Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_receiver.c | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | drbd is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation; either version 2, or (at your option) | |
13 | any later version. | |
14 | ||
15 | drbd is distributed in the hope that it will be useful, | |
16 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | GNU General Public License for more details. | |
19 | ||
20 | You should have received a copy of the GNU General Public License | |
21 | along with drbd; see the file COPYING. If not, write to | |
22 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | */ | |
24 | ||
25 | ||
b411b363 PR |
26 | #include <linux/module.h> |
27 | ||
28 | #include <asm/uaccess.h> | |
29 | #include <net/sock.h> | |
30 | ||
b411b363 PR |
31 | #include <linux/drbd.h> |
32 | #include <linux/fs.h> | |
33 | #include <linux/file.h> | |
34 | #include <linux/in.h> | |
35 | #include <linux/mm.h> | |
36 | #include <linux/memcontrol.h> | |
37 | #include <linux/mm_inline.h> | |
38 | #include <linux/slab.h> | |
b411b363 PR |
39 | #include <linux/pkt_sched.h> |
40 | #define __KERNEL_SYSCALLS__ | |
41 | #include <linux/unistd.h> | |
42 | #include <linux/vmalloc.h> | |
43 | #include <linux/random.h> | |
b411b363 PR |
44 | #include <linux/string.h> |
45 | #include <linux/scatterlist.h> | |
46 | #include "drbd_int.h" | |
b411b363 PR |
47 | #include "drbd_req.h" |
48 | ||
49 | #include "drbd_vli.h" | |
50 | ||
77351055 PR |
51 | struct packet_info { |
52 | enum drbd_packet cmd; | |
e2857216 AG |
53 | unsigned int size; |
54 | unsigned int vnr; | |
e658983a | 55 | void *data; |
77351055 PR |
56 | }; |
57 | ||
b411b363 PR |
58 | enum finish_epoch { |
59 | FE_STILL_LIVE, | |
60 | FE_DESTROYED, | |
61 | FE_RECYCLED, | |
62 | }; | |
63 | ||
6038178e | 64 | static int drbd_do_features(struct drbd_tconn *tconn); |
13e6037d | 65 | static int drbd_do_auth(struct drbd_tconn *tconn); |
c141ebda | 66 | static int drbd_disconnected(struct drbd_conf *mdev); |
b411b363 | 67 | |
1e9dd291 | 68 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_tconn *, struct drbd_epoch *, enum epoch_event); |
99920dc5 | 69 | static int e_end_block(struct drbd_work *, int); |
b411b363 | 70 | |
b411b363 PR |
71 | |
72 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
73 | ||
45bb912b LE |
74 | /* |
75 | * some helper functions to deal with single linked page lists, | |
76 | * page->private being our "next" pointer. | |
77 | */ | |
78 | ||
79 | /* If at least n pages are linked at head, get n pages off. | |
80 | * Otherwise, don't modify head, and return NULL. | |
81 | * Locking is the responsibility of the caller. | |
82 | */ | |
83 | static struct page *page_chain_del(struct page **head, int n) | |
84 | { | |
85 | struct page *page; | |
86 | struct page *tmp; | |
87 | ||
88 | BUG_ON(!n); | |
89 | BUG_ON(!head); | |
90 | ||
91 | page = *head; | |
23ce4227 PR |
92 | |
93 | if (!page) | |
94 | return NULL; | |
95 | ||
45bb912b LE |
96 | while (page) { |
97 | tmp = page_chain_next(page); | |
98 | if (--n == 0) | |
99 | break; /* found sufficient pages */ | |
100 | if (tmp == NULL) | |
101 | /* insufficient pages, don't use any of them. */ | |
102 | return NULL; | |
103 | page = tmp; | |
104 | } | |
105 | ||
106 | /* add end of list marker for the returned list */ | |
107 | set_page_private(page, 0); | |
108 | /* actual return value, and adjustment of head */ | |
109 | page = *head; | |
110 | *head = tmp; | |
111 | return page; | |
112 | } | |
113 | ||
114 | /* may be used outside of locks to find the tail of a (usually short) | |
115 | * "private" page chain, before adding it back to a global chain head | |
116 | * with page_chain_add() under a spinlock. */ | |
117 | static struct page *page_chain_tail(struct page *page, int *len) | |
118 | { | |
119 | struct page *tmp; | |
120 | int i = 1; | |
121 | while ((tmp = page_chain_next(page))) | |
122 | ++i, page = tmp; | |
123 | if (len) | |
124 | *len = i; | |
125 | return page; | |
126 | } | |
127 | ||
128 | static int page_chain_free(struct page *page) | |
129 | { | |
130 | struct page *tmp; | |
131 | int i = 0; | |
132 | page_chain_for_each_safe(page, tmp) { | |
133 | put_page(page); | |
134 | ++i; | |
135 | } | |
136 | return i; | |
137 | } | |
138 | ||
139 | static void page_chain_add(struct page **head, | |
140 | struct page *chain_first, struct page *chain_last) | |
141 | { | |
142 | #if 1 | |
143 | struct page *tmp; | |
144 | tmp = page_chain_tail(chain_first, NULL); | |
145 | BUG_ON(tmp != chain_last); | |
146 | #endif | |
147 | ||
148 | /* add chain to head */ | |
149 | set_page_private(chain_last, (unsigned long)*head); | |
150 | *head = chain_first; | |
151 | } | |
152 | ||
18c2d522 AG |
153 | static struct page *__drbd_alloc_pages(struct drbd_conf *mdev, |
154 | unsigned int number) | |
b411b363 PR |
155 | { |
156 | struct page *page = NULL; | |
45bb912b | 157 | struct page *tmp = NULL; |
18c2d522 | 158 | unsigned int i = 0; |
b411b363 PR |
159 | |
160 | /* Yes, testing drbd_pp_vacant outside the lock is racy. | |
161 | * So what. It saves a spin_lock. */ | |
45bb912b | 162 | if (drbd_pp_vacant >= number) { |
b411b363 | 163 | spin_lock(&drbd_pp_lock); |
45bb912b LE |
164 | page = page_chain_del(&drbd_pp_pool, number); |
165 | if (page) | |
166 | drbd_pp_vacant -= number; | |
b411b363 | 167 | spin_unlock(&drbd_pp_lock); |
45bb912b LE |
168 | if (page) |
169 | return page; | |
b411b363 | 170 | } |
45bb912b | 171 | |
b411b363 PR |
172 | /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD |
173 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
174 | * which in turn might block on the other node at this very place. */ | |
45bb912b LE |
175 | for (i = 0; i < number; i++) { |
176 | tmp = alloc_page(GFP_TRY); | |
177 | if (!tmp) | |
178 | break; | |
179 | set_page_private(tmp, (unsigned long)page); | |
180 | page = tmp; | |
181 | } | |
182 | ||
183 | if (i == number) | |
184 | return page; | |
185 | ||
186 | /* Not enough pages immediately available this time. | |
c37c8ecf | 187 | * No need to jump around here, drbd_alloc_pages will retry this |
45bb912b LE |
188 | * function "soon". */ |
189 | if (page) { | |
190 | tmp = page_chain_tail(page, NULL); | |
191 | spin_lock(&drbd_pp_lock); | |
192 | page_chain_add(&drbd_pp_pool, page, tmp); | |
193 | drbd_pp_vacant += i; | |
194 | spin_unlock(&drbd_pp_lock); | |
195 | } | |
196 | return NULL; | |
b411b363 PR |
197 | } |
198 | ||
a990be46 AG |
199 | static void reclaim_finished_net_peer_reqs(struct drbd_conf *mdev, |
200 | struct list_head *to_be_freed) | |
b411b363 | 201 | { |
db830c46 | 202 | struct drbd_peer_request *peer_req; |
b411b363 PR |
203 | struct list_head *le, *tle; |
204 | ||
205 | /* The EEs are always appended to the end of the list. Since | |
206 | they are sent in order over the wire, they have to finish | |
207 | in order. As soon as we see the first not finished we can | |
208 | stop to examine the list... */ | |
209 | ||
210 | list_for_each_safe(le, tle, &mdev->net_ee) { | |
db830c46 | 211 | peer_req = list_entry(le, struct drbd_peer_request, w.list); |
045417f7 | 212 | if (drbd_peer_req_has_active_page(peer_req)) |
b411b363 PR |
213 | break; |
214 | list_move(le, to_be_freed); | |
215 | } | |
216 | } | |
217 | ||
218 | static void drbd_kick_lo_and_reclaim_net(struct drbd_conf *mdev) | |
219 | { | |
220 | LIST_HEAD(reclaimed); | |
db830c46 | 221 | struct drbd_peer_request *peer_req, *t; |
b411b363 | 222 | |
87eeee41 | 223 | spin_lock_irq(&mdev->tconn->req_lock); |
a990be46 | 224 | reclaim_finished_net_peer_reqs(mdev, &reclaimed); |
87eeee41 | 225 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 226 | |
db830c46 | 227 | list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) |
3967deb1 | 228 | drbd_free_net_peer_req(mdev, peer_req); |
b411b363 PR |
229 | } |
230 | ||
231 | /** | |
c37c8ecf | 232 | * drbd_alloc_pages() - Returns @number pages, retries forever (or until signalled) |
b411b363 | 233 | * @mdev: DRBD device. |
45bb912b LE |
234 | * @number: number of pages requested |
235 | * @retry: whether to retry, if not enough pages are available right now | |
236 | * | |
237 | * Tries to allocate number pages, first from our own page pool, then from | |
238 | * the kernel, unless this allocation would exceed the max_buffers setting. | |
239 | * Possibly retry until DRBD frees sufficient pages somewhere else. | |
b411b363 | 240 | * |
45bb912b | 241 | * Returns a page chain linked via page->private. |
b411b363 | 242 | */ |
c37c8ecf AG |
243 | struct page *drbd_alloc_pages(struct drbd_conf *mdev, unsigned int number, |
244 | bool retry) | |
b411b363 PR |
245 | { |
246 | struct page *page = NULL; | |
44ed167d | 247 | struct net_conf *nc; |
b411b363 | 248 | DEFINE_WAIT(wait); |
44ed167d | 249 | int mxb; |
b411b363 | 250 | |
45bb912b LE |
251 | /* Yes, we may run up to @number over max_buffers. If we |
252 | * follow it strictly, the admin will get it wrong anyways. */ | |
44ed167d PR |
253 | rcu_read_lock(); |
254 | nc = rcu_dereference(mdev->tconn->net_conf); | |
255 | mxb = nc ? nc->max_buffers : 1000000; | |
256 | rcu_read_unlock(); | |
257 | ||
258 | if (atomic_read(&mdev->pp_in_use) < mxb) | |
18c2d522 | 259 | page = __drbd_alloc_pages(mdev, number); |
b411b363 | 260 | |
45bb912b | 261 | while (page == NULL) { |
b411b363 PR |
262 | prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE); |
263 | ||
264 | drbd_kick_lo_and_reclaim_net(mdev); | |
265 | ||
44ed167d | 266 | if (atomic_read(&mdev->pp_in_use) < mxb) { |
18c2d522 | 267 | page = __drbd_alloc_pages(mdev, number); |
b411b363 PR |
268 | if (page) |
269 | break; | |
270 | } | |
271 | ||
272 | if (!retry) | |
273 | break; | |
274 | ||
275 | if (signal_pending(current)) { | |
c37c8ecf | 276 | dev_warn(DEV, "drbd_alloc_pages interrupted!\n"); |
b411b363 PR |
277 | break; |
278 | } | |
279 | ||
280 | schedule(); | |
281 | } | |
282 | finish_wait(&drbd_pp_wait, &wait); | |
283 | ||
45bb912b LE |
284 | if (page) |
285 | atomic_add(number, &mdev->pp_in_use); | |
b411b363 PR |
286 | return page; |
287 | } | |
288 | ||
c37c8ecf | 289 | /* Must not be used from irq, as that may deadlock: see drbd_alloc_pages. |
87eeee41 | 290 | * Is also used from inside an other spin_lock_irq(&mdev->tconn->req_lock); |
45bb912b LE |
291 | * Either links the page chain back to the global pool, |
292 | * or returns all pages to the system. */ | |
5cc287e0 | 293 | static void drbd_free_pages(struct drbd_conf *mdev, struct page *page, int is_net) |
b411b363 | 294 | { |
435f0740 | 295 | atomic_t *a = is_net ? &mdev->pp_in_use_by_net : &mdev->pp_in_use; |
b411b363 | 296 | int i; |
435f0740 | 297 | |
a73ff323 LE |
298 | if (page == NULL) |
299 | return; | |
300 | ||
81a5d60e | 301 | if (drbd_pp_vacant > (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count) |
45bb912b LE |
302 | i = page_chain_free(page); |
303 | else { | |
304 | struct page *tmp; | |
305 | tmp = page_chain_tail(page, &i); | |
306 | spin_lock(&drbd_pp_lock); | |
307 | page_chain_add(&drbd_pp_pool, page, tmp); | |
308 | drbd_pp_vacant += i; | |
309 | spin_unlock(&drbd_pp_lock); | |
b411b363 | 310 | } |
435f0740 | 311 | i = atomic_sub_return(i, a); |
45bb912b | 312 | if (i < 0) |
435f0740 LE |
313 | dev_warn(DEV, "ASSERTION FAILED: %s: %d < 0\n", |
314 | is_net ? "pp_in_use_by_net" : "pp_in_use", i); | |
b411b363 PR |
315 | wake_up(&drbd_pp_wait); |
316 | } | |
317 | ||
318 | /* | |
319 | You need to hold the req_lock: | |
320 | _drbd_wait_ee_list_empty() | |
321 | ||
322 | You must not have the req_lock: | |
3967deb1 | 323 | drbd_free_peer_req() |
0db55363 | 324 | drbd_alloc_peer_req() |
7721f567 | 325 | drbd_free_peer_reqs() |
b411b363 | 326 | drbd_ee_fix_bhs() |
a990be46 | 327 | drbd_finish_peer_reqs() |
b411b363 PR |
328 | drbd_clear_done_ee() |
329 | drbd_wait_ee_list_empty() | |
330 | */ | |
331 | ||
f6ffca9f | 332 | struct drbd_peer_request * |
0db55363 AG |
333 | drbd_alloc_peer_req(struct drbd_conf *mdev, u64 id, sector_t sector, |
334 | unsigned int data_size, gfp_t gfp_mask) __must_hold(local) | |
b411b363 | 335 | { |
db830c46 | 336 | struct drbd_peer_request *peer_req; |
a73ff323 | 337 | struct page *page = NULL; |
45bb912b | 338 | unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b411b363 | 339 | |
0cf9d27e | 340 | if (drbd_insert_fault(mdev, DRBD_FAULT_AL_EE)) |
b411b363 PR |
341 | return NULL; |
342 | ||
db830c46 AG |
343 | peer_req = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM); |
344 | if (!peer_req) { | |
b411b363 | 345 | if (!(gfp_mask & __GFP_NOWARN)) |
0db55363 | 346 | dev_err(DEV, "%s: allocation failed\n", __func__); |
b411b363 PR |
347 | return NULL; |
348 | } | |
349 | ||
a73ff323 | 350 | if (data_size) { |
81a3537a | 351 | page = drbd_alloc_pages(mdev, nr_pages, (gfp_mask & __GFP_WAIT)); |
a73ff323 LE |
352 | if (!page) |
353 | goto fail; | |
354 | } | |
b411b363 | 355 | |
db830c46 AG |
356 | drbd_clear_interval(&peer_req->i); |
357 | peer_req->i.size = data_size; | |
358 | peer_req->i.sector = sector; | |
359 | peer_req->i.local = false; | |
360 | peer_req->i.waiting = false; | |
361 | ||
362 | peer_req->epoch = NULL; | |
a21e9298 | 363 | peer_req->w.mdev = mdev; |
db830c46 AG |
364 | peer_req->pages = page; |
365 | atomic_set(&peer_req->pending_bios, 0); | |
366 | peer_req->flags = 0; | |
9a8e7753 AG |
367 | /* |
368 | * The block_id is opaque to the receiver. It is not endianness | |
369 | * converted, and sent back to the sender unchanged. | |
370 | */ | |
db830c46 | 371 | peer_req->block_id = id; |
b411b363 | 372 | |
db830c46 | 373 | return peer_req; |
b411b363 | 374 | |
45bb912b | 375 | fail: |
db830c46 | 376 | mempool_free(peer_req, drbd_ee_mempool); |
b411b363 PR |
377 | return NULL; |
378 | } | |
379 | ||
3967deb1 | 380 | void __drbd_free_peer_req(struct drbd_conf *mdev, struct drbd_peer_request *peer_req, |
f6ffca9f | 381 | int is_net) |
b411b363 | 382 | { |
db830c46 AG |
383 | if (peer_req->flags & EE_HAS_DIGEST) |
384 | kfree(peer_req->digest); | |
5cc287e0 | 385 | drbd_free_pages(mdev, peer_req->pages, is_net); |
db830c46 AG |
386 | D_ASSERT(atomic_read(&peer_req->pending_bios) == 0); |
387 | D_ASSERT(drbd_interval_empty(&peer_req->i)); | |
388 | mempool_free(peer_req, drbd_ee_mempool); | |
b411b363 PR |
389 | } |
390 | ||
7721f567 | 391 | int drbd_free_peer_reqs(struct drbd_conf *mdev, struct list_head *list) |
b411b363 PR |
392 | { |
393 | LIST_HEAD(work_list); | |
db830c46 | 394 | struct drbd_peer_request *peer_req, *t; |
b411b363 | 395 | int count = 0; |
435f0740 | 396 | int is_net = list == &mdev->net_ee; |
b411b363 | 397 | |
87eeee41 | 398 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 399 | list_splice_init(list, &work_list); |
87eeee41 | 400 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 401 | |
db830c46 | 402 | list_for_each_entry_safe(peer_req, t, &work_list, w.list) { |
3967deb1 | 403 | __drbd_free_peer_req(mdev, peer_req, is_net); |
b411b363 PR |
404 | count++; |
405 | } | |
406 | return count; | |
407 | } | |
408 | ||
b411b363 | 409 | /* |
a990be46 | 410 | * See also comments in _req_mod(,BARRIER_ACKED) and receive_Barrier. |
b411b363 | 411 | */ |
a990be46 | 412 | static int drbd_finish_peer_reqs(struct drbd_conf *mdev) |
b411b363 PR |
413 | { |
414 | LIST_HEAD(work_list); | |
415 | LIST_HEAD(reclaimed); | |
db830c46 | 416 | struct drbd_peer_request *peer_req, *t; |
e2b3032b | 417 | int err = 0; |
b411b363 | 418 | |
87eeee41 | 419 | spin_lock_irq(&mdev->tconn->req_lock); |
a990be46 | 420 | reclaim_finished_net_peer_reqs(mdev, &reclaimed); |
b411b363 | 421 | list_splice_init(&mdev->done_ee, &work_list); |
87eeee41 | 422 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 423 | |
db830c46 | 424 | list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) |
3967deb1 | 425 | drbd_free_net_peer_req(mdev, peer_req); |
b411b363 PR |
426 | |
427 | /* possible callbacks here: | |
d4dabbe2 | 428 | * e_end_block, and e_end_resync_block, e_send_superseded. |
b411b363 PR |
429 | * all ignore the last argument. |
430 | */ | |
db830c46 | 431 | list_for_each_entry_safe(peer_req, t, &work_list, w.list) { |
e2b3032b AG |
432 | int err2; |
433 | ||
b411b363 | 434 | /* list_del not necessary, next/prev members not touched */ |
e2b3032b AG |
435 | err2 = peer_req->w.cb(&peer_req->w, !!err); |
436 | if (!err) | |
437 | err = err2; | |
3967deb1 | 438 | drbd_free_peer_req(mdev, peer_req); |
b411b363 PR |
439 | } |
440 | wake_up(&mdev->ee_wait); | |
441 | ||
e2b3032b | 442 | return err; |
b411b363 PR |
443 | } |
444 | ||
d4da1537 AG |
445 | static void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, |
446 | struct list_head *head) | |
b411b363 PR |
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); | |
87eeee41 | 454 | spin_unlock_irq(&mdev->tconn->req_lock); |
7eaceacc | 455 | io_schedule(); |
b411b363 | 456 | finish_wait(&mdev->ee_wait, &wait); |
87eeee41 | 457 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
458 | } |
459 | } | |
460 | ||
d4da1537 AG |
461 | static void drbd_wait_ee_list_empty(struct drbd_conf *mdev, |
462 | struct list_head *head) | |
b411b363 | 463 | { |
87eeee41 | 464 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 465 | _drbd_wait_ee_list_empty(mdev, head); |
87eeee41 | 466 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
467 | } |
468 | ||
dbd9eea0 | 469 | static int drbd_recv_short(struct socket *sock, void *buf, size_t size, int flags) |
b411b363 PR |
470 | { |
471 | mm_segment_t oldfs; | |
472 | struct kvec iov = { | |
473 | .iov_base = buf, | |
474 | .iov_len = size, | |
475 | }; | |
476 | struct msghdr msg = { | |
477 | .msg_iovlen = 1, | |
478 | .msg_iov = (struct iovec *)&iov, | |
479 | .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) | |
480 | }; | |
481 | int rv; | |
482 | ||
483 | oldfs = get_fs(); | |
484 | set_fs(KERNEL_DS); | |
485 | rv = sock_recvmsg(sock, &msg, size, msg.msg_flags); | |
486 | set_fs(oldfs); | |
487 | ||
488 | return rv; | |
489 | } | |
490 | ||
de0ff338 | 491 | static int drbd_recv(struct drbd_tconn *tconn, void *buf, size_t size) |
b411b363 | 492 | { |
b411b363 PR |
493 | int rv; |
494 | ||
1393b59f | 495 | rv = drbd_recv_short(tconn->data.socket, buf, size, 0); |
b411b363 | 496 | |
dbd0820c PR |
497 | if (rv < 0) { |
498 | if (rv == -ECONNRESET) | |
155522df | 499 | conn_info(tconn, "sock was reset by peer\n"); |
dbd0820c | 500 | else if (rv != -ERESTARTSYS) |
155522df | 501 | conn_err(tconn, "sock_recvmsg returned %d\n", rv); |
dbd0820c | 502 | } else if (rv == 0) { |
b66623e3 PR |
503 | if (test_bit(DISCONNECT_SENT, &tconn->flags)) { |
504 | long t; | |
505 | rcu_read_lock(); | |
506 | t = rcu_dereference(tconn->net_conf)->ping_timeo * HZ/10; | |
507 | rcu_read_unlock(); | |
508 | ||
509 | t = wait_event_timeout(tconn->ping_wait, tconn->cstate < C_WF_REPORT_PARAMS, t); | |
510 | ||
599377ac PR |
511 | if (t) |
512 | goto out; | |
513 | } | |
b66623e3 | 514 | conn_info(tconn, "sock was shut down by peer\n"); |
599377ac PR |
515 | } |
516 | ||
b411b363 | 517 | if (rv != size) |
bbeb641c | 518 | conn_request_state(tconn, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 | 519 | |
599377ac | 520 | out: |
b411b363 PR |
521 | return rv; |
522 | } | |
523 | ||
c6967746 AG |
524 | static int drbd_recv_all(struct drbd_tconn *tconn, void *buf, size_t size) |
525 | { | |
526 | int err; | |
527 | ||
528 | err = drbd_recv(tconn, buf, size); | |
529 | if (err != size) { | |
530 | if (err >= 0) | |
531 | err = -EIO; | |
532 | } else | |
533 | err = 0; | |
534 | return err; | |
535 | } | |
536 | ||
a5c31904 AG |
537 | static int drbd_recv_all_warn(struct drbd_tconn *tconn, void *buf, size_t size) |
538 | { | |
539 | int err; | |
540 | ||
541 | err = drbd_recv_all(tconn, buf, size); | |
542 | if (err && !signal_pending(current)) | |
543 | conn_warn(tconn, "short read (expected size %d)\n", (int)size); | |
544 | return err; | |
545 | } | |
546 | ||
5dbf1673 LE |
547 | /* quoting tcp(7): |
548 | * On individual connections, the socket buffer size must be set prior to the | |
549 | * listen(2) or connect(2) calls in order to have it take effect. | |
550 | * This is our wrapper to do so. | |
551 | */ | |
552 | static void drbd_setbufsize(struct socket *sock, unsigned int snd, | |
553 | unsigned int rcv) | |
554 | { | |
555 | /* open coded SO_SNDBUF, SO_RCVBUF */ | |
556 | if (snd) { | |
557 | sock->sk->sk_sndbuf = snd; | |
558 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
559 | } | |
560 | if (rcv) { | |
561 | sock->sk->sk_rcvbuf = rcv; | |
562 | sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
563 | } | |
564 | } | |
565 | ||
eac3e990 | 566 | static struct socket *drbd_try_connect(struct drbd_tconn *tconn) |
b411b363 PR |
567 | { |
568 | const char *what; | |
569 | struct socket *sock; | |
570 | struct sockaddr_in6 src_in6; | |
44ed167d PR |
571 | struct sockaddr_in6 peer_in6; |
572 | struct net_conf *nc; | |
573 | int err, peer_addr_len, my_addr_len; | |
69ef82de | 574 | int sndbuf_size, rcvbuf_size, connect_int; |
b411b363 PR |
575 | int disconnect_on_error = 1; |
576 | ||
44ed167d PR |
577 | rcu_read_lock(); |
578 | nc = rcu_dereference(tconn->net_conf); | |
579 | if (!nc) { | |
580 | rcu_read_unlock(); | |
b411b363 | 581 | return NULL; |
44ed167d | 582 | } |
44ed167d PR |
583 | sndbuf_size = nc->sndbuf_size; |
584 | rcvbuf_size = nc->rcvbuf_size; | |
69ef82de | 585 | connect_int = nc->connect_int; |
089c075d | 586 | rcu_read_unlock(); |
44ed167d | 587 | |
089c075d AG |
588 | my_addr_len = min_t(int, tconn->my_addr_len, sizeof(src_in6)); |
589 | memcpy(&src_in6, &tconn->my_addr, my_addr_len); | |
44ed167d | 590 | |
089c075d | 591 | if (((struct sockaddr *)&tconn->my_addr)->sa_family == AF_INET6) |
44ed167d PR |
592 | src_in6.sin6_port = 0; |
593 | else | |
594 | ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ | |
595 | ||
089c075d AG |
596 | peer_addr_len = min_t(int, tconn->peer_addr_len, sizeof(src_in6)); |
597 | memcpy(&peer_in6, &tconn->peer_addr, peer_addr_len); | |
b411b363 PR |
598 | |
599 | what = "sock_create_kern"; | |
44ed167d PR |
600 | err = sock_create_kern(((struct sockaddr *)&src_in6)->sa_family, |
601 | SOCK_STREAM, IPPROTO_TCP, &sock); | |
b411b363 PR |
602 | if (err < 0) { |
603 | sock = NULL; | |
604 | goto out; | |
605 | } | |
606 | ||
607 | sock->sk->sk_rcvtimeo = | |
69ef82de | 608 | sock->sk->sk_sndtimeo = connect_int * HZ; |
44ed167d | 609 | drbd_setbufsize(sock, sndbuf_size, rcvbuf_size); |
b411b363 PR |
610 | |
611 | /* explicitly bind to the configured IP as source IP | |
612 | * for the outgoing connections. | |
613 | * This is needed for multihomed hosts and to be | |
614 | * able to use lo: interfaces for drbd. | |
615 | * Make sure to use 0 as port number, so linux selects | |
616 | * a free one dynamically. | |
617 | */ | |
b411b363 | 618 | what = "bind before connect"; |
44ed167d | 619 | err = sock->ops->bind(sock, (struct sockaddr *) &src_in6, my_addr_len); |
b411b363 PR |
620 | if (err < 0) |
621 | goto out; | |
622 | ||
623 | /* connect may fail, peer not yet available. | |
624 | * stay C_WF_CONNECTION, don't go Disconnecting! */ | |
625 | disconnect_on_error = 0; | |
626 | what = "connect"; | |
44ed167d | 627 | err = sock->ops->connect(sock, (struct sockaddr *) &peer_in6, peer_addr_len, 0); |
b411b363 PR |
628 | |
629 | out: | |
630 | if (err < 0) { | |
631 | if (sock) { | |
632 | sock_release(sock); | |
633 | sock = NULL; | |
634 | } | |
635 | switch (-err) { | |
636 | /* timeout, busy, signal pending */ | |
637 | case ETIMEDOUT: case EAGAIN: case EINPROGRESS: | |
638 | case EINTR: case ERESTARTSYS: | |
639 | /* peer not (yet) available, network problem */ | |
640 | case ECONNREFUSED: case ENETUNREACH: | |
641 | case EHOSTDOWN: case EHOSTUNREACH: | |
642 | disconnect_on_error = 0; | |
643 | break; | |
644 | default: | |
eac3e990 | 645 | conn_err(tconn, "%s failed, err = %d\n", what, err); |
b411b363 PR |
646 | } |
647 | if (disconnect_on_error) | |
bbeb641c | 648 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 649 | } |
44ed167d | 650 | |
b411b363 PR |
651 | return sock; |
652 | } | |
653 | ||
7a426fd8 PR |
654 | struct accept_wait_data { |
655 | struct drbd_tconn *tconn; | |
656 | struct socket *s_listen; | |
657 | struct completion door_bell; | |
658 | void (*original_sk_state_change)(struct sock *sk); | |
659 | ||
660 | }; | |
661 | ||
715306f6 | 662 | static void drbd_incoming_connection(struct sock *sk) |
7a426fd8 PR |
663 | { |
664 | struct accept_wait_data *ad = sk->sk_user_data; | |
715306f6 | 665 | void (*state_change)(struct sock *sk); |
7a426fd8 | 666 | |
715306f6 AG |
667 | state_change = ad->original_sk_state_change; |
668 | if (sk->sk_state == TCP_ESTABLISHED) | |
669 | complete(&ad->door_bell); | |
670 | state_change(sk); | |
7a426fd8 PR |
671 | } |
672 | ||
673 | static int prepare_listen_socket(struct drbd_tconn *tconn, struct accept_wait_data *ad) | |
b411b363 | 674 | { |
1f3e509b | 675 | int err, sndbuf_size, rcvbuf_size, my_addr_len; |
44ed167d | 676 | struct sockaddr_in6 my_addr; |
1f3e509b | 677 | struct socket *s_listen; |
44ed167d | 678 | struct net_conf *nc; |
b411b363 PR |
679 | const char *what; |
680 | ||
44ed167d PR |
681 | rcu_read_lock(); |
682 | nc = rcu_dereference(tconn->net_conf); | |
683 | if (!nc) { | |
684 | rcu_read_unlock(); | |
7a426fd8 | 685 | return -EIO; |
44ed167d | 686 | } |
44ed167d PR |
687 | sndbuf_size = nc->sndbuf_size; |
688 | rcvbuf_size = nc->rcvbuf_size; | |
44ed167d | 689 | rcu_read_unlock(); |
b411b363 | 690 | |
089c075d AG |
691 | my_addr_len = min_t(int, tconn->my_addr_len, sizeof(struct sockaddr_in6)); |
692 | memcpy(&my_addr, &tconn->my_addr, my_addr_len); | |
b411b363 PR |
693 | |
694 | what = "sock_create_kern"; | |
44ed167d | 695 | err = sock_create_kern(((struct sockaddr *)&my_addr)->sa_family, |
1f3e509b | 696 | SOCK_STREAM, IPPROTO_TCP, &s_listen); |
b411b363 PR |
697 | if (err) { |
698 | s_listen = NULL; | |
699 | goto out; | |
700 | } | |
701 | ||
98683650 | 702 | s_listen->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ |
44ed167d | 703 | drbd_setbufsize(s_listen, sndbuf_size, rcvbuf_size); |
b411b363 PR |
704 | |
705 | what = "bind before listen"; | |
44ed167d | 706 | err = s_listen->ops->bind(s_listen, (struct sockaddr *)&my_addr, my_addr_len); |
b411b363 PR |
707 | if (err < 0) |
708 | goto out; | |
709 | ||
7a426fd8 PR |
710 | ad->s_listen = s_listen; |
711 | write_lock_bh(&s_listen->sk->sk_callback_lock); | |
712 | ad->original_sk_state_change = s_listen->sk->sk_state_change; | |
715306f6 | 713 | s_listen->sk->sk_state_change = drbd_incoming_connection; |
7a426fd8 PR |
714 | s_listen->sk->sk_user_data = ad; |
715 | write_unlock_bh(&s_listen->sk->sk_callback_lock); | |
b411b363 | 716 | |
2820fd39 PR |
717 | what = "listen"; |
718 | err = s_listen->ops->listen(s_listen, 5); | |
719 | if (err < 0) | |
720 | goto out; | |
721 | ||
7a426fd8 | 722 | return 0; |
b411b363 PR |
723 | out: |
724 | if (s_listen) | |
725 | sock_release(s_listen); | |
726 | if (err < 0) { | |
727 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
1f3e509b PR |
728 | conn_err(tconn, "%s failed, err = %d\n", what, err); |
729 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); | |
b411b363 PR |
730 | } |
731 | } | |
b411b363 | 732 | |
7a426fd8 | 733 | return -EIO; |
b411b363 PR |
734 | } |
735 | ||
715306f6 | 736 | static void unregister_state_change(struct sock *sk, struct accept_wait_data *ad) |
b411b363 | 737 | { |
715306f6 AG |
738 | write_lock_bh(&sk->sk_callback_lock); |
739 | sk->sk_state_change = ad->original_sk_state_change; | |
740 | sk->sk_user_data = NULL; | |
741 | write_unlock_bh(&sk->sk_callback_lock); | |
b411b363 PR |
742 | } |
743 | ||
7a426fd8 | 744 | static struct socket *drbd_wait_for_connect(struct drbd_tconn *tconn, struct accept_wait_data *ad) |
b411b363 | 745 | { |
1f3e509b PR |
746 | int timeo, connect_int, err = 0; |
747 | struct socket *s_estab = NULL; | |
1f3e509b PR |
748 | struct net_conf *nc; |
749 | ||
750 | rcu_read_lock(); | |
751 | nc = rcu_dereference(tconn->net_conf); | |
752 | if (!nc) { | |
753 | rcu_read_unlock(); | |
754 | return NULL; | |
755 | } | |
756 | connect_int = nc->connect_int; | |
757 | rcu_read_unlock(); | |
758 | ||
759 | timeo = connect_int * HZ; | |
760 | timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */ | |
761 | ||
7a426fd8 PR |
762 | err = wait_for_completion_interruptible_timeout(&ad->door_bell, timeo); |
763 | if (err <= 0) | |
764 | return NULL; | |
b411b363 | 765 | |
7a426fd8 | 766 | err = kernel_accept(ad->s_listen, &s_estab, 0); |
b411b363 PR |
767 | if (err < 0) { |
768 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
1f3e509b | 769 | conn_err(tconn, "accept failed, err = %d\n", err); |
bbeb641c | 770 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
771 | } |
772 | } | |
b411b363 | 773 | |
715306f6 AG |
774 | if (s_estab) |
775 | unregister_state_change(s_estab->sk, ad); | |
b411b363 | 776 | |
b411b363 PR |
777 | return s_estab; |
778 | } | |
b411b363 | 779 | |
e658983a | 780 | static int decode_header(struct drbd_tconn *, void *, struct packet_info *); |
b411b363 | 781 | |
9f5bdc33 AG |
782 | static int send_first_packet(struct drbd_tconn *tconn, struct drbd_socket *sock, |
783 | enum drbd_packet cmd) | |
784 | { | |
785 | if (!conn_prepare_command(tconn, sock)) | |
786 | return -EIO; | |
e658983a | 787 | return conn_send_command(tconn, sock, cmd, 0, NULL, 0); |
b411b363 PR |
788 | } |
789 | ||
9f5bdc33 | 790 | static int receive_first_packet(struct drbd_tconn *tconn, struct socket *sock) |
b411b363 | 791 | { |
9f5bdc33 AG |
792 | unsigned int header_size = drbd_header_size(tconn); |
793 | struct packet_info pi; | |
794 | int err; | |
b411b363 | 795 | |
9f5bdc33 AG |
796 | err = drbd_recv_short(sock, tconn->data.rbuf, header_size, 0); |
797 | if (err != header_size) { | |
798 | if (err >= 0) | |
799 | err = -EIO; | |
800 | return err; | |
801 | } | |
802 | err = decode_header(tconn, tconn->data.rbuf, &pi); | |
803 | if (err) | |
804 | return err; | |
805 | return pi.cmd; | |
b411b363 PR |
806 | } |
807 | ||
808 | /** | |
809 | * drbd_socket_okay() - Free the socket if its connection is not okay | |
b411b363 PR |
810 | * @sock: pointer to the pointer to the socket. |
811 | */ | |
dbd9eea0 | 812 | static int drbd_socket_okay(struct socket **sock) |
b411b363 PR |
813 | { |
814 | int rr; | |
815 | char tb[4]; | |
816 | ||
817 | if (!*sock) | |
81e84650 | 818 | return false; |
b411b363 | 819 | |
dbd9eea0 | 820 | rr = drbd_recv_short(*sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); |
b411b363 PR |
821 | |
822 | if (rr > 0 || rr == -EAGAIN) { | |
81e84650 | 823 | return true; |
b411b363 PR |
824 | } else { |
825 | sock_release(*sock); | |
826 | *sock = NULL; | |
81e84650 | 827 | return false; |
b411b363 PR |
828 | } |
829 | } | |
2325eb66 PR |
830 | /* Gets called if a connection is established, or if a new minor gets created |
831 | in a connection */ | |
c141ebda | 832 | int drbd_connected(struct drbd_conf *mdev) |
907599e0 | 833 | { |
0829f5ed | 834 | int err; |
907599e0 PR |
835 | |
836 | atomic_set(&mdev->packet_seq, 0); | |
837 | mdev->peer_seq = 0; | |
838 | ||
8410da8f PR |
839 | mdev->state_mutex = mdev->tconn->agreed_pro_version < 100 ? |
840 | &mdev->tconn->cstate_mutex : | |
841 | &mdev->own_state_mutex; | |
842 | ||
0829f5ed AG |
843 | err = drbd_send_sync_param(mdev); |
844 | if (!err) | |
845 | err = drbd_send_sizes(mdev, 0, 0); | |
846 | if (!err) | |
847 | err = drbd_send_uuids(mdev); | |
848 | if (!err) | |
43de7c85 | 849 | err = drbd_send_current_state(mdev); |
907599e0 PR |
850 | clear_bit(USE_DEGR_WFC_T, &mdev->flags); |
851 | clear_bit(RESIZE_PENDING, &mdev->flags); | |
8b924f1d | 852 | mod_timer(&mdev->request_timer, jiffies + HZ); /* just start it here. */ |
0829f5ed | 853 | return err; |
907599e0 | 854 | } |
b411b363 PR |
855 | |
856 | /* | |
857 | * return values: | |
858 | * 1 yes, we have a valid connection | |
859 | * 0 oops, did not work out, please try again | |
860 | * -1 peer talks different language, | |
861 | * no point in trying again, please go standalone. | |
862 | * -2 We do not have a network config... | |
863 | */ | |
81fa2e67 | 864 | static int conn_connect(struct drbd_tconn *tconn) |
b411b363 | 865 | { |
7da35862 | 866 | struct drbd_socket sock, msock; |
c141ebda | 867 | struct drbd_conf *mdev; |
44ed167d | 868 | struct net_conf *nc; |
92f14951 | 869 | int vnr, timeout, h, ok; |
08b165ba | 870 | bool discard_my_data; |
197296ff | 871 | enum drbd_state_rv rv; |
7a426fd8 PR |
872 | struct accept_wait_data ad = { |
873 | .tconn = tconn, | |
874 | .door_bell = COMPLETION_INITIALIZER_ONSTACK(ad.door_bell), | |
875 | }; | |
b411b363 | 876 | |
b66623e3 | 877 | clear_bit(DISCONNECT_SENT, &tconn->flags); |
bbeb641c | 878 | if (conn_request_state(tconn, NS(conn, C_WF_CONNECTION), CS_VERBOSE) < SS_SUCCESS) |
b411b363 PR |
879 | return -2; |
880 | ||
7da35862 PR |
881 | mutex_init(&sock.mutex); |
882 | sock.sbuf = tconn->data.sbuf; | |
883 | sock.rbuf = tconn->data.rbuf; | |
884 | sock.socket = NULL; | |
885 | mutex_init(&msock.mutex); | |
886 | msock.sbuf = tconn->meta.sbuf; | |
887 | msock.rbuf = tconn->meta.rbuf; | |
888 | msock.socket = NULL; | |
889 | ||
0916e0e3 AG |
890 | /* Assume that the peer only understands protocol 80 until we know better. */ |
891 | tconn->agreed_pro_version = 80; | |
b411b363 | 892 | |
7a426fd8 PR |
893 | if (prepare_listen_socket(tconn, &ad)) |
894 | return 0; | |
b411b363 PR |
895 | |
896 | do { | |
2bf89621 | 897 | struct socket *s; |
b411b363 | 898 | |
92f14951 | 899 | s = drbd_try_connect(tconn); |
b411b363 | 900 | if (s) { |
7da35862 PR |
901 | if (!sock.socket) { |
902 | sock.socket = s; | |
903 | send_first_packet(tconn, &sock, P_INITIAL_DATA); | |
904 | } else if (!msock.socket) { | |
427c0434 | 905 | clear_bit(RESOLVE_CONFLICTS, &tconn->flags); |
7da35862 PR |
906 | msock.socket = s; |
907 | send_first_packet(tconn, &msock, P_INITIAL_META); | |
b411b363 | 908 | } else { |
81fa2e67 | 909 | conn_err(tconn, "Logic error in conn_connect()\n"); |
b411b363 PR |
910 | goto out_release_sockets; |
911 | } | |
912 | } | |
913 | ||
7da35862 PR |
914 | if (sock.socket && msock.socket) { |
915 | rcu_read_lock(); | |
916 | nc = rcu_dereference(tconn->net_conf); | |
917 | timeout = nc->ping_timeo * HZ / 10; | |
918 | rcu_read_unlock(); | |
919 | schedule_timeout_interruptible(timeout); | |
920 | ok = drbd_socket_okay(&sock.socket); | |
921 | ok = drbd_socket_okay(&msock.socket) && ok; | |
b411b363 PR |
922 | if (ok) |
923 | break; | |
924 | } | |
925 | ||
926 | retry: | |
7a426fd8 | 927 | s = drbd_wait_for_connect(tconn, &ad); |
b411b363 | 928 | if (s) { |
92f14951 | 929 | int fp = receive_first_packet(tconn, s); |
7da35862 PR |
930 | drbd_socket_okay(&sock.socket); |
931 | drbd_socket_okay(&msock.socket); | |
92f14951 | 932 | switch (fp) { |
e5d6f33a | 933 | case P_INITIAL_DATA: |
7da35862 | 934 | if (sock.socket) { |
907599e0 | 935 | conn_warn(tconn, "initial packet S crossed\n"); |
7da35862 | 936 | sock_release(sock.socket); |
80c6eed4 PR |
937 | sock.socket = s; |
938 | goto randomize; | |
b411b363 | 939 | } |
7da35862 | 940 | sock.socket = s; |
b411b363 | 941 | break; |
e5d6f33a | 942 | case P_INITIAL_META: |
427c0434 | 943 | set_bit(RESOLVE_CONFLICTS, &tconn->flags); |
7da35862 | 944 | if (msock.socket) { |
907599e0 | 945 | conn_warn(tconn, "initial packet M crossed\n"); |
7da35862 | 946 | sock_release(msock.socket); |
80c6eed4 PR |
947 | msock.socket = s; |
948 | goto randomize; | |
b411b363 | 949 | } |
7da35862 | 950 | msock.socket = s; |
b411b363 PR |
951 | break; |
952 | default: | |
907599e0 | 953 | conn_warn(tconn, "Error receiving initial packet\n"); |
b411b363 | 954 | sock_release(s); |
80c6eed4 | 955 | randomize: |
b411b363 PR |
956 | if (random32() & 1) |
957 | goto retry; | |
958 | } | |
959 | } | |
960 | ||
bbeb641c | 961 | if (tconn->cstate <= C_DISCONNECTING) |
b411b363 PR |
962 | goto out_release_sockets; |
963 | if (signal_pending(current)) { | |
964 | flush_signals(current); | |
965 | smp_rmb(); | |
907599e0 | 966 | if (get_t_state(&tconn->receiver) == EXITING) |
b411b363 PR |
967 | goto out_release_sockets; |
968 | } | |
969 | ||
b666dbf8 PR |
970 | ok = drbd_socket_okay(&sock.socket); |
971 | ok = drbd_socket_okay(&msock.socket) && ok; | |
972 | } while (!ok); | |
b411b363 | 973 | |
7a426fd8 PR |
974 | if (ad.s_listen) |
975 | sock_release(ad.s_listen); | |
b411b363 | 976 | |
98683650 PR |
977 | sock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ |
978 | msock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ | |
b411b363 | 979 | |
7da35862 PR |
980 | sock.socket->sk->sk_allocation = GFP_NOIO; |
981 | msock.socket->sk->sk_allocation = GFP_NOIO; | |
b411b363 | 982 | |
7da35862 PR |
983 | sock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; |
984 | msock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE; | |
b411b363 | 985 | |
b411b363 | 986 | /* NOT YET ... |
7da35862 PR |
987 | * sock.socket->sk->sk_sndtimeo = tconn->net_conf->timeout*HZ/10; |
988 | * sock.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; | |
6038178e | 989 | * first set it to the P_CONNECTION_FEATURES timeout, |
b411b363 | 990 | * which we set to 4x the configured ping_timeout. */ |
44ed167d PR |
991 | rcu_read_lock(); |
992 | nc = rcu_dereference(tconn->net_conf); | |
993 | ||
7da35862 PR |
994 | sock.socket->sk->sk_sndtimeo = |
995 | sock.socket->sk->sk_rcvtimeo = nc->ping_timeo*4*HZ/10; | |
b411b363 | 996 | |
7da35862 | 997 | msock.socket->sk->sk_rcvtimeo = nc->ping_int*HZ; |
44ed167d | 998 | timeout = nc->timeout * HZ / 10; |
08b165ba | 999 | discard_my_data = nc->discard_my_data; |
44ed167d | 1000 | rcu_read_unlock(); |
b411b363 | 1001 | |
7da35862 | 1002 | msock.socket->sk->sk_sndtimeo = timeout; |
b411b363 PR |
1003 | |
1004 | /* we don't want delays. | |
25985edc | 1005 | * we use TCP_CORK where appropriate, though */ |
7da35862 PR |
1006 | drbd_tcp_nodelay(sock.socket); |
1007 | drbd_tcp_nodelay(msock.socket); | |
b411b363 | 1008 | |
7da35862 PR |
1009 | tconn->data.socket = sock.socket; |
1010 | tconn->meta.socket = msock.socket; | |
907599e0 | 1011 | tconn->last_received = jiffies; |
b411b363 | 1012 | |
6038178e | 1013 | h = drbd_do_features(tconn); |
b411b363 PR |
1014 | if (h <= 0) |
1015 | return h; | |
1016 | ||
907599e0 | 1017 | if (tconn->cram_hmac_tfm) { |
b411b363 | 1018 | /* drbd_request_state(mdev, NS(conn, WFAuth)); */ |
907599e0 | 1019 | switch (drbd_do_auth(tconn)) { |
b10d96cb | 1020 | case -1: |
907599e0 | 1021 | conn_err(tconn, "Authentication of peer failed\n"); |
b411b363 | 1022 | return -1; |
b10d96cb | 1023 | case 0: |
907599e0 | 1024 | conn_err(tconn, "Authentication of peer failed, trying again.\n"); |
b10d96cb | 1025 | return 0; |
b411b363 PR |
1026 | } |
1027 | } | |
1028 | ||
7da35862 PR |
1029 | tconn->data.socket->sk->sk_sndtimeo = timeout; |
1030 | tconn->data.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; | |
b411b363 | 1031 | |
387eb308 | 1032 | if (drbd_send_protocol(tconn) == -EOPNOTSUPP) |
7e2455c1 | 1033 | return -1; |
b411b363 | 1034 | |
a1096a6e PR |
1035 | set_bit(STATE_SENT, &tconn->flags); |
1036 | ||
c141ebda PR |
1037 | rcu_read_lock(); |
1038 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { | |
1039 | kref_get(&mdev->kref); | |
13c76aba PR |
1040 | /* Prevent a race between resync-handshake and |
1041 | * being promoted to Primary. | |
1042 | * | |
1043 | * Grab and release the state mutex, so we know that any current | |
1044 | * drbd_set_role() is finished, and any incoming drbd_set_role | |
1045 | * will see the STATE_SENT flag, and wait for it to be cleared. | |
1046 | */ | |
1047 | mutex_lock(mdev->state_mutex); | |
1048 | mutex_unlock(mdev->state_mutex); | |
1049 | ||
c141ebda | 1050 | rcu_read_unlock(); |
08b165ba PR |
1051 | |
1052 | if (discard_my_data) | |
1053 | set_bit(DISCARD_MY_DATA, &mdev->flags); | |
1054 | else | |
1055 | clear_bit(DISCARD_MY_DATA, &mdev->flags); | |
1056 | ||
c141ebda PR |
1057 | drbd_connected(mdev); |
1058 | kref_put(&mdev->kref, &drbd_minor_destroy); | |
1059 | rcu_read_lock(); | |
1060 | } | |
1061 | rcu_read_unlock(); | |
1062 | ||
a1096a6e | 1063 | rv = conn_request_state(tconn, NS(conn, C_WF_REPORT_PARAMS), CS_VERBOSE); |
ed635cb0 | 1064 | if (rv < SS_SUCCESS || tconn->cstate != C_WF_REPORT_PARAMS) { |
a1096a6e | 1065 | clear_bit(STATE_SENT, &tconn->flags); |
1e86ac48 | 1066 | return 0; |
a1096a6e | 1067 | } |
1e86ac48 | 1068 | |
823bd832 | 1069 | drbd_thread_start(&tconn->asender); |
b411b363 | 1070 | |
08b165ba PR |
1071 | mutex_lock(&tconn->conf_update); |
1072 | /* The discard_my_data flag is a single-shot modifier to the next | |
1073 | * connection attempt, the handshake of which is now well underway. | |
1074 | * No need for rcu style copying of the whole struct | |
1075 | * just to clear a single value. */ | |
1076 | tconn->net_conf->discard_my_data = 0; | |
1077 | mutex_unlock(&tconn->conf_update); | |
1078 | ||
d3fcb490 | 1079 | return h; |
b411b363 PR |
1080 | |
1081 | out_release_sockets: | |
7a426fd8 PR |
1082 | if (ad.s_listen) |
1083 | sock_release(ad.s_listen); | |
7da35862 PR |
1084 | if (sock.socket) |
1085 | sock_release(sock.socket); | |
1086 | if (msock.socket) | |
1087 | sock_release(msock.socket); | |
b411b363 PR |
1088 | return -1; |
1089 | } | |
1090 | ||
e658983a | 1091 | static int decode_header(struct drbd_tconn *tconn, void *header, struct packet_info *pi) |
b411b363 | 1092 | { |
e658983a AG |
1093 | unsigned int header_size = drbd_header_size(tconn); |
1094 | ||
0c8e36d9 AG |
1095 | if (header_size == sizeof(struct p_header100) && |
1096 | *(__be32 *)header == cpu_to_be32(DRBD_MAGIC_100)) { | |
1097 | struct p_header100 *h = header; | |
1098 | if (h->pad != 0) { | |
1099 | conn_err(tconn, "Header padding is not zero\n"); | |
1100 | return -EINVAL; | |
1101 | } | |
1102 | pi->vnr = be16_to_cpu(h->volume); | |
1103 | pi->cmd = be16_to_cpu(h->command); | |
1104 | pi->size = be32_to_cpu(h->length); | |
1105 | } else if (header_size == sizeof(struct p_header95) && | |
1106 | *(__be16 *)header == cpu_to_be16(DRBD_MAGIC_BIG)) { | |
e658983a | 1107 | struct p_header95 *h = header; |
e658983a | 1108 | pi->cmd = be16_to_cpu(h->command); |
b55d84ba AG |
1109 | pi->size = be32_to_cpu(h->length); |
1110 | pi->vnr = 0; | |
e658983a AG |
1111 | } else if (header_size == sizeof(struct p_header80) && |
1112 | *(__be32 *)header == cpu_to_be32(DRBD_MAGIC)) { | |
1113 | struct p_header80 *h = header; | |
1114 | pi->cmd = be16_to_cpu(h->command); | |
1115 | pi->size = be16_to_cpu(h->length); | |
77351055 | 1116 | pi->vnr = 0; |
02918be2 | 1117 | } else { |
e658983a AG |
1118 | conn_err(tconn, "Wrong magic value 0x%08x in protocol version %d\n", |
1119 | be32_to_cpu(*(__be32 *)header), | |
1120 | tconn->agreed_pro_version); | |
8172f3e9 | 1121 | return -EINVAL; |
b411b363 | 1122 | } |
e658983a | 1123 | pi->data = header + header_size; |
8172f3e9 | 1124 | return 0; |
257d0af6 | 1125 | } |
b411b363 | 1126 | |
9ba7aa00 | 1127 | static int drbd_recv_header(struct drbd_tconn *tconn, struct packet_info *pi) |
257d0af6 | 1128 | { |
e658983a | 1129 | void *buffer = tconn->data.rbuf; |
69bc7bc3 | 1130 | int err; |
257d0af6 | 1131 | |
e658983a | 1132 | err = drbd_recv_all_warn(tconn, buffer, drbd_header_size(tconn)); |
a5c31904 | 1133 | if (err) |
69bc7bc3 | 1134 | return err; |
257d0af6 | 1135 | |
e658983a | 1136 | err = decode_header(tconn, buffer, pi); |
9ba7aa00 | 1137 | tconn->last_received = jiffies; |
b411b363 | 1138 | |
69bc7bc3 | 1139 | return err; |
b411b363 PR |
1140 | } |
1141 | ||
4b0007c0 | 1142 | static void drbd_flush(struct drbd_tconn *tconn) |
b411b363 PR |
1143 | { |
1144 | int rv; | |
4b0007c0 PR |
1145 | struct drbd_conf *mdev; |
1146 | int vnr; | |
1147 | ||
1148 | if (tconn->write_ordering >= WO_bdev_flush) { | |
615e087f | 1149 | rcu_read_lock(); |
4b0007c0 | 1150 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
615e087f LE |
1151 | if (!get_ldev(mdev)) |
1152 | continue; | |
1153 | kref_get(&mdev->kref); | |
1154 | rcu_read_unlock(); | |
1155 | ||
1156 | rv = blkdev_issue_flush(mdev->ldev->backing_bdev, | |
1157 | GFP_NOIO, NULL); | |
1158 | if (rv) { | |
1159 | dev_info(DEV, "local disk flush failed with status %d\n", rv); | |
1160 | /* would rather check on EOPNOTSUPP, but that is not reliable. | |
1161 | * don't try again for ANY return value != 0 | |
1162 | * if (rv == -EOPNOTSUPP) */ | |
1163 | drbd_bump_write_ordering(tconn, WO_drain_io); | |
4b0007c0 | 1164 | } |
615e087f LE |
1165 | put_ldev(mdev); |
1166 | kref_put(&mdev->kref, &drbd_minor_destroy); | |
b411b363 | 1167 | |
615e087f LE |
1168 | rcu_read_lock(); |
1169 | if (rv) | |
1170 | break; | |
b411b363 | 1171 | } |
615e087f | 1172 | rcu_read_unlock(); |
b411b363 | 1173 | } |
b411b363 PR |
1174 | } |
1175 | ||
1176 | /** | |
1177 | * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. | |
1178 | * @mdev: DRBD device. | |
1179 | * @epoch: Epoch object. | |
1180 | * @ev: Epoch event. | |
1181 | */ | |
1e9dd291 | 1182 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_tconn *tconn, |
b411b363 PR |
1183 | struct drbd_epoch *epoch, |
1184 | enum epoch_event ev) | |
1185 | { | |
2451fc3b | 1186 | int epoch_size; |
b411b363 | 1187 | struct drbd_epoch *next_epoch; |
b411b363 PR |
1188 | enum finish_epoch rv = FE_STILL_LIVE; |
1189 | ||
12038a3a | 1190 | spin_lock(&tconn->epoch_lock); |
b411b363 PR |
1191 | do { |
1192 | next_epoch = NULL; | |
b411b363 PR |
1193 | |
1194 | epoch_size = atomic_read(&epoch->epoch_size); | |
1195 | ||
1196 | switch (ev & ~EV_CLEANUP) { | |
1197 | case EV_PUT: | |
1198 | atomic_dec(&epoch->active); | |
1199 | break; | |
1200 | case EV_GOT_BARRIER_NR: | |
1201 | set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); | |
b411b363 PR |
1202 | break; |
1203 | case EV_BECAME_LAST: | |
1204 | /* nothing to do*/ | |
1205 | break; | |
1206 | } | |
1207 | ||
b411b363 PR |
1208 | if (epoch_size != 0 && |
1209 | atomic_read(&epoch->active) == 0 && | |
80f9fd55 | 1210 | (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) || ev & EV_CLEANUP)) { |
b411b363 | 1211 | if (!(ev & EV_CLEANUP)) { |
12038a3a | 1212 | spin_unlock(&tconn->epoch_lock); |
9ed57dcb | 1213 | drbd_send_b_ack(epoch->tconn, epoch->barrier_nr, epoch_size); |
12038a3a | 1214 | spin_lock(&tconn->epoch_lock); |
b411b363 | 1215 | } |
9ed57dcb LE |
1216 | #if 0 |
1217 | /* FIXME: dec unacked on connection, once we have | |
1218 | * something to count pending connection packets in. */ | |
80f9fd55 | 1219 | if (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) |
9ed57dcb LE |
1220 | dec_unacked(epoch->tconn); |
1221 | #endif | |
b411b363 | 1222 | |
12038a3a | 1223 | if (tconn->current_epoch != epoch) { |
b411b363 PR |
1224 | next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); |
1225 | list_del(&epoch->list); | |
1226 | ev = EV_BECAME_LAST | (ev & EV_CLEANUP); | |
12038a3a | 1227 | tconn->epochs--; |
b411b363 PR |
1228 | kfree(epoch); |
1229 | ||
1230 | if (rv == FE_STILL_LIVE) | |
1231 | rv = FE_DESTROYED; | |
1232 | } else { | |
1233 | epoch->flags = 0; | |
1234 | atomic_set(&epoch->epoch_size, 0); | |
698f9315 | 1235 | /* atomic_set(&epoch->active, 0); is already zero */ |
b411b363 PR |
1236 | if (rv == FE_STILL_LIVE) |
1237 | rv = FE_RECYCLED; | |
1238 | } | |
1239 | } | |
1240 | ||
1241 | if (!next_epoch) | |
1242 | break; | |
1243 | ||
1244 | epoch = next_epoch; | |
1245 | } while (1); | |
1246 | ||
12038a3a | 1247 | spin_unlock(&tconn->epoch_lock); |
b411b363 | 1248 | |
b411b363 PR |
1249 | return rv; |
1250 | } | |
1251 | ||
1252 | /** | |
1253 | * drbd_bump_write_ordering() - Fall back to an other write ordering method | |
4b0007c0 | 1254 | * @tconn: DRBD connection. |
b411b363 PR |
1255 | * @wo: Write ordering method to try. |
1256 | */ | |
4b0007c0 | 1257 | void drbd_bump_write_ordering(struct drbd_tconn *tconn, enum write_ordering_e wo) |
b411b363 | 1258 | { |
daeda1cc | 1259 | struct disk_conf *dc; |
4b0007c0 | 1260 | struct drbd_conf *mdev; |
b411b363 | 1261 | enum write_ordering_e pwo; |
4b0007c0 | 1262 | int vnr; |
b411b363 PR |
1263 | static char *write_ordering_str[] = { |
1264 | [WO_none] = "none", | |
1265 | [WO_drain_io] = "drain", | |
1266 | [WO_bdev_flush] = "flush", | |
b411b363 PR |
1267 | }; |
1268 | ||
4b0007c0 | 1269 | pwo = tconn->write_ordering; |
b411b363 | 1270 | wo = min(pwo, wo); |
daeda1cc | 1271 | rcu_read_lock(); |
4b0007c0 | 1272 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
27eb13e9 | 1273 | if (!get_ldev_if_state(mdev, D_ATTACHING)) |
4b0007c0 PR |
1274 | continue; |
1275 | dc = rcu_dereference(mdev->ldev->disk_conf); | |
1276 | ||
1277 | if (wo == WO_bdev_flush && !dc->disk_flushes) | |
1278 | wo = WO_drain_io; | |
1279 | if (wo == WO_drain_io && !dc->disk_drain) | |
1280 | wo = WO_none; | |
1281 | put_ldev(mdev); | |
1282 | } | |
daeda1cc | 1283 | rcu_read_unlock(); |
4b0007c0 PR |
1284 | tconn->write_ordering = wo; |
1285 | if (pwo != tconn->write_ordering || wo == WO_bdev_flush) | |
1286 | conn_info(tconn, "Method to ensure write ordering: %s\n", write_ordering_str[tconn->write_ordering]); | |
b411b363 PR |
1287 | } |
1288 | ||
45bb912b | 1289 | /** |
fbe29dec | 1290 | * drbd_submit_peer_request() |
45bb912b | 1291 | * @mdev: DRBD device. |
db830c46 | 1292 | * @peer_req: peer request |
45bb912b | 1293 | * @rw: flag field, see bio->bi_rw |
10f6d992 LE |
1294 | * |
1295 | * May spread the pages to multiple bios, | |
1296 | * depending on bio_add_page restrictions. | |
1297 | * | |
1298 | * Returns 0 if all bios have been submitted, | |
1299 | * -ENOMEM if we could not allocate enough bios, | |
1300 | * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a | |
1301 | * single page to an empty bio (which should never happen and likely indicates | |
1302 | * that the lower level IO stack is in some way broken). This has been observed | |
1303 | * on certain Xen deployments. | |
45bb912b LE |
1304 | */ |
1305 | /* TODO allocate from our own bio_set. */ | |
fbe29dec AG |
1306 | int drbd_submit_peer_request(struct drbd_conf *mdev, |
1307 | struct drbd_peer_request *peer_req, | |
1308 | const unsigned rw, const int fault_type) | |
45bb912b LE |
1309 | { |
1310 | struct bio *bios = NULL; | |
1311 | struct bio *bio; | |
db830c46 AG |
1312 | struct page *page = peer_req->pages; |
1313 | sector_t sector = peer_req->i.sector; | |
1314 | unsigned ds = peer_req->i.size; | |
45bb912b LE |
1315 | unsigned n_bios = 0; |
1316 | unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT; | |
10f6d992 | 1317 | int err = -ENOMEM; |
45bb912b LE |
1318 | |
1319 | /* In most cases, we will only need one bio. But in case the lower | |
1320 | * level restrictions happen to be different at this offset on this | |
1321 | * side than those of the sending peer, we may need to submit the | |
9476f39d LE |
1322 | * request in more than one bio. |
1323 | * | |
1324 | * Plain bio_alloc is good enough here, this is no DRBD internally | |
1325 | * generated bio, but a bio allocated on behalf of the peer. | |
1326 | */ | |
45bb912b LE |
1327 | next_bio: |
1328 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
1329 | if (!bio) { | |
1330 | dev_err(DEV, "submit_ee: Allocation of a bio failed\n"); | |
1331 | goto fail; | |
1332 | } | |
db830c46 | 1333 | /* > peer_req->i.sector, unless this is the first bio */ |
45bb912b LE |
1334 | bio->bi_sector = sector; |
1335 | bio->bi_bdev = mdev->ldev->backing_bdev; | |
45bb912b | 1336 | bio->bi_rw = rw; |
db830c46 | 1337 | bio->bi_private = peer_req; |
fcefa62e | 1338 | bio->bi_end_io = drbd_peer_request_endio; |
45bb912b LE |
1339 | |
1340 | bio->bi_next = bios; | |
1341 | bios = bio; | |
1342 | ++n_bios; | |
1343 | ||
1344 | page_chain_for_each(page) { | |
1345 | unsigned len = min_t(unsigned, ds, PAGE_SIZE); | |
1346 | if (!bio_add_page(bio, page, len, 0)) { | |
10f6d992 LE |
1347 | /* A single page must always be possible! |
1348 | * But in case it fails anyways, | |
1349 | * we deal with it, and complain (below). */ | |
1350 | if (bio->bi_vcnt == 0) { | |
1351 | dev_err(DEV, | |
1352 | "bio_add_page failed for len=%u, " | |
1353 | "bi_vcnt=0 (bi_sector=%llu)\n", | |
1354 | len, (unsigned long long)bio->bi_sector); | |
1355 | err = -ENOSPC; | |
1356 | goto fail; | |
1357 | } | |
45bb912b LE |
1358 | goto next_bio; |
1359 | } | |
1360 | ds -= len; | |
1361 | sector += len >> 9; | |
1362 | --nr_pages; | |
1363 | } | |
1364 | D_ASSERT(page == NULL); | |
1365 | D_ASSERT(ds == 0); | |
1366 | ||
db830c46 | 1367 | atomic_set(&peer_req->pending_bios, n_bios); |
45bb912b LE |
1368 | do { |
1369 | bio = bios; | |
1370 | bios = bios->bi_next; | |
1371 | bio->bi_next = NULL; | |
1372 | ||
45bb912b | 1373 | drbd_generic_make_request(mdev, fault_type, bio); |
45bb912b | 1374 | } while (bios); |
45bb912b LE |
1375 | return 0; |
1376 | ||
1377 | fail: | |
1378 | while (bios) { | |
1379 | bio = bios; | |
1380 | bios = bios->bi_next; | |
1381 | bio_put(bio); | |
1382 | } | |
10f6d992 | 1383 | return err; |
45bb912b LE |
1384 | } |
1385 | ||
53840641 | 1386 | static void drbd_remove_epoch_entry_interval(struct drbd_conf *mdev, |
db830c46 | 1387 | struct drbd_peer_request *peer_req) |
53840641 | 1388 | { |
db830c46 | 1389 | struct drbd_interval *i = &peer_req->i; |
53840641 AG |
1390 | |
1391 | drbd_remove_interval(&mdev->write_requests, i); | |
1392 | drbd_clear_interval(i); | |
1393 | ||
6c852bec | 1394 | /* Wake up any processes waiting for this peer request to complete. */ |
53840641 AG |
1395 | if (i->waiting) |
1396 | wake_up(&mdev->misc_wait); | |
1397 | } | |
1398 | ||
77fede51 PR |
1399 | void conn_wait_active_ee_empty(struct drbd_tconn *tconn) |
1400 | { | |
1401 | struct drbd_conf *mdev; | |
1402 | int vnr; | |
1403 | ||
1404 | rcu_read_lock(); | |
1405 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { | |
1406 | kref_get(&mdev->kref); | |
1407 | rcu_read_unlock(); | |
1408 | drbd_wait_ee_list_empty(mdev, &mdev->active_ee); | |
1409 | kref_put(&mdev->kref, &drbd_minor_destroy); | |
1410 | rcu_read_lock(); | |
1411 | } | |
1412 | rcu_read_unlock(); | |
1413 | } | |
1414 | ||
4a76b161 | 1415 | static int receive_Barrier(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 1416 | { |
2451fc3b | 1417 | int rv; |
e658983a | 1418 | struct p_barrier *p = pi->data; |
b411b363 PR |
1419 | struct drbd_epoch *epoch; |
1420 | ||
9ed57dcb LE |
1421 | /* FIXME these are unacked on connection, |
1422 | * not a specific (peer)device. | |
1423 | */ | |
12038a3a | 1424 | tconn->current_epoch->barrier_nr = p->barrier; |
9ed57dcb | 1425 | tconn->current_epoch->tconn = tconn; |
1e9dd291 | 1426 | rv = drbd_may_finish_epoch(tconn, tconn->current_epoch, EV_GOT_BARRIER_NR); |
b411b363 PR |
1427 | |
1428 | /* P_BARRIER_ACK may imply that the corresponding extent is dropped from | |
1429 | * the activity log, which means it would not be resynced in case the | |
1430 | * R_PRIMARY crashes now. | |
1431 | * Therefore we must send the barrier_ack after the barrier request was | |
1432 | * completed. */ | |
4b0007c0 | 1433 | switch (tconn->write_ordering) { |
b411b363 PR |
1434 | case WO_none: |
1435 | if (rv == FE_RECYCLED) | |
82bc0194 | 1436 | return 0; |
2451fc3b PR |
1437 | |
1438 | /* receiver context, in the writeout path of the other node. | |
1439 | * avoid potential distributed deadlock */ | |
1440 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1441 | if (epoch) | |
1442 | break; | |
1443 | else | |
9ed57dcb | 1444 | conn_warn(tconn, "Allocation of an epoch failed, slowing down\n"); |
2451fc3b | 1445 | /* Fall through */ |
b411b363 PR |
1446 | |
1447 | case WO_bdev_flush: | |
1448 | case WO_drain_io: | |
77fede51 | 1449 | conn_wait_active_ee_empty(tconn); |
4b0007c0 | 1450 | drbd_flush(tconn); |
2451fc3b | 1451 | |
12038a3a | 1452 | if (atomic_read(&tconn->current_epoch->epoch_size)) { |
2451fc3b PR |
1453 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); |
1454 | if (epoch) | |
1455 | break; | |
b411b363 PR |
1456 | } |
1457 | ||
82bc0194 | 1458 | return 0; |
2451fc3b | 1459 | default: |
9ed57dcb | 1460 | conn_err(tconn, "Strangeness in tconn->write_ordering %d\n", tconn->write_ordering); |
82bc0194 | 1461 | return -EIO; |
b411b363 PR |
1462 | } |
1463 | ||
1464 | epoch->flags = 0; | |
1465 | atomic_set(&epoch->epoch_size, 0); | |
1466 | atomic_set(&epoch->active, 0); | |
1467 | ||
12038a3a PR |
1468 | spin_lock(&tconn->epoch_lock); |
1469 | if (atomic_read(&tconn->current_epoch->epoch_size)) { | |
1470 | list_add(&epoch->list, &tconn->current_epoch->list); | |
1471 | tconn->current_epoch = epoch; | |
1472 | tconn->epochs++; | |
b411b363 PR |
1473 | } else { |
1474 | /* The current_epoch got recycled while we allocated this one... */ | |
1475 | kfree(epoch); | |
1476 | } | |
12038a3a | 1477 | spin_unlock(&tconn->epoch_lock); |
b411b363 | 1478 | |
82bc0194 | 1479 | return 0; |
b411b363 PR |
1480 | } |
1481 | ||
1482 | /* used from receive_RSDataReply (recv_resync_read) | |
1483 | * and from receive_Data */ | |
f6ffca9f AG |
1484 | static struct drbd_peer_request * |
1485 | read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, | |
1486 | int data_size) __must_hold(local) | |
b411b363 | 1487 | { |
6666032a | 1488 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); |
db830c46 | 1489 | struct drbd_peer_request *peer_req; |
b411b363 | 1490 | struct page *page; |
a5c31904 | 1491 | int dgs, ds, err; |
a0638456 PR |
1492 | void *dig_in = mdev->tconn->int_dig_in; |
1493 | void *dig_vv = mdev->tconn->int_dig_vv; | |
6b4388ac | 1494 | unsigned long *data; |
b411b363 | 1495 | |
88104ca4 AG |
1496 | dgs = 0; |
1497 | if (mdev->tconn->peer_integrity_tfm) { | |
1498 | dgs = crypto_hash_digestsize(mdev->tconn->peer_integrity_tfm); | |
9f5bdc33 AG |
1499 | /* |
1500 | * FIXME: Receive the incoming digest into the receive buffer | |
1501 | * here, together with its struct p_data? | |
1502 | */ | |
a5c31904 AG |
1503 | err = drbd_recv_all_warn(mdev->tconn, dig_in, dgs); |
1504 | if (err) | |
b411b363 | 1505 | return NULL; |
88104ca4 | 1506 | data_size -= dgs; |
b411b363 PR |
1507 | } |
1508 | ||
841ce241 AG |
1509 | if (!expect(IS_ALIGNED(data_size, 512))) |
1510 | return NULL; | |
1511 | if (!expect(data_size <= DRBD_MAX_BIO_SIZE)) | |
1512 | return NULL; | |
b411b363 | 1513 | |
6666032a LE |
1514 | /* even though we trust out peer, |
1515 | * we sometimes have to double check. */ | |
1516 | if (sector + (data_size>>9) > capacity) { | |
fdda6544 LE |
1517 | dev_err(DEV, "request from peer beyond end of local disk: " |
1518 | "capacity: %llus < sector: %llus + size: %u\n", | |
6666032a LE |
1519 | (unsigned long long)capacity, |
1520 | (unsigned long long)sector, data_size); | |
1521 | return NULL; | |
1522 | } | |
1523 | ||
b411b363 PR |
1524 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD |
1525 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
1526 | * which in turn might block on the other node at this very place. */ | |
0db55363 | 1527 | peer_req = drbd_alloc_peer_req(mdev, id, sector, data_size, GFP_NOIO); |
db830c46 | 1528 | if (!peer_req) |
b411b363 | 1529 | return NULL; |
45bb912b | 1530 | |
a73ff323 | 1531 | if (!data_size) |
81a3537a | 1532 | return peer_req; |
a73ff323 | 1533 | |
b411b363 | 1534 | ds = data_size; |
db830c46 | 1535 | page = peer_req->pages; |
45bb912b LE |
1536 | page_chain_for_each(page) { |
1537 | unsigned len = min_t(int, ds, PAGE_SIZE); | |
6b4388ac | 1538 | data = kmap(page); |
a5c31904 | 1539 | err = drbd_recv_all_warn(mdev->tconn, data, len); |
0cf9d27e | 1540 | if (drbd_insert_fault(mdev, DRBD_FAULT_RECEIVE)) { |
6b4388ac PR |
1541 | dev_err(DEV, "Fault injection: Corrupting data on receive\n"); |
1542 | data[0] = data[0] ^ (unsigned long)-1; | |
1543 | } | |
b411b363 | 1544 | kunmap(page); |
a5c31904 | 1545 | if (err) { |
3967deb1 | 1546 | drbd_free_peer_req(mdev, peer_req); |
b411b363 PR |
1547 | return NULL; |
1548 | } | |
a5c31904 | 1549 | ds -= len; |
b411b363 PR |
1550 | } |
1551 | ||
1552 | if (dgs) { | |
5b614abe | 1553 | drbd_csum_ee(mdev, mdev->tconn->peer_integrity_tfm, peer_req, dig_vv); |
b411b363 | 1554 | if (memcmp(dig_in, dig_vv, dgs)) { |
470be44a LE |
1555 | dev_err(DEV, "Digest integrity check FAILED: %llus +%u\n", |
1556 | (unsigned long long)sector, data_size); | |
3967deb1 | 1557 | drbd_free_peer_req(mdev, peer_req); |
b411b363 PR |
1558 | return NULL; |
1559 | } | |
1560 | } | |
1561 | mdev->recv_cnt += data_size>>9; | |
db830c46 | 1562 | return peer_req; |
b411b363 PR |
1563 | } |
1564 | ||
1565 | /* drbd_drain_block() just takes a data block | |
1566 | * out of the socket input buffer, and discards it. | |
1567 | */ | |
1568 | static int drbd_drain_block(struct drbd_conf *mdev, int data_size) | |
1569 | { | |
1570 | struct page *page; | |
a5c31904 | 1571 | int err = 0; |
b411b363 PR |
1572 | void *data; |
1573 | ||
c3470cde | 1574 | if (!data_size) |
fc5be839 | 1575 | return 0; |
c3470cde | 1576 | |
c37c8ecf | 1577 | page = drbd_alloc_pages(mdev, 1, 1); |
b411b363 PR |
1578 | |
1579 | data = kmap(page); | |
1580 | while (data_size) { | |
fc5be839 AG |
1581 | unsigned int len = min_t(int, data_size, PAGE_SIZE); |
1582 | ||
a5c31904 AG |
1583 | err = drbd_recv_all_warn(mdev->tconn, data, len); |
1584 | if (err) | |
b411b363 | 1585 | break; |
a5c31904 | 1586 | data_size -= len; |
b411b363 PR |
1587 | } |
1588 | kunmap(page); | |
5cc287e0 | 1589 | drbd_free_pages(mdev, page, 0); |
fc5be839 | 1590 | return err; |
b411b363 PR |
1591 | } |
1592 | ||
1593 | static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req, | |
1594 | sector_t sector, int data_size) | |
1595 | { | |
1596 | struct bio_vec *bvec; | |
1597 | struct bio *bio; | |
a5c31904 | 1598 | int dgs, err, i, expect; |
a0638456 PR |
1599 | void *dig_in = mdev->tconn->int_dig_in; |
1600 | void *dig_vv = mdev->tconn->int_dig_vv; | |
b411b363 | 1601 | |
88104ca4 AG |
1602 | dgs = 0; |
1603 | if (mdev->tconn->peer_integrity_tfm) { | |
1604 | dgs = crypto_hash_digestsize(mdev->tconn->peer_integrity_tfm); | |
a5c31904 AG |
1605 | err = drbd_recv_all_warn(mdev->tconn, dig_in, dgs); |
1606 | if (err) | |
1607 | return err; | |
88104ca4 | 1608 | data_size -= dgs; |
b411b363 PR |
1609 | } |
1610 | ||
b411b363 PR |
1611 | /* optimistically update recv_cnt. if receiving fails below, |
1612 | * we disconnect anyways, and counters will be reset. */ | |
1613 | mdev->recv_cnt += data_size>>9; | |
1614 | ||
1615 | bio = req->master_bio; | |
1616 | D_ASSERT(sector == bio->bi_sector); | |
1617 | ||
1618 | bio_for_each_segment(bvec, bio, i) { | |
a5c31904 | 1619 | void *mapped = kmap(bvec->bv_page) + bvec->bv_offset; |
b411b363 | 1620 | expect = min_t(int, data_size, bvec->bv_len); |
a5c31904 | 1621 | err = drbd_recv_all_warn(mdev->tconn, mapped, expect); |
b411b363 | 1622 | kunmap(bvec->bv_page); |
a5c31904 AG |
1623 | if (err) |
1624 | return err; | |
1625 | data_size -= expect; | |
b411b363 PR |
1626 | } |
1627 | ||
1628 | if (dgs) { | |
5b614abe | 1629 | drbd_csum_bio(mdev, mdev->tconn->peer_integrity_tfm, bio, dig_vv); |
b411b363 PR |
1630 | if (memcmp(dig_in, dig_vv, dgs)) { |
1631 | dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n"); | |
28284cef | 1632 | return -EINVAL; |
b411b363 PR |
1633 | } |
1634 | } | |
1635 | ||
1636 | D_ASSERT(data_size == 0); | |
28284cef | 1637 | return 0; |
b411b363 PR |
1638 | } |
1639 | ||
a990be46 AG |
1640 | /* |
1641 | * e_end_resync_block() is called in asender context via | |
1642 | * drbd_finish_peer_reqs(). | |
1643 | */ | |
99920dc5 | 1644 | static int e_end_resync_block(struct drbd_work *w, int unused) |
b411b363 | 1645 | { |
8050e6d0 AG |
1646 | struct drbd_peer_request *peer_req = |
1647 | container_of(w, struct drbd_peer_request, w); | |
00d56944 | 1648 | struct drbd_conf *mdev = w->mdev; |
db830c46 | 1649 | sector_t sector = peer_req->i.sector; |
99920dc5 | 1650 | int err; |
b411b363 | 1651 | |
db830c46 | 1652 | D_ASSERT(drbd_interval_empty(&peer_req->i)); |
b411b363 | 1653 | |
db830c46 AG |
1654 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
1655 | drbd_set_in_sync(mdev, sector, peer_req->i.size); | |
99920dc5 | 1656 | err = drbd_send_ack(mdev, P_RS_WRITE_ACK, peer_req); |
b411b363 PR |
1657 | } else { |
1658 | /* Record failure to sync */ | |
db830c46 | 1659 | drbd_rs_failed_io(mdev, sector, peer_req->i.size); |
b411b363 | 1660 | |
99920dc5 | 1661 | err = drbd_send_ack(mdev, P_NEG_ACK, peer_req); |
b411b363 PR |
1662 | } |
1663 | dec_unacked(mdev); | |
1664 | ||
99920dc5 | 1665 | return err; |
b411b363 PR |
1666 | } |
1667 | ||
1668 | static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local) | |
1669 | { | |
db830c46 | 1670 | struct drbd_peer_request *peer_req; |
b411b363 | 1671 | |
db830c46 AG |
1672 | peer_req = read_in_block(mdev, ID_SYNCER, sector, data_size); |
1673 | if (!peer_req) | |
45bb912b | 1674 | goto fail; |
b411b363 PR |
1675 | |
1676 | dec_rs_pending(mdev); | |
1677 | ||
b411b363 PR |
1678 | inc_unacked(mdev); |
1679 | /* corresponding dec_unacked() in e_end_resync_block() | |
1680 | * respective _drbd_clear_done_ee */ | |
1681 | ||
db830c46 | 1682 | peer_req->w.cb = e_end_resync_block; |
45bb912b | 1683 | |
87eeee41 | 1684 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 1685 | list_add(&peer_req->w.list, &mdev->sync_ee); |
87eeee41 | 1686 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 1687 | |
0f0601f4 | 1688 | atomic_add(data_size >> 9, &mdev->rs_sect_ev); |
fbe29dec | 1689 | if (drbd_submit_peer_request(mdev, peer_req, WRITE, DRBD_FAULT_RS_WR) == 0) |
e1c1b0fc | 1690 | return 0; |
b411b363 | 1691 | |
10f6d992 LE |
1692 | /* don't care for the reason here */ |
1693 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 1694 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 1695 | list_del(&peer_req->w.list); |
87eeee41 | 1696 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 | 1697 | |
3967deb1 | 1698 | drbd_free_peer_req(mdev, peer_req); |
45bb912b LE |
1699 | fail: |
1700 | put_ldev(mdev); | |
e1c1b0fc | 1701 | return -EIO; |
b411b363 PR |
1702 | } |
1703 | ||
668eebc6 | 1704 | static struct drbd_request * |
bc9c5c41 AG |
1705 | find_request(struct drbd_conf *mdev, struct rb_root *root, u64 id, |
1706 | sector_t sector, bool missing_ok, const char *func) | |
51624585 | 1707 | { |
51624585 AG |
1708 | struct drbd_request *req; |
1709 | ||
bc9c5c41 AG |
1710 | /* Request object according to our peer */ |
1711 | req = (struct drbd_request *)(unsigned long)id; | |
5e472264 | 1712 | if (drbd_contains_interval(root, sector, &req->i) && req->i.local) |
668eebc6 | 1713 | return req; |
c3afd8f5 | 1714 | if (!missing_ok) { |
5af172ed | 1715 | dev_err(DEV, "%s: failed to find request 0x%lx, sector %llus\n", func, |
c3afd8f5 AG |
1716 | (unsigned long)id, (unsigned long long)sector); |
1717 | } | |
51624585 | 1718 | return NULL; |
b411b363 PR |
1719 | } |
1720 | ||
4a76b161 | 1721 | static int receive_DataReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 1722 | { |
4a76b161 | 1723 | struct drbd_conf *mdev; |
b411b363 PR |
1724 | struct drbd_request *req; |
1725 | sector_t sector; | |
82bc0194 | 1726 | int err; |
e658983a | 1727 | struct p_data *p = pi->data; |
4a76b161 AG |
1728 | |
1729 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
1730 | if (!mdev) | |
1731 | return -EIO; | |
b411b363 PR |
1732 | |
1733 | sector = be64_to_cpu(p->sector); | |
1734 | ||
87eeee41 | 1735 | spin_lock_irq(&mdev->tconn->req_lock); |
bc9c5c41 | 1736 | req = find_request(mdev, &mdev->read_requests, p->block_id, sector, false, __func__); |
87eeee41 | 1737 | spin_unlock_irq(&mdev->tconn->req_lock); |
c3afd8f5 | 1738 | if (unlikely(!req)) |
82bc0194 | 1739 | return -EIO; |
b411b363 | 1740 | |
24c4830c | 1741 | /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid |
b411b363 PR |
1742 | * special casing it there for the various failure cases. |
1743 | * still no race with drbd_fail_pending_reads */ | |
e2857216 | 1744 | err = recv_dless_read(mdev, req, sector, pi->size); |
82bc0194 | 1745 | if (!err) |
8554df1c | 1746 | req_mod(req, DATA_RECEIVED); |
b411b363 PR |
1747 | /* else: nothing. handled from drbd_disconnect... |
1748 | * I don't think we may complete this just yet | |
1749 | * in case we are "on-disconnect: freeze" */ | |
1750 | ||
82bc0194 | 1751 | return err; |
b411b363 PR |
1752 | } |
1753 | ||
4a76b161 | 1754 | static int receive_RSDataReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 1755 | { |
4a76b161 | 1756 | struct drbd_conf *mdev; |
b411b363 | 1757 | sector_t sector; |
82bc0194 | 1758 | int err; |
e658983a | 1759 | struct p_data *p = pi->data; |
4a76b161 AG |
1760 | |
1761 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
1762 | if (!mdev) | |
1763 | return -EIO; | |
b411b363 PR |
1764 | |
1765 | sector = be64_to_cpu(p->sector); | |
1766 | D_ASSERT(p->block_id == ID_SYNCER); | |
1767 | ||
1768 | if (get_ldev(mdev)) { | |
1769 | /* data is submitted to disk within recv_resync_read. | |
1770 | * corresponding put_ldev done below on error, | |
fcefa62e | 1771 | * or in drbd_peer_request_endio. */ |
e2857216 | 1772 | err = recv_resync_read(mdev, sector, pi->size); |
b411b363 PR |
1773 | } else { |
1774 | if (__ratelimit(&drbd_ratelimit_state)) | |
1775 | dev_err(DEV, "Can not write resync data to local disk.\n"); | |
1776 | ||
e2857216 | 1777 | err = drbd_drain_block(mdev, pi->size); |
b411b363 | 1778 | |
e2857216 | 1779 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, pi->size); |
b411b363 PR |
1780 | } |
1781 | ||
e2857216 | 1782 | atomic_add(pi->size >> 9, &mdev->rs_sect_in); |
778f271d | 1783 | |
82bc0194 | 1784 | return err; |
b411b363 PR |
1785 | } |
1786 | ||
7be8da07 AG |
1787 | static void restart_conflicting_writes(struct drbd_conf *mdev, |
1788 | sector_t sector, int size) | |
b411b363 | 1789 | { |
7be8da07 AG |
1790 | struct drbd_interval *i; |
1791 | struct drbd_request *req; | |
1792 | ||
1793 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
1794 | if (!i->local) | |
1795 | continue; | |
1796 | req = container_of(i, struct drbd_request, i); | |
1797 | if (req->rq_state & RQ_LOCAL_PENDING || | |
1798 | !(req->rq_state & RQ_POSTPONED)) | |
1799 | continue; | |
2312f0b3 LE |
1800 | /* as it is RQ_POSTPONED, this will cause it to |
1801 | * be queued on the retry workqueue. */ | |
d4dabbe2 | 1802 | __req_mod(req, CONFLICT_RESOLVED, NULL); |
7be8da07 AG |
1803 | } |
1804 | } | |
b411b363 | 1805 | |
a990be46 AG |
1806 | /* |
1807 | * e_end_block() is called in asender context via drbd_finish_peer_reqs(). | |
b411b363 | 1808 | */ |
99920dc5 | 1809 | static int e_end_block(struct drbd_work *w, int cancel) |
b411b363 | 1810 | { |
8050e6d0 AG |
1811 | struct drbd_peer_request *peer_req = |
1812 | container_of(w, struct drbd_peer_request, w); | |
00d56944 | 1813 | struct drbd_conf *mdev = w->mdev; |
db830c46 | 1814 | sector_t sector = peer_req->i.sector; |
99920dc5 | 1815 | int err = 0, pcmd; |
b411b363 | 1816 | |
303d1448 | 1817 | if (peer_req->flags & EE_SEND_WRITE_ACK) { |
db830c46 | 1818 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1819 | pcmd = (mdev->state.conn >= C_SYNC_SOURCE && |
1820 | mdev->state.conn <= C_PAUSED_SYNC_T && | |
db830c46 | 1821 | peer_req->flags & EE_MAY_SET_IN_SYNC) ? |
b411b363 | 1822 | P_RS_WRITE_ACK : P_WRITE_ACK; |
99920dc5 | 1823 | err = drbd_send_ack(mdev, pcmd, peer_req); |
b411b363 | 1824 | if (pcmd == P_RS_WRITE_ACK) |
db830c46 | 1825 | drbd_set_in_sync(mdev, sector, peer_req->i.size); |
b411b363 | 1826 | } else { |
99920dc5 | 1827 | err = drbd_send_ack(mdev, P_NEG_ACK, peer_req); |
b411b363 PR |
1828 | /* we expect it to be marked out of sync anyways... |
1829 | * maybe assert this? */ | |
1830 | } | |
1831 | dec_unacked(mdev); | |
1832 | } | |
1833 | /* we delete from the conflict detection hash _after_ we sent out the | |
1834 | * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ | |
302bdeae | 1835 | if (peer_req->flags & EE_IN_INTERVAL_TREE) { |
87eeee41 | 1836 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 AG |
1837 | D_ASSERT(!drbd_interval_empty(&peer_req->i)); |
1838 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
7be8da07 AG |
1839 | if (peer_req->flags & EE_RESTART_REQUESTS) |
1840 | restart_conflicting_writes(mdev, sector, peer_req->i.size); | |
87eeee41 | 1841 | spin_unlock_irq(&mdev->tconn->req_lock); |
bb3bfe96 | 1842 | } else |
db830c46 | 1843 | D_ASSERT(drbd_interval_empty(&peer_req->i)); |
b411b363 | 1844 | |
1e9dd291 | 1845 | drbd_may_finish_epoch(mdev->tconn, peer_req->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); |
b411b363 | 1846 | |
99920dc5 | 1847 | return err; |
b411b363 PR |
1848 | } |
1849 | ||
7be8da07 | 1850 | static int e_send_ack(struct drbd_work *w, enum drbd_packet ack) |
b411b363 | 1851 | { |
7be8da07 | 1852 | struct drbd_conf *mdev = w->mdev; |
8050e6d0 AG |
1853 | struct drbd_peer_request *peer_req = |
1854 | container_of(w, struct drbd_peer_request, w); | |
99920dc5 | 1855 | int err; |
b411b363 | 1856 | |
99920dc5 | 1857 | err = drbd_send_ack(mdev, ack, peer_req); |
b411b363 PR |
1858 | dec_unacked(mdev); |
1859 | ||
99920dc5 | 1860 | return err; |
b411b363 PR |
1861 | } |
1862 | ||
d4dabbe2 | 1863 | static int e_send_superseded(struct drbd_work *w, int unused) |
7be8da07 | 1864 | { |
d4dabbe2 | 1865 | return e_send_ack(w, P_SUPERSEDED); |
7be8da07 AG |
1866 | } |
1867 | ||
99920dc5 | 1868 | static int e_send_retry_write(struct drbd_work *w, int unused) |
7be8da07 AG |
1869 | { |
1870 | struct drbd_tconn *tconn = w->mdev->tconn; | |
1871 | ||
1872 | return e_send_ack(w, tconn->agreed_pro_version >= 100 ? | |
d4dabbe2 | 1873 | P_RETRY_WRITE : P_SUPERSEDED); |
7be8da07 | 1874 | } |
b411b363 | 1875 | |
3e394da1 AG |
1876 | static bool seq_greater(u32 a, u32 b) |
1877 | { | |
1878 | /* | |
1879 | * We assume 32-bit wrap-around here. | |
1880 | * For 24-bit wrap-around, we would have to shift: | |
1881 | * a <<= 8; b <<= 8; | |
1882 | */ | |
1883 | return (s32)a - (s32)b > 0; | |
1884 | } | |
b411b363 | 1885 | |
3e394da1 AG |
1886 | static u32 seq_max(u32 a, u32 b) |
1887 | { | |
1888 | return seq_greater(a, b) ? a : b; | |
b411b363 PR |
1889 | } |
1890 | ||
7be8da07 | 1891 | static bool need_peer_seq(struct drbd_conf *mdev) |
b411b363 | 1892 | { |
7be8da07 | 1893 | struct drbd_tconn *tconn = mdev->tconn; |
302bdeae | 1894 | int tp; |
b411b363 | 1895 | |
7be8da07 AG |
1896 | /* |
1897 | * We only need to keep track of the last packet_seq number of our peer | |
427c0434 | 1898 | * if we are in dual-primary mode and we have the resolve-conflicts flag set; see |
7be8da07 AG |
1899 | * handle_write_conflicts(). |
1900 | */ | |
b411b363 | 1901 | |
302bdeae PR |
1902 | rcu_read_lock(); |
1903 | tp = rcu_dereference(mdev->tconn->net_conf)->two_primaries; | |
1904 | rcu_read_unlock(); | |
b411b363 | 1905 | |
427c0434 | 1906 | return tp && test_bit(RESOLVE_CONFLICTS, &tconn->flags); |
7be8da07 | 1907 | } |
b411b363 | 1908 | |
43ae077d | 1909 | static void update_peer_seq(struct drbd_conf *mdev, unsigned int peer_seq) |
3e394da1 | 1910 | { |
3c13b680 | 1911 | unsigned int newest_peer_seq; |
3e394da1 | 1912 | |
7be8da07 AG |
1913 | if (need_peer_seq(mdev)) { |
1914 | spin_lock(&mdev->peer_seq_lock); | |
3c13b680 LE |
1915 | newest_peer_seq = seq_max(mdev->peer_seq, peer_seq); |
1916 | mdev->peer_seq = newest_peer_seq; | |
7be8da07 | 1917 | spin_unlock(&mdev->peer_seq_lock); |
3c13b680 LE |
1918 | /* wake up only if we actually changed mdev->peer_seq */ |
1919 | if (peer_seq == newest_peer_seq) | |
7be8da07 AG |
1920 | wake_up(&mdev->seq_wait); |
1921 | } | |
b411b363 PR |
1922 | } |
1923 | ||
d93f6302 | 1924 | static inline int overlaps(sector_t s1, int l1, sector_t s2, int l2) |
b6a370ba | 1925 | { |
d93f6302 LE |
1926 | return !((s1 + (l1>>9) <= s2) || (s1 >= s2 + (l2>>9))); |
1927 | } | |
b6a370ba | 1928 | |
d93f6302 | 1929 | /* maybe change sync_ee into interval trees as well? */ |
3ea35df8 | 1930 | static bool overlapping_resync_write(struct drbd_conf *mdev, struct drbd_peer_request *peer_req) |
d93f6302 LE |
1931 | { |
1932 | struct drbd_peer_request *rs_req; | |
b6a370ba PR |
1933 | bool rv = 0; |
1934 | ||
d93f6302 LE |
1935 | spin_lock_irq(&mdev->tconn->req_lock); |
1936 | list_for_each_entry(rs_req, &mdev->sync_ee, w.list) { | |
1937 | if (overlaps(peer_req->i.sector, peer_req->i.size, | |
1938 | rs_req->i.sector, rs_req->i.size)) { | |
b6a370ba PR |
1939 | rv = 1; |
1940 | break; | |
1941 | } | |
1942 | } | |
d93f6302 | 1943 | spin_unlock_irq(&mdev->tconn->req_lock); |
b6a370ba PR |
1944 | |
1945 | return rv; | |
1946 | } | |
1947 | ||
b411b363 PR |
1948 | /* Called from receive_Data. |
1949 | * Synchronize packets on sock with packets on msock. | |
1950 | * | |
1951 | * This is here so even when a P_DATA packet traveling via sock overtook an Ack | |
1952 | * packet traveling on msock, they are still processed in the order they have | |
1953 | * been sent. | |
1954 | * | |
1955 | * Note: we don't care for Ack packets overtaking P_DATA packets. | |
1956 | * | |
1957 | * In case packet_seq is larger than mdev->peer_seq number, there are | |
1958 | * outstanding packets on the msock. We wait for them to arrive. | |
1959 | * In case we are the logically next packet, we update mdev->peer_seq | |
1960 | * ourselves. Correctly handles 32bit wrap around. | |
1961 | * | |
1962 | * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, | |
1963 | * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds | |
1964 | * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have | |
1965 | * 1<<9 == 512 seconds aka ages for the 32bit wrap around... | |
1966 | * | |
1967 | * returns 0 if we may process the packet, | |
1968 | * -ERESTARTSYS if we were interrupted (by disconnect signal). */ | |
7be8da07 | 1969 | static int wait_for_and_update_peer_seq(struct drbd_conf *mdev, const u32 peer_seq) |
b411b363 PR |
1970 | { |
1971 | DEFINE_WAIT(wait); | |
b411b363 | 1972 | long timeout; |
7be8da07 AG |
1973 | int ret; |
1974 | ||
1975 | if (!need_peer_seq(mdev)) | |
1976 | return 0; | |
1977 | ||
b411b363 PR |
1978 | spin_lock(&mdev->peer_seq_lock); |
1979 | for (;;) { | |
7be8da07 AG |
1980 | if (!seq_greater(peer_seq - 1, mdev->peer_seq)) { |
1981 | mdev->peer_seq = seq_max(mdev->peer_seq, peer_seq); | |
1982 | ret = 0; | |
b411b363 | 1983 | break; |
7be8da07 | 1984 | } |
b411b363 PR |
1985 | if (signal_pending(current)) { |
1986 | ret = -ERESTARTSYS; | |
1987 | break; | |
1988 | } | |
7be8da07 | 1989 | prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE); |
b411b363 | 1990 | spin_unlock(&mdev->peer_seq_lock); |
44ed167d PR |
1991 | rcu_read_lock(); |
1992 | timeout = rcu_dereference(mdev->tconn->net_conf)->ping_timeo*HZ/10; | |
1993 | rcu_read_unlock(); | |
71b1c1eb | 1994 | timeout = schedule_timeout(timeout); |
b411b363 | 1995 | spin_lock(&mdev->peer_seq_lock); |
7be8da07 | 1996 | if (!timeout) { |
b411b363 | 1997 | ret = -ETIMEDOUT; |
71b1c1eb | 1998 | dev_err(DEV, "Timed out waiting for missing ack packets; disconnecting\n"); |
b411b363 PR |
1999 | break; |
2000 | } | |
2001 | } | |
b411b363 | 2002 | spin_unlock(&mdev->peer_seq_lock); |
7be8da07 | 2003 | finish_wait(&mdev->seq_wait, &wait); |
b411b363 PR |
2004 | return ret; |
2005 | } | |
2006 | ||
688593c5 LE |
2007 | /* see also bio_flags_to_wire() |
2008 | * DRBD_REQ_*, because we need to semantically map the flags to data packet | |
2009 | * flags and back. We may replicate to other kernel versions. */ | |
2010 | static unsigned long wire_flags_to_bio(struct drbd_conf *mdev, u32 dpf) | |
76d2e7ec | 2011 | { |
688593c5 LE |
2012 | return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | |
2013 | (dpf & DP_FUA ? REQ_FUA : 0) | | |
2014 | (dpf & DP_FLUSH ? REQ_FLUSH : 0) | | |
2015 | (dpf & DP_DISCARD ? REQ_DISCARD : 0); | |
76d2e7ec PR |
2016 | } |
2017 | ||
7be8da07 AG |
2018 | static void fail_postponed_requests(struct drbd_conf *mdev, sector_t sector, |
2019 | unsigned int size) | |
2020 | { | |
2021 | struct drbd_interval *i; | |
2022 | ||
2023 | repeat: | |
2024 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
2025 | struct drbd_request *req; | |
2026 | struct bio_and_error m; | |
2027 | ||
2028 | if (!i->local) | |
2029 | continue; | |
2030 | req = container_of(i, struct drbd_request, i); | |
2031 | if (!(req->rq_state & RQ_POSTPONED)) | |
2032 | continue; | |
2033 | req->rq_state &= ~RQ_POSTPONED; | |
2034 | __req_mod(req, NEG_ACKED, &m); | |
2035 | spin_unlock_irq(&mdev->tconn->req_lock); | |
2036 | if (m.bio) | |
2037 | complete_master_bio(mdev, &m); | |
2038 | spin_lock_irq(&mdev->tconn->req_lock); | |
2039 | goto repeat; | |
2040 | } | |
2041 | } | |
2042 | ||
2043 | static int handle_write_conflicts(struct drbd_conf *mdev, | |
2044 | struct drbd_peer_request *peer_req) | |
2045 | { | |
2046 | struct drbd_tconn *tconn = mdev->tconn; | |
427c0434 | 2047 | bool resolve_conflicts = test_bit(RESOLVE_CONFLICTS, &tconn->flags); |
7be8da07 AG |
2048 | sector_t sector = peer_req->i.sector; |
2049 | const unsigned int size = peer_req->i.size; | |
2050 | struct drbd_interval *i; | |
2051 | bool equal; | |
2052 | int err; | |
2053 | ||
2054 | /* | |
2055 | * Inserting the peer request into the write_requests tree will prevent | |
2056 | * new conflicting local requests from being added. | |
2057 | */ | |
2058 | drbd_insert_interval(&mdev->write_requests, &peer_req->i); | |
2059 | ||
2060 | repeat: | |
2061 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
2062 | if (i == &peer_req->i) | |
2063 | continue; | |
2064 | ||
2065 | if (!i->local) { | |
2066 | /* | |
2067 | * Our peer has sent a conflicting remote request; this | |
2068 | * should not happen in a two-node setup. Wait for the | |
2069 | * earlier peer request to complete. | |
2070 | */ | |
2071 | err = drbd_wait_misc(mdev, i); | |
2072 | if (err) | |
2073 | goto out; | |
2074 | goto repeat; | |
2075 | } | |
2076 | ||
2077 | equal = i->sector == sector && i->size == size; | |
2078 | if (resolve_conflicts) { | |
2079 | /* | |
2080 | * If the peer request is fully contained within the | |
d4dabbe2 LE |
2081 | * overlapping request, it can be considered overwritten |
2082 | * and thus superseded; otherwise, it will be retried | |
2083 | * once all overlapping requests have completed. | |
7be8da07 | 2084 | */ |
d4dabbe2 | 2085 | bool superseded = i->sector <= sector && i->sector + |
7be8da07 AG |
2086 | (i->size >> 9) >= sector + (size >> 9); |
2087 | ||
2088 | if (!equal) | |
2089 | dev_alert(DEV, "Concurrent writes detected: " | |
2090 | "local=%llus +%u, remote=%llus +%u, " | |
2091 | "assuming %s came first\n", | |
2092 | (unsigned long long)i->sector, i->size, | |
2093 | (unsigned long long)sector, size, | |
d4dabbe2 | 2094 | superseded ? "local" : "remote"); |
7be8da07 AG |
2095 | |
2096 | inc_unacked(mdev); | |
d4dabbe2 | 2097 | peer_req->w.cb = superseded ? e_send_superseded : |
7be8da07 AG |
2098 | e_send_retry_write; |
2099 | list_add_tail(&peer_req->w.list, &mdev->done_ee); | |
2100 | wake_asender(mdev->tconn); | |
2101 | ||
2102 | err = -ENOENT; | |
2103 | goto out; | |
2104 | } else { | |
2105 | struct drbd_request *req = | |
2106 | container_of(i, struct drbd_request, i); | |
2107 | ||
2108 | if (!equal) | |
2109 | dev_alert(DEV, "Concurrent writes detected: " | |
2110 | "local=%llus +%u, remote=%llus +%u\n", | |
2111 | (unsigned long long)i->sector, i->size, | |
2112 | (unsigned long long)sector, size); | |
2113 | ||
2114 | if (req->rq_state & RQ_LOCAL_PENDING || | |
2115 | !(req->rq_state & RQ_POSTPONED)) { | |
2116 | /* | |
2117 | * Wait for the node with the discard flag to | |
d4dabbe2 LE |
2118 | * decide if this request has been superseded |
2119 | * or needs to be retried. | |
2120 | * Requests that have been superseded will | |
7be8da07 AG |
2121 | * disappear from the write_requests tree. |
2122 | * | |
2123 | * In addition, wait for the conflicting | |
2124 | * request to finish locally before submitting | |
2125 | * the conflicting peer request. | |
2126 | */ | |
2127 | err = drbd_wait_misc(mdev, &req->i); | |
2128 | if (err) { | |
2129 | _conn_request_state(mdev->tconn, | |
2130 | NS(conn, C_TIMEOUT), | |
2131 | CS_HARD); | |
2132 | fail_postponed_requests(mdev, sector, size); | |
2133 | goto out; | |
2134 | } | |
2135 | goto repeat; | |
2136 | } | |
2137 | /* | |
2138 | * Remember to restart the conflicting requests after | |
2139 | * the new peer request has completed. | |
2140 | */ | |
2141 | peer_req->flags |= EE_RESTART_REQUESTS; | |
2142 | } | |
2143 | } | |
2144 | err = 0; | |
2145 | ||
2146 | out: | |
2147 | if (err) | |
2148 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
2149 | return err; | |
2150 | } | |
2151 | ||
b411b363 | 2152 | /* mirrored write */ |
4a76b161 | 2153 | static int receive_Data(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 2154 | { |
4a76b161 | 2155 | struct drbd_conf *mdev; |
b411b363 | 2156 | sector_t sector; |
db830c46 | 2157 | struct drbd_peer_request *peer_req; |
e658983a | 2158 | struct p_data *p = pi->data; |
7be8da07 | 2159 | u32 peer_seq = be32_to_cpu(p->seq_num); |
b411b363 PR |
2160 | int rw = WRITE; |
2161 | u32 dp_flags; | |
302bdeae | 2162 | int err, tp; |
b411b363 | 2163 | |
4a76b161 AG |
2164 | mdev = vnr_to_mdev(tconn, pi->vnr); |
2165 | if (!mdev) | |
2166 | return -EIO; | |
b411b363 | 2167 | |
7be8da07 | 2168 | if (!get_ldev(mdev)) { |
82bc0194 AG |
2169 | int err2; |
2170 | ||
7be8da07 | 2171 | err = wait_for_and_update_peer_seq(mdev, peer_seq); |
e2857216 | 2172 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, pi->size); |
12038a3a | 2173 | atomic_inc(&tconn->current_epoch->epoch_size); |
e2857216 | 2174 | err2 = drbd_drain_block(mdev, pi->size); |
82bc0194 AG |
2175 | if (!err) |
2176 | err = err2; | |
2177 | return err; | |
b411b363 PR |
2178 | } |
2179 | ||
fcefa62e AG |
2180 | /* |
2181 | * Corresponding put_ldev done either below (on various errors), or in | |
2182 | * drbd_peer_request_endio, if we successfully submit the data at the | |
2183 | * end of this function. | |
2184 | */ | |
b411b363 PR |
2185 | |
2186 | sector = be64_to_cpu(p->sector); | |
e2857216 | 2187 | peer_req = read_in_block(mdev, p->block_id, sector, pi->size); |
db830c46 | 2188 | if (!peer_req) { |
b411b363 | 2189 | put_ldev(mdev); |
82bc0194 | 2190 | return -EIO; |
b411b363 PR |
2191 | } |
2192 | ||
db830c46 | 2193 | peer_req->w.cb = e_end_block; |
b411b363 | 2194 | |
688593c5 LE |
2195 | dp_flags = be32_to_cpu(p->dp_flags); |
2196 | rw |= wire_flags_to_bio(mdev, dp_flags); | |
81a3537a LE |
2197 | if (peer_req->pages == NULL) { |
2198 | D_ASSERT(peer_req->i.size == 0); | |
a73ff323 LE |
2199 | D_ASSERT(dp_flags & DP_FLUSH); |
2200 | } | |
688593c5 LE |
2201 | |
2202 | if (dp_flags & DP_MAY_SET_IN_SYNC) | |
db830c46 | 2203 | peer_req->flags |= EE_MAY_SET_IN_SYNC; |
688593c5 | 2204 | |
12038a3a PR |
2205 | spin_lock(&tconn->epoch_lock); |
2206 | peer_req->epoch = tconn->current_epoch; | |
db830c46 AG |
2207 | atomic_inc(&peer_req->epoch->epoch_size); |
2208 | atomic_inc(&peer_req->epoch->active); | |
12038a3a | 2209 | spin_unlock(&tconn->epoch_lock); |
b411b363 | 2210 | |
302bdeae PR |
2211 | rcu_read_lock(); |
2212 | tp = rcu_dereference(mdev->tconn->net_conf)->two_primaries; | |
2213 | rcu_read_unlock(); | |
2214 | if (tp) { | |
2215 | peer_req->flags |= EE_IN_INTERVAL_TREE; | |
7be8da07 AG |
2216 | err = wait_for_and_update_peer_seq(mdev, peer_seq); |
2217 | if (err) | |
b411b363 | 2218 | goto out_interrupted; |
87eeee41 | 2219 | spin_lock_irq(&mdev->tconn->req_lock); |
7be8da07 AG |
2220 | err = handle_write_conflicts(mdev, peer_req); |
2221 | if (err) { | |
2222 | spin_unlock_irq(&mdev->tconn->req_lock); | |
2223 | if (err == -ENOENT) { | |
b411b363 | 2224 | put_ldev(mdev); |
82bc0194 | 2225 | return 0; |
b411b363 | 2226 | } |
7be8da07 | 2227 | goto out_interrupted; |
b411b363 | 2228 | } |
7be8da07 AG |
2229 | } else |
2230 | spin_lock_irq(&mdev->tconn->req_lock); | |
db830c46 | 2231 | list_add(&peer_req->w.list, &mdev->active_ee); |
87eeee41 | 2232 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 2233 | |
d93f6302 | 2234 | if (mdev->state.conn == C_SYNC_TARGET) |
3ea35df8 | 2235 | wait_event(mdev->ee_wait, !overlapping_resync_write(mdev, peer_req)); |
b411b363 | 2236 | |
303d1448 | 2237 | if (mdev->tconn->agreed_pro_version < 100) { |
44ed167d PR |
2238 | rcu_read_lock(); |
2239 | switch (rcu_dereference(mdev->tconn->net_conf)->wire_protocol) { | |
303d1448 PR |
2240 | case DRBD_PROT_C: |
2241 | dp_flags |= DP_SEND_WRITE_ACK; | |
2242 | break; | |
2243 | case DRBD_PROT_B: | |
2244 | dp_flags |= DP_SEND_RECEIVE_ACK; | |
2245 | break; | |
b411b363 | 2246 | } |
44ed167d | 2247 | rcu_read_unlock(); |
b411b363 PR |
2248 | } |
2249 | ||
303d1448 PR |
2250 | if (dp_flags & DP_SEND_WRITE_ACK) { |
2251 | peer_req->flags |= EE_SEND_WRITE_ACK; | |
b411b363 PR |
2252 | inc_unacked(mdev); |
2253 | /* corresponding dec_unacked() in e_end_block() | |
2254 | * respective _drbd_clear_done_ee */ | |
303d1448 PR |
2255 | } |
2256 | ||
2257 | if (dp_flags & DP_SEND_RECEIVE_ACK) { | |
b411b363 PR |
2258 | /* I really don't like it that the receiver thread |
2259 | * sends on the msock, but anyways */ | |
db830c46 | 2260 | drbd_send_ack(mdev, P_RECV_ACK, peer_req); |
b411b363 PR |
2261 | } |
2262 | ||
6719fb03 | 2263 | if (mdev->state.pdsk < D_INCONSISTENT) { |
b411b363 | 2264 | /* In case we have the only disk of the cluster, */ |
db830c46 AG |
2265 | drbd_set_out_of_sync(mdev, peer_req->i.sector, peer_req->i.size); |
2266 | peer_req->flags |= EE_CALL_AL_COMPLETE_IO; | |
2267 | peer_req->flags &= ~EE_MAY_SET_IN_SYNC; | |
181286ad | 2268 | drbd_al_begin_io(mdev, &peer_req->i); |
b411b363 PR |
2269 | } |
2270 | ||
82bc0194 AG |
2271 | err = drbd_submit_peer_request(mdev, peer_req, rw, DRBD_FAULT_DT_WR); |
2272 | if (!err) | |
2273 | return 0; | |
b411b363 | 2274 | |
10f6d992 LE |
2275 | /* don't care for the reason here */ |
2276 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 2277 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 AG |
2278 | list_del(&peer_req->w.list); |
2279 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
87eeee41 | 2280 | spin_unlock_irq(&mdev->tconn->req_lock); |
db830c46 | 2281 | if (peer_req->flags & EE_CALL_AL_COMPLETE_IO) |
181286ad | 2282 | drbd_al_complete_io(mdev, &peer_req->i); |
22cc37a9 | 2283 | |
b411b363 | 2284 | out_interrupted: |
1e9dd291 | 2285 | drbd_may_finish_epoch(tconn, peer_req->epoch, EV_PUT + EV_CLEANUP); |
b411b363 | 2286 | put_ldev(mdev); |
3967deb1 | 2287 | drbd_free_peer_req(mdev, peer_req); |
82bc0194 | 2288 | return err; |
b411b363 PR |
2289 | } |
2290 | ||
0f0601f4 LE |
2291 | /* We may throttle resync, if the lower device seems to be busy, |
2292 | * and current sync rate is above c_min_rate. | |
2293 | * | |
2294 | * To decide whether or not the lower device is busy, we use a scheme similar | |
2295 | * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" | |
2296 | * (more than 64 sectors) of activity we cannot account for with our own resync | |
2297 | * activity, it obviously is "busy". | |
2298 | * | |
2299 | * The current sync rate used here uses only the most recent two step marks, | |
2300 | * to have a short time average so we can react faster. | |
2301 | */ | |
e3555d85 | 2302 | int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector) |
0f0601f4 LE |
2303 | { |
2304 | struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk; | |
2305 | unsigned long db, dt, dbdt; | |
e3555d85 | 2306 | struct lc_element *tmp; |
0f0601f4 LE |
2307 | int curr_events; |
2308 | int throttle = 0; | |
daeda1cc PR |
2309 | unsigned int c_min_rate; |
2310 | ||
2311 | rcu_read_lock(); | |
2312 | c_min_rate = rcu_dereference(mdev->ldev->disk_conf)->c_min_rate; | |
2313 | rcu_read_unlock(); | |
0f0601f4 LE |
2314 | |
2315 | /* feature disabled? */ | |
daeda1cc | 2316 | if (c_min_rate == 0) |
0f0601f4 LE |
2317 | return 0; |
2318 | ||
e3555d85 PR |
2319 | spin_lock_irq(&mdev->al_lock); |
2320 | tmp = lc_find(mdev->resync, BM_SECT_TO_EXT(sector)); | |
2321 | if (tmp) { | |
2322 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
2323 | if (test_bit(BME_PRIORITY, &bm_ext->flags)) { | |
2324 | spin_unlock_irq(&mdev->al_lock); | |
2325 | return 0; | |
2326 | } | |
2327 | /* Do not slow down if app IO is already waiting for this extent */ | |
2328 | } | |
2329 | spin_unlock_irq(&mdev->al_lock); | |
2330 | ||
0f0601f4 LE |
2331 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
2332 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
2333 | atomic_read(&mdev->rs_sect_ev); | |
e3555d85 | 2334 | |
0f0601f4 LE |
2335 | if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) { |
2336 | unsigned long rs_left; | |
2337 | int i; | |
2338 | ||
2339 | mdev->rs_last_events = curr_events; | |
2340 | ||
2341 | /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, | |
2342 | * approx. */ | |
2649f080 LE |
2343 | i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS; |
2344 | ||
2345 | if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T) | |
2346 | rs_left = mdev->ov_left; | |
2347 | else | |
2348 | rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; | |
0f0601f4 LE |
2349 | |
2350 | dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ; | |
2351 | if (!dt) | |
2352 | dt++; | |
2353 | db = mdev->rs_mark_left[i] - rs_left; | |
2354 | dbdt = Bit2KB(db/dt); | |
2355 | ||
daeda1cc | 2356 | if (dbdt > c_min_rate) |
0f0601f4 LE |
2357 | throttle = 1; |
2358 | } | |
2359 | return throttle; | |
2360 | } | |
2361 | ||
2362 | ||
4a76b161 | 2363 | static int receive_DataRequest(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 2364 | { |
4a76b161 | 2365 | struct drbd_conf *mdev; |
b411b363 | 2366 | sector_t sector; |
4a76b161 | 2367 | sector_t capacity; |
db830c46 | 2368 | struct drbd_peer_request *peer_req; |
b411b363 | 2369 | struct digest_info *di = NULL; |
b18b37be | 2370 | int size, verb; |
b411b363 | 2371 | unsigned int fault_type; |
e658983a | 2372 | struct p_block_req *p = pi->data; |
4a76b161 AG |
2373 | |
2374 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
2375 | if (!mdev) | |
2376 | return -EIO; | |
2377 | capacity = drbd_get_capacity(mdev->this_bdev); | |
b411b363 PR |
2378 | |
2379 | sector = be64_to_cpu(p->sector); | |
2380 | size = be32_to_cpu(p->blksize); | |
2381 | ||
c670a398 | 2382 | if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { |
b411b363 PR |
2383 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, |
2384 | (unsigned long long)sector, size); | |
82bc0194 | 2385 | return -EINVAL; |
b411b363 PR |
2386 | } |
2387 | if (sector + (size>>9) > capacity) { | |
2388 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, | |
2389 | (unsigned long long)sector, size); | |
82bc0194 | 2390 | return -EINVAL; |
b411b363 PR |
2391 | } |
2392 | ||
2393 | if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) { | |
b18b37be | 2394 | verb = 1; |
e2857216 | 2395 | switch (pi->cmd) { |
b18b37be PR |
2396 | case P_DATA_REQUEST: |
2397 | drbd_send_ack_rp(mdev, P_NEG_DREPLY, p); | |
2398 | break; | |
2399 | case P_RS_DATA_REQUEST: | |
2400 | case P_CSUM_RS_REQUEST: | |
2401 | case P_OV_REQUEST: | |
2402 | drbd_send_ack_rp(mdev, P_NEG_RS_DREPLY , p); | |
2403 | break; | |
2404 | case P_OV_REPLY: | |
2405 | verb = 0; | |
2406 | dec_rs_pending(mdev); | |
2407 | drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, ID_IN_SYNC); | |
2408 | break; | |
2409 | default: | |
49ba9b1b | 2410 | BUG(); |
b18b37be PR |
2411 | } |
2412 | if (verb && __ratelimit(&drbd_ratelimit_state)) | |
b411b363 PR |
2413 | dev_err(DEV, "Can not satisfy peer's read request, " |
2414 | "no local data.\n"); | |
b18b37be | 2415 | |
a821cc4a | 2416 | /* drain possibly payload */ |
e2857216 | 2417 | return drbd_drain_block(mdev, pi->size); |
b411b363 PR |
2418 | } |
2419 | ||
2420 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD | |
2421 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
2422 | * which in turn might block on the other node at this very place. */ | |
0db55363 | 2423 | peer_req = drbd_alloc_peer_req(mdev, p->block_id, sector, size, GFP_NOIO); |
db830c46 | 2424 | if (!peer_req) { |
b411b363 | 2425 | put_ldev(mdev); |
82bc0194 | 2426 | return -ENOMEM; |
b411b363 PR |
2427 | } |
2428 | ||
e2857216 | 2429 | switch (pi->cmd) { |
b411b363 | 2430 | case P_DATA_REQUEST: |
db830c46 | 2431 | peer_req->w.cb = w_e_end_data_req; |
b411b363 | 2432 | fault_type = DRBD_FAULT_DT_RD; |
80a40e43 LE |
2433 | /* application IO, don't drbd_rs_begin_io */ |
2434 | goto submit; | |
2435 | ||
b411b363 | 2436 | case P_RS_DATA_REQUEST: |
db830c46 | 2437 | peer_req->w.cb = w_e_end_rsdata_req; |
b411b363 | 2438 | fault_type = DRBD_FAULT_RS_RD; |
5f9915bb LE |
2439 | /* used in the sector offset progress display */ |
2440 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
b411b363 PR |
2441 | break; |
2442 | ||
2443 | case P_OV_REPLY: | |
2444 | case P_CSUM_RS_REQUEST: | |
2445 | fault_type = DRBD_FAULT_RS_RD; | |
e2857216 | 2446 | di = kmalloc(sizeof(*di) + pi->size, GFP_NOIO); |
b411b363 PR |
2447 | if (!di) |
2448 | goto out_free_e; | |
2449 | ||
e2857216 | 2450 | di->digest_size = pi->size; |
b411b363 PR |
2451 | di->digest = (((char *)di)+sizeof(struct digest_info)); |
2452 | ||
db830c46 AG |
2453 | peer_req->digest = di; |
2454 | peer_req->flags |= EE_HAS_DIGEST; | |
c36c3ced | 2455 | |
e2857216 | 2456 | if (drbd_recv_all(mdev->tconn, di->digest, pi->size)) |
b411b363 PR |
2457 | goto out_free_e; |
2458 | ||
e2857216 | 2459 | if (pi->cmd == P_CSUM_RS_REQUEST) { |
31890f4a | 2460 | D_ASSERT(mdev->tconn->agreed_pro_version >= 89); |
db830c46 | 2461 | peer_req->w.cb = w_e_end_csum_rs_req; |
5f9915bb LE |
2462 | /* used in the sector offset progress display */ |
2463 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
e2857216 | 2464 | } else if (pi->cmd == P_OV_REPLY) { |
2649f080 LE |
2465 | /* track progress, we may need to throttle */ |
2466 | atomic_add(size >> 9, &mdev->rs_sect_in); | |
db830c46 | 2467 | peer_req->w.cb = w_e_end_ov_reply; |
b411b363 | 2468 | dec_rs_pending(mdev); |
0f0601f4 LE |
2469 | /* drbd_rs_begin_io done when we sent this request, |
2470 | * but accounting still needs to be done. */ | |
2471 | goto submit_for_resync; | |
b411b363 PR |
2472 | } |
2473 | break; | |
2474 | ||
2475 | case P_OV_REQUEST: | |
b411b363 | 2476 | if (mdev->ov_start_sector == ~(sector_t)0 && |
31890f4a | 2477 | mdev->tconn->agreed_pro_version >= 90) { |
de228bba LE |
2478 | unsigned long now = jiffies; |
2479 | int i; | |
b411b363 PR |
2480 | mdev->ov_start_sector = sector; |
2481 | mdev->ov_position = sector; | |
30b743a2 LE |
2482 | mdev->ov_left = drbd_bm_bits(mdev) - BM_SECT_TO_BIT(sector); |
2483 | mdev->rs_total = mdev->ov_left; | |
de228bba LE |
2484 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
2485 | mdev->rs_mark_left[i] = mdev->ov_left; | |
2486 | mdev->rs_mark_time[i] = now; | |
2487 | } | |
b411b363 PR |
2488 | dev_info(DEV, "Online Verify start sector: %llu\n", |
2489 | (unsigned long long)sector); | |
2490 | } | |
db830c46 | 2491 | peer_req->w.cb = w_e_end_ov_req; |
b411b363 | 2492 | fault_type = DRBD_FAULT_RS_RD; |
b411b363 PR |
2493 | break; |
2494 | ||
b411b363 | 2495 | default: |
49ba9b1b | 2496 | BUG(); |
b411b363 PR |
2497 | } |
2498 | ||
0f0601f4 LE |
2499 | /* Throttle, drbd_rs_begin_io and submit should become asynchronous |
2500 | * wrt the receiver, but it is not as straightforward as it may seem. | |
2501 | * Various places in the resync start and stop logic assume resync | |
2502 | * requests are processed in order, requeuing this on the worker thread | |
2503 | * introduces a bunch of new code for synchronization between threads. | |
2504 | * | |
2505 | * Unlimited throttling before drbd_rs_begin_io may stall the resync | |
2506 | * "forever", throttling after drbd_rs_begin_io will lock that extent | |
2507 | * for application writes for the same time. For now, just throttle | |
2508 | * here, where the rest of the code expects the receiver to sleep for | |
2509 | * a while, anyways. | |
2510 | */ | |
2511 | ||
2512 | /* Throttle before drbd_rs_begin_io, as that locks out application IO; | |
2513 | * this defers syncer requests for some time, before letting at least | |
2514 | * on request through. The resync controller on the receiving side | |
2515 | * will adapt to the incoming rate accordingly. | |
2516 | * | |
2517 | * We cannot throttle here if remote is Primary/SyncTarget: | |
2518 | * we would also throttle its application reads. | |
2519 | * In that case, throttling is done on the SyncTarget only. | |
2520 | */ | |
e3555d85 PR |
2521 | if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev, sector)) |
2522 | schedule_timeout_uninterruptible(HZ/10); | |
2523 | if (drbd_rs_begin_io(mdev, sector)) | |
80a40e43 | 2524 | goto out_free_e; |
b411b363 | 2525 | |
0f0601f4 LE |
2526 | submit_for_resync: |
2527 | atomic_add(size >> 9, &mdev->rs_sect_ev); | |
2528 | ||
80a40e43 | 2529 | submit: |
b411b363 | 2530 | inc_unacked(mdev); |
87eeee41 | 2531 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 2532 | list_add_tail(&peer_req->w.list, &mdev->read_ee); |
87eeee41 | 2533 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 2534 | |
fbe29dec | 2535 | if (drbd_submit_peer_request(mdev, peer_req, READ, fault_type) == 0) |
82bc0194 | 2536 | return 0; |
b411b363 | 2537 | |
10f6d992 LE |
2538 | /* don't care for the reason here */ |
2539 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 2540 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 2541 | list_del(&peer_req->w.list); |
87eeee41 | 2542 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 LE |
2543 | /* no drbd_rs_complete_io(), we are dropping the connection anyways */ |
2544 | ||
b411b363 | 2545 | out_free_e: |
b411b363 | 2546 | put_ldev(mdev); |
3967deb1 | 2547 | drbd_free_peer_req(mdev, peer_req); |
82bc0194 | 2548 | return -EIO; |
b411b363 PR |
2549 | } |
2550 | ||
2551 | static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local) | |
2552 | { | |
2553 | int self, peer, rv = -100; | |
2554 | unsigned long ch_self, ch_peer; | |
44ed167d | 2555 | enum drbd_after_sb_p after_sb_0p; |
b411b363 PR |
2556 | |
2557 | self = mdev->ldev->md.uuid[UI_BITMAP] & 1; | |
2558 | peer = mdev->p_uuid[UI_BITMAP] & 1; | |
2559 | ||
2560 | ch_peer = mdev->p_uuid[UI_SIZE]; | |
2561 | ch_self = mdev->comm_bm_set; | |
2562 | ||
44ed167d PR |
2563 | rcu_read_lock(); |
2564 | after_sb_0p = rcu_dereference(mdev->tconn->net_conf)->after_sb_0p; | |
2565 | rcu_read_unlock(); | |
2566 | switch (after_sb_0p) { | |
b411b363 PR |
2567 | case ASB_CONSENSUS: |
2568 | case ASB_DISCARD_SECONDARY: | |
2569 | case ASB_CALL_HELPER: | |
44ed167d | 2570 | case ASB_VIOLENTLY: |
b411b363 PR |
2571 | dev_err(DEV, "Configuration error.\n"); |
2572 | break; | |
2573 | case ASB_DISCONNECT: | |
2574 | break; | |
2575 | case ASB_DISCARD_YOUNGER_PRI: | |
2576 | if (self == 0 && peer == 1) { | |
2577 | rv = -1; | |
2578 | break; | |
2579 | } | |
2580 | if (self == 1 && peer == 0) { | |
2581 | rv = 1; | |
2582 | break; | |
2583 | } | |
2584 | /* Else fall through to one of the other strategies... */ | |
2585 | case ASB_DISCARD_OLDER_PRI: | |
2586 | if (self == 0 && peer == 1) { | |
2587 | rv = 1; | |
2588 | break; | |
2589 | } | |
2590 | if (self == 1 && peer == 0) { | |
2591 | rv = -1; | |
2592 | break; | |
2593 | } | |
2594 | /* Else fall through to one of the other strategies... */ | |
ad19bf6e | 2595 | dev_warn(DEV, "Discard younger/older primary did not find a decision\n" |
b411b363 PR |
2596 | "Using discard-least-changes instead\n"); |
2597 | case ASB_DISCARD_ZERO_CHG: | |
2598 | if (ch_peer == 0 && ch_self == 0) { | |
427c0434 | 2599 | rv = test_bit(RESOLVE_CONFLICTS, &mdev->tconn->flags) |
b411b363 PR |
2600 | ? -1 : 1; |
2601 | break; | |
2602 | } else { | |
2603 | if (ch_peer == 0) { rv = 1; break; } | |
2604 | if (ch_self == 0) { rv = -1; break; } | |
2605 | } | |
44ed167d | 2606 | if (after_sb_0p == ASB_DISCARD_ZERO_CHG) |
b411b363 PR |
2607 | break; |
2608 | case ASB_DISCARD_LEAST_CHG: | |
2609 | if (ch_self < ch_peer) | |
2610 | rv = -1; | |
2611 | else if (ch_self > ch_peer) | |
2612 | rv = 1; | |
2613 | else /* ( ch_self == ch_peer ) */ | |
2614 | /* Well, then use something else. */ | |
427c0434 | 2615 | rv = test_bit(RESOLVE_CONFLICTS, &mdev->tconn->flags) |
b411b363 PR |
2616 | ? -1 : 1; |
2617 | break; | |
2618 | case ASB_DISCARD_LOCAL: | |
2619 | rv = -1; | |
2620 | break; | |
2621 | case ASB_DISCARD_REMOTE: | |
2622 | rv = 1; | |
2623 | } | |
2624 | ||
2625 | return rv; | |
2626 | } | |
2627 | ||
2628 | static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local) | |
2629 | { | |
6184ea21 | 2630 | int hg, rv = -100; |
44ed167d | 2631 | enum drbd_after_sb_p after_sb_1p; |
b411b363 | 2632 | |
44ed167d PR |
2633 | rcu_read_lock(); |
2634 | after_sb_1p = rcu_dereference(mdev->tconn->net_conf)->after_sb_1p; | |
2635 | rcu_read_unlock(); | |
2636 | switch (after_sb_1p) { | |
b411b363 PR |
2637 | case ASB_DISCARD_YOUNGER_PRI: |
2638 | case ASB_DISCARD_OLDER_PRI: | |
2639 | case ASB_DISCARD_LEAST_CHG: | |
2640 | case ASB_DISCARD_LOCAL: | |
2641 | case ASB_DISCARD_REMOTE: | |
44ed167d | 2642 | case ASB_DISCARD_ZERO_CHG: |
b411b363 PR |
2643 | dev_err(DEV, "Configuration error.\n"); |
2644 | break; | |
2645 | case ASB_DISCONNECT: | |
2646 | break; | |
2647 | case ASB_CONSENSUS: | |
2648 | hg = drbd_asb_recover_0p(mdev); | |
2649 | if (hg == -1 && mdev->state.role == R_SECONDARY) | |
2650 | rv = hg; | |
2651 | if (hg == 1 && mdev->state.role == R_PRIMARY) | |
2652 | rv = hg; | |
2653 | break; | |
2654 | case ASB_VIOLENTLY: | |
2655 | rv = drbd_asb_recover_0p(mdev); | |
2656 | break; | |
2657 | case ASB_DISCARD_SECONDARY: | |
2658 | return mdev->state.role == R_PRIMARY ? 1 : -1; | |
2659 | case ASB_CALL_HELPER: | |
2660 | hg = drbd_asb_recover_0p(mdev); | |
2661 | if (hg == -1 && mdev->state.role == R_PRIMARY) { | |
bb437946 AG |
2662 | enum drbd_state_rv rv2; |
2663 | ||
2664 | drbd_set_role(mdev, R_SECONDARY, 0); | |
b411b363 PR |
2665 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2666 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2667 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2668 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2669 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2670 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2671 | } else { | |
2672 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2673 | rv = hg; | |
2674 | } | |
2675 | } else | |
2676 | rv = hg; | |
2677 | } | |
2678 | ||
2679 | return rv; | |
2680 | } | |
2681 | ||
2682 | static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local) | |
2683 | { | |
6184ea21 | 2684 | int hg, rv = -100; |
44ed167d | 2685 | enum drbd_after_sb_p after_sb_2p; |
b411b363 | 2686 | |
44ed167d PR |
2687 | rcu_read_lock(); |
2688 | after_sb_2p = rcu_dereference(mdev->tconn->net_conf)->after_sb_2p; | |
2689 | rcu_read_unlock(); | |
2690 | switch (after_sb_2p) { | |
b411b363 PR |
2691 | case ASB_DISCARD_YOUNGER_PRI: |
2692 | case ASB_DISCARD_OLDER_PRI: | |
2693 | case ASB_DISCARD_LEAST_CHG: | |
2694 | case ASB_DISCARD_LOCAL: | |
2695 | case ASB_DISCARD_REMOTE: | |
2696 | case ASB_CONSENSUS: | |
2697 | case ASB_DISCARD_SECONDARY: | |
44ed167d | 2698 | case ASB_DISCARD_ZERO_CHG: |
b411b363 PR |
2699 | dev_err(DEV, "Configuration error.\n"); |
2700 | break; | |
2701 | case ASB_VIOLENTLY: | |
2702 | rv = drbd_asb_recover_0p(mdev); | |
2703 | break; | |
2704 | case ASB_DISCONNECT: | |
2705 | break; | |
2706 | case ASB_CALL_HELPER: | |
2707 | hg = drbd_asb_recover_0p(mdev); | |
2708 | if (hg == -1) { | |
bb437946 AG |
2709 | enum drbd_state_rv rv2; |
2710 | ||
b411b363 PR |
2711 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2712 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2713 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2714 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2715 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2716 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2717 | } else { | |
2718 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2719 | rv = hg; | |
2720 | } | |
2721 | } else | |
2722 | rv = hg; | |
2723 | } | |
2724 | ||
2725 | return rv; | |
2726 | } | |
2727 | ||
2728 | static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid, | |
2729 | u64 bits, u64 flags) | |
2730 | { | |
2731 | if (!uuid) { | |
2732 | dev_info(DEV, "%s uuid info vanished while I was looking!\n", text); | |
2733 | return; | |
2734 | } | |
2735 | dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", | |
2736 | text, | |
2737 | (unsigned long long)uuid[UI_CURRENT], | |
2738 | (unsigned long long)uuid[UI_BITMAP], | |
2739 | (unsigned long long)uuid[UI_HISTORY_START], | |
2740 | (unsigned long long)uuid[UI_HISTORY_END], | |
2741 | (unsigned long long)bits, | |
2742 | (unsigned long long)flags); | |
2743 | } | |
2744 | ||
2745 | /* | |
2746 | 100 after split brain try auto recover | |
2747 | 2 C_SYNC_SOURCE set BitMap | |
2748 | 1 C_SYNC_SOURCE use BitMap | |
2749 | 0 no Sync | |
2750 | -1 C_SYNC_TARGET use BitMap | |
2751 | -2 C_SYNC_TARGET set BitMap | |
2752 | -100 after split brain, disconnect | |
2753 | -1000 unrelated data | |
4a23f264 PR |
2754 | -1091 requires proto 91 |
2755 | -1096 requires proto 96 | |
b411b363 PR |
2756 | */ |
2757 | static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local) | |
2758 | { | |
2759 | u64 self, peer; | |
2760 | int i, j; | |
2761 | ||
2762 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2763 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2764 | ||
2765 | *rule_nr = 10; | |
2766 | if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) | |
2767 | return 0; | |
2768 | ||
2769 | *rule_nr = 20; | |
2770 | if ((self == UUID_JUST_CREATED || self == (u64)0) && | |
2771 | peer != UUID_JUST_CREATED) | |
2772 | return -2; | |
2773 | ||
2774 | *rule_nr = 30; | |
2775 | if (self != UUID_JUST_CREATED && | |
2776 | (peer == UUID_JUST_CREATED || peer == (u64)0)) | |
2777 | return 2; | |
2778 | ||
2779 | if (self == peer) { | |
2780 | int rct, dc; /* roles at crash time */ | |
2781 | ||
2782 | if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) { | |
2783 | ||
31890f4a | 2784 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2785 | return -1091; |
b411b363 PR |
2786 | |
2787 | if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) && | |
2788 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { | |
2789 | dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n"); | |
9f2247bb PR |
2790 | drbd_uuid_move_history(mdev); |
2791 | mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[UI_BITMAP]; | |
2792 | mdev->ldev->md.uuid[UI_BITMAP] = 0; | |
b411b363 PR |
2793 | |
2794 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, | |
2795 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2796 | *rule_nr = 34; | |
2797 | } else { | |
2798 | dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n"); | |
2799 | *rule_nr = 36; | |
2800 | } | |
2801 | ||
2802 | return 1; | |
2803 | } | |
2804 | ||
2805 | if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) { | |
2806 | ||
31890f4a | 2807 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2808 | return -1091; |
b411b363 PR |
2809 | |
2810 | if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) && | |
2811 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) { | |
2812 | dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); | |
2813 | ||
2814 | mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START]; | |
2815 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP]; | |
2816 | mdev->p_uuid[UI_BITMAP] = 0UL; | |
2817 | ||
2818 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2819 | *rule_nr = 35; | |
2820 | } else { | |
2821 | dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n"); | |
2822 | *rule_nr = 37; | |
2823 | } | |
2824 | ||
2825 | return -1; | |
2826 | } | |
2827 | ||
2828 | /* Common power [off|failure] */ | |
2829 | rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) + | |
2830 | (mdev->p_uuid[UI_FLAGS] & 2); | |
2831 | /* lowest bit is set when we were primary, | |
2832 | * next bit (weight 2) is set when peer was primary */ | |
2833 | *rule_nr = 40; | |
2834 | ||
2835 | switch (rct) { | |
2836 | case 0: /* !self_pri && !peer_pri */ return 0; | |
2837 | case 1: /* self_pri && !peer_pri */ return 1; | |
2838 | case 2: /* !self_pri && peer_pri */ return -1; | |
2839 | case 3: /* self_pri && peer_pri */ | |
427c0434 | 2840 | dc = test_bit(RESOLVE_CONFLICTS, &mdev->tconn->flags); |
b411b363 PR |
2841 | return dc ? -1 : 1; |
2842 | } | |
2843 | } | |
2844 | ||
2845 | *rule_nr = 50; | |
2846 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2847 | if (self == peer) | |
2848 | return -1; | |
2849 | ||
2850 | *rule_nr = 51; | |
2851 | peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); | |
2852 | if (self == peer) { | |
31890f4a | 2853 | if (mdev->tconn->agreed_pro_version < 96 ? |
4a23f264 PR |
2854 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == |
2855 | (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) : | |
2856 | peer + UUID_NEW_BM_OFFSET == (mdev->p_uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2857 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2858 | resync as sync source modifications of the peer's UUIDs. */ | |
2859 | ||
31890f4a | 2860 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2861 | return -1091; |
b411b363 PR |
2862 | |
2863 | mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START]; | |
2864 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1]; | |
4a23f264 | 2865 | |
92b4ca29 | 2866 | dev_info(DEV, "Lost last syncUUID packet, corrected:\n"); |
4a23f264 PR |
2867 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); |
2868 | ||
b411b363 PR |
2869 | return -1; |
2870 | } | |
2871 | } | |
2872 | ||
2873 | *rule_nr = 60; | |
2874 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2875 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2876 | peer = mdev->p_uuid[i] & ~((u64)1); | |
2877 | if (self == peer) | |
2878 | return -2; | |
2879 | } | |
2880 | ||
2881 | *rule_nr = 70; | |
2882 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2883 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2884 | if (self == peer) | |
2885 | return 1; | |
2886 | ||
2887 | *rule_nr = 71; | |
2888 | self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); | |
2889 | if (self == peer) { | |
31890f4a | 2890 | if (mdev->tconn->agreed_pro_version < 96 ? |
4a23f264 PR |
2891 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == |
2892 | (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) : | |
2893 | self + UUID_NEW_BM_OFFSET == (mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2894 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2895 | resync as sync source modifications of our UUIDs. */ | |
2896 | ||
31890f4a | 2897 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2898 | return -1091; |
b411b363 | 2899 | |
9f2247bb PR |
2900 | __drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]); |
2901 | __drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]); | |
b411b363 | 2902 | |
4a23f264 | 2903 | dev_info(DEV, "Last syncUUID did not get through, corrected:\n"); |
b411b363 PR |
2904 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, |
2905 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2906 | ||
2907 | return 1; | |
2908 | } | |
2909 | } | |
2910 | ||
2911 | ||
2912 | *rule_nr = 80; | |
d8c2a36b | 2913 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); |
b411b363 PR |
2914 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
2915 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2916 | if (self == peer) | |
2917 | return 2; | |
2918 | } | |
2919 | ||
2920 | *rule_nr = 90; | |
2921 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2922 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2923 | if (self == peer && self != ((u64)0)) | |
2924 | return 100; | |
2925 | ||
2926 | *rule_nr = 100; | |
2927 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2928 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2929 | for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { | |
2930 | peer = mdev->p_uuid[j] & ~((u64)1); | |
2931 | if (self == peer) | |
2932 | return -100; | |
2933 | } | |
2934 | } | |
2935 | ||
2936 | return -1000; | |
2937 | } | |
2938 | ||
2939 | /* drbd_sync_handshake() returns the new conn state on success, or | |
2940 | CONN_MASK (-1) on failure. | |
2941 | */ | |
2942 | static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role, | |
2943 | enum drbd_disk_state peer_disk) __must_hold(local) | |
2944 | { | |
b411b363 PR |
2945 | enum drbd_conns rv = C_MASK; |
2946 | enum drbd_disk_state mydisk; | |
44ed167d | 2947 | struct net_conf *nc; |
6dff2902 | 2948 | int hg, rule_nr, rr_conflict, tentative; |
b411b363 PR |
2949 | |
2950 | mydisk = mdev->state.disk; | |
2951 | if (mydisk == D_NEGOTIATING) | |
2952 | mydisk = mdev->new_state_tmp.disk; | |
2953 | ||
2954 | dev_info(DEV, "drbd_sync_handshake:\n"); | |
9f2247bb PR |
2955 | |
2956 | spin_lock_irq(&mdev->ldev->md.uuid_lock); | |
b411b363 PR |
2957 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0); |
2958 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, | |
2959 | mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2960 | ||
2961 | hg = drbd_uuid_compare(mdev, &rule_nr); | |
9f2247bb | 2962 | spin_unlock_irq(&mdev->ldev->md.uuid_lock); |
b411b363 PR |
2963 | |
2964 | dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr); | |
2965 | ||
2966 | if (hg == -1000) { | |
2967 | dev_alert(DEV, "Unrelated data, aborting!\n"); | |
2968 | return C_MASK; | |
2969 | } | |
4a23f264 PR |
2970 | if (hg < -1000) { |
2971 | dev_alert(DEV, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000); | |
b411b363 PR |
2972 | return C_MASK; |
2973 | } | |
2974 | ||
2975 | if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || | |
2976 | (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { | |
2977 | int f = (hg == -100) || abs(hg) == 2; | |
2978 | hg = mydisk > D_INCONSISTENT ? 1 : -1; | |
2979 | if (f) | |
2980 | hg = hg*2; | |
2981 | dev_info(DEV, "Becoming sync %s due to disk states.\n", | |
2982 | hg > 0 ? "source" : "target"); | |
2983 | } | |
2984 | ||
3a11a487 AG |
2985 | if (abs(hg) == 100) |
2986 | drbd_khelper(mdev, "initial-split-brain"); | |
2987 | ||
44ed167d PR |
2988 | rcu_read_lock(); |
2989 | nc = rcu_dereference(mdev->tconn->net_conf); | |
2990 | ||
2991 | if (hg == 100 || (hg == -100 && nc->always_asbp)) { | |
b411b363 PR |
2992 | int pcount = (mdev->state.role == R_PRIMARY) |
2993 | + (peer_role == R_PRIMARY); | |
2994 | int forced = (hg == -100); | |
2995 | ||
2996 | switch (pcount) { | |
2997 | case 0: | |
2998 | hg = drbd_asb_recover_0p(mdev); | |
2999 | break; | |
3000 | case 1: | |
3001 | hg = drbd_asb_recover_1p(mdev); | |
3002 | break; | |
3003 | case 2: | |
3004 | hg = drbd_asb_recover_2p(mdev); | |
3005 | break; | |
3006 | } | |
3007 | if (abs(hg) < 100) { | |
3008 | dev_warn(DEV, "Split-Brain detected, %d primaries, " | |
3009 | "automatically solved. Sync from %s node\n", | |
3010 | pcount, (hg < 0) ? "peer" : "this"); | |
3011 | if (forced) { | |
3012 | dev_warn(DEV, "Doing a full sync, since" | |
3013 | " UUIDs where ambiguous.\n"); | |
3014 | hg = hg*2; | |
3015 | } | |
3016 | } | |
3017 | } | |
3018 | ||
3019 | if (hg == -100) { | |
08b165ba | 3020 | if (test_bit(DISCARD_MY_DATA, &mdev->flags) && !(mdev->p_uuid[UI_FLAGS]&1)) |
b411b363 | 3021 | hg = -1; |
08b165ba | 3022 | if (!test_bit(DISCARD_MY_DATA, &mdev->flags) && (mdev->p_uuid[UI_FLAGS]&1)) |
b411b363 PR |
3023 | hg = 1; |
3024 | ||
3025 | if (abs(hg) < 100) | |
3026 | dev_warn(DEV, "Split-Brain detected, manually solved. " | |
3027 | "Sync from %s node\n", | |
3028 | (hg < 0) ? "peer" : "this"); | |
3029 | } | |
44ed167d | 3030 | rr_conflict = nc->rr_conflict; |
6dff2902 | 3031 | tentative = nc->tentative; |
44ed167d | 3032 | rcu_read_unlock(); |
b411b363 PR |
3033 | |
3034 | if (hg == -100) { | |
580b9767 LE |
3035 | /* FIXME this log message is not correct if we end up here |
3036 | * after an attempted attach on a diskless node. | |
3037 | * We just refuse to attach -- well, we drop the "connection" | |
3038 | * to that disk, in a way... */ | |
3a11a487 | 3039 | dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n"); |
b411b363 PR |
3040 | drbd_khelper(mdev, "split-brain"); |
3041 | return C_MASK; | |
3042 | } | |
3043 | ||
3044 | if (hg > 0 && mydisk <= D_INCONSISTENT) { | |
3045 | dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n"); | |
3046 | return C_MASK; | |
3047 | } | |
3048 | ||
3049 | if (hg < 0 && /* by intention we do not use mydisk here. */ | |
3050 | mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) { | |
44ed167d | 3051 | switch (rr_conflict) { |
b411b363 PR |
3052 | case ASB_CALL_HELPER: |
3053 | drbd_khelper(mdev, "pri-lost"); | |
3054 | /* fall through */ | |
3055 | case ASB_DISCONNECT: | |
3056 | dev_err(DEV, "I shall become SyncTarget, but I am primary!\n"); | |
3057 | return C_MASK; | |
3058 | case ASB_VIOLENTLY: | |
3059 | dev_warn(DEV, "Becoming SyncTarget, violating the stable-data" | |
3060 | "assumption\n"); | |
3061 | } | |
3062 | } | |
3063 | ||
6dff2902 | 3064 | if (tentative || test_bit(CONN_DRY_RUN, &mdev->tconn->flags)) { |
cf14c2e9 PR |
3065 | if (hg == 0) |
3066 | dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n"); | |
3067 | else | |
3068 | dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.", | |
3069 | drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), | |
3070 | abs(hg) >= 2 ? "full" : "bit-map based"); | |
3071 | return C_MASK; | |
3072 | } | |
3073 | ||
b411b363 PR |
3074 | if (abs(hg) >= 2) { |
3075 | dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); | |
20ceb2b2 LE |
3076 | if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake", |
3077 | BM_LOCKED_SET_ALLOWED)) | |
b411b363 PR |
3078 | return C_MASK; |
3079 | } | |
3080 | ||
3081 | if (hg > 0) { /* become sync source. */ | |
3082 | rv = C_WF_BITMAP_S; | |
3083 | } else if (hg < 0) { /* become sync target */ | |
3084 | rv = C_WF_BITMAP_T; | |
3085 | } else { | |
3086 | rv = C_CONNECTED; | |
3087 | if (drbd_bm_total_weight(mdev)) { | |
3088 | dev_info(DEV, "No resync, but %lu bits in bitmap!\n", | |
3089 | drbd_bm_total_weight(mdev)); | |
3090 | } | |
3091 | } | |
3092 | ||
3093 | return rv; | |
3094 | } | |
3095 | ||
f179d76d | 3096 | static enum drbd_after_sb_p convert_after_sb(enum drbd_after_sb_p peer) |
b411b363 PR |
3097 | { |
3098 | /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ | |
f179d76d PR |
3099 | if (peer == ASB_DISCARD_REMOTE) |
3100 | return ASB_DISCARD_LOCAL; | |
b411b363 PR |
3101 | |
3102 | /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ | |
f179d76d PR |
3103 | if (peer == ASB_DISCARD_LOCAL) |
3104 | return ASB_DISCARD_REMOTE; | |
b411b363 PR |
3105 | |
3106 | /* everything else is valid if they are equal on both sides. */ | |
f179d76d | 3107 | return peer; |
b411b363 PR |
3108 | } |
3109 | ||
e2857216 | 3110 | static int receive_protocol(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3111 | { |
e658983a | 3112 | struct p_protocol *p = pi->data; |
036b17ea PR |
3113 | enum drbd_after_sb_p p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; |
3114 | int p_proto, p_discard_my_data, p_two_primaries, cf; | |
3115 | struct net_conf *nc, *old_net_conf, *new_net_conf = NULL; | |
3116 | char integrity_alg[SHARED_SECRET_MAX] = ""; | |
accdbcc5 | 3117 | struct crypto_hash *peer_integrity_tfm = NULL; |
7aca6c75 | 3118 | void *int_dig_in = NULL, *int_dig_vv = NULL; |
b411b363 | 3119 | |
b411b363 PR |
3120 | p_proto = be32_to_cpu(p->protocol); |
3121 | p_after_sb_0p = be32_to_cpu(p->after_sb_0p); | |
3122 | p_after_sb_1p = be32_to_cpu(p->after_sb_1p); | |
3123 | p_after_sb_2p = be32_to_cpu(p->after_sb_2p); | |
b411b363 | 3124 | p_two_primaries = be32_to_cpu(p->two_primaries); |
cf14c2e9 | 3125 | cf = be32_to_cpu(p->conn_flags); |
6139f60d | 3126 | p_discard_my_data = cf & CF_DISCARD_MY_DATA; |
cf14c2e9 | 3127 | |
86db0618 AG |
3128 | if (tconn->agreed_pro_version >= 87) { |
3129 | int err; | |
cf14c2e9 | 3130 | |
88104ca4 | 3131 | if (pi->size > sizeof(integrity_alg)) |
86db0618 | 3132 | return -EIO; |
88104ca4 | 3133 | err = drbd_recv_all(tconn, integrity_alg, pi->size); |
86db0618 AG |
3134 | if (err) |
3135 | return err; | |
036b17ea | 3136 | integrity_alg[SHARED_SECRET_MAX - 1] = 0; |
b411b363 PR |
3137 | } |
3138 | ||
7d4c782c | 3139 | if (pi->cmd != P_PROTOCOL_UPDATE) { |
fbc12f45 | 3140 | clear_bit(CONN_DRY_RUN, &tconn->flags); |
b411b363 | 3141 | |
fbc12f45 AG |
3142 | if (cf & CF_DRY_RUN) |
3143 | set_bit(CONN_DRY_RUN, &tconn->flags); | |
b411b363 | 3144 | |
fbc12f45 AG |
3145 | rcu_read_lock(); |
3146 | nc = rcu_dereference(tconn->net_conf); | |
b411b363 | 3147 | |
fbc12f45 | 3148 | if (p_proto != nc->wire_protocol) { |
d505d9be | 3149 | conn_err(tconn, "incompatible %s settings\n", "protocol"); |
fbc12f45 AG |
3150 | goto disconnect_rcu_unlock; |
3151 | } | |
b411b363 | 3152 | |
fbc12f45 | 3153 | if (convert_after_sb(p_after_sb_0p) != nc->after_sb_0p) { |
d505d9be | 3154 | conn_err(tconn, "incompatible %s settings\n", "after-sb-0pri"); |
fbc12f45 AG |
3155 | goto disconnect_rcu_unlock; |
3156 | } | |
b411b363 | 3157 | |
fbc12f45 | 3158 | if (convert_after_sb(p_after_sb_1p) != nc->after_sb_1p) { |
d505d9be | 3159 | conn_err(tconn, "incompatible %s settings\n", "after-sb-1pri"); |
fbc12f45 AG |
3160 | goto disconnect_rcu_unlock; |
3161 | } | |
b411b363 | 3162 | |
fbc12f45 | 3163 | if (convert_after_sb(p_after_sb_2p) != nc->after_sb_2p) { |
d505d9be | 3164 | conn_err(tconn, "incompatible %s settings\n", "after-sb-2pri"); |
fbc12f45 AG |
3165 | goto disconnect_rcu_unlock; |
3166 | } | |
b411b363 | 3167 | |
fbc12f45 | 3168 | if (p_discard_my_data && nc->discard_my_data) { |
d505d9be | 3169 | conn_err(tconn, "incompatible %s settings\n", "discard-my-data"); |
fbc12f45 AG |
3170 | goto disconnect_rcu_unlock; |
3171 | } | |
b411b363 | 3172 | |
fbc12f45 | 3173 | if (p_two_primaries != nc->two_primaries) { |
d505d9be | 3174 | conn_err(tconn, "incompatible %s settings\n", "allow-two-primaries"); |
fbc12f45 AG |
3175 | goto disconnect_rcu_unlock; |
3176 | } | |
b411b363 | 3177 | |
fbc12f45 | 3178 | if (strcmp(integrity_alg, nc->integrity_alg)) { |
d505d9be | 3179 | conn_err(tconn, "incompatible %s settings\n", "data-integrity-alg"); |
fbc12f45 AG |
3180 | goto disconnect_rcu_unlock; |
3181 | } | |
b411b363 | 3182 | |
fbc12f45 | 3183 | rcu_read_unlock(); |
b411b363 PR |
3184 | } |
3185 | ||
7d4c782c AG |
3186 | if (integrity_alg[0]) { |
3187 | int hash_size; | |
3188 | ||
3189 | /* | |
3190 | * We can only change the peer data integrity algorithm | |
3191 | * here. Changing our own data integrity algorithm | |
3192 | * requires that we send a P_PROTOCOL_UPDATE packet at | |
3193 | * the same time; otherwise, the peer has no way to | |
3194 | * tell between which packets the algorithm should | |
3195 | * change. | |
3196 | */ | |
b411b363 | 3197 | |
7d4c782c AG |
3198 | peer_integrity_tfm = crypto_alloc_hash(integrity_alg, 0, CRYPTO_ALG_ASYNC); |
3199 | if (!peer_integrity_tfm) { | |
3200 | conn_err(tconn, "peer data-integrity-alg %s not supported\n", | |
3201 | integrity_alg); | |
3202 | goto disconnect; | |
3203 | } | |
b411b363 | 3204 | |
7d4c782c AG |
3205 | hash_size = crypto_hash_digestsize(peer_integrity_tfm); |
3206 | int_dig_in = kmalloc(hash_size, GFP_KERNEL); | |
3207 | int_dig_vv = kmalloc(hash_size, GFP_KERNEL); | |
3208 | if (!(int_dig_in && int_dig_vv)) { | |
3209 | conn_err(tconn, "Allocation of buffers for data integrity checking failed\n"); | |
b411b363 PR |
3210 | goto disconnect; |
3211 | } | |
b411b363 PR |
3212 | } |
3213 | ||
7d4c782c AG |
3214 | new_net_conf = kmalloc(sizeof(struct net_conf), GFP_KERNEL); |
3215 | if (!new_net_conf) { | |
3216 | conn_err(tconn, "Allocation of new net_conf failed\n"); | |
3217 | goto disconnect; | |
3218 | } | |
3219 | ||
3220 | mutex_lock(&tconn->data.mutex); | |
3221 | mutex_lock(&tconn->conf_update); | |
3222 | old_net_conf = tconn->net_conf; | |
3223 | *new_net_conf = *old_net_conf; | |
3224 | ||
3225 | new_net_conf->wire_protocol = p_proto; | |
3226 | new_net_conf->after_sb_0p = convert_after_sb(p_after_sb_0p); | |
3227 | new_net_conf->after_sb_1p = convert_after_sb(p_after_sb_1p); | |
3228 | new_net_conf->after_sb_2p = convert_after_sb(p_after_sb_2p); | |
3229 | new_net_conf->two_primaries = p_two_primaries; | |
3230 | ||
3231 | rcu_assign_pointer(tconn->net_conf, new_net_conf); | |
3232 | mutex_unlock(&tconn->conf_update); | |
3233 | mutex_unlock(&tconn->data.mutex); | |
3234 | ||
3235 | crypto_free_hash(tconn->peer_integrity_tfm); | |
3236 | kfree(tconn->int_dig_in); | |
3237 | kfree(tconn->int_dig_vv); | |
3238 | tconn->peer_integrity_tfm = peer_integrity_tfm; | |
3239 | tconn->int_dig_in = int_dig_in; | |
3240 | tconn->int_dig_vv = int_dig_vv; | |
3241 | ||
3242 | if (strcmp(old_net_conf->integrity_alg, integrity_alg)) | |
3243 | conn_info(tconn, "peer data-integrity-alg: %s\n", | |
3244 | integrity_alg[0] ? integrity_alg : "(none)"); | |
3245 | ||
3246 | synchronize_rcu(); | |
3247 | kfree(old_net_conf); | |
82bc0194 | 3248 | return 0; |
b411b363 | 3249 | |
44ed167d PR |
3250 | disconnect_rcu_unlock: |
3251 | rcu_read_unlock(); | |
b411b363 | 3252 | disconnect: |
b792c35c | 3253 | crypto_free_hash(peer_integrity_tfm); |
036b17ea PR |
3254 | kfree(int_dig_in); |
3255 | kfree(int_dig_vv); | |
7204624c | 3256 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3257 | return -EIO; |
b411b363 PR |
3258 | } |
3259 | ||
3260 | /* helper function | |
3261 | * input: alg name, feature name | |
3262 | * return: NULL (alg name was "") | |
3263 | * ERR_PTR(error) if something goes wrong | |
3264 | * or the crypto hash ptr, if it worked out ok. */ | |
3265 | struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev, | |
3266 | const char *alg, const char *name) | |
3267 | { | |
3268 | struct crypto_hash *tfm; | |
3269 | ||
3270 | if (!alg[0]) | |
3271 | return NULL; | |
3272 | ||
3273 | tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); | |
3274 | if (IS_ERR(tfm)) { | |
3275 | dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n", | |
3276 | alg, name, PTR_ERR(tfm)); | |
3277 | return tfm; | |
3278 | } | |
b411b363 PR |
3279 | return tfm; |
3280 | } | |
3281 | ||
4a76b161 AG |
3282 | static int ignore_remaining_packet(struct drbd_tconn *tconn, struct packet_info *pi) |
3283 | { | |
3284 | void *buffer = tconn->data.rbuf; | |
3285 | int size = pi->size; | |
3286 | ||
3287 | while (size) { | |
3288 | int s = min_t(int, size, DRBD_SOCKET_BUFFER_SIZE); | |
3289 | s = drbd_recv(tconn, buffer, s); | |
3290 | if (s <= 0) { | |
3291 | if (s < 0) | |
3292 | return s; | |
3293 | break; | |
3294 | } | |
3295 | size -= s; | |
3296 | } | |
3297 | if (size) | |
3298 | return -EIO; | |
3299 | return 0; | |
3300 | } | |
3301 | ||
3302 | /* | |
3303 | * config_unknown_volume - device configuration command for unknown volume | |
3304 | * | |
3305 | * When a device is added to an existing connection, the node on which the | |
3306 | * device is added first will send configuration commands to its peer but the | |
3307 | * peer will not know about the device yet. It will warn and ignore these | |
3308 | * commands. Once the device is added on the second node, the second node will | |
3309 | * send the same device configuration commands, but in the other direction. | |
3310 | * | |
3311 | * (We can also end up here if drbd is misconfigured.) | |
3312 | */ | |
3313 | static int config_unknown_volume(struct drbd_tconn *tconn, struct packet_info *pi) | |
3314 | { | |
2fcb8f30 AG |
3315 | conn_warn(tconn, "%s packet received for volume %u, which is not configured locally\n", |
3316 | cmdname(pi->cmd), pi->vnr); | |
4a76b161 AG |
3317 | return ignore_remaining_packet(tconn, pi); |
3318 | } | |
3319 | ||
3320 | static int receive_SyncParam(struct drbd_tconn *tconn, struct packet_info *pi) | |
b411b363 | 3321 | { |
4a76b161 | 3322 | struct drbd_conf *mdev; |
e658983a | 3323 | struct p_rs_param_95 *p; |
b411b363 PR |
3324 | unsigned int header_size, data_size, exp_max_sz; |
3325 | struct crypto_hash *verify_tfm = NULL; | |
3326 | struct crypto_hash *csums_tfm = NULL; | |
2ec91e0e | 3327 | struct net_conf *old_net_conf, *new_net_conf = NULL; |
813472ce | 3328 | struct disk_conf *old_disk_conf = NULL, *new_disk_conf = NULL; |
4a76b161 | 3329 | const int apv = tconn->agreed_pro_version; |
813472ce | 3330 | struct fifo_buffer *old_plan = NULL, *new_plan = NULL; |
778f271d | 3331 | int fifo_size = 0; |
82bc0194 | 3332 | int err; |
b411b363 | 3333 | |
4a76b161 AG |
3334 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3335 | if (!mdev) | |
3336 | return config_unknown_volume(tconn, pi); | |
b411b363 PR |
3337 | |
3338 | exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) | |
3339 | : apv == 88 ? sizeof(struct p_rs_param) | |
3340 | + SHARED_SECRET_MAX | |
8e26f9cc PR |
3341 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
3342 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 3343 | |
e2857216 | 3344 | if (pi->size > exp_max_sz) { |
b411b363 | 3345 | dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n", |
e2857216 | 3346 | pi->size, exp_max_sz); |
82bc0194 | 3347 | return -EIO; |
b411b363 PR |
3348 | } |
3349 | ||
3350 | if (apv <= 88) { | |
e658983a | 3351 | header_size = sizeof(struct p_rs_param); |
e2857216 | 3352 | data_size = pi->size - header_size; |
8e26f9cc | 3353 | } else if (apv <= 94) { |
e658983a | 3354 | header_size = sizeof(struct p_rs_param_89); |
e2857216 | 3355 | data_size = pi->size - header_size; |
b411b363 | 3356 | D_ASSERT(data_size == 0); |
8e26f9cc | 3357 | } else { |
e658983a | 3358 | header_size = sizeof(struct p_rs_param_95); |
e2857216 | 3359 | data_size = pi->size - header_size; |
b411b363 PR |
3360 | D_ASSERT(data_size == 0); |
3361 | } | |
3362 | ||
3363 | /* initialize verify_alg and csums_alg */ | |
e658983a | 3364 | p = pi->data; |
b411b363 PR |
3365 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); |
3366 | ||
e658983a | 3367 | err = drbd_recv_all(mdev->tconn, p, header_size); |
82bc0194 AG |
3368 | if (err) |
3369 | return err; | |
b411b363 | 3370 | |
daeda1cc PR |
3371 | mutex_lock(&mdev->tconn->conf_update); |
3372 | old_net_conf = mdev->tconn->net_conf; | |
813472ce PR |
3373 | if (get_ldev(mdev)) { |
3374 | new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); | |
3375 | if (!new_disk_conf) { | |
3376 | put_ldev(mdev); | |
3377 | mutex_unlock(&mdev->tconn->conf_update); | |
3378 | dev_err(DEV, "Allocation of new disk_conf failed\n"); | |
3379 | return -ENOMEM; | |
3380 | } | |
daeda1cc | 3381 | |
813472ce PR |
3382 | old_disk_conf = mdev->ldev->disk_conf; |
3383 | *new_disk_conf = *old_disk_conf; | |
b411b363 | 3384 | |
6394b935 | 3385 | new_disk_conf->resync_rate = be32_to_cpu(p->resync_rate); |
813472ce | 3386 | } |
b411b363 PR |
3387 | |
3388 | if (apv >= 88) { | |
3389 | if (apv == 88) { | |
5de73827 PR |
3390 | if (data_size > SHARED_SECRET_MAX || data_size == 0) { |
3391 | dev_err(DEV, "verify-alg of wrong size, " | |
3392 | "peer wants %u, accepting only up to %u byte\n", | |
3393 | data_size, SHARED_SECRET_MAX); | |
813472ce PR |
3394 | err = -EIO; |
3395 | goto reconnect; | |
b411b363 PR |
3396 | } |
3397 | ||
82bc0194 | 3398 | err = drbd_recv_all(mdev->tconn, p->verify_alg, data_size); |
813472ce PR |
3399 | if (err) |
3400 | goto reconnect; | |
b411b363 PR |
3401 | /* we expect NUL terminated string */ |
3402 | /* but just in case someone tries to be evil */ | |
3403 | D_ASSERT(p->verify_alg[data_size-1] == 0); | |
3404 | p->verify_alg[data_size-1] = 0; | |
3405 | ||
3406 | } else /* apv >= 89 */ { | |
3407 | /* we still expect NUL terminated strings */ | |
3408 | /* but just in case someone tries to be evil */ | |
3409 | D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0); | |
3410 | D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0); | |
3411 | p->verify_alg[SHARED_SECRET_MAX-1] = 0; | |
3412 | p->csums_alg[SHARED_SECRET_MAX-1] = 0; | |
3413 | } | |
3414 | ||
2ec91e0e | 3415 | if (strcmp(old_net_conf->verify_alg, p->verify_alg)) { |
b411b363 PR |
3416 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { |
3417 | dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", | |
2ec91e0e | 3418 | old_net_conf->verify_alg, p->verify_alg); |
b411b363 PR |
3419 | goto disconnect; |
3420 | } | |
3421 | verify_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
3422 | p->verify_alg, "verify-alg"); | |
3423 | if (IS_ERR(verify_tfm)) { | |
3424 | verify_tfm = NULL; | |
3425 | goto disconnect; | |
3426 | } | |
3427 | } | |
3428 | ||
2ec91e0e | 3429 | if (apv >= 89 && strcmp(old_net_conf->csums_alg, p->csums_alg)) { |
b411b363 PR |
3430 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { |
3431 | dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", | |
2ec91e0e | 3432 | old_net_conf->csums_alg, p->csums_alg); |
b411b363 PR |
3433 | goto disconnect; |
3434 | } | |
3435 | csums_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
3436 | p->csums_alg, "csums-alg"); | |
3437 | if (IS_ERR(csums_tfm)) { | |
3438 | csums_tfm = NULL; | |
3439 | goto disconnect; | |
3440 | } | |
3441 | } | |
3442 | ||
813472ce | 3443 | if (apv > 94 && new_disk_conf) { |
daeda1cc PR |
3444 | new_disk_conf->c_plan_ahead = be32_to_cpu(p->c_plan_ahead); |
3445 | new_disk_conf->c_delay_target = be32_to_cpu(p->c_delay_target); | |
3446 | new_disk_conf->c_fill_target = be32_to_cpu(p->c_fill_target); | |
3447 | new_disk_conf->c_max_rate = be32_to_cpu(p->c_max_rate); | |
778f271d | 3448 | |
daeda1cc | 3449 | fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ; |
9958c857 | 3450 | if (fifo_size != mdev->rs_plan_s->size) { |
813472ce PR |
3451 | new_plan = fifo_alloc(fifo_size); |
3452 | if (!new_plan) { | |
778f271d | 3453 | dev_err(DEV, "kmalloc of fifo_buffer failed"); |
f399002e | 3454 | put_ldev(mdev); |
778f271d PR |
3455 | goto disconnect; |
3456 | } | |
3457 | } | |
8e26f9cc | 3458 | } |
b411b363 | 3459 | |
91fd4dad | 3460 | if (verify_tfm || csums_tfm) { |
2ec91e0e PR |
3461 | new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL); |
3462 | if (!new_net_conf) { | |
91fd4dad PR |
3463 | dev_err(DEV, "Allocation of new net_conf failed\n"); |
3464 | goto disconnect; | |
3465 | } | |
3466 | ||
2ec91e0e | 3467 | *new_net_conf = *old_net_conf; |
91fd4dad PR |
3468 | |
3469 | if (verify_tfm) { | |
2ec91e0e PR |
3470 | strcpy(new_net_conf->verify_alg, p->verify_alg); |
3471 | new_net_conf->verify_alg_len = strlen(p->verify_alg) + 1; | |
91fd4dad PR |
3472 | crypto_free_hash(mdev->tconn->verify_tfm); |
3473 | mdev->tconn->verify_tfm = verify_tfm; | |
3474 | dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg); | |
3475 | } | |
3476 | if (csums_tfm) { | |
2ec91e0e PR |
3477 | strcpy(new_net_conf->csums_alg, p->csums_alg); |
3478 | new_net_conf->csums_alg_len = strlen(p->csums_alg) + 1; | |
91fd4dad PR |
3479 | crypto_free_hash(mdev->tconn->csums_tfm); |
3480 | mdev->tconn->csums_tfm = csums_tfm; | |
3481 | dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg); | |
3482 | } | |
2ec91e0e | 3483 | rcu_assign_pointer(tconn->net_conf, new_net_conf); |
778f271d | 3484 | } |
b411b363 PR |
3485 | } |
3486 | ||
813472ce PR |
3487 | if (new_disk_conf) { |
3488 | rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf); | |
3489 | put_ldev(mdev); | |
3490 | } | |
3491 | ||
3492 | if (new_plan) { | |
3493 | old_plan = mdev->rs_plan_s; | |
3494 | rcu_assign_pointer(mdev->rs_plan_s, new_plan); | |
b411b363 | 3495 | } |
daeda1cc PR |
3496 | |
3497 | mutex_unlock(&mdev->tconn->conf_update); | |
3498 | synchronize_rcu(); | |
3499 | if (new_net_conf) | |
3500 | kfree(old_net_conf); | |
3501 | kfree(old_disk_conf); | |
813472ce | 3502 | kfree(old_plan); |
daeda1cc | 3503 | |
82bc0194 | 3504 | return 0; |
b411b363 | 3505 | |
813472ce PR |
3506 | reconnect: |
3507 | if (new_disk_conf) { | |
3508 | put_ldev(mdev); | |
3509 | kfree(new_disk_conf); | |
3510 | } | |
3511 | mutex_unlock(&mdev->tconn->conf_update); | |
3512 | return -EIO; | |
3513 | ||
b411b363 | 3514 | disconnect: |
813472ce PR |
3515 | kfree(new_plan); |
3516 | if (new_disk_conf) { | |
3517 | put_ldev(mdev); | |
3518 | kfree(new_disk_conf); | |
3519 | } | |
a0095508 | 3520 | mutex_unlock(&mdev->tconn->conf_update); |
b411b363 PR |
3521 | /* just for completeness: actually not needed, |
3522 | * as this is not reached if csums_tfm was ok. */ | |
3523 | crypto_free_hash(csums_tfm); | |
3524 | /* but free the verify_tfm again, if csums_tfm did not work out */ | |
3525 | crypto_free_hash(verify_tfm); | |
38fa9988 | 3526 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3527 | return -EIO; |
b411b363 PR |
3528 | } |
3529 | ||
b411b363 PR |
3530 | /* warn if the arguments differ by more than 12.5% */ |
3531 | static void warn_if_differ_considerably(struct drbd_conf *mdev, | |
3532 | const char *s, sector_t a, sector_t b) | |
3533 | { | |
3534 | sector_t d; | |
3535 | if (a == 0 || b == 0) | |
3536 | return; | |
3537 | d = (a > b) ? (a - b) : (b - a); | |
3538 | if (d > (a>>3) || d > (b>>3)) | |
3539 | dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s, | |
3540 | (unsigned long long)a, (unsigned long long)b); | |
3541 | } | |
3542 | ||
4a76b161 | 3543 | static int receive_sizes(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3544 | { |
4a76b161 | 3545 | struct drbd_conf *mdev; |
e658983a | 3546 | struct p_sizes *p = pi->data; |
b411b363 | 3547 | enum determine_dev_size dd = unchanged; |
b411b363 PR |
3548 | sector_t p_size, p_usize, my_usize; |
3549 | int ldsc = 0; /* local disk size changed */ | |
e89b591c | 3550 | enum dds_flags ddsf; |
b411b363 | 3551 | |
4a76b161 AG |
3552 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3553 | if (!mdev) | |
3554 | return config_unknown_volume(tconn, pi); | |
3555 | ||
b411b363 PR |
3556 | p_size = be64_to_cpu(p->d_size); |
3557 | p_usize = be64_to_cpu(p->u_size); | |
3558 | ||
b411b363 PR |
3559 | /* just store the peer's disk size for now. |
3560 | * we still need to figure out whether we accept that. */ | |
3561 | mdev->p_size = p_size; | |
3562 | ||
b411b363 | 3563 | if (get_ldev(mdev)) { |
daeda1cc PR |
3564 | rcu_read_lock(); |
3565 | my_usize = rcu_dereference(mdev->ldev->disk_conf)->disk_size; | |
3566 | rcu_read_unlock(); | |
3567 | ||
b411b363 PR |
3568 | warn_if_differ_considerably(mdev, "lower level device sizes", |
3569 | p_size, drbd_get_max_capacity(mdev->ldev)); | |
3570 | warn_if_differ_considerably(mdev, "user requested size", | |
daeda1cc | 3571 | p_usize, my_usize); |
b411b363 PR |
3572 | |
3573 | /* if this is the first connect, or an otherwise expected | |
3574 | * param exchange, choose the minimum */ | |
3575 | if (mdev->state.conn == C_WF_REPORT_PARAMS) | |
daeda1cc | 3576 | p_usize = min_not_zero(my_usize, p_usize); |
b411b363 PR |
3577 | |
3578 | /* Never shrink a device with usable data during connect. | |
3579 | But allow online shrinking if we are connected. */ | |
ef5e44a6 | 3580 | if (drbd_new_dev_size(mdev, mdev->ldev, p_usize, 0) < |
daeda1cc PR |
3581 | drbd_get_capacity(mdev->this_bdev) && |
3582 | mdev->state.disk >= D_OUTDATED && | |
3583 | mdev->state.conn < C_CONNECTED) { | |
b411b363 | 3584 | dev_err(DEV, "The peer's disk size is too small!\n"); |
38fa9988 | 3585 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 3586 | put_ldev(mdev); |
82bc0194 | 3587 | return -EIO; |
b411b363 | 3588 | } |
daeda1cc PR |
3589 | |
3590 | if (my_usize != p_usize) { | |
3591 | struct disk_conf *old_disk_conf, *new_disk_conf = NULL; | |
3592 | ||
3593 | new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); | |
3594 | if (!new_disk_conf) { | |
3595 | dev_err(DEV, "Allocation of new disk_conf failed\n"); | |
3596 | put_ldev(mdev); | |
3597 | return -ENOMEM; | |
3598 | } | |
3599 | ||
3600 | mutex_lock(&mdev->tconn->conf_update); | |
3601 | old_disk_conf = mdev->ldev->disk_conf; | |
3602 | *new_disk_conf = *old_disk_conf; | |
3603 | new_disk_conf->disk_size = p_usize; | |
3604 | ||
3605 | rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf); | |
3606 | mutex_unlock(&mdev->tconn->conf_update); | |
3607 | synchronize_rcu(); | |
3608 | kfree(old_disk_conf); | |
3609 | ||
3610 | dev_info(DEV, "Peer sets u_size to %lu sectors\n", | |
3611 | (unsigned long)my_usize); | |
b411b363 | 3612 | } |
daeda1cc | 3613 | |
b411b363 PR |
3614 | put_ldev(mdev); |
3615 | } | |
b411b363 | 3616 | |
e89b591c | 3617 | ddsf = be16_to_cpu(p->dds_flags); |
b411b363 | 3618 | if (get_ldev(mdev)) { |
24c4830c | 3619 | dd = drbd_determine_dev_size(mdev, ddsf); |
b411b363 PR |
3620 | put_ldev(mdev); |
3621 | if (dd == dev_size_error) | |
82bc0194 | 3622 | return -EIO; |
b411b363 PR |
3623 | drbd_md_sync(mdev); |
3624 | } else { | |
3625 | /* I am diskless, need to accept the peer's size. */ | |
3626 | drbd_set_my_capacity(mdev, p_size); | |
3627 | } | |
3628 | ||
99432fcc PR |
3629 | mdev->peer_max_bio_size = be32_to_cpu(p->max_bio_size); |
3630 | drbd_reconsider_max_bio_size(mdev); | |
3631 | ||
b411b363 PR |
3632 | if (get_ldev(mdev)) { |
3633 | if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) { | |
3634 | mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); | |
3635 | ldsc = 1; | |
3636 | } | |
3637 | ||
b411b363 PR |
3638 | put_ldev(mdev); |
3639 | } | |
3640 | ||
3641 | if (mdev->state.conn > C_WF_REPORT_PARAMS) { | |
3642 | if (be64_to_cpu(p->c_size) != | |
3643 | drbd_get_capacity(mdev->this_bdev) || ldsc) { | |
3644 | /* we have different sizes, probably peer | |
3645 | * needs to know my new size... */ | |
e89b591c | 3646 | drbd_send_sizes(mdev, 0, ddsf); |
b411b363 PR |
3647 | } |
3648 | if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) || | |
3649 | (dd == grew && mdev->state.conn == C_CONNECTED)) { | |
3650 | if (mdev->state.pdsk >= D_INCONSISTENT && | |
e89b591c PR |
3651 | mdev->state.disk >= D_INCONSISTENT) { |
3652 | if (ddsf & DDSF_NO_RESYNC) | |
3653 | dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n"); | |
3654 | else | |
3655 | resync_after_online_grow(mdev); | |
3656 | } else | |
b411b363 PR |
3657 | set_bit(RESYNC_AFTER_NEG, &mdev->flags); |
3658 | } | |
3659 | } | |
3660 | ||
82bc0194 | 3661 | return 0; |
b411b363 PR |
3662 | } |
3663 | ||
4a76b161 | 3664 | static int receive_uuids(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3665 | { |
4a76b161 | 3666 | struct drbd_conf *mdev; |
e658983a | 3667 | struct p_uuids *p = pi->data; |
b411b363 | 3668 | u64 *p_uuid; |
62b0da3a | 3669 | int i, updated_uuids = 0; |
b411b363 | 3670 | |
4a76b161 AG |
3671 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3672 | if (!mdev) | |
3673 | return config_unknown_volume(tconn, pi); | |
3674 | ||
b411b363 | 3675 | p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); |
063eacf8 JW |
3676 | if (!p_uuid) { |
3677 | dev_err(DEV, "kmalloc of p_uuid failed\n"); | |
3678 | return false; | |
3679 | } | |
b411b363 PR |
3680 | |
3681 | for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) | |
3682 | p_uuid[i] = be64_to_cpu(p->uuid[i]); | |
3683 | ||
3684 | kfree(mdev->p_uuid); | |
3685 | mdev->p_uuid = p_uuid; | |
3686 | ||
3687 | if (mdev->state.conn < C_CONNECTED && | |
3688 | mdev->state.disk < D_INCONSISTENT && | |
3689 | mdev->state.role == R_PRIMARY && | |
3690 | (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { | |
3691 | dev_err(DEV, "Can only connect to data with current UUID=%016llX\n", | |
3692 | (unsigned long long)mdev->ed_uuid); | |
38fa9988 | 3693 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3694 | return -EIO; |
b411b363 PR |
3695 | } |
3696 | ||
3697 | if (get_ldev(mdev)) { | |
3698 | int skip_initial_sync = | |
3699 | mdev->state.conn == C_CONNECTED && | |
31890f4a | 3700 | mdev->tconn->agreed_pro_version >= 90 && |
b411b363 PR |
3701 | mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && |
3702 | (p_uuid[UI_FLAGS] & 8); | |
3703 | if (skip_initial_sync) { | |
3704 | dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n"); | |
3705 | drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, | |
20ceb2b2 LE |
3706 | "clear_n_write from receive_uuids", |
3707 | BM_LOCKED_TEST_ALLOWED); | |
b411b363 PR |
3708 | _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]); |
3709 | _drbd_uuid_set(mdev, UI_BITMAP, 0); | |
3710 | _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), | |
3711 | CS_VERBOSE, NULL); | |
3712 | drbd_md_sync(mdev); | |
62b0da3a | 3713 | updated_uuids = 1; |
b411b363 PR |
3714 | } |
3715 | put_ldev(mdev); | |
18a50fa2 PR |
3716 | } else if (mdev->state.disk < D_INCONSISTENT && |
3717 | mdev->state.role == R_PRIMARY) { | |
3718 | /* I am a diskless primary, the peer just created a new current UUID | |
3719 | for me. */ | |
62b0da3a | 3720 | updated_uuids = drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
b411b363 PR |
3721 | } |
3722 | ||
3723 | /* Before we test for the disk state, we should wait until an eventually | |
3724 | ongoing cluster wide state change is finished. That is important if | |
3725 | we are primary and are detaching from our disk. We need to see the | |
3726 | new disk state... */ | |
8410da8f PR |
3727 | mutex_lock(mdev->state_mutex); |
3728 | mutex_unlock(mdev->state_mutex); | |
b411b363 | 3729 | if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT) |
62b0da3a LE |
3730 | updated_uuids |= drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
3731 | ||
3732 | if (updated_uuids) | |
3733 | drbd_print_uuids(mdev, "receiver updated UUIDs to"); | |
b411b363 | 3734 | |
82bc0194 | 3735 | return 0; |
b411b363 PR |
3736 | } |
3737 | ||
3738 | /** | |
3739 | * convert_state() - Converts the peer's view of the cluster state to our point of view | |
3740 | * @ps: The state as seen by the peer. | |
3741 | */ | |
3742 | static union drbd_state convert_state(union drbd_state ps) | |
3743 | { | |
3744 | union drbd_state ms; | |
3745 | ||
3746 | static enum drbd_conns c_tab[] = { | |
369bea63 | 3747 | [C_WF_REPORT_PARAMS] = C_WF_REPORT_PARAMS, |
b411b363 PR |
3748 | [C_CONNECTED] = C_CONNECTED, |
3749 | ||
3750 | [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, | |
3751 | [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, | |
3752 | [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ | |
3753 | [C_VERIFY_S] = C_VERIFY_T, | |
3754 | [C_MASK] = C_MASK, | |
3755 | }; | |
3756 | ||
3757 | ms.i = ps.i; | |
3758 | ||
3759 | ms.conn = c_tab[ps.conn]; | |
3760 | ms.peer = ps.role; | |
3761 | ms.role = ps.peer; | |
3762 | ms.pdsk = ps.disk; | |
3763 | ms.disk = ps.pdsk; | |
3764 | ms.peer_isp = (ps.aftr_isp | ps.user_isp); | |
3765 | ||
3766 | return ms; | |
3767 | } | |
3768 | ||
4a76b161 | 3769 | static int receive_req_state(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3770 | { |
4a76b161 | 3771 | struct drbd_conf *mdev; |
e658983a | 3772 | struct p_req_state *p = pi->data; |
b411b363 | 3773 | union drbd_state mask, val; |
bf885f8a | 3774 | enum drbd_state_rv rv; |
b411b363 | 3775 | |
4a76b161 AG |
3776 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3777 | if (!mdev) | |
3778 | return -EIO; | |
3779 | ||
b411b363 PR |
3780 | mask.i = be32_to_cpu(p->mask); |
3781 | val.i = be32_to_cpu(p->val); | |
3782 | ||
427c0434 | 3783 | if (test_bit(RESOLVE_CONFLICTS, &mdev->tconn->flags) && |
8410da8f | 3784 | mutex_is_locked(mdev->state_mutex)) { |
b411b363 | 3785 | drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG); |
82bc0194 | 3786 | return 0; |
b411b363 PR |
3787 | } |
3788 | ||
3789 | mask = convert_state(mask); | |
3790 | val = convert_state(val); | |
3791 | ||
dfafcc8a PR |
3792 | rv = drbd_change_state(mdev, CS_VERBOSE, mask, val); |
3793 | drbd_send_sr_reply(mdev, rv); | |
b411b363 | 3794 | |
b411b363 PR |
3795 | drbd_md_sync(mdev); |
3796 | ||
82bc0194 | 3797 | return 0; |
b411b363 PR |
3798 | } |
3799 | ||
e2857216 | 3800 | static int receive_req_conn_state(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3801 | { |
e658983a | 3802 | struct p_req_state *p = pi->data; |
b411b363 | 3803 | union drbd_state mask, val; |
bf885f8a | 3804 | enum drbd_state_rv rv; |
b411b363 | 3805 | |
b411b363 PR |
3806 | mask.i = be32_to_cpu(p->mask); |
3807 | val.i = be32_to_cpu(p->val); | |
3808 | ||
427c0434 | 3809 | if (test_bit(RESOLVE_CONFLICTS, &tconn->flags) && |
dfafcc8a PR |
3810 | mutex_is_locked(&tconn->cstate_mutex)) { |
3811 | conn_send_sr_reply(tconn, SS_CONCURRENT_ST_CHG); | |
82bc0194 | 3812 | return 0; |
b411b363 PR |
3813 | } |
3814 | ||
3815 | mask = convert_state(mask); | |
3816 | val = convert_state(val); | |
3817 | ||
778bcf2e | 3818 | rv = conn_request_state(tconn, mask, val, CS_VERBOSE | CS_LOCAL_ONLY | CS_IGN_OUTD_FAIL); |
dfafcc8a | 3819 | conn_send_sr_reply(tconn, rv); |
b411b363 | 3820 | |
82bc0194 | 3821 | return 0; |
b411b363 PR |
3822 | } |
3823 | ||
4a76b161 | 3824 | static int receive_state(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3825 | { |
4a76b161 | 3826 | struct drbd_conf *mdev; |
e658983a | 3827 | struct p_state *p = pi->data; |
4ac4aada | 3828 | union drbd_state os, ns, peer_state; |
b411b363 | 3829 | enum drbd_disk_state real_peer_disk; |
65d922c3 | 3830 | enum chg_state_flags cs_flags; |
b411b363 PR |
3831 | int rv; |
3832 | ||
4a76b161 AG |
3833 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3834 | if (!mdev) | |
3835 | return config_unknown_volume(tconn, pi); | |
3836 | ||
b411b363 PR |
3837 | peer_state.i = be32_to_cpu(p->state); |
3838 | ||
3839 | real_peer_disk = peer_state.disk; | |
3840 | if (peer_state.disk == D_NEGOTIATING) { | |
3841 | real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; | |
3842 | dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); | |
3843 | } | |
3844 | ||
87eeee41 | 3845 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 3846 | retry: |
78bae59b | 3847 | os = ns = drbd_read_state(mdev); |
87eeee41 | 3848 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 3849 | |
545752d5 LE |
3850 | /* If some other part of the code (asender thread, timeout) |
3851 | * already decided to close the connection again, | |
3852 | * we must not "re-establish" it here. */ | |
3853 | if (os.conn <= C_TEAR_DOWN) | |
58ffa580 | 3854 | return -ECONNRESET; |
545752d5 | 3855 | |
40424e4a LE |
3856 | /* If this is the "end of sync" confirmation, usually the peer disk |
3857 | * transitions from D_INCONSISTENT to D_UP_TO_DATE. For empty (0 bits | |
3858 | * set) resync started in PausedSyncT, or if the timing of pause-/ | |
3859 | * unpause-sync events has been "just right", the peer disk may | |
3860 | * transition from D_CONSISTENT to D_UP_TO_DATE as well. | |
3861 | */ | |
3862 | if ((os.pdsk == D_INCONSISTENT || os.pdsk == D_CONSISTENT) && | |
3863 | real_peer_disk == D_UP_TO_DATE && | |
e9ef7bb6 LE |
3864 | os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { |
3865 | /* If we are (becoming) SyncSource, but peer is still in sync | |
3866 | * preparation, ignore its uptodate-ness to avoid flapping, it | |
3867 | * will change to inconsistent once the peer reaches active | |
3868 | * syncing states. | |
3869 | * It may have changed syncer-paused flags, however, so we | |
3870 | * cannot ignore this completely. */ | |
3871 | if (peer_state.conn > C_CONNECTED && | |
3872 | peer_state.conn < C_SYNC_SOURCE) | |
3873 | real_peer_disk = D_INCONSISTENT; | |
3874 | ||
3875 | /* if peer_state changes to connected at the same time, | |
3876 | * it explicitly notifies us that it finished resync. | |
3877 | * Maybe we should finish it up, too? */ | |
3878 | else if (os.conn >= C_SYNC_SOURCE && | |
3879 | peer_state.conn == C_CONNECTED) { | |
3880 | if (drbd_bm_total_weight(mdev) <= mdev->rs_failed) | |
3881 | drbd_resync_finished(mdev); | |
82bc0194 | 3882 | return 0; |
e9ef7bb6 LE |
3883 | } |
3884 | } | |
3885 | ||
02b91b55 LE |
3886 | /* explicit verify finished notification, stop sector reached. */ |
3887 | if (os.conn == C_VERIFY_T && os.disk == D_UP_TO_DATE && | |
3888 | peer_state.conn == C_CONNECTED && real_peer_disk == D_UP_TO_DATE) { | |
58ffa580 | 3889 | ov_out_of_sync_print(mdev); |
02b91b55 | 3890 | drbd_resync_finished(mdev); |
58ffa580 | 3891 | return 0; |
02b91b55 LE |
3892 | } |
3893 | ||
e9ef7bb6 LE |
3894 | /* peer says his disk is inconsistent, while we think it is uptodate, |
3895 | * and this happens while the peer still thinks we have a sync going on, | |
3896 | * but we think we are already done with the sync. | |
3897 | * We ignore this to avoid flapping pdsk. | |
3898 | * This should not happen, if the peer is a recent version of drbd. */ | |
3899 | if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && | |
3900 | os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) | |
3901 | real_peer_disk = D_UP_TO_DATE; | |
3902 | ||
4ac4aada LE |
3903 | if (ns.conn == C_WF_REPORT_PARAMS) |
3904 | ns.conn = C_CONNECTED; | |
b411b363 | 3905 | |
67531718 PR |
3906 | if (peer_state.conn == C_AHEAD) |
3907 | ns.conn = C_BEHIND; | |
3908 | ||
b411b363 PR |
3909 | if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING && |
3910 | get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3911 | int cr; /* consider resync */ | |
3912 | ||
3913 | /* if we established a new connection */ | |
4ac4aada | 3914 | cr = (os.conn < C_CONNECTED); |
b411b363 PR |
3915 | /* if we had an established connection |
3916 | * and one of the nodes newly attaches a disk */ | |
4ac4aada | 3917 | cr |= (os.conn == C_CONNECTED && |
b411b363 | 3918 | (peer_state.disk == D_NEGOTIATING || |
4ac4aada | 3919 | os.disk == D_NEGOTIATING)); |
b411b363 PR |
3920 | /* if we have both been inconsistent, and the peer has been |
3921 | * forced to be UpToDate with --overwrite-data */ | |
3922 | cr |= test_bit(CONSIDER_RESYNC, &mdev->flags); | |
3923 | /* if we had been plain connected, and the admin requested to | |
3924 | * start a sync by "invalidate" or "invalidate-remote" */ | |
4ac4aada | 3925 | cr |= (os.conn == C_CONNECTED && |
b411b363 PR |
3926 | (peer_state.conn >= C_STARTING_SYNC_S && |
3927 | peer_state.conn <= C_WF_BITMAP_T)); | |
3928 | ||
3929 | if (cr) | |
4ac4aada | 3930 | ns.conn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk); |
b411b363 PR |
3931 | |
3932 | put_ldev(mdev); | |
4ac4aada LE |
3933 | if (ns.conn == C_MASK) { |
3934 | ns.conn = C_CONNECTED; | |
b411b363 | 3935 | if (mdev->state.disk == D_NEGOTIATING) { |
82f59cc6 | 3936 | drbd_force_state(mdev, NS(disk, D_FAILED)); |
b411b363 PR |
3937 | } else if (peer_state.disk == D_NEGOTIATING) { |
3938 | dev_err(DEV, "Disk attach process on the peer node was aborted.\n"); | |
3939 | peer_state.disk = D_DISKLESS; | |
580b9767 | 3940 | real_peer_disk = D_DISKLESS; |
b411b363 | 3941 | } else { |
8169e41b | 3942 | if (test_and_clear_bit(CONN_DRY_RUN, &mdev->tconn->flags)) |
82bc0194 | 3943 | return -EIO; |
4ac4aada | 3944 | D_ASSERT(os.conn == C_WF_REPORT_PARAMS); |
38fa9988 | 3945 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3946 | return -EIO; |
b411b363 PR |
3947 | } |
3948 | } | |
3949 | } | |
3950 | ||
87eeee41 | 3951 | spin_lock_irq(&mdev->tconn->req_lock); |
78bae59b | 3952 | if (os.i != drbd_read_state(mdev).i) |
b411b363 PR |
3953 | goto retry; |
3954 | clear_bit(CONSIDER_RESYNC, &mdev->flags); | |
b411b363 PR |
3955 | ns.peer = peer_state.role; |
3956 | ns.pdsk = real_peer_disk; | |
3957 | ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); | |
4ac4aada | 3958 | if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) |
b411b363 | 3959 | ns.disk = mdev->new_state_tmp.disk; |
4ac4aada | 3960 | cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); |
2aebfabb | 3961 | if (ns.pdsk == D_CONSISTENT && drbd_suspended(mdev) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && |
481c6f50 | 3962 | test_bit(NEW_CUR_UUID, &mdev->flags)) { |
8554df1c | 3963 | /* Do not allow tl_restart(RESEND) for a rebooted peer. We can only allow this |
481c6f50 | 3964 | for temporal network outages! */ |
87eeee41 | 3965 | spin_unlock_irq(&mdev->tconn->req_lock); |
481c6f50 | 3966 | dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); |
2f5cdd0b | 3967 | tl_clear(mdev->tconn); |
481c6f50 PR |
3968 | drbd_uuid_new_current(mdev); |
3969 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
38fa9988 | 3970 | conn_request_state(mdev->tconn, NS2(conn, C_PROTOCOL_ERROR, susp, 0), CS_HARD); |
82bc0194 | 3971 | return -EIO; |
481c6f50 | 3972 | } |
65d922c3 | 3973 | rv = _drbd_set_state(mdev, ns, cs_flags, NULL); |
78bae59b | 3974 | ns = drbd_read_state(mdev); |
87eeee41 | 3975 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
3976 | |
3977 | if (rv < SS_SUCCESS) { | |
38fa9988 | 3978 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3979 | return -EIO; |
b411b363 PR |
3980 | } |
3981 | ||
4ac4aada LE |
3982 | if (os.conn > C_WF_REPORT_PARAMS) { |
3983 | if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && | |
b411b363 PR |
3984 | peer_state.disk != D_NEGOTIATING ) { |
3985 | /* we want resync, peer has not yet decided to sync... */ | |
3986 | /* Nowadays only used when forcing a node into primary role and | |
3987 | setting its disk to UpToDate with that */ | |
3988 | drbd_send_uuids(mdev); | |
f479ea06 | 3989 | drbd_send_current_state(mdev); |
b411b363 PR |
3990 | } |
3991 | } | |
3992 | ||
08b165ba | 3993 | clear_bit(DISCARD_MY_DATA, &mdev->flags); |
b411b363 PR |
3994 | |
3995 | drbd_md_sync(mdev); /* update connected indicator, la_size, ... */ | |
3996 | ||
82bc0194 | 3997 | return 0; |
b411b363 PR |
3998 | } |
3999 | ||
4a76b161 | 4000 | static int receive_sync_uuid(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4001 | { |
4a76b161 | 4002 | struct drbd_conf *mdev; |
e658983a | 4003 | struct p_rs_uuid *p = pi->data; |
4a76b161 AG |
4004 | |
4005 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
4006 | if (!mdev) | |
4007 | return -EIO; | |
b411b363 PR |
4008 | |
4009 | wait_event(mdev->misc_wait, | |
4010 | mdev->state.conn == C_WF_SYNC_UUID || | |
c4752ef1 | 4011 | mdev->state.conn == C_BEHIND || |
b411b363 PR |
4012 | mdev->state.conn < C_CONNECTED || |
4013 | mdev->state.disk < D_NEGOTIATING); | |
4014 | ||
4015 | /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */ | |
4016 | ||
b411b363 PR |
4017 | /* Here the _drbd_uuid_ functions are right, current should |
4018 | _not_ be rotated into the history */ | |
4019 | if (get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
4020 | _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid)); | |
4021 | _drbd_uuid_set(mdev, UI_BITMAP, 0UL); | |
4022 | ||
62b0da3a | 4023 | drbd_print_uuids(mdev, "updated sync uuid"); |
b411b363 PR |
4024 | drbd_start_resync(mdev, C_SYNC_TARGET); |
4025 | ||
4026 | put_ldev(mdev); | |
4027 | } else | |
4028 | dev_err(DEV, "Ignoring SyncUUID packet!\n"); | |
4029 | ||
82bc0194 | 4030 | return 0; |
b411b363 PR |
4031 | } |
4032 | ||
2c46407d AG |
4033 | /** |
4034 | * receive_bitmap_plain | |
4035 | * | |
4036 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4037 | * code upon failure. | |
4038 | */ | |
4039 | static int | |
50d0b1ad | 4040 | receive_bitmap_plain(struct drbd_conf *mdev, unsigned int size, |
e658983a | 4041 | unsigned long *p, struct bm_xfer_ctx *c) |
b411b363 | 4042 | { |
50d0b1ad AG |
4043 | unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - |
4044 | drbd_header_size(mdev->tconn); | |
e658983a | 4045 | unsigned int num_words = min_t(size_t, data_size / sizeof(*p), |
50d0b1ad | 4046 | c->bm_words - c->word_offset); |
e658983a | 4047 | unsigned int want = num_words * sizeof(*p); |
2c46407d | 4048 | int err; |
b411b363 | 4049 | |
50d0b1ad AG |
4050 | if (want != size) { |
4051 | dev_err(DEV, "%s:want (%u) != size (%u)\n", __func__, want, size); | |
2c46407d | 4052 | return -EIO; |
b411b363 PR |
4053 | } |
4054 | if (want == 0) | |
2c46407d | 4055 | return 0; |
e658983a | 4056 | err = drbd_recv_all(mdev->tconn, p, want); |
82bc0194 | 4057 | if (err) |
2c46407d | 4058 | return err; |
b411b363 | 4059 | |
e658983a | 4060 | drbd_bm_merge_lel(mdev, c->word_offset, num_words, p); |
b411b363 PR |
4061 | |
4062 | c->word_offset += num_words; | |
4063 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
4064 | if (c->bit_offset > c->bm_bits) | |
4065 | c->bit_offset = c->bm_bits; | |
4066 | ||
2c46407d | 4067 | return 1; |
b411b363 PR |
4068 | } |
4069 | ||
a02d1240 AG |
4070 | static enum drbd_bitmap_code dcbp_get_code(struct p_compressed_bm *p) |
4071 | { | |
4072 | return (enum drbd_bitmap_code)(p->encoding & 0x0f); | |
4073 | } | |
4074 | ||
4075 | static int dcbp_get_start(struct p_compressed_bm *p) | |
4076 | { | |
4077 | return (p->encoding & 0x80) != 0; | |
4078 | } | |
4079 | ||
4080 | static int dcbp_get_pad_bits(struct p_compressed_bm *p) | |
4081 | { | |
4082 | return (p->encoding >> 4) & 0x7; | |
4083 | } | |
4084 | ||
2c46407d AG |
4085 | /** |
4086 | * recv_bm_rle_bits | |
4087 | * | |
4088 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4089 | * code upon failure. | |
4090 | */ | |
4091 | static int | |
b411b363 PR |
4092 | recv_bm_rle_bits(struct drbd_conf *mdev, |
4093 | struct p_compressed_bm *p, | |
c6d25cfe PR |
4094 | struct bm_xfer_ctx *c, |
4095 | unsigned int len) | |
b411b363 PR |
4096 | { |
4097 | struct bitstream bs; | |
4098 | u64 look_ahead; | |
4099 | u64 rl; | |
4100 | u64 tmp; | |
4101 | unsigned long s = c->bit_offset; | |
4102 | unsigned long e; | |
a02d1240 | 4103 | int toggle = dcbp_get_start(p); |
b411b363 PR |
4104 | int have; |
4105 | int bits; | |
4106 | ||
a02d1240 | 4107 | bitstream_init(&bs, p->code, len, dcbp_get_pad_bits(p)); |
b411b363 PR |
4108 | |
4109 | bits = bitstream_get_bits(&bs, &look_ahead, 64); | |
4110 | if (bits < 0) | |
2c46407d | 4111 | return -EIO; |
b411b363 PR |
4112 | |
4113 | for (have = bits; have > 0; s += rl, toggle = !toggle) { | |
4114 | bits = vli_decode_bits(&rl, look_ahead); | |
4115 | if (bits <= 0) | |
2c46407d | 4116 | return -EIO; |
b411b363 PR |
4117 | |
4118 | if (toggle) { | |
4119 | e = s + rl -1; | |
4120 | if (e >= c->bm_bits) { | |
4121 | dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); | |
2c46407d | 4122 | return -EIO; |
b411b363 PR |
4123 | } |
4124 | _drbd_bm_set_bits(mdev, s, e); | |
4125 | } | |
4126 | ||
4127 | if (have < bits) { | |
4128 | dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", | |
4129 | have, bits, look_ahead, | |
4130 | (unsigned int)(bs.cur.b - p->code), | |
4131 | (unsigned int)bs.buf_len); | |
2c46407d | 4132 | return -EIO; |
b411b363 PR |
4133 | } |
4134 | look_ahead >>= bits; | |
4135 | have -= bits; | |
4136 | ||
4137 | bits = bitstream_get_bits(&bs, &tmp, 64 - have); | |
4138 | if (bits < 0) | |
2c46407d | 4139 | return -EIO; |
b411b363 PR |
4140 | look_ahead |= tmp << have; |
4141 | have += bits; | |
4142 | } | |
4143 | ||
4144 | c->bit_offset = s; | |
4145 | bm_xfer_ctx_bit_to_word_offset(c); | |
4146 | ||
2c46407d | 4147 | return (s != c->bm_bits); |
b411b363 PR |
4148 | } |
4149 | ||
2c46407d AG |
4150 | /** |
4151 | * decode_bitmap_c | |
4152 | * | |
4153 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4154 | * code upon failure. | |
4155 | */ | |
4156 | static int | |
b411b363 PR |
4157 | decode_bitmap_c(struct drbd_conf *mdev, |
4158 | struct p_compressed_bm *p, | |
c6d25cfe PR |
4159 | struct bm_xfer_ctx *c, |
4160 | unsigned int len) | |
b411b363 | 4161 | { |
a02d1240 | 4162 | if (dcbp_get_code(p) == RLE_VLI_Bits) |
e658983a | 4163 | return recv_bm_rle_bits(mdev, p, c, len - sizeof(*p)); |
b411b363 PR |
4164 | |
4165 | /* other variants had been implemented for evaluation, | |
4166 | * but have been dropped as this one turned out to be "best" | |
4167 | * during all our tests. */ | |
4168 | ||
4169 | dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding); | |
38fa9988 | 4170 | conn_request_state(mdev->tconn, NS(conn, C_PROTOCOL_ERROR), CS_HARD); |
2c46407d | 4171 | return -EIO; |
b411b363 PR |
4172 | } |
4173 | ||
4174 | void INFO_bm_xfer_stats(struct drbd_conf *mdev, | |
4175 | const char *direction, struct bm_xfer_ctx *c) | |
4176 | { | |
4177 | /* what would it take to transfer it "plaintext" */ | |
50d0b1ad AG |
4178 | unsigned int header_size = drbd_header_size(mdev->tconn); |
4179 | unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; | |
4180 | unsigned int plain = | |
4181 | header_size * (DIV_ROUND_UP(c->bm_words, data_size) + 1) + | |
4182 | c->bm_words * sizeof(unsigned long); | |
4183 | unsigned int total = c->bytes[0] + c->bytes[1]; | |
4184 | unsigned int r; | |
b411b363 PR |
4185 | |
4186 | /* total can not be zero. but just in case: */ | |
4187 | if (total == 0) | |
4188 | return; | |
4189 | ||
4190 | /* don't report if not compressed */ | |
4191 | if (total >= plain) | |
4192 | return; | |
4193 | ||
4194 | /* total < plain. check for overflow, still */ | |
4195 | r = (total > UINT_MAX/1000) ? (total / (plain/1000)) | |
4196 | : (1000 * total / plain); | |
4197 | ||
4198 | if (r > 1000) | |
4199 | r = 1000; | |
4200 | ||
4201 | r = 1000 - r; | |
4202 | dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " | |
4203 | "total %u; compression: %u.%u%%\n", | |
4204 | direction, | |
4205 | c->bytes[1], c->packets[1], | |
4206 | c->bytes[0], c->packets[0], | |
4207 | total, r/10, r % 10); | |
4208 | } | |
4209 | ||
4210 | /* Since we are processing the bitfield from lower addresses to higher, | |
4211 | it does not matter if the process it in 32 bit chunks or 64 bit | |
4212 | chunks as long as it is little endian. (Understand it as byte stream, | |
4213 | beginning with the lowest byte...) If we would use big endian | |
4214 | we would need to process it from the highest address to the lowest, | |
4215 | in order to be agnostic to the 32 vs 64 bits issue. | |
4216 | ||
4217 | returns 0 on failure, 1 if we successfully received it. */ | |
4a76b161 | 4218 | static int receive_bitmap(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4219 | { |
4a76b161 | 4220 | struct drbd_conf *mdev; |
b411b363 | 4221 | struct bm_xfer_ctx c; |
2c46407d | 4222 | int err; |
4a76b161 AG |
4223 | |
4224 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
4225 | if (!mdev) | |
4226 | return -EIO; | |
b411b363 | 4227 | |
20ceb2b2 LE |
4228 | drbd_bm_lock(mdev, "receive bitmap", BM_LOCKED_SET_ALLOWED); |
4229 | /* you are supposed to send additional out-of-sync information | |
4230 | * if you actually set bits during this phase */ | |
b411b363 | 4231 | |
b411b363 PR |
4232 | c = (struct bm_xfer_ctx) { |
4233 | .bm_bits = drbd_bm_bits(mdev), | |
4234 | .bm_words = drbd_bm_words(mdev), | |
4235 | }; | |
4236 | ||
2c46407d | 4237 | for(;;) { |
e658983a AG |
4238 | if (pi->cmd == P_BITMAP) |
4239 | err = receive_bitmap_plain(mdev, pi->size, pi->data, &c); | |
4240 | else if (pi->cmd == P_COMPRESSED_BITMAP) { | |
b411b363 PR |
4241 | /* MAYBE: sanity check that we speak proto >= 90, |
4242 | * and the feature is enabled! */ | |
e658983a | 4243 | struct p_compressed_bm *p = pi->data; |
b411b363 | 4244 | |
50d0b1ad | 4245 | if (pi->size > DRBD_SOCKET_BUFFER_SIZE - drbd_header_size(tconn)) { |
b411b363 | 4246 | dev_err(DEV, "ReportCBitmap packet too large\n"); |
82bc0194 | 4247 | err = -EIO; |
b411b363 PR |
4248 | goto out; |
4249 | } | |
e658983a | 4250 | if (pi->size <= sizeof(*p)) { |
e2857216 | 4251 | dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", pi->size); |
82bc0194 | 4252 | err = -EIO; |
78fcbdae | 4253 | goto out; |
b411b363 | 4254 | } |
e658983a AG |
4255 | err = drbd_recv_all(mdev->tconn, p, pi->size); |
4256 | if (err) | |
4257 | goto out; | |
e2857216 | 4258 | err = decode_bitmap_c(mdev, p, &c, pi->size); |
b411b363 | 4259 | } else { |
e2857216 | 4260 | dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", pi->cmd); |
82bc0194 | 4261 | err = -EIO; |
b411b363 PR |
4262 | goto out; |
4263 | } | |
4264 | ||
e2857216 | 4265 | c.packets[pi->cmd == P_BITMAP]++; |
50d0b1ad | 4266 | c.bytes[pi->cmd == P_BITMAP] += drbd_header_size(tconn) + pi->size; |
b411b363 | 4267 | |
2c46407d AG |
4268 | if (err <= 0) { |
4269 | if (err < 0) | |
4270 | goto out; | |
b411b363 | 4271 | break; |
2c46407d | 4272 | } |
e2857216 | 4273 | err = drbd_recv_header(mdev->tconn, pi); |
82bc0194 | 4274 | if (err) |
b411b363 | 4275 | goto out; |
2c46407d | 4276 | } |
b411b363 PR |
4277 | |
4278 | INFO_bm_xfer_stats(mdev, "receive", &c); | |
4279 | ||
4280 | if (mdev->state.conn == C_WF_BITMAP_T) { | |
de1f8e4a AG |
4281 | enum drbd_state_rv rv; |
4282 | ||
82bc0194 AG |
4283 | err = drbd_send_bitmap(mdev); |
4284 | if (err) | |
b411b363 PR |
4285 | goto out; |
4286 | /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ | |
de1f8e4a AG |
4287 | rv = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); |
4288 | D_ASSERT(rv == SS_SUCCESS); | |
b411b363 PR |
4289 | } else if (mdev->state.conn != C_WF_BITMAP_S) { |
4290 | /* admin may have requested C_DISCONNECTING, | |
4291 | * other threads may have noticed network errors */ | |
4292 | dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n", | |
4293 | drbd_conn_str(mdev->state.conn)); | |
4294 | } | |
82bc0194 | 4295 | err = 0; |
b411b363 | 4296 | |
b411b363 | 4297 | out: |
20ceb2b2 | 4298 | drbd_bm_unlock(mdev); |
82bc0194 | 4299 | if (!err && mdev->state.conn == C_WF_BITMAP_S) |
b411b363 | 4300 | drbd_start_resync(mdev, C_SYNC_SOURCE); |
82bc0194 | 4301 | return err; |
b411b363 PR |
4302 | } |
4303 | ||
4a76b161 | 4304 | static int receive_skip(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4305 | { |
4a76b161 | 4306 | conn_warn(tconn, "skipping unknown optional packet type %d, l: %d!\n", |
e2857216 | 4307 | pi->cmd, pi->size); |
b411b363 | 4308 | |
4a76b161 | 4309 | return ignore_remaining_packet(tconn, pi); |
b411b363 PR |
4310 | } |
4311 | ||
4a76b161 | 4312 | static int receive_UnplugRemote(struct drbd_tconn *tconn, struct packet_info *pi) |
0ced55a3 | 4313 | { |
e7f52dfb LE |
4314 | /* Make sure we've acked all the TCP data associated |
4315 | * with the data requests being unplugged */ | |
4a76b161 | 4316 | drbd_tcp_quickack(tconn->data.socket); |
0ced55a3 | 4317 | |
82bc0194 | 4318 | return 0; |
0ced55a3 PR |
4319 | } |
4320 | ||
4a76b161 | 4321 | static int receive_out_of_sync(struct drbd_tconn *tconn, struct packet_info *pi) |
73a01a18 | 4322 | { |
4a76b161 | 4323 | struct drbd_conf *mdev; |
e658983a | 4324 | struct p_block_desc *p = pi->data; |
4a76b161 AG |
4325 | |
4326 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
4327 | if (!mdev) | |
4328 | return -EIO; | |
73a01a18 | 4329 | |
f735e363 LE |
4330 | switch (mdev->state.conn) { |
4331 | case C_WF_SYNC_UUID: | |
4332 | case C_WF_BITMAP_T: | |
4333 | case C_BEHIND: | |
4334 | break; | |
4335 | default: | |
4336 | dev_err(DEV, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n", | |
4337 | drbd_conn_str(mdev->state.conn)); | |
4338 | } | |
4339 | ||
73a01a18 PR |
4340 | drbd_set_out_of_sync(mdev, be64_to_cpu(p->sector), be32_to_cpu(p->blksize)); |
4341 | ||
82bc0194 | 4342 | return 0; |
73a01a18 PR |
4343 | } |
4344 | ||
02918be2 PR |
4345 | struct data_cmd { |
4346 | int expect_payload; | |
4347 | size_t pkt_size; | |
4a76b161 | 4348 | int (*fn)(struct drbd_tconn *, struct packet_info *); |
02918be2 PR |
4349 | }; |
4350 | ||
4351 | static struct data_cmd drbd_cmd_handler[] = { | |
4352 | [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, | |
4353 | [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, | |
4354 | [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , | |
4355 | [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , | |
e658983a AG |
4356 | [P_BITMAP] = { 1, 0, receive_bitmap } , |
4357 | [P_COMPRESSED_BITMAP] = { 1, 0, receive_bitmap } , | |
4358 | [P_UNPLUG_REMOTE] = { 0, 0, receive_UnplugRemote }, | |
02918be2 PR |
4359 | [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, |
4360 | [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
e658983a AG |
4361 | [P_SYNC_PARAM] = { 1, 0, receive_SyncParam }, |
4362 | [P_SYNC_PARAM89] = { 1, 0, receive_SyncParam }, | |
02918be2 PR |
4363 | [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, |
4364 | [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, | |
4365 | [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, | |
4366 | [P_STATE] = { 0, sizeof(struct p_state), receive_state }, | |
4367 | [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, | |
4368 | [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, | |
4369 | [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
4370 | [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
4371 | [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
4372 | [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, | |
73a01a18 | 4373 | [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync }, |
4a76b161 | 4374 | [P_CONN_ST_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_conn_state }, |
036b17ea | 4375 | [P_PROTOCOL_UPDATE] = { 1, sizeof(struct p_protocol), receive_protocol }, |
b411b363 PR |
4376 | }; |
4377 | ||
eefc2f7d | 4378 | static void drbdd(struct drbd_tconn *tconn) |
b411b363 | 4379 | { |
77351055 | 4380 | struct packet_info pi; |
02918be2 | 4381 | size_t shs; /* sub header size */ |
82bc0194 | 4382 | int err; |
b411b363 | 4383 | |
eefc2f7d | 4384 | while (get_t_state(&tconn->receiver) == RUNNING) { |
deebe195 | 4385 | struct data_cmd *cmd; |
b411b363 | 4386 | |
eefc2f7d | 4387 | drbd_thread_current_set_cpu(&tconn->receiver); |
69bc7bc3 | 4388 | if (drbd_recv_header(tconn, &pi)) |
02918be2 | 4389 | goto err_out; |
b411b363 | 4390 | |
deebe195 | 4391 | cmd = &drbd_cmd_handler[pi.cmd]; |
4a76b161 | 4392 | if (unlikely(pi.cmd >= ARRAY_SIZE(drbd_cmd_handler) || !cmd->fn)) { |
2fcb8f30 AG |
4393 | conn_err(tconn, "Unexpected data packet %s (0x%04x)", |
4394 | cmdname(pi.cmd), pi.cmd); | |
02918be2 | 4395 | goto err_out; |
0b33a916 | 4396 | } |
b411b363 | 4397 | |
e658983a AG |
4398 | shs = cmd->pkt_size; |
4399 | if (pi.size > shs && !cmd->expect_payload) { | |
2fcb8f30 AG |
4400 | conn_err(tconn, "No payload expected %s l:%d\n", |
4401 | cmdname(pi.cmd), pi.size); | |
02918be2 | 4402 | goto err_out; |
b411b363 | 4403 | } |
b411b363 | 4404 | |
c13f7e1a | 4405 | if (shs) { |
e658983a | 4406 | err = drbd_recv_all_warn(tconn, pi.data, shs); |
a5c31904 | 4407 | if (err) |
c13f7e1a | 4408 | goto err_out; |
e2857216 | 4409 | pi.size -= shs; |
c13f7e1a LE |
4410 | } |
4411 | ||
4a76b161 AG |
4412 | err = cmd->fn(tconn, &pi); |
4413 | if (err) { | |
9f5bdc33 AG |
4414 | conn_err(tconn, "error receiving %s, e: %d l: %d!\n", |
4415 | cmdname(pi.cmd), err, pi.size); | |
02918be2 | 4416 | goto err_out; |
b411b363 PR |
4417 | } |
4418 | } | |
82bc0194 | 4419 | return; |
b411b363 | 4420 | |
82bc0194 AG |
4421 | err_out: |
4422 | conn_request_state(tconn, NS(conn, C_PROTOCOL_ERROR), CS_HARD); | |
b411b363 PR |
4423 | } |
4424 | ||
0e29d163 | 4425 | void conn_flush_workqueue(struct drbd_tconn *tconn) |
b411b363 PR |
4426 | { |
4427 | struct drbd_wq_barrier barr; | |
4428 | ||
4429 | barr.w.cb = w_prev_work_done; | |
0e29d163 | 4430 | barr.w.tconn = tconn; |
b411b363 | 4431 | init_completion(&barr.done); |
d5b27b01 | 4432 | drbd_queue_work(&tconn->sender_work, &barr.w); |
b411b363 PR |
4433 | wait_for_completion(&barr.done); |
4434 | } | |
4435 | ||
81fa2e67 | 4436 | static void conn_disconnect(struct drbd_tconn *tconn) |
b411b363 | 4437 | { |
c141ebda | 4438 | struct drbd_conf *mdev; |
bbeb641c | 4439 | enum drbd_conns oc; |
376694a0 | 4440 | int vnr; |
b411b363 | 4441 | |
bbeb641c | 4442 | if (tconn->cstate == C_STANDALONE) |
b411b363 | 4443 | return; |
b411b363 | 4444 | |
545752d5 LE |
4445 | /* We are about to start the cleanup after connection loss. |
4446 | * Make sure drbd_make_request knows about that. | |
4447 | * Usually we should be in some network failure state already, | |
4448 | * but just in case we are not, we fix it up here. | |
4449 | */ | |
b8853dbd | 4450 | conn_request_state(tconn, NS(conn, C_NETWORK_FAILURE), CS_HARD); |
545752d5 | 4451 | |
b411b363 | 4452 | /* asender does not clean up anything. it must not interfere, either */ |
360cc740 PR |
4453 | drbd_thread_stop(&tconn->asender); |
4454 | drbd_free_sock(tconn); | |
4455 | ||
c141ebda PR |
4456 | rcu_read_lock(); |
4457 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { | |
4458 | kref_get(&mdev->kref); | |
4459 | rcu_read_unlock(); | |
4460 | drbd_disconnected(mdev); | |
4461 | kref_put(&mdev->kref, &drbd_minor_destroy); | |
4462 | rcu_read_lock(); | |
4463 | } | |
4464 | rcu_read_unlock(); | |
4465 | ||
12038a3a PR |
4466 | if (!list_empty(&tconn->current_epoch->list)) |
4467 | conn_err(tconn, "ASSERTION FAILED: tconn->current_epoch->list not empty\n"); | |
4468 | /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ | |
4469 | atomic_set(&tconn->current_epoch->epoch_size, 0); | |
b6dd1a89 | 4470 | tconn->send.seen_any_write_yet = false; |
12038a3a | 4471 | |
360cc740 PR |
4472 | conn_info(tconn, "Connection closed\n"); |
4473 | ||
cb703454 PR |
4474 | if (conn_highest_role(tconn) == R_PRIMARY && conn_highest_pdsk(tconn) >= D_UNKNOWN) |
4475 | conn_try_outdate_peer_async(tconn); | |
4476 | ||
360cc740 | 4477 | spin_lock_irq(&tconn->req_lock); |
bbeb641c PR |
4478 | oc = tconn->cstate; |
4479 | if (oc >= C_UNCONNECTED) | |
376694a0 | 4480 | _conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE); |
bbeb641c | 4481 | |
360cc740 PR |
4482 | spin_unlock_irq(&tconn->req_lock); |
4483 | ||
f3dfa40a | 4484 | if (oc == C_DISCONNECTING) |
d9cc6e23 | 4485 | conn_request_state(tconn, NS(conn, C_STANDALONE), CS_VERBOSE | CS_HARD); |
360cc740 PR |
4486 | } |
4487 | ||
c141ebda | 4488 | static int drbd_disconnected(struct drbd_conf *mdev) |
360cc740 | 4489 | { |
360cc740 | 4490 | unsigned int i; |
b411b363 | 4491 | |
85719573 | 4492 | /* wait for current activity to cease. */ |
87eeee41 | 4493 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4494 | _drbd_wait_ee_list_empty(mdev, &mdev->active_ee); |
4495 | _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee); | |
4496 | _drbd_wait_ee_list_empty(mdev, &mdev->read_ee); | |
87eeee41 | 4497 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4498 | |
4499 | /* We do not have data structures that would allow us to | |
4500 | * get the rs_pending_cnt down to 0 again. | |
4501 | * * On C_SYNC_TARGET we do not have any data structures describing | |
4502 | * the pending RSDataRequest's we have sent. | |
4503 | * * On C_SYNC_SOURCE there is no data structure that tracks | |
4504 | * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. | |
4505 | * And no, it is not the sum of the reference counts in the | |
4506 | * resync_LRU. The resync_LRU tracks the whole operation including | |
4507 | * the disk-IO, while the rs_pending_cnt only tracks the blocks | |
4508 | * on the fly. */ | |
4509 | drbd_rs_cancel_all(mdev); | |
4510 | mdev->rs_total = 0; | |
4511 | mdev->rs_failed = 0; | |
4512 | atomic_set(&mdev->rs_pending_cnt, 0); | |
4513 | wake_up(&mdev->misc_wait); | |
4514 | ||
b411b363 | 4515 | del_timer_sync(&mdev->resync_timer); |
b411b363 PR |
4516 | resync_timer_fn((unsigned long)mdev); |
4517 | ||
b411b363 PR |
4518 | /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, |
4519 | * w_make_resync_request etc. which may still be on the worker queue | |
4520 | * to be "canceled" */ | |
4521 | drbd_flush_workqueue(mdev); | |
4522 | ||
a990be46 | 4523 | drbd_finish_peer_reqs(mdev); |
b411b363 | 4524 | |
d10b4ea3 PR |
4525 | /* This second workqueue flush is necessary, since drbd_finish_peer_reqs() |
4526 | might have issued a work again. The one before drbd_finish_peer_reqs() is | |
4527 | necessary to reclain net_ee in drbd_finish_peer_reqs(). */ | |
4528 | drbd_flush_workqueue(mdev); | |
4529 | ||
08332d73 LE |
4530 | /* need to do it again, drbd_finish_peer_reqs() may have populated it |
4531 | * again via drbd_try_clear_on_disk_bm(). */ | |
4532 | drbd_rs_cancel_all(mdev); | |
b411b363 PR |
4533 | |
4534 | kfree(mdev->p_uuid); | |
4535 | mdev->p_uuid = NULL; | |
4536 | ||
2aebfabb | 4537 | if (!drbd_suspended(mdev)) |
2f5cdd0b | 4538 | tl_clear(mdev->tconn); |
b411b363 PR |
4539 | |
4540 | drbd_md_sync(mdev); | |
4541 | ||
20ceb2b2 LE |
4542 | /* serialize with bitmap writeout triggered by the state change, |
4543 | * if any. */ | |
4544 | wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags)); | |
4545 | ||
b411b363 PR |
4546 | /* tcp_close and release of sendpage pages can be deferred. I don't |
4547 | * want to use SO_LINGER, because apparently it can be deferred for | |
4548 | * more than 20 seconds (longest time I checked). | |
4549 | * | |
4550 | * Actually we don't care for exactly when the network stack does its | |
4551 | * put_page(), but release our reference on these pages right here. | |
4552 | */ | |
7721f567 | 4553 | i = drbd_free_peer_reqs(mdev, &mdev->net_ee); |
b411b363 PR |
4554 | if (i) |
4555 | dev_info(DEV, "net_ee not empty, killed %u entries\n", i); | |
435f0740 LE |
4556 | i = atomic_read(&mdev->pp_in_use_by_net); |
4557 | if (i) | |
4558 | dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i); | |
b411b363 PR |
4559 | i = atomic_read(&mdev->pp_in_use); |
4560 | if (i) | |
45bb912b | 4561 | dev_info(DEV, "pp_in_use = %d, expected 0\n", i); |
b411b363 PR |
4562 | |
4563 | D_ASSERT(list_empty(&mdev->read_ee)); | |
4564 | D_ASSERT(list_empty(&mdev->active_ee)); | |
4565 | D_ASSERT(list_empty(&mdev->sync_ee)); | |
4566 | D_ASSERT(list_empty(&mdev->done_ee)); | |
4567 | ||
360cc740 | 4568 | return 0; |
b411b363 PR |
4569 | } |
4570 | ||
4571 | /* | |
4572 | * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version | |
4573 | * we can agree on is stored in agreed_pro_version. | |
4574 | * | |
4575 | * feature flags and the reserved array should be enough room for future | |
4576 | * enhancements of the handshake protocol, and possible plugins... | |
4577 | * | |
4578 | * for now, they are expected to be zero, but ignored. | |
4579 | */ | |
6038178e | 4580 | static int drbd_send_features(struct drbd_tconn *tconn) |
b411b363 | 4581 | { |
9f5bdc33 AG |
4582 | struct drbd_socket *sock; |
4583 | struct p_connection_features *p; | |
b411b363 | 4584 | |
9f5bdc33 AG |
4585 | sock = &tconn->data; |
4586 | p = conn_prepare_command(tconn, sock); | |
4587 | if (!p) | |
e8d17b01 | 4588 | return -EIO; |
b411b363 PR |
4589 | memset(p, 0, sizeof(*p)); |
4590 | p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); | |
4591 | p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); | |
9f5bdc33 | 4592 | return conn_send_command(tconn, sock, P_CONNECTION_FEATURES, sizeof(*p), NULL, 0); |
b411b363 PR |
4593 | } |
4594 | ||
4595 | /* | |
4596 | * return values: | |
4597 | * 1 yes, we have a valid connection | |
4598 | * 0 oops, did not work out, please try again | |
4599 | * -1 peer talks different language, | |
4600 | * no point in trying again, please go standalone. | |
4601 | */ | |
6038178e | 4602 | static int drbd_do_features(struct drbd_tconn *tconn) |
b411b363 | 4603 | { |
65d11ed6 | 4604 | /* ASSERT current == tconn->receiver ... */ |
e658983a AG |
4605 | struct p_connection_features *p; |
4606 | const int expect = sizeof(struct p_connection_features); | |
77351055 | 4607 | struct packet_info pi; |
a5c31904 | 4608 | int err; |
b411b363 | 4609 | |
6038178e | 4610 | err = drbd_send_features(tconn); |
e8d17b01 | 4611 | if (err) |
b411b363 PR |
4612 | return 0; |
4613 | ||
69bc7bc3 AG |
4614 | err = drbd_recv_header(tconn, &pi); |
4615 | if (err) | |
b411b363 PR |
4616 | return 0; |
4617 | ||
6038178e AG |
4618 | if (pi.cmd != P_CONNECTION_FEATURES) { |
4619 | conn_err(tconn, "expected ConnectionFeatures packet, received: %s (0x%04x)\n", | |
2fcb8f30 | 4620 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4621 | return -1; |
4622 | } | |
4623 | ||
77351055 | 4624 | if (pi.size != expect) { |
6038178e | 4625 | conn_err(tconn, "expected ConnectionFeatures length: %u, received: %u\n", |
77351055 | 4626 | expect, pi.size); |
b411b363 PR |
4627 | return -1; |
4628 | } | |
4629 | ||
e658983a AG |
4630 | p = pi.data; |
4631 | err = drbd_recv_all_warn(tconn, p, expect); | |
a5c31904 | 4632 | if (err) |
b411b363 | 4633 | return 0; |
b411b363 | 4634 | |
b411b363 PR |
4635 | p->protocol_min = be32_to_cpu(p->protocol_min); |
4636 | p->protocol_max = be32_to_cpu(p->protocol_max); | |
4637 | if (p->protocol_max == 0) | |
4638 | p->protocol_max = p->protocol_min; | |
4639 | ||
4640 | if (PRO_VERSION_MAX < p->protocol_min || | |
4641 | PRO_VERSION_MIN > p->protocol_max) | |
4642 | goto incompat; | |
4643 | ||
65d11ed6 | 4644 | tconn->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); |
b411b363 | 4645 | |
65d11ed6 PR |
4646 | conn_info(tconn, "Handshake successful: " |
4647 | "Agreed network protocol version %d\n", tconn->agreed_pro_version); | |
b411b363 PR |
4648 | |
4649 | return 1; | |
4650 | ||
4651 | incompat: | |
65d11ed6 | 4652 | conn_err(tconn, "incompatible DRBD dialects: " |
b411b363 PR |
4653 | "I support %d-%d, peer supports %d-%d\n", |
4654 | PRO_VERSION_MIN, PRO_VERSION_MAX, | |
4655 | p->protocol_min, p->protocol_max); | |
4656 | return -1; | |
4657 | } | |
4658 | ||
4659 | #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) | |
13e6037d | 4660 | static int drbd_do_auth(struct drbd_tconn *tconn) |
b411b363 PR |
4661 | { |
4662 | dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); | |
4663 | dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); | |
b10d96cb | 4664 | return -1; |
b411b363 PR |
4665 | } |
4666 | #else | |
4667 | #define CHALLENGE_LEN 64 | |
b10d96cb JT |
4668 | |
4669 | /* Return value: | |
4670 | 1 - auth succeeded, | |
4671 | 0 - failed, try again (network error), | |
4672 | -1 - auth failed, don't try again. | |
4673 | */ | |
4674 | ||
13e6037d | 4675 | static int drbd_do_auth(struct drbd_tconn *tconn) |
b411b363 | 4676 | { |
9f5bdc33 | 4677 | struct drbd_socket *sock; |
b411b363 PR |
4678 | char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ |
4679 | struct scatterlist sg; | |
4680 | char *response = NULL; | |
4681 | char *right_response = NULL; | |
4682 | char *peers_ch = NULL; | |
44ed167d PR |
4683 | unsigned int key_len; |
4684 | char secret[SHARED_SECRET_MAX]; /* 64 byte */ | |
b411b363 PR |
4685 | unsigned int resp_size; |
4686 | struct hash_desc desc; | |
77351055 | 4687 | struct packet_info pi; |
44ed167d | 4688 | struct net_conf *nc; |
69bc7bc3 | 4689 | int err, rv; |
b411b363 | 4690 | |
9f5bdc33 | 4691 | /* FIXME: Put the challenge/response into the preallocated socket buffer. */ |
b411b363 | 4692 | |
44ed167d PR |
4693 | rcu_read_lock(); |
4694 | nc = rcu_dereference(tconn->net_conf); | |
4695 | key_len = strlen(nc->shared_secret); | |
4696 | memcpy(secret, nc->shared_secret, key_len); | |
4697 | rcu_read_unlock(); | |
4698 | ||
13e6037d | 4699 | desc.tfm = tconn->cram_hmac_tfm; |
b411b363 PR |
4700 | desc.flags = 0; |
4701 | ||
44ed167d | 4702 | rv = crypto_hash_setkey(tconn->cram_hmac_tfm, (u8 *)secret, key_len); |
b411b363 | 4703 | if (rv) { |
13e6037d | 4704 | conn_err(tconn, "crypto_hash_setkey() failed with %d\n", rv); |
b10d96cb | 4705 | rv = -1; |
b411b363 PR |
4706 | goto fail; |
4707 | } | |
4708 | ||
4709 | get_random_bytes(my_challenge, CHALLENGE_LEN); | |
4710 | ||
9f5bdc33 AG |
4711 | sock = &tconn->data; |
4712 | if (!conn_prepare_command(tconn, sock)) { | |
4713 | rv = 0; | |
4714 | goto fail; | |
4715 | } | |
e658983a | 4716 | rv = !conn_send_command(tconn, sock, P_AUTH_CHALLENGE, 0, |
9f5bdc33 | 4717 | my_challenge, CHALLENGE_LEN); |
b411b363 PR |
4718 | if (!rv) |
4719 | goto fail; | |
4720 | ||
69bc7bc3 AG |
4721 | err = drbd_recv_header(tconn, &pi); |
4722 | if (err) { | |
4723 | rv = 0; | |
b411b363 | 4724 | goto fail; |
69bc7bc3 | 4725 | } |
b411b363 | 4726 | |
77351055 | 4727 | if (pi.cmd != P_AUTH_CHALLENGE) { |
13e6037d | 4728 | conn_err(tconn, "expected AuthChallenge packet, received: %s (0x%04x)\n", |
2fcb8f30 | 4729 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4730 | rv = 0; |
4731 | goto fail; | |
4732 | } | |
4733 | ||
77351055 | 4734 | if (pi.size > CHALLENGE_LEN * 2) { |
13e6037d | 4735 | conn_err(tconn, "expected AuthChallenge payload too big.\n"); |
b10d96cb | 4736 | rv = -1; |
b411b363 PR |
4737 | goto fail; |
4738 | } | |
4739 | ||
77351055 | 4740 | peers_ch = kmalloc(pi.size, GFP_NOIO); |
b411b363 | 4741 | if (peers_ch == NULL) { |
13e6037d | 4742 | conn_err(tconn, "kmalloc of peers_ch failed\n"); |
b10d96cb | 4743 | rv = -1; |
b411b363 PR |
4744 | goto fail; |
4745 | } | |
4746 | ||
a5c31904 AG |
4747 | err = drbd_recv_all_warn(tconn, peers_ch, pi.size); |
4748 | if (err) { | |
b411b363 PR |
4749 | rv = 0; |
4750 | goto fail; | |
4751 | } | |
4752 | ||
13e6037d | 4753 | resp_size = crypto_hash_digestsize(tconn->cram_hmac_tfm); |
b411b363 PR |
4754 | response = kmalloc(resp_size, GFP_NOIO); |
4755 | if (response == NULL) { | |
13e6037d | 4756 | conn_err(tconn, "kmalloc of response failed\n"); |
b10d96cb | 4757 | rv = -1; |
b411b363 PR |
4758 | goto fail; |
4759 | } | |
4760 | ||
4761 | sg_init_table(&sg, 1); | |
77351055 | 4762 | sg_set_buf(&sg, peers_ch, pi.size); |
b411b363 PR |
4763 | |
4764 | rv = crypto_hash_digest(&desc, &sg, sg.length, response); | |
4765 | if (rv) { | |
13e6037d | 4766 | conn_err(tconn, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 4767 | rv = -1; |
b411b363 PR |
4768 | goto fail; |
4769 | } | |
4770 | ||
9f5bdc33 AG |
4771 | if (!conn_prepare_command(tconn, sock)) { |
4772 | rv = 0; | |
b411b363 | 4773 | goto fail; |
9f5bdc33 | 4774 | } |
e658983a | 4775 | rv = !conn_send_command(tconn, sock, P_AUTH_RESPONSE, 0, |
9f5bdc33 | 4776 | response, resp_size); |
b411b363 PR |
4777 | if (!rv) |
4778 | goto fail; | |
4779 | ||
69bc7bc3 AG |
4780 | err = drbd_recv_header(tconn, &pi); |
4781 | if (err) { | |
b411b363 PR |
4782 | rv = 0; |
4783 | goto fail; | |
4784 | } | |
4785 | ||
77351055 | 4786 | if (pi.cmd != P_AUTH_RESPONSE) { |
13e6037d | 4787 | conn_err(tconn, "expected AuthResponse packet, received: %s (0x%04x)\n", |
2fcb8f30 | 4788 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4789 | rv = 0; |
4790 | goto fail; | |
4791 | } | |
4792 | ||
77351055 | 4793 | if (pi.size != resp_size) { |
13e6037d | 4794 | conn_err(tconn, "expected AuthResponse payload of wrong size\n"); |
b411b363 PR |
4795 | rv = 0; |
4796 | goto fail; | |
4797 | } | |
b411b363 | 4798 | |
a5c31904 AG |
4799 | err = drbd_recv_all_warn(tconn, response , resp_size); |
4800 | if (err) { | |
b411b363 PR |
4801 | rv = 0; |
4802 | goto fail; | |
4803 | } | |
4804 | ||
4805 | right_response = kmalloc(resp_size, GFP_NOIO); | |
2d1ee87d | 4806 | if (right_response == NULL) { |
13e6037d | 4807 | conn_err(tconn, "kmalloc of right_response failed\n"); |
b10d96cb | 4808 | rv = -1; |
b411b363 PR |
4809 | goto fail; |
4810 | } | |
4811 | ||
4812 | sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); | |
4813 | ||
4814 | rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); | |
4815 | if (rv) { | |
13e6037d | 4816 | conn_err(tconn, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 4817 | rv = -1; |
b411b363 PR |
4818 | goto fail; |
4819 | } | |
4820 | ||
4821 | rv = !memcmp(response, right_response, resp_size); | |
4822 | ||
4823 | if (rv) | |
44ed167d PR |
4824 | conn_info(tconn, "Peer authenticated using %d bytes HMAC\n", |
4825 | resp_size); | |
b10d96cb JT |
4826 | else |
4827 | rv = -1; | |
b411b363 PR |
4828 | |
4829 | fail: | |
4830 | kfree(peers_ch); | |
4831 | kfree(response); | |
4832 | kfree(right_response); | |
4833 | ||
4834 | return rv; | |
4835 | } | |
4836 | #endif | |
4837 | ||
4838 | int drbdd_init(struct drbd_thread *thi) | |
4839 | { | |
392c8801 | 4840 | struct drbd_tconn *tconn = thi->tconn; |
b411b363 PR |
4841 | int h; |
4842 | ||
4d641dd7 | 4843 | conn_info(tconn, "receiver (re)started\n"); |
b411b363 PR |
4844 | |
4845 | do { | |
81fa2e67 | 4846 | h = conn_connect(tconn); |
b411b363 | 4847 | if (h == 0) { |
81fa2e67 | 4848 | conn_disconnect(tconn); |
20ee6390 | 4849 | schedule_timeout_interruptible(HZ); |
b411b363 PR |
4850 | } |
4851 | if (h == -1) { | |
4d641dd7 | 4852 | conn_warn(tconn, "Discarding network configuration.\n"); |
bbeb641c | 4853 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
4854 | } |
4855 | } while (h == 0); | |
4856 | ||
91fd4dad PR |
4857 | if (h > 0) |
4858 | drbdd(tconn); | |
b411b363 | 4859 | |
81fa2e67 | 4860 | conn_disconnect(tconn); |
b411b363 | 4861 | |
4d641dd7 | 4862 | conn_info(tconn, "receiver terminated\n"); |
b411b363 PR |
4863 | return 0; |
4864 | } | |
4865 | ||
4866 | /* ********* acknowledge sender ******** */ | |
4867 | ||
e05e1e59 | 4868 | static int got_conn_RqSReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4869 | { |
e658983a | 4870 | struct p_req_state_reply *p = pi->data; |
e4f78ede PR |
4871 | int retcode = be32_to_cpu(p->retcode); |
4872 | ||
4873 | if (retcode >= SS_SUCCESS) { | |
4874 | set_bit(CONN_WD_ST_CHG_OKAY, &tconn->flags); | |
4875 | } else { | |
4876 | set_bit(CONN_WD_ST_CHG_FAIL, &tconn->flags); | |
4877 | conn_err(tconn, "Requested state change failed by peer: %s (%d)\n", | |
4878 | drbd_set_st_err_str(retcode), retcode); | |
4879 | } | |
4880 | wake_up(&tconn->ping_wait); | |
4881 | ||
2735a594 | 4882 | return 0; |
e4f78ede | 4883 | } |
b411b363 | 4884 | |
1952e916 | 4885 | static int got_RqSReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4886 | { |
1952e916 | 4887 | struct drbd_conf *mdev; |
e658983a | 4888 | struct p_req_state_reply *p = pi->data; |
b411b363 PR |
4889 | int retcode = be32_to_cpu(p->retcode); |
4890 | ||
1952e916 AG |
4891 | mdev = vnr_to_mdev(tconn, pi->vnr); |
4892 | if (!mdev) | |
2735a594 | 4893 | return -EIO; |
1952e916 | 4894 | |
4d0fc3fd PR |
4895 | if (test_bit(CONN_WD_ST_CHG_REQ, &tconn->flags)) { |
4896 | D_ASSERT(tconn->agreed_pro_version < 100); | |
4897 | return got_conn_RqSReply(tconn, pi); | |
4898 | } | |
4899 | ||
b411b363 | 4900 | if (retcode >= SS_SUCCESS) { |
e4f78ede | 4901 | set_bit(CL_ST_CHG_SUCCESS, &mdev->flags); |
b411b363 | 4902 | } else { |
e4f78ede | 4903 | set_bit(CL_ST_CHG_FAIL, &mdev->flags); |
b411b363 | 4904 | dev_err(DEV, "Requested state change failed by peer: %s (%d)\n", |
e4f78ede | 4905 | drbd_set_st_err_str(retcode), retcode); |
b411b363 PR |
4906 | } |
4907 | wake_up(&mdev->state_wait); | |
4908 | ||
2735a594 | 4909 | return 0; |
b411b363 PR |
4910 | } |
4911 | ||
e05e1e59 | 4912 | static int got_Ping(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4913 | { |
2735a594 | 4914 | return drbd_send_ping_ack(tconn); |
b411b363 PR |
4915 | |
4916 | } | |
4917 | ||
e05e1e59 | 4918 | static int got_PingAck(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 PR |
4919 | { |
4920 | /* restore idle timeout */ | |
2a67d8b9 PR |
4921 | tconn->meta.socket->sk->sk_rcvtimeo = tconn->net_conf->ping_int*HZ; |
4922 | if (!test_and_set_bit(GOT_PING_ACK, &tconn->flags)) | |
4923 | wake_up(&tconn->ping_wait); | |
b411b363 | 4924 | |
2735a594 | 4925 | return 0; |
b411b363 PR |
4926 | } |
4927 | ||
1952e916 | 4928 | static int got_IsInSync(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4929 | { |
1952e916 | 4930 | struct drbd_conf *mdev; |
e658983a | 4931 | struct p_block_ack *p = pi->data; |
b411b363 PR |
4932 | sector_t sector = be64_to_cpu(p->sector); |
4933 | int blksize = be32_to_cpu(p->blksize); | |
4934 | ||
1952e916 AG |
4935 | mdev = vnr_to_mdev(tconn, pi->vnr); |
4936 | if (!mdev) | |
2735a594 | 4937 | return -EIO; |
1952e916 | 4938 | |
31890f4a | 4939 | D_ASSERT(mdev->tconn->agreed_pro_version >= 89); |
b411b363 PR |
4940 | |
4941 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4942 | ||
1d53f09e LE |
4943 | if (get_ldev(mdev)) { |
4944 | drbd_rs_complete_io(mdev, sector); | |
4945 | drbd_set_in_sync(mdev, sector, blksize); | |
4946 | /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ | |
4947 | mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); | |
4948 | put_ldev(mdev); | |
4949 | } | |
b411b363 | 4950 | dec_rs_pending(mdev); |
778f271d | 4951 | atomic_add(blksize >> 9, &mdev->rs_sect_in); |
b411b363 | 4952 | |
2735a594 | 4953 | return 0; |
b411b363 PR |
4954 | } |
4955 | ||
bc9c5c41 AG |
4956 | static int |
4957 | validate_req_change_req_state(struct drbd_conf *mdev, u64 id, sector_t sector, | |
4958 | struct rb_root *root, const char *func, | |
4959 | enum drbd_req_event what, bool missing_ok) | |
b411b363 PR |
4960 | { |
4961 | struct drbd_request *req; | |
4962 | struct bio_and_error m; | |
4963 | ||
87eeee41 | 4964 | spin_lock_irq(&mdev->tconn->req_lock); |
bc9c5c41 | 4965 | req = find_request(mdev, root, id, sector, missing_ok, func); |
b411b363 | 4966 | if (unlikely(!req)) { |
87eeee41 | 4967 | spin_unlock_irq(&mdev->tconn->req_lock); |
85997675 | 4968 | return -EIO; |
b411b363 PR |
4969 | } |
4970 | __req_mod(req, what, &m); | |
87eeee41 | 4971 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4972 | |
4973 | if (m.bio) | |
4974 | complete_master_bio(mdev, &m); | |
85997675 | 4975 | return 0; |
b411b363 PR |
4976 | } |
4977 | ||
1952e916 | 4978 | static int got_BlockAck(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4979 | { |
1952e916 | 4980 | struct drbd_conf *mdev; |
e658983a | 4981 | struct p_block_ack *p = pi->data; |
b411b363 PR |
4982 | sector_t sector = be64_to_cpu(p->sector); |
4983 | int blksize = be32_to_cpu(p->blksize); | |
4984 | enum drbd_req_event what; | |
4985 | ||
1952e916 AG |
4986 | mdev = vnr_to_mdev(tconn, pi->vnr); |
4987 | if (!mdev) | |
2735a594 | 4988 | return -EIO; |
1952e916 | 4989 | |
b411b363 PR |
4990 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); |
4991 | ||
579b57ed | 4992 | if (p->block_id == ID_SYNCER) { |
b411b363 PR |
4993 | drbd_set_in_sync(mdev, sector, blksize); |
4994 | dec_rs_pending(mdev); | |
2735a594 | 4995 | return 0; |
b411b363 | 4996 | } |
e05e1e59 | 4997 | switch (pi->cmd) { |
b411b363 | 4998 | case P_RS_WRITE_ACK: |
8554df1c | 4999 | what = WRITE_ACKED_BY_PEER_AND_SIS; |
b411b363 PR |
5000 | break; |
5001 | case P_WRITE_ACK: | |
8554df1c | 5002 | what = WRITE_ACKED_BY_PEER; |
b411b363 PR |
5003 | break; |
5004 | case P_RECV_ACK: | |
8554df1c | 5005 | what = RECV_ACKED_BY_PEER; |
b411b363 | 5006 | break; |
d4dabbe2 LE |
5007 | case P_SUPERSEDED: |
5008 | what = CONFLICT_RESOLVED; | |
b411b363 | 5009 | break; |
7be8da07 | 5010 | case P_RETRY_WRITE: |
7be8da07 | 5011 | what = POSTPONE_WRITE; |
b411b363 PR |
5012 | break; |
5013 | default: | |
2735a594 | 5014 | BUG(); |
b411b363 PR |
5015 | } |
5016 | ||
5017 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
2735a594 AG |
5018 | &mdev->write_requests, __func__, |
5019 | what, false); | |
b411b363 PR |
5020 | } |
5021 | ||
1952e916 | 5022 | static int got_NegAck(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 5023 | { |
1952e916 | 5024 | struct drbd_conf *mdev; |
e658983a | 5025 | struct p_block_ack *p = pi->data; |
b411b363 | 5026 | sector_t sector = be64_to_cpu(p->sector); |
2deb8336 | 5027 | int size = be32_to_cpu(p->blksize); |
85997675 | 5028 | int err; |
b411b363 | 5029 | |
1952e916 AG |
5030 | mdev = vnr_to_mdev(tconn, pi->vnr); |
5031 | if (!mdev) | |
2735a594 | 5032 | return -EIO; |
b411b363 PR |
5033 | |
5034 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
5035 | ||
579b57ed | 5036 | if (p->block_id == ID_SYNCER) { |
b411b363 PR |
5037 | dec_rs_pending(mdev); |
5038 | drbd_rs_failed_io(mdev, sector, size); | |
2735a594 | 5039 | return 0; |
b411b363 | 5040 | } |
2deb8336 | 5041 | |
85997675 AG |
5042 | err = validate_req_change_req_state(mdev, p->block_id, sector, |
5043 | &mdev->write_requests, __func__, | |
303d1448 | 5044 | NEG_ACKED, true); |
85997675 | 5045 | if (err) { |
c3afd8f5 AG |
5046 | /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs. |
5047 | The master bio might already be completed, therefore the | |
5048 | request is no longer in the collision hash. */ | |
5049 | /* In Protocol B we might already have got a P_RECV_ACK | |
5050 | but then get a P_NEG_ACK afterwards. */ | |
c3afd8f5 | 5051 | drbd_set_out_of_sync(mdev, sector, size); |
2deb8336 | 5052 | } |
2735a594 | 5053 | return 0; |
b411b363 PR |
5054 | } |
5055 | ||
1952e916 | 5056 | static int got_NegDReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 5057 | { |
1952e916 | 5058 | struct drbd_conf *mdev; |
e658983a | 5059 | struct p_block_ack *p = pi->data; |
b411b363 PR |
5060 | sector_t sector = be64_to_cpu(p->sector); |
5061 | ||
1952e916 AG |
5062 | mdev = vnr_to_mdev(tconn, pi->vnr); |
5063 | if (!mdev) | |
2735a594 | 5064 | return -EIO; |
1952e916 | 5065 | |
b411b363 | 5066 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); |
7be8da07 | 5067 | |
380207d0 | 5068 | dev_err(DEV, "Got NegDReply; Sector %llus, len %u.\n", |
b411b363 PR |
5069 | (unsigned long long)sector, be32_to_cpu(p->blksize)); |
5070 | ||
5071 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
2735a594 AG |
5072 | &mdev->read_requests, __func__, |
5073 | NEG_ACKED, false); | |
b411b363 PR |
5074 | } |
5075 | ||
1952e916 | 5076 | static int got_NegRSDReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 5077 | { |
1952e916 | 5078 | struct drbd_conf *mdev; |
b411b363 PR |
5079 | sector_t sector; |
5080 | int size; | |
e658983a | 5081 | struct p_block_ack *p = pi->data; |
1952e916 AG |
5082 | |
5083 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
5084 | if (!mdev) | |
2735a594 | 5085 | return -EIO; |
b411b363 PR |
5086 | |
5087 | sector = be64_to_cpu(p->sector); | |
5088 | size = be32_to_cpu(p->blksize); | |
b411b363 PR |
5089 | |
5090 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
5091 | ||
5092 | dec_rs_pending(mdev); | |
5093 | ||
5094 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
5095 | drbd_rs_complete_io(mdev, sector); | |
e05e1e59 | 5096 | switch (pi->cmd) { |
d612d309 PR |
5097 | case P_NEG_RS_DREPLY: |
5098 | drbd_rs_failed_io(mdev, sector, size); | |
5099 | case P_RS_CANCEL: | |
5100 | break; | |
5101 | default: | |
2735a594 | 5102 | BUG(); |
d612d309 | 5103 | } |
b411b363 PR |
5104 | put_ldev(mdev); |
5105 | } | |
5106 | ||
2735a594 | 5107 | return 0; |
b411b363 PR |
5108 | } |
5109 | ||
1952e916 | 5110 | static int got_BarrierAck(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 5111 | { |
e658983a | 5112 | struct p_barrier_ack *p = pi->data; |
9ed57dcb LE |
5113 | struct drbd_conf *mdev; |
5114 | int vnr; | |
1952e916 | 5115 | |
9ed57dcb | 5116 | tl_release(tconn, p->barrier, be32_to_cpu(p->set_size)); |
b411b363 | 5117 | |
9ed57dcb LE |
5118 | rcu_read_lock(); |
5119 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { | |
5120 | if (mdev->state.conn == C_AHEAD && | |
5121 | atomic_read(&mdev->ap_in_flight) == 0 && | |
5122 | !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &mdev->flags)) { | |
5123 | mdev->start_resync_timer.expires = jiffies + HZ; | |
5124 | add_timer(&mdev->start_resync_timer); | |
5125 | } | |
c4752ef1 | 5126 | } |
9ed57dcb | 5127 | rcu_read_unlock(); |
c4752ef1 | 5128 | |
2735a594 | 5129 | return 0; |
b411b363 PR |
5130 | } |
5131 | ||
1952e916 | 5132 | static int got_OVResult(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 5133 | { |
1952e916 | 5134 | struct drbd_conf *mdev; |
e658983a | 5135 | struct p_block_ack *p = pi->data; |
b411b363 PR |
5136 | struct drbd_work *w; |
5137 | sector_t sector; | |
5138 | int size; | |
5139 | ||
1952e916 AG |
5140 | mdev = vnr_to_mdev(tconn, pi->vnr); |
5141 | if (!mdev) | |
2735a594 | 5142 | return -EIO; |
1952e916 | 5143 | |
b411b363 PR |
5144 | sector = be64_to_cpu(p->sector); |
5145 | size = be32_to_cpu(p->blksize); | |
5146 | ||
5147 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
5148 | ||
5149 | if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) | |
8f7bed77 | 5150 | drbd_ov_out_of_sync_found(mdev, sector, size); |
b411b363 | 5151 | else |
8f7bed77 | 5152 | ov_out_of_sync_print(mdev); |
b411b363 | 5153 | |
1d53f09e | 5154 | if (!get_ldev(mdev)) |
2735a594 | 5155 | return 0; |
1d53f09e | 5156 | |
b411b363 PR |
5157 | drbd_rs_complete_io(mdev, sector); |
5158 | dec_rs_pending(mdev); | |
5159 | ||
ea5442af LE |
5160 | --mdev->ov_left; |
5161 | ||
5162 | /* let's advance progress step marks only for every other megabyte */ | |
5163 | if ((mdev->ov_left & 0x200) == 0x200) | |
5164 | drbd_advance_rs_marks(mdev, mdev->ov_left); | |
5165 | ||
5166 | if (mdev->ov_left == 0) { | |
b411b363 PR |
5167 | w = kmalloc(sizeof(*w), GFP_NOIO); |
5168 | if (w) { | |
5169 | w->cb = w_ov_finished; | |
a21e9298 | 5170 | w->mdev = mdev; |
d5b27b01 | 5171 | drbd_queue_work(&mdev->tconn->sender_work, w); |
b411b363 PR |
5172 | } else { |
5173 | dev_err(DEV, "kmalloc(w) failed."); | |
8f7bed77 | 5174 | ov_out_of_sync_print(mdev); |
b411b363 PR |
5175 | drbd_resync_finished(mdev); |
5176 | } | |
5177 | } | |
1d53f09e | 5178 | put_ldev(mdev); |
2735a594 | 5179 | return 0; |
b411b363 PR |
5180 | } |
5181 | ||
1952e916 | 5182 | static int got_skip(struct drbd_tconn *tconn, struct packet_info *pi) |
0ced55a3 | 5183 | { |
2735a594 | 5184 | return 0; |
b411b363 PR |
5185 | } |
5186 | ||
a990be46 | 5187 | static int tconn_finish_peer_reqs(struct drbd_tconn *tconn) |
0ced55a3 | 5188 | { |
082a3439 | 5189 | struct drbd_conf *mdev; |
c141ebda | 5190 | int vnr, not_empty = 0; |
32862ec7 PR |
5191 | |
5192 | do { | |
5193 | clear_bit(SIGNAL_ASENDER, &tconn->flags); | |
5194 | flush_signals(current); | |
c141ebda PR |
5195 | |
5196 | rcu_read_lock(); | |
5197 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { | |
5198 | kref_get(&mdev->kref); | |
5199 | rcu_read_unlock(); | |
d3fcb490 | 5200 | if (drbd_finish_peer_reqs(mdev)) { |
c141ebda PR |
5201 | kref_put(&mdev->kref, &drbd_minor_destroy); |
5202 | return 1; | |
d3fcb490 | 5203 | } |
c141ebda PR |
5204 | kref_put(&mdev->kref, &drbd_minor_destroy); |
5205 | rcu_read_lock(); | |
082a3439 | 5206 | } |
32862ec7 | 5207 | set_bit(SIGNAL_ASENDER, &tconn->flags); |
082a3439 PR |
5208 | |
5209 | spin_lock_irq(&tconn->req_lock); | |
c141ebda | 5210 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
082a3439 PR |
5211 | not_empty = !list_empty(&mdev->done_ee); |
5212 | if (not_empty) | |
5213 | break; | |
5214 | } | |
5215 | spin_unlock_irq(&tconn->req_lock); | |
c141ebda | 5216 | rcu_read_unlock(); |
32862ec7 PR |
5217 | } while (not_empty); |
5218 | ||
5219 | return 0; | |
0ced55a3 PR |
5220 | } |
5221 | ||
b411b363 PR |
5222 | struct asender_cmd { |
5223 | size_t pkt_size; | |
1952e916 | 5224 | int (*fn)(struct drbd_tconn *tconn, struct packet_info *); |
b411b363 PR |
5225 | }; |
5226 | ||
7201b972 | 5227 | static struct asender_cmd asender_tbl[] = { |
e658983a AG |
5228 | [P_PING] = { 0, got_Ping }, |
5229 | [P_PING_ACK] = { 0, got_PingAck }, | |
b411b363 PR |
5230 | [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, |
5231 | [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
5232 | [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
d4dabbe2 | 5233 | [P_SUPERSEDED] = { sizeof(struct p_block_ack), got_BlockAck }, |
b411b363 PR |
5234 | [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, |
5235 | [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, | |
1952e916 | 5236 | [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply }, |
b411b363 PR |
5237 | [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, |
5238 | [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, | |
5239 | [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, | |
5240 | [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, | |
02918be2 | 5241 | [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, |
1952e916 AG |
5242 | [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply }, |
5243 | [P_CONN_ST_CHG_REPLY]={ sizeof(struct p_req_state_reply), got_conn_RqSReply }, | |
5244 | [P_RETRY_WRITE] = { sizeof(struct p_block_ack), got_BlockAck }, | |
7201b972 | 5245 | }; |
b411b363 PR |
5246 | |
5247 | int drbd_asender(struct drbd_thread *thi) | |
5248 | { | |
392c8801 | 5249 | struct drbd_tconn *tconn = thi->tconn; |
b411b363 | 5250 | struct asender_cmd *cmd = NULL; |
77351055 | 5251 | struct packet_info pi; |
257d0af6 | 5252 | int rv; |
e658983a | 5253 | void *buf = tconn->meta.rbuf; |
b411b363 | 5254 | int received = 0; |
52b061a4 AG |
5255 | unsigned int header_size = drbd_header_size(tconn); |
5256 | int expect = header_size; | |
44ed167d PR |
5257 | bool ping_timeout_active = false; |
5258 | struct net_conf *nc; | |
bb77d34e | 5259 | int ping_timeo, tcp_cork, ping_int; |
b411b363 PR |
5260 | |
5261 | current->policy = SCHED_RR; /* Make this a realtime task! */ | |
5262 | current->rt_priority = 2; /* more important than all other tasks */ | |
5263 | ||
e77a0a5c | 5264 | while (get_t_state(thi) == RUNNING) { |
80822284 | 5265 | drbd_thread_current_set_cpu(thi); |
b411b363 | 5266 | |
44ed167d PR |
5267 | rcu_read_lock(); |
5268 | nc = rcu_dereference(tconn->net_conf); | |
5269 | ping_timeo = nc->ping_timeo; | |
bb77d34e | 5270 | tcp_cork = nc->tcp_cork; |
44ed167d PR |
5271 | ping_int = nc->ping_int; |
5272 | rcu_read_unlock(); | |
5273 | ||
32862ec7 | 5274 | if (test_and_clear_bit(SEND_PING, &tconn->flags)) { |
a17647aa | 5275 | if (drbd_send_ping(tconn)) { |
32862ec7 | 5276 | conn_err(tconn, "drbd_send_ping has failed\n"); |
b411b363 | 5277 | goto reconnect; |
841ce241 | 5278 | } |
44ed167d PR |
5279 | tconn->meta.socket->sk->sk_rcvtimeo = ping_timeo * HZ / 10; |
5280 | ping_timeout_active = true; | |
b411b363 PR |
5281 | } |
5282 | ||
32862ec7 PR |
5283 | /* TODO: conditionally cork; it may hurt latency if we cork without |
5284 | much to send */ | |
bb77d34e | 5285 | if (tcp_cork) |
32862ec7 | 5286 | drbd_tcp_cork(tconn->meta.socket); |
a990be46 AG |
5287 | if (tconn_finish_peer_reqs(tconn)) { |
5288 | conn_err(tconn, "tconn_finish_peer_reqs() failed\n"); | |
32862ec7 | 5289 | goto reconnect; |
b411b363 PR |
5290 | } |
5291 | /* but unconditionally uncork unless disabled */ | |
bb77d34e | 5292 | if (tcp_cork) |
32862ec7 | 5293 | drbd_tcp_uncork(tconn->meta.socket); |
b411b363 PR |
5294 | |
5295 | /* short circuit, recv_msg would return EINTR anyways. */ | |
5296 | if (signal_pending(current)) | |
5297 | continue; | |
5298 | ||
32862ec7 PR |
5299 | rv = drbd_recv_short(tconn->meta.socket, buf, expect-received, 0); |
5300 | clear_bit(SIGNAL_ASENDER, &tconn->flags); | |
b411b363 PR |
5301 | |
5302 | flush_signals(current); | |
5303 | ||
5304 | /* Note: | |
5305 | * -EINTR (on meta) we got a signal | |
5306 | * -EAGAIN (on meta) rcvtimeo expired | |
5307 | * -ECONNRESET other side closed the connection | |
5308 | * -ERESTARTSYS (on data) we got a signal | |
5309 | * rv < 0 other than above: unexpected error! | |
5310 | * rv == expected: full header or command | |
5311 | * rv < expected: "woken" by signal during receive | |
5312 | * rv == 0 : "connection shut down by peer" | |
5313 | */ | |
5314 | if (likely(rv > 0)) { | |
5315 | received += rv; | |
5316 | buf += rv; | |
5317 | } else if (rv == 0) { | |
b66623e3 PR |
5318 | if (test_bit(DISCONNECT_SENT, &tconn->flags)) { |
5319 | long t; | |
5320 | rcu_read_lock(); | |
5321 | t = rcu_dereference(tconn->net_conf)->ping_timeo * HZ/10; | |
5322 | rcu_read_unlock(); | |
5323 | ||
5324 | t = wait_event_timeout(tconn->ping_wait, | |
5325 | tconn->cstate < C_WF_REPORT_PARAMS, | |
5326 | t); | |
599377ac PR |
5327 | if (t) |
5328 | break; | |
5329 | } | |
32862ec7 | 5330 | conn_err(tconn, "meta connection shut down by peer.\n"); |
b411b363 PR |
5331 | goto reconnect; |
5332 | } else if (rv == -EAGAIN) { | |
cb6518cb LE |
5333 | /* If the data socket received something meanwhile, |
5334 | * that is good enough: peer is still alive. */ | |
32862ec7 PR |
5335 | if (time_after(tconn->last_received, |
5336 | jiffies - tconn->meta.socket->sk->sk_rcvtimeo)) | |
cb6518cb | 5337 | continue; |
f36af18c | 5338 | if (ping_timeout_active) { |
32862ec7 | 5339 | conn_err(tconn, "PingAck did not arrive in time.\n"); |
b411b363 PR |
5340 | goto reconnect; |
5341 | } | |
32862ec7 | 5342 | set_bit(SEND_PING, &tconn->flags); |
b411b363 PR |
5343 | continue; |
5344 | } else if (rv == -EINTR) { | |
5345 | continue; | |
5346 | } else { | |
32862ec7 | 5347 | conn_err(tconn, "sock_recvmsg returned %d\n", rv); |
b411b363 PR |
5348 | goto reconnect; |
5349 | } | |
5350 | ||
5351 | if (received == expect && cmd == NULL) { | |
e658983a | 5352 | if (decode_header(tconn, tconn->meta.rbuf, &pi)) |
b411b363 | 5353 | goto reconnect; |
7201b972 | 5354 | cmd = &asender_tbl[pi.cmd]; |
1952e916 | 5355 | if (pi.cmd >= ARRAY_SIZE(asender_tbl) || !cmd->fn) { |
2fcb8f30 AG |
5356 | conn_err(tconn, "Unexpected meta packet %s (0x%04x)\n", |
5357 | cmdname(pi.cmd), pi.cmd); | |
b411b363 PR |
5358 | goto disconnect; |
5359 | } | |
e658983a | 5360 | expect = header_size + cmd->pkt_size; |
52b061a4 | 5361 | if (pi.size != expect - header_size) { |
32862ec7 | 5362 | conn_err(tconn, "Wrong packet size on meta (c: %d, l: %d)\n", |
77351055 | 5363 | pi.cmd, pi.size); |
b411b363 | 5364 | goto reconnect; |
257d0af6 | 5365 | } |
b411b363 PR |
5366 | } |
5367 | if (received == expect) { | |
2735a594 | 5368 | bool err; |
a4fbda8e | 5369 | |
2735a594 AG |
5370 | err = cmd->fn(tconn, &pi); |
5371 | if (err) { | |
1952e916 | 5372 | conn_err(tconn, "%pf failed\n", cmd->fn); |
b411b363 | 5373 | goto reconnect; |
1952e916 | 5374 | } |
b411b363 | 5375 | |
a4fbda8e | 5376 | tconn->last_received = jiffies; |
f36af18c | 5377 | |
44ed167d PR |
5378 | if (cmd == &asender_tbl[P_PING_ACK]) { |
5379 | /* restore idle timeout */ | |
5380 | tconn->meta.socket->sk->sk_rcvtimeo = ping_int * HZ; | |
5381 | ping_timeout_active = false; | |
5382 | } | |
f36af18c | 5383 | |
e658983a | 5384 | buf = tconn->meta.rbuf; |
b411b363 | 5385 | received = 0; |
52b061a4 | 5386 | expect = header_size; |
b411b363 PR |
5387 | cmd = NULL; |
5388 | } | |
5389 | } | |
5390 | ||
5391 | if (0) { | |
5392 | reconnect: | |
bbeb641c | 5393 | conn_request_state(tconn, NS(conn, C_NETWORK_FAILURE), CS_HARD); |
19fffd7b | 5394 | conn_md_sync(tconn); |
b411b363 PR |
5395 | } |
5396 | if (0) { | |
5397 | disconnect: | |
bbeb641c | 5398 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 5399 | } |
32862ec7 | 5400 | clear_bit(SIGNAL_ASENDER, &tconn->flags); |
b411b363 | 5401 | |
32862ec7 | 5402 | conn_info(tconn, "asender terminated\n"); |
b411b363 PR |
5403 | |
5404 | return 0; | |
5405 | } |