1226 | spin_unlock(&mdev->epoch_lock); | |
1227 | ||
81e84650 | 1228 | return true; |
b411b363 PR |
1229 | } |
1230 | ||
1231 | /* used from receive_RSDataReply (recv_resync_read) | |
1232 | * and from receive_Data */ | |
1233 | static struct drbd_epoch_entry * | |
1234 | read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, int data_size) __must_hold(local) | |
1235 | { | |
6666032a | 1236 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); |
b411b363 | 1237 | struct drbd_epoch_entry *e; |
b411b363 | 1238 | struct page *page; |
45bb912b | 1239 | int dgs, ds, rr; |
b411b363 PR |
1240 | void *dig_in = mdev->int_dig_in; |
1241 | void *dig_vv = mdev->int_dig_vv; | |
6b4388ac | 1242 | unsigned long *data; |
b411b363 PR |
1243 | |
1244 | dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ? | |
1245 | crypto_hash_digestsize(mdev->integrity_r_tfm) : 0; | |
1246 | ||
1247 | if (dgs) { | |
1248 | rr = drbd_recv(mdev, dig_in, dgs); | |
1249 | if (rr != dgs) { | |
0ddc5549 LE |
1250 | if (!signal_pending(current)) |
1251 | dev_warn(DEV, | |
1252 | "short read receiving data digest: read %d expected %d\n", | |
1253 | rr, dgs); | |
b411b363 PR |
1254 | return NULL; |
1255 | } | |
1256 | } | |
1257 | ||
1258 | data_size -= dgs; | |
1259 | ||
d07c9c10 | 1260 | ERR_IF(data_size == 0) return NULL; |
b411b363 | 1261 | ERR_IF(data_size & 0x1ff) return NULL; |
1816a2b4 | 1262 | ERR_IF(data_size > DRBD_MAX_BIO_SIZE) return NULL; |
b411b363 | 1263 | |
6666032a LE |
1264 | /* even though we trust out peer, |
1265 | * we sometimes have to double check. */ | |
1266 | if (sector + (data_size>>9) > capacity) { | |
fdda6544 LE |
1267 | dev_err(DEV, "request from peer beyond end of local disk: " |
1268 | "capacity: %llus < sector: %llus + size: %u\n", | |
6666032a LE |
1269 | (unsigned long long)capacity, |
1270 | (unsigned long long)sector, data_size); | |
1271 | return NULL; | |
1272 | } | |
1273 | ||
b411b363 PR |
1274 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD |
1275 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
1276 | * which in turn might block on the other node at this very place. */ | |
1277 | e = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO); | |
1278 | if (!e) | |
1279 | return NULL; | |
45bb912b | 1280 | |
b411b363 | 1281 | ds = data_size; |
45bb912b LE |
1282 | page = e->pages; |
1283 | page_chain_for_each(page) { | |
1284 | unsigned len = min_t(int, ds, PAGE_SIZE); | |
6b4388ac | 1285 | data = kmap(page); |
45bb912b | 1286 | rr = drbd_recv(mdev, data, len); |
0cf9d27e | 1287 | if (drbd_insert_fault(mdev, DRBD_FAULT_RECEIVE)) { |
6b4388ac PR |
1288 | dev_err(DEV, "Fault injection: Corrupting data on receive\n"); |
1289 | data[0] = data[0] ^ (unsigned long)-1; | |
1290 | } | |
b411b363 | 1291 | kunmap(page); |
45bb912b | 1292 | if (rr != len) { |
b411b363 | 1293 | drbd_free_ee(mdev, e); |
0ddc5549 LE |
1294 | if (!signal_pending(current)) |
1295 | dev_warn(DEV, "short read receiving data: read %d expected %d\n", | |
1296 | rr, len); | |
b411b363 PR |
1297 | return NULL; |
1298 | } | |
1299 | ds -= rr; | |
1300 | } | |
1301 | ||
1302 | if (dgs) { | |
45bb912b | 1303 | drbd_csum_ee(mdev, mdev->integrity_r_tfm, e, dig_vv); |
b411b363 | 1304 | if (memcmp(dig_in, dig_vv, dgs)) { |
470be44a LE |
1305 | dev_err(DEV, "Digest integrity check FAILED: %llus +%u\n", |
1306 | (unsigned long long)sector, data_size); | |
b411b363 PR |
1307 | drbd_bcast_ee(mdev, "digest failed", |
1308 | dgs, dig_in, dig_vv, e); | |
1309 | drbd_free_ee(mdev, e); | |
1310 | return NULL; | |
1311 | } | |
1312 | } | |
1313 | mdev->recv_cnt += data_size>>9; | |
1314 | return e; | |
1315 | } | |
1316 | ||
1317 | /* drbd_drain_block() just takes a data block | |
1318 | * out of the socket input buffer, and discards it. | |
1319 | */ | |
1320 | static int drbd_drain_block(struct drbd_conf *mdev, int data_size) | |
1321 | { | |
1322 | struct page *page; | |
1323 | int rr, rv = 1; | |
1324 | void *data; | |
1325 | ||
c3470cde | 1326 | if (!data_size) |
81e84650 | 1327 | return true; |
c3470cde | 1328 | |
45bb912b | 1329 | page = drbd_pp_alloc(mdev, 1, 1); |
b411b363 PR |
1330 | |
1331 | data = kmap(page); | |
1332 | while (data_size) { | |
1333 | rr = drbd_recv(mdev, data, min_t(int, data_size, PAGE_SIZE)); | |
1334 | if (rr != min_t(int, data_size, PAGE_SIZE)) { | |
1335 | rv = 0; | |
0ddc5549 LE |
1336 | if (!signal_pending(current)) |
1337 | dev_warn(DEV, | |
1338 | "short read receiving data: read %d expected %d\n", | |
1339 | rr, min_t(int, data_size, PAGE_SIZE)); | |
b411b363 PR |
1340 | break; |
1341 | } | |
1342 | data_size -= rr; | |
1343 | } | |
1344 | kunmap(page); | |
435f0740 | 1345 | drbd_pp_free(mdev, page, 0); |
b411b363 PR |
1346 | return rv; |
1347 | } | |
1348 | ||
1349 | static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req, | |
1350 | sector_t sector, int data_size) | |
1351 | { | |
1352 | struct bio_vec *bvec; | |
1353 | struct bio *bio; | |
1354 | int dgs, rr, i, expect; | |
1355 | void *dig_in = mdev->int_dig_in; | |
1356 | void *dig_vv = mdev->int_dig_vv; | |
1357 | ||
1358 | dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ? | |
1359 | crypto_hash_digestsize(mdev->integrity_r_tfm) : 0; | |
1360 | ||
1361 | if (dgs) { | |
1362 | rr = drbd_recv(mdev, dig_in, dgs); | |
1363 | if (rr != dgs) { | |
0ddc5549 LE |
1364 | if (!signal_pending(current)) |
1365 | dev_warn(DEV, | |
1366 | "short read receiving data reply digest: read %d expected %d\n", | |
1367 | rr, dgs); | |
b411b363 PR |
1368 | return 0; |
1369 | } | |
1370 | } | |
1371 | ||
1372 | data_size -= dgs; | |
1373 | ||
1374 | /* optimistically update recv_cnt. if receiving fails below, | |
1375 | * we disconnect anyways, and counters will be reset. */ | |
1376 | mdev->recv_cnt += data_size>>9; | |
1377 | ||
1378 | bio = req->master_bio; | |
1379 | D_ASSERT(sector == bio->bi_sector); | |
1380 | ||
1381 | bio_for_each_segment(bvec, bio, i) { | |
1382 | expect = min_t(int, data_size, bvec->bv_len); | |
1383 | rr = drbd_recv(mdev, | |
1384 | kmap(bvec->bv_page)+bvec->bv_offset, | |
1385 | expect); | |
1386 | kunmap(bvec->bv_page); | |
1387 | if (rr != expect) { | |
0ddc5549 LE |
1388 | if (!signal_pending(current)) |
1389 | dev_warn(DEV, "short read receiving data reply: " | |
1390 | "read %d expected %d\n", | |
1391 | rr, expect); | |
b411b363 PR |
1392 | return 0; |
1393 | } | |
1394 | data_size -= rr; | |
1395 | } | |
1396 | ||
1397 | if (dgs) { | |
45bb912b | 1398 | drbd_csum_bio(mdev, mdev->integrity_r_tfm, bio, dig_vv); |
b411b363 PR |
1399 | if (memcmp(dig_in, dig_vv, dgs)) { |
1400 | dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n"); | |
1401 | return 0; | |
1402 | } | |
1403 | } | |
1404 | ||
1405 | D_ASSERT(data_size == 0); | |
1406 | return 1; | |
1407 | } | |
1408 | ||
1409 | /* e_end_resync_block() is called via | |
1410 | * drbd_process_done_ee() by asender only */ | |
1411 | static int e_end_resync_block(struct drbd_conf *mdev, struct drbd_work *w, int unused) | |
1412 | { | |
1413 | struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; | |
1414 | sector_t sector = e->sector; | |
1415 | int ok; | |
1416 | ||
24c4830c | 1417 | D_ASSERT(hlist_unhashed(&e->collision)); |
b411b363 | 1418 | |
45bb912b | 1419 | if (likely((e->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1420 | drbd_set_in_sync(mdev, sector, e->size); |
1421 | ok = drbd_send_ack(mdev, P_RS_WRITE_ACK, e); | |
1422 | } else { | |
1423 | /* Record failure to sync */ | |
1424 | drbd_rs_failed_io(mdev, sector, e->size); | |
1425 | ||
1426 | ok = drbd_send_ack(mdev, P_NEG_ACK, e); | |
1427 | } | |
1428 | dec_unacked(mdev); | |
1429 | ||
1430 | return ok; | |
1431 | } | |
1432 | ||
1433 | static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local) | |
1434 | { | |
1435 | struct drbd_epoch_entry *e; | |
1436 | ||
1437 | e = read_in_block(mdev, ID_SYNCER, sector, data_size); | |
45bb912b LE |
1438 | if (!e) |
1439 | goto fail; | |
b411b363 PR |
1440 | |
1441 | dec_rs_pending(mdev); | |
1442 | ||
b411b363 PR |
1443 | inc_unacked(mdev); |
1444 | /* corresponding dec_unacked() in e_end_resync_block() | |
1445 | * respective _drbd_clear_done_ee */ | |
1446 | ||
45bb912b LE |
1447 | e->w.cb = e_end_resync_block; |
1448 | ||
b411b363 PR |
1449 | spin_lock_irq(&mdev->req_lock); |
1450 | list_add(&e->w.list, &mdev->sync_ee); | |
1451 | spin_unlock_irq(&mdev->req_lock); | |
1452 | ||
0f0601f4 | 1453 | atomic_add(data_size >> 9, &mdev->rs_sect_ev); |
45bb912b | 1454 | if (drbd_submit_ee(mdev, e, WRITE, DRBD_FAULT_RS_WR) == 0) |
81e84650 | 1455 | return true; |
b411b363 | 1456 | |
10f6d992 LE |
1457 | /* don't care for the reason here */ |
1458 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
22cc37a9 LE |
1459 | spin_lock_irq(&mdev->req_lock); |
1460 | list_del(&e->w.list); | |
1461 | spin_unlock_irq(&mdev->req_lock); | |
1462 | ||
45bb912b LE |
1463 | drbd_free_ee(mdev, e); |
1464 | fail: | |
1465 | put_ldev(mdev); | |
81e84650 | 1466 | return false; |
b411b363 PR |
1467 | } |
1468 | ||
02918be2 | 1469 | static int receive_DataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
1470 | { |
1471 | struct drbd_request *req; | |
1472 | sector_t sector; | |
b411b363 | 1473 | int ok; |
02918be2 | 1474 | struct p_data *p = &mdev->data.rbuf.data; |
b411b363 PR |
1475 | |
1476 | sector = be64_to_cpu(p->sector); | |
1477 | ||
1478 | spin_lock_irq(&mdev->req_lock); | |
1479 | req = _ar_id_to_req(mdev, p->block_id, sector); | |
1480 | spin_unlock_irq(&mdev->req_lock); | |
1481 | if (unlikely(!req)) { | |
1482 | dev_err(DEV, "Got a corrupt block_id/sector pair(1).\n"); | |
81e84650 | 1483 | return false; |
b411b363 PR |
1484 | } |
1485 | ||
24c4830c | 1486 | /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid |
b411b363 PR |
1487 | * special casing it there for the various failure cases. |
1488 | * still no race with drbd_fail_pending_reads */ | |
1489 | ok = recv_dless_read(mdev, req, sector, data_size); | |
1490 | ||
1491 | if (ok) | |
1492 | req_mod(req, data_received); | |
1493 | /* else: nothing. handled from drbd_disconnect... | |
1494 | * I don't think we may complete this just yet | |
1495 | * in case we are "on-disconnect: freeze" */ | |
1496 | ||
1497 | return ok; | |
1498 | } | |
1499 | ||
02918be2 | 1500 | static int receive_RSDataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
1501 | { |
1502 | sector_t sector; | |
b411b363 | 1503 | int ok; |
02918be2 | 1504 | struct p_data *p = &mdev->data.rbuf.data; |
b411b363 PR |
1505 | |
1506 | sector = be64_to_cpu(p->sector); | |
1507 | D_ASSERT(p->block_id == ID_SYNCER); | |
1508 | ||
1509 | if (get_ldev(mdev)) { | |
1510 | /* data is submitted to disk within recv_resync_read. | |
1511 | * corresponding put_ldev done below on error, | |
1512 | * or in drbd_endio_write_sec. */ | |
1513 | ok = recv_resync_read(mdev, sector, data_size); | |
1514 | } else { | |
1515 | if (__ratelimit(&drbd_ratelimit_state)) | |
1516 | dev_err(DEV, "Can not write resync data to local disk.\n"); | |
1517 | ||
1518 | ok = drbd_drain_block(mdev, data_size); | |
1519 | ||
2b2bf214 | 1520 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size); |
b411b363 PR |
1521 | } |
1522 | ||
778f271d PR |
1523 | atomic_add(data_size >> 9, &mdev->rs_sect_in); |
1524 | ||
b411b363 PR |
1525 | return ok; |
1526 | } | |
1527 | ||
1528 | /* e_end_block() is called via drbd_process_done_ee(). | |
1529 | * this means this function only runs in the asender thread | |
1530 | */ | |
1531 | static int e_end_block(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
1532 | { | |
1533 | struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; | |
1534 | sector_t sector = e->sector; | |
b411b363 PR |
1535 | int ok = 1, pcmd; |
1536 | ||
b411b363 | 1537 | if (mdev->net_conf->wire_protocol == DRBD_PROT_C) { |
45bb912b | 1538 | if (likely((e->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1539 | pcmd = (mdev->state.conn >= C_SYNC_SOURCE && |
1540 | mdev->state.conn <= C_PAUSED_SYNC_T && | |
1541 | e->flags & EE_MAY_SET_IN_SYNC) ? | |
1542 | P_RS_WRITE_ACK : P_WRITE_ACK; | |
1543 | ok &= drbd_send_ack(mdev, pcmd, e); | |
1544 | if (pcmd == P_RS_WRITE_ACK) | |
1545 | drbd_set_in_sync(mdev, sector, e->size); | |
1546 | } else { | |
1547 | ok = drbd_send_ack(mdev, P_NEG_ACK, e); | |
1548 | /* we expect it to be marked out of sync anyways... | |
1549 | * maybe assert this? */ | |
1550 | } | |
1551 | dec_unacked(mdev); | |
1552 | } | |
1553 | /* we delete from the conflict detection hash _after_ we sent out the | |
1554 | * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ | |
1555 | if (mdev->net_conf->two_primaries) { | |
1556 | spin_lock_irq(&mdev->req_lock); | |
24c4830c BVA |
1557 | D_ASSERT(!hlist_unhashed(&e->collision)); |
1558 | hlist_del_init(&e->collision); | |
b411b363 PR |
1559 | spin_unlock_irq(&mdev->req_lock); |
1560 | } else { | |
24c4830c | 1561 | D_ASSERT(hlist_unhashed(&e->collision)); |
b411b363 PR |
1562 | } |
1563 | ||
1564 | drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); | |
1565 | ||
1566 | return ok; | |
1567 | } | |
1568 | ||
1569 | static int e_send_discard_ack(struct drbd_conf *mdev, struct drbd_work *w, int unused) | |
1570 | { | |
1571 | struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; | |
1572 | int ok = 1; | |
1573 | ||
1574 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
1575 | ok = drbd_send_ack(mdev, P_DISCARD_ACK, e); | |
1576 | ||
1577 | spin_lock_irq(&mdev->req_lock); | |
24c4830c BVA |
1578 | D_ASSERT(!hlist_unhashed(&e->collision)); |
1579 | hlist_del_init(&e->collision); | |
b411b363 PR |
1580 | spin_unlock_irq(&mdev->req_lock); |
1581 | ||
1582 | dec_unacked(mdev); | |
1583 | ||
1584 | return ok; | |
1585 | } | |
1586 | ||
1587 | /* Called from receive_Data. | |
1588 | * Synchronize packets on sock with packets on msock. | |
1589 | * | |
1590 | * This is here so even when a P_DATA packet traveling via sock overtook an Ack | |
1591 | * packet traveling on msock, they are still processed in the order they have | |
1592 | * been sent. | |
1593 | * | |
1594 | * Note: we don't care for Ack packets overtaking P_DATA packets. | |
1595 | * | |
1596 | * In case packet_seq is larger than mdev->peer_seq number, there are | |
1597 | * outstanding packets on the msock. We wait for them to arrive. | |
1598 | * In case we are the logically next packet, we update mdev->peer_seq | |
1599 | * ourselves. Correctly handles 32bit wrap around. | |
1600 | * | |
1601 | * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, | |
1602 | * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds | |
1603 | * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have | |
1604 | * 1<<9 == 512 seconds aka ages for the 32bit wrap around... | |
1605 | * | |
1606 | * returns 0 if we may process the packet, | |
1607 | * -ERESTARTSYS if we were interrupted (by disconnect signal). */ | |
1608 | static int drbd_wait_peer_seq(struct drbd_conf *mdev, const u32 packet_seq) | |
1609 | { | |
1610 | DEFINE_WAIT(wait); | |
1611 | unsigned int p_seq; | |
1612 | long timeout; | |
1613 | int ret = 0; | |
1614 | spin_lock(&mdev->peer_seq_lock); | |
1615 | for (;;) { | |
1616 | prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE); | |
1617 | if (seq_le(packet_seq, mdev->peer_seq+1)) | |
1618 | break; | |
1619 | if (signal_pending(current)) { | |
1620 | ret = -ERESTARTSYS; | |
1621 | break; | |
1622 | } | |
1623 | p_seq = mdev->peer_seq; | |
1624 | spin_unlock(&mdev->peer_seq_lock); | |
1625 | timeout = schedule_timeout(30*HZ); | |
1626 | spin_lock(&mdev->peer_seq_lock); | |
1627 | if (timeout == 0 && p_seq == mdev->peer_seq) { | |
1628 | ret = -ETIMEDOUT; | |
1629 | dev_err(DEV, "ASSERT FAILED waited 30 seconds for sequence update, forcing reconnect\n"); | |
1630 | break; | |
1631 | } | |
1632 | } | |
1633 | finish_wait(&mdev->seq_wait, &wait); | |
1634 | if (mdev->peer_seq+1 == packet_seq) | |
1635 | mdev->peer_seq++; | |
1636 | spin_unlock(&mdev->peer_seq_lock); | |
1637 | return ret; | |
1638 | } | |
1639 | ||
688593c5 LE |
1640 | /* see also bio_flags_to_wire() |
1641 | * DRBD_REQ_*, because we need to semantically map the flags to data packet | |
1642 | * flags and back. We may replicate to other kernel versions. */ | |
1643 | static unsigned long wire_flags_to_bio(struct drbd_conf *mdev, u32 dpf) | |
76d2e7ec | 1644 | { |
688593c5 LE |
1645 | return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | |
1646 | (dpf & DP_FUA ? REQ_FUA : 0) | | |
1647 | (dpf & DP_FLUSH ? REQ_FLUSH : 0) | | |
1648 | (dpf & DP_DISCARD ? REQ_DISCARD : 0); | |
76d2e7ec PR |
1649 | } |
1650 | ||
b411b363 | 1651 | /* mirrored write */ |
02918be2 | 1652 | static int receive_Data(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
1653 | { |
1654 | sector_t sector; | |
1655 | struct drbd_epoch_entry *e; | |
02918be2 | 1656 | struct p_data *p = &mdev->data.rbuf.data; |
b411b363 PR |
1657 | int rw = WRITE; |
1658 | u32 dp_flags; | |
1659 | ||
b411b363 | 1660 | if (!get_ldev(mdev)) { |
b411b363 PR |
1661 | spin_lock(&mdev->peer_seq_lock); |
1662 | if (mdev->peer_seq+1 == be32_to_cpu(p->seq_num)) | |
1663 | mdev->peer_seq++; | |
1664 | spin_unlock(&mdev->peer_seq_lock); | |
1665 | ||
2b2bf214 | 1666 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size); |
b411b363 PR |
1667 | atomic_inc(&mdev->current_epoch->epoch_size); |
1668 | return drbd_drain_block(mdev, data_size); | |
1669 | } | |
1670 | ||
1671 | /* get_ldev(mdev) successful. | |
1672 | * Corresponding put_ldev done either below (on various errors), | |
1673 | * or in drbd_endio_write_sec, if we successfully submit the data at | |
1674 | * the end of this function. */ | |
1675 | ||
1676 | sector = be64_to_cpu(p->sector); | |
1677 | e = read_in_block(mdev, p->block_id, sector, data_size); | |
1678 | if (!e) { | |
1679 | put_ldev(mdev); | |
81e84650 | 1680 | return false; |
b411b363 PR |
1681 | } |
1682 | ||
b411b363 PR |
1683 | e->w.cb = e_end_block; |
1684 | ||
688593c5 LE |
1685 | dp_flags = be32_to_cpu(p->dp_flags); |
1686 | rw |= wire_flags_to_bio(mdev, dp_flags); | |
1687 | ||
1688 | if (dp_flags & DP_MAY_SET_IN_SYNC) | |
1689 | e->flags |= EE_MAY_SET_IN_SYNC; | |
1690 | ||
b411b363 PR |
1691 | spin_lock(&mdev->epoch_lock); |
1692 | e->epoch = mdev->current_epoch; | |
1693 | atomic_inc(&e->epoch->epoch_size); | |
1694 | atomic_inc(&e->epoch->active); | |
b411b363 PR |
1695 | spin_unlock(&mdev->epoch_lock); |
1696 | ||
b411b363 PR |
1697 | /* I'm the receiver, I do hold a net_cnt reference. */ |
1698 | if (!mdev->net_conf->two_primaries) { | |
1699 | spin_lock_irq(&mdev->req_lock); | |
1700 | } else { | |
1701 | /* don't get the req_lock yet, | |
1702 | * we may sleep in drbd_wait_peer_seq */ | |
1703 | const int size = e->size; | |
1704 | const int discard = test_bit(DISCARD_CONCURRENT, &mdev->flags); | |
1705 | DEFINE_WAIT(wait); | |
1706 | struct drbd_request *i; | |
1707 | struct hlist_node *n; | |
1708 | struct hlist_head *slot; | |
1709 | int first; | |
1710 | ||
1711 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
1712 | BUG_ON(mdev->ee_hash == NULL); | |
1713 | BUG_ON(mdev->tl_hash == NULL); | |
1714 | ||
1715 | /* conflict detection and handling: | |
1716 | * 1. wait on the sequence number, | |
1717 | * in case this data packet overtook ACK packets. | |
1718 | * 2. check our hash tables for conflicting requests. | |
1719 | * we only need to walk the tl_hash, since an ee can not | |
1720 | * have a conflict with an other ee: on the submitting | |
1721 | * node, the corresponding req had already been conflicting, | |
1722 | * and a conflicting req is never sent. | |
1723 | * | |
1724 | * Note: for two_primaries, we are protocol C, | |
1725 | * so there cannot be any request that is DONE | |
1726 | * but still on the transfer log. | |
1727 | * | |
1728 | * unconditionally add to the ee_hash. | |
1729 | * | |
1730 | * if no conflicting request is found: | |
1731 | * submit. | |
1732 | * | |
1733 | * if any conflicting request is found | |
1734 | * that has not yet been acked, | |
1735 | * AND I have the "discard concurrent writes" flag: | |
1736 | * queue (via done_ee) the P_DISCARD_ACK; OUT. | |
1737 | * | |
1738 | * if any conflicting request is found: | |
1739 | * block the receiver, waiting on misc_wait | |
1740 | * until no more conflicting requests are there, | |
1741 | * or we get interrupted (disconnect). | |
1742 | * | |
1743 | * we do not just write after local io completion of those | |
1744 | * requests, but only after req is done completely, i.e. | |
1745 | * we wait for the P_DISCARD_ACK to arrive! | |
1746 | * | |
1747 | * then proceed normally, i.e. submit. | |
1748 | */ | |
1749 | if (drbd_wait_peer_seq(mdev, be32_to_cpu(p->seq_num))) | |
1750 | goto out_interrupted; | |
1751 | ||
1752 | spin_lock_irq(&mdev->req_lock); | |
1753 | ||
24c4830c | 1754 | hlist_add_head(&e->collision, ee_hash_slot(mdev, sector)); |
b411b363 PR |
1755 | |
1756 | #define OVERLAPS overlaps(i->sector, i->size, sector, size) | |
1757 | slot = tl_hash_slot(mdev, sector); | |
1758 | first = 1; | |
1759 | for (;;) { | |
1760 | int have_unacked = 0; | |
1761 | int have_conflict = 0; | |
1762 | prepare_to_wait(&mdev->misc_wait, &wait, | |
1763 | TASK_INTERRUPTIBLE); | |
24c4830c | 1764 | hlist_for_each_entry(i, n, slot, collision) { |
b411b363 PR |
1765 | if (OVERLAPS) { |
1766 | /* only ALERT on first iteration, | |
1767 | * we may be woken up early... */ | |
1768 | if (first) | |
1769 | dev_alert(DEV, "%s[%u] Concurrent local write detected!" | |
1770 | " new: %llus +%u; pending: %llus +%u\n", | |
1771 | current->comm, current->pid, | |
1772 | (unsigned long long)sector, size, | |
1773 | (unsigned long long)i->sector, i->size); | |
1774 | if (i->rq_state & RQ_NET_PENDING) | |
1775 | ++have_unacked; | |
1776 | ++have_conflict; | |
1777 | } | |
1778 | } | |
1779 | #undef OVERLAPS | |
1780 | if (!have_conflict) | |
1781 | break; | |
1782 | ||
1783 | /* Discard Ack only for the _first_ iteration */ | |
1784 | if (first && discard && have_unacked) { | |
1785 | dev_alert(DEV, "Concurrent write! [DISCARD BY FLAG] sec=%llus\n", | |
1786 | (unsigned long long)sector); | |
1787 | inc_unacked(mdev); | |
1788 | e->w.cb = e_send_discard_ack; | |
1789 | list_add_tail(&e->w.list, &mdev->done_ee); | |
1790 | ||
1791 | spin_unlock_irq(&mdev->req_lock); | |
1792 | ||
1793 | /* we could probably send that P_DISCARD_ACK ourselves, | |
1794 | * but I don't like the receiver using the msock */ | |
1795 | ||
1796 | put_ldev(mdev); | |
1797 | wake_asender(mdev); | |
1798 | finish_wait(&mdev->misc_wait, &wait); | |
81e84650 | 1799 | return true; |
b411b363 PR |
1800 | } |
1801 | ||
1802 | if (signal_pending(current)) { | |
24c4830c | 1803 | hlist_del_init(&e->collision); |
b411b363 PR |
1804 | |
1805 | spin_unlock_irq(&mdev->req_lock); | |
1806 | ||
1807 | finish_wait(&mdev->misc_wait, &wait); | |
1808 | goto out_interrupted; | |
1809 | } | |
1810 | ||
1811 | spin_unlock_irq(&mdev->req_lock); | |
1812 | if (first) { | |
1813 | first = 0; | |
1814 | dev_alert(DEV, "Concurrent write! [W AFTERWARDS] " | |
1815 | "sec=%llus\n", (unsigned long long)sector); | |
1816 | } else if (discard) { | |
1817 | /* we had none on the first iteration. | |
1818 | * there must be none now. */ | |
1819 | D_ASSERT(have_unacked == 0); | |
1820 | } | |
1821 | schedule(); | |
1822 | spin_lock_irq(&mdev->req_lock); | |
1823 | } | |
1824 | finish_wait(&mdev->misc_wait, &wait); | |
1825 | } | |
1826 | ||
1827 | list_add(&e->w.list, &mdev->active_ee); | |
1828 | spin_unlock_irq(&mdev->req_lock); | |
1829 | ||
1830 | switch (mdev->net_conf->wire_protocol) { | |
1831 | case DRBD_PROT_C: | |
1832 | inc_unacked(mdev); | |
1833 | /* corresponding dec_unacked() in e_end_block() | |
1834 | * respective _drbd_clear_done_ee */ | |
1835 | break; | |
1836 | case DRBD_PROT_B: | |
1837 | /* I really don't like it that the receiver thread | |
1838 | * sends on the msock, but anyways */ | |
1839 | drbd_send_ack(mdev, P_RECV_ACK, e); | |
1840 | break; | |
1841 | case DRBD_PROT_A: | |
1842 | /* nothing to do */ | |
1843 | break; | |
1844 | } | |
1845 | ||
6719fb03 | 1846 | if (mdev->state.pdsk < D_INCONSISTENT) { |
b411b363 PR |
1847 | /* In case we have the only disk of the cluster, */ |
1848 | drbd_set_out_of_sync(mdev, e->sector, e->size); | |
1849 | e->flags |= EE_CALL_AL_COMPLETE_IO; | |
6719fb03 | 1850 | e->flags &= ~EE_MAY_SET_IN_SYNC; |
b411b363 PR |
1851 | drbd_al_begin_io(mdev, e->sector); |
1852 | } | |
1853 | ||
45bb912b | 1854 | if (drbd_submit_ee(mdev, e, rw, DRBD_FAULT_DT_WR) == 0) |
81e84650 | 1855 | return true; |
b411b363 | 1856 | |
10f6d992 LE |
1857 | /* don't care for the reason here */ |
1858 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
22cc37a9 LE |
1859 | spin_lock_irq(&mdev->req_lock); |
1860 | list_del(&e->w.list); | |
24c4830c | 1861 | hlist_del_init(&e->collision); |
22cc37a9 LE |
1862 | spin_unlock_irq(&mdev->req_lock); |
1863 | if (e->flags & EE_CALL_AL_COMPLETE_IO) | |
1864 | drbd_al_complete_io(mdev, e->sector); | |
1865 | ||
b411b363 | 1866 | out_interrupted: |
10f6d992 | 1867 | drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + EV_CLEANUP); |
b411b363 PR |
1868 | put_ldev(mdev); |
1869 | drbd_free_ee(mdev, e); | |
81e84650 | 1870 | return false; |
b411b363 PR |
1871 | } |
1872 | ||
0f0601f4 LE |
1873 | /* We may throttle resync, if the lower device seems to be busy, |
1874 | * and current sync rate is above c_min_rate. | |
1875 | * | |
1876 | * To decide whether or not the lower device is busy, we use a scheme similar | |
1877 | * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" | |
1878 | * (more than 64 sectors) of activity we cannot account for with our own resync | |
1879 | * activity, it obviously is "busy". | |
1880 | * | |
1881 | * The current sync rate used here uses only the most recent two step marks, | |
1882 | * to have a short time average so we can react faster. | |
1883 | */ | |
e3555d85 | 1884 | int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector) |
0f0601f4 LE |
1885 | { |
1886 | struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk; | |
1887 | unsigned long db, dt, dbdt; | |
e3555d85 | 1888 | struct lc_element *tmp; |
0f0601f4 LE |
1889 | int curr_events; |
1890 | int throttle = 0; | |
1891 | ||
1892 | /* feature disabled? */ | |
1893 | if (mdev->sync_conf.c_min_rate == 0) | |
1894 | return 0; | |
1895 | ||
e3555d85 PR |
1896 | spin_lock_irq(&mdev->al_lock); |
1897 | tmp = lc_find(mdev->resync, BM_SECT_TO_EXT(sector)); | |
1898 | if (tmp) { | |
1899 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
1900 | if (test_bit(BME_PRIORITY, &bm_ext->flags)) { | |
1901 | spin_unlock_irq(&mdev->al_lock); | |
1902 | return 0; | |
1903 | } | |
1904 | /* Do not slow down if app IO is already waiting for this extent */ | |
1905 | } | |
1906 | spin_unlock_irq(&mdev->al_lock); | |
1907 | ||
0f0601f4 LE |
1908 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
1909 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
1910 | atomic_read(&mdev->rs_sect_ev); | |
e3555d85 | 1911 | |
0f0601f4 LE |
1912 | if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) { |
1913 | unsigned long rs_left; | |
1914 | int i; | |
1915 | ||
1916 | mdev->rs_last_events = curr_events; | |
1917 | ||
1918 | /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, | |
1919 | * approx. */ | |
2649f080 LE |
1920 | i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS; |
1921 | ||
1922 | if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T) | |
1923 | rs_left = mdev->ov_left; | |
1924 | else | |
1925 | rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; | |
0f0601f4 LE |
1926 | |
1927 | dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ; | |
1928 | if (!dt) | |
1929 | dt++; | |
1930 | db = mdev->rs_mark_left[i] - rs_left; | |
1931 | dbdt = Bit2KB(db/dt); | |
1932 | ||
1933 | if (dbdt > mdev->sync_conf.c_min_rate) | |
1934 | throttle = 1; | |
1935 | } | |
1936 | return throttle; | |
1937 | } | |
1938 | ||
1939 | ||
02918be2 | 1940 | static int receive_DataRequest(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int digest_size) |
b411b363 PR |
1941 | { |
1942 | sector_t sector; | |
1943 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); | |
1944 | struct drbd_epoch_entry *e; | |
1945 | struct digest_info *di = NULL; | |
b18b37be | 1946 | int size, verb; |
b411b363 | 1947 | unsigned int fault_type; |
02918be2 | 1948 | struct p_block_req *p = &mdev->data.rbuf.block_req; |
b411b363 PR |
1949 | |
1950 | sector = be64_to_cpu(p->sector); | |
1951 | size = be32_to_cpu(p->blksize); | |
1952 | ||
1816a2b4 | 1953 | if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_BIO_SIZE) { |
b411b363 PR |
1954 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, |
1955 | (unsigned long long)sector, size); | |
81e84650 | 1956 | return false; |
b411b363 PR |
1957 | } |
1958 | if (sector + (size>>9) > capacity) { | |
1959 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, | |
1960 | (unsigned long long)sector, size); | |
81e84650 | 1961 | return false; |
b411b363 PR |
1962 | } |
1963 | ||
1964 | if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) { | |
b18b37be PR |
1965 | verb = 1; |
1966 | switch (cmd) { | |
1967 | case P_DATA_REQUEST: | |
1968 | drbd_send_ack_rp(mdev, P_NEG_DREPLY, p); | |
1969 | break; | |
1970 | case P_RS_DATA_REQUEST: | |
1971 | case P_CSUM_RS_REQUEST: | |
1972 | case P_OV_REQUEST: | |
1973 | drbd_send_ack_rp(mdev, P_NEG_RS_DREPLY , p); | |
1974 | break; | |
1975 | case P_OV_REPLY: | |
1976 | verb = 0; | |
1977 | dec_rs_pending(mdev); | |
1978 | drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, ID_IN_SYNC); | |
1979 | break; | |
1980 | default: | |
1981 | dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", | |
1982 | cmdname(cmd)); | |
1983 | } | |
1984 | if (verb && __ratelimit(&drbd_ratelimit_state)) | |
b411b363 PR |
1985 | dev_err(DEV, "Can not satisfy peer's read request, " |
1986 | "no local data.\n"); | |
b18b37be | 1987 | |
a821cc4a LE |
1988 | /* drain possibly payload */ |
1989 | return drbd_drain_block(mdev, digest_size); | |
b411b363 PR |
1990 | } |
1991 | ||
1992 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD | |
1993 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
1994 | * which in turn might block on the other node at this very place. */ | |
1995 | e = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO); | |
1996 | if (!e) { | |
1997 | put_ldev(mdev); | |
81e84650 | 1998 | return false; |
b411b363 PR |
1999 | } |
2000 | ||
02918be2 | 2001 | switch (cmd) { |
b411b363 PR |
2002 | case P_DATA_REQUEST: |
2003 | e->w.cb = w_e_end_data_req; | |
2004 | fault_type = DRBD_FAULT_DT_RD; | |
80a40e43 LE |
2005 | /* application IO, don't drbd_rs_begin_io */ |
2006 | goto submit; | |
2007 | ||
b411b363 PR |
2008 | case P_RS_DATA_REQUEST: |
2009 | e->w.cb = w_e_end_rsdata_req; | |
2010 | fault_type = DRBD_FAULT_RS_RD; | |
5f9915bb LE |
2011 | /* used in the sector offset progress display */ |
2012 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
b411b363 PR |
2013 | break; |
2014 | ||
2015 | case P_OV_REPLY: | |
2016 | case P_CSUM_RS_REQUEST: | |
2017 | fault_type = DRBD_FAULT_RS_RD; | |
b411b363 PR |
2018 | di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO); |
2019 | if (!di) | |
2020 | goto out_free_e; | |
2021 | ||
2022 | di->digest_size = digest_size; | |
2023 | di->digest = (((char *)di)+sizeof(struct digest_info)); | |
2024 | ||
c36c3ced LE |
2025 | e->digest = di; |
2026 | e->flags |= EE_HAS_DIGEST; | |
2027 | ||
b411b363 PR |
2028 | if (drbd_recv(mdev, di->digest, digest_size) != digest_size) |
2029 | goto out_free_e; | |
2030 | ||
02918be2 | 2031 | if (cmd == P_CSUM_RS_REQUEST) { |
b411b363 PR |
2032 | D_ASSERT(mdev->agreed_pro_version >= 89); |
2033 | e->w.cb = w_e_end_csum_rs_req; | |
5f9915bb LE |
2034 | /* used in the sector offset progress display */ |
2035 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
02918be2 | 2036 | } else if (cmd == P_OV_REPLY) { |
2649f080 LE |
2037 | /* track progress, we may need to throttle */ |
2038 | atomic_add(size >> 9, &mdev->rs_sect_in); | |
b411b363 PR |
2039 | e->w.cb = w_e_end_ov_reply; |
2040 | dec_rs_pending(mdev); | |
0f0601f4 LE |
2041 | /* drbd_rs_begin_io done when we sent this request, |
2042 | * but accounting still needs to be done. */ | |
2043 | goto submit_for_resync; | |
b411b363 PR |
2044 | } |
2045 | break; | |
2046 | ||
2047 | case P_OV_REQUEST: | |
b411b363 PR |
2048 | if (mdev->ov_start_sector == ~(sector_t)0 && |
2049 | mdev->agreed_pro_version >= 90) { | |
de228bba LE |
2050 | unsigned long now = jiffies; |
2051 | int i; | |
b411b363 PR |
2052 | mdev->ov_start_sector = sector; |
2053 | mdev->ov_position = sector; | |
30b743a2 LE |
2054 | mdev->ov_left = drbd_bm_bits(mdev) - BM_SECT_TO_BIT(sector); |
2055 | mdev->rs_total = mdev->ov_left; | |
de228bba LE |
2056 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
2057 | mdev->rs_mark_left[i] = mdev->ov_left; | |
2058 | mdev->rs_mark_time[i] = now; | |
2059 | } | |
b411b363 PR |
2060 | dev_info(DEV, "Online Verify start sector: %llu\n", |
2061 | (unsigned long long)sector); | |
2062 | } | |
2063 | e->w.cb = w_e_end_ov_req; | |
2064 | fault_type = DRBD_FAULT_RS_RD; | |
b411b363 PR |
2065 | break; |
2066 | ||
b411b363 PR |
2067 | default: |
2068 | dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", | |
02918be2 | 2069 | cmdname(cmd)); |
b411b363 | 2070 | fault_type = DRBD_FAULT_MAX; |
80a40e43 | 2071 | goto out_free_e; |
b411b363 PR |
2072 | } |
2073 | ||
0f0601f4 LE |
2074 | /* Throttle, drbd_rs_begin_io and submit should become asynchronous |
2075 | * wrt the receiver, but it is not as straightforward as it may seem. | |
2076 | * Various places in the resync start and stop logic assume resync | |
2077 | * requests are processed in order, requeuing this on the worker thread | |
2078 | * introduces a bunch of new code for synchronization between threads. | |
2079 | * | |
2080 | * Unlimited throttling before drbd_rs_begin_io may stall the resync | |
2081 | * "forever", throttling after drbd_rs_begin_io will lock that extent | |
2082 | * for application writes for the same time. For now, just throttle | |
2083 | * here, where the rest of the code expects the receiver to sleep for | |
2084 | * a while, anyways. | |
2085 | */ | |
2086 | ||
2087 | /* Throttle before drbd_rs_begin_io, as that locks out application IO; | |
2088 | * this defers syncer requests for some time, before letting at least | |
2089 | * on request through. The resync controller on the receiving side | |
2090 | * will adapt to the incoming rate accordingly. | |
2091 | * | |
2092 | * We cannot throttle here if remote is Primary/SyncTarget: | |
2093 | * we would also throttle its application reads. | |
2094 | * In that case, throttling is done on the SyncTarget only. | |
2095 | */ | |
e3555d85 PR |
2096 | if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev, sector)) |
2097 | schedule_timeout_uninterruptible(HZ/10); | |
2098 | if (drbd_rs_begin_io(mdev, sector)) | |
80a40e43 | 2099 | goto out_free_e; |
b411b363 | 2100 | |
0f0601f4 LE |
2101 | submit_for_resync: |
2102 | atomic_add(size >> 9, &mdev->rs_sect_ev); | |
2103 | ||
80a40e43 | 2104 | submit: |
b411b363 | 2105 | inc_unacked(mdev); |
80a40e43 LE |
2106 | spin_lock_irq(&mdev->req_lock); |
2107 | list_add_tail(&e->w.list, &mdev->read_ee); | |
2108 | spin_unlock_irq(&mdev->req_lock); | |
b411b363 | 2109 | |
45bb912b | 2110 | if (drbd_submit_ee(mdev, e, READ, fault_type) == 0) |
81e84650 | 2111 | return true; |
b411b363 | 2112 | |
10f6d992 LE |
2113 | /* don't care for the reason here */ |
2114 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
22cc37a9 LE |
2115 | spin_lock_irq(&mdev->req_lock); |
2116 | list_del(&e->w.list); | |
2117 | spin_unlock_irq(&mdev->req_lock); | |
2118 | /* no drbd_rs_complete_io(), we are dropping the connection anyways */ | |
2119 | ||
b411b363 | 2120 | out_free_e: |
b411b363 PR |
2121 | put_ldev(mdev); |
2122 | drbd_free_ee(mdev, e); | |
81e84650 | 2123 | return false; |
b411b363 PR |
2124 | } |
2125 | ||
2126 | static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local) | |
2127 | { | |
2128 | int self, peer, rv = -100; | |
2129 | unsigned long ch_self, ch_peer; | |
2130 | ||
2131 | self = mdev->ldev->md.uuid[UI_BITMAP] & 1; | |
2132 | peer = mdev->p_uuid[UI_BITMAP] & 1; | |
2133 | ||
2134 | ch_peer = mdev->p_uuid[UI_SIZE]; | |
2135 | ch_self = mdev->comm_bm_set; | |
2136 | ||
2137 | switch (mdev->net_conf->after_sb_0p) { | |
2138 | case ASB_CONSENSUS: | |
2139 | case ASB_DISCARD_SECONDARY: | |
2140 | case ASB_CALL_HELPER: | |
2141 | dev_err(DEV, "Configuration error.\n"); | |
2142 | break; | |
2143 | case ASB_DISCONNECT: | |
2144 | break; | |
2145 | case ASB_DISCARD_YOUNGER_PRI: | |
2146 | if (self == 0 && peer == 1) { | |
2147 | rv = -1; | |
2148 | break; | |
2149 | } | |
2150 | if (self == 1 && peer == 0) { | |
2151 | rv = 1; | |
2152 | break; | |
2153 | } | |
2154 | /* Else fall through to one of the other strategies... */ | |
2155 | case ASB_DISCARD_OLDER_PRI: | |
2156 | if (self == 0 && peer == 1) { | |
2157 | rv = 1; | |
2158 | break; | |
2159 | } | |
2160 | if (self == 1 && peer == 0) { | |
2161 | rv = -1; | |
2162 | break; | |
2163 | } | |
2164 | /* Else fall through to one of the other strategies... */ | |
ad19bf6e | 2165 | dev_warn(DEV, "Discard younger/older primary did not find a decision\n" |
b411b363 PR |
2166 | "Using discard-least-changes instead\n"); |
2167 | case ASB_DISCARD_ZERO_CHG: | |
2168 | if (ch_peer == 0 && ch_self == 0) { | |
2169 | rv = test_bit(DISCARD_CONCURRENT, &mdev->flags) | |
2170 | ? -1 : 1; | |
2171 | break; | |
2172 | } else { | |
2173 | if (ch_peer == 0) { rv = 1; break; } | |
2174 | if (ch_self == 0) { rv = -1; break; } | |
2175 | } | |
2176 | if (mdev->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG) | |
2177 | break; | |
2178 | case ASB_DISCARD_LEAST_CHG: | |
2179 | if (ch_self < ch_peer) | |
2180 | rv = -1; | |
2181 | else if (ch_self > ch_peer) | |
2182 | rv = 1; | |
2183 | else /* ( ch_self == ch_peer ) */ | |
2184 | /* Well, then use something else. */ | |
2185 | rv = test_bit(DISCARD_CONCURRENT, &mdev->flags) | |
2186 | ? -1 : 1; | |
2187 | break; | |
2188 | case ASB_DISCARD_LOCAL: | |
2189 | rv = -1; | |
2190 | break; | |
2191 | case ASB_DISCARD_REMOTE: | |
2192 | rv = 1; | |
2193 | } | |
2194 | ||
2195 | return rv; | |
2196 | } | |
2197 | ||
2198 | static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local) | |
2199 | { | |
6184ea21 | 2200 | int hg, rv = -100; |
b411b363 PR |
2201 | |
2202 | switch (mdev->net_conf->after_sb_1p) { | |
2203 | case ASB_DISCARD_YOUNGER_PRI: | |
2204 | case ASB_DISCARD_OLDER_PRI: | |
2205 | case ASB_DISCARD_LEAST_CHG: | |
2206 | case ASB_DISCARD_LOCAL: | |
2207 | case ASB_DISCARD_REMOTE: | |
2208 | dev_err(DEV, "Configuration error.\n"); | |
2209 | break; | |
2210 | case ASB_DISCONNECT: | |
2211 | break; | |
2212 | case ASB_CONSENSUS: | |
2213 | hg = drbd_asb_recover_0p(mdev); | |
2214 | if (hg == -1 && mdev->state.role == R_SECONDARY) | |
2215 | rv = hg; | |
2216 | if (hg == 1 && mdev->state.role == R_PRIMARY) | |
2217 | rv = hg; | |
2218 | break; | |
2219 | case ASB_VIOLENTLY: | |
2220 | rv = drbd_asb_recover_0p(mdev); | |
2221 | break; | |
2222 | case ASB_DISCARD_SECONDARY: | |
2223 | return mdev->state.role == R_PRIMARY ? 1 : -1; | |
2224 | case ASB_CALL_HELPER: | |
2225 | hg = drbd_asb_recover_0p(mdev); | |
2226 | if (hg == -1 && mdev->state.role == R_PRIMARY) { | |
bb437946 AG |
2227 | enum drbd_state_rv rv2; |
2228 | ||
2229 | drbd_set_role(mdev, R_SECONDARY, 0); | |
b411b363 PR |
2230 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2231 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2232 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2233 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2234 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2235 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2236 | } else { | |
2237 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2238 | rv = hg; | |
2239 | } | |
2240 | } else | |
2241 | rv = hg; | |
2242 | } | |
2243 | ||
2244 | return rv; | |
2245 | } | |
2246 | ||
2247 | static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local) | |
2248 | { | |
6184ea21 | 2249 | int hg, rv = -100; |
b411b363 PR |
2250 | |
2251 | switch (mdev->net_conf->after_sb_2p) { | |
2252 | case ASB_DISCARD_YOUNGER_PRI: | |
2253 | case ASB_DISCARD_OLDER_PRI: | |
2254 | case ASB_DISCARD_LEAST_CHG: | |
2255 | case ASB_DISCARD_LOCAL: | |
2256 | case ASB_DISCARD_REMOTE: | |
2257 | case ASB_CONSENSUS: | |
2258 | case ASB_DISCARD_SECONDARY: | |
2259 | dev_err(DEV, "Configuration error.\n"); | |
2260 | break; | |
2261 | case ASB_VIOLENTLY: | |
2262 | rv = drbd_asb_recover_0p(mdev); | |
2263 | break; | |
2264 | case ASB_DISCONNECT: | |
2265 | break; | |
2266 | case ASB_CALL_HELPER: | |
2267 | hg = drbd_asb_recover_0p(mdev); | |
2268 | if (hg == -1) { | |
bb437946 AG |
2269 | enum drbd_state_rv rv2; |
2270 | ||
b411b363 PR |
2271 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2272 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2273 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2274 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2275 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2276 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2277 | } else { | |
2278 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2279 | rv = hg; | |
2280 | } | |
2281 | } else | |
2282 | rv = hg; | |
2283 | } | |
2284 | ||
2285 | return rv; | |
2286 | } | |
2287 | ||
2288 | static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid, | |
2289 | u64 bits, u64 flags) | |
2290 | { | |
2291 | if (!uuid) { | |
2292 | dev_info(DEV, "%s uuid info vanished while I was looking!\n", text); | |
2293 | return; | |
2294 | } | |
2295 | dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", | |
2296 | text, | |
2297 | (unsigned long long)uuid[UI_CURRENT], | |
2298 | (unsigned long long)uuid[UI_BITMAP], | |
2299 | (unsigned long long)uuid[UI_HISTORY_START], | |
2300 | (unsigned long long)uuid[UI_HISTORY_END], | |
2301 | (unsigned long long)bits, | |
2302 | (unsigned long long)flags); | |
2303 | } | |
2304 | ||
2305 | /* | |
2306 | 100 after split brain try auto recover | |
2307 | 2 C_SYNC_SOURCE set BitMap | |
2308 | 1 C_SYNC_SOURCE use BitMap | |
2309 | 0 no Sync | |
2310 | -1 C_SYNC_TARGET use BitMap | |
2311 | -2 C_SYNC_TARGET set BitMap | |
2312 | -100 after split brain, disconnect | |
2313 | -1000 unrelated data | |
4a23f264 PR |
2314 | -1091 requires proto 91 |
2315 | -1096 requires proto 96 | |
b411b363 PR |
2316 | */ |
2317 | static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local) | |
2318 | { | |
2319 | u64 self, peer; | |
2320 | int i, j; | |
2321 | ||
2322 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2323 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2324 | ||
2325 | *rule_nr = 10; | |
2326 | if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) | |
2327 | return 0; | |
2328 | ||
2329 | *rule_nr = 20; | |
2330 | if ((self == UUID_JUST_CREATED || self == (u64)0) && | |
2331 | peer != UUID_JUST_CREATED) | |
2332 | return -2; | |
2333 | ||
2334 | *rule_nr = 30; | |
2335 | if (self != UUID_JUST_CREATED && | |
2336 | (peer == UUID_JUST_CREATED || peer == (u64)0)) | |
2337 | return 2; | |
2338 | ||
2339 | if (self == peer) { | |
2340 | int rct, dc; /* roles at crash time */ | |
2341 | ||
2342 | if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) { | |
2343 | ||
2344 | if (mdev->agreed_pro_version < 91) | |
4a23f264 | 2345 | return -1091; |
b411b363 PR |
2346 | |
2347 | if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) && | |
2348 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { | |
2349 | dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n"); | |
2350 | drbd_uuid_set_bm(mdev, 0UL); | |
2351 | ||
2352 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, | |
2353 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2354 | *rule_nr = 34; | |
2355 | } else { | |
2356 | dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n"); | |
2357 | *rule_nr = 36; | |
2358 | } | |
2359 | ||
2360 | return 1; | |
2361 | } | |
2362 | ||
2363 | if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) { | |
2364 | ||
2365 | if (mdev->agreed_pro_version < 91) | |
4a23f264 | 2366 | return -1091; |
b411b363 PR |
2367 | |
2368 | if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) && | |
2369 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) { | |
2370 | dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); | |
2371 | ||
2372 | mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START]; | |
2373 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP]; | |
2374 | mdev->p_uuid[UI_BITMAP] = 0UL; | |
2375 | ||
2376 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2377 | *rule_nr = 35; | |
2378 | } else { | |
2379 | dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n"); | |
2380 | *rule_nr = 37; | |
2381 | } | |
2382 | ||
2383 | return -1; | |
2384 | } | |
2385 | ||
2386 | /* Common power [off|failure] */ | |
2387 | rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) + | |
2388 | (mdev->p_uuid[UI_FLAGS] & 2); | |
2389 | /* lowest bit is set when we were primary, | |
2390 | * next bit (weight 2) is set when peer was primary */ | |
2391 | *rule_nr = 40; | |
2392 | ||
2393 | switch (rct) { | |
2394 | case 0: /* !self_pri && !peer_pri */ return 0; | |
2395 | case 1: /* self_pri && !peer_pri */ return 1; | |
2396 | case 2: /* !self_pri && peer_pri */ return -1; | |
2397 | case 3: /* self_pri && peer_pri */ | |
2398 | dc = test_bit(DISCARD_CONCURRENT, &mdev->flags); | |
2399 | return dc ? -1 : 1; | |
2400 | } | |
2401 | } | |
2402 | ||
2403 | *rule_nr = 50; | |
2404 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2405 | if (self == peer) | |
2406 | return -1; | |
2407 | ||
2408 | *rule_nr = 51; | |
2409 | peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); | |
2410 | if (self == peer) { | |
4a23f264 PR |
2411 | if (mdev->agreed_pro_version < 96 ? |
2412 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == | |
2413 | (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) : | |
2414 | peer + UUID_NEW_BM_OFFSET == (mdev->p_uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2415 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2416 | resync as sync source modifications of the peer's UUIDs. */ | |
2417 | ||
2418 | if (mdev->agreed_pro_version < 91) | |
4a23f264 | 2419 | return -1091; |
b411b363 PR |
2420 | |
2421 | mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START]; | |
2422 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1]; | |
4a23f264 PR |
2423 | |
2424 | dev_info(DEV, "Did not got last syncUUID packet, corrected:\n"); | |
2425 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2426 | ||
b411b363 PR |
2427 | return -1; |
2428 | } | |
2429 | } | |
2430 | ||
2431 | *rule_nr = 60; | |
2432 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2433 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2434 | peer = mdev->p_uuid[i] & ~((u64)1); | |
2435 | if (self == peer) | |
2436 | return -2; | |
2437 | } | |
2438 | ||
2439 | *rule_nr = 70; | |
2440 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2441 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2442 | if (self == peer) | |
2443 | return 1; | |
2444 | ||
2445 | *rule_nr = 71; | |
2446 | self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); | |
2447 | if (self == peer) { | |
4a23f264 PR |
2448 | if (mdev->agreed_pro_version < 96 ? |
2449 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == | |
2450 | (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) : | |
2451 | self + UUID_NEW_BM_OFFSET == (mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2452 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2453 | resync as sync source modifications of our UUIDs. */ | |
2454 | ||
2455 | if (mdev->agreed_pro_version < 91) | |
4a23f264 | 2456 | return -1091; |
b411b363 PR |
2457 | |
2458 | _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]); | |
2459 | _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]); | |
2460 | ||
4a23f264 | 2461 | dev_info(DEV, "Last syncUUID did not get through, corrected:\n"); |
b411b363 PR |
2462 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, |
2463 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2464 | ||
2465 | return 1; | |
2466 | } | |
2467 | } | |
2468 | ||
2469 | ||
2470 | *rule_nr = 80; | |
d8c2a36b | 2471 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); |
b411b363 PR |
2472 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
2473 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2474 | if (self == peer) | |
2475 | return 2; | |
2476 | } | |
2477 | ||
2478 | *rule_nr = 90; | |
2479 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2480 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2481 | if (self == peer && self != ((u64)0)) | |
2482 | return 100; | |
2483 | ||
2484 | *rule_nr = 100; | |
2485 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2486 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2487 | for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { | |
2488 | peer = mdev->p_uuid[j] & ~((u64)1); | |
2489 | if (self == peer) | |
2490 | return -100; | |
2491 | } | |
2492 | } | |
2493 | ||
2494 | return -1000; | |
2495 | } | |
2496 | ||
2497 | /* drbd_sync_handshake() returns the new conn state on success, or | |
2498 | CONN_MASK (-1) on failure. | |
2499 | */ | |
2500 | static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role, | |
2501 | enum drbd_disk_state peer_disk) __must_hold(local) | |
2502 | { | |
2503 | int hg, rule_nr; | |
2504 | enum drbd_conns rv = C_MASK; | |
2505 | enum drbd_disk_state mydisk; | |
2506 | ||
2507 | mydisk = mdev->state.disk; | |
2508 | if (mydisk == D_NEGOTIATING) | |
2509 | mydisk = mdev->new_state_tmp.disk; | |
2510 | ||
2511 | dev_info(DEV, "drbd_sync_handshake:\n"); | |
2512 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0); | |
2513 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, | |
2514 | mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2515 | ||
2516 | hg = drbd_uuid_compare(mdev, &rule_nr); | |
2517 | ||
2518 | dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr); | |
2519 | ||
2520 | if (hg == -1000) { | |
2521 | dev_alert(DEV, "Unrelated data, aborting!\n"); | |
2522 | return C_MASK; | |
2523 | } | |
4a23f264 PR |
2524 | if (hg < -1000) { |
2525 | dev_alert(DEV, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000); | |
b411b363 PR |
2526 | return C_MASK; |
2527 | } | |
2528 | ||
2529 | if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || | |
2530 | (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { | |
2531 | int f = (hg == -100) || abs(hg) == 2; | |
2532 | hg = mydisk > D_INCONSISTENT ? 1 : -1; | |
2533 | if (f) | |
2534 | hg = hg*2; | |
2535 | dev_info(DEV, "Becoming sync %s due to disk states.\n", | |
2536 | hg > 0 ? "source" : "target"); | |
2537 | } | |
2538 | ||
3a11a487 AG |
2539 | if (abs(hg) == 100) |
2540 | drbd_khelper(mdev, "initial-split-brain"); | |
2541 | ||
b411b363 PR |
2542 | if (hg == 100 || (hg == -100 && mdev->net_conf->always_asbp)) { |
2543 | int pcount = (mdev->state.role == R_PRIMARY) | |
2544 | + (peer_role == R_PRIMARY); | |
2545 | int forced = (hg == -100); | |
2546 | ||
2547 | switch (pcount) { | |
2548 | case 0: | |
2549 | hg = drbd_asb_recover_0p(mdev); | |
2550 | break; | |
2551 | case 1: | |
2552 | hg = drbd_asb_recover_1p(mdev); | |
2553 | break; | |
2554 | case 2: | |
2555 | hg = drbd_asb_recover_2p(mdev); | |
2556 | break; | |
2557 | } | |
2558 | if (abs(hg) < 100) { | |
2559 | dev_warn(DEV, "Split-Brain detected, %d primaries, " | |
2560 | "automatically solved. Sync from %s node\n", | |
2561 | pcount, (hg < 0) ? "peer" : "this"); | |
2562 | if (forced) { | |
2563 | dev_warn(DEV, "Doing a full sync, since" | |
2564 | " UUIDs where ambiguous.\n"); | |
2565 | hg = hg*2; | |
2566 | } | |
2567 | } | |
2568 | } | |
2569 | ||
2570 | if (hg == -100) { | |
2571 | if (mdev->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1)) | |
2572 | hg = -1; | |
2573 | if (!mdev->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1)) | |
2574 | hg = 1; | |
2575 | ||
2576 | if (abs(hg) < 100) | |
2577 | dev_warn(DEV, "Split-Brain detected, manually solved. " | |
2578 | "Sync from %s node\n", | |
2579 | (hg < 0) ? "peer" : "this"); | |
2580 | } | |
2581 | ||
2582 | if (hg == -100) { | |
580b9767 LE |
2583 | /* FIXME this log message is not correct if we end up here |
2584 | * after an attempted attach on a diskless node. | |
2585 | * We just refuse to attach -- well, we drop the "connection" | |
2586 | * to that disk, in a way... */ | |
3a11a487 | 2587 | dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n"); |
b411b363 PR |
2588 | drbd_khelper(mdev, "split-brain"); |
2589 | return C_MASK; | |
2590 | } | |
2591 | ||
2592 | if (hg > 0 && mydisk <= D_INCONSISTENT) { | |
2593 | dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n"); | |
2594 | return C_MASK; | |
2595 | } | |
2596 | ||
2597 | if (hg < 0 && /* by intention we do not use mydisk here. */ | |
2598 | mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) { | |
2599 | switch (mdev->net_conf->rr_conflict) { | |
2600 | case ASB_CALL_HELPER: | |
2601 | drbd_khelper(mdev, "pri-lost"); | |
2602 | /* fall through */ | |
2603 | case ASB_DISCONNECT: | |
2604 | dev_err(DEV, "I shall become SyncTarget, but I am primary!\n"); | |
2605 | return C_MASK; | |
2606 | case ASB_VIOLENTLY: | |
2607 | dev_warn(DEV, "Becoming SyncTarget, violating the stable-data" | |
2608 | "assumption\n"); | |
2609 | } | |
2610 | } | |
2611 | ||
cf14c2e9 PR |
2612 | if (mdev->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->flags)) { |
2613 | if (hg == 0) | |
2614 | dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n"); | |
2615 | else | |
2616 | dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.", | |
2617 | drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), | |
2618 | abs(hg) >= 2 ? "full" : "bit-map based"); | |
2619 | return C_MASK; | |
2620 | } | |
2621 | ||
b411b363 PR |
2622 | if (abs(hg) >= 2) { |
2623 | dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); | |
20ceb2b2 LE |
2624 | if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake", |
2625 | BM_LOCKED_SET_ALLOWED)) | |
b411b363 PR |
2626 | return C_MASK; |
2627 | } | |
2628 | ||
2629 | if (hg > 0) { /* become sync source. */ | |
2630 | rv = C_WF_BITMAP_S; | |
2631 | } else if (hg < 0) { /* become sync target */ | |
2632 | rv = C_WF_BITMAP_T; | |
2633 | } else { | |
2634 | rv = C_CONNECTED; | |
2635 | if (drbd_bm_total_weight(mdev)) { | |
2636 | dev_info(DEV, "No resync, but %lu bits in bitmap!\n", | |
2637 | drbd_bm_total_weight(mdev)); | |
2638 | } | |
2639 | } | |
2640 | ||
2641 | return rv; | |
2642 | } | |
2643 | ||
2644 | /* returns 1 if invalid */ | |
2645 | static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self) | |
2646 | { | |
2647 | /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ | |
2648 | if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) || | |
2649 | (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL)) | |
2650 | return 0; | |
2651 | ||
2652 | /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ | |
2653 | if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL || | |
2654 | self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL) | |
2655 | return 1; | |
2656 | ||
2657 | /* everything else is valid if they are equal on both sides. */ | |
2658 | if (peer == self) | |
2659 | return 0; | |
2660 | ||
2661 | /* everything es is invalid. */ | |
2662 | return 1; | |
2663 | } | |
2664 | ||
02918be2 | 2665 | static int receive_protocol(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 2666 | { |
02918be2 | 2667 | struct p_protocol *p = &mdev->data.rbuf.protocol; |
b411b363 | 2668 | int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; |
cf14c2e9 | 2669 | int p_want_lose, p_two_primaries, cf; |
b411b363 PR |
2670 | char p_integrity_alg[SHARED_SECRET_MAX] = ""; |
2671 | ||
b411b363 PR |
2672 | p_proto = be32_to_cpu(p->protocol); |
2673 | p_after_sb_0p = be32_to_cpu(p->after_sb_0p); | |
2674 | p_after_sb_1p = be32_to_cpu(p->after_sb_1p); | |
2675 | p_after_sb_2p = be32_to_cpu(p->after_sb_2p); | |
b411b363 | 2676 | p_two_primaries = be32_to_cpu(p->two_primaries); |
cf14c2e9 PR |
2677 | cf = be32_to_cpu(p->conn_flags); |
2678 | p_want_lose = cf & CF_WANT_LOSE; | |
2679 | ||
2680 | clear_bit(CONN_DRY_RUN, &mdev->flags); | |
2681 | ||
2682 | if (cf & CF_DRY_RUN) | |
2683 | set_bit(CONN_DRY_RUN, &mdev->flags); | |
b411b363 PR |
2684 | |
2685 | if (p_proto != mdev->net_conf->wire_protocol) { | |
2686 | dev_err(DEV, "incompatible communication protocols\n"); | |
2687 | goto disconnect; | |
2688 | } | |
2689 | ||
2690 | if (cmp_after_sb(p_after_sb_0p, mdev->net_conf->after_sb_0p)) { | |
2691 | dev_err(DEV, "incompatible after-sb-0pri settings\n"); | |
2692 | goto disconnect; | |
2693 | } | |
2694 | ||
2695 | if (cmp_after_sb(p_after_sb_1p, mdev->net_conf->after_sb_1p)) { | |
2696 | dev_err(DEV, "incompatible after-sb-1pri settings\n"); | |
2697 | goto disconnect; | |
2698 | } | |
2699 | ||
2700 | if (cmp_after_sb(p_after_sb_2p, mdev->net_conf->after_sb_2p)) { | |
2701 | dev_err(DEV, "incompatible after-sb-2pri settings\n"); | |
2702 | goto disconnect; | |
2703 | } | |
2704 | ||
2705 | if (p_want_lose && mdev->net_conf->want_lose) { | |
2706 | dev_err(DEV, "both sides have the 'want_lose' flag set\n"); | |
2707 | goto disconnect; | |
2708 | } | |
2709 | ||
2710 | if (p_two_primaries != mdev->net_conf->two_primaries) { | |
2711 | dev_err(DEV, "incompatible setting of the two-primaries options\n"); | |
2712 | goto disconnect; | |
2713 | } | |
2714 | ||
2715 | if (mdev->agreed_pro_version >= 87) { | |
2716 | unsigned char *my_alg = mdev->net_conf->integrity_alg; | |
2717 | ||
2718 | if (drbd_recv(mdev, p_integrity_alg, data_size) != data_size) | |
81e84650 | 2719 | return false; |
b411b363 PR |
2720 | |
2721 | p_integrity_alg[SHARED_SECRET_MAX-1] = 0; | |
2722 | if (strcmp(p_integrity_alg, my_alg)) { | |
2723 | dev_err(DEV, "incompatible setting of the data-integrity-alg\n"); | |
2724 | goto disconnect; | |
2725 | } | |
2726 | dev_info(DEV, "data-integrity-alg: %s\n", | |
2727 | my_alg[0] ? my_alg : (unsigned char *)"<not-used>"); | |
2728 | } | |
2729 | ||
81e84650 | 2730 | return true; |
b411b363 PR |
2731 | |
2732 | disconnect: | |
2733 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 2734 | return false; |
b411b363 PR |
2735 | } |
2736 | ||
2737 | /* helper function | |
2738 | * input: alg name, feature name | |
2739 | * return: NULL (alg name was "") | |
2740 | * ERR_PTR(error) if something goes wrong | |
2741 | * or the crypto hash ptr, if it worked out ok. */ | |
2742 | struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev, | |
2743 | const char *alg, const char *name) | |
2744 | { | |
2745 | struct crypto_hash *tfm; | |
2746 | ||
2747 | if (!alg[0]) | |
2748 | return NULL; | |
2749 | ||
2750 | tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); | |
2751 | if (IS_ERR(tfm)) { | |
2752 | dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n", | |
2753 | alg, name, PTR_ERR(tfm)); | |
2754 | return tfm; | |
2755 | } | |
2756 | if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) { | |
2757 | crypto_free_hash(tfm); | |
2758 | dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name); | |
2759 | return ERR_PTR(-EINVAL); | |
2760 | } | |
2761 | return tfm; | |
2762 | } | |
2763 | ||
02918be2 | 2764 | static int receive_SyncParam(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int packet_size) |
b411b363 | 2765 | { |
81e84650 | 2766 | int ok = true; |
02918be2 | 2767 | struct p_rs_param_95 *p = &mdev->data.rbuf.rs_param_95; |
b411b363 PR |
2768 | unsigned int header_size, data_size, exp_max_sz; |
2769 | struct crypto_hash *verify_tfm = NULL; | |
2770 | struct crypto_hash *csums_tfm = NULL; | |
2771 | const int apv = mdev->agreed_pro_version; | |
778f271d PR |
2772 | int *rs_plan_s = NULL; |
2773 | int fifo_size = 0; | |
b411b363 PR |
2774 | |
2775 | exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) | |
2776 | : apv == 88 ? sizeof(struct p_rs_param) | |
2777 | + SHARED_SECRET_MAX | |
8e26f9cc PR |
2778 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
2779 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 2780 | |
02918be2 | 2781 | if (packet_size > exp_max_sz) { |
b411b363 | 2782 | dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n", |
02918be2 | 2783 | packet_size, exp_max_sz); |
81e84650 | 2784 | return false; |
b411b363 PR |
2785 | } |
2786 | ||
2787 | if (apv <= 88) { | |
02918be2 PR |
2788 | header_size = sizeof(struct p_rs_param) - sizeof(struct p_header80); |
2789 | data_size = packet_size - header_size; | |
8e26f9cc | 2790 | } else if (apv <= 94) { |
02918be2 PR |
2791 | header_size = sizeof(struct p_rs_param_89) - sizeof(struct p_header80); |
2792 | data_size = packet_size - header_size; | |
b411b363 | 2793 | D_ASSERT(data_size == 0); |
8e26f9cc | 2794 | } else { |
02918be2 PR |
2795 | header_size = sizeof(struct p_rs_param_95) - sizeof(struct p_header80); |
2796 | data_size = packet_size - header_size; | |
b411b363 PR |
2797 | D_ASSERT(data_size == 0); |
2798 | } | |
2799 | ||
2800 | /* initialize verify_alg and csums_alg */ | |
2801 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); | |
2802 | ||
02918be2 | 2803 | if (drbd_recv(mdev, &p->head.payload, header_size) != header_size) |
81e84650 | 2804 | return false; |
b411b363 PR |
2805 | |
2806 | mdev->sync_conf.rate = be32_to_cpu(p->rate); | |
2807 | ||
2808 | if (apv >= 88) { | |
2809 | if (apv == 88) { | |
2810 | if (data_size > SHARED_SECRET_MAX) { | |
2811 | dev_err(DEV, "verify-alg too long, " | |
2812 | "peer wants %u, accepting only %u byte\n", | |
2813 | data_size, SHARED_SECRET_MAX); | |
81e84650 | 2814 | return false; |
b411b363 PR |
2815 | } |
2816 | ||
2817 | if (drbd_recv(mdev, p->verify_alg, data_size) != data_size) | |
81e84650 | 2818 | return false; |
b411b363 PR |
2819 | |
2820 | /* we expect NUL terminated string */ | |
2821 | /* but just in case someone tries to be evil */ | |
2822 | D_ASSERT(p->verify_alg[data_size-1] == 0); | |
2823 | p->verify_alg[data_size-1] = 0; | |
2824 | ||
2825 | } else /* apv >= 89 */ { | |
2826 | /* we still expect NUL terminated strings */ | |
2827 | /* but just in case someone tries to be evil */ | |
2828 | D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0); | |
2829 | D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0); | |
2830 | p->verify_alg[SHARED_SECRET_MAX-1] = 0; | |
2831 | p->csums_alg[SHARED_SECRET_MAX-1] = 0; | |
2832 | } | |
2833 | ||
2834 | if (strcmp(mdev->sync_conf.verify_alg, p->verify_alg)) { | |
2835 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { | |
2836 | dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", | |
2837 | mdev->sync_conf.verify_alg, p->verify_alg); | |
2838 | goto disconnect; | |
2839 | } | |
2840 | verify_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
2841 | p->verify_alg, "verify-alg"); | |
2842 | if (IS_ERR(verify_tfm)) { | |
2843 | verify_tfm = NULL; | |
2844 | goto disconnect; | |
2845 | } | |
2846 | } | |
2847 | ||
2848 | if (apv >= 89 && strcmp(mdev->sync_conf.csums_alg, p->csums_alg)) { | |
2849 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { | |
2850 | dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", | |
2851 | mdev->sync_conf.csums_alg, p->csums_alg); | |
2852 | goto disconnect; | |
2853 | } | |
2854 | csums_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
2855 | p->csums_alg, "csums-alg"); | |
2856 | if (IS_ERR(csums_tfm)) { | |
2857 | csums_tfm = NULL; | |
2858 | goto disconnect; | |
2859 | } | |
2860 | } | |
2861 | ||
8e26f9cc PR |
2862 | if (apv > 94) { |
2863 | mdev->sync_conf.rate = be32_to_cpu(p->rate); | |
2864 | mdev->sync_conf.c_plan_ahead = be32_to_cpu(p->c_plan_ahead); | |
2865 | mdev->sync_conf.c_delay_target = be32_to_cpu(p->c_delay_target); | |
2866 | mdev->sync_conf.c_fill_target = be32_to_cpu(p->c_fill_target); | |
2867 | mdev->sync_conf.c_max_rate = be32_to_cpu(p->c_max_rate); | |
778f271d PR |
2868 | |
2869 | fifo_size = (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ; | |
2870 | if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) { | |
2871 | rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL); | |
2872 | if (!rs_plan_s) { | |
2873 | dev_err(DEV, "kmalloc of fifo_buffer failed"); | |
2874 | goto disconnect; | |
2875 | } | |
2876 | } | |
8e26f9cc | 2877 | } |
b411b363 PR |
2878 | |
2879 | spin_lock(&mdev->peer_seq_lock); | |
2880 | /* lock against drbd_nl_syncer_conf() */ | |
2881 | if (verify_tfm) { | |
2882 | strcpy(mdev->sync_conf.verify_alg, p->verify_alg); | |
2883 | mdev->sync_conf.verify_alg_len = strlen(p->verify_alg) + 1; | |
2884 | crypto_free_hash(mdev->verify_tfm); | |
2885 | mdev->verify_tfm = verify_tfm; | |
2886 | dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg); | |
2887 | } | |
2888 | if (csums_tfm) { | |
2889 | strcpy(mdev->sync_conf.csums_alg, p->csums_alg); | |
2890 | mdev->sync_conf.csums_alg_len = strlen(p->csums_alg) + 1; | |
2891 | crypto_free_hash(mdev->csums_tfm); | |
2892 | mdev->csums_tfm = csums_tfm; | |
2893 | dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg); | |
2894 | } | |
778f271d PR |
2895 | if (fifo_size != mdev->rs_plan_s.size) { |
2896 | kfree(mdev->rs_plan_s.values); | |
2897 | mdev->rs_plan_s.values = rs_plan_s; | |
2898 | mdev->rs_plan_s.size = fifo_size; | |
2899 | mdev->rs_planed = 0; | |
2900 | } | |
b411b363 PR |
2901 | spin_unlock(&mdev->peer_seq_lock); |
2902 | } | |
2903 | ||
2904 | return ok; | |
2905 | disconnect: | |
2906 | /* just for completeness: actually not needed, | |
2907 | * as this is not reached if csums_tfm was ok. */ | |
2908 | crypto_free_hash(csums_tfm); | |
2909 | /* but free the verify_tfm again, if csums_tfm did not work out */ | |
2910 | crypto_free_hash(verify_tfm); | |
2911 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 2912 | return false; |
b411b363 PR |
2913 | } |
2914 | ||
b411b363 PR |
2915 | /* warn if the arguments differ by more than 12.5% */ |
2916 | static void warn_if_differ_considerably(struct drbd_conf *mdev, | |
2917 | const char *s, sector_t a, sector_t b) | |
2918 | { | |
2919 | sector_t d; | |
2920 | if (a == 0 || b == 0) | |
2921 | return; | |
2922 | d = (a > b) ? (a - b) : (b - a); | |
2923 | if (d > (a>>3) || d > (b>>3)) | |
2924 | dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s, | |
2925 | (unsigned long long)a, (unsigned long long)b); | |
2926 | } | |
2927 | ||
02918be2 | 2928 | static int receive_sizes(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 2929 | { |
02918be2 | 2930 | struct p_sizes *p = &mdev->data.rbuf.sizes; |
b411b363 | 2931 | enum determine_dev_size dd = unchanged; |
b411b363 PR |
2932 | sector_t p_size, p_usize, my_usize; |
2933 | int ldsc = 0; /* local disk size changed */ | |
e89b591c | 2934 | enum dds_flags ddsf; |
b411b363 | 2935 | |
b411b363 PR |
2936 | p_size = be64_to_cpu(p->d_size); |
2937 | p_usize = be64_to_cpu(p->u_size); | |
2938 | ||
2939 | if (p_size == 0 && mdev->state.disk == D_DISKLESS) { | |
2940 | dev_err(DEV, "some backing storage is needed\n"); | |
2941 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 2942 | return false; |
b411b363 PR |
2943 | } |
2944 | ||
2945 | /* just store the peer's disk size for now. | |
2946 | * we still need to figure out whether we accept that. */ | |
2947 | mdev->p_size = p_size; | |
2948 | ||
b411b363 PR |
2949 | if (get_ldev(mdev)) { |
2950 | warn_if_differ_considerably(mdev, "lower level device sizes", | |
2951 | p_size, drbd_get_max_capacity(mdev->ldev)); | |
2952 | warn_if_differ_considerably(mdev, "user requested size", | |
2953 | p_usize, mdev->ldev->dc.disk_size); | |
2954 | ||
2955 | /* if this is the first connect, or an otherwise expected | |
2956 | * param exchange, choose the minimum */ | |
2957 | if (mdev->state.conn == C_WF_REPORT_PARAMS) | |
2958 | p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size, | |
2959 | p_usize); | |
2960 | ||
2961 | my_usize = mdev->ldev->dc.disk_size; | |
2962 | ||
2963 | if (mdev->ldev->dc.disk_size != p_usize) { | |
2964 | mdev->ldev->dc.disk_size = p_usize; | |
2965 | dev_info(DEV, "Peer sets u_size to %lu sectors\n", | |
2966 | (unsigned long)mdev->ldev->dc.disk_size); | |
2967 | } | |
2968 | ||
2969 | /* Never shrink a device with usable data during connect. | |
2970 | But allow online shrinking if we are connected. */ | |
a393db6f | 2971 | if (drbd_new_dev_size(mdev, mdev->ldev, 0) < |
b411b363 PR |
2972 | drbd_get_capacity(mdev->this_bdev) && |
2973 | mdev->state.disk >= D_OUTDATED && | |
2974 | mdev->state.conn < C_CONNECTED) { | |
2975 | dev_err(DEV, "The peer's disk size is too small!\n"); | |
2976 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
2977 | mdev->ldev->dc.disk_size = my_usize; | |
2978 | put_ldev(mdev); | |
81e84650 | 2979 | return false; |
b411b363 PR |
2980 | } |
2981 | put_ldev(mdev); | |
2982 | } | |
b411b363 | 2983 | |
e89b591c | 2984 | ddsf = be16_to_cpu(p->dds_flags); |
b411b363 | 2985 | if (get_ldev(mdev)) { |
24c4830c | 2986 | dd = drbd_determine_dev_size(mdev, ddsf); |
b411b363 PR |
2987 | put_ldev(mdev); |
2988 | if (dd == dev_size_error) | |
81e84650 | 2989 | return false; |
b411b363 PR |
2990 | drbd_md_sync(mdev); |
2991 | } else { | |
2992 | /* I am diskless, need to accept the peer's size. */ | |
2993 | drbd_set_my_capacity(mdev, p_size); | |
2994 | } | |
2995 | ||
99432fcc PR |
2996 | mdev->peer_max_bio_size = be32_to_cpu(p->max_bio_size); |
2997 | drbd_reconsider_max_bio_size(mdev); | |
2998 | ||
b411b363 PR |
2999 | if (get_ldev(mdev)) { |
3000 | if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) { | |
3001 | mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); | |
3002 | ldsc = 1; | |
3003 | } | |
3004 | ||
b411b363 PR |
3005 | put_ldev(mdev); |
3006 | } | |
3007 | ||
3008 | if (mdev->state.conn > C_WF_REPORT_PARAMS) { | |
3009 | if (be64_to_cpu(p->c_size) != | |
3010 | drbd_get_capacity(mdev->this_bdev) || ldsc) { | |
3011 | /* we have different sizes, probably peer | |
3012 | * needs to know my new size... */ | |
e89b591c | 3013 | drbd_send_sizes(mdev, 0, ddsf); |
b411b363 PR |
3014 | } |
3015 | if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) || | |
3016 | (dd == grew && mdev->state.conn == C_CONNECTED)) { | |
3017 | if (mdev->state.pdsk >= D_INCONSISTENT && | |
e89b591c PR |
3018 | mdev->state.disk >= D_INCONSISTENT) { |
3019 | if (ddsf & DDSF_NO_RESYNC) | |
3020 | dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n"); | |
3021 | else | |
3022 | resync_after_online_grow(mdev); | |
3023 | } else | |
b411b363 PR |
3024 | set_bit(RESYNC_AFTER_NEG, &mdev->flags); |
3025 | } | |
3026 | } | |
3027 | ||
81e84650 | 3028 | return true; |
b411b363 PR |
3029 | } |
3030 | ||
02918be2 | 3031 | static int receive_uuids(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 3032 | { |
02918be2 | 3033 | struct p_uuids *p = &mdev->data.rbuf.uuids; |
b411b363 | 3034 | u64 *p_uuid; |
62b0da3a | 3035 | int i, updated_uuids = 0; |
b411b363 | 3036 | |
b411b363 PR |
3037 | p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); |
3038 | ||
3039 | for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) | |
3040 | p_uuid[i] = be64_to_cpu(p->uuid[i]); | |
3041 | ||
3042 | kfree(mdev->p_uuid); | |
3043 | mdev->p_uuid = p_uuid; | |
3044 | ||
3045 | if (mdev->state.conn < C_CONNECTED && | |
3046 | mdev->state.disk < D_INCONSISTENT && | |
3047 | mdev->state.role == R_PRIMARY && | |
3048 | (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { | |
3049 | dev_err(DEV, "Can only connect to data with current UUID=%016llX\n", | |
3050 | (unsigned long long)mdev->ed_uuid); | |
3051 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 3052 | return false; |
b411b363 PR |
3053 | } |
3054 | ||
3055 | if (get_ldev(mdev)) { | |
3056 | int skip_initial_sync = | |
3057 | mdev->state.conn == C_CONNECTED && | |
3058 | mdev->agreed_pro_version >= 90 && | |
3059 | mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && | |
3060 | (p_uuid[UI_FLAGS] & 8); | |
3061 | if (skip_initial_sync) { | |
3062 | dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n"); | |
3063 | drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, | |
20ceb2b2 LE |
3064 | "clear_n_write from receive_uuids", |
3065 | BM_LOCKED_TEST_ALLOWED); | |
b411b363 PR |
3066 | _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]); |
3067 | _drbd_uuid_set(mdev, UI_BITMAP, 0); | |
3068 | _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), | |
3069 | CS_VERBOSE, NULL); | |
3070 | drbd_md_sync(mdev); | |
62b0da3a | 3071 | updated_uuids = 1; |
b411b363 PR |
3072 | } |
3073 | put_ldev(mdev); | |
18a50fa2 PR |
3074 | } else if (mdev->state.disk < D_INCONSISTENT && |
3075 | mdev->state.role == R_PRIMARY) { | |
3076 | /* I am a diskless primary, the peer just created a new current UUID | |
3077 | for me. */ | |
62b0da3a | 3078 | updated_uuids = drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
b411b363 PR |
3079 | } |
3080 | ||
3081 | /* Before we test for the disk state, we should wait until an eventually | |
3082 | ongoing cluster wide state change is finished. That is important if | |
3083 | we are primary and are detaching from our disk. We need to see the | |
3084 | new disk state... */ | |
3085 | wait_event(mdev->misc_wait, !test_bit(CLUSTER_ST_CHANGE, &mdev->flags)); | |
3086 | if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT) | |
62b0da3a LE |
3087 | updated_uuids |= drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
3088 | ||
3089 | if (updated_uuids) | |
3090 | drbd_print_uuids(mdev, "receiver updated UUIDs to"); | |
b411b363 | 3091 | |
81e84650 | 3092 | return true; |
b411b363 PR |
3093 | } |
3094 | ||
3095 | /** | |
3096 | * convert_state() - Converts the peer's view of the cluster state to our point of view | |
3097 | * @ps: The state as seen by the peer. | |
3098 | */ | |
3099 | static union drbd_state convert_state(union drbd_state ps) | |
3100 | { | |
3101 | union drbd_state ms; | |
3102 | ||
3103 | static enum drbd_conns c_tab[] = { | |
3104 | [C_CONNECTED] = C_CONNECTED, | |
3105 | ||
3106 | [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, | |
3107 | [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, | |
3108 | [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ | |
3109 | [C_VERIFY_S] = C_VERIFY_T, | |
3110 | [C_MASK] = C_MASK, | |
3111 | }; | |
3112 | ||
3113 | ms.i = ps.i; | |
3114 | ||
3115 | ms.conn = c_tab[ps.conn]; | |
3116 | ms.peer = ps.role; | |
3117 | ms.role = ps.peer; | |
3118 | ms.pdsk = ps.disk; | |
3119 | ms.disk = ps.pdsk; | |
3120 | ms.peer_isp = (ps.aftr_isp | ps.user_isp); | |
3121 | ||
3122 | return ms; | |
3123 | } | |
3124 | ||
02918be2 | 3125 | static int receive_req_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 3126 | { |
02918be2 | 3127 | struct p_req_state *p = &mdev->data.rbuf.req_state; |
b411b363 | 3128 | union drbd_state mask, val; |
bf885f8a | 3129 | enum drbd_state_rv rv; |
b411b363 | 3130 | |
b411b363 PR |
3131 | mask.i = be32_to_cpu(p->mask); |
3132 | val.i = be32_to_cpu(p->val); | |
3133 | ||
3134 | if (test_bit(DISCARD_CONCURRENT, &mdev->flags) && | |
3135 | test_bit(CLUSTER_ST_CHANGE, &mdev->flags)) { | |
3136 | drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG); | |
81e84650 | 3137 | return true; |
b411b363 PR |
3138 | } |
3139 | ||
3140 | mask = convert_state(mask); | |
3141 | val = convert_state(val); | |
3142 | ||
3143 | rv = drbd_change_state(mdev, CS_VERBOSE, mask, val); | |
3144 | ||
3145 | drbd_send_sr_reply(mdev, rv); | |
3146 | drbd_md_sync(mdev); | |
3147 | ||
81e84650 | 3148 | return true; |
b411b363 PR |
3149 | } |
3150 | ||
02918be2 | 3151 | static int receive_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 3152 | { |
02918be2 | 3153 | struct p_state *p = &mdev->data.rbuf.state; |
4ac4aada | 3154 | union drbd_state os, ns, peer_state; |
b411b363 | 3155 | enum drbd_disk_state real_peer_disk; |
65d922c3 | 3156 | enum chg_state_flags cs_flags; |
b411b363 PR |
3157 | int rv; |
3158 | ||
b411b363 PR |
3159 | peer_state.i = be32_to_cpu(p->state); |
3160 | ||
3161 | real_peer_disk = peer_state.disk; | |
3162 | if (peer_state.disk == D_NEGOTIATING) { | |
3163 | real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; | |
3164 | dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); | |
3165 | } | |
3166 | ||
3167 | spin_lock_irq(&mdev->req_lock); | |
3168 | retry: | |
4ac4aada | 3169 | os = ns = mdev->state; |
b411b363 PR |
3170 | spin_unlock_irq(&mdev->req_lock); |
3171 | ||
40424e4a LE |
3172 | /* If this is the "end of sync" confirmation, usually the peer disk |
3173 | * transitions from D_INCONSISTENT to D_UP_TO_DATE. For empty (0 bits | |
3174 | * set) resync started in PausedSyncT, or if the timing of pause-/ | |
3175 | * unpause-sync events has been "just right", the peer disk may | |
3176 | * transition from D_CONSISTENT to D_UP_TO_DATE as well. | |
3177 | */ | |
3178 | if ((os.pdsk == D_INCONSISTENT || os.pdsk == D_CONSISTENT) && | |
3179 | real_peer_disk == D_UP_TO_DATE && | |
e9ef7bb6 LE |
3180 | os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { |
3181 | /* If we are (becoming) SyncSource, but peer is still in sync | |
3182 | * preparation, ignore its uptodate-ness to avoid flapping, it | |
3183 | * will change to inconsistent once the peer reaches active | |
3184 | * syncing states. | |
3185 | * It may have changed syncer-paused flags, however, so we | |
3186 | * cannot ignore this completely. */ | |
3187 | if (peer_state.conn > C_CONNECTED && | |
3188 | peer_state.conn < C_SYNC_SOURCE) | |
3189 | real_peer_disk = D_INCONSISTENT; | |
3190 | ||
3191 | /* if peer_state changes to connected at the same time, | |
3192 | * it explicitly notifies us that it finished resync. | |
3193 | * Maybe we should finish it up, too? */ | |
3194 | else if (os.conn >= C_SYNC_SOURCE && | |
3195 | peer_state.conn == C_CONNECTED) { | |
3196 | if (drbd_bm_total_weight(mdev) <= mdev->rs_failed) | |
3197 | drbd_resync_finished(mdev); | |
81e84650 | 3198 | return true; |
e9ef7bb6 LE |
3199 | } |
3200 | } | |
3201 | ||
3202 | /* peer says his disk is inconsistent, while we think it is uptodate, | |
3203 | * and this happens while the peer still thinks we have a sync going on, | |
3204 | * but we think we are already done with the sync. | |
3205 | * We ignore this to avoid flapping pdsk. | |
3206 | * This should not happen, if the peer is a recent version of drbd. */ | |
3207 | if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && | |
3208 | os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) | |
3209 | real_peer_disk = D_UP_TO_DATE; | |
3210 | ||
4ac4aada LE |
3211 | if (ns.conn == C_WF_REPORT_PARAMS) |
3212 | ns.conn = C_CONNECTED; | |
b411b363 | 3213 | |
67531718 PR |
3214 | if (peer_state.conn == C_AHEAD) |
3215 | ns.conn = C_BEHIND; | |
3216 | ||
b411b363 PR |
3217 | if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING && |
3218 | get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3219 | int cr; /* consider resync */ | |
3220 | ||
3221 | /* if we established a new connection */ | |
4ac4aada | 3222 | cr = (os.conn < C_CONNECTED); |
b411b363 PR |
3223 | /* if we had an established connection |
3224 | * and one of the nodes newly attaches a disk */ | |
4ac4aada | 3225 | cr |= (os.conn == C_CONNECTED && |
b411b363 | 3226 | (peer_state.disk == D_NEGOTIATING || |
4ac4aada | 3227 | os.disk == D_NEGOTIATING)); |
b411b363 PR |
3228 | /* if we have both been inconsistent, and the peer has been |
3229 | * forced to be UpToDate with --overwrite-data */ | |
3230 | cr |= test_bit(CONSIDER_RESYNC, &mdev->flags); | |
3231 | /* if we had been plain connected, and the admin requested to | |
3232 | * start a sync by "invalidate" or "invalidate-remote" */ | |
4ac4aada | 3233 | cr |= (os.conn == C_CONNECTED && |
b411b363 PR |
3234 | (peer_state.conn >= C_STARTING_SYNC_S && |
3235 | peer_state.conn <= C_WF_BITMAP_T)); | |
3236 | ||
3237 | if (cr) | |
4ac4aada | 3238 | ns.conn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk); |
b411b363 PR |
3239 | |
3240 | put_ldev(mdev); | |
4ac4aada LE |
3241 | if (ns.conn == C_MASK) { |
3242 | ns.conn = C_CONNECTED; | |
b411b363 | 3243 | if (mdev->state.disk == D_NEGOTIATING) { |
82f59cc6 | 3244 | drbd_force_state(mdev, NS(disk, D_FAILED)); |
b411b363 PR |
3245 | } else if (peer_state.disk == D_NEGOTIATING) { |
3246 | dev_err(DEV, "Disk attach process on the peer node was aborted.\n"); | |
3247 | peer_state.disk = D_DISKLESS; | |
580b9767 | 3248 | real_peer_disk = D_DISKLESS; |
b411b363 | 3249 | } else { |
cf14c2e9 | 3250 | if (test_and_clear_bit(CONN_DRY_RUN, &mdev->flags)) |
81e84650 | 3251 | return false; |
4ac4aada | 3252 | D_ASSERT(os.conn == C_WF_REPORT_PARAMS); |
b411b363 | 3253 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); |
81e84650 | 3254 | return false; |
b411b363 PR |
3255 | } |
3256 | } | |
3257 | } | |
3258 | ||
3259 | spin_lock_irq(&mdev->req_lock); | |
4ac4aada | 3260 | if (mdev->state.i != os.i) |
b411b363 PR |
3261 | goto retry; |
3262 | clear_bit(CONSIDER_RESYNC, &mdev->flags); | |
b411b363 PR |
3263 | ns.peer = peer_state.role; |
3264 | ns.pdsk = real_peer_disk; | |
3265 | ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); | |
4ac4aada | 3266 | if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) |
b411b363 | 3267 | ns.disk = mdev->new_state_tmp.disk; |
4ac4aada LE |
3268 | cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); |
3269 | if (ns.pdsk == D_CONSISTENT && is_susp(ns) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && | |
481c6f50 PR |
3270 | test_bit(NEW_CUR_UUID, &mdev->flags)) { |
3271 | /* Do not allow tl_restart(resend) for a rebooted peer. We can only allow this | |
3272 | for temporal network outages! */ | |
3273 | spin_unlock_irq(&mdev->req_lock); | |
3274 | dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); | |
3275 | tl_clear(mdev); | |
3276 | drbd_uuid_new_current(mdev); | |
3277 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
3278 | drbd_force_state(mdev, NS2(conn, C_PROTOCOL_ERROR, susp, 0)); | |
81e84650 | 3279 | return false; |
481c6f50 | 3280 | } |
65d922c3 | 3281 | rv = _drbd_set_state(mdev, ns, cs_flags, NULL); |
b411b363 PR |
3282 | ns = mdev->state; |
3283 | spin_unlock_irq(&mdev->req_lock); | |
3284 | ||
3285 | if (rv < SS_SUCCESS) { | |
3286 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 3287 | return false; |
b411b363 PR |
3288 | } |
3289 | ||
4ac4aada LE |
3290 | if (os.conn > C_WF_REPORT_PARAMS) { |
3291 | if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && | |
b411b363 PR |
3292 | peer_state.disk != D_NEGOTIATING ) { |
3293 | /* we want resync, peer has not yet decided to sync... */ | |
3294 | /* Nowadays only used when forcing a node into primary role and | |
3295 | setting its disk to UpToDate with that */ | |
3296 | drbd_send_uuids(mdev); | |
3297 | drbd_send_state(mdev); | |
3298 | } | |
3299 | } | |
3300 | ||
3301 | mdev->net_conf->want_lose = 0; | |
3302 | ||
3303 | drbd_md_sync(mdev); /* update connected indicator, la_size, ... */ | |
3304 | ||
81e84650 | 3305 | return true; |
b411b363 PR |
3306 | } |
3307 | ||
02918be2 | 3308 | static int receive_sync_uuid(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 3309 | { |
02918be2 | 3310 | struct p_rs_uuid *p = &mdev->data.rbuf.rs_uuid; |
b411b363 PR |
3311 | |
3312 | wait_event(mdev->misc_wait, | |
3313 | mdev->state.conn == C_WF_SYNC_UUID || | |
c4752ef1 | 3314 | mdev->state.conn == C_BEHIND || |
b411b363 PR |
3315 | mdev->state.conn < C_CONNECTED || |
3316 | mdev->state.disk < D_NEGOTIATING); | |
3317 | ||
3318 | /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */ | |
3319 | ||
b411b363 PR |
3320 | /* Here the _drbd_uuid_ functions are right, current should |
3321 | _not_ be rotated into the history */ | |
3322 | if (get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3323 | _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid)); | |
3324 | _drbd_uuid_set(mdev, UI_BITMAP, 0UL); | |
3325 | ||
62b0da3a | 3326 | drbd_print_uuids(mdev, "updated sync uuid"); |
b411b363 PR |
3327 | drbd_start_resync(mdev, C_SYNC_TARGET); |
3328 | ||
3329 | put_ldev(mdev); | |
3330 | } else | |
3331 | dev_err(DEV, "Ignoring SyncUUID packet!\n"); | |
3332 | ||
81e84650 | 3333 | return true; |
b411b363 PR |
3334 | } |
3335 | ||
2c46407d AG |
3336 | /** |
3337 | * receive_bitmap_plain | |
3338 | * | |
3339 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3340 | * code upon failure. | |
3341 | */ | |
3342 | static int | |
02918be2 PR |
3343 | receive_bitmap_plain(struct drbd_conf *mdev, unsigned int data_size, |
3344 | unsigned long *buffer, struct bm_xfer_ctx *c) | |
b411b363 PR |
3345 | { |
3346 | unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset); | |
3347 | unsigned want = num_words * sizeof(long); | |
2c46407d | 3348 | int err; |
b411b363 | 3349 | |
02918be2 PR |
3350 | if (want != data_size) { |
3351 | dev_err(DEV, "%s:want (%u) != data_size (%u)\n", __func__, want, data_size); | |
2c46407d | 3352 | return -EIO; |
b411b363 PR |
3353 | } |
3354 | if (want == 0) | |
2c46407d AG |
3355 | return 0; |
3356 | err = drbd_recv(mdev, buffer, want); | |
3357 | if (err != want) { | |
3358 | if (err >= 0) | |
3359 | err = -EIO; | |
3360 | return err; | |
3361 | } | |
b411b363 PR |
3362 | |
3363 | drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer); | |
3364 | ||
3365 | c->word_offset += num_words; | |
3366 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
3367 | if (c->bit_offset > c->bm_bits) | |
3368 | c->bit_offset = c->bm_bits; | |
3369 | ||
2c46407d | 3370 | return 1; |
b411b363 PR |
3371 | } |
3372 | ||
2c46407d AG |
3373 | /** |
3374 | * recv_bm_rle_bits | |
3375 | * | |
3376 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3377 | * code upon failure. | |
3378 | */ | |
3379 | static int | |
b411b363 PR |
3380 | recv_bm_rle_bits(struct drbd_conf *mdev, |
3381 | struct p_compressed_bm *p, | |
3382 | struct bm_xfer_ctx *c) | |
3383 | { | |
3384 | struct bitstream bs; | |
3385 | u64 look_ahead; | |
3386 | u64 rl; | |
3387 | u64 tmp; | |
3388 | unsigned long s = c->bit_offset; | |
3389 | unsigned long e; | |
004352fa | 3390 | int len = be16_to_cpu(p->head.length) - (sizeof(*p) - sizeof(p->head)); |
b411b363 PR |
3391 | int toggle = DCBP_get_start(p); |
3392 | int have; | |
3393 | int bits; | |
3394 | ||
3395 | bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p)); | |
3396 | ||
3397 | bits = bitstream_get_bits(&bs, &look_ahead, 64); | |
3398 | if (bits < 0) | |
2c46407d | 3399 | return -EIO; |
b411b363 PR |
3400 | |
3401 | for (have = bits; have > 0; s += rl, toggle = !toggle) { | |
3402 | bits = vli_decode_bits(&rl, look_ahead); | |
3403 | if (bits <= 0) | |
2c46407d | 3404 | return -EIO; |
b411b363 PR |
3405 | |
3406 | if (toggle) { | |
3407 | e = s + rl -1; | |
3408 | if (e >= c->bm_bits) { | |
3409 | dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); | |
2c46407d | 3410 | return -EIO; |
b411b363 PR |
3411 | } |
3412 | _drbd_bm_set_bits(mdev, s, e); | |
3413 | } | |
3414 | ||
3415 | if (have < bits) { | |
3416 | dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", | |
3417 | have, bits, look_ahead, | |
3418 | (unsigned int)(bs.cur.b - p->code), | |
3419 | (unsigned int)bs.buf_len); | |
2c46407d | 3420 | return -EIO; |
b411b363 PR |
3421 | } |
3422 | look_ahead >>= bits; | |
3423 | have -= bits; | |
3424 | ||
3425 | bits = bitstream_get_bits(&bs, &tmp, 64 - have); | |
3426 | if (bits < 0) | |
2c46407d | 3427 | return -EIO; |
b411b363 PR |
3428 | look_ahead |= tmp << have; |
3429 | have += bits; | |
3430 | } | |
3431 | ||
3432 | c->bit_offset = s; | |
3433 | bm_xfer_ctx_bit_to_word_offset(c); | |
3434 | ||
2c46407d | 3435 | return (s != c->bm_bits); |
b411b363 PR |
3436 | } |
3437 | ||
2c46407d AG |
3438 | /** |
3439 | * decode_bitmap_c | |
3440 | * | |
3441 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3442 | * code upon failure. | |
3443 | */ | |
3444 | static int | |
b411b363 PR |
3445 | decode_bitmap_c(struct drbd_conf *mdev, |
3446 | struct p_compressed_bm *p, | |
3447 | struct bm_xfer_ctx *c) | |
3448 | { | |
3449 | if (DCBP_get_code(p) == RLE_VLI_Bits) | |
3450 | return recv_bm_rle_bits(mdev, p, c); | |
3451 | ||
3452 | /* other variants had been implemented for evaluation, | |
3453 | * but have been dropped as this one turned out to be "best" | |
3454 | * during all our tests. */ | |
3455 | ||
3456 | dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding); | |
3457 | drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); | |
2c46407d | 3458 | return -EIO; |
b411b363 PR |
3459 | } |
3460 | ||
3461 | void INFO_bm_xfer_stats(struct drbd_conf *mdev, | |
3462 | const char *direction, struct bm_xfer_ctx *c) | |
3463 | { | |
3464 | /* what would it take to transfer it "plaintext" */ | |
0b70a13d | 3465 | unsigned plain = sizeof(struct p_header80) * |
b411b363 PR |
3466 | ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1) |
3467 | + c->bm_words * sizeof(long); | |
3468 | unsigned total = c->bytes[0] + c->bytes[1]; | |
3469 | unsigned r; | |
3470 | ||
3471 | /* total can not be zero. but just in case: */ | |
3472 | if (total == 0) | |
3473 | return; | |
3474 | ||
3475 | /* don't report if not compressed */ | |
3476 | if (total >= plain) | |
3477 | return; | |
3478 | ||
3479 | /* total < plain. check for overflow, still */ | |
3480 | r = (total > UINT_MAX/1000) ? (total / (plain/1000)) | |
3481 | : (1000 * total / plain); | |
3482 | ||
3483 | if (r > 1000) | |
3484 | r = 1000; | |
3485 | ||
3486 | r = 1000 - r; | |
3487 | dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " | |
3488 | "total %u; compression: %u.%u%%\n", | |
3489 | direction, | |
3490 | c->bytes[1], c->packets[1], | |
3491 | c->bytes[0], c->packets[0], | |
3492 | total, r/10, r % 10); | |
3493 | } | |
3494 | ||
3495 | /* Since we are processing the bitfield from lower addresses to higher, | |
3496 | it does not matter if the process it in 32 bit chunks or 64 bit | |
3497 | chunks as long as it is little endian. (Understand it as byte stream, | |
3498 | beginning with the lowest byte...) If we would use big endian | |
3499 | we would need to process it from the highest address to the lowest, | |
3500 | in order to be agnostic to the 32 vs 64 bits issue. | |
3501 | ||
3502 | returns 0 on failure, 1 if we successfully received it. */ | |
02918be2 | 3503 | static int receive_bitmap(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
3504 | { |
3505 | struct bm_xfer_ctx c; | |
3506 | void *buffer; | |
2c46407d | 3507 | int err; |
81e84650 | 3508 | int ok = false; |
02918be2 | 3509 | struct p_header80 *h = &mdev->data.rbuf.header.h80; |
b411b363 | 3510 | |
20ceb2b2 LE |
3511 | drbd_bm_lock(mdev, "receive bitmap", BM_LOCKED_SET_ALLOWED); |
3512 | /* you are supposed to send additional out-of-sync information | |
3513 | * if you actually set bits during this phase */ | |
b411b363 PR |
3514 | |
3515 | /* maybe we should use some per thread scratch page, | |
3516 | * and allocate that during initial device creation? */ | |
3517 | buffer = (unsigned long *) __get_free_page(GFP_NOIO); | |
3518 | if (!buffer) { | |
3519 | dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__); | |
3520 | goto out; | |
3521 | } | |
3522 | ||
3523 | c = (struct bm_xfer_ctx) { | |
3524 | .bm_bits = drbd_bm_bits(mdev), | |
3525 | .bm_words = drbd_bm_words(mdev), | |
3526 | }; | |
3527 | ||
2c46407d | 3528 | for(;;) { |
02918be2 | 3529 | if (cmd == P_BITMAP) { |
2c46407d | 3530 | err = receive_bitmap_plain(mdev, data_size, buffer, &c); |
02918be2 | 3531 | } else if (cmd == P_COMPRESSED_BITMAP) { |
b411b363 PR |
3532 | /* MAYBE: sanity check that we speak proto >= 90, |
3533 | * and the feature is enabled! */ | |
3534 | struct p_compressed_bm *p; | |
3535 | ||
02918be2 | 3536 | if (data_size > BM_PACKET_PAYLOAD_BYTES) { |
b411b363 PR |
3537 | dev_err(DEV, "ReportCBitmap packet too large\n"); |
3538 | goto out; | |
3539 | } | |
3540 | /* use the page buff */ | |
3541 | p = buffer; | |
3542 | memcpy(p, h, sizeof(*h)); | |
02918be2 | 3543 | if (drbd_recv(mdev, p->head.payload, data_size) != data_size) |
b411b363 | 3544 | goto out; |
004352fa LE |
3545 | if (data_size <= (sizeof(*p) - sizeof(p->head))) { |
3546 | dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", data_size); | |
78fcbdae | 3547 | goto out; |
b411b363 | 3548 | } |
2c46407d | 3549 | err = decode_bitmap_c(mdev, p, &c); |
b411b363 | 3550 | } else { |
02918be2 | 3551 | dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", cmd); |
b411b363 PR |
3552 | goto out; |
3553 | } | |
3554 | ||
02918be2 PR |
3555 | c.packets[cmd == P_BITMAP]++; |
3556 | c.bytes[cmd == P_BITMAP] += sizeof(struct p_header80) + data_size; | |
b411b363 | 3557 | |
2c46407d AG |
3558 | if (err <= 0) { |
3559 | if (err < 0) | |
3560 | goto out; | |
b411b363 | 3561 | break; |
2c46407d | 3562 | } |
02918be2 | 3563 | if (!drbd_recv_header(mdev, &cmd, &data_size)) |
b411b363 | 3564 | goto out; |
2c46407d | 3565 | } |
b411b363 PR |
3566 | |
3567 | INFO_bm_xfer_stats(mdev, "receive", &c); | |
3568 | ||
3569 | if (mdev->state.conn == C_WF_BITMAP_T) { | |
de1f8e4a AG |
3570 | enum drbd_state_rv rv; |
3571 | ||
b411b363 PR |
3572 | ok = !drbd_send_bitmap(mdev); |
3573 | if (!ok) | |
3574 | goto out; | |
3575 | /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ | |
de1f8e4a AG |
3576 | rv = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); |
3577 | D_ASSERT(rv == SS_SUCCESS); | |
b411b363 PR |
3578 | } else if (mdev->state.conn != C_WF_BITMAP_S) { |
3579 | /* admin may have requested C_DISCONNECTING, | |
3580 | * other threads may have noticed network errors */ | |
3581 | dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n", | |
3582 | drbd_conn_str(mdev->state.conn)); | |
3583 | } | |
3584 | ||
81e84650 | 3585 | ok = true; |
b411b363 | 3586 | out: |
20ceb2b2 | 3587 | drbd_bm_unlock(mdev); |
b411b363 PR |
3588 | if (ok && mdev->state.conn == C_WF_BITMAP_S) |
3589 | drbd_start_resync(mdev, C_SYNC_SOURCE); | |
3590 | free_page((unsigned long) buffer); | |
3591 | return ok; | |
3592 | } | |
3593 | ||
02918be2 | 3594 | static int receive_skip(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
3595 | { |
3596 | /* TODO zero copy sink :) */ | |
3597 | static char sink[128]; | |
3598 | int size, want, r; | |
3599 | ||
02918be2 PR |
3600 | dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n", |
3601 | cmd, data_size); | |
b411b363 | 3602 | |
02918be2 | 3603 | size = data_size; |
b411b363 PR |
3604 | while (size > 0) { |
3605 | want = min_t(int, size, sizeof(sink)); | |
3606 | r = drbd_recv(mdev, sink, want); | |
3607 | ERR_IF(r <= 0) break; | |
3608 | size -= r; | |
3609 | } | |
3610 | return size == 0; | |
3611 | } | |
3612 | ||
02918be2 | 3613 | static int receive_UnplugRemote(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
0ced55a3 | 3614 | { |
e7f52dfb LE |
3615 | /* Make sure we've acked all the TCP data associated |
3616 | * with the data requests being unplugged */ | |
3617 | drbd_tcp_quickack(mdev->data.socket); | |
0ced55a3 | 3618 | |
81e84650 | 3619 | return true; |
0ced55a3 PR |
3620 | } |
3621 | ||
73a01a18 PR |
3622 | static int receive_out_of_sync(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
3623 | { | |
3624 | struct p_block_desc *p = &mdev->data.rbuf.block_desc; | |
3625 | ||
f735e363 LE |
3626 | switch (mdev->state.conn) { |
3627 | case C_WF_SYNC_UUID: | |
3628 | case C_WF_BITMAP_T: | |
3629 | case C_BEHIND: | |
3630 | break; | |
3631 | default: | |
3632 | dev_err(DEV, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n", | |
3633 | drbd_conn_str(mdev->state.conn)); | |
3634 | } | |
3635 | ||
73a01a18 PR |
3636 | drbd_set_out_of_sync(mdev, be64_to_cpu(p->sector), be32_to_cpu(p->blksize)); |
3637 | ||
81e84650 | 3638 | return true; |
73a01a18 PR |
3639 | } |
3640 | ||
02918be2 PR |
3641 | typedef int (*drbd_cmd_handler_f)(struct drbd_conf *, enum drbd_packets cmd, unsigned int to_receive); |
3642 | ||
3643 | struct data_cmd { | |
3644 | int expect_payload; | |
3645 | size_t pkt_size; | |
3646 | drbd_cmd_handler_f function; | |
3647 | }; | |
3648 | ||
3649 | static struct data_cmd drbd_cmd_handler[] = { | |
3650 | [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, | |
3651 | [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, | |
3652 | [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , | |
3653 | [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , | |
3654 | [P_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } , | |
3655 | [P_COMPRESSED_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } , | |
3656 | [P_UNPLUG_REMOTE] = { 0, sizeof(struct p_header80), receive_UnplugRemote }, | |
3657 | [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
3658 | [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
3659 | [P_SYNC_PARAM] = { 1, sizeof(struct p_header80), receive_SyncParam }, | |
3660 | [P_SYNC_PARAM89] = { 1, sizeof(struct p_header80), receive_SyncParam }, | |
3661 | [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, | |
3662 | [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, | |
3663 | [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, | |
3664 | [P_STATE] = { 0, sizeof(struct p_state), receive_state }, | |
3665 | [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, | |
3666 | [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, | |
3667 | [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
3668 | [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
3669 | [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
3670 | [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, | |
73a01a18 | 3671 | [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync }, |
b411b363 PR |
3672 | /* anything missing from this table is in |
3673 | * the asender_tbl, see get_asender_cmd */ | |
02918be2 | 3674 | [P_MAX_CMD] = { 0, 0, NULL }, |
b411b363 PR |
3675 | }; |
3676 | ||
02918be2 PR |
3677 | /* All handler functions that expect a sub-header get that sub-heder in |
3678 | mdev->data.rbuf.header.head.payload. | |
3679 | ||
3680 | Usually in mdev->data.rbuf.header.head the callback can find the usual | |
3681 | p_header, but they may not rely on that. Since there is also p_header95 ! | |
3682 | */ | |
b411b363 PR |
3683 | |
3684 | static void drbdd(struct drbd_conf *mdev) | |
3685 | { | |
02918be2 PR |
3686 | union p_header *header = &mdev->data.rbuf.header; |
3687 | unsigned int packet_size; | |
3688 | enum drbd_packets cmd; | |
3689 | size_t shs; /* sub header size */ | |
3690 | int rv; | |
b411b363 PR |
3691 | |
3692 | while (get_t_state(&mdev->receiver) == Running) { | |
3693 | drbd_thread_current_set_cpu(mdev); | |
02918be2 PR |
3694 | if (!drbd_recv_header(mdev, &cmd, &packet_size)) |
3695 | goto err_out; | |
b411b363 | 3696 | |
02918be2 PR |
3697 | if (unlikely(cmd >= P_MAX_CMD || !drbd_cmd_handler[cmd].function)) { |
3698 | dev_err(DEV, "unknown packet type %d, l: %d!\n", cmd, packet_size); | |
3699 | goto err_out; | |
0b33a916 | 3700 | } |
b411b363 | 3701 | |
02918be2 | 3702 | shs = drbd_cmd_handler[cmd].pkt_size - sizeof(union p_header); |
02918be2 PR |
3703 | if (packet_size - shs > 0 && !drbd_cmd_handler[cmd].expect_payload) { |
3704 | dev_err(DEV, "No payload expected %s l:%d\n", cmdname(cmd), packet_size); | |
3705 | goto err_out; | |
b411b363 | 3706 | } |
b411b363 | 3707 | |
c13f7e1a LE |
3708 | if (shs) { |
3709 | rv = drbd_recv(mdev, &header->h80.payload, shs); | |
3710 | if (unlikely(rv != shs)) { | |
0ddc5549 LE |
3711 | if (!signal_pending(current)) |
3712 | dev_warn(DEV, "short read while reading sub header: rv=%d\n", rv); | |
c13f7e1a LE |
3713 | goto err_out; |
3714 | } | |
3715 | } | |
3716 | ||
02918be2 | 3717 | rv = drbd_cmd_handler[cmd].function(mdev, cmd, packet_size - shs); |
b411b363 | 3718 | |
02918be2 | 3719 | if (unlikely(!rv)) { |
b411b363 | 3720 | dev_err(DEV, "error receiving %s, l: %d!\n", |
02918be2 PR |
3721 | cmdname(cmd), packet_size); |
3722 | goto err_out; | |
b411b363 PR |
3723 | } |
3724 | } | |
b411b363 | 3725 | |
02918be2 PR |
3726 | if (0) { |
3727 | err_out: | |
3728 | drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); | |
3729 | } | |
856c50c7 LE |
3730 | /* If we leave here, we probably want to update at least the |
3731 | * "Connected" indicator on stable storage. Do so explicitly here. */ | |
3732 | drbd_md_sync(mdev); | |
b411b363 PR |
3733 | } |
3734 | ||
3735 | void drbd_flush_workqueue(struct drbd_conf *mdev) | |
3736 | { | |
3737 | struct drbd_wq_barrier barr; | |
3738 | ||
3739 | barr.w.cb = w_prev_work_done; | |
3740 | init_completion(&barr.done); | |
3741 | drbd_queue_work(&mdev->data.work, &barr.w); | |
3742 | wait_for_completion(&barr.done); | |
3743 | } | |
3744 | ||
f70b3511 PR |
3745 | void drbd_free_tl_hash(struct drbd_conf *mdev) |
3746 | { | |
3747 | struct hlist_head *h; | |
3748 | ||
3749 | spin_lock_irq(&mdev->req_lock); | |
3750 | ||
3751 | if (!mdev->tl_hash || mdev->state.conn != C_STANDALONE) { | |
3752 | spin_unlock_irq(&mdev->req_lock); | |
3753 | return; | |
3754 | } | |
3755 | /* paranoia code */ | |
3756 | for (h = mdev->ee_hash; h < mdev->ee_hash + mdev->ee_hash_s; h++) | |
3757 | if (h->first) | |
3758 | dev_err(DEV, "ASSERT FAILED ee_hash[%u].first == %p, expected NULL\n", | |
3759 | (int)(h - mdev->ee_hash), h->first); | |
3760 | kfree(mdev->ee_hash); | |
3761 | mdev->ee_hash = NULL; | |
3762 | mdev->ee_hash_s = 0; | |
3763 | ||
3764 | /* paranoia code */ | |
3765 | for (h = mdev->tl_hash; h < mdev->tl_hash + mdev->tl_hash_s; h++) | |
3766 | if (h->first) | |
3767 | dev_err(DEV, "ASSERT FAILED tl_hash[%u] == %p, expected NULL\n", | |
3768 | (int)(h - mdev->tl_hash), h->first); | |
3769 | kfree(mdev->tl_hash); | |
3770 | mdev->tl_hash = NULL; | |
3771 | mdev->tl_hash_s = 0; | |
3772 | spin_unlock_irq(&mdev->req_lock); | |
3773 | } | |
3774 | ||
b411b363 PR |
3775 | static void drbd_disconnect(struct drbd_conf *mdev) |
3776 | { | |
3777 | enum drbd_fencing_p fp; | |
3778 | union drbd_state os, ns; | |
3779 | int rv = SS_UNKNOWN_ERROR; | |
3780 | unsigned int i; | |
3781 | ||
3782 | if (mdev->state.conn == C_STANDALONE) | |
3783 | return; | |
b411b363 PR |
3784 | |
3785 | /* asender does not clean up anything. it must not interfere, either */ | |
3786 | drbd_thread_stop(&mdev->asender); | |
b411b363 | 3787 | drbd_free_sock(mdev); |
b411b363 | 3788 | |
85719573 | 3789 | /* wait for current activity to cease. */ |
b411b363 PR |
3790 | spin_lock_irq(&mdev->req_lock); |
3791 | _drbd_wait_ee_list_empty(mdev, &mdev->active_ee); | |
3792 | _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee); | |
3793 | _drbd_wait_ee_list_empty(mdev, &mdev->read_ee); | |
3794 | spin_unlock_irq(&mdev->req_lock); | |
3795 | ||
3796 | /* We do not have data structures that would allow us to | |
3797 | * get the rs_pending_cnt down to 0 again. | |
3798 | * * On C_SYNC_TARGET we do not have any data structures describing | |
3799 | * the pending RSDataRequest's we have sent. | |
3800 | * * On C_SYNC_SOURCE there is no data structure that tracks | |
3801 | * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. | |
3802 | * And no, it is not the sum of the reference counts in the | |
3803 | * resync_LRU. The resync_LRU tracks the whole operation including | |
3804 | * the disk-IO, while the rs_pending_cnt only tracks the blocks | |
3805 | * on the fly. */ | |
3806 | drbd_rs_cancel_all(mdev); | |
3807 | mdev->rs_total = 0; | |
3808 | mdev->rs_failed = 0; | |
3809 | atomic_set(&mdev->rs_pending_cnt, 0); | |
3810 | wake_up(&mdev->misc_wait); | |
3811 | ||
3812 | /* make sure syncer is stopped and w_resume_next_sg queued */ | |
3813 | del_timer_sync(&mdev->resync_timer); | |
b411b363 PR |
3814 | resync_timer_fn((unsigned long)mdev); |
3815 | ||
b411b363 PR |
3816 | /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, |
3817 | * w_make_resync_request etc. which may still be on the worker queue | |
3818 | * to be "canceled" */ | |
3819 | drbd_flush_workqueue(mdev); | |
3820 | ||
3821 | /* This also does reclaim_net_ee(). If we do this too early, we might | |
3822 | * miss some resync ee and pages.*/ | |
3823 | drbd_process_done_ee(mdev); | |
3824 | ||
3825 | kfree(mdev->p_uuid); | |
3826 | mdev->p_uuid = NULL; | |
3827 | ||
fb22c402 | 3828 | if (!is_susp(mdev->state)) |
b411b363 PR |
3829 | tl_clear(mdev); |
3830 | ||
b411b363 PR |
3831 | dev_info(DEV, "Connection closed\n"); |
3832 | ||
3833 | drbd_md_sync(mdev); | |
3834 | ||
3835 | fp = FP_DONT_CARE; | |
3836 | if (get_ldev(mdev)) { | |
3837 | fp = mdev->ldev->dc.fencing; | |
3838 | put_ldev(mdev); | |
3839 | } | |
3840 | ||
87f7be4c PR |
3841 | if (mdev->state.role == R_PRIMARY && fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN) |
3842 | drbd_try_outdate_peer_async(mdev); | |
b411b363 PR |
3843 | |
3844 | spin_lock_irq(&mdev->req_lock); | |
3845 | os = mdev->state; | |
3846 | if (os.conn >= C_UNCONNECTED) { | |
3847 | /* Do not restart in case we are C_DISCONNECTING */ | |
3848 | ns = os; | |
3849 | ns.conn = C_UNCONNECTED; | |
3850 | rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); | |
3851 | } | |
3852 | spin_unlock_irq(&mdev->req_lock); | |
3853 | ||
3854 | if (os.conn == C_DISCONNECTING) { | |
84dfb9f5 | 3855 | wait_event(mdev->net_cnt_wait, atomic_read(&mdev->net_cnt) == 0); |
b411b363 | 3856 | |
b411b363 PR |
3857 | crypto_free_hash(mdev->cram_hmac_tfm); |
3858 | mdev->cram_hmac_tfm = NULL; | |
3859 | ||
3860 | kfree(mdev->net_conf); | |
3861 | mdev->net_conf = NULL; | |
3862 | drbd_request_state(mdev, NS(conn, C_STANDALONE)); | |
3863 | } | |
3864 | ||
20ceb2b2 LE |
3865 | /* serialize with bitmap writeout triggered by the state change, |
3866 | * if any. */ | |
3867 | wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags)); | |
3868 | ||
b411b363 PR |
3869 | /* tcp_close and release of sendpage pages can be deferred. I don't |
3870 | * want to use SO_LINGER, because apparently it can be deferred for | |
3871 | * more than 20 seconds (longest time I checked). | |
3872 | * | |
3873 | * Actually we don't care for exactly when the network stack does its | |
3874 | * put_page(), but release our reference on these pages right here. | |
3875 | */ | |
3876 | i = drbd_release_ee(mdev, &mdev->net_ee); | |
3877 | if (i) | |
3878 | dev_info(DEV, "net_ee not empty, killed %u entries\n", i); | |
435f0740 LE |
3879 | i = atomic_read(&mdev->pp_in_use_by_net); |
3880 | if (i) | |
3881 | dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i); | |
b411b363 PR |
3882 | i = atomic_read(&mdev->pp_in_use); |
3883 | if (i) | |
45bb912b | 3884 | dev_info(DEV, "pp_in_use = %d, expected 0\n", i); |
b411b363 PR |
3885 | |
3886 | D_ASSERT(list_empty(&mdev->read_ee)); | |
3887 | D_ASSERT(list_empty(&mdev->active_ee)); | |
3888 | D_ASSERT(list_empty(&mdev->sync_ee)); | |
3889 | D_ASSERT(list_empty(&mdev->done_ee)); | |
3890 | ||
3891 | /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ | |
3892 | atomic_set(&mdev->current_epoch->epoch_size, 0); | |
3893 | D_ASSERT(list_empty(&mdev->current_epoch->list)); | |
3894 | } | |
3895 | ||
3896 | /* | |
3897 | * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version | |
3898 | * we can agree on is stored in agreed_pro_version. | |
3899 | * | |
3900 | * feature flags and the reserved array should be enough room for future | |
3901 | * enhancements of the handshake protocol, and possible plugins... | |
3902 | * | |
3903 | * for now, they are expected to be zero, but ignored. | |
3904 | */ | |
3905 | static int drbd_send_handshake(struct drbd_conf *mdev) | |
3906 | { | |
3907 | /* ASSERT current == mdev->receiver ... */ | |
3908 | struct p_handshake *p = &mdev->data.sbuf.handshake; | |
3909 | int ok; | |
3910 | ||
3911 | if (mutex_lock_interruptible(&mdev->data.mutex)) { | |
3912 | dev_err(DEV, "interrupted during initial handshake\n"); | |
3913 | return 0; /* interrupted. not ok. */ | |
3914 | } | |
3915 | ||
3916 | if (mdev->data.socket == NULL) { | |
3917 | mutex_unlock(&mdev->data.mutex); | |
3918 | return 0; | |
3919 | } | |
3920 | ||
3921 | memset(p, 0, sizeof(*p)); | |
3922 | p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); | |
3923 | p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); | |
3924 | ok = _drbd_send_cmd( mdev, mdev->data.socket, P_HAND_SHAKE, | |
0b70a13d | 3925 | (struct p_header80 *)p, sizeof(*p), 0 ); |
b411b363 PR |
3926 | mutex_unlock(&mdev->data.mutex); |
3927 | return ok; | |
3928 | } | |
3929 | ||
3930 | /* | |
3931 | * return values: | |
3932 | * 1 yes, we have a valid connection | |
3933 | * 0 oops, did not work out, please try again | |
3934 | * -1 peer talks different language, | |
3935 | * no point in trying again, please go standalone. | |
3936 | */ | |
3937 | static int drbd_do_handshake(struct drbd_conf *mdev) | |
3938 | { | |
3939 | /* ASSERT current == mdev->receiver ... */ | |
3940 | struct p_handshake *p = &mdev->data.rbuf.handshake; | |
02918be2 PR |
3941 | const int expect = sizeof(struct p_handshake) - sizeof(struct p_header80); |
3942 | unsigned int length; | |
3943 | enum drbd_packets cmd; | |
b411b363 PR |
3944 | int rv; |
3945 | ||
3946 | rv = drbd_send_handshake(mdev); | |
3947 | if (!rv) | |
3948 | return 0; | |
3949 | ||
02918be2 | 3950 | rv = drbd_recv_header(mdev, &cmd, &length); |
b411b363 PR |
3951 | if (!rv) |
3952 | return 0; | |
3953 | ||
02918be2 | 3954 | if (cmd != P_HAND_SHAKE) { |
b411b363 | 3955 | dev_err(DEV, "expected HandShake packet, received: %s (0x%04x)\n", |
02918be2 | 3956 | cmdname(cmd), cmd); |
b411b363 PR |
3957 | return -1; |
3958 | } | |
3959 | ||
02918be2 | 3960 | if (length != expect) { |
b411b363 | 3961 | dev_err(DEV, "expected HandShake length: %u, received: %u\n", |
02918be2 | 3962 | expect, length); |
b411b363 PR |
3963 | return -1; |
3964 | } | |
3965 | ||
3966 | rv = drbd_recv(mdev, &p->head.payload, expect); | |
3967 | ||
3968 | if (rv != expect) { | |
0ddc5549 LE |
3969 | if (!signal_pending(current)) |
3970 | dev_warn(DEV, "short read receiving handshake packet: l=%u\n", rv); | |
b411b363 PR |
3971 | return 0; |
3972 | } | |
3973 | ||
b411b363 PR |
3974 | p->protocol_min = be32_to_cpu(p->protocol_min); |
3975 | p->protocol_max = be32_to_cpu(p->protocol_max); | |
3976 | if (p->protocol_max == 0) | |
3977 | p->protocol_max = p->protocol_min; | |
3978 | ||
3979 | if (PRO_VERSION_MAX < p->protocol_min || | |
3980 | PRO_VERSION_MIN > p->protocol_max) | |
3981 | goto incompat; | |
3982 | ||
3983 | mdev->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); | |
3984 | ||
3985 | dev_info(DEV, "Handshake successful: " | |
3986 | "Agreed network protocol version %d\n", mdev->agreed_pro_version); | |
3987 | ||
3988 | return 1; | |
3989 | ||
3990 | incompat: | |
3991 | dev_err(DEV, "incompatible DRBD dialects: " | |
3992 | "I support %d-%d, peer supports %d-%d\n", | |
3993 | PRO_VERSION_MIN, PRO_VERSION_MAX, | |
3994 | p->protocol_min, p->protocol_max); | |
3995 | return -1; | |
3996 | } | |
3997 | ||
3998 | #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) | |
3999 | static int drbd_do_auth(struct drbd_conf *mdev) | |
4000 | { | |
4001 | dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); | |
4002 | dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); | |
b10d96cb | 4003 | return -1; |
b411b363 PR |
4004 | } |
4005 | #else | |
4006 | #define CHALLENGE_LEN 64 | |
b10d96cb JT |
4007 | |
4008 | /* Return value: | |
4009 | 1 - auth succeeded, | |
4010 | 0 - failed, try again (network error), | |
4011 | -1 - auth failed, don't try again. | |
4012 | */ | |
4013 | ||
b411b363 PR |
4014 | static int drbd_do_auth(struct drbd_conf *mdev) |
4015 | { | |
4016 | char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ | |
4017 | struct scatterlist sg; | |
4018 | char *response = NULL; | |
4019 | char *right_response = NULL; | |
4020 | char *peers_ch = NULL; | |
b411b363 PR |
4021 | unsigned int key_len = strlen(mdev->net_conf->shared_secret); |
4022 | unsigned int resp_size; | |
4023 | struct hash_desc desc; | |
02918be2 PR |
4024 | enum drbd_packets cmd; |
4025 | unsigned int length; | |
b411b363 PR |
4026 | int rv; |
4027 | ||
4028 | desc.tfm = mdev->cram_hmac_tfm; | |
4029 | desc.flags = 0; | |
4030 | ||
4031 | rv = crypto_hash_setkey(mdev->cram_hmac_tfm, | |
4032 | (u8 *)mdev->net_conf->shared_secret, key_len); | |
4033 | if (rv) { | |
4034 | dev_err(DEV, "crypto_hash_setkey() failed with %d\n", rv); | |
b10d96cb | 4035 | rv = -1; |
b411b363 PR |
4036 | goto fail; |
4037 | } | |
4038 | ||
4039 | get_random_bytes(my_challenge, CHALLENGE_LEN); | |
4040 | ||
4041 | rv = drbd_send_cmd2(mdev, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN); | |
4042 | if (!rv) | |
4043 | goto fail; | |
4044 | ||
02918be2 | 4045 | rv = drbd_recv_header(mdev, &cmd, &length); |
b411b363 PR |
4046 | if (!rv) |
4047 | goto fail; | |
4048 | ||
02918be2 | 4049 | if (cmd != P_AUTH_CHALLENGE) { |
b411b363 | 4050 | dev_err(DEV, "expected AuthChallenge packet, received: %s (0x%04x)\n", |
02918be2 | 4051 | cmdname(cmd), cmd); |
b411b363 PR |
4052 | rv = 0; |
4053 | goto fail; | |
4054 | } | |
4055 | ||
02918be2 | 4056 | if (length > CHALLENGE_LEN * 2) { |
b411b363 | 4057 | dev_err(DEV, "expected AuthChallenge payload too big.\n"); |
b10d96cb | 4058 | rv = -1; |
b411b363 PR |
4059 | goto fail; |
4060 | } | |
4061 | ||
02918be2 | 4062 | peers_ch = kmalloc(length, GFP_NOIO); |
b411b363 PR |
4063 | if (peers_ch == NULL) { |
4064 | dev_err(DEV, "kmalloc of peers_ch failed\n"); | |
b10d96cb | 4065 | rv = -1; |
b411b363 PR |
4066 | goto fail; |
4067 | } | |
4068 | ||
02918be2 | 4069 | rv = drbd_recv(mdev, peers_ch, length); |
b411b363 | 4070 | |
02918be2 | 4071 | if (rv != length) { |
0ddc5549 LE |
4072 | if (!signal_pending(current)) |
4073 | dev_warn(DEV, "short read AuthChallenge: l=%u\n", rv); | |
b411b363 PR |
4074 | rv = 0; |
4075 | goto fail; | |
4076 | } | |
4077 | ||
4078 | resp_size = crypto_hash_digestsize(mdev->cram_hmac_tfm); | |
4079 | response = kmalloc(resp_size, GFP_NOIO); | |
4080 | if (response == NULL) { | |
4081 | dev_err(DEV, "kmalloc of response failed\n"); | |
b10d96cb | 4082 | rv = -1; |
b411b363 PR |
4083 | goto fail; |
4084 | } | |
4085 | ||
4086 | sg_init_table(&sg, 1); | |
02918be2 | 4087 | sg_set_buf(&sg, peers_ch, length); |
b411b363 PR |
4088 | |
4089 | rv = crypto_hash_digest(&desc, &sg, sg.length, response); | |
4090 | if (rv) { | |
4091 | dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv); | |
b10d96cb | 4092 | rv = -1; |
b411b363 PR |
4093 | goto fail; |
4094 | } | |
4095 | ||
4096 | rv = drbd_send_cmd2(mdev, P_AUTH_RESPONSE, response, resp_size); | |
4097 | if (!rv) | |
4098 | goto fail; | |
4099 | ||
02918be2 | 4100 | rv = drbd_recv_header(mdev, &cmd, &length); |
b411b363 PR |
4101 | if (!rv) |
4102 | goto fail; | |
4103 | ||
02918be2 | 4104 | if (cmd != P_AUTH_RESPONSE) { |
b411b363 | 4105 | dev_err(DEV, "expected AuthResponse packet, received: %s (0x%04x)\n", |
02918be2 | 4106 | cmdname(cmd), cmd); |
b411b363 PR |
4107 | rv = 0; |
4108 | goto fail; | |
4109 | } | |
4110 | ||
02918be2 | 4111 | if (length != resp_size) { |
b411b363 PR |
4112 | dev_err(DEV, "expected AuthResponse payload of wrong size\n"); |
4113 | rv = 0; | |
4114 | goto fail; | |
4115 | } | |
4116 | ||
4117 | rv = drbd_recv(mdev, response , resp_size); | |
4118 | ||
4119 | if (rv != resp_size) { | |
0ddc5549 LE |
4120 | if (!signal_pending(current)) |
4121 | dev_warn(DEV, "short read receiving AuthResponse: l=%u\n", rv); | |
b411b363 PR |
4122 | rv = 0; |
4123 | goto fail; | |
4124 | } | |
4125 | ||
4126 | right_response = kmalloc(resp_size, GFP_NOIO); | |
2d1ee87d | 4127 | if (right_response == NULL) { |
b411b363 | 4128 | dev_err(DEV, "kmalloc of right_response failed\n"); |
b10d96cb | 4129 | rv = -1; |
b411b363 PR |
4130 | goto fail; |
4131 | } | |
4132 | ||
4133 | sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); | |
4134 | ||
4135 | rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); | |
4136 | if (rv) { | |
4137 | dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv); | |
b10d96cb | 4138 | rv = -1; |
b411b363 PR |
4139 | goto fail; |
4140 | } | |
4141 | ||
4142 | rv = !memcmp(response, right_response, resp_size); | |
4143 | ||
4144 | if (rv) | |
4145 | dev_info(DEV, "Peer authenticated using %d bytes of '%s' HMAC\n", | |
4146 | resp_size, mdev->net_conf->cram_hmac_alg); | |
b10d96cb JT |
4147 | else |
4148 | rv = -1; | |
b411b363 PR |
4149 | |
4150 | fail: | |
4151 | kfree(peers_ch); | |
4152 | kfree(response); | |
4153 | kfree(right_response); | |
4154 | ||
4155 | return rv; | |
4156 | } | |
4157 | #endif | |
4158 | ||
4159 | int drbdd_init(struct drbd_thread *thi) | |
4160 | { | |
4161 | struct drbd_conf *mdev = thi->mdev; | |
4162 | unsigned int minor = mdev_to_minor(mdev); | |
4163 | int h; | |
4164 | ||
4165 | sprintf(current->comm, "drbd%d_receiver", minor); | |
4166 | ||
4167 | dev_info(DEV, "receiver (re)started\n"); | |
4168 | ||
4169 | do { | |
4170 | h = drbd_connect(mdev); | |
4171 | if (h == 0) { | |
4172 | drbd_disconnect(mdev); | |
20ee6390 | 4173 | schedule_timeout_interruptible(HZ); |
b411b363 PR |
4174 | } |
4175 | if (h == -1) { | |
4176 | dev_warn(DEV, "Discarding network configuration.\n"); | |
4177 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
4178 | } | |
4179 | } while (h == 0); | |
4180 | ||
4181 | if (h > 0) { | |
4182 | if (get_net_conf(mdev)) { | |
4183 | drbdd(mdev); | |
4184 | put_net_conf(mdev); | |
4185 | } | |
4186 | } | |
4187 | ||
4188 | drbd_disconnect(mdev); | |
4189 | ||
4190 | dev_info(DEV, "receiver terminated\n"); | |
4191 | return 0; | |
4192 | } | |
4193 | ||
4194 | /* ********* acknowledge sender ******** */ | |
4195 | ||
0b70a13d | 4196 | static int got_RqSReply(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4197 | { |
4198 | struct p_req_state_reply *p = (struct p_req_state_reply *)h; | |
4199 | ||
4200 | int retcode = be32_to_cpu(p->retcode); | |
4201 | ||
4202 | if (retcode >= SS_SUCCESS) { | |
4203 | set_bit(CL_ST_CHG_SUCCESS, &mdev->flags); | |
4204 | } else { | |
4205 | set_bit(CL_ST_CHG_FAIL, &mdev->flags); | |
4206 | dev_err(DEV, "Requested state change failed by peer: %s (%d)\n", | |
4207 | drbd_set_st_err_str(retcode), retcode); | |
4208 | } | |
4209 | wake_up(&mdev->state_wait); | |
4210 | ||
81e84650 | 4211 | return true; |
b411b363 PR |
4212 | } |
4213 | ||
0b70a13d | 4214 | static int got_Ping(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4215 | { |
4216 | return drbd_send_ping_ack(mdev); | |
4217 | ||
4218 | } | |
4219 | ||
0b70a13d | 4220 | static int got_PingAck(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4221 | { |
4222 | /* restore idle timeout */ | |
4223 | mdev->meta.socket->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ; | |
309d1608 PR |
4224 | if (!test_and_set_bit(GOT_PING_ACK, &mdev->flags)) |
4225 | wake_up(&mdev->misc_wait); | |
b411b363 | 4226 | |
81e84650 | 4227 | return true; |
b411b363 PR |
4228 | } |
4229 | ||
0b70a13d | 4230 | static int got_IsInSync(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4231 | { |
4232 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4233 | sector_t sector = be64_to_cpu(p->sector); | |
4234 | int blksize = be32_to_cpu(p->blksize); | |
4235 | ||
4236 | D_ASSERT(mdev->agreed_pro_version >= 89); | |
4237 | ||
4238 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4239 | ||
1d53f09e LE |
4240 | if (get_ldev(mdev)) { |
4241 | drbd_rs_complete_io(mdev, sector); | |
4242 | drbd_set_in_sync(mdev, sector, blksize); | |
4243 | /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ | |
4244 | mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); | |
4245 | put_ldev(mdev); | |
4246 | } | |
b411b363 | 4247 | dec_rs_pending(mdev); |
778f271d | 4248 | atomic_add(blksize >> 9, &mdev->rs_sect_in); |
b411b363 | 4249 | |
81e84650 | 4250 | return true; |
b411b363 PR |
4251 | } |
4252 | ||
4253 | /* when we receive the ACK for a write request, | |
4254 | * verify that we actually know about it */ | |
4255 | static struct drbd_request *_ack_id_to_req(struct drbd_conf *mdev, | |
4256 | u64 id, sector_t sector) | |
4257 | { | |
4258 | struct hlist_head *slot = tl_hash_slot(mdev, sector); | |
4259 | struct hlist_node *n; | |
4260 | struct drbd_request *req; | |
4261 | ||
24c4830c | 4262 | hlist_for_each_entry(req, n, slot, collision) { |
b411b363 PR |
4263 | if ((unsigned long)req == (unsigned long)id) { |
4264 | if (req->sector != sector) { | |
4265 | dev_err(DEV, "_ack_id_to_req: found req %p but it has " | |
4266 | "wrong sector (%llus versus %llus)\n", req, | |
4267 | (unsigned long long)req->sector, | |
4268 | (unsigned long long)sector); | |
4269 | break; | |
4270 | } | |
4271 | return req; | |
4272 | } | |
4273 | } | |
b411b363 PR |
4274 | return NULL; |
4275 | } | |
4276 | ||
4277 | typedef struct drbd_request *(req_validator_fn) | |
4278 | (struct drbd_conf *mdev, u64 id, sector_t sector); | |
4279 | ||
4280 | static int validate_req_change_req_state(struct drbd_conf *mdev, | |
4281 | u64 id, sector_t sector, req_validator_fn validator, | |
4282 | const char *func, enum drbd_req_event what) | |
4283 | { | |
4284 | struct drbd_request *req; | |
4285 | struct bio_and_error m; | |
4286 | ||
4287 | spin_lock_irq(&mdev->req_lock); | |
4288 | req = validator(mdev, id, sector); | |
4289 | if (unlikely(!req)) { | |
4290 | spin_unlock_irq(&mdev->req_lock); | |
2deb8336 PR |
4291 | |
4292 | dev_err(DEV, "%s: failed to find req %p, sector %llus\n", func, | |
4293 | (void *)(unsigned long)id, (unsigned long long)sector); | |
81e84650 | 4294 | return false; |
b411b363 PR |
4295 | } |
4296 | __req_mod(req, what, &m); | |
4297 | spin_unlock_irq(&mdev->req_lock); | |
4298 | ||
4299 | if (m.bio) | |
4300 | complete_master_bio(mdev, &m); | |
81e84650 | 4301 | return true; |
b411b363 PR |
4302 | } |
4303 | ||
0b70a13d | 4304 | static int got_BlockAck(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4305 | { |
4306 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4307 | sector_t sector = be64_to_cpu(p->sector); | |
4308 | int blksize = be32_to_cpu(p->blksize); | |
4309 | enum drbd_req_event what; | |
4310 | ||
4311 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4312 | ||
4313 | if (is_syncer_block_id(p->block_id)) { | |
4314 | drbd_set_in_sync(mdev, sector, blksize); | |
4315 | dec_rs_pending(mdev); | |
81e84650 | 4316 | return true; |
b411b363 PR |
4317 | } |
4318 | switch (be16_to_cpu(h->command)) { | |
4319 | case P_RS_WRITE_ACK: | |
4320 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
4321 | what = write_acked_by_peer_and_sis; | |
4322 | break; | |
4323 | case P_WRITE_ACK: | |
4324 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
4325 | what = write_acked_by_peer; | |
4326 | break; | |
4327 | case P_RECV_ACK: | |
4328 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_B); | |
4329 | what = recv_acked_by_peer; | |
4330 | break; | |
4331 | case P_DISCARD_ACK: | |
4332 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
4333 | what = conflict_discarded_by_peer; | |
4334 | break; | |
4335 | default: | |
4336 | D_ASSERT(0); | |
81e84650 | 4337 | return false; |
b411b363 PR |
4338 | } |
4339 | ||
4340 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
4341 | _ack_id_to_req, __func__ , what); | |
4342 | } | |
4343 | ||
0b70a13d | 4344 | static int got_NegAck(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4345 | { |
4346 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4347 | sector_t sector = be64_to_cpu(p->sector); | |
2deb8336 PR |
4348 | int size = be32_to_cpu(p->blksize); |
4349 | struct drbd_request *req; | |
4350 | struct bio_and_error m; | |
b411b363 PR |
4351 | |
4352 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4353 | ||
4354 | if (is_syncer_block_id(p->block_id)) { | |
b411b363 PR |
4355 | dec_rs_pending(mdev); |
4356 | drbd_rs_failed_io(mdev, sector, size); | |
81e84650 | 4357 | return true; |
b411b363 | 4358 | } |
2deb8336 PR |
4359 | |
4360 | spin_lock_irq(&mdev->req_lock); | |
4361 | req = _ack_id_to_req(mdev, p->block_id, sector); | |
4362 | if (!req) { | |
4363 | spin_unlock_irq(&mdev->req_lock); | |
4364 | if (mdev->net_conf->wire_protocol == DRBD_PROT_A || | |
4365 | mdev->net_conf->wire_protocol == DRBD_PROT_B) { | |
4366 | /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs. | |
4367 | The master bio might already be completed, therefore the | |
4368 | request is no longer in the collision hash. | |
4369 | => Do not try to validate block_id as request. */ | |
4370 | /* In Protocol B we might already have got a P_RECV_ACK | |
4371 | but then get a P_NEG_ACK after wards. */ | |
4372 | drbd_set_out_of_sync(mdev, sector, size); | |
4373 | return true; | |
4374 | } else { | |
4375 | dev_err(DEV, "%s: failed to find req %p, sector %llus\n", __func__, | |
4376 | (void *)(unsigned long)p->block_id, (unsigned long long)sector); | |
4377 | return false; | |
4378 | } | |
4379 | } | |
4380 | __req_mod(req, neg_acked, &m); | |
4381 | spin_unlock_irq(&mdev->req_lock); | |
4382 | ||
4383 | if (m.bio) | |
4384 | complete_master_bio(mdev, &m); | |
4385 | return true; | |
b411b363 PR |
4386 | } |
4387 | ||
0b70a13d | 4388 | static int got_NegDReply(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4389 | { |
4390 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4391 | sector_t sector = be64_to_cpu(p->sector); | |
4392 | ||
4393 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4394 | dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n", | |
4395 | (unsigned long long)sector, be32_to_cpu(p->blksize)); | |
4396 | ||
4397 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
4398 | _ar_id_to_req, __func__ , neg_acked); | |
4399 | } | |
4400 | ||
0b70a13d | 4401 | static int got_NegRSDReply(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4402 | { |
4403 | sector_t sector; | |
4404 | int size; | |
4405 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4406 | ||
4407 | sector = be64_to_cpu(p->sector); | |
4408 | size = be32_to_cpu(p->blksize); | |
b411b363 PR |
4409 | |
4410 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4411 | ||
4412 | dec_rs_pending(mdev); | |
4413 | ||
4414 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
4415 | drbd_rs_complete_io(mdev, sector); | |
d612d309 PR |
4416 | switch (be16_to_cpu(h->command)) { |
4417 | case P_NEG_RS_DREPLY: | |
4418 | drbd_rs_failed_io(mdev, sector, size); | |
4419 | case P_RS_CANCEL: | |
4420 | break; | |
4421 | default: | |
4422 | D_ASSERT(0); | |
4423 | put_ldev(mdev); | |
4424 | return false; | |
4425 | } | |
b411b363 PR |
4426 | put_ldev(mdev); |
4427 | } | |
4428 | ||
81e84650 | 4429 | return true; |
b411b363 PR |
4430 | } |
4431 | ||
0b70a13d | 4432 | static int got_BarrierAck(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4433 | { |
4434 | struct p_barrier_ack *p = (struct p_barrier_ack *)h; | |
4435 | ||
4436 | tl_release(mdev, p->barrier, be32_to_cpu(p->set_size)); | |
4437 | ||
c4752ef1 PR |
4438 | if (mdev->state.conn == C_AHEAD && |
4439 | atomic_read(&mdev->ap_in_flight) == 0 && | |
370a43e7 PR |
4440 | !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &mdev->current_epoch->flags)) { |
4441 | mdev->start_resync_timer.expires = jiffies + HZ; | |
4442 | add_timer(&mdev->start_resync_timer); | |
c4752ef1 PR |
4443 | } |
4444 | ||
81e84650 | 4445 | return true; |
b411b363 PR |
4446 | } |
4447 | ||
0b70a13d | 4448 | static int got_OVResult(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4449 | { |
4450 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4451 | struct drbd_work *w; | |
4452 | sector_t sector; | |
4453 | int size; | |
4454 | ||
4455 | sector = be64_to_cpu(p->sector); | |
4456 | size = be32_to_cpu(p->blksize); | |
4457 | ||
4458 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4459 | ||
4460 | if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) | |
4461 | drbd_ov_oos_found(mdev, sector, size); | |
4462 | else | |
4463 | ov_oos_print(mdev); | |
4464 | ||
1d53f09e | 4465 | if (!get_ldev(mdev)) |
81e84650 | 4466 | return true; |
1d53f09e | 4467 | |
b411b363 PR |
4468 | drbd_rs_complete_io(mdev, sector); |
4469 | dec_rs_pending(mdev); | |
4470 | ||
ea5442af LE |
4471 | --mdev->ov_left; |
4472 | ||
4473 | /* let's advance progress step marks only for every other megabyte */ | |
4474 | if ((mdev->ov_left & 0x200) == 0x200) | |
4475 | drbd_advance_rs_marks(mdev, mdev->ov_left); | |
4476 | ||
4477 | if (mdev->ov_left == 0) { | |
b411b363 PR |
4478 | w = kmalloc(sizeof(*w), GFP_NOIO); |
4479 | if (w) { | |
4480 | w->cb = w_ov_finished; | |
4481 | drbd_queue_work_front(&mdev->data.work, w); | |
4482 | } else { | |
4483 | dev_err(DEV, "kmalloc(w) failed."); | |
4484 | ov_oos_print(mdev); | |
4485 | drbd_resync_finished(mdev); | |
4486 | } | |
4487 | } | |
1d53f09e | 4488 | put_ldev(mdev); |
81e84650 | 4489 | return true; |
b411b363 PR |
4490 | } |
4491 | ||
02918be2 | 4492 | static int got_skip(struct drbd_conf *mdev, struct p_header80 *h) |
0ced55a3 | 4493 | { |
81e84650 | 4494 | return true; |
0ced55a3 PR |
4495 | } |
4496 | ||
b411b363 PR |
4497 | struct asender_cmd { |
4498 | size_t pkt_size; | |
0b70a13d | 4499 | int (*process)(struct drbd_conf *mdev, struct p_header80 *h); |
b411b363 PR |
4500 | }; |
4501 | ||
4502 | static struct asender_cmd *get_asender_cmd(int cmd) | |
4503 | { | |
4504 | static struct asender_cmd asender_tbl[] = { | |
4505 | /* anything missing from this table is in | |
4506 | * the drbd_cmd_handler (drbd_default_handler) table, | |
4507 | * see the beginning of drbdd() */ | |
0b70a13d PR |
4508 | [P_PING] = { sizeof(struct p_header80), got_Ping }, |
4509 | [P_PING_ACK] = { sizeof(struct p_header80), got_PingAck }, | |
b411b363 PR |
4510 | [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, |
4511 | [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4512 | [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4513 | [P_DISCARD_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4514 | [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, | |
4515 | [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, | |
4516 | [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply}, | |
4517 | [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, | |
4518 | [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, | |
4519 | [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, | |
4520 | [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, | |
02918be2 | 4521 | [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, |
d612d309 | 4522 | [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply}, |
b411b363 PR |
4523 | [P_MAX_CMD] = { 0, NULL }, |
4524 | }; | |
4525 | if (cmd > P_MAX_CMD || asender_tbl[cmd].process == NULL) | |
4526 | return NULL; | |
4527 | return &asender_tbl[cmd]; | |
4528 | } | |
4529 | ||
4530 | int drbd_asender(struct drbd_thread *thi) | |
4531 | { | |
4532 | struct drbd_conf *mdev = thi->mdev; | |
02918be2 | 4533 | struct p_header80 *h = &mdev->meta.rbuf.header.h80; |
b411b363 PR |
4534 | struct asender_cmd *cmd = NULL; |
4535 | ||
4536 | int rv, len; | |
4537 | void *buf = h; | |
4538 | int received = 0; | |
0b70a13d | 4539 | int expect = sizeof(struct p_header80); |
b411b363 | 4540 | int empty; |
f36af18c | 4541 | int ping_timeout_active = 0; |
b411b363 PR |
4542 | |
4543 | sprintf(current->comm, "drbd%d_asender", mdev_to_minor(mdev)); | |
4544 | ||
4545 | current->policy = SCHED_RR; /* Make this a realtime task! */ | |
4546 | current->rt_priority = 2; /* more important than all other tasks */ | |
4547 | ||
4548 | while (get_t_state(thi) == Running) { | |
4549 | drbd_thread_current_set_cpu(mdev); | |
4550 | if (test_and_clear_bit(SEND_PING, &mdev->flags)) { | |
4551 | ERR_IF(!drbd_send_ping(mdev)) goto reconnect; | |
4552 | mdev->meta.socket->sk->sk_rcvtimeo = | |
4553 | mdev->net_conf->ping_timeo*HZ/10; | |
f36af18c | 4554 | ping_timeout_active = 1; |
b411b363 PR |
4555 | } |
4556 | ||
4557 | /* conditionally cork; | |
4558 | * it may hurt latency if we cork without much to send */ | |
4559 | if (!mdev->net_conf->no_cork && | |
4560 | 3 < atomic_read(&mdev->unacked_cnt)) | |
4561 | drbd_tcp_cork(mdev->meta.socket); | |
4562 | while (1) { | |
4563 | clear_bit(SIGNAL_ASENDER, &mdev->flags); | |
4564 | flush_signals(current); | |
0f8488e1 | 4565 | if (!drbd_process_done_ee(mdev)) |
b411b363 | 4566 | goto reconnect; |
b411b363 PR |
4567 | /* to avoid race with newly queued ACKs */ |
4568 | set_bit(SIGNAL_ASENDER, &mdev->flags); | |
4569 | spin_lock_irq(&mdev->req_lock); | |
4570 | empty = list_empty(&mdev->done_ee); | |
4571 | spin_unlock_irq(&mdev->req_lock); | |
4572 | /* new ack may have been queued right here, | |
4573 | * but then there is also a signal pending, | |
4574 | * and we start over... */ | |
4575 | if (empty) | |
4576 | break; | |
4577 | } | |
4578 | /* but unconditionally uncork unless disabled */ | |
4579 | if (!mdev->net_conf->no_cork) | |
4580 | drbd_tcp_uncork(mdev->meta.socket); | |
4581 | ||
4582 | /* short circuit, recv_msg would return EINTR anyways. */ | |
4583 | if (signal_pending(current)) | |
4584 | continue; | |
4585 | ||
4586 | rv = drbd_recv_short(mdev, mdev->meta.socket, | |
4587 | buf, expect-received, 0); | |
4588 | clear_bit(SIGNAL_ASENDER, &mdev->flags); | |
4589 | ||
4590 | flush_signals(current); | |
4591 | ||
4592 | /* Note: | |
4593 | * -EINTR (on meta) we got a signal | |
4594 | * -EAGAIN (on meta) rcvtimeo expired | |
4595 | * -ECONNRESET other side closed the connection | |
4596 | * -ERESTARTSYS (on data) we got a signal | |
4597 | * rv < 0 other than above: unexpected error! | |
4598 | * rv == expected: full header or command | |
4599 | * rv < expected: "woken" by signal during receive | |
4600 | * rv == 0 : "connection shut down by peer" | |
4601 | */ | |
4602 | if (likely(rv > 0)) { | |
4603 | received += rv; | |
4604 | buf += rv; | |
4605 | } else if (rv == 0) { | |
4606 | dev_err(DEV, "meta connection shut down by peer.\n"); | |
4607 | goto reconnect; | |
4608 | } else if (rv == -EAGAIN) { | |
cb6518cb LE |
4609 | /* If the data socket received something meanwhile, |
4610 | * that is good enough: peer is still alive. */ | |
4611 | if (time_after(mdev->last_received, | |
4612 | jiffies - mdev->meta.socket->sk->sk_rcvtimeo)) | |
4613 | continue; | |
f36af18c | 4614 | if (ping_timeout_active) { |
b411b363 PR |
4615 | dev_err(DEV, "PingAck did not arrive in time.\n"); |
4616 | goto reconnect; | |
4617 | } | |
4618 | set_bit(SEND_PING, &mdev->flags); | |
4619 | continue; | |
4620 | } else if (rv == -EINTR) { | |
4621 | continue; | |
4622 | } else { | |
4623 | dev_err(DEV, "sock_recvmsg returned %d\n", rv); | |
4624 | goto reconnect; | |
4625 | } | |
4626 | ||
4627 | if (received == expect && cmd == NULL) { | |
4628 | if (unlikely(h->magic != BE_DRBD_MAGIC)) { | |
004352fa LE |
4629 | dev_err(DEV, "magic?? on meta m: 0x%08x c: %d l: %d\n", |
4630 | be32_to_cpu(h->magic), | |
4631 | be16_to_cpu(h->command), | |
4632 | be16_to_cpu(h->length)); | |
b411b363 PR |
4633 | goto reconnect; |
4634 | } | |
4635 | cmd = get_asender_cmd(be16_to_cpu(h->command)); | |
4636 | len = be16_to_cpu(h->length); | |
4637 | if (unlikely(cmd == NULL)) { | |
004352fa LE |
4638 | dev_err(DEV, "unknown command?? on meta m: 0x%08x c: %d l: %d\n", |
4639 | be32_to_cpu(h->magic), | |
4640 | be16_to_cpu(h->command), | |
4641 | be16_to_cpu(h->length)); | |
b411b363 PR |
4642 | goto disconnect; |
4643 | } | |
4644 | expect = cmd->pkt_size; | |
0b70a13d | 4645 | ERR_IF(len != expect-sizeof(struct p_header80)) |
b411b363 | 4646 | goto reconnect; |
b411b363 PR |
4647 | } |
4648 | if (received == expect) { | |
cb6518cb | 4649 | mdev->last_received = jiffies; |
b411b363 | 4650 | D_ASSERT(cmd != NULL); |
b411b363 PR |
4651 | if (!cmd->process(mdev, h)) |
4652 | goto reconnect; | |
4653 | ||
f36af18c LE |
4654 | /* the idle_timeout (ping-int) |
4655 | * has been restored in got_PingAck() */ | |
4656 | if (cmd == get_asender_cmd(P_PING_ACK)) | |
4657 | ping_timeout_active = 0; | |
4658 | ||
b411b363 PR |
4659 | buf = h; |
4660 | received = 0; | |
0b70a13d | 4661 | expect = sizeof(struct p_header80); |
b411b363 PR |
4662 | cmd = NULL; |
4663 | } | |
4664 | } | |
4665 | ||
4666 | if (0) { | |
4667 | reconnect: | |
4668 | drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE)); | |
856c50c7 | 4669 | drbd_md_sync(mdev); |
b411b363 PR |
4670 | } |
4671 | if (0) { | |
4672 | disconnect: | |
4673 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
856c50c7 | 4674 | drbd_md_sync(mdev); |
b411b363 PR |
4675 | } |
4676 | clear_bit(SIGNAL_ASENDER, &mdev->flags); | |
4677 | ||
4678 | D_ASSERT(mdev->state.conn < C_CONNECTED); | |
4679 | dev_info(DEV, "asender terminated\n"); | |
4680 | ||
4681 | return 0; | |
4682 | } |