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> | |
39 | #include <linux/smp_lock.h> | |
40 | #include <linux/pkt_sched.h> | |
41 | #define __KERNEL_SYSCALLS__ | |
42 | #include <linux/unistd.h> | |
43 | #include <linux/vmalloc.h> | |
44 | #include <linux/random.h> | |
b411b363 PR |
45 | #include <linux/string.h> |
46 | #include <linux/scatterlist.h> | |
47 | #include "drbd_int.h" | |
b411b363 PR |
48 | #include "drbd_req.h" |
49 | ||
50 | #include "drbd_vli.h" | |
51 | ||
52 | struct flush_work { | |
53 | struct drbd_work w; | |
54 | struct drbd_epoch *epoch; | |
55 | }; | |
56 | ||
57 | enum finish_epoch { | |
58 | FE_STILL_LIVE, | |
59 | FE_DESTROYED, | |
60 | FE_RECYCLED, | |
61 | }; | |
62 | ||
63 | static int drbd_do_handshake(struct drbd_conf *mdev); | |
64 | static int drbd_do_auth(struct drbd_conf *mdev); | |
65 | ||
66 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *, struct drbd_epoch *, enum epoch_event); | |
67 | static int e_end_block(struct drbd_conf *, struct drbd_work *, int); | |
68 | ||
69 | static struct drbd_epoch *previous_epoch(struct drbd_conf *mdev, struct drbd_epoch *epoch) | |
70 | { | |
71 | struct drbd_epoch *prev; | |
72 | spin_lock(&mdev->epoch_lock); | |
73 | prev = list_entry(epoch->list.prev, struct drbd_epoch, list); | |
74 | if (prev == epoch || prev == mdev->current_epoch) | |
75 | prev = NULL; | |
76 | spin_unlock(&mdev->epoch_lock); | |
77 | return prev; | |
78 | } | |
79 | ||
80 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
81 | ||
45bb912b LE |
82 | /* |
83 | * some helper functions to deal with single linked page lists, | |
84 | * page->private being our "next" pointer. | |
85 | */ | |
86 | ||
87 | /* If at least n pages are linked at head, get n pages off. | |
88 | * Otherwise, don't modify head, and return NULL. | |
89 | * Locking is the responsibility of the caller. | |
90 | */ | |
91 | static struct page *page_chain_del(struct page **head, int n) | |
92 | { | |
93 | struct page *page; | |
94 | struct page *tmp; | |
95 | ||
96 | BUG_ON(!n); | |
97 | BUG_ON(!head); | |
98 | ||
99 | page = *head; | |
23ce4227 PR |
100 | |
101 | if (!page) | |
102 | return NULL; | |
103 | ||
45bb912b LE |
104 | while (page) { |
105 | tmp = page_chain_next(page); | |
106 | if (--n == 0) | |
107 | break; /* found sufficient pages */ | |
108 | if (tmp == NULL) | |
109 | /* insufficient pages, don't use any of them. */ | |
110 | return NULL; | |
111 | page = tmp; | |
112 | } | |
113 | ||
114 | /* add end of list marker for the returned list */ | |
115 | set_page_private(page, 0); | |
116 | /* actual return value, and adjustment of head */ | |
117 | page = *head; | |
118 | *head = tmp; | |
119 | return page; | |
120 | } | |
121 | ||
122 | /* may be used outside of locks to find the tail of a (usually short) | |
123 | * "private" page chain, before adding it back to a global chain head | |
124 | * with page_chain_add() under a spinlock. */ | |
125 | static struct page *page_chain_tail(struct page *page, int *len) | |
126 | { | |
127 | struct page *tmp; | |
128 | int i = 1; | |
129 | while ((tmp = page_chain_next(page))) | |
130 | ++i, page = tmp; | |
131 | if (len) | |
132 | *len = i; | |
133 | return page; | |
134 | } | |
135 | ||
136 | static int page_chain_free(struct page *page) | |
137 | { | |
138 | struct page *tmp; | |
139 | int i = 0; | |
140 | page_chain_for_each_safe(page, tmp) { | |
141 | put_page(page); | |
142 | ++i; | |
143 | } | |
144 | return i; | |
145 | } | |
146 | ||
147 | static void page_chain_add(struct page **head, | |
148 | struct page *chain_first, struct page *chain_last) | |
149 | { | |
150 | #if 1 | |
151 | struct page *tmp; | |
152 | tmp = page_chain_tail(chain_first, NULL); | |
153 | BUG_ON(tmp != chain_last); | |
154 | #endif | |
155 | ||
156 | /* add chain to head */ | |
157 | set_page_private(chain_last, (unsigned long)*head); | |
158 | *head = chain_first; | |
159 | } | |
160 | ||
161 | static struct page *drbd_pp_first_pages_or_try_alloc(struct drbd_conf *mdev, int number) | |
b411b363 PR |
162 | { |
163 | struct page *page = NULL; | |
45bb912b LE |
164 | struct page *tmp = NULL; |
165 | int i = 0; | |
b411b363 PR |
166 | |
167 | /* Yes, testing drbd_pp_vacant outside the lock is racy. | |
168 | * So what. It saves a spin_lock. */ | |
45bb912b | 169 | if (drbd_pp_vacant >= number) { |
b411b363 | 170 | spin_lock(&drbd_pp_lock); |
45bb912b LE |
171 | page = page_chain_del(&drbd_pp_pool, number); |
172 | if (page) | |
173 | drbd_pp_vacant -= number; | |
b411b363 | 174 | spin_unlock(&drbd_pp_lock); |
45bb912b LE |
175 | if (page) |
176 | return page; | |
b411b363 | 177 | } |
45bb912b | 178 | |
b411b363 PR |
179 | /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD |
180 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
181 | * which in turn might block on the other node at this very place. */ | |
45bb912b LE |
182 | for (i = 0; i < number; i++) { |
183 | tmp = alloc_page(GFP_TRY); | |
184 | if (!tmp) | |
185 | break; | |
186 | set_page_private(tmp, (unsigned long)page); | |
187 | page = tmp; | |
188 | } | |
189 | ||
190 | if (i == number) | |
191 | return page; | |
192 | ||
193 | /* Not enough pages immediately available this time. | |
194 | * No need to jump around here, drbd_pp_alloc will retry this | |
195 | * function "soon". */ | |
196 | if (page) { | |
197 | tmp = page_chain_tail(page, NULL); | |
198 | spin_lock(&drbd_pp_lock); | |
199 | page_chain_add(&drbd_pp_pool, page, tmp); | |
200 | drbd_pp_vacant += i; | |
201 | spin_unlock(&drbd_pp_lock); | |
202 | } | |
203 | return NULL; | |
b411b363 PR |
204 | } |
205 | ||
206 | /* kick lower level device, if we have more than (arbitrary number) | |
207 | * reference counts on it, which typically are locally submitted io | |
208 | * requests. don't use unacked_cnt, so we speed up proto A and B, too. */ | |
209 | static void maybe_kick_lo(struct drbd_conf *mdev) | |
210 | { | |
211 | if (atomic_read(&mdev->local_cnt) >= mdev->net_conf->unplug_watermark) | |
212 | drbd_kick_lo(mdev); | |
213 | } | |
214 | ||
215 | static void reclaim_net_ee(struct drbd_conf *mdev, struct list_head *to_be_freed) | |
216 | { | |
217 | struct drbd_epoch_entry *e; | |
218 | struct list_head *le, *tle; | |
219 | ||
220 | /* The EEs are always appended to the end of the list. Since | |
221 | they are sent in order over the wire, they have to finish | |
222 | in order. As soon as we see the first not finished we can | |
223 | stop to examine the list... */ | |
224 | ||
225 | list_for_each_safe(le, tle, &mdev->net_ee) { | |
226 | e = list_entry(le, struct drbd_epoch_entry, w.list); | |
45bb912b | 227 | if (drbd_ee_has_active_page(e)) |
b411b363 PR |
228 | break; |
229 | list_move(le, to_be_freed); | |
230 | } | |
231 | } | |
232 | ||
233 | static void drbd_kick_lo_and_reclaim_net(struct drbd_conf *mdev) | |
234 | { | |
235 | LIST_HEAD(reclaimed); | |
236 | struct drbd_epoch_entry *e, *t; | |
237 | ||
238 | maybe_kick_lo(mdev); | |
239 | spin_lock_irq(&mdev->req_lock); | |
240 | reclaim_net_ee(mdev, &reclaimed); | |
241 | spin_unlock_irq(&mdev->req_lock); | |
242 | ||
243 | list_for_each_entry_safe(e, t, &reclaimed, w.list) | |
435f0740 | 244 | drbd_free_net_ee(mdev, e); |
b411b363 PR |
245 | } |
246 | ||
247 | /** | |
45bb912b | 248 | * drbd_pp_alloc() - Returns @number pages, retries forever (or until signalled) |
b411b363 | 249 | * @mdev: DRBD device. |
45bb912b LE |
250 | * @number: number of pages requested |
251 | * @retry: whether to retry, if not enough pages are available right now | |
252 | * | |
253 | * Tries to allocate number pages, first from our own page pool, then from | |
254 | * the kernel, unless this allocation would exceed the max_buffers setting. | |
255 | * Possibly retry until DRBD frees sufficient pages somewhere else. | |
b411b363 | 256 | * |
45bb912b | 257 | * Returns a page chain linked via page->private. |
b411b363 | 258 | */ |
45bb912b | 259 | static struct page *drbd_pp_alloc(struct drbd_conf *mdev, unsigned number, bool retry) |
b411b363 PR |
260 | { |
261 | struct page *page = NULL; | |
262 | DEFINE_WAIT(wait); | |
263 | ||
45bb912b LE |
264 | /* Yes, we may run up to @number over max_buffers. If we |
265 | * follow it strictly, the admin will get it wrong anyways. */ | |
266 | if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) | |
267 | page = drbd_pp_first_pages_or_try_alloc(mdev, number); | |
b411b363 | 268 | |
45bb912b | 269 | while (page == NULL) { |
b411b363 PR |
270 | prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE); |
271 | ||
272 | drbd_kick_lo_and_reclaim_net(mdev); | |
273 | ||
274 | if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) { | |
45bb912b | 275 | page = drbd_pp_first_pages_or_try_alloc(mdev, number); |
b411b363 PR |
276 | if (page) |
277 | break; | |
278 | } | |
279 | ||
280 | if (!retry) | |
281 | break; | |
282 | ||
283 | if (signal_pending(current)) { | |
284 | dev_warn(DEV, "drbd_pp_alloc interrupted!\n"); | |
285 | break; | |
286 | } | |
287 | ||
288 | schedule(); | |
289 | } | |
290 | finish_wait(&drbd_pp_wait, &wait); | |
291 | ||
45bb912b LE |
292 | if (page) |
293 | atomic_add(number, &mdev->pp_in_use); | |
b411b363 PR |
294 | return page; |
295 | } | |
296 | ||
297 | /* Must not be used from irq, as that may deadlock: see drbd_pp_alloc. | |
45bb912b LE |
298 | * Is also used from inside an other spin_lock_irq(&mdev->req_lock); |
299 | * Either links the page chain back to the global pool, | |
300 | * or returns all pages to the system. */ | |
435f0740 | 301 | static void drbd_pp_free(struct drbd_conf *mdev, struct page *page, int is_net) |
b411b363 | 302 | { |
435f0740 | 303 | atomic_t *a = is_net ? &mdev->pp_in_use_by_net : &mdev->pp_in_use; |
b411b363 | 304 | int i; |
435f0740 | 305 | |
45bb912b LE |
306 | if (drbd_pp_vacant > (DRBD_MAX_SEGMENT_SIZE/PAGE_SIZE)*minor_count) |
307 | i = page_chain_free(page); | |
308 | else { | |
309 | struct page *tmp; | |
310 | tmp = page_chain_tail(page, &i); | |
311 | spin_lock(&drbd_pp_lock); | |
312 | page_chain_add(&drbd_pp_pool, page, tmp); | |
313 | drbd_pp_vacant += i; | |
314 | spin_unlock(&drbd_pp_lock); | |
b411b363 | 315 | } |
435f0740 | 316 | i = atomic_sub_return(i, a); |
45bb912b | 317 | if (i < 0) |
435f0740 LE |
318 | dev_warn(DEV, "ASSERTION FAILED: %s: %d < 0\n", |
319 | is_net ? "pp_in_use_by_net" : "pp_in_use", i); | |
b411b363 PR |
320 | wake_up(&drbd_pp_wait); |
321 | } | |
322 | ||
323 | /* | |
324 | You need to hold the req_lock: | |
325 | _drbd_wait_ee_list_empty() | |
326 | ||
327 | You must not have the req_lock: | |
328 | drbd_free_ee() | |
329 | drbd_alloc_ee() | |
330 | drbd_init_ee() | |
331 | drbd_release_ee() | |
332 | drbd_ee_fix_bhs() | |
333 | drbd_process_done_ee() | |
334 | drbd_clear_done_ee() | |
335 | drbd_wait_ee_list_empty() | |
336 | */ | |
337 | ||
338 | struct drbd_epoch_entry *drbd_alloc_ee(struct drbd_conf *mdev, | |
339 | u64 id, | |
340 | sector_t sector, | |
341 | unsigned int data_size, | |
342 | gfp_t gfp_mask) __must_hold(local) | |
343 | { | |
b411b363 PR |
344 | struct drbd_epoch_entry *e; |
345 | struct page *page; | |
45bb912b | 346 | unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b411b363 PR |
347 | |
348 | if (FAULT_ACTIVE(mdev, DRBD_FAULT_AL_EE)) | |
349 | return NULL; | |
350 | ||
351 | e = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM); | |
352 | if (!e) { | |
353 | if (!(gfp_mask & __GFP_NOWARN)) | |
354 | dev_err(DEV, "alloc_ee: Allocation of an EE failed\n"); | |
355 | return NULL; | |
356 | } | |
357 | ||
45bb912b LE |
358 | page = drbd_pp_alloc(mdev, nr_pages, (gfp_mask & __GFP_WAIT)); |
359 | if (!page) | |
360 | goto fail; | |
b411b363 | 361 | |
b411b363 PR |
362 | INIT_HLIST_NODE(&e->colision); |
363 | e->epoch = NULL; | |
45bb912b LE |
364 | e->mdev = mdev; |
365 | e->pages = page; | |
366 | atomic_set(&e->pending_bios, 0); | |
367 | e->size = data_size; | |
b411b363 | 368 | e->flags = 0; |
45bb912b | 369 | e->sector = sector; |
45bb912b | 370 | e->block_id = id; |
b411b363 | 371 | |
b411b363 PR |
372 | return e; |
373 | ||
45bb912b | 374 | fail: |
b411b363 | 375 | mempool_free(e, drbd_ee_mempool); |
b411b363 PR |
376 | return NULL; |
377 | } | |
378 | ||
435f0740 | 379 | void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, int is_net) |
b411b363 | 380 | { |
c36c3ced LE |
381 | if (e->flags & EE_HAS_DIGEST) |
382 | kfree(e->digest); | |
435f0740 | 383 | drbd_pp_free(mdev, e->pages, is_net); |
45bb912b | 384 | D_ASSERT(atomic_read(&e->pending_bios) == 0); |
b411b363 PR |
385 | D_ASSERT(hlist_unhashed(&e->colision)); |
386 | mempool_free(e, drbd_ee_mempool); | |
387 | } | |
388 | ||
389 | int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list) | |
390 | { | |
391 | LIST_HEAD(work_list); | |
392 | struct drbd_epoch_entry *e, *t; | |
393 | int count = 0; | |
435f0740 | 394 | int is_net = list == &mdev->net_ee; |
b411b363 PR |
395 | |
396 | spin_lock_irq(&mdev->req_lock); | |
397 | list_splice_init(list, &work_list); | |
398 | spin_unlock_irq(&mdev->req_lock); | |
399 | ||
400 | list_for_each_entry_safe(e, t, &work_list, w.list) { | |
435f0740 | 401 | drbd_free_some_ee(mdev, e, is_net); |
b411b363 PR |
402 | count++; |
403 | } | |
404 | return count; | |
405 | } | |
406 | ||
407 | ||
408 | /* | |
409 | * This function is called from _asender only_ | |
410 | * but see also comments in _req_mod(,barrier_acked) | |
411 | * and receive_Barrier. | |
412 | * | |
413 | * Move entries from net_ee to done_ee, if ready. | |
414 | * Grab done_ee, call all callbacks, free the entries. | |
415 | * The callbacks typically send out ACKs. | |
416 | */ | |
417 | static int drbd_process_done_ee(struct drbd_conf *mdev) | |
418 | { | |
419 | LIST_HEAD(work_list); | |
420 | LIST_HEAD(reclaimed); | |
421 | struct drbd_epoch_entry *e, *t; | |
422 | int ok = (mdev->state.conn >= C_WF_REPORT_PARAMS); | |
423 | ||
424 | spin_lock_irq(&mdev->req_lock); | |
425 | reclaim_net_ee(mdev, &reclaimed); | |
426 | list_splice_init(&mdev->done_ee, &work_list); | |
427 | spin_unlock_irq(&mdev->req_lock); | |
428 | ||
429 | list_for_each_entry_safe(e, t, &reclaimed, w.list) | |
435f0740 | 430 | drbd_free_net_ee(mdev, e); |
b411b363 PR |
431 | |
432 | /* possible callbacks here: | |
433 | * e_end_block, and e_end_resync_block, e_send_discard_ack. | |
434 | * all ignore the last argument. | |
435 | */ | |
436 | list_for_each_entry_safe(e, t, &work_list, w.list) { | |
b411b363 PR |
437 | /* list_del not necessary, next/prev members not touched */ |
438 | ok = e->w.cb(mdev, &e->w, !ok) && ok; | |
439 | drbd_free_ee(mdev, e); | |
440 | } | |
441 | wake_up(&mdev->ee_wait); | |
442 | ||
443 | return ok; | |
444 | } | |
445 | ||
446 | void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) | |
447 | { | |
448 | DEFINE_WAIT(wait); | |
449 | ||
450 | /* avoids spin_lock/unlock | |
451 | * and calling prepare_to_wait in the fast path */ | |
452 | while (!list_empty(head)) { | |
453 | prepare_to_wait(&mdev->ee_wait, &wait, TASK_UNINTERRUPTIBLE); | |
454 | spin_unlock_irq(&mdev->req_lock); | |
455 | drbd_kick_lo(mdev); | |
456 | schedule(); | |
457 | finish_wait(&mdev->ee_wait, &wait); | |
458 | spin_lock_irq(&mdev->req_lock); | |
459 | } | |
460 | } | |
461 | ||
462 | void drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) | |
463 | { | |
464 | spin_lock_irq(&mdev->req_lock); | |
465 | _drbd_wait_ee_list_empty(mdev, head); | |
466 | spin_unlock_irq(&mdev->req_lock); | |
467 | } | |
468 | ||
469 | /* see also kernel_accept; which is only present since 2.6.18. | |
470 | * also we want to log which part of it failed, exactly */ | |
471 | static int drbd_accept(struct drbd_conf *mdev, const char **what, | |
472 | struct socket *sock, struct socket **newsock) | |
473 | { | |
474 | struct sock *sk = sock->sk; | |
475 | int err = 0; | |
476 | ||
477 | *what = "listen"; | |
478 | err = sock->ops->listen(sock, 5); | |
479 | if (err < 0) | |
480 | goto out; | |
481 | ||
482 | *what = "sock_create_lite"; | |
483 | err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol, | |
484 | newsock); | |
485 | if (err < 0) | |
486 | goto out; | |
487 | ||
488 | *what = "accept"; | |
489 | err = sock->ops->accept(sock, *newsock, 0); | |
490 | if (err < 0) { | |
491 | sock_release(*newsock); | |
492 | *newsock = NULL; | |
493 | goto out; | |
494 | } | |
495 | (*newsock)->ops = sock->ops; | |
496 | ||
497 | out: | |
498 | return err; | |
499 | } | |
500 | ||
501 | static int drbd_recv_short(struct drbd_conf *mdev, struct socket *sock, | |
502 | void *buf, size_t size, int flags) | |
503 | { | |
504 | mm_segment_t oldfs; | |
505 | struct kvec iov = { | |
506 | .iov_base = buf, | |
507 | .iov_len = size, | |
508 | }; | |
509 | struct msghdr msg = { | |
510 | .msg_iovlen = 1, | |
511 | .msg_iov = (struct iovec *)&iov, | |
512 | .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) | |
513 | }; | |
514 | int rv; | |
515 | ||
516 | oldfs = get_fs(); | |
517 | set_fs(KERNEL_DS); | |
518 | rv = sock_recvmsg(sock, &msg, size, msg.msg_flags); | |
519 | set_fs(oldfs); | |
520 | ||
521 | return rv; | |
522 | } | |
523 | ||
524 | static int drbd_recv(struct drbd_conf *mdev, void *buf, size_t size) | |
525 | { | |
526 | mm_segment_t oldfs; | |
527 | struct kvec iov = { | |
528 | .iov_base = buf, | |
529 | .iov_len = size, | |
530 | }; | |
531 | struct msghdr msg = { | |
532 | .msg_iovlen = 1, | |
533 | .msg_iov = (struct iovec *)&iov, | |
534 | .msg_flags = MSG_WAITALL | MSG_NOSIGNAL | |
535 | }; | |
536 | int rv; | |
537 | ||
538 | oldfs = get_fs(); | |
539 | set_fs(KERNEL_DS); | |
540 | ||
541 | for (;;) { | |
542 | rv = sock_recvmsg(mdev->data.socket, &msg, size, msg.msg_flags); | |
543 | if (rv == size) | |
544 | break; | |
545 | ||
546 | /* Note: | |
547 | * ECONNRESET other side closed the connection | |
548 | * ERESTARTSYS (on sock) we got a signal | |
549 | */ | |
550 | ||
551 | if (rv < 0) { | |
552 | if (rv == -ECONNRESET) | |
553 | dev_info(DEV, "sock was reset by peer\n"); | |
554 | else if (rv != -ERESTARTSYS) | |
555 | dev_err(DEV, "sock_recvmsg returned %d\n", rv); | |
556 | break; | |
557 | } else if (rv == 0) { | |
558 | dev_info(DEV, "sock was shut down by peer\n"); | |
559 | break; | |
560 | } else { | |
561 | /* signal came in, or peer/link went down, | |
562 | * after we read a partial message | |
563 | */ | |
564 | /* D_ASSERT(signal_pending(current)); */ | |
565 | break; | |
566 | } | |
567 | }; | |
568 | ||
569 | set_fs(oldfs); | |
570 | ||
571 | if (rv != size) | |
572 | drbd_force_state(mdev, NS(conn, C_BROKEN_PIPE)); | |
573 | ||
574 | return rv; | |
575 | } | |
576 | ||
5dbf1673 LE |
577 | /* quoting tcp(7): |
578 | * On individual connections, the socket buffer size must be set prior to the | |
579 | * listen(2) or connect(2) calls in order to have it take effect. | |
580 | * This is our wrapper to do so. | |
581 | */ | |
582 | static void drbd_setbufsize(struct socket *sock, unsigned int snd, | |
583 | unsigned int rcv) | |
584 | { | |
585 | /* open coded SO_SNDBUF, SO_RCVBUF */ | |
586 | if (snd) { | |
587 | sock->sk->sk_sndbuf = snd; | |
588 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
589 | } | |
590 | if (rcv) { | |
591 | sock->sk->sk_rcvbuf = rcv; | |
592 | sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
593 | } | |
594 | } | |
595 | ||
b411b363 PR |
596 | static struct socket *drbd_try_connect(struct drbd_conf *mdev) |
597 | { | |
598 | const char *what; | |
599 | struct socket *sock; | |
600 | struct sockaddr_in6 src_in6; | |
601 | int err; | |
602 | int disconnect_on_error = 1; | |
603 | ||
604 | if (!get_net_conf(mdev)) | |
605 | return NULL; | |
606 | ||
607 | what = "sock_create_kern"; | |
608 | err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family, | |
609 | SOCK_STREAM, IPPROTO_TCP, &sock); | |
610 | if (err < 0) { | |
611 | sock = NULL; | |
612 | goto out; | |
613 | } | |
614 | ||
615 | sock->sk->sk_rcvtimeo = | |
616 | sock->sk->sk_sndtimeo = mdev->net_conf->try_connect_int*HZ; | |
5dbf1673 LE |
617 | drbd_setbufsize(sock, mdev->net_conf->sndbuf_size, |
618 | mdev->net_conf->rcvbuf_size); | |
b411b363 PR |
619 | |
620 | /* explicitly bind to the configured IP as source IP | |
621 | * for the outgoing connections. | |
622 | * This is needed for multihomed hosts and to be | |
623 | * able to use lo: interfaces for drbd. | |
624 | * Make sure to use 0 as port number, so linux selects | |
625 | * a free one dynamically. | |
626 | */ | |
627 | memcpy(&src_in6, mdev->net_conf->my_addr, | |
628 | min_t(int, mdev->net_conf->my_addr_len, sizeof(src_in6))); | |
629 | if (((struct sockaddr *)mdev->net_conf->my_addr)->sa_family == AF_INET6) | |
630 | src_in6.sin6_port = 0; | |
631 | else | |
632 | ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ | |
633 | ||
634 | what = "bind before connect"; | |
635 | err = sock->ops->bind(sock, | |
636 | (struct sockaddr *) &src_in6, | |
637 | mdev->net_conf->my_addr_len); | |
638 | if (err < 0) | |
639 | goto out; | |
640 | ||
641 | /* connect may fail, peer not yet available. | |
642 | * stay C_WF_CONNECTION, don't go Disconnecting! */ | |
643 | disconnect_on_error = 0; | |
644 | what = "connect"; | |
645 | err = sock->ops->connect(sock, | |
646 | (struct sockaddr *)mdev->net_conf->peer_addr, | |
647 | mdev->net_conf->peer_addr_len, 0); | |
648 | ||
649 | out: | |
650 | if (err < 0) { | |
651 | if (sock) { | |
652 | sock_release(sock); | |
653 | sock = NULL; | |
654 | } | |
655 | switch (-err) { | |
656 | /* timeout, busy, signal pending */ | |
657 | case ETIMEDOUT: case EAGAIN: case EINPROGRESS: | |
658 | case EINTR: case ERESTARTSYS: | |
659 | /* peer not (yet) available, network problem */ | |
660 | case ECONNREFUSED: case ENETUNREACH: | |
661 | case EHOSTDOWN: case EHOSTUNREACH: | |
662 | disconnect_on_error = 0; | |
663 | break; | |
664 | default: | |
665 | dev_err(DEV, "%s failed, err = %d\n", what, err); | |
666 | } | |
667 | if (disconnect_on_error) | |
668 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
669 | } | |
670 | put_net_conf(mdev); | |
671 | return sock; | |
672 | } | |
673 | ||
674 | static struct socket *drbd_wait_for_connect(struct drbd_conf *mdev) | |
675 | { | |
676 | int timeo, err; | |
677 | struct socket *s_estab = NULL, *s_listen; | |
678 | const char *what; | |
679 | ||
680 | if (!get_net_conf(mdev)) | |
681 | return NULL; | |
682 | ||
683 | what = "sock_create_kern"; | |
684 | err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family, | |
685 | SOCK_STREAM, IPPROTO_TCP, &s_listen); | |
686 | if (err) { | |
687 | s_listen = NULL; | |
688 | goto out; | |
689 | } | |
690 | ||
691 | timeo = mdev->net_conf->try_connect_int * HZ; | |
692 | timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */ | |
693 | ||
694 | s_listen->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
695 | s_listen->sk->sk_rcvtimeo = timeo; | |
696 | s_listen->sk->sk_sndtimeo = timeo; | |
5dbf1673 LE |
697 | drbd_setbufsize(s_listen, mdev->net_conf->sndbuf_size, |
698 | mdev->net_conf->rcvbuf_size); | |
b411b363 PR |
699 | |
700 | what = "bind before listen"; | |
701 | err = s_listen->ops->bind(s_listen, | |
702 | (struct sockaddr *) mdev->net_conf->my_addr, | |
703 | mdev->net_conf->my_addr_len); | |
704 | if (err < 0) | |
705 | goto out; | |
706 | ||
707 | err = drbd_accept(mdev, &what, s_listen, &s_estab); | |
708 | ||
709 | out: | |
710 | if (s_listen) | |
711 | sock_release(s_listen); | |
712 | if (err < 0) { | |
713 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
714 | dev_err(DEV, "%s failed, err = %d\n", what, err); | |
715 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
716 | } | |
717 | } | |
718 | put_net_conf(mdev); | |
719 | ||
720 | return s_estab; | |
721 | } | |
722 | ||
723 | static int drbd_send_fp(struct drbd_conf *mdev, | |
724 | struct socket *sock, enum drbd_packets cmd) | |
725 | { | |
02918be2 | 726 | struct p_header80 *h = &mdev->data.sbuf.header.h80; |
b411b363 PR |
727 | |
728 | return _drbd_send_cmd(mdev, sock, cmd, h, sizeof(*h), 0); | |
729 | } | |
730 | ||
731 | static enum drbd_packets drbd_recv_fp(struct drbd_conf *mdev, struct socket *sock) | |
732 | { | |
02918be2 | 733 | struct p_header80 *h = &mdev->data.rbuf.header.h80; |
b411b363 PR |
734 | int rr; |
735 | ||
736 | rr = drbd_recv_short(mdev, sock, h, sizeof(*h), 0); | |
737 | ||
738 | if (rr == sizeof(*h) && h->magic == BE_DRBD_MAGIC) | |
739 | return be16_to_cpu(h->command); | |
740 | ||
741 | return 0xffff; | |
742 | } | |
743 | ||
744 | /** | |
745 | * drbd_socket_okay() - Free the socket if its connection is not okay | |
746 | * @mdev: DRBD device. | |
747 | * @sock: pointer to the pointer to the socket. | |
748 | */ | |
749 | static int drbd_socket_okay(struct drbd_conf *mdev, struct socket **sock) | |
750 | { | |
751 | int rr; | |
752 | char tb[4]; | |
753 | ||
754 | if (!*sock) | |
755 | return FALSE; | |
756 | ||
757 | rr = drbd_recv_short(mdev, *sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); | |
758 | ||
759 | if (rr > 0 || rr == -EAGAIN) { | |
760 | return TRUE; | |
761 | } else { | |
762 | sock_release(*sock); | |
763 | *sock = NULL; | |
764 | return FALSE; | |
765 | } | |
766 | } | |
767 | ||
768 | /* | |
769 | * return values: | |
770 | * 1 yes, we have a valid connection | |
771 | * 0 oops, did not work out, please try again | |
772 | * -1 peer talks different language, | |
773 | * no point in trying again, please go standalone. | |
774 | * -2 We do not have a network config... | |
775 | */ | |
776 | static int drbd_connect(struct drbd_conf *mdev) | |
777 | { | |
778 | struct socket *s, *sock, *msock; | |
779 | int try, h, ok; | |
780 | ||
781 | D_ASSERT(!mdev->data.socket); | |
782 | ||
b411b363 PR |
783 | if (drbd_request_state(mdev, NS(conn, C_WF_CONNECTION)) < SS_SUCCESS) |
784 | return -2; | |
785 | ||
786 | clear_bit(DISCARD_CONCURRENT, &mdev->flags); | |
787 | ||
788 | sock = NULL; | |
789 | msock = NULL; | |
790 | ||
791 | do { | |
792 | for (try = 0;;) { | |
793 | /* 3 tries, this should take less than a second! */ | |
794 | s = drbd_try_connect(mdev); | |
795 | if (s || ++try >= 3) | |
796 | break; | |
797 | /* give the other side time to call bind() & listen() */ | |
798 | __set_current_state(TASK_INTERRUPTIBLE); | |
799 | schedule_timeout(HZ / 10); | |
800 | } | |
801 | ||
802 | if (s) { | |
803 | if (!sock) { | |
804 | drbd_send_fp(mdev, s, P_HAND_SHAKE_S); | |
805 | sock = s; | |
806 | s = NULL; | |
807 | } else if (!msock) { | |
808 | drbd_send_fp(mdev, s, P_HAND_SHAKE_M); | |
809 | msock = s; | |
810 | s = NULL; | |
811 | } else { | |
812 | dev_err(DEV, "Logic error in drbd_connect()\n"); | |
813 | goto out_release_sockets; | |
814 | } | |
815 | } | |
816 | ||
817 | if (sock && msock) { | |
818 | __set_current_state(TASK_INTERRUPTIBLE); | |
819 | schedule_timeout(HZ / 10); | |
820 | ok = drbd_socket_okay(mdev, &sock); | |
821 | ok = drbd_socket_okay(mdev, &msock) && ok; | |
822 | if (ok) | |
823 | break; | |
824 | } | |
825 | ||
826 | retry: | |
827 | s = drbd_wait_for_connect(mdev); | |
828 | if (s) { | |
829 | try = drbd_recv_fp(mdev, s); | |
830 | drbd_socket_okay(mdev, &sock); | |
831 | drbd_socket_okay(mdev, &msock); | |
832 | switch (try) { | |
833 | case P_HAND_SHAKE_S: | |
834 | if (sock) { | |
835 | dev_warn(DEV, "initial packet S crossed\n"); | |
836 | sock_release(sock); | |
837 | } | |
838 | sock = s; | |
839 | break; | |
840 | case P_HAND_SHAKE_M: | |
841 | if (msock) { | |
842 | dev_warn(DEV, "initial packet M crossed\n"); | |
843 | sock_release(msock); | |
844 | } | |
845 | msock = s; | |
846 | set_bit(DISCARD_CONCURRENT, &mdev->flags); | |
847 | break; | |
848 | default: | |
849 | dev_warn(DEV, "Error receiving initial packet\n"); | |
850 | sock_release(s); | |
851 | if (random32() & 1) | |
852 | goto retry; | |
853 | } | |
854 | } | |
855 | ||
856 | if (mdev->state.conn <= C_DISCONNECTING) | |
857 | goto out_release_sockets; | |
858 | if (signal_pending(current)) { | |
859 | flush_signals(current); | |
860 | smp_rmb(); | |
861 | if (get_t_state(&mdev->receiver) == Exiting) | |
862 | goto out_release_sockets; | |
863 | } | |
864 | ||
865 | if (sock && msock) { | |
866 | ok = drbd_socket_okay(mdev, &sock); | |
867 | ok = drbd_socket_okay(mdev, &msock) && ok; | |
868 | if (ok) | |
869 | break; | |
870 | } | |
871 | } while (1); | |
872 | ||
873 | msock->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
874 | sock->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
875 | ||
876 | sock->sk->sk_allocation = GFP_NOIO; | |
877 | msock->sk->sk_allocation = GFP_NOIO; | |
878 | ||
879 | sock->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; | |
880 | msock->sk->sk_priority = TC_PRIO_INTERACTIVE; | |
881 | ||
b411b363 PR |
882 | /* NOT YET ... |
883 | * sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; | |
884 | * sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; | |
885 | * first set it to the P_HAND_SHAKE timeout, | |
886 | * which we set to 4x the configured ping_timeout. */ | |
887 | sock->sk->sk_sndtimeo = | |
888 | sock->sk->sk_rcvtimeo = mdev->net_conf->ping_timeo*4*HZ/10; | |
889 | ||
890 | msock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; | |
891 | msock->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ; | |
892 | ||
893 | /* we don't want delays. | |
894 | * we use TCP_CORK where apropriate, though */ | |
895 | drbd_tcp_nodelay(sock); | |
896 | drbd_tcp_nodelay(msock); | |
897 | ||
898 | mdev->data.socket = sock; | |
899 | mdev->meta.socket = msock; | |
900 | mdev->last_received = jiffies; | |
901 | ||
902 | D_ASSERT(mdev->asender.task == NULL); | |
903 | ||
904 | h = drbd_do_handshake(mdev); | |
905 | if (h <= 0) | |
906 | return h; | |
907 | ||
908 | if (mdev->cram_hmac_tfm) { | |
909 | /* drbd_request_state(mdev, NS(conn, WFAuth)); */ | |
b10d96cb JT |
910 | switch (drbd_do_auth(mdev)) { |
911 | case -1: | |
b411b363 PR |
912 | dev_err(DEV, "Authentication of peer failed\n"); |
913 | return -1; | |
b10d96cb JT |
914 | case 0: |
915 | dev_err(DEV, "Authentication of peer failed, trying again.\n"); | |
916 | return 0; | |
b411b363 PR |
917 | } |
918 | } | |
919 | ||
920 | if (drbd_request_state(mdev, NS(conn, C_WF_REPORT_PARAMS)) < SS_SUCCESS) | |
921 | return 0; | |
922 | ||
923 | sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; | |
924 | sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; | |
925 | ||
926 | atomic_set(&mdev->packet_seq, 0); | |
927 | mdev->peer_seq = 0; | |
928 | ||
929 | drbd_thread_start(&mdev->asender); | |
930 | ||
d5373389 PR |
931 | if (mdev->agreed_pro_version < 95 && get_ldev(mdev)) { |
932 | drbd_setup_queue_param(mdev, DRBD_MAX_SIZE_H80_PACKET); | |
933 | put_ldev(mdev); | |
934 | } | |
935 | ||
7e2455c1 PR |
936 | if (!drbd_send_protocol(mdev)) |
937 | return -1; | |
b411b363 | 938 | drbd_send_sync_param(mdev, &mdev->sync_conf); |
e89b591c | 939 | drbd_send_sizes(mdev, 0, 0); |
b411b363 PR |
940 | drbd_send_uuids(mdev); |
941 | drbd_send_state(mdev); | |
942 | clear_bit(USE_DEGR_WFC_T, &mdev->flags); | |
943 | clear_bit(RESIZE_PENDING, &mdev->flags); | |
944 | ||
945 | return 1; | |
946 | ||
947 | out_release_sockets: | |
948 | if (sock) | |
949 | sock_release(sock); | |
950 | if (msock) | |
951 | sock_release(msock); | |
952 | return -1; | |
953 | } | |
954 | ||
02918be2 | 955 | static int drbd_recv_header(struct drbd_conf *mdev, enum drbd_packets *cmd, unsigned int *packet_size) |
b411b363 | 956 | { |
02918be2 | 957 | union p_header *h = &mdev->data.rbuf.header; |
b411b363 PR |
958 | int r; |
959 | ||
960 | r = drbd_recv(mdev, h, sizeof(*h)); | |
b411b363 PR |
961 | if (unlikely(r != sizeof(*h))) { |
962 | dev_err(DEV, "short read expecting header on sock: r=%d\n", r); | |
963 | return FALSE; | |
02918be2 PR |
964 | } |
965 | ||
966 | if (likely(h->h80.magic == BE_DRBD_MAGIC)) { | |
967 | *cmd = be16_to_cpu(h->h80.command); | |
968 | *packet_size = be16_to_cpu(h->h80.length); | |
969 | } else if (h->h95.magic == BE_DRBD_MAGIC_BIG) { | |
970 | *cmd = be16_to_cpu(h->h95.command); | |
971 | *packet_size = be32_to_cpu(h->h95.length); | |
972 | } else { | |
004352fa LE |
973 | dev_err(DEV, "magic?? on data m: 0x%08x c: %d l: %d\n", |
974 | be32_to_cpu(h->h80.magic), | |
975 | be16_to_cpu(h->h80.command), | |
976 | be16_to_cpu(h->h80.length)); | |
b411b363 PR |
977 | return FALSE; |
978 | } | |
979 | mdev->last_received = jiffies; | |
980 | ||
981 | return TRUE; | |
982 | } | |
983 | ||
984 | static enum finish_epoch drbd_flush_after_epoch(struct drbd_conf *mdev, struct drbd_epoch *epoch) | |
985 | { | |
986 | int rv; | |
987 | ||
988 | if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) { | |
fbd9b09a DM |
989 | rv = blkdev_issue_flush(mdev->ldev->backing_bdev, GFP_KERNEL, |
990 | NULL, BLKDEV_IFL_WAIT); | |
b411b363 PR |
991 | if (rv) { |
992 | dev_err(DEV, "local disk flush failed with status %d\n", rv); | |
993 | /* would rather check on EOPNOTSUPP, but that is not reliable. | |
994 | * don't try again for ANY return value != 0 | |
995 | * if (rv == -EOPNOTSUPP) */ | |
996 | drbd_bump_write_ordering(mdev, WO_drain_io); | |
997 | } | |
998 | put_ldev(mdev); | |
999 | } | |
1000 | ||
1001 | return drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE); | |
1002 | } | |
1003 | ||
1004 | static int w_flush(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
1005 | { | |
1006 | struct flush_work *fw = (struct flush_work *)w; | |
1007 | struct drbd_epoch *epoch = fw->epoch; | |
1008 | ||
1009 | kfree(w); | |
1010 | ||
1011 | if (!test_and_set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags)) | |
1012 | drbd_flush_after_epoch(mdev, epoch); | |
1013 | ||
1014 | drbd_may_finish_epoch(mdev, epoch, EV_PUT | | |
1015 | (mdev->state.conn < C_CONNECTED ? EV_CLEANUP : 0)); | |
1016 | ||
1017 | return 1; | |
1018 | } | |
1019 | ||
1020 | /** | |
1021 | * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. | |
1022 | * @mdev: DRBD device. | |
1023 | * @epoch: Epoch object. | |
1024 | * @ev: Epoch event. | |
1025 | */ | |
1026 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev, | |
1027 | struct drbd_epoch *epoch, | |
1028 | enum epoch_event ev) | |
1029 | { | |
1030 | int finish, epoch_size; | |
1031 | struct drbd_epoch *next_epoch; | |
1032 | int schedule_flush = 0; | |
1033 | enum finish_epoch rv = FE_STILL_LIVE; | |
1034 | ||
1035 | spin_lock(&mdev->epoch_lock); | |
1036 | do { | |
1037 | next_epoch = NULL; | |
1038 | finish = 0; | |
1039 | ||
1040 | epoch_size = atomic_read(&epoch->epoch_size); | |
1041 | ||
1042 | switch (ev & ~EV_CLEANUP) { | |
1043 | case EV_PUT: | |
1044 | atomic_dec(&epoch->active); | |
1045 | break; | |
1046 | case EV_GOT_BARRIER_NR: | |
1047 | set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); | |
1048 | ||
1049 | /* Special case: If we just switched from WO_bio_barrier to | |
1050 | WO_bdev_flush we should not finish the current epoch */ | |
1051 | if (test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags) && epoch_size == 1 && | |
1052 | mdev->write_ordering != WO_bio_barrier && | |
1053 | epoch == mdev->current_epoch) | |
1054 | clear_bit(DE_CONTAINS_A_BARRIER, &epoch->flags); | |
1055 | break; | |
1056 | case EV_BARRIER_DONE: | |
1057 | set_bit(DE_BARRIER_IN_NEXT_EPOCH_DONE, &epoch->flags); | |
1058 | break; | |
1059 | case EV_BECAME_LAST: | |
1060 | /* nothing to do*/ | |
1061 | break; | |
1062 | } | |
1063 | ||
b411b363 PR |
1064 | if (epoch_size != 0 && |
1065 | atomic_read(&epoch->active) == 0 && | |
1066 | test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) && | |
1067 | epoch->list.prev == &mdev->current_epoch->list && | |
1068 | !test_bit(DE_IS_FINISHING, &epoch->flags)) { | |
1069 | /* Nearly all conditions are met to finish that epoch... */ | |
1070 | if (test_bit(DE_BARRIER_IN_NEXT_EPOCH_DONE, &epoch->flags) || | |
1071 | mdev->write_ordering == WO_none || | |
1072 | (epoch_size == 1 && test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags)) || | |
1073 | ev & EV_CLEANUP) { | |
1074 | finish = 1; | |
1075 | set_bit(DE_IS_FINISHING, &epoch->flags); | |
1076 | } else if (!test_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags) && | |
1077 | mdev->write_ordering == WO_bio_barrier) { | |
1078 | atomic_inc(&epoch->active); | |
1079 | schedule_flush = 1; | |
1080 | } | |
1081 | } | |
1082 | if (finish) { | |
1083 | if (!(ev & EV_CLEANUP)) { | |
1084 | spin_unlock(&mdev->epoch_lock); | |
1085 | drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size); | |
1086 | spin_lock(&mdev->epoch_lock); | |
1087 | } | |
1088 | dec_unacked(mdev); | |
1089 | ||
1090 | if (mdev->current_epoch != epoch) { | |
1091 | next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); | |
1092 | list_del(&epoch->list); | |
1093 | ev = EV_BECAME_LAST | (ev & EV_CLEANUP); | |
1094 | mdev->epochs--; | |
b411b363 PR |
1095 | kfree(epoch); |
1096 | ||
1097 | if (rv == FE_STILL_LIVE) | |
1098 | rv = FE_DESTROYED; | |
1099 | } else { | |
1100 | epoch->flags = 0; | |
1101 | atomic_set(&epoch->epoch_size, 0); | |
698f9315 | 1102 | /* atomic_set(&epoch->active, 0); is already zero */ |
b411b363 PR |
1103 | if (rv == FE_STILL_LIVE) |
1104 | rv = FE_RECYCLED; | |
1105 | } | |
1106 | } | |
1107 | ||
1108 | if (!next_epoch) | |
1109 | break; | |
1110 | ||
1111 | epoch = next_epoch; | |
1112 | } while (1); | |
1113 | ||
1114 | spin_unlock(&mdev->epoch_lock); | |
1115 | ||
1116 | if (schedule_flush) { | |
1117 | struct flush_work *fw; | |
1118 | fw = kmalloc(sizeof(*fw), GFP_ATOMIC); | |
1119 | if (fw) { | |
b411b363 PR |
1120 | fw->w.cb = w_flush; |
1121 | fw->epoch = epoch; | |
1122 | drbd_queue_work(&mdev->data.work, &fw->w); | |
1123 | } else { | |
1124 | dev_warn(DEV, "Could not kmalloc a flush_work obj\n"); | |
1125 | set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags); | |
1126 | /* That is not a recursion, only one level */ | |
1127 | drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE); | |
1128 | drbd_may_finish_epoch(mdev, epoch, EV_PUT); | |
1129 | } | |
1130 | } | |
1131 | ||
1132 | return rv; | |
1133 | } | |
1134 | ||
1135 | /** | |
1136 | * drbd_bump_write_ordering() - Fall back to an other write ordering method | |
1137 | * @mdev: DRBD device. | |
1138 | * @wo: Write ordering method to try. | |
1139 | */ | |
1140 | void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local) | |
1141 | { | |
1142 | enum write_ordering_e pwo; | |
1143 | static char *write_ordering_str[] = { | |
1144 | [WO_none] = "none", | |
1145 | [WO_drain_io] = "drain", | |
1146 | [WO_bdev_flush] = "flush", | |
1147 | [WO_bio_barrier] = "barrier", | |
1148 | }; | |
1149 | ||
1150 | pwo = mdev->write_ordering; | |
1151 | wo = min(pwo, wo); | |
1152 | if (wo == WO_bio_barrier && mdev->ldev->dc.no_disk_barrier) | |
1153 | wo = WO_bdev_flush; | |
1154 | if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush) | |
1155 | wo = WO_drain_io; | |
1156 | if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain) | |
1157 | wo = WO_none; | |
1158 | mdev->write_ordering = wo; | |
1159 | if (pwo != mdev->write_ordering || wo == WO_bio_barrier) | |
1160 | dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]); | |
1161 | } | |
1162 | ||
45bb912b LE |
1163 | /** |
1164 | * drbd_submit_ee() | |
1165 | * @mdev: DRBD device. | |
1166 | * @e: epoch entry | |
1167 | * @rw: flag field, see bio->bi_rw | |
1168 | */ | |
1169 | /* TODO allocate from our own bio_set. */ | |
1170 | int drbd_submit_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, | |
1171 | const unsigned rw, const int fault_type) | |
1172 | { | |
1173 | struct bio *bios = NULL; | |
1174 | struct bio *bio; | |
1175 | struct page *page = e->pages; | |
1176 | sector_t sector = e->sector; | |
1177 | unsigned ds = e->size; | |
1178 | unsigned n_bios = 0; | |
1179 | unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT; | |
1180 | ||
1181 | /* In most cases, we will only need one bio. But in case the lower | |
1182 | * level restrictions happen to be different at this offset on this | |
1183 | * side than those of the sending peer, we may need to submit the | |
1184 | * request in more than one bio. */ | |
1185 | next_bio: | |
1186 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
1187 | if (!bio) { | |
1188 | dev_err(DEV, "submit_ee: Allocation of a bio failed\n"); | |
1189 | goto fail; | |
1190 | } | |
1191 | /* > e->sector, unless this is the first bio */ | |
1192 | bio->bi_sector = sector; | |
1193 | bio->bi_bdev = mdev->ldev->backing_bdev; | |
1194 | /* we special case some flags in the multi-bio case, see below | |
7b6d91da | 1195 | * (REQ_UNPLUG, REQ_HARDBARRIER) */ |
45bb912b LE |
1196 | bio->bi_rw = rw; |
1197 | bio->bi_private = e; | |
1198 | bio->bi_end_io = drbd_endio_sec; | |
1199 | ||
1200 | bio->bi_next = bios; | |
1201 | bios = bio; | |
1202 | ++n_bios; | |
1203 | ||
1204 | page_chain_for_each(page) { | |
1205 | unsigned len = min_t(unsigned, ds, PAGE_SIZE); | |
1206 | if (!bio_add_page(bio, page, len, 0)) { | |
1207 | /* a single page must always be possible! */ | |
1208 | BUG_ON(bio->bi_vcnt == 0); | |
1209 | goto next_bio; | |
1210 | } | |
1211 | ds -= len; | |
1212 | sector += len >> 9; | |
1213 | --nr_pages; | |
1214 | } | |
1215 | D_ASSERT(page == NULL); | |
1216 | D_ASSERT(ds == 0); | |
1217 | ||
1218 | atomic_set(&e->pending_bios, n_bios); | |
1219 | do { | |
1220 | bio = bios; | |
1221 | bios = bios->bi_next; | |
1222 | bio->bi_next = NULL; | |
1223 | ||
7b6d91da | 1224 | /* strip off REQ_UNPLUG unless it is the last bio */ |
45bb912b | 1225 | if (bios) |
7b6d91da | 1226 | bio->bi_rw &= ~REQ_UNPLUG; |
45bb912b LE |
1227 | |
1228 | drbd_generic_make_request(mdev, fault_type, bio); | |
1229 | ||
7b6d91da | 1230 | /* strip off REQ_HARDBARRIER, |
45bb912b LE |
1231 | * unless it is the first or last bio */ |
1232 | if (bios && bios->bi_next) | |
7b6d91da | 1233 | bios->bi_rw &= ~REQ_HARDBARRIER; |
45bb912b LE |
1234 | } while (bios); |
1235 | maybe_kick_lo(mdev); | |
1236 | return 0; | |
1237 | ||
1238 | fail: | |
1239 | while (bios) { | |
1240 | bio = bios; | |
1241 | bios = bios->bi_next; | |
1242 | bio_put(bio); | |
1243 | } | |
1244 | return -ENOMEM; | |
1245 | } | |
1246 | ||
b411b363 | 1247 | /** |
7b6d91da | 1248 | * w_e_reissue() - Worker callback; Resubmit a bio, without REQ_HARDBARRIER set |
b411b363 PR |
1249 | * @mdev: DRBD device. |
1250 | * @w: work object. | |
1251 | * @cancel: The connection will be closed anyways (unused in this callback) | |
1252 | */ | |
1253 | int w_e_reissue(struct drbd_conf *mdev, struct drbd_work *w, int cancel) __releases(local) | |
1254 | { | |
1255 | struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; | |
b411b363 PR |
1256 | /* We leave DE_CONTAINS_A_BARRIER and EE_IS_BARRIER in place, |
1257 | (and DE_BARRIER_IN_NEXT_EPOCH_ISSUED in the previous Epoch) | |
1258 | so that we can finish that epoch in drbd_may_finish_epoch(). | |
1259 | That is necessary if we already have a long chain of Epochs, before | |
7b6d91da | 1260 | we realize that REQ_HARDBARRIER is actually not supported */ |
b411b363 PR |
1261 | |
1262 | /* As long as the -ENOTSUPP on the barrier is reported immediately | |
1263 | that will never trigger. If it is reported late, we will just | |
1264 | print that warning and continue correctly for all future requests | |
1265 | with WO_bdev_flush */ | |
1266 | if (previous_epoch(mdev, e->epoch)) | |
1267 | dev_warn(DEV, "Write ordering was not enforced (one time event)\n"); | |
1268 | ||
b411b363 PR |
1269 | /* we still have a local reference, |
1270 | * get_ldev was done in receive_Data. */ | |
b411b363 PR |
1271 | |
1272 | e->w.cb = e_end_block; | |
45bb912b LE |
1273 | if (drbd_submit_ee(mdev, e, WRITE, DRBD_FAULT_DT_WR) != 0) { |
1274 | /* drbd_submit_ee fails for one reason only: | |
1275 | * if was not able to allocate sufficient bios. | |
1276 | * requeue, try again later. */ | |
1277 | e->w.cb = w_e_reissue; | |
1278 | drbd_queue_work(&mdev->data.work, &e->w); | |
1279 | } | |
b411b363 PR |
1280 | return 1; |
1281 | } | |
1282 | ||
02918be2 | 1283 | static int receive_Barrier(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
1284 | { |
1285 | int rv, issue_flush; | |
02918be2 | 1286 | struct p_barrier *p = &mdev->data.rbuf.barrier; |
b411b363 PR |
1287 | struct drbd_epoch *epoch; |
1288 | ||
b411b363 PR |
1289 | inc_unacked(mdev); |
1290 | ||
1291 | if (mdev->net_conf->wire_protocol != DRBD_PROT_C) | |
1292 | drbd_kick_lo(mdev); | |
1293 | ||
1294 | mdev->current_epoch->barrier_nr = p->barrier; | |
1295 | rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR); | |
1296 | ||
1297 | /* P_BARRIER_ACK may imply that the corresponding extent is dropped from | |
1298 | * the activity log, which means it would not be resynced in case the | |
1299 | * R_PRIMARY crashes now. | |
1300 | * Therefore we must send the barrier_ack after the barrier request was | |
1301 | * completed. */ | |
1302 | switch (mdev->write_ordering) { | |
1303 | case WO_bio_barrier: | |
1304 | case WO_none: | |
1305 | if (rv == FE_RECYCLED) | |
1306 | return TRUE; | |
1307 | break; | |
1308 | ||
1309 | case WO_bdev_flush: | |
1310 | case WO_drain_io: | |
367a8d73 PR |
1311 | if (rv == FE_STILL_LIVE) { |
1312 | set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &mdev->current_epoch->flags); | |
1313 | drbd_wait_ee_list_empty(mdev, &mdev->active_ee); | |
1314 | rv = drbd_flush_after_epoch(mdev, mdev->current_epoch); | |
1315 | } | |
b411b363 PR |
1316 | if (rv == FE_RECYCLED) |
1317 | return TRUE; | |
1318 | ||
1319 | /* The asender will send all the ACKs and barrier ACKs out, since | |
1320 | all EEs moved from the active_ee to the done_ee. We need to | |
1321 | provide a new epoch object for the EEs that come in soon */ | |
1322 | break; | |
1323 | } | |
1324 | ||
1325 | /* receiver context, in the writeout path of the other node. | |
1326 | * avoid potential distributed deadlock */ | |
1327 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1328 | if (!epoch) { | |
1329 | dev_warn(DEV, "Allocation of an epoch failed, slowing down\n"); | |
d3db7b48 | 1330 | issue_flush = !test_and_set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &mdev->current_epoch->flags); |
b411b363 PR |
1331 | drbd_wait_ee_list_empty(mdev, &mdev->active_ee); |
1332 | if (issue_flush) { | |
1333 | rv = drbd_flush_after_epoch(mdev, mdev->current_epoch); | |
1334 | if (rv == FE_RECYCLED) | |
1335 | return TRUE; | |
1336 | } | |
1337 | ||
1338 | drbd_wait_ee_list_empty(mdev, &mdev->done_ee); | |
1339 | ||
1340 | return TRUE; | |
1341 | } | |
1342 | ||
1343 | epoch->flags = 0; | |
1344 | atomic_set(&epoch->epoch_size, 0); | |
1345 | atomic_set(&epoch->active, 0); | |
1346 | ||
1347 | spin_lock(&mdev->epoch_lock); | |
1348 | if (atomic_read(&mdev->current_epoch->epoch_size)) { | |
1349 | list_add(&epoch->list, &mdev->current_epoch->list); | |
1350 | mdev->current_epoch = epoch; | |
1351 | mdev->epochs++; | |
b411b363 PR |
1352 | } else { |
1353 | /* The current_epoch got recycled while we allocated this one... */ | |
1354 | kfree(epoch); | |
1355 | } | |
1356 | spin_unlock(&mdev->epoch_lock); | |
1357 | ||
1358 | return TRUE; | |
1359 | } | |
1360 | ||
1361 | /* used from receive_RSDataReply (recv_resync_read) | |
1362 | * and from receive_Data */ | |
1363 | static struct drbd_epoch_entry * | |
1364 | read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, int data_size) __must_hold(local) | |
1365 | { | |
6666032a | 1366 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); |
b411b363 | 1367 | struct drbd_epoch_entry *e; |
b411b363 | 1368 | struct page *page; |
45bb912b | 1369 | int dgs, ds, rr; |
b411b363 PR |
1370 | void *dig_in = mdev->int_dig_in; |
1371 | void *dig_vv = mdev->int_dig_vv; | |
6b4388ac | 1372 | unsigned long *data; |
b411b363 PR |
1373 | |
1374 | dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ? | |
1375 | crypto_hash_digestsize(mdev->integrity_r_tfm) : 0; | |
1376 | ||
1377 | if (dgs) { | |
1378 | rr = drbd_recv(mdev, dig_in, dgs); | |
1379 | if (rr != dgs) { | |
1380 | dev_warn(DEV, "short read receiving data digest: read %d expected %d\n", | |
1381 | rr, dgs); | |
1382 | return NULL; | |
1383 | } | |
1384 | } | |
1385 | ||
1386 | data_size -= dgs; | |
1387 | ||
1388 | ERR_IF(data_size & 0x1ff) return NULL; | |
1389 | ERR_IF(data_size > DRBD_MAX_SEGMENT_SIZE) return NULL; | |
1390 | ||
6666032a LE |
1391 | /* even though we trust out peer, |
1392 | * we sometimes have to double check. */ | |
1393 | if (sector + (data_size>>9) > capacity) { | |
1394 | dev_err(DEV, "capacity: %llus < sector: %llus + size: %u\n", | |
1395 | (unsigned long long)capacity, | |
1396 | (unsigned long long)sector, data_size); | |
1397 | return NULL; | |
1398 | } | |
1399 | ||
b411b363 PR |
1400 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD |
1401 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
1402 | * which in turn might block on the other node at this very place. */ | |
1403 | e = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO); | |
1404 | if (!e) | |
1405 | return NULL; | |
45bb912b | 1406 | |
b411b363 | 1407 | ds = data_size; |
45bb912b LE |
1408 | page = e->pages; |
1409 | page_chain_for_each(page) { | |
1410 | unsigned len = min_t(int, ds, PAGE_SIZE); | |
6b4388ac | 1411 | data = kmap(page); |
45bb912b | 1412 | rr = drbd_recv(mdev, data, len); |
6b4388ac PR |
1413 | if (FAULT_ACTIVE(mdev, DRBD_FAULT_RECEIVE)) { |
1414 | dev_err(DEV, "Fault injection: Corrupting data on receive\n"); | |
1415 | data[0] = data[0] ^ (unsigned long)-1; | |
1416 | } | |
b411b363 | 1417 | kunmap(page); |
45bb912b | 1418 | if (rr != len) { |
b411b363 PR |
1419 | drbd_free_ee(mdev, e); |
1420 | dev_warn(DEV, "short read receiving data: read %d expected %d\n", | |
45bb912b | 1421 | rr, len); |
b411b363 PR |
1422 | return NULL; |
1423 | } | |
1424 | ds -= rr; | |
1425 | } | |
1426 | ||
1427 | if (dgs) { | |
45bb912b | 1428 | drbd_csum_ee(mdev, mdev->integrity_r_tfm, e, dig_vv); |
b411b363 PR |
1429 | if (memcmp(dig_in, dig_vv, dgs)) { |
1430 | dev_err(DEV, "Digest integrity check FAILED.\n"); | |
1431 | drbd_bcast_ee(mdev, "digest failed", | |
1432 | dgs, dig_in, dig_vv, e); | |
1433 | drbd_free_ee(mdev, e); | |
1434 | return NULL; | |
1435 | } | |
1436 | } | |
1437 | mdev->recv_cnt += data_size>>9; | |
1438 | return e; | |
1439 | } | |
1440 | ||
1441 | /* drbd_drain_block() just takes a data block | |
1442 | * out of the socket input buffer, and discards it. | |
1443 | */ | |
1444 | static int drbd_drain_block(struct drbd_conf *mdev, int data_size) | |
1445 | { | |
1446 | struct page *page; | |
1447 | int rr, rv = 1; | |
1448 | void *data; | |
1449 | ||
c3470cde LE |
1450 | if (!data_size) |
1451 | return TRUE; | |
1452 | ||
45bb912b | 1453 | page = drbd_pp_alloc(mdev, 1, 1); |
b411b363 PR |
1454 | |
1455 | data = kmap(page); | |
1456 | while (data_size) { | |
1457 | rr = drbd_recv(mdev, data, min_t(int, data_size, PAGE_SIZE)); | |
1458 | if (rr != min_t(int, data_size, PAGE_SIZE)) { | |
1459 | rv = 0; | |
1460 | dev_warn(DEV, "short read receiving data: read %d expected %d\n", | |
1461 | rr, min_t(int, data_size, PAGE_SIZE)); | |
1462 | break; | |
1463 | } | |
1464 | data_size -= rr; | |
1465 | } | |
1466 | kunmap(page); | |
435f0740 | 1467 | drbd_pp_free(mdev, page, 0); |
b411b363 PR |
1468 | return rv; |
1469 | } | |
1470 | ||
1471 | static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req, | |
1472 | sector_t sector, int data_size) | |
1473 | { | |
1474 | struct bio_vec *bvec; | |
1475 | struct bio *bio; | |
1476 | int dgs, rr, i, expect; | |
1477 | void *dig_in = mdev->int_dig_in; | |
1478 | void *dig_vv = mdev->int_dig_vv; | |
1479 | ||
1480 | dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ? | |
1481 | crypto_hash_digestsize(mdev->integrity_r_tfm) : 0; | |
1482 | ||
1483 | if (dgs) { | |
1484 | rr = drbd_recv(mdev, dig_in, dgs); | |
1485 | if (rr != dgs) { | |
1486 | dev_warn(DEV, "short read receiving data reply digest: read %d expected %d\n", | |
1487 | rr, dgs); | |
1488 | return 0; | |
1489 | } | |
1490 | } | |
1491 | ||
1492 | data_size -= dgs; | |
1493 | ||
1494 | /* optimistically update recv_cnt. if receiving fails below, | |
1495 | * we disconnect anyways, and counters will be reset. */ | |
1496 | mdev->recv_cnt += data_size>>9; | |
1497 | ||
1498 | bio = req->master_bio; | |
1499 | D_ASSERT(sector == bio->bi_sector); | |
1500 | ||
1501 | bio_for_each_segment(bvec, bio, i) { | |
1502 | expect = min_t(int, data_size, bvec->bv_len); | |
1503 | rr = drbd_recv(mdev, | |
1504 | kmap(bvec->bv_page)+bvec->bv_offset, | |
1505 | expect); | |
1506 | kunmap(bvec->bv_page); | |
1507 | if (rr != expect) { | |
1508 | dev_warn(DEV, "short read receiving data reply: " | |
1509 | "read %d expected %d\n", | |
1510 | rr, expect); | |
1511 | return 0; | |
1512 | } | |
1513 | data_size -= rr; | |
1514 | } | |
1515 | ||
1516 | if (dgs) { | |
45bb912b | 1517 | drbd_csum_bio(mdev, mdev->integrity_r_tfm, bio, dig_vv); |
b411b363 PR |
1518 | if (memcmp(dig_in, dig_vv, dgs)) { |
1519 | dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n"); | |
1520 | return 0; | |
1521 | } | |
1522 | } | |
1523 | ||
1524 | D_ASSERT(data_size == 0); | |
1525 | return 1; | |
1526 | } | |
1527 | ||
1528 | /* e_end_resync_block() is called via | |
1529 | * drbd_process_done_ee() by asender only */ | |
1530 | static int e_end_resync_block(struct drbd_conf *mdev, struct drbd_work *w, int unused) | |
1531 | { | |
1532 | struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; | |
1533 | sector_t sector = e->sector; | |
1534 | int ok; | |
1535 | ||
1536 | D_ASSERT(hlist_unhashed(&e->colision)); | |
1537 | ||
45bb912b | 1538 | if (likely((e->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1539 | drbd_set_in_sync(mdev, sector, e->size); |
1540 | ok = drbd_send_ack(mdev, P_RS_WRITE_ACK, e); | |
1541 | } else { | |
1542 | /* Record failure to sync */ | |
1543 | drbd_rs_failed_io(mdev, sector, e->size); | |
1544 | ||
1545 | ok = drbd_send_ack(mdev, P_NEG_ACK, e); | |
1546 | } | |
1547 | dec_unacked(mdev); | |
1548 | ||
1549 | return ok; | |
1550 | } | |
1551 | ||
1552 | static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local) | |
1553 | { | |
1554 | struct drbd_epoch_entry *e; | |
1555 | ||
1556 | e = read_in_block(mdev, ID_SYNCER, sector, data_size); | |
45bb912b LE |
1557 | if (!e) |
1558 | goto fail; | |
b411b363 PR |
1559 | |
1560 | dec_rs_pending(mdev); | |
1561 | ||
b411b363 PR |
1562 | inc_unacked(mdev); |
1563 | /* corresponding dec_unacked() in e_end_resync_block() | |
1564 | * respective _drbd_clear_done_ee */ | |
1565 | ||
45bb912b LE |
1566 | e->w.cb = e_end_resync_block; |
1567 | ||
b411b363 PR |
1568 | spin_lock_irq(&mdev->req_lock); |
1569 | list_add(&e->w.list, &mdev->sync_ee); | |
1570 | spin_unlock_irq(&mdev->req_lock); | |
1571 | ||
0f0601f4 | 1572 | atomic_add(data_size >> 9, &mdev->rs_sect_ev); |
45bb912b LE |
1573 | if (drbd_submit_ee(mdev, e, WRITE, DRBD_FAULT_RS_WR) == 0) |
1574 | return TRUE; | |
b411b363 | 1575 | |
45bb912b LE |
1576 | drbd_free_ee(mdev, e); |
1577 | fail: | |
1578 | put_ldev(mdev); | |
1579 | return FALSE; | |
b411b363 PR |
1580 | } |
1581 | ||
02918be2 | 1582 | static int receive_DataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
1583 | { |
1584 | struct drbd_request *req; | |
1585 | sector_t sector; | |
b411b363 | 1586 | int ok; |
02918be2 | 1587 | struct p_data *p = &mdev->data.rbuf.data; |
b411b363 PR |
1588 | |
1589 | sector = be64_to_cpu(p->sector); | |
1590 | ||
1591 | spin_lock_irq(&mdev->req_lock); | |
1592 | req = _ar_id_to_req(mdev, p->block_id, sector); | |
1593 | spin_unlock_irq(&mdev->req_lock); | |
1594 | if (unlikely(!req)) { | |
1595 | dev_err(DEV, "Got a corrupt block_id/sector pair(1).\n"); | |
1596 | return FALSE; | |
1597 | } | |
1598 | ||
1599 | /* hlist_del(&req->colision) is done in _req_may_be_done, to avoid | |
1600 | * special casing it there for the various failure cases. | |
1601 | * still no race with drbd_fail_pending_reads */ | |
1602 | ok = recv_dless_read(mdev, req, sector, data_size); | |
1603 | ||
1604 | if (ok) | |
1605 | req_mod(req, data_received); | |
1606 | /* else: nothing. handled from drbd_disconnect... | |
1607 | * I don't think we may complete this just yet | |
1608 | * in case we are "on-disconnect: freeze" */ | |
1609 | ||
1610 | return ok; | |
1611 | } | |
1612 | ||
02918be2 | 1613 | static int receive_RSDataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
1614 | { |
1615 | sector_t sector; | |
b411b363 | 1616 | int ok; |
02918be2 | 1617 | struct p_data *p = &mdev->data.rbuf.data; |
b411b363 PR |
1618 | |
1619 | sector = be64_to_cpu(p->sector); | |
1620 | D_ASSERT(p->block_id == ID_SYNCER); | |
1621 | ||
1622 | if (get_ldev(mdev)) { | |
1623 | /* data is submitted to disk within recv_resync_read. | |
1624 | * corresponding put_ldev done below on error, | |
1625 | * or in drbd_endio_write_sec. */ | |
1626 | ok = recv_resync_read(mdev, sector, data_size); | |
1627 | } else { | |
1628 | if (__ratelimit(&drbd_ratelimit_state)) | |
1629 | dev_err(DEV, "Can not write resync data to local disk.\n"); | |
1630 | ||
1631 | ok = drbd_drain_block(mdev, data_size); | |
1632 | ||
1633 | drbd_send_ack_dp(mdev, P_NEG_ACK, p); | |
1634 | } | |
1635 | ||
778f271d PR |
1636 | atomic_add(data_size >> 9, &mdev->rs_sect_in); |
1637 | ||
b411b363 PR |
1638 | return ok; |
1639 | } | |
1640 | ||
1641 | /* e_end_block() is called via drbd_process_done_ee(). | |
1642 | * this means this function only runs in the asender thread | |
1643 | */ | |
1644 | static int e_end_block(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
1645 | { | |
1646 | struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; | |
1647 | sector_t sector = e->sector; | |
1648 | struct drbd_epoch *epoch; | |
1649 | int ok = 1, pcmd; | |
1650 | ||
1651 | if (e->flags & EE_IS_BARRIER) { | |
1652 | epoch = previous_epoch(mdev, e->epoch); | |
1653 | if (epoch) | |
1654 | drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE + (cancel ? EV_CLEANUP : 0)); | |
1655 | } | |
1656 | ||
1657 | if (mdev->net_conf->wire_protocol == DRBD_PROT_C) { | |
45bb912b | 1658 | if (likely((e->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1659 | pcmd = (mdev->state.conn >= C_SYNC_SOURCE && |
1660 | mdev->state.conn <= C_PAUSED_SYNC_T && | |
1661 | e->flags & EE_MAY_SET_IN_SYNC) ? | |
1662 | P_RS_WRITE_ACK : P_WRITE_ACK; | |
1663 | ok &= drbd_send_ack(mdev, pcmd, e); | |
1664 | if (pcmd == P_RS_WRITE_ACK) | |
1665 | drbd_set_in_sync(mdev, sector, e->size); | |
1666 | } else { | |
1667 | ok = drbd_send_ack(mdev, P_NEG_ACK, e); | |
1668 | /* we expect it to be marked out of sync anyways... | |
1669 | * maybe assert this? */ | |
1670 | } | |
1671 | dec_unacked(mdev); | |
1672 | } | |
1673 | /* we delete from the conflict detection hash _after_ we sent out the | |
1674 | * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ | |
1675 | if (mdev->net_conf->two_primaries) { | |
1676 | spin_lock_irq(&mdev->req_lock); | |
1677 | D_ASSERT(!hlist_unhashed(&e->colision)); | |
1678 | hlist_del_init(&e->colision); | |
1679 | spin_unlock_irq(&mdev->req_lock); | |
1680 | } else { | |
1681 | D_ASSERT(hlist_unhashed(&e->colision)); | |
1682 | } | |
1683 | ||
1684 | drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); | |
1685 | ||
1686 | return ok; | |
1687 | } | |
1688 | ||
1689 | static int e_send_discard_ack(struct drbd_conf *mdev, struct drbd_work *w, int unused) | |
1690 | { | |
1691 | struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; | |
1692 | int ok = 1; | |
1693 | ||
1694 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
1695 | ok = drbd_send_ack(mdev, P_DISCARD_ACK, e); | |
1696 | ||
1697 | spin_lock_irq(&mdev->req_lock); | |
1698 | D_ASSERT(!hlist_unhashed(&e->colision)); | |
1699 | hlist_del_init(&e->colision); | |
1700 | spin_unlock_irq(&mdev->req_lock); | |
1701 | ||
1702 | dec_unacked(mdev); | |
1703 | ||
1704 | return ok; | |
1705 | } | |
1706 | ||
1707 | /* Called from receive_Data. | |
1708 | * Synchronize packets on sock with packets on msock. | |
1709 | * | |
1710 | * This is here so even when a P_DATA packet traveling via sock overtook an Ack | |
1711 | * packet traveling on msock, they are still processed in the order they have | |
1712 | * been sent. | |
1713 | * | |
1714 | * Note: we don't care for Ack packets overtaking P_DATA packets. | |
1715 | * | |
1716 | * In case packet_seq is larger than mdev->peer_seq number, there are | |
1717 | * outstanding packets on the msock. We wait for them to arrive. | |
1718 | * In case we are the logically next packet, we update mdev->peer_seq | |
1719 | * ourselves. Correctly handles 32bit wrap around. | |
1720 | * | |
1721 | * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, | |
1722 | * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds | |
1723 | * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have | |
1724 | * 1<<9 == 512 seconds aka ages for the 32bit wrap around... | |
1725 | * | |
1726 | * returns 0 if we may process the packet, | |
1727 | * -ERESTARTSYS if we were interrupted (by disconnect signal). */ | |
1728 | static int drbd_wait_peer_seq(struct drbd_conf *mdev, const u32 packet_seq) | |
1729 | { | |
1730 | DEFINE_WAIT(wait); | |
1731 | unsigned int p_seq; | |
1732 | long timeout; | |
1733 | int ret = 0; | |
1734 | spin_lock(&mdev->peer_seq_lock); | |
1735 | for (;;) { | |
1736 | prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE); | |
1737 | if (seq_le(packet_seq, mdev->peer_seq+1)) | |
1738 | break; | |
1739 | if (signal_pending(current)) { | |
1740 | ret = -ERESTARTSYS; | |
1741 | break; | |
1742 | } | |
1743 | p_seq = mdev->peer_seq; | |
1744 | spin_unlock(&mdev->peer_seq_lock); | |
1745 | timeout = schedule_timeout(30*HZ); | |
1746 | spin_lock(&mdev->peer_seq_lock); | |
1747 | if (timeout == 0 && p_seq == mdev->peer_seq) { | |
1748 | ret = -ETIMEDOUT; | |
1749 | dev_err(DEV, "ASSERT FAILED waited 30 seconds for sequence update, forcing reconnect\n"); | |
1750 | break; | |
1751 | } | |
1752 | } | |
1753 | finish_wait(&mdev->seq_wait, &wait); | |
1754 | if (mdev->peer_seq+1 == packet_seq) | |
1755 | mdev->peer_seq++; | |
1756 | spin_unlock(&mdev->peer_seq_lock); | |
1757 | return ret; | |
1758 | } | |
1759 | ||
76d2e7ec PR |
1760 | static unsigned long write_flags_to_bio(struct drbd_conf *mdev, u32 dpf) |
1761 | { | |
1762 | if (mdev->agreed_pro_version >= 95) | |
1763 | return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | | |
1764 | (dpf & DP_UNPLUG ? REQ_UNPLUG : 0) | | |
1765 | (dpf & DP_FUA ? REQ_FUA : 0) | | |
1766 | (dpf & DP_FLUSH ? REQ_FUA : 0) | | |
1767 | (dpf & DP_DISCARD ? REQ_DISCARD : 0); | |
1768 | else | |
1769 | return dpf & DP_RW_SYNC ? (REQ_SYNC | REQ_UNPLUG) : 0; | |
1770 | } | |
1771 | ||
b411b363 | 1772 | /* mirrored write */ |
02918be2 | 1773 | static int receive_Data(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
1774 | { |
1775 | sector_t sector; | |
1776 | struct drbd_epoch_entry *e; | |
02918be2 | 1777 | struct p_data *p = &mdev->data.rbuf.data; |
b411b363 PR |
1778 | int rw = WRITE; |
1779 | u32 dp_flags; | |
1780 | ||
b411b363 PR |
1781 | if (!get_ldev(mdev)) { |
1782 | if (__ratelimit(&drbd_ratelimit_state)) | |
1783 | dev_err(DEV, "Can not write mirrored data block " | |
1784 | "to local disk.\n"); | |
1785 | spin_lock(&mdev->peer_seq_lock); | |
1786 | if (mdev->peer_seq+1 == be32_to_cpu(p->seq_num)) | |
1787 | mdev->peer_seq++; | |
1788 | spin_unlock(&mdev->peer_seq_lock); | |
1789 | ||
1790 | drbd_send_ack_dp(mdev, P_NEG_ACK, p); | |
1791 | atomic_inc(&mdev->current_epoch->epoch_size); | |
1792 | return drbd_drain_block(mdev, data_size); | |
1793 | } | |
1794 | ||
1795 | /* get_ldev(mdev) successful. | |
1796 | * Corresponding put_ldev done either below (on various errors), | |
1797 | * or in drbd_endio_write_sec, if we successfully submit the data at | |
1798 | * the end of this function. */ | |
1799 | ||
1800 | sector = be64_to_cpu(p->sector); | |
1801 | e = read_in_block(mdev, p->block_id, sector, data_size); | |
1802 | if (!e) { | |
1803 | put_ldev(mdev); | |
1804 | return FALSE; | |
1805 | } | |
1806 | ||
b411b363 PR |
1807 | e->w.cb = e_end_block; |
1808 | ||
1809 | spin_lock(&mdev->epoch_lock); | |
1810 | e->epoch = mdev->current_epoch; | |
1811 | atomic_inc(&e->epoch->epoch_size); | |
1812 | atomic_inc(&e->epoch->active); | |
1813 | ||
1814 | if (mdev->write_ordering == WO_bio_barrier && atomic_read(&e->epoch->epoch_size) == 1) { | |
1815 | struct drbd_epoch *epoch; | |
1816 | /* Issue a barrier if we start a new epoch, and the previous epoch | |
1817 | was not a epoch containing a single request which already was | |
1818 | a Barrier. */ | |
1819 | epoch = list_entry(e->epoch->list.prev, struct drbd_epoch, list); | |
1820 | if (epoch == e->epoch) { | |
1821 | set_bit(DE_CONTAINS_A_BARRIER, &e->epoch->flags); | |
7b6d91da | 1822 | rw |= REQ_HARDBARRIER; |
b411b363 PR |
1823 | e->flags |= EE_IS_BARRIER; |
1824 | } else { | |
1825 | if (atomic_read(&epoch->epoch_size) > 1 || | |
1826 | !test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags)) { | |
1827 | set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags); | |
b411b363 | 1828 | set_bit(DE_CONTAINS_A_BARRIER, &e->epoch->flags); |
7b6d91da | 1829 | rw |= REQ_HARDBARRIER; |
b411b363 PR |
1830 | e->flags |= EE_IS_BARRIER; |
1831 | } | |
1832 | } | |
1833 | } | |
1834 | spin_unlock(&mdev->epoch_lock); | |
1835 | ||
1836 | dp_flags = be32_to_cpu(p->dp_flags); | |
76d2e7ec PR |
1837 | rw |= write_flags_to_bio(mdev, dp_flags); |
1838 | ||
b411b363 PR |
1839 | if (dp_flags & DP_MAY_SET_IN_SYNC) |
1840 | e->flags |= EE_MAY_SET_IN_SYNC; | |
1841 | ||
1842 | /* I'm the receiver, I do hold a net_cnt reference. */ | |
1843 | if (!mdev->net_conf->two_primaries) { | |
1844 | spin_lock_irq(&mdev->req_lock); | |
1845 | } else { | |
1846 | /* don't get the req_lock yet, | |
1847 | * we may sleep in drbd_wait_peer_seq */ | |
1848 | const int size = e->size; | |
1849 | const int discard = test_bit(DISCARD_CONCURRENT, &mdev->flags); | |
1850 | DEFINE_WAIT(wait); | |
1851 | struct drbd_request *i; | |
1852 | struct hlist_node *n; | |
1853 | struct hlist_head *slot; | |
1854 | int first; | |
1855 | ||
1856 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
1857 | BUG_ON(mdev->ee_hash == NULL); | |
1858 | BUG_ON(mdev->tl_hash == NULL); | |
1859 | ||
1860 | /* conflict detection and handling: | |
1861 | * 1. wait on the sequence number, | |
1862 | * in case this data packet overtook ACK packets. | |
1863 | * 2. check our hash tables for conflicting requests. | |
1864 | * we only need to walk the tl_hash, since an ee can not | |
1865 | * have a conflict with an other ee: on the submitting | |
1866 | * node, the corresponding req had already been conflicting, | |
1867 | * and a conflicting req is never sent. | |
1868 | * | |
1869 | * Note: for two_primaries, we are protocol C, | |
1870 | * so there cannot be any request that is DONE | |
1871 | * but still on the transfer log. | |
1872 | * | |
1873 | * unconditionally add to the ee_hash. | |
1874 | * | |
1875 | * if no conflicting request is found: | |
1876 | * submit. | |
1877 | * | |
1878 | * if any conflicting request is found | |
1879 | * that has not yet been acked, | |
1880 | * AND I have the "discard concurrent writes" flag: | |
1881 | * queue (via done_ee) the P_DISCARD_ACK; OUT. | |
1882 | * | |
1883 | * if any conflicting request is found: | |
1884 | * block the receiver, waiting on misc_wait | |
1885 | * until no more conflicting requests are there, | |
1886 | * or we get interrupted (disconnect). | |
1887 | * | |
1888 | * we do not just write after local io completion of those | |
1889 | * requests, but only after req is done completely, i.e. | |
1890 | * we wait for the P_DISCARD_ACK to arrive! | |
1891 | * | |
1892 | * then proceed normally, i.e. submit. | |
1893 | */ | |
1894 | if (drbd_wait_peer_seq(mdev, be32_to_cpu(p->seq_num))) | |
1895 | goto out_interrupted; | |
1896 | ||
1897 | spin_lock_irq(&mdev->req_lock); | |
1898 | ||
1899 | hlist_add_head(&e->colision, ee_hash_slot(mdev, sector)); | |
1900 | ||
1901 | #define OVERLAPS overlaps(i->sector, i->size, sector, size) | |
1902 | slot = tl_hash_slot(mdev, sector); | |
1903 | first = 1; | |
1904 | for (;;) { | |
1905 | int have_unacked = 0; | |
1906 | int have_conflict = 0; | |
1907 | prepare_to_wait(&mdev->misc_wait, &wait, | |
1908 | TASK_INTERRUPTIBLE); | |
1909 | hlist_for_each_entry(i, n, slot, colision) { | |
1910 | if (OVERLAPS) { | |
1911 | /* only ALERT on first iteration, | |
1912 | * we may be woken up early... */ | |
1913 | if (first) | |
1914 | dev_alert(DEV, "%s[%u] Concurrent local write detected!" | |
1915 | " new: %llus +%u; pending: %llus +%u\n", | |
1916 | current->comm, current->pid, | |
1917 | (unsigned long long)sector, size, | |
1918 | (unsigned long long)i->sector, i->size); | |
1919 | if (i->rq_state & RQ_NET_PENDING) | |
1920 | ++have_unacked; | |
1921 | ++have_conflict; | |
1922 | } | |
1923 | } | |
1924 | #undef OVERLAPS | |
1925 | if (!have_conflict) | |
1926 | break; | |
1927 | ||
1928 | /* Discard Ack only for the _first_ iteration */ | |
1929 | if (first && discard && have_unacked) { | |
1930 | dev_alert(DEV, "Concurrent write! [DISCARD BY FLAG] sec=%llus\n", | |
1931 | (unsigned long long)sector); | |
1932 | inc_unacked(mdev); | |
1933 | e->w.cb = e_send_discard_ack; | |
1934 | list_add_tail(&e->w.list, &mdev->done_ee); | |
1935 | ||
1936 | spin_unlock_irq(&mdev->req_lock); | |
1937 | ||
1938 | /* we could probably send that P_DISCARD_ACK ourselves, | |
1939 | * but I don't like the receiver using the msock */ | |
1940 | ||
1941 | put_ldev(mdev); | |
1942 | wake_asender(mdev); | |
1943 | finish_wait(&mdev->misc_wait, &wait); | |
1944 | return TRUE; | |
1945 | } | |
1946 | ||
1947 | if (signal_pending(current)) { | |
1948 | hlist_del_init(&e->colision); | |
1949 | ||
1950 | spin_unlock_irq(&mdev->req_lock); | |
1951 | ||
1952 | finish_wait(&mdev->misc_wait, &wait); | |
1953 | goto out_interrupted; | |
1954 | } | |
1955 | ||
1956 | spin_unlock_irq(&mdev->req_lock); | |
1957 | if (first) { | |
1958 | first = 0; | |
1959 | dev_alert(DEV, "Concurrent write! [W AFTERWARDS] " | |
1960 | "sec=%llus\n", (unsigned long long)sector); | |
1961 | } else if (discard) { | |
1962 | /* we had none on the first iteration. | |
1963 | * there must be none now. */ | |
1964 | D_ASSERT(have_unacked == 0); | |
1965 | } | |
1966 | schedule(); | |
1967 | spin_lock_irq(&mdev->req_lock); | |
1968 | } | |
1969 | finish_wait(&mdev->misc_wait, &wait); | |
1970 | } | |
1971 | ||
1972 | list_add(&e->w.list, &mdev->active_ee); | |
1973 | spin_unlock_irq(&mdev->req_lock); | |
1974 | ||
1975 | switch (mdev->net_conf->wire_protocol) { | |
1976 | case DRBD_PROT_C: | |
1977 | inc_unacked(mdev); | |
1978 | /* corresponding dec_unacked() in e_end_block() | |
1979 | * respective _drbd_clear_done_ee */ | |
1980 | break; | |
1981 | case DRBD_PROT_B: | |
1982 | /* I really don't like it that the receiver thread | |
1983 | * sends on the msock, but anyways */ | |
1984 | drbd_send_ack(mdev, P_RECV_ACK, e); | |
1985 | break; | |
1986 | case DRBD_PROT_A: | |
1987 | /* nothing to do */ | |
1988 | break; | |
1989 | } | |
1990 | ||
1991 | if (mdev->state.pdsk == D_DISKLESS) { | |
1992 | /* In case we have the only disk of the cluster, */ | |
1993 | drbd_set_out_of_sync(mdev, e->sector, e->size); | |
1994 | e->flags |= EE_CALL_AL_COMPLETE_IO; | |
1995 | drbd_al_begin_io(mdev, e->sector); | |
1996 | } | |
1997 | ||
45bb912b LE |
1998 | if (drbd_submit_ee(mdev, e, rw, DRBD_FAULT_DT_WR) == 0) |
1999 | return TRUE; | |
b411b363 PR |
2000 | |
2001 | out_interrupted: | |
2002 | /* yes, the epoch_size now is imbalanced. | |
2003 | * but we drop the connection anyways, so we don't have a chance to | |
2004 | * receive a barrier... atomic_inc(&mdev->epoch_size); */ | |
2005 | put_ldev(mdev); | |
2006 | drbd_free_ee(mdev, e); | |
2007 | return FALSE; | |
2008 | } | |
2009 | ||
0f0601f4 LE |
2010 | /* We may throttle resync, if the lower device seems to be busy, |
2011 | * and current sync rate is above c_min_rate. | |
2012 | * | |
2013 | * To decide whether or not the lower device is busy, we use a scheme similar | |
2014 | * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" | |
2015 | * (more than 64 sectors) of activity we cannot account for with our own resync | |
2016 | * activity, it obviously is "busy". | |
2017 | * | |
2018 | * The current sync rate used here uses only the most recent two step marks, | |
2019 | * to have a short time average so we can react faster. | |
2020 | */ | |
2021 | int drbd_rs_should_slow_down(struct drbd_conf *mdev) | |
2022 | { | |
2023 | struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk; | |
2024 | unsigned long db, dt, dbdt; | |
2025 | int curr_events; | |
2026 | int throttle = 0; | |
2027 | ||
2028 | /* feature disabled? */ | |
2029 | if (mdev->sync_conf.c_min_rate == 0) | |
2030 | return 0; | |
2031 | ||
2032 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + | |
2033 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
2034 | atomic_read(&mdev->rs_sect_ev); | |
2035 | if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) { | |
2036 | unsigned long rs_left; | |
2037 | int i; | |
2038 | ||
2039 | mdev->rs_last_events = curr_events; | |
2040 | ||
2041 | /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, | |
2042 | * approx. */ | |
2043 | i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-2) % DRBD_SYNC_MARKS; | |
2044 | rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; | |
2045 | ||
2046 | dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ; | |
2047 | if (!dt) | |
2048 | dt++; | |
2049 | db = mdev->rs_mark_left[i] - rs_left; | |
2050 | dbdt = Bit2KB(db/dt); | |
2051 | ||
2052 | if (dbdt > mdev->sync_conf.c_min_rate) | |
2053 | throttle = 1; | |
2054 | } | |
2055 | return throttle; | |
2056 | } | |
2057 | ||
2058 | ||
02918be2 | 2059 | static int receive_DataRequest(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int digest_size) |
b411b363 PR |
2060 | { |
2061 | sector_t sector; | |
2062 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); | |
2063 | struct drbd_epoch_entry *e; | |
2064 | struct digest_info *di = NULL; | |
02918be2 | 2065 | int size; |
b411b363 | 2066 | unsigned int fault_type; |
02918be2 | 2067 | struct p_block_req *p = &mdev->data.rbuf.block_req; |
b411b363 PR |
2068 | |
2069 | sector = be64_to_cpu(p->sector); | |
2070 | size = be32_to_cpu(p->blksize); | |
2071 | ||
2072 | if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_SEGMENT_SIZE) { | |
2073 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, | |
2074 | (unsigned long long)sector, size); | |
2075 | return FALSE; | |
2076 | } | |
2077 | if (sector + (size>>9) > capacity) { | |
2078 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, | |
2079 | (unsigned long long)sector, size); | |
2080 | return FALSE; | |
2081 | } | |
2082 | ||
2083 | if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) { | |
2084 | if (__ratelimit(&drbd_ratelimit_state)) | |
2085 | dev_err(DEV, "Can not satisfy peer's read request, " | |
2086 | "no local data.\n"); | |
02918be2 | 2087 | drbd_send_ack_rp(mdev, cmd == P_DATA_REQUEST ? P_NEG_DREPLY : |
b411b363 | 2088 | P_NEG_RS_DREPLY , p); |
a821cc4a LE |
2089 | /* drain possibly payload */ |
2090 | return drbd_drain_block(mdev, digest_size); | |
b411b363 PR |
2091 | } |
2092 | ||
2093 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD | |
2094 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
2095 | * which in turn might block on the other node at this very place. */ | |
2096 | e = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO); | |
2097 | if (!e) { | |
2098 | put_ldev(mdev); | |
2099 | return FALSE; | |
2100 | } | |
2101 | ||
02918be2 | 2102 | switch (cmd) { |
b411b363 PR |
2103 | case P_DATA_REQUEST: |
2104 | e->w.cb = w_e_end_data_req; | |
2105 | fault_type = DRBD_FAULT_DT_RD; | |
80a40e43 LE |
2106 | /* application IO, don't drbd_rs_begin_io */ |
2107 | goto submit; | |
2108 | ||
b411b363 PR |
2109 | case P_RS_DATA_REQUEST: |
2110 | e->w.cb = w_e_end_rsdata_req; | |
2111 | fault_type = DRBD_FAULT_RS_RD; | |
b411b363 PR |
2112 | break; |
2113 | ||
2114 | case P_OV_REPLY: | |
2115 | case P_CSUM_RS_REQUEST: | |
2116 | fault_type = DRBD_FAULT_RS_RD; | |
b411b363 PR |
2117 | di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO); |
2118 | if (!di) | |
2119 | goto out_free_e; | |
2120 | ||
2121 | di->digest_size = digest_size; | |
2122 | di->digest = (((char *)di)+sizeof(struct digest_info)); | |
2123 | ||
c36c3ced LE |
2124 | e->digest = di; |
2125 | e->flags |= EE_HAS_DIGEST; | |
2126 | ||
b411b363 PR |
2127 | if (drbd_recv(mdev, di->digest, digest_size) != digest_size) |
2128 | goto out_free_e; | |
2129 | ||
02918be2 | 2130 | if (cmd == P_CSUM_RS_REQUEST) { |
b411b363 PR |
2131 | D_ASSERT(mdev->agreed_pro_version >= 89); |
2132 | e->w.cb = w_e_end_csum_rs_req; | |
02918be2 | 2133 | } else if (cmd == P_OV_REPLY) { |
b411b363 PR |
2134 | e->w.cb = w_e_end_ov_reply; |
2135 | dec_rs_pending(mdev); | |
0f0601f4 LE |
2136 | /* drbd_rs_begin_io done when we sent this request, |
2137 | * but accounting still needs to be done. */ | |
2138 | goto submit_for_resync; | |
b411b363 PR |
2139 | } |
2140 | break; | |
2141 | ||
2142 | case P_OV_REQUEST: | |
2143 | if (mdev->state.conn >= C_CONNECTED && | |
2144 | mdev->state.conn != C_VERIFY_T) | |
2145 | dev_warn(DEV, "ASSERT FAILED: got P_OV_REQUEST while being %s\n", | |
2146 | drbd_conn_str(mdev->state.conn)); | |
2147 | if (mdev->ov_start_sector == ~(sector_t)0 && | |
2148 | mdev->agreed_pro_version >= 90) { | |
2149 | mdev->ov_start_sector = sector; | |
2150 | mdev->ov_position = sector; | |
2151 | mdev->ov_left = mdev->rs_total - BM_SECT_TO_BIT(sector); | |
2152 | dev_info(DEV, "Online Verify start sector: %llu\n", | |
2153 | (unsigned long long)sector); | |
2154 | } | |
2155 | e->w.cb = w_e_end_ov_req; | |
2156 | fault_type = DRBD_FAULT_RS_RD; | |
b411b363 PR |
2157 | break; |
2158 | ||
b411b363 PR |
2159 | default: |
2160 | dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", | |
02918be2 | 2161 | cmdname(cmd)); |
b411b363 | 2162 | fault_type = DRBD_FAULT_MAX; |
80a40e43 | 2163 | goto out_free_e; |
b411b363 PR |
2164 | } |
2165 | ||
0f0601f4 LE |
2166 | /* Throttle, drbd_rs_begin_io and submit should become asynchronous |
2167 | * wrt the receiver, but it is not as straightforward as it may seem. | |
2168 | * Various places in the resync start and stop logic assume resync | |
2169 | * requests are processed in order, requeuing this on the worker thread | |
2170 | * introduces a bunch of new code for synchronization between threads. | |
2171 | * | |
2172 | * Unlimited throttling before drbd_rs_begin_io may stall the resync | |
2173 | * "forever", throttling after drbd_rs_begin_io will lock that extent | |
2174 | * for application writes for the same time. For now, just throttle | |
2175 | * here, where the rest of the code expects the receiver to sleep for | |
2176 | * a while, anyways. | |
2177 | */ | |
2178 | ||
2179 | /* Throttle before drbd_rs_begin_io, as that locks out application IO; | |
2180 | * this defers syncer requests for some time, before letting at least | |
2181 | * on request through. The resync controller on the receiving side | |
2182 | * will adapt to the incoming rate accordingly. | |
2183 | * | |
2184 | * We cannot throttle here if remote is Primary/SyncTarget: | |
2185 | * we would also throttle its application reads. | |
2186 | * In that case, throttling is done on the SyncTarget only. | |
2187 | */ | |
2188 | if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev)) | |
2189 | msleep(100); | |
80a40e43 LE |
2190 | if (drbd_rs_begin_io(mdev, e->sector)) |
2191 | goto out_free_e; | |
b411b363 | 2192 | |
0f0601f4 LE |
2193 | submit_for_resync: |
2194 | atomic_add(size >> 9, &mdev->rs_sect_ev); | |
2195 | ||
80a40e43 | 2196 | submit: |
b411b363 | 2197 | inc_unacked(mdev); |
80a40e43 LE |
2198 | spin_lock_irq(&mdev->req_lock); |
2199 | list_add_tail(&e->w.list, &mdev->read_ee); | |
2200 | spin_unlock_irq(&mdev->req_lock); | |
b411b363 | 2201 | |
45bb912b LE |
2202 | if (drbd_submit_ee(mdev, e, READ, fault_type) == 0) |
2203 | return TRUE; | |
b411b363 PR |
2204 | |
2205 | out_free_e: | |
b411b363 PR |
2206 | put_ldev(mdev); |
2207 | drbd_free_ee(mdev, e); | |
2208 | return FALSE; | |
2209 | } | |
2210 | ||
2211 | static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local) | |
2212 | { | |
2213 | int self, peer, rv = -100; | |
2214 | unsigned long ch_self, ch_peer; | |
2215 | ||
2216 | self = mdev->ldev->md.uuid[UI_BITMAP] & 1; | |
2217 | peer = mdev->p_uuid[UI_BITMAP] & 1; | |
2218 | ||
2219 | ch_peer = mdev->p_uuid[UI_SIZE]; | |
2220 | ch_self = mdev->comm_bm_set; | |
2221 | ||
2222 | switch (mdev->net_conf->after_sb_0p) { | |
2223 | case ASB_CONSENSUS: | |
2224 | case ASB_DISCARD_SECONDARY: | |
2225 | case ASB_CALL_HELPER: | |
2226 | dev_err(DEV, "Configuration error.\n"); | |
2227 | break; | |
2228 | case ASB_DISCONNECT: | |
2229 | break; | |
2230 | case ASB_DISCARD_YOUNGER_PRI: | |
2231 | if (self == 0 && peer == 1) { | |
2232 | rv = -1; | |
2233 | break; | |
2234 | } | |
2235 | if (self == 1 && peer == 0) { | |
2236 | rv = 1; | |
2237 | break; | |
2238 | } | |
2239 | /* Else fall through to one of the other strategies... */ | |
2240 | case ASB_DISCARD_OLDER_PRI: | |
2241 | if (self == 0 && peer == 1) { | |
2242 | rv = 1; | |
2243 | break; | |
2244 | } | |
2245 | if (self == 1 && peer == 0) { | |
2246 | rv = -1; | |
2247 | break; | |
2248 | } | |
2249 | /* Else fall through to one of the other strategies... */ | |
ad19bf6e | 2250 | dev_warn(DEV, "Discard younger/older primary did not find a decision\n" |
b411b363 PR |
2251 | "Using discard-least-changes instead\n"); |
2252 | case ASB_DISCARD_ZERO_CHG: | |
2253 | if (ch_peer == 0 && ch_self == 0) { | |
2254 | rv = test_bit(DISCARD_CONCURRENT, &mdev->flags) | |
2255 | ? -1 : 1; | |
2256 | break; | |
2257 | } else { | |
2258 | if (ch_peer == 0) { rv = 1; break; } | |
2259 | if (ch_self == 0) { rv = -1; break; } | |
2260 | } | |
2261 | if (mdev->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG) | |
2262 | break; | |
2263 | case ASB_DISCARD_LEAST_CHG: | |
2264 | if (ch_self < ch_peer) | |
2265 | rv = -1; | |
2266 | else if (ch_self > ch_peer) | |
2267 | rv = 1; | |
2268 | else /* ( ch_self == ch_peer ) */ | |
2269 | /* Well, then use something else. */ | |
2270 | rv = test_bit(DISCARD_CONCURRENT, &mdev->flags) | |
2271 | ? -1 : 1; | |
2272 | break; | |
2273 | case ASB_DISCARD_LOCAL: | |
2274 | rv = -1; | |
2275 | break; | |
2276 | case ASB_DISCARD_REMOTE: | |
2277 | rv = 1; | |
2278 | } | |
2279 | ||
2280 | return rv; | |
2281 | } | |
2282 | ||
2283 | static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local) | |
2284 | { | |
2285 | int self, peer, hg, rv = -100; | |
2286 | ||
2287 | self = mdev->ldev->md.uuid[UI_BITMAP] & 1; | |
2288 | peer = mdev->p_uuid[UI_BITMAP] & 1; | |
2289 | ||
2290 | switch (mdev->net_conf->after_sb_1p) { | |
2291 | case ASB_DISCARD_YOUNGER_PRI: | |
2292 | case ASB_DISCARD_OLDER_PRI: | |
2293 | case ASB_DISCARD_LEAST_CHG: | |
2294 | case ASB_DISCARD_LOCAL: | |
2295 | case ASB_DISCARD_REMOTE: | |
2296 | dev_err(DEV, "Configuration error.\n"); | |
2297 | break; | |
2298 | case ASB_DISCONNECT: | |
2299 | break; | |
2300 | case ASB_CONSENSUS: | |
2301 | hg = drbd_asb_recover_0p(mdev); | |
2302 | if (hg == -1 && mdev->state.role == R_SECONDARY) | |
2303 | rv = hg; | |
2304 | if (hg == 1 && mdev->state.role == R_PRIMARY) | |
2305 | rv = hg; | |
2306 | break; | |
2307 | case ASB_VIOLENTLY: | |
2308 | rv = drbd_asb_recover_0p(mdev); | |
2309 | break; | |
2310 | case ASB_DISCARD_SECONDARY: | |
2311 | return mdev->state.role == R_PRIMARY ? 1 : -1; | |
2312 | case ASB_CALL_HELPER: | |
2313 | hg = drbd_asb_recover_0p(mdev); | |
2314 | if (hg == -1 && mdev->state.role == R_PRIMARY) { | |
2315 | self = drbd_set_role(mdev, R_SECONDARY, 0); | |
2316 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, | |
2317 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2318 | * we do not need to wait for the after state change work either. */ | |
2319 | self = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); | |
2320 | if (self != SS_SUCCESS) { | |
2321 | drbd_khelper(mdev, "pri-lost-after-sb"); | |
2322 | } else { | |
2323 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2324 | rv = hg; | |
2325 | } | |
2326 | } else | |
2327 | rv = hg; | |
2328 | } | |
2329 | ||
2330 | return rv; | |
2331 | } | |
2332 | ||
2333 | static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local) | |
2334 | { | |
2335 | int self, peer, hg, rv = -100; | |
2336 | ||
2337 | self = mdev->ldev->md.uuid[UI_BITMAP] & 1; | |
2338 | peer = mdev->p_uuid[UI_BITMAP] & 1; | |
2339 | ||
2340 | switch (mdev->net_conf->after_sb_2p) { | |
2341 | case ASB_DISCARD_YOUNGER_PRI: | |
2342 | case ASB_DISCARD_OLDER_PRI: | |
2343 | case ASB_DISCARD_LEAST_CHG: | |
2344 | case ASB_DISCARD_LOCAL: | |
2345 | case ASB_DISCARD_REMOTE: | |
2346 | case ASB_CONSENSUS: | |
2347 | case ASB_DISCARD_SECONDARY: | |
2348 | dev_err(DEV, "Configuration error.\n"); | |
2349 | break; | |
2350 | case ASB_VIOLENTLY: | |
2351 | rv = drbd_asb_recover_0p(mdev); | |
2352 | break; | |
2353 | case ASB_DISCONNECT: | |
2354 | break; | |
2355 | case ASB_CALL_HELPER: | |
2356 | hg = drbd_asb_recover_0p(mdev); | |
2357 | if (hg == -1) { | |
2358 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, | |
2359 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2360 | * we do not need to wait for the after state change work either. */ | |
2361 | self = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); | |
2362 | if (self != SS_SUCCESS) { | |
2363 | drbd_khelper(mdev, "pri-lost-after-sb"); | |
2364 | } else { | |
2365 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2366 | rv = hg; | |
2367 | } | |
2368 | } else | |
2369 | rv = hg; | |
2370 | } | |
2371 | ||
2372 | return rv; | |
2373 | } | |
2374 | ||
2375 | static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid, | |
2376 | u64 bits, u64 flags) | |
2377 | { | |
2378 | if (!uuid) { | |
2379 | dev_info(DEV, "%s uuid info vanished while I was looking!\n", text); | |
2380 | return; | |
2381 | } | |
2382 | dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", | |
2383 | text, | |
2384 | (unsigned long long)uuid[UI_CURRENT], | |
2385 | (unsigned long long)uuid[UI_BITMAP], | |
2386 | (unsigned long long)uuid[UI_HISTORY_START], | |
2387 | (unsigned long long)uuid[UI_HISTORY_END], | |
2388 | (unsigned long long)bits, | |
2389 | (unsigned long long)flags); | |
2390 | } | |
2391 | ||
2392 | /* | |
2393 | 100 after split brain try auto recover | |
2394 | 2 C_SYNC_SOURCE set BitMap | |
2395 | 1 C_SYNC_SOURCE use BitMap | |
2396 | 0 no Sync | |
2397 | -1 C_SYNC_TARGET use BitMap | |
2398 | -2 C_SYNC_TARGET set BitMap | |
2399 | -100 after split brain, disconnect | |
2400 | -1000 unrelated data | |
2401 | */ | |
2402 | static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local) | |
2403 | { | |
2404 | u64 self, peer; | |
2405 | int i, j; | |
2406 | ||
2407 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2408 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2409 | ||
2410 | *rule_nr = 10; | |
2411 | if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) | |
2412 | return 0; | |
2413 | ||
2414 | *rule_nr = 20; | |
2415 | if ((self == UUID_JUST_CREATED || self == (u64)0) && | |
2416 | peer != UUID_JUST_CREATED) | |
2417 | return -2; | |
2418 | ||
2419 | *rule_nr = 30; | |
2420 | if (self != UUID_JUST_CREATED && | |
2421 | (peer == UUID_JUST_CREATED || peer == (u64)0)) | |
2422 | return 2; | |
2423 | ||
2424 | if (self == peer) { | |
2425 | int rct, dc; /* roles at crash time */ | |
2426 | ||
2427 | if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) { | |
2428 | ||
2429 | if (mdev->agreed_pro_version < 91) | |
2430 | return -1001; | |
2431 | ||
2432 | if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) && | |
2433 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { | |
2434 | dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n"); | |
2435 | drbd_uuid_set_bm(mdev, 0UL); | |
2436 | ||
2437 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, | |
2438 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2439 | *rule_nr = 34; | |
2440 | } else { | |
2441 | dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n"); | |
2442 | *rule_nr = 36; | |
2443 | } | |
2444 | ||
2445 | return 1; | |
2446 | } | |
2447 | ||
2448 | if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) { | |
2449 | ||
2450 | if (mdev->agreed_pro_version < 91) | |
2451 | return -1001; | |
2452 | ||
2453 | if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) && | |
2454 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) { | |
2455 | dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); | |
2456 | ||
2457 | mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START]; | |
2458 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP]; | |
2459 | mdev->p_uuid[UI_BITMAP] = 0UL; | |
2460 | ||
2461 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2462 | *rule_nr = 35; | |
2463 | } else { | |
2464 | dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n"); | |
2465 | *rule_nr = 37; | |
2466 | } | |
2467 | ||
2468 | return -1; | |
2469 | } | |
2470 | ||
2471 | /* Common power [off|failure] */ | |
2472 | rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) + | |
2473 | (mdev->p_uuid[UI_FLAGS] & 2); | |
2474 | /* lowest bit is set when we were primary, | |
2475 | * next bit (weight 2) is set when peer was primary */ | |
2476 | *rule_nr = 40; | |
2477 | ||
2478 | switch (rct) { | |
2479 | case 0: /* !self_pri && !peer_pri */ return 0; | |
2480 | case 1: /* self_pri && !peer_pri */ return 1; | |
2481 | case 2: /* !self_pri && peer_pri */ return -1; | |
2482 | case 3: /* self_pri && peer_pri */ | |
2483 | dc = test_bit(DISCARD_CONCURRENT, &mdev->flags); | |
2484 | return dc ? -1 : 1; | |
2485 | } | |
2486 | } | |
2487 | ||
2488 | *rule_nr = 50; | |
2489 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2490 | if (self == peer) | |
2491 | return -1; | |
2492 | ||
2493 | *rule_nr = 51; | |
2494 | peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); | |
2495 | if (self == peer) { | |
2496 | self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); | |
2497 | peer = mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1); | |
2498 | if (self == peer) { | |
2499 | /* The last P_SYNC_UUID did not get though. Undo the last start of | |
2500 | resync as sync source modifications of the peer's UUIDs. */ | |
2501 | ||
2502 | if (mdev->agreed_pro_version < 91) | |
2503 | return -1001; | |
2504 | ||
2505 | mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START]; | |
2506 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1]; | |
2507 | return -1; | |
2508 | } | |
2509 | } | |
2510 | ||
2511 | *rule_nr = 60; | |
2512 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2513 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2514 | peer = mdev->p_uuid[i] & ~((u64)1); | |
2515 | if (self == peer) | |
2516 | return -2; | |
2517 | } | |
2518 | ||
2519 | *rule_nr = 70; | |
2520 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2521 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2522 | if (self == peer) | |
2523 | return 1; | |
2524 | ||
2525 | *rule_nr = 71; | |
2526 | self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); | |
2527 | if (self == peer) { | |
2528 | self = mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1); | |
2529 | peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); | |
2530 | if (self == peer) { | |
2531 | /* The last P_SYNC_UUID did not get though. Undo the last start of | |
2532 | resync as sync source modifications of our UUIDs. */ | |
2533 | ||
2534 | if (mdev->agreed_pro_version < 91) | |
2535 | return -1001; | |
2536 | ||
2537 | _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]); | |
2538 | _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]); | |
2539 | ||
2540 | dev_info(DEV, "Undid last start of resync:\n"); | |
2541 | ||
2542 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, | |
2543 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2544 | ||
2545 | return 1; | |
2546 | } | |
2547 | } | |
2548 | ||
2549 | ||
2550 | *rule_nr = 80; | |
d8c2a36b | 2551 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); |
b411b363 PR |
2552 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
2553 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2554 | if (self == peer) | |
2555 | return 2; | |
2556 | } | |
2557 | ||
2558 | *rule_nr = 90; | |
2559 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2560 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2561 | if (self == peer && self != ((u64)0)) | |
2562 | return 100; | |
2563 | ||
2564 | *rule_nr = 100; | |
2565 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2566 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2567 | for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { | |
2568 | peer = mdev->p_uuid[j] & ~((u64)1); | |
2569 | if (self == peer) | |
2570 | return -100; | |
2571 | } | |
2572 | } | |
2573 | ||
2574 | return -1000; | |
2575 | } | |
2576 | ||
2577 | /* drbd_sync_handshake() returns the new conn state on success, or | |
2578 | CONN_MASK (-1) on failure. | |
2579 | */ | |
2580 | static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role, | |
2581 | enum drbd_disk_state peer_disk) __must_hold(local) | |
2582 | { | |
2583 | int hg, rule_nr; | |
2584 | enum drbd_conns rv = C_MASK; | |
2585 | enum drbd_disk_state mydisk; | |
2586 | ||
2587 | mydisk = mdev->state.disk; | |
2588 | if (mydisk == D_NEGOTIATING) | |
2589 | mydisk = mdev->new_state_tmp.disk; | |
2590 | ||
2591 | dev_info(DEV, "drbd_sync_handshake:\n"); | |
2592 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0); | |
2593 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, | |
2594 | mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2595 | ||
2596 | hg = drbd_uuid_compare(mdev, &rule_nr); | |
2597 | ||
2598 | dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr); | |
2599 | ||
2600 | if (hg == -1000) { | |
2601 | dev_alert(DEV, "Unrelated data, aborting!\n"); | |
2602 | return C_MASK; | |
2603 | } | |
2604 | if (hg == -1001) { | |
2605 | dev_alert(DEV, "To resolve this both sides have to support at least protocol\n"); | |
2606 | return C_MASK; | |
2607 | } | |
2608 | ||
2609 | if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || | |
2610 | (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { | |
2611 | int f = (hg == -100) || abs(hg) == 2; | |
2612 | hg = mydisk > D_INCONSISTENT ? 1 : -1; | |
2613 | if (f) | |
2614 | hg = hg*2; | |
2615 | dev_info(DEV, "Becoming sync %s due to disk states.\n", | |
2616 | hg > 0 ? "source" : "target"); | |
2617 | } | |
2618 | ||
3a11a487 AG |
2619 | if (abs(hg) == 100) |
2620 | drbd_khelper(mdev, "initial-split-brain"); | |
2621 | ||
b411b363 PR |
2622 | if (hg == 100 || (hg == -100 && mdev->net_conf->always_asbp)) { |
2623 | int pcount = (mdev->state.role == R_PRIMARY) | |
2624 | + (peer_role == R_PRIMARY); | |
2625 | int forced = (hg == -100); | |
2626 | ||
2627 | switch (pcount) { | |
2628 | case 0: | |
2629 | hg = drbd_asb_recover_0p(mdev); | |
2630 | break; | |
2631 | case 1: | |
2632 | hg = drbd_asb_recover_1p(mdev); | |
2633 | break; | |
2634 | case 2: | |
2635 | hg = drbd_asb_recover_2p(mdev); | |
2636 | break; | |
2637 | } | |
2638 | if (abs(hg) < 100) { | |
2639 | dev_warn(DEV, "Split-Brain detected, %d primaries, " | |
2640 | "automatically solved. Sync from %s node\n", | |
2641 | pcount, (hg < 0) ? "peer" : "this"); | |
2642 | if (forced) { | |
2643 | dev_warn(DEV, "Doing a full sync, since" | |
2644 | " UUIDs where ambiguous.\n"); | |
2645 | hg = hg*2; | |
2646 | } | |
2647 | } | |
2648 | } | |
2649 | ||
2650 | if (hg == -100) { | |
2651 | if (mdev->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1)) | |
2652 | hg = -1; | |
2653 | if (!mdev->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1)) | |
2654 | hg = 1; | |
2655 | ||
2656 | if (abs(hg) < 100) | |
2657 | dev_warn(DEV, "Split-Brain detected, manually solved. " | |
2658 | "Sync from %s node\n", | |
2659 | (hg < 0) ? "peer" : "this"); | |
2660 | } | |
2661 | ||
2662 | if (hg == -100) { | |
580b9767 LE |
2663 | /* FIXME this log message is not correct if we end up here |
2664 | * after an attempted attach on a diskless node. | |
2665 | * We just refuse to attach -- well, we drop the "connection" | |
2666 | * to that disk, in a way... */ | |
3a11a487 | 2667 | dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n"); |
b411b363 PR |
2668 | drbd_khelper(mdev, "split-brain"); |
2669 | return C_MASK; | |
2670 | } | |
2671 | ||
2672 | if (hg > 0 && mydisk <= D_INCONSISTENT) { | |
2673 | dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n"); | |
2674 | return C_MASK; | |
2675 | } | |
2676 | ||
2677 | if (hg < 0 && /* by intention we do not use mydisk here. */ | |
2678 | mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) { | |
2679 | switch (mdev->net_conf->rr_conflict) { | |
2680 | case ASB_CALL_HELPER: | |
2681 | drbd_khelper(mdev, "pri-lost"); | |
2682 | /* fall through */ | |
2683 | case ASB_DISCONNECT: | |
2684 | dev_err(DEV, "I shall become SyncTarget, but I am primary!\n"); | |
2685 | return C_MASK; | |
2686 | case ASB_VIOLENTLY: | |
2687 | dev_warn(DEV, "Becoming SyncTarget, violating the stable-data" | |
2688 | "assumption\n"); | |
2689 | } | |
2690 | } | |
2691 | ||
cf14c2e9 PR |
2692 | if (mdev->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->flags)) { |
2693 | if (hg == 0) | |
2694 | dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n"); | |
2695 | else | |
2696 | dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.", | |
2697 | drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), | |
2698 | abs(hg) >= 2 ? "full" : "bit-map based"); | |
2699 | return C_MASK; | |
2700 | } | |
2701 | ||
b411b363 PR |
2702 | if (abs(hg) >= 2) { |
2703 | dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); | |
2704 | if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake")) | |
2705 | return C_MASK; | |
2706 | } | |
2707 | ||
2708 | if (hg > 0) { /* become sync source. */ | |
2709 | rv = C_WF_BITMAP_S; | |
2710 | } else if (hg < 0) { /* become sync target */ | |
2711 | rv = C_WF_BITMAP_T; | |
2712 | } else { | |
2713 | rv = C_CONNECTED; | |
2714 | if (drbd_bm_total_weight(mdev)) { | |
2715 | dev_info(DEV, "No resync, but %lu bits in bitmap!\n", | |
2716 | drbd_bm_total_weight(mdev)); | |
2717 | } | |
2718 | } | |
2719 | ||
2720 | return rv; | |
2721 | } | |
2722 | ||
2723 | /* returns 1 if invalid */ | |
2724 | static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self) | |
2725 | { | |
2726 | /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ | |
2727 | if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) || | |
2728 | (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL)) | |
2729 | return 0; | |
2730 | ||
2731 | /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ | |
2732 | if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL || | |
2733 | self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL) | |
2734 | return 1; | |
2735 | ||
2736 | /* everything else is valid if they are equal on both sides. */ | |
2737 | if (peer == self) | |
2738 | return 0; | |
2739 | ||
2740 | /* everything es is invalid. */ | |
2741 | return 1; | |
2742 | } | |
2743 | ||
02918be2 | 2744 | static int receive_protocol(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 2745 | { |
02918be2 | 2746 | struct p_protocol *p = &mdev->data.rbuf.protocol; |
b411b363 | 2747 | int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; |
cf14c2e9 | 2748 | int p_want_lose, p_two_primaries, cf; |
b411b363 PR |
2749 | char p_integrity_alg[SHARED_SECRET_MAX] = ""; |
2750 | ||
b411b363 PR |
2751 | p_proto = be32_to_cpu(p->protocol); |
2752 | p_after_sb_0p = be32_to_cpu(p->after_sb_0p); | |
2753 | p_after_sb_1p = be32_to_cpu(p->after_sb_1p); | |
2754 | p_after_sb_2p = be32_to_cpu(p->after_sb_2p); | |
b411b363 | 2755 | p_two_primaries = be32_to_cpu(p->two_primaries); |
cf14c2e9 PR |
2756 | cf = be32_to_cpu(p->conn_flags); |
2757 | p_want_lose = cf & CF_WANT_LOSE; | |
2758 | ||
2759 | clear_bit(CONN_DRY_RUN, &mdev->flags); | |
2760 | ||
2761 | if (cf & CF_DRY_RUN) | |
2762 | set_bit(CONN_DRY_RUN, &mdev->flags); | |
b411b363 PR |
2763 | |
2764 | if (p_proto != mdev->net_conf->wire_protocol) { | |
2765 | dev_err(DEV, "incompatible communication protocols\n"); | |
2766 | goto disconnect; | |
2767 | } | |
2768 | ||
2769 | if (cmp_after_sb(p_after_sb_0p, mdev->net_conf->after_sb_0p)) { | |
2770 | dev_err(DEV, "incompatible after-sb-0pri settings\n"); | |
2771 | goto disconnect; | |
2772 | } | |
2773 | ||
2774 | if (cmp_after_sb(p_after_sb_1p, mdev->net_conf->after_sb_1p)) { | |
2775 | dev_err(DEV, "incompatible after-sb-1pri settings\n"); | |
2776 | goto disconnect; | |
2777 | } | |
2778 | ||
2779 | if (cmp_after_sb(p_after_sb_2p, mdev->net_conf->after_sb_2p)) { | |
2780 | dev_err(DEV, "incompatible after-sb-2pri settings\n"); | |
2781 | goto disconnect; | |
2782 | } | |
2783 | ||
2784 | if (p_want_lose && mdev->net_conf->want_lose) { | |
2785 | dev_err(DEV, "both sides have the 'want_lose' flag set\n"); | |
2786 | goto disconnect; | |
2787 | } | |
2788 | ||
2789 | if (p_two_primaries != mdev->net_conf->two_primaries) { | |
2790 | dev_err(DEV, "incompatible setting of the two-primaries options\n"); | |
2791 | goto disconnect; | |
2792 | } | |
2793 | ||
2794 | if (mdev->agreed_pro_version >= 87) { | |
2795 | unsigned char *my_alg = mdev->net_conf->integrity_alg; | |
2796 | ||
2797 | if (drbd_recv(mdev, p_integrity_alg, data_size) != data_size) | |
2798 | return FALSE; | |
2799 | ||
2800 | p_integrity_alg[SHARED_SECRET_MAX-1] = 0; | |
2801 | if (strcmp(p_integrity_alg, my_alg)) { | |
2802 | dev_err(DEV, "incompatible setting of the data-integrity-alg\n"); | |
2803 | goto disconnect; | |
2804 | } | |
2805 | dev_info(DEV, "data-integrity-alg: %s\n", | |
2806 | my_alg[0] ? my_alg : (unsigned char *)"<not-used>"); | |
2807 | } | |
2808 | ||
2809 | return TRUE; | |
2810 | ||
2811 | disconnect: | |
2812 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
2813 | return FALSE; | |
2814 | } | |
2815 | ||
2816 | /* helper function | |
2817 | * input: alg name, feature name | |
2818 | * return: NULL (alg name was "") | |
2819 | * ERR_PTR(error) if something goes wrong | |
2820 | * or the crypto hash ptr, if it worked out ok. */ | |
2821 | struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev, | |
2822 | const char *alg, const char *name) | |
2823 | { | |
2824 | struct crypto_hash *tfm; | |
2825 | ||
2826 | if (!alg[0]) | |
2827 | return NULL; | |
2828 | ||
2829 | tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); | |
2830 | if (IS_ERR(tfm)) { | |
2831 | dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n", | |
2832 | alg, name, PTR_ERR(tfm)); | |
2833 | return tfm; | |
2834 | } | |
2835 | if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) { | |
2836 | crypto_free_hash(tfm); | |
2837 | dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name); | |
2838 | return ERR_PTR(-EINVAL); | |
2839 | } | |
2840 | return tfm; | |
2841 | } | |
2842 | ||
02918be2 | 2843 | static int receive_SyncParam(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int packet_size) |
b411b363 PR |
2844 | { |
2845 | int ok = TRUE; | |
02918be2 | 2846 | struct p_rs_param_95 *p = &mdev->data.rbuf.rs_param_95; |
b411b363 PR |
2847 | unsigned int header_size, data_size, exp_max_sz; |
2848 | struct crypto_hash *verify_tfm = NULL; | |
2849 | struct crypto_hash *csums_tfm = NULL; | |
2850 | const int apv = mdev->agreed_pro_version; | |
778f271d PR |
2851 | int *rs_plan_s = NULL; |
2852 | int fifo_size = 0; | |
b411b363 PR |
2853 | |
2854 | exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) | |
2855 | : apv == 88 ? sizeof(struct p_rs_param) | |
2856 | + SHARED_SECRET_MAX | |
8e26f9cc PR |
2857 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
2858 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 2859 | |
02918be2 | 2860 | if (packet_size > exp_max_sz) { |
b411b363 | 2861 | dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n", |
02918be2 | 2862 | packet_size, exp_max_sz); |
b411b363 PR |
2863 | return FALSE; |
2864 | } | |
2865 | ||
2866 | if (apv <= 88) { | |
02918be2 PR |
2867 | header_size = sizeof(struct p_rs_param) - sizeof(struct p_header80); |
2868 | data_size = packet_size - header_size; | |
8e26f9cc | 2869 | } else if (apv <= 94) { |
02918be2 PR |
2870 | header_size = sizeof(struct p_rs_param_89) - sizeof(struct p_header80); |
2871 | data_size = packet_size - header_size; | |
b411b363 | 2872 | D_ASSERT(data_size == 0); |
8e26f9cc | 2873 | } else { |
02918be2 PR |
2874 | header_size = sizeof(struct p_rs_param_95) - sizeof(struct p_header80); |
2875 | data_size = packet_size - header_size; | |
8e26f9cc | 2876 | D_ASSERT(data_size == 0); |
b411b363 PR |
2877 | } |
2878 | ||
2879 | /* initialize verify_alg and csums_alg */ | |
2880 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); | |
2881 | ||
02918be2 | 2882 | if (drbd_recv(mdev, &p->head.payload, header_size) != header_size) |
b411b363 PR |
2883 | return FALSE; |
2884 | ||
2885 | mdev->sync_conf.rate = be32_to_cpu(p->rate); | |
2886 | ||
2887 | if (apv >= 88) { | |
2888 | if (apv == 88) { | |
2889 | if (data_size > SHARED_SECRET_MAX) { | |
2890 | dev_err(DEV, "verify-alg too long, " | |
2891 | "peer wants %u, accepting only %u byte\n", | |
2892 | data_size, SHARED_SECRET_MAX); | |
2893 | return FALSE; | |
2894 | } | |
2895 | ||
2896 | if (drbd_recv(mdev, p->verify_alg, data_size) != data_size) | |
2897 | return FALSE; | |
2898 | ||
2899 | /* we expect NUL terminated string */ | |
2900 | /* but just in case someone tries to be evil */ | |
2901 | D_ASSERT(p->verify_alg[data_size-1] == 0); | |
2902 | p->verify_alg[data_size-1] = 0; | |
2903 | ||
2904 | } else /* apv >= 89 */ { | |
2905 | /* we still expect NUL terminated strings */ | |
2906 | /* but just in case someone tries to be evil */ | |
2907 | D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0); | |
2908 | D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0); | |
2909 | p->verify_alg[SHARED_SECRET_MAX-1] = 0; | |
2910 | p->csums_alg[SHARED_SECRET_MAX-1] = 0; | |
2911 | } | |
2912 | ||
2913 | if (strcmp(mdev->sync_conf.verify_alg, p->verify_alg)) { | |
2914 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { | |
2915 | dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", | |
2916 | mdev->sync_conf.verify_alg, p->verify_alg); | |
2917 | goto disconnect; | |
2918 | } | |
2919 | verify_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
2920 | p->verify_alg, "verify-alg"); | |
2921 | if (IS_ERR(verify_tfm)) { | |
2922 | verify_tfm = NULL; | |
2923 | goto disconnect; | |
2924 | } | |
2925 | } | |
2926 | ||
2927 | if (apv >= 89 && strcmp(mdev->sync_conf.csums_alg, p->csums_alg)) { | |
2928 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { | |
2929 | dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", | |
2930 | mdev->sync_conf.csums_alg, p->csums_alg); | |
2931 | goto disconnect; | |
2932 | } | |
2933 | csums_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
2934 | p->csums_alg, "csums-alg"); | |
2935 | if (IS_ERR(csums_tfm)) { | |
2936 | csums_tfm = NULL; | |
2937 | goto disconnect; | |
2938 | } | |
2939 | } | |
2940 | ||
8e26f9cc PR |
2941 | if (apv > 94) { |
2942 | mdev->sync_conf.rate = be32_to_cpu(p->rate); | |
2943 | mdev->sync_conf.c_plan_ahead = be32_to_cpu(p->c_plan_ahead); | |
2944 | mdev->sync_conf.c_delay_target = be32_to_cpu(p->c_delay_target); | |
2945 | mdev->sync_conf.c_fill_target = be32_to_cpu(p->c_fill_target); | |
2946 | mdev->sync_conf.c_max_rate = be32_to_cpu(p->c_max_rate); | |
778f271d PR |
2947 | |
2948 | fifo_size = (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ; | |
2949 | if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) { | |
2950 | rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL); | |
2951 | if (!rs_plan_s) { | |
2952 | dev_err(DEV, "kmalloc of fifo_buffer failed"); | |
2953 | goto disconnect; | |
2954 | } | |
2955 | } | |
8e26f9cc | 2956 | } |
b411b363 PR |
2957 | |
2958 | spin_lock(&mdev->peer_seq_lock); | |
2959 | /* lock against drbd_nl_syncer_conf() */ | |
2960 | if (verify_tfm) { | |
2961 | strcpy(mdev->sync_conf.verify_alg, p->verify_alg); | |
2962 | mdev->sync_conf.verify_alg_len = strlen(p->verify_alg) + 1; | |
2963 | crypto_free_hash(mdev->verify_tfm); | |
2964 | mdev->verify_tfm = verify_tfm; | |
2965 | dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg); | |
2966 | } | |
2967 | if (csums_tfm) { | |
2968 | strcpy(mdev->sync_conf.csums_alg, p->csums_alg); | |
2969 | mdev->sync_conf.csums_alg_len = strlen(p->csums_alg) + 1; | |
2970 | crypto_free_hash(mdev->csums_tfm); | |
2971 | mdev->csums_tfm = csums_tfm; | |
2972 | dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg); | |
2973 | } | |
778f271d PR |
2974 | if (fifo_size != mdev->rs_plan_s.size) { |
2975 | kfree(mdev->rs_plan_s.values); | |
2976 | mdev->rs_plan_s.values = rs_plan_s; | |
2977 | mdev->rs_plan_s.size = fifo_size; | |
2978 | mdev->rs_planed = 0; | |
2979 | } | |
b411b363 PR |
2980 | spin_unlock(&mdev->peer_seq_lock); |
2981 | } | |
2982 | ||
2983 | return ok; | |
2984 | disconnect: | |
2985 | /* just for completeness: actually not needed, | |
2986 | * as this is not reached if csums_tfm was ok. */ | |
2987 | crypto_free_hash(csums_tfm); | |
2988 | /* but free the verify_tfm again, if csums_tfm did not work out */ | |
2989 | crypto_free_hash(verify_tfm); | |
2990 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
2991 | return FALSE; | |
2992 | } | |
2993 | ||
2994 | static void drbd_setup_order_type(struct drbd_conf *mdev, int peer) | |
2995 | { | |
2996 | /* sorry, we currently have no working implementation | |
2997 | * of distributed TCQ */ | |
2998 | } | |
2999 | ||
3000 | /* warn if the arguments differ by more than 12.5% */ | |
3001 | static void warn_if_differ_considerably(struct drbd_conf *mdev, | |
3002 | const char *s, sector_t a, sector_t b) | |
3003 | { | |
3004 | sector_t d; | |
3005 | if (a == 0 || b == 0) | |
3006 | return; | |
3007 | d = (a > b) ? (a - b) : (b - a); | |
3008 | if (d > (a>>3) || d > (b>>3)) | |
3009 | dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s, | |
3010 | (unsigned long long)a, (unsigned long long)b); | |
3011 | } | |
3012 | ||
02918be2 | 3013 | static int receive_sizes(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 3014 | { |
02918be2 | 3015 | struct p_sizes *p = &mdev->data.rbuf.sizes; |
b411b363 PR |
3016 | enum determine_dev_size dd = unchanged; |
3017 | unsigned int max_seg_s; | |
3018 | sector_t p_size, p_usize, my_usize; | |
3019 | int ldsc = 0; /* local disk size changed */ | |
e89b591c | 3020 | enum dds_flags ddsf; |
b411b363 | 3021 | |
b411b363 PR |
3022 | p_size = be64_to_cpu(p->d_size); |
3023 | p_usize = be64_to_cpu(p->u_size); | |
3024 | ||
3025 | if (p_size == 0 && mdev->state.disk == D_DISKLESS) { | |
3026 | dev_err(DEV, "some backing storage is needed\n"); | |
3027 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
3028 | return FALSE; | |
3029 | } | |
3030 | ||
3031 | /* just store the peer's disk size for now. | |
3032 | * we still need to figure out whether we accept that. */ | |
3033 | mdev->p_size = p_size; | |
3034 | ||
3035 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
3036 | if (get_ldev(mdev)) { | |
3037 | warn_if_differ_considerably(mdev, "lower level device sizes", | |
3038 | p_size, drbd_get_max_capacity(mdev->ldev)); | |
3039 | warn_if_differ_considerably(mdev, "user requested size", | |
3040 | p_usize, mdev->ldev->dc.disk_size); | |
3041 | ||
3042 | /* if this is the first connect, or an otherwise expected | |
3043 | * param exchange, choose the minimum */ | |
3044 | if (mdev->state.conn == C_WF_REPORT_PARAMS) | |
3045 | p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size, | |
3046 | p_usize); | |
3047 | ||
3048 | my_usize = mdev->ldev->dc.disk_size; | |
3049 | ||
3050 | if (mdev->ldev->dc.disk_size != p_usize) { | |
3051 | mdev->ldev->dc.disk_size = p_usize; | |
3052 | dev_info(DEV, "Peer sets u_size to %lu sectors\n", | |
3053 | (unsigned long)mdev->ldev->dc.disk_size); | |
3054 | } | |
3055 | ||
3056 | /* Never shrink a device with usable data during connect. | |
3057 | But allow online shrinking if we are connected. */ | |
a393db6f | 3058 | if (drbd_new_dev_size(mdev, mdev->ldev, 0) < |
b411b363 PR |
3059 | drbd_get_capacity(mdev->this_bdev) && |
3060 | mdev->state.disk >= D_OUTDATED && | |
3061 | mdev->state.conn < C_CONNECTED) { | |
3062 | dev_err(DEV, "The peer's disk size is too small!\n"); | |
3063 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
3064 | mdev->ldev->dc.disk_size = my_usize; | |
3065 | put_ldev(mdev); | |
3066 | return FALSE; | |
3067 | } | |
3068 | put_ldev(mdev); | |
3069 | } | |
3070 | #undef min_not_zero | |
3071 | ||
e89b591c | 3072 | ddsf = be16_to_cpu(p->dds_flags); |
b411b363 | 3073 | if (get_ldev(mdev)) { |
e89b591c | 3074 | dd = drbd_determin_dev_size(mdev, ddsf); |
b411b363 PR |
3075 | put_ldev(mdev); |
3076 | if (dd == dev_size_error) | |
3077 | return FALSE; | |
3078 | drbd_md_sync(mdev); | |
3079 | } else { | |
3080 | /* I am diskless, need to accept the peer's size. */ | |
3081 | drbd_set_my_capacity(mdev, p_size); | |
3082 | } | |
3083 | ||
b411b363 PR |
3084 | if (get_ldev(mdev)) { |
3085 | if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) { | |
3086 | mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); | |
3087 | ldsc = 1; | |
3088 | } | |
3089 | ||
a1c88d0d LE |
3090 | if (mdev->agreed_pro_version < 94) |
3091 | max_seg_s = be32_to_cpu(p->max_segment_size); | |
8979d9c9 LE |
3092 | else if (mdev->agreed_pro_version == 94) |
3093 | max_seg_s = DRBD_MAX_SIZE_H80_PACKET; | |
a1c88d0d LE |
3094 | else /* drbd 8.3.8 onwards */ |
3095 | max_seg_s = DRBD_MAX_SEGMENT_SIZE; | |
3096 | ||
b411b363 PR |
3097 | if (max_seg_s != queue_max_segment_size(mdev->rq_queue)) |
3098 | drbd_setup_queue_param(mdev, max_seg_s); | |
3099 | ||
e89b591c | 3100 | drbd_setup_order_type(mdev, be16_to_cpu(p->queue_order_type)); |
b411b363 PR |
3101 | put_ldev(mdev); |
3102 | } | |
3103 | ||
3104 | if (mdev->state.conn > C_WF_REPORT_PARAMS) { | |
3105 | if (be64_to_cpu(p->c_size) != | |
3106 | drbd_get_capacity(mdev->this_bdev) || ldsc) { | |
3107 | /* we have different sizes, probably peer | |
3108 | * needs to know my new size... */ | |
e89b591c | 3109 | drbd_send_sizes(mdev, 0, ddsf); |
b411b363 PR |
3110 | } |
3111 | if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) || | |
3112 | (dd == grew && mdev->state.conn == C_CONNECTED)) { | |
3113 | if (mdev->state.pdsk >= D_INCONSISTENT && | |
e89b591c PR |
3114 | mdev->state.disk >= D_INCONSISTENT) { |
3115 | if (ddsf & DDSF_NO_RESYNC) | |
3116 | dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n"); | |
3117 | else | |
3118 | resync_after_online_grow(mdev); | |
3119 | } else | |
b411b363 PR |
3120 | set_bit(RESYNC_AFTER_NEG, &mdev->flags); |
3121 | } | |
3122 | } | |
3123 | ||
3124 | return TRUE; | |
3125 | } | |
3126 | ||
02918be2 | 3127 | static int receive_uuids(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 3128 | { |
02918be2 | 3129 | struct p_uuids *p = &mdev->data.rbuf.uuids; |
b411b363 PR |
3130 | u64 *p_uuid; |
3131 | int i; | |
3132 | ||
b411b363 PR |
3133 | p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); |
3134 | ||
3135 | for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) | |
3136 | p_uuid[i] = be64_to_cpu(p->uuid[i]); | |
3137 | ||
3138 | kfree(mdev->p_uuid); | |
3139 | mdev->p_uuid = p_uuid; | |
3140 | ||
3141 | if (mdev->state.conn < C_CONNECTED && | |
3142 | mdev->state.disk < D_INCONSISTENT && | |
3143 | mdev->state.role == R_PRIMARY && | |
3144 | (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { | |
3145 | dev_err(DEV, "Can only connect to data with current UUID=%016llX\n", | |
3146 | (unsigned long long)mdev->ed_uuid); | |
3147 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
3148 | return FALSE; | |
3149 | } | |
3150 | ||
3151 | if (get_ldev(mdev)) { | |
3152 | int skip_initial_sync = | |
3153 | mdev->state.conn == C_CONNECTED && | |
3154 | mdev->agreed_pro_version >= 90 && | |
3155 | mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && | |
3156 | (p_uuid[UI_FLAGS] & 8); | |
3157 | if (skip_initial_sync) { | |
3158 | dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n"); | |
3159 | drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, | |
3160 | "clear_n_write from receive_uuids"); | |
3161 | _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]); | |
3162 | _drbd_uuid_set(mdev, UI_BITMAP, 0); | |
3163 | _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), | |
3164 | CS_VERBOSE, NULL); | |
3165 | drbd_md_sync(mdev); | |
3166 | } | |
3167 | put_ldev(mdev); | |
18a50fa2 PR |
3168 | } else if (mdev->state.disk < D_INCONSISTENT && |
3169 | mdev->state.role == R_PRIMARY) { | |
3170 | /* I am a diskless primary, the peer just created a new current UUID | |
3171 | for me. */ | |
3172 | drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); | |
b411b363 PR |
3173 | } |
3174 | ||
3175 | /* Before we test for the disk state, we should wait until an eventually | |
3176 | ongoing cluster wide state change is finished. That is important if | |
3177 | we are primary and are detaching from our disk. We need to see the | |
3178 | new disk state... */ | |
3179 | wait_event(mdev->misc_wait, !test_bit(CLUSTER_ST_CHANGE, &mdev->flags)); | |
3180 | if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT) | |
3181 | drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); | |
3182 | ||
3183 | return TRUE; | |
3184 | } | |
3185 | ||
3186 | /** | |
3187 | * convert_state() - Converts the peer's view of the cluster state to our point of view | |
3188 | * @ps: The state as seen by the peer. | |
3189 | */ | |
3190 | static union drbd_state convert_state(union drbd_state ps) | |
3191 | { | |
3192 | union drbd_state ms; | |
3193 | ||
3194 | static enum drbd_conns c_tab[] = { | |
3195 | [C_CONNECTED] = C_CONNECTED, | |
3196 | ||
3197 | [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, | |
3198 | [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, | |
3199 | [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ | |
3200 | [C_VERIFY_S] = C_VERIFY_T, | |
3201 | [C_MASK] = C_MASK, | |
3202 | }; | |
3203 | ||
3204 | ms.i = ps.i; | |
3205 | ||
3206 | ms.conn = c_tab[ps.conn]; | |
3207 | ms.peer = ps.role; | |
3208 | ms.role = ps.peer; | |
3209 | ms.pdsk = ps.disk; | |
3210 | ms.disk = ps.pdsk; | |
3211 | ms.peer_isp = (ps.aftr_isp | ps.user_isp); | |
3212 | ||
3213 | return ms; | |
3214 | } | |
3215 | ||
02918be2 | 3216 | static int receive_req_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 3217 | { |
02918be2 | 3218 | struct p_req_state *p = &mdev->data.rbuf.req_state; |
b411b363 PR |
3219 | union drbd_state mask, val; |
3220 | int rv; | |
3221 | ||
b411b363 PR |
3222 | mask.i = be32_to_cpu(p->mask); |
3223 | val.i = be32_to_cpu(p->val); | |
3224 | ||
3225 | if (test_bit(DISCARD_CONCURRENT, &mdev->flags) && | |
3226 | test_bit(CLUSTER_ST_CHANGE, &mdev->flags)) { | |
3227 | drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG); | |
3228 | return TRUE; | |
3229 | } | |
3230 | ||
3231 | mask = convert_state(mask); | |
3232 | val = convert_state(val); | |
3233 | ||
3234 | rv = drbd_change_state(mdev, CS_VERBOSE, mask, val); | |
3235 | ||
3236 | drbd_send_sr_reply(mdev, rv); | |
3237 | drbd_md_sync(mdev); | |
3238 | ||
3239 | return TRUE; | |
3240 | } | |
3241 | ||
02918be2 | 3242 | static int receive_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 3243 | { |
02918be2 | 3244 | struct p_state *p = &mdev->data.rbuf.state; |
b411b363 PR |
3245 | enum drbd_conns nconn, oconn; |
3246 | union drbd_state ns, peer_state; | |
3247 | enum drbd_disk_state real_peer_disk; | |
65d922c3 | 3248 | enum chg_state_flags cs_flags; |
b411b363 PR |
3249 | int rv; |
3250 | ||
b411b363 PR |
3251 | peer_state.i = be32_to_cpu(p->state); |
3252 | ||
3253 | real_peer_disk = peer_state.disk; | |
3254 | if (peer_state.disk == D_NEGOTIATING) { | |
3255 | real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; | |
3256 | dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); | |
3257 | } | |
3258 | ||
3259 | spin_lock_irq(&mdev->req_lock); | |
3260 | retry: | |
3261 | oconn = nconn = mdev->state.conn; | |
3262 | spin_unlock_irq(&mdev->req_lock); | |
3263 | ||
3264 | if (nconn == C_WF_REPORT_PARAMS) | |
3265 | nconn = C_CONNECTED; | |
3266 | ||
3267 | if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING && | |
3268 | get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3269 | int cr; /* consider resync */ | |
3270 | ||
3271 | /* if we established a new connection */ | |
3272 | cr = (oconn < C_CONNECTED); | |
3273 | /* if we had an established connection | |
3274 | * and one of the nodes newly attaches a disk */ | |
3275 | cr |= (oconn == C_CONNECTED && | |
3276 | (peer_state.disk == D_NEGOTIATING || | |
3277 | mdev->state.disk == D_NEGOTIATING)); | |
3278 | /* if we have both been inconsistent, and the peer has been | |
3279 | * forced to be UpToDate with --overwrite-data */ | |
3280 | cr |= test_bit(CONSIDER_RESYNC, &mdev->flags); | |
3281 | /* if we had been plain connected, and the admin requested to | |
3282 | * start a sync by "invalidate" or "invalidate-remote" */ | |
3283 | cr |= (oconn == C_CONNECTED && | |
3284 | (peer_state.conn >= C_STARTING_SYNC_S && | |
3285 | peer_state.conn <= C_WF_BITMAP_T)); | |
3286 | ||
3287 | if (cr) | |
3288 | nconn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk); | |
3289 | ||
3290 | put_ldev(mdev); | |
3291 | if (nconn == C_MASK) { | |
580b9767 | 3292 | nconn = C_CONNECTED; |
b411b363 PR |
3293 | if (mdev->state.disk == D_NEGOTIATING) { |
3294 | drbd_force_state(mdev, NS(disk, D_DISKLESS)); | |
b411b363 PR |
3295 | } else if (peer_state.disk == D_NEGOTIATING) { |
3296 | dev_err(DEV, "Disk attach process on the peer node was aborted.\n"); | |
3297 | peer_state.disk = D_DISKLESS; | |
580b9767 | 3298 | real_peer_disk = D_DISKLESS; |
b411b363 | 3299 | } else { |
cf14c2e9 PR |
3300 | if (test_and_clear_bit(CONN_DRY_RUN, &mdev->flags)) |
3301 | return FALSE; | |
b411b363 PR |
3302 | D_ASSERT(oconn == C_WF_REPORT_PARAMS); |
3303 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
3304 | return FALSE; | |
3305 | } | |
3306 | } | |
3307 | } | |
3308 | ||
3309 | spin_lock_irq(&mdev->req_lock); | |
3310 | if (mdev->state.conn != oconn) | |
3311 | goto retry; | |
3312 | clear_bit(CONSIDER_RESYNC, &mdev->flags); | |
3313 | ns.i = mdev->state.i; | |
3314 | ns.conn = nconn; | |
3315 | ns.peer = peer_state.role; | |
3316 | ns.pdsk = real_peer_disk; | |
3317 | ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); | |
3318 | if ((nconn == C_CONNECTED || nconn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) | |
3319 | ns.disk = mdev->new_state_tmp.disk; | |
65d922c3 | 3320 | cs_flags = CS_VERBOSE + (oconn < C_CONNECTED && nconn >= C_CONNECTED ? 0 : CS_HARD); |
fb22c402 | 3321 | if (ns.pdsk == D_CONSISTENT && is_susp(ns) && nconn == C_CONNECTED && oconn < C_CONNECTED && |
481c6f50 PR |
3322 | test_bit(NEW_CUR_UUID, &mdev->flags)) { |
3323 | /* Do not allow tl_restart(resend) for a rebooted peer. We can only allow this | |
3324 | for temporal network outages! */ | |
3325 | spin_unlock_irq(&mdev->req_lock); | |
3326 | dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); | |
3327 | tl_clear(mdev); | |
3328 | drbd_uuid_new_current(mdev); | |
3329 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
3330 | drbd_force_state(mdev, NS2(conn, C_PROTOCOL_ERROR, susp, 0)); | |
3331 | return FALSE; | |
3332 | } | |
65d922c3 | 3333 | rv = _drbd_set_state(mdev, ns, cs_flags, NULL); |
b411b363 PR |
3334 | ns = mdev->state; |
3335 | spin_unlock_irq(&mdev->req_lock); | |
3336 | ||
3337 | if (rv < SS_SUCCESS) { | |
3338 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
3339 | return FALSE; | |
3340 | } | |
3341 | ||
3342 | if (oconn > C_WF_REPORT_PARAMS) { | |
3343 | if (nconn > C_CONNECTED && peer_state.conn <= C_CONNECTED && | |
3344 | peer_state.disk != D_NEGOTIATING ) { | |
3345 | /* we want resync, peer has not yet decided to sync... */ | |
3346 | /* Nowadays only used when forcing a node into primary role and | |
3347 | setting its disk to UpToDate with that */ | |
3348 | drbd_send_uuids(mdev); | |
3349 | drbd_send_state(mdev); | |
3350 | } | |
3351 | } | |
3352 | ||
3353 | mdev->net_conf->want_lose = 0; | |
3354 | ||
3355 | drbd_md_sync(mdev); /* update connected indicator, la_size, ... */ | |
3356 | ||
3357 | return TRUE; | |
3358 | } | |
3359 | ||
02918be2 | 3360 | static int receive_sync_uuid(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 | 3361 | { |
02918be2 | 3362 | struct p_rs_uuid *p = &mdev->data.rbuf.rs_uuid; |
b411b363 PR |
3363 | |
3364 | wait_event(mdev->misc_wait, | |
3365 | mdev->state.conn == C_WF_SYNC_UUID || | |
3366 | mdev->state.conn < C_CONNECTED || | |
3367 | mdev->state.disk < D_NEGOTIATING); | |
3368 | ||
3369 | /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */ | |
3370 | ||
b411b363 PR |
3371 | /* Here the _drbd_uuid_ functions are right, current should |
3372 | _not_ be rotated into the history */ | |
3373 | if (get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3374 | _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid)); | |
3375 | _drbd_uuid_set(mdev, UI_BITMAP, 0UL); | |
3376 | ||
3377 | drbd_start_resync(mdev, C_SYNC_TARGET); | |
3378 | ||
3379 | put_ldev(mdev); | |
3380 | } else | |
3381 | dev_err(DEV, "Ignoring SyncUUID packet!\n"); | |
3382 | ||
3383 | return TRUE; | |
3384 | } | |
3385 | ||
3386 | enum receive_bitmap_ret { OK, DONE, FAILED }; | |
3387 | ||
3388 | static enum receive_bitmap_ret | |
02918be2 PR |
3389 | receive_bitmap_plain(struct drbd_conf *mdev, unsigned int data_size, |
3390 | unsigned long *buffer, struct bm_xfer_ctx *c) | |
b411b363 PR |
3391 | { |
3392 | unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset); | |
3393 | unsigned want = num_words * sizeof(long); | |
3394 | ||
02918be2 PR |
3395 | if (want != data_size) { |
3396 | dev_err(DEV, "%s:want (%u) != data_size (%u)\n", __func__, want, data_size); | |
b411b363 PR |
3397 | return FAILED; |
3398 | } | |
3399 | if (want == 0) | |
3400 | return DONE; | |
3401 | if (drbd_recv(mdev, buffer, want) != want) | |
3402 | return FAILED; | |
3403 | ||
3404 | drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer); | |
3405 | ||
3406 | c->word_offset += num_words; | |
3407 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
3408 | if (c->bit_offset > c->bm_bits) | |
3409 | c->bit_offset = c->bm_bits; | |
3410 | ||
3411 | return OK; | |
3412 | } | |
3413 | ||
3414 | static enum receive_bitmap_ret | |
3415 | recv_bm_rle_bits(struct drbd_conf *mdev, | |
3416 | struct p_compressed_bm *p, | |
3417 | struct bm_xfer_ctx *c) | |
3418 | { | |
3419 | struct bitstream bs; | |
3420 | u64 look_ahead; | |
3421 | u64 rl; | |
3422 | u64 tmp; | |
3423 | unsigned long s = c->bit_offset; | |
3424 | unsigned long e; | |
004352fa | 3425 | int len = be16_to_cpu(p->head.length) - (sizeof(*p) - sizeof(p->head)); |
b411b363 PR |
3426 | int toggle = DCBP_get_start(p); |
3427 | int have; | |
3428 | int bits; | |
3429 | ||
3430 | bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p)); | |
3431 | ||
3432 | bits = bitstream_get_bits(&bs, &look_ahead, 64); | |
3433 | if (bits < 0) | |
3434 | return FAILED; | |
3435 | ||
3436 | for (have = bits; have > 0; s += rl, toggle = !toggle) { | |
3437 | bits = vli_decode_bits(&rl, look_ahead); | |
3438 | if (bits <= 0) | |
3439 | return FAILED; | |
3440 | ||
3441 | if (toggle) { | |
3442 | e = s + rl -1; | |
3443 | if (e >= c->bm_bits) { | |
3444 | dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); | |
3445 | return FAILED; | |
3446 | } | |
3447 | _drbd_bm_set_bits(mdev, s, e); | |
3448 | } | |
3449 | ||
3450 | if (have < bits) { | |
3451 | dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", | |
3452 | have, bits, look_ahead, | |
3453 | (unsigned int)(bs.cur.b - p->code), | |
3454 | (unsigned int)bs.buf_len); | |
3455 | return FAILED; | |
3456 | } | |
3457 | look_ahead >>= bits; | |
3458 | have -= bits; | |
3459 | ||
3460 | bits = bitstream_get_bits(&bs, &tmp, 64 - have); | |
3461 | if (bits < 0) | |
3462 | return FAILED; | |
3463 | look_ahead |= tmp << have; | |
3464 | have += bits; | |
3465 | } | |
3466 | ||
3467 | c->bit_offset = s; | |
3468 | bm_xfer_ctx_bit_to_word_offset(c); | |
3469 | ||
3470 | return (s == c->bm_bits) ? DONE : OK; | |
3471 | } | |
3472 | ||
3473 | static enum receive_bitmap_ret | |
3474 | decode_bitmap_c(struct drbd_conf *mdev, | |
3475 | struct p_compressed_bm *p, | |
3476 | struct bm_xfer_ctx *c) | |
3477 | { | |
3478 | if (DCBP_get_code(p) == RLE_VLI_Bits) | |
3479 | return recv_bm_rle_bits(mdev, p, c); | |
3480 | ||
3481 | /* other variants had been implemented for evaluation, | |
3482 | * but have been dropped as this one turned out to be "best" | |
3483 | * during all our tests. */ | |
3484 | ||
3485 | dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding); | |
3486 | drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); | |
3487 | return FAILED; | |
3488 | } | |
3489 | ||
3490 | void INFO_bm_xfer_stats(struct drbd_conf *mdev, | |
3491 | const char *direction, struct bm_xfer_ctx *c) | |
3492 | { | |
3493 | /* what would it take to transfer it "plaintext" */ | |
0b70a13d | 3494 | unsigned plain = sizeof(struct p_header80) * |
b411b363 PR |
3495 | ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1) |
3496 | + c->bm_words * sizeof(long); | |
3497 | unsigned total = c->bytes[0] + c->bytes[1]; | |
3498 | unsigned r; | |
3499 | ||
3500 | /* total can not be zero. but just in case: */ | |
3501 | if (total == 0) | |
3502 | return; | |
3503 | ||
3504 | /* don't report if not compressed */ | |
3505 | if (total >= plain) | |
3506 | return; | |
3507 | ||
3508 | /* total < plain. check for overflow, still */ | |
3509 | r = (total > UINT_MAX/1000) ? (total / (plain/1000)) | |
3510 | : (1000 * total / plain); | |
3511 | ||
3512 | if (r > 1000) | |
3513 | r = 1000; | |
3514 | ||
3515 | r = 1000 - r; | |
3516 | dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " | |
3517 | "total %u; compression: %u.%u%%\n", | |
3518 | direction, | |
3519 | c->bytes[1], c->packets[1], | |
3520 | c->bytes[0], c->packets[0], | |
3521 | total, r/10, r % 10); | |
3522 | } | |
3523 | ||
3524 | /* Since we are processing the bitfield from lower addresses to higher, | |
3525 | it does not matter if the process it in 32 bit chunks or 64 bit | |
3526 | chunks as long as it is little endian. (Understand it as byte stream, | |
3527 | beginning with the lowest byte...) If we would use big endian | |
3528 | we would need to process it from the highest address to the lowest, | |
3529 | in order to be agnostic to the 32 vs 64 bits issue. | |
3530 | ||
3531 | returns 0 on failure, 1 if we successfully received it. */ | |
02918be2 | 3532 | static int receive_bitmap(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
3533 | { |
3534 | struct bm_xfer_ctx c; | |
3535 | void *buffer; | |
3536 | enum receive_bitmap_ret ret; | |
3537 | int ok = FALSE; | |
02918be2 | 3538 | struct p_header80 *h = &mdev->data.rbuf.header.h80; |
b411b363 PR |
3539 | |
3540 | wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt)); | |
3541 | ||
3542 | drbd_bm_lock(mdev, "receive bitmap"); | |
3543 | ||
3544 | /* maybe we should use some per thread scratch page, | |
3545 | * and allocate that during initial device creation? */ | |
3546 | buffer = (unsigned long *) __get_free_page(GFP_NOIO); | |
3547 | if (!buffer) { | |
3548 | dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__); | |
3549 | goto out; | |
3550 | } | |
3551 | ||
3552 | c = (struct bm_xfer_ctx) { | |
3553 | .bm_bits = drbd_bm_bits(mdev), | |
3554 | .bm_words = drbd_bm_words(mdev), | |
3555 | }; | |
3556 | ||
3557 | do { | |
02918be2 PR |
3558 | if (cmd == P_BITMAP) { |
3559 | ret = receive_bitmap_plain(mdev, data_size, buffer, &c); | |
3560 | } else if (cmd == P_COMPRESSED_BITMAP) { | |
b411b363 PR |
3561 | /* MAYBE: sanity check that we speak proto >= 90, |
3562 | * and the feature is enabled! */ | |
3563 | struct p_compressed_bm *p; | |
3564 | ||
02918be2 | 3565 | if (data_size > BM_PACKET_PAYLOAD_BYTES) { |
b411b363 PR |
3566 | dev_err(DEV, "ReportCBitmap packet too large\n"); |
3567 | goto out; | |
3568 | } | |
3569 | /* use the page buff */ | |
3570 | p = buffer; | |
3571 | memcpy(p, h, sizeof(*h)); | |
02918be2 | 3572 | if (drbd_recv(mdev, p->head.payload, data_size) != data_size) |
b411b363 | 3573 | goto out; |
004352fa LE |
3574 | if (data_size <= (sizeof(*p) - sizeof(p->head))) { |
3575 | dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", data_size); | |
b411b363 PR |
3576 | return FAILED; |
3577 | } | |
3578 | ret = decode_bitmap_c(mdev, p, &c); | |
3579 | } else { | |
02918be2 | 3580 | dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", cmd); |
b411b363 PR |
3581 | goto out; |
3582 | } | |
3583 | ||
02918be2 PR |
3584 | c.packets[cmd == P_BITMAP]++; |
3585 | c.bytes[cmd == P_BITMAP] += sizeof(struct p_header80) + data_size; | |
b411b363 PR |
3586 | |
3587 | if (ret != OK) | |
3588 | break; | |
3589 | ||
02918be2 | 3590 | if (!drbd_recv_header(mdev, &cmd, &data_size)) |
b411b363 PR |
3591 | goto out; |
3592 | } while (ret == OK); | |
3593 | if (ret == FAILED) | |
3594 | goto out; | |
3595 | ||
3596 | INFO_bm_xfer_stats(mdev, "receive", &c); | |
3597 | ||
3598 | if (mdev->state.conn == C_WF_BITMAP_T) { | |
3599 | ok = !drbd_send_bitmap(mdev); | |
3600 | if (!ok) | |
3601 | goto out; | |
3602 | /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ | |
3603 | ok = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); | |
3604 | D_ASSERT(ok == SS_SUCCESS); | |
3605 | } else if (mdev->state.conn != C_WF_BITMAP_S) { | |
3606 | /* admin may have requested C_DISCONNECTING, | |
3607 | * other threads may have noticed network errors */ | |
3608 | dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n", | |
3609 | drbd_conn_str(mdev->state.conn)); | |
3610 | } | |
3611 | ||
3612 | ok = TRUE; | |
3613 | out: | |
3614 | drbd_bm_unlock(mdev); | |
3615 | if (ok && mdev->state.conn == C_WF_BITMAP_S) | |
3616 | drbd_start_resync(mdev, C_SYNC_SOURCE); | |
3617 | free_page((unsigned long) buffer); | |
3618 | return ok; | |
3619 | } | |
3620 | ||
02918be2 | 3621 | static int receive_skip(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
b411b363 PR |
3622 | { |
3623 | /* TODO zero copy sink :) */ | |
3624 | static char sink[128]; | |
3625 | int size, want, r; | |
3626 | ||
02918be2 PR |
3627 | dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n", |
3628 | cmd, data_size); | |
b411b363 | 3629 | |
02918be2 | 3630 | size = data_size; |
b411b363 PR |
3631 | while (size > 0) { |
3632 | want = min_t(int, size, sizeof(sink)); | |
3633 | r = drbd_recv(mdev, sink, want); | |
3634 | ERR_IF(r <= 0) break; | |
3635 | size -= r; | |
3636 | } | |
3637 | return size == 0; | |
3638 | } | |
3639 | ||
02918be2 | 3640 | static int receive_UnplugRemote(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) |
0ced55a3 | 3641 | { |
e7f52dfb LE |
3642 | if (mdev->state.disk >= D_INCONSISTENT) |
3643 | drbd_kick_lo(mdev); | |
0ced55a3 | 3644 | |
e7f52dfb LE |
3645 | /* Make sure we've acked all the TCP data associated |
3646 | * with the data requests being unplugged */ | |
3647 | drbd_tcp_quickack(mdev->data.socket); | |
0ced55a3 | 3648 | |
0ced55a3 PR |
3649 | return TRUE; |
3650 | } | |
3651 | ||
02918be2 PR |
3652 | typedef int (*drbd_cmd_handler_f)(struct drbd_conf *, enum drbd_packets cmd, unsigned int to_receive); |
3653 | ||
3654 | struct data_cmd { | |
3655 | int expect_payload; | |
3656 | size_t pkt_size; | |
3657 | drbd_cmd_handler_f function; | |
3658 | }; | |
3659 | ||
3660 | static struct data_cmd drbd_cmd_handler[] = { | |
3661 | [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, | |
3662 | [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, | |
3663 | [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , | |
3664 | [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , | |
3665 | [P_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } , | |
3666 | [P_COMPRESSED_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } , | |
3667 | [P_UNPLUG_REMOTE] = { 0, sizeof(struct p_header80), receive_UnplugRemote }, | |
3668 | [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
3669 | [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
3670 | [P_SYNC_PARAM] = { 1, sizeof(struct p_header80), receive_SyncParam }, | |
3671 | [P_SYNC_PARAM89] = { 1, sizeof(struct p_header80), receive_SyncParam }, | |
3672 | [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, | |
3673 | [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, | |
3674 | [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, | |
3675 | [P_STATE] = { 0, sizeof(struct p_state), receive_state }, | |
3676 | [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, | |
3677 | [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, | |
3678 | [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
3679 | [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
3680 | [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
3681 | [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, | |
b411b363 PR |
3682 | /* anything missing from this table is in |
3683 | * the asender_tbl, see get_asender_cmd */ | |
02918be2 | 3684 | [P_MAX_CMD] = { 0, 0, NULL }, |
b411b363 PR |
3685 | }; |
3686 | ||
02918be2 PR |
3687 | /* All handler functions that expect a sub-header get that sub-heder in |
3688 | mdev->data.rbuf.header.head.payload. | |
3689 | ||
3690 | Usually in mdev->data.rbuf.header.head the callback can find the usual | |
3691 | p_header, but they may not rely on that. Since there is also p_header95 ! | |
3692 | */ | |
b411b363 PR |
3693 | |
3694 | static void drbdd(struct drbd_conf *mdev) | |
3695 | { | |
02918be2 PR |
3696 | union p_header *header = &mdev->data.rbuf.header; |
3697 | unsigned int packet_size; | |
3698 | enum drbd_packets cmd; | |
3699 | size_t shs; /* sub header size */ | |
3700 | int rv; | |
b411b363 PR |
3701 | |
3702 | while (get_t_state(&mdev->receiver) == Running) { | |
3703 | drbd_thread_current_set_cpu(mdev); | |
02918be2 PR |
3704 | if (!drbd_recv_header(mdev, &cmd, &packet_size)) |
3705 | goto err_out; | |
3706 | ||
3707 | if (unlikely(cmd >= P_MAX_CMD || !drbd_cmd_handler[cmd].function)) { | |
3708 | dev_err(DEV, "unknown packet type %d, l: %d!\n", cmd, packet_size); | |
3709 | goto err_out; | |
0b33a916 | 3710 | } |
b411b363 | 3711 | |
02918be2 PR |
3712 | shs = drbd_cmd_handler[cmd].pkt_size - sizeof(union p_header); |
3713 | rv = drbd_recv(mdev, &header->h80.payload, shs); | |
3714 | if (unlikely(rv != shs)) { | |
3715 | dev_err(DEV, "short read while reading sub header: rv=%d\n", rv); | |
3716 | goto err_out; | |
3717 | } | |
b411b363 | 3718 | |
02918be2 PR |
3719 | if (packet_size - shs > 0 && !drbd_cmd_handler[cmd].expect_payload) { |
3720 | dev_err(DEV, "No payload expected %s l:%d\n", cmdname(cmd), packet_size); | |
3721 | goto err_out; | |
b411b363 | 3722 | } |
02918be2 PR |
3723 | |
3724 | rv = drbd_cmd_handler[cmd].function(mdev, cmd, packet_size - shs); | |
3725 | ||
3726 | if (unlikely(!rv)) { | |
b411b363 | 3727 | dev_err(DEV, "error receiving %s, l: %d!\n", |
02918be2 PR |
3728 | cmdname(cmd), packet_size); |
3729 | goto err_out; | |
b411b363 | 3730 | } |
b411b363 | 3731 | } |
02918be2 PR |
3732 | |
3733 | if (0) { | |
3734 | err_out: | |
3735 | drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); | |
3736 | } | |
b411b363 PR |
3737 | } |
3738 | ||
b411b363 PR |
3739 | void drbd_flush_workqueue(struct drbd_conf *mdev) |
3740 | { | |
3741 | struct drbd_wq_barrier barr; | |
3742 | ||
3743 | barr.w.cb = w_prev_work_done; | |
3744 | init_completion(&barr.done); | |
3745 | drbd_queue_work(&mdev->data.work, &barr.w); | |
3746 | wait_for_completion(&barr.done); | |
3747 | } | |
3748 | ||
f70b3511 PR |
3749 | void drbd_free_tl_hash(struct drbd_conf *mdev) |
3750 | { | |
3751 | struct hlist_head *h; | |
3752 | ||
3753 | spin_lock_irq(&mdev->req_lock); | |
3754 | ||
3755 | if (!mdev->tl_hash || mdev->state.conn != C_STANDALONE) { | |
3756 | spin_unlock_irq(&mdev->req_lock); | |
3757 | return; | |
3758 | } | |
3759 | /* paranoia code */ | |
3760 | for (h = mdev->ee_hash; h < mdev->ee_hash + mdev->ee_hash_s; h++) | |
3761 | if (h->first) | |
3762 | dev_err(DEV, "ASSERT FAILED ee_hash[%u].first == %p, expected NULL\n", | |
3763 | (int)(h - mdev->ee_hash), h->first); | |
3764 | kfree(mdev->ee_hash); | |
3765 | mdev->ee_hash = NULL; | |
3766 | mdev->ee_hash_s = 0; | |
3767 | ||
3768 | /* paranoia code */ | |
3769 | for (h = mdev->tl_hash; h < mdev->tl_hash + mdev->tl_hash_s; h++) | |
3770 | if (h->first) | |
3771 | dev_err(DEV, "ASSERT FAILED tl_hash[%u] == %p, expected NULL\n", | |
3772 | (int)(h - mdev->tl_hash), h->first); | |
3773 | kfree(mdev->tl_hash); | |
3774 | mdev->tl_hash = NULL; | |
3775 | mdev->tl_hash_s = 0; | |
3776 | spin_unlock_irq(&mdev->req_lock); | |
3777 | } | |
3778 | ||
b411b363 PR |
3779 | static void drbd_disconnect(struct drbd_conf *mdev) |
3780 | { | |
3781 | enum drbd_fencing_p fp; | |
3782 | union drbd_state os, ns; | |
3783 | int rv = SS_UNKNOWN_ERROR; | |
3784 | unsigned int i; | |
3785 | ||
3786 | if (mdev->state.conn == C_STANDALONE) | |
3787 | return; | |
3788 | if (mdev->state.conn >= C_WF_CONNECTION) | |
3789 | dev_err(DEV, "ASSERT FAILED cstate = %s, expected < WFConnection\n", | |
3790 | drbd_conn_str(mdev->state.conn)); | |
3791 | ||
3792 | /* asender does not clean up anything. it must not interfere, either */ | |
3793 | drbd_thread_stop(&mdev->asender); | |
b411b363 | 3794 | drbd_free_sock(mdev); |
b411b363 | 3795 | |
85719573 | 3796 | /* wait for current activity to cease. */ |
b411b363 PR |
3797 | spin_lock_irq(&mdev->req_lock); |
3798 | _drbd_wait_ee_list_empty(mdev, &mdev->active_ee); | |
3799 | _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee); | |
3800 | _drbd_wait_ee_list_empty(mdev, &mdev->read_ee); | |
3801 | spin_unlock_irq(&mdev->req_lock); | |
3802 | ||
3803 | /* We do not have data structures that would allow us to | |
3804 | * get the rs_pending_cnt down to 0 again. | |
3805 | * * On C_SYNC_TARGET we do not have any data structures describing | |
3806 | * the pending RSDataRequest's we have sent. | |
3807 | * * On C_SYNC_SOURCE there is no data structure that tracks | |
3808 | * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. | |
3809 | * And no, it is not the sum of the reference counts in the | |
3810 | * resync_LRU. The resync_LRU tracks the whole operation including | |
3811 | * the disk-IO, while the rs_pending_cnt only tracks the blocks | |
3812 | * on the fly. */ | |
3813 | drbd_rs_cancel_all(mdev); | |
3814 | mdev->rs_total = 0; | |
3815 | mdev->rs_failed = 0; | |
3816 | atomic_set(&mdev->rs_pending_cnt, 0); | |
3817 | wake_up(&mdev->misc_wait); | |
3818 | ||
3819 | /* make sure syncer is stopped and w_resume_next_sg queued */ | |
3820 | del_timer_sync(&mdev->resync_timer); | |
b411b363 PR |
3821 | resync_timer_fn((unsigned long)mdev); |
3822 | ||
b411b363 PR |
3823 | /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, |
3824 | * w_make_resync_request etc. which may still be on the worker queue | |
3825 | * to be "canceled" */ | |
3826 | drbd_flush_workqueue(mdev); | |
3827 | ||
3828 | /* This also does reclaim_net_ee(). If we do this too early, we might | |
3829 | * miss some resync ee and pages.*/ | |
3830 | drbd_process_done_ee(mdev); | |
3831 | ||
3832 | kfree(mdev->p_uuid); | |
3833 | mdev->p_uuid = NULL; | |
3834 | ||
fb22c402 | 3835 | if (!is_susp(mdev->state)) |
b411b363 PR |
3836 | tl_clear(mdev); |
3837 | ||
b411b363 PR |
3838 | dev_info(DEV, "Connection closed\n"); |
3839 | ||
3840 | drbd_md_sync(mdev); | |
3841 | ||
3842 | fp = FP_DONT_CARE; | |
3843 | if (get_ldev(mdev)) { | |
3844 | fp = mdev->ldev->dc.fencing; | |
3845 | put_ldev(mdev); | |
3846 | } | |
3847 | ||
87f7be4c PR |
3848 | if (mdev->state.role == R_PRIMARY && fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN) |
3849 | drbd_try_outdate_peer_async(mdev); | |
b411b363 PR |
3850 | |
3851 | spin_lock_irq(&mdev->req_lock); | |
3852 | os = mdev->state; | |
3853 | if (os.conn >= C_UNCONNECTED) { | |
3854 | /* Do not restart in case we are C_DISCONNECTING */ | |
3855 | ns = os; | |
3856 | ns.conn = C_UNCONNECTED; | |
3857 | rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); | |
3858 | } | |
3859 | spin_unlock_irq(&mdev->req_lock); | |
3860 | ||
3861 | if (os.conn == C_DISCONNECTING) { | |
84dfb9f5 | 3862 | wait_event(mdev->net_cnt_wait, atomic_read(&mdev->net_cnt) == 0); |
b411b363 | 3863 | |
fb22c402 | 3864 | if (!is_susp(mdev->state)) { |
f70b3511 PR |
3865 | /* we must not free the tl_hash |
3866 | * while application io is still on the fly */ | |
3867 | wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt)); | |
3868 | drbd_free_tl_hash(mdev); | |
3869 | } | |
b411b363 PR |
3870 | |
3871 | crypto_free_hash(mdev->cram_hmac_tfm); | |
3872 | mdev->cram_hmac_tfm = NULL; | |
3873 | ||
3874 | kfree(mdev->net_conf); | |
3875 | mdev->net_conf = NULL; | |
3876 | drbd_request_state(mdev, NS(conn, C_STANDALONE)); | |
3877 | } | |
3878 | ||
3879 | /* tcp_close and release of sendpage pages can be deferred. I don't | |
3880 | * want to use SO_LINGER, because apparently it can be deferred for | |
3881 | * more than 20 seconds (longest time I checked). | |
3882 | * | |
3883 | * Actually we don't care for exactly when the network stack does its | |
3884 | * put_page(), but release our reference on these pages right here. | |
3885 | */ | |
3886 | i = drbd_release_ee(mdev, &mdev->net_ee); | |
3887 | if (i) | |
3888 | dev_info(DEV, "net_ee not empty, killed %u entries\n", i); | |
435f0740 LE |
3889 | i = atomic_read(&mdev->pp_in_use_by_net); |
3890 | if (i) | |
3891 | dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i); | |
b411b363 PR |
3892 | i = atomic_read(&mdev->pp_in_use); |
3893 | if (i) | |
45bb912b | 3894 | dev_info(DEV, "pp_in_use = %d, expected 0\n", i); |
b411b363 PR |
3895 | |
3896 | D_ASSERT(list_empty(&mdev->read_ee)); | |
3897 | D_ASSERT(list_empty(&mdev->active_ee)); | |
3898 | D_ASSERT(list_empty(&mdev->sync_ee)); | |
3899 | D_ASSERT(list_empty(&mdev->done_ee)); | |
3900 | ||
3901 | /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ | |
3902 | atomic_set(&mdev->current_epoch->epoch_size, 0); | |
3903 | D_ASSERT(list_empty(&mdev->current_epoch->list)); | |
3904 | } | |
3905 | ||
3906 | /* | |
3907 | * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version | |
3908 | * we can agree on is stored in agreed_pro_version. | |
3909 | * | |
3910 | * feature flags and the reserved array should be enough room for future | |
3911 | * enhancements of the handshake protocol, and possible plugins... | |
3912 | * | |
3913 | * for now, they are expected to be zero, but ignored. | |
3914 | */ | |
3915 | static int drbd_send_handshake(struct drbd_conf *mdev) | |
3916 | { | |
3917 | /* ASSERT current == mdev->receiver ... */ | |
3918 | struct p_handshake *p = &mdev->data.sbuf.handshake; | |
3919 | int ok; | |
3920 | ||
3921 | if (mutex_lock_interruptible(&mdev->data.mutex)) { | |
3922 | dev_err(DEV, "interrupted during initial handshake\n"); | |
3923 | return 0; /* interrupted. not ok. */ | |
3924 | } | |
3925 | ||
3926 | if (mdev->data.socket == NULL) { | |
3927 | mutex_unlock(&mdev->data.mutex); | |
3928 | return 0; | |
3929 | } | |
3930 | ||
3931 | memset(p, 0, sizeof(*p)); | |
3932 | p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); | |
3933 | p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); | |
3934 | ok = _drbd_send_cmd( mdev, mdev->data.socket, P_HAND_SHAKE, | |
0b70a13d | 3935 | (struct p_header80 *)p, sizeof(*p), 0 ); |
b411b363 PR |
3936 | mutex_unlock(&mdev->data.mutex); |
3937 | return ok; | |
3938 | } | |
3939 | ||
3940 | /* | |
3941 | * return values: | |
3942 | * 1 yes, we have a valid connection | |
3943 | * 0 oops, did not work out, please try again | |
3944 | * -1 peer talks different language, | |
3945 | * no point in trying again, please go standalone. | |
3946 | */ | |
3947 | static int drbd_do_handshake(struct drbd_conf *mdev) | |
3948 | { | |
3949 | /* ASSERT current == mdev->receiver ... */ | |
3950 | struct p_handshake *p = &mdev->data.rbuf.handshake; | |
02918be2 PR |
3951 | const int expect = sizeof(struct p_handshake) - sizeof(struct p_header80); |
3952 | unsigned int length; | |
3953 | enum drbd_packets cmd; | |
b411b363 PR |
3954 | int rv; |
3955 | ||
3956 | rv = drbd_send_handshake(mdev); | |
3957 | if (!rv) | |
3958 | return 0; | |
3959 | ||
02918be2 | 3960 | rv = drbd_recv_header(mdev, &cmd, &length); |
b411b363 PR |
3961 | if (!rv) |
3962 | return 0; | |
3963 | ||
02918be2 | 3964 | if (cmd != P_HAND_SHAKE) { |
b411b363 | 3965 | dev_err(DEV, "expected HandShake packet, received: %s (0x%04x)\n", |
02918be2 | 3966 | cmdname(cmd), cmd); |
b411b363 PR |
3967 | return -1; |
3968 | } | |
3969 | ||
02918be2 | 3970 | if (length != expect) { |
b411b363 | 3971 | dev_err(DEV, "expected HandShake length: %u, received: %u\n", |
02918be2 | 3972 | expect, length); |
b411b363 PR |
3973 | return -1; |
3974 | } | |
3975 | ||
3976 | rv = drbd_recv(mdev, &p->head.payload, expect); | |
3977 | ||
3978 | if (rv != expect) { | |
3979 | dev_err(DEV, "short read receiving handshake packet: l=%u\n", rv); | |
3980 | return 0; | |
3981 | } | |
3982 | ||
b411b363 PR |
3983 | p->protocol_min = be32_to_cpu(p->protocol_min); |
3984 | p->protocol_max = be32_to_cpu(p->protocol_max); | |
3985 | if (p->protocol_max == 0) | |
3986 | p->protocol_max = p->protocol_min; | |
3987 | ||
3988 | if (PRO_VERSION_MAX < p->protocol_min || | |
3989 | PRO_VERSION_MIN > p->protocol_max) | |
3990 | goto incompat; | |
3991 | ||
3992 | mdev->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); | |
3993 | ||
3994 | dev_info(DEV, "Handshake successful: " | |
3995 | "Agreed network protocol version %d\n", mdev->agreed_pro_version); | |
3996 | ||
3997 | return 1; | |
3998 | ||
3999 | incompat: | |
4000 | dev_err(DEV, "incompatible DRBD dialects: " | |
4001 | "I support %d-%d, peer supports %d-%d\n", | |
4002 | PRO_VERSION_MIN, PRO_VERSION_MAX, | |
4003 | p->protocol_min, p->protocol_max); | |
4004 | return -1; | |
4005 | } | |
4006 | ||
4007 | #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) | |
4008 | static int drbd_do_auth(struct drbd_conf *mdev) | |
4009 | { | |
4010 | dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); | |
4011 | dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); | |
b10d96cb | 4012 | return -1; |
b411b363 PR |
4013 | } |
4014 | #else | |
4015 | #define CHALLENGE_LEN 64 | |
b10d96cb JT |
4016 | |
4017 | /* Return value: | |
4018 | 1 - auth succeeded, | |
4019 | 0 - failed, try again (network error), | |
4020 | -1 - auth failed, don't try again. | |
4021 | */ | |
4022 | ||
b411b363 PR |
4023 | static int drbd_do_auth(struct drbd_conf *mdev) |
4024 | { | |
4025 | char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ | |
4026 | struct scatterlist sg; | |
4027 | char *response = NULL; | |
4028 | char *right_response = NULL; | |
4029 | char *peers_ch = NULL; | |
b411b363 PR |
4030 | unsigned int key_len = strlen(mdev->net_conf->shared_secret); |
4031 | unsigned int resp_size; | |
4032 | struct hash_desc desc; | |
02918be2 PR |
4033 | enum drbd_packets cmd; |
4034 | unsigned int length; | |
b411b363 PR |
4035 | int rv; |
4036 | ||
4037 | desc.tfm = mdev->cram_hmac_tfm; | |
4038 | desc.flags = 0; | |
4039 | ||
4040 | rv = crypto_hash_setkey(mdev->cram_hmac_tfm, | |
4041 | (u8 *)mdev->net_conf->shared_secret, key_len); | |
4042 | if (rv) { | |
4043 | dev_err(DEV, "crypto_hash_setkey() failed with %d\n", rv); | |
b10d96cb | 4044 | rv = -1; |
b411b363 PR |
4045 | goto fail; |
4046 | } | |
4047 | ||
4048 | get_random_bytes(my_challenge, CHALLENGE_LEN); | |
4049 | ||
4050 | rv = drbd_send_cmd2(mdev, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN); | |
4051 | if (!rv) | |
4052 | goto fail; | |
4053 | ||
02918be2 | 4054 | rv = drbd_recv_header(mdev, &cmd, &length); |
b411b363 PR |
4055 | if (!rv) |
4056 | goto fail; | |
4057 | ||
02918be2 | 4058 | if (cmd != P_AUTH_CHALLENGE) { |
b411b363 | 4059 | dev_err(DEV, "expected AuthChallenge packet, received: %s (0x%04x)\n", |
02918be2 | 4060 | cmdname(cmd), cmd); |
b411b363 PR |
4061 | rv = 0; |
4062 | goto fail; | |
4063 | } | |
4064 | ||
02918be2 | 4065 | if (length > CHALLENGE_LEN * 2) { |
b411b363 | 4066 | dev_err(DEV, "expected AuthChallenge payload too big.\n"); |
b10d96cb | 4067 | rv = -1; |
b411b363 PR |
4068 | goto fail; |
4069 | } | |
4070 | ||
02918be2 | 4071 | peers_ch = kmalloc(length, GFP_NOIO); |
b411b363 PR |
4072 | if (peers_ch == NULL) { |
4073 | dev_err(DEV, "kmalloc of peers_ch failed\n"); | |
b10d96cb | 4074 | rv = -1; |
b411b363 PR |
4075 | goto fail; |
4076 | } | |
4077 | ||
02918be2 | 4078 | rv = drbd_recv(mdev, peers_ch, length); |
b411b363 | 4079 | |
02918be2 | 4080 | if (rv != length) { |
b411b363 PR |
4081 | dev_err(DEV, "short read AuthChallenge: l=%u\n", rv); |
4082 | rv = 0; | |
4083 | goto fail; | |
4084 | } | |
4085 | ||
4086 | resp_size = crypto_hash_digestsize(mdev->cram_hmac_tfm); | |
4087 | response = kmalloc(resp_size, GFP_NOIO); | |
4088 | if (response == NULL) { | |
4089 | dev_err(DEV, "kmalloc of response failed\n"); | |
b10d96cb | 4090 | rv = -1; |
b411b363 PR |
4091 | goto fail; |
4092 | } | |
4093 | ||
4094 | sg_init_table(&sg, 1); | |
02918be2 | 4095 | sg_set_buf(&sg, peers_ch, length); |
b411b363 PR |
4096 | |
4097 | rv = crypto_hash_digest(&desc, &sg, sg.length, response); | |
4098 | if (rv) { | |
4099 | dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv); | |
b10d96cb | 4100 | rv = -1; |
b411b363 PR |
4101 | goto fail; |
4102 | } | |
4103 | ||
4104 | rv = drbd_send_cmd2(mdev, P_AUTH_RESPONSE, response, resp_size); | |
4105 | if (!rv) | |
4106 | goto fail; | |
4107 | ||
02918be2 | 4108 | rv = drbd_recv_header(mdev, &cmd, &length); |
b411b363 PR |
4109 | if (!rv) |
4110 | goto fail; | |
4111 | ||
02918be2 | 4112 | if (cmd != P_AUTH_RESPONSE) { |
b411b363 | 4113 | dev_err(DEV, "expected AuthResponse packet, received: %s (0x%04x)\n", |
02918be2 | 4114 | cmdname(cmd), cmd); |
b411b363 PR |
4115 | rv = 0; |
4116 | goto fail; | |
4117 | } | |
4118 | ||
02918be2 | 4119 | if (length != resp_size) { |
b411b363 PR |
4120 | dev_err(DEV, "expected AuthResponse payload of wrong size\n"); |
4121 | rv = 0; | |
4122 | goto fail; | |
4123 | } | |
4124 | ||
4125 | rv = drbd_recv(mdev, response , resp_size); | |
4126 | ||
4127 | if (rv != resp_size) { | |
4128 | dev_err(DEV, "short read receiving AuthResponse: l=%u\n", rv); | |
4129 | rv = 0; | |
4130 | goto fail; | |
4131 | } | |
4132 | ||
4133 | right_response = kmalloc(resp_size, GFP_NOIO); | |
2d1ee87d | 4134 | if (right_response == NULL) { |
b411b363 | 4135 | dev_err(DEV, "kmalloc of right_response failed\n"); |
b10d96cb | 4136 | rv = -1; |
b411b363 PR |
4137 | goto fail; |
4138 | } | |
4139 | ||
4140 | sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); | |
4141 | ||
4142 | rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); | |
4143 | if (rv) { | |
4144 | dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv); | |
b10d96cb | 4145 | rv = -1; |
b411b363 PR |
4146 | goto fail; |
4147 | } | |
4148 | ||
4149 | rv = !memcmp(response, right_response, resp_size); | |
4150 | ||
4151 | if (rv) | |
4152 | dev_info(DEV, "Peer authenticated using %d bytes of '%s' HMAC\n", | |
4153 | resp_size, mdev->net_conf->cram_hmac_alg); | |
b10d96cb JT |
4154 | else |
4155 | rv = -1; | |
b411b363 PR |
4156 | |
4157 | fail: | |
4158 | kfree(peers_ch); | |
4159 | kfree(response); | |
4160 | kfree(right_response); | |
4161 | ||
4162 | return rv; | |
4163 | } | |
4164 | #endif | |
4165 | ||
4166 | int drbdd_init(struct drbd_thread *thi) | |
4167 | { | |
4168 | struct drbd_conf *mdev = thi->mdev; | |
4169 | unsigned int minor = mdev_to_minor(mdev); | |
4170 | int h; | |
4171 | ||
4172 | sprintf(current->comm, "drbd%d_receiver", minor); | |
4173 | ||
4174 | dev_info(DEV, "receiver (re)started\n"); | |
4175 | ||
4176 | do { | |
4177 | h = drbd_connect(mdev); | |
4178 | if (h == 0) { | |
4179 | drbd_disconnect(mdev); | |
4180 | __set_current_state(TASK_INTERRUPTIBLE); | |
4181 | schedule_timeout(HZ); | |
4182 | } | |
4183 | if (h == -1) { | |
4184 | dev_warn(DEV, "Discarding network configuration.\n"); | |
4185 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
4186 | } | |
4187 | } while (h == 0); | |
4188 | ||
4189 | if (h > 0) { | |
4190 | if (get_net_conf(mdev)) { | |
4191 | drbdd(mdev); | |
4192 | put_net_conf(mdev); | |
4193 | } | |
4194 | } | |
4195 | ||
4196 | drbd_disconnect(mdev); | |
4197 | ||
4198 | dev_info(DEV, "receiver terminated\n"); | |
4199 | return 0; | |
4200 | } | |
4201 | ||
4202 | /* ********* acknowledge sender ******** */ | |
4203 | ||
0b70a13d | 4204 | static int got_RqSReply(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4205 | { |
4206 | struct p_req_state_reply *p = (struct p_req_state_reply *)h; | |
4207 | ||
4208 | int retcode = be32_to_cpu(p->retcode); | |
4209 | ||
4210 | if (retcode >= SS_SUCCESS) { | |
4211 | set_bit(CL_ST_CHG_SUCCESS, &mdev->flags); | |
4212 | } else { | |
4213 | set_bit(CL_ST_CHG_FAIL, &mdev->flags); | |
4214 | dev_err(DEV, "Requested state change failed by peer: %s (%d)\n", | |
4215 | drbd_set_st_err_str(retcode), retcode); | |
4216 | } | |
4217 | wake_up(&mdev->state_wait); | |
4218 | ||
4219 | return TRUE; | |
4220 | } | |
4221 | ||
0b70a13d | 4222 | static int got_Ping(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4223 | { |
4224 | return drbd_send_ping_ack(mdev); | |
4225 | ||
4226 | } | |
4227 | ||
0b70a13d | 4228 | static int got_PingAck(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4229 | { |
4230 | /* restore idle timeout */ | |
4231 | mdev->meta.socket->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ; | |
309d1608 PR |
4232 | if (!test_and_set_bit(GOT_PING_ACK, &mdev->flags)) |
4233 | wake_up(&mdev->misc_wait); | |
b411b363 PR |
4234 | |
4235 | return TRUE; | |
4236 | } | |
4237 | ||
0b70a13d | 4238 | static int got_IsInSync(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4239 | { |
4240 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4241 | sector_t sector = be64_to_cpu(p->sector); | |
4242 | int blksize = be32_to_cpu(p->blksize); | |
4243 | ||
4244 | D_ASSERT(mdev->agreed_pro_version >= 89); | |
4245 | ||
4246 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4247 | ||
1d53f09e LE |
4248 | if (get_ldev(mdev)) { |
4249 | drbd_rs_complete_io(mdev, sector); | |
4250 | drbd_set_in_sync(mdev, sector, blksize); | |
4251 | /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ | |
4252 | mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); | |
4253 | put_ldev(mdev); | |
4254 | } | |
b411b363 | 4255 | dec_rs_pending(mdev); |
778f271d | 4256 | atomic_add(blksize >> 9, &mdev->rs_sect_in); |
b411b363 PR |
4257 | |
4258 | return TRUE; | |
4259 | } | |
4260 | ||
4261 | /* when we receive the ACK for a write request, | |
4262 | * verify that we actually know about it */ | |
4263 | static struct drbd_request *_ack_id_to_req(struct drbd_conf *mdev, | |
4264 | u64 id, sector_t sector) | |
4265 | { | |
4266 | struct hlist_head *slot = tl_hash_slot(mdev, sector); | |
4267 | struct hlist_node *n; | |
4268 | struct drbd_request *req; | |
4269 | ||
4270 | hlist_for_each_entry(req, n, slot, colision) { | |
4271 | if ((unsigned long)req == (unsigned long)id) { | |
4272 | if (req->sector != sector) { | |
4273 | dev_err(DEV, "_ack_id_to_req: found req %p but it has " | |
4274 | "wrong sector (%llus versus %llus)\n", req, | |
4275 | (unsigned long long)req->sector, | |
4276 | (unsigned long long)sector); | |
4277 | break; | |
4278 | } | |
4279 | return req; | |
4280 | } | |
4281 | } | |
4282 | dev_err(DEV, "_ack_id_to_req: failed to find req %p, sector %llus in list\n", | |
4283 | (void *)(unsigned long)id, (unsigned long long)sector); | |
4284 | return NULL; | |
4285 | } | |
4286 | ||
4287 | typedef struct drbd_request *(req_validator_fn) | |
4288 | (struct drbd_conf *mdev, u64 id, sector_t sector); | |
4289 | ||
4290 | static int validate_req_change_req_state(struct drbd_conf *mdev, | |
4291 | u64 id, sector_t sector, req_validator_fn validator, | |
4292 | const char *func, enum drbd_req_event what) | |
4293 | { | |
4294 | struct drbd_request *req; | |
4295 | struct bio_and_error m; | |
4296 | ||
4297 | spin_lock_irq(&mdev->req_lock); | |
4298 | req = validator(mdev, id, sector); | |
4299 | if (unlikely(!req)) { | |
4300 | spin_unlock_irq(&mdev->req_lock); | |
4301 | dev_err(DEV, "%s: got a corrupt block_id/sector pair\n", func); | |
4302 | return FALSE; | |
4303 | } | |
4304 | __req_mod(req, what, &m); | |
4305 | spin_unlock_irq(&mdev->req_lock); | |
4306 | ||
4307 | if (m.bio) | |
4308 | complete_master_bio(mdev, &m); | |
4309 | return TRUE; | |
4310 | } | |
4311 | ||
0b70a13d | 4312 | static int got_BlockAck(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4313 | { |
4314 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4315 | sector_t sector = be64_to_cpu(p->sector); | |
4316 | int blksize = be32_to_cpu(p->blksize); | |
4317 | enum drbd_req_event what; | |
4318 | ||
4319 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4320 | ||
4321 | if (is_syncer_block_id(p->block_id)) { | |
4322 | drbd_set_in_sync(mdev, sector, blksize); | |
4323 | dec_rs_pending(mdev); | |
4324 | return TRUE; | |
4325 | } | |
4326 | switch (be16_to_cpu(h->command)) { | |
4327 | case P_RS_WRITE_ACK: | |
4328 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
4329 | what = write_acked_by_peer_and_sis; | |
4330 | break; | |
4331 | case P_WRITE_ACK: | |
4332 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
4333 | what = write_acked_by_peer; | |
4334 | break; | |
4335 | case P_RECV_ACK: | |
4336 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_B); | |
4337 | what = recv_acked_by_peer; | |
4338 | break; | |
4339 | case P_DISCARD_ACK: | |
4340 | D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); | |
4341 | what = conflict_discarded_by_peer; | |
4342 | break; | |
4343 | default: | |
4344 | D_ASSERT(0); | |
4345 | return FALSE; | |
4346 | } | |
4347 | ||
4348 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
4349 | _ack_id_to_req, __func__ , what); | |
4350 | } | |
4351 | ||
0b70a13d | 4352 | static int got_NegAck(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4353 | { |
4354 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4355 | sector_t sector = be64_to_cpu(p->sector); | |
4356 | ||
4357 | if (__ratelimit(&drbd_ratelimit_state)) | |
4358 | dev_warn(DEV, "Got NegAck packet. Peer is in troubles?\n"); | |
4359 | ||
4360 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4361 | ||
4362 | if (is_syncer_block_id(p->block_id)) { | |
4363 | int size = be32_to_cpu(p->blksize); | |
4364 | dec_rs_pending(mdev); | |
4365 | drbd_rs_failed_io(mdev, sector, size); | |
4366 | return TRUE; | |
4367 | } | |
4368 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
4369 | _ack_id_to_req, __func__ , neg_acked); | |
4370 | } | |
4371 | ||
0b70a13d | 4372 | static int got_NegDReply(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4373 | { |
4374 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4375 | sector_t sector = be64_to_cpu(p->sector); | |
4376 | ||
4377 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4378 | dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n", | |
4379 | (unsigned long long)sector, be32_to_cpu(p->blksize)); | |
4380 | ||
4381 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
4382 | _ar_id_to_req, __func__ , neg_acked); | |
4383 | } | |
4384 | ||
0b70a13d | 4385 | static int got_NegRSDReply(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4386 | { |
4387 | sector_t sector; | |
4388 | int size; | |
4389 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4390 | ||
4391 | sector = be64_to_cpu(p->sector); | |
4392 | size = be32_to_cpu(p->blksize); | |
b411b363 PR |
4393 | |
4394 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4395 | ||
4396 | dec_rs_pending(mdev); | |
4397 | ||
4398 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
4399 | drbd_rs_complete_io(mdev, sector); | |
4400 | drbd_rs_failed_io(mdev, sector, size); | |
4401 | put_ldev(mdev); | |
4402 | } | |
4403 | ||
4404 | return TRUE; | |
4405 | } | |
4406 | ||
0b70a13d | 4407 | static int got_BarrierAck(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4408 | { |
4409 | struct p_barrier_ack *p = (struct p_barrier_ack *)h; | |
4410 | ||
4411 | tl_release(mdev, p->barrier, be32_to_cpu(p->set_size)); | |
4412 | ||
4413 | return TRUE; | |
4414 | } | |
4415 | ||
0b70a13d | 4416 | static int got_OVResult(struct drbd_conf *mdev, struct p_header80 *h) |
b411b363 PR |
4417 | { |
4418 | struct p_block_ack *p = (struct p_block_ack *)h; | |
4419 | struct drbd_work *w; | |
4420 | sector_t sector; | |
4421 | int size; | |
4422 | ||
4423 | sector = be64_to_cpu(p->sector); | |
4424 | size = be32_to_cpu(p->blksize); | |
4425 | ||
4426 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4427 | ||
4428 | if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) | |
4429 | drbd_ov_oos_found(mdev, sector, size); | |
4430 | else | |
4431 | ov_oos_print(mdev); | |
4432 | ||
1d53f09e LE |
4433 | if (!get_ldev(mdev)) |
4434 | return TRUE; | |
4435 | ||
b411b363 PR |
4436 | drbd_rs_complete_io(mdev, sector); |
4437 | dec_rs_pending(mdev); | |
4438 | ||
4439 | if (--mdev->ov_left == 0) { | |
4440 | w = kmalloc(sizeof(*w), GFP_NOIO); | |
4441 | if (w) { | |
4442 | w->cb = w_ov_finished; | |
4443 | drbd_queue_work_front(&mdev->data.work, w); | |
4444 | } else { | |
4445 | dev_err(DEV, "kmalloc(w) failed."); | |
4446 | ov_oos_print(mdev); | |
4447 | drbd_resync_finished(mdev); | |
4448 | } | |
4449 | } | |
1d53f09e | 4450 | put_ldev(mdev); |
b411b363 PR |
4451 | return TRUE; |
4452 | } | |
4453 | ||
02918be2 | 4454 | static int got_skip(struct drbd_conf *mdev, struct p_header80 *h) |
0ced55a3 | 4455 | { |
0ced55a3 PR |
4456 | return TRUE; |
4457 | } | |
4458 | ||
b411b363 PR |
4459 | struct asender_cmd { |
4460 | size_t pkt_size; | |
0b70a13d | 4461 | int (*process)(struct drbd_conf *mdev, struct p_header80 *h); |
b411b363 PR |
4462 | }; |
4463 | ||
4464 | static struct asender_cmd *get_asender_cmd(int cmd) | |
4465 | { | |
4466 | static struct asender_cmd asender_tbl[] = { | |
4467 | /* anything missing from this table is in | |
4468 | * the drbd_cmd_handler (drbd_default_handler) table, | |
4469 | * see the beginning of drbdd() */ | |
0b70a13d PR |
4470 | [P_PING] = { sizeof(struct p_header80), got_Ping }, |
4471 | [P_PING_ACK] = { sizeof(struct p_header80), got_PingAck }, | |
b411b363 PR |
4472 | [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, |
4473 | [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4474 | [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4475 | [P_DISCARD_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4476 | [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, | |
4477 | [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, | |
4478 | [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply}, | |
4479 | [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, | |
4480 | [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, | |
4481 | [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, | |
4482 | [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, | |
02918be2 | 4483 | [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, |
b411b363 PR |
4484 | [P_MAX_CMD] = { 0, NULL }, |
4485 | }; | |
4486 | if (cmd > P_MAX_CMD || asender_tbl[cmd].process == NULL) | |
4487 | return NULL; | |
4488 | return &asender_tbl[cmd]; | |
4489 | } | |
4490 | ||
4491 | int drbd_asender(struct drbd_thread *thi) | |
4492 | { | |
4493 | struct drbd_conf *mdev = thi->mdev; | |
02918be2 | 4494 | struct p_header80 *h = &mdev->meta.rbuf.header.h80; |
b411b363 PR |
4495 | struct asender_cmd *cmd = NULL; |
4496 | ||
4497 | int rv, len; | |
4498 | void *buf = h; | |
4499 | int received = 0; | |
0b70a13d | 4500 | int expect = sizeof(struct p_header80); |
b411b363 PR |
4501 | int empty; |
4502 | ||
4503 | sprintf(current->comm, "drbd%d_asender", mdev_to_minor(mdev)); | |
4504 | ||
4505 | current->policy = SCHED_RR; /* Make this a realtime task! */ | |
4506 | current->rt_priority = 2; /* more important than all other tasks */ | |
4507 | ||
4508 | while (get_t_state(thi) == Running) { | |
4509 | drbd_thread_current_set_cpu(mdev); | |
4510 | if (test_and_clear_bit(SEND_PING, &mdev->flags)) { | |
4511 | ERR_IF(!drbd_send_ping(mdev)) goto reconnect; | |
4512 | mdev->meta.socket->sk->sk_rcvtimeo = | |
4513 | mdev->net_conf->ping_timeo*HZ/10; | |
4514 | } | |
4515 | ||
4516 | /* conditionally cork; | |
4517 | * it may hurt latency if we cork without much to send */ | |
4518 | if (!mdev->net_conf->no_cork && | |
4519 | 3 < atomic_read(&mdev->unacked_cnt)) | |
4520 | drbd_tcp_cork(mdev->meta.socket); | |
4521 | while (1) { | |
4522 | clear_bit(SIGNAL_ASENDER, &mdev->flags); | |
4523 | flush_signals(current); | |
4524 | if (!drbd_process_done_ee(mdev)) { | |
4525 | dev_err(DEV, "process_done_ee() = NOT_OK\n"); | |
4526 | goto reconnect; | |
4527 | } | |
4528 | /* to avoid race with newly queued ACKs */ | |
4529 | set_bit(SIGNAL_ASENDER, &mdev->flags); | |
4530 | spin_lock_irq(&mdev->req_lock); | |
4531 | empty = list_empty(&mdev->done_ee); | |
4532 | spin_unlock_irq(&mdev->req_lock); | |
4533 | /* new ack may have been queued right here, | |
4534 | * but then there is also a signal pending, | |
4535 | * and we start over... */ | |
4536 | if (empty) | |
4537 | break; | |
4538 | } | |
4539 | /* but unconditionally uncork unless disabled */ | |
4540 | if (!mdev->net_conf->no_cork) | |
4541 | drbd_tcp_uncork(mdev->meta.socket); | |
4542 | ||
4543 | /* short circuit, recv_msg would return EINTR anyways. */ | |
4544 | if (signal_pending(current)) | |
4545 | continue; | |
4546 | ||
4547 | rv = drbd_recv_short(mdev, mdev->meta.socket, | |
4548 | buf, expect-received, 0); | |
4549 | clear_bit(SIGNAL_ASENDER, &mdev->flags); | |
4550 | ||
4551 | flush_signals(current); | |
4552 | ||
4553 | /* Note: | |
4554 | * -EINTR (on meta) we got a signal | |
4555 | * -EAGAIN (on meta) rcvtimeo expired | |
4556 | * -ECONNRESET other side closed the connection | |
4557 | * -ERESTARTSYS (on data) we got a signal | |
4558 | * rv < 0 other than above: unexpected error! | |
4559 | * rv == expected: full header or command | |
4560 | * rv < expected: "woken" by signal during receive | |
4561 | * rv == 0 : "connection shut down by peer" | |
4562 | */ | |
4563 | if (likely(rv > 0)) { | |
4564 | received += rv; | |
4565 | buf += rv; | |
4566 | } else if (rv == 0) { | |
4567 | dev_err(DEV, "meta connection shut down by peer.\n"); | |
4568 | goto reconnect; | |
4569 | } else if (rv == -EAGAIN) { | |
4570 | if (mdev->meta.socket->sk->sk_rcvtimeo == | |
4571 | mdev->net_conf->ping_timeo*HZ/10) { | |
4572 | dev_err(DEV, "PingAck did not arrive in time.\n"); | |
4573 | goto reconnect; | |
4574 | } | |
4575 | set_bit(SEND_PING, &mdev->flags); | |
4576 | continue; | |
4577 | } else if (rv == -EINTR) { | |
4578 | continue; | |
4579 | } else { | |
4580 | dev_err(DEV, "sock_recvmsg returned %d\n", rv); | |
4581 | goto reconnect; | |
4582 | } | |
4583 | ||
4584 | if (received == expect && cmd == NULL) { | |
4585 | if (unlikely(h->magic != BE_DRBD_MAGIC)) { | |
004352fa LE |
4586 | dev_err(DEV, "magic?? on meta m: 0x%08x c: %d l: %d\n", |
4587 | be32_to_cpu(h->magic), | |
4588 | be16_to_cpu(h->command), | |
4589 | be16_to_cpu(h->length)); | |
b411b363 PR |
4590 | goto reconnect; |
4591 | } | |
4592 | cmd = get_asender_cmd(be16_to_cpu(h->command)); | |
4593 | len = be16_to_cpu(h->length); | |
4594 | if (unlikely(cmd == NULL)) { | |
004352fa LE |
4595 | dev_err(DEV, "unknown command?? on meta m: 0x%08x c: %d l: %d\n", |
4596 | be32_to_cpu(h->magic), | |
4597 | be16_to_cpu(h->command), | |
4598 | be16_to_cpu(h->length)); | |
b411b363 PR |
4599 | goto disconnect; |
4600 | } | |
4601 | expect = cmd->pkt_size; | |
0b70a13d | 4602 | ERR_IF(len != expect-sizeof(struct p_header80)) |
b411b363 | 4603 | goto reconnect; |
b411b363 PR |
4604 | } |
4605 | if (received == expect) { | |
4606 | D_ASSERT(cmd != NULL); | |
b411b363 PR |
4607 | if (!cmd->process(mdev, h)) |
4608 | goto reconnect; | |
4609 | ||
4610 | buf = h; | |
4611 | received = 0; | |
0b70a13d | 4612 | expect = sizeof(struct p_header80); |
b411b363 PR |
4613 | cmd = NULL; |
4614 | } | |
4615 | } | |
4616 | ||
4617 | if (0) { | |
4618 | reconnect: | |
4619 | drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE)); | |
4620 | } | |
4621 | if (0) { | |
4622 | disconnect: | |
4623 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
4624 | } | |
4625 | clear_bit(SIGNAL_ASENDER, &mdev->flags); | |
4626 | ||
4627 | D_ASSERT(mdev->state.conn < C_CONNECTED); | |
4628 | dev_info(DEV, "asender terminated\n"); | |
4629 | ||
4630 | return 0; | |
4631 | } |