]>
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; | |
38b682b2 AM |
760 | /* 28.5% random jitter */ |
761 | timeo += (prandom_u32() & 1) ? timeo / 7 : -timeo / 7; | |
1f3e509b | 762 | |
7a426fd8 PR |
763 | err = wait_for_completion_interruptible_timeout(&ad->door_bell, timeo); |
764 | if (err <= 0) | |
765 | return NULL; | |
b411b363 | 766 | |
7a426fd8 | 767 | err = kernel_accept(ad->s_listen, &s_estab, 0); |
b411b363 PR |
768 | if (err < 0) { |
769 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
1f3e509b | 770 | conn_err(tconn, "accept failed, err = %d\n", err); |
bbeb641c | 771 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
772 | } |
773 | } | |
b411b363 | 774 | |
715306f6 AG |
775 | if (s_estab) |
776 | unregister_state_change(s_estab->sk, ad); | |
b411b363 | 777 | |
b411b363 PR |
778 | return s_estab; |
779 | } | |
b411b363 | 780 | |
e658983a | 781 | static int decode_header(struct drbd_tconn *, void *, struct packet_info *); |
b411b363 | 782 | |
9f5bdc33 AG |
783 | static int send_first_packet(struct drbd_tconn *tconn, struct drbd_socket *sock, |
784 | enum drbd_packet cmd) | |
785 | { | |
786 | if (!conn_prepare_command(tconn, sock)) | |
787 | return -EIO; | |
e658983a | 788 | return conn_send_command(tconn, sock, cmd, 0, NULL, 0); |
b411b363 PR |
789 | } |
790 | ||
9f5bdc33 | 791 | static int receive_first_packet(struct drbd_tconn *tconn, struct socket *sock) |
b411b363 | 792 | { |
9f5bdc33 AG |
793 | unsigned int header_size = drbd_header_size(tconn); |
794 | struct packet_info pi; | |
795 | int err; | |
b411b363 | 796 | |
9f5bdc33 AG |
797 | err = drbd_recv_short(sock, tconn->data.rbuf, header_size, 0); |
798 | if (err != header_size) { | |
799 | if (err >= 0) | |
800 | err = -EIO; | |
801 | return err; | |
802 | } | |
803 | err = decode_header(tconn, tconn->data.rbuf, &pi); | |
804 | if (err) | |
805 | return err; | |
806 | return pi.cmd; | |
b411b363 PR |
807 | } |
808 | ||
809 | /** | |
810 | * drbd_socket_okay() - Free the socket if its connection is not okay | |
b411b363 PR |
811 | * @sock: pointer to the pointer to the socket. |
812 | */ | |
dbd9eea0 | 813 | static int drbd_socket_okay(struct socket **sock) |
b411b363 PR |
814 | { |
815 | int rr; | |
816 | char tb[4]; | |
817 | ||
818 | if (!*sock) | |
81e84650 | 819 | return false; |
b411b363 | 820 | |
dbd9eea0 | 821 | rr = drbd_recv_short(*sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); |
b411b363 PR |
822 | |
823 | if (rr > 0 || rr == -EAGAIN) { | |
81e84650 | 824 | return true; |
b411b363 PR |
825 | } else { |
826 | sock_release(*sock); | |
827 | *sock = NULL; | |
81e84650 | 828 | return false; |
b411b363 PR |
829 | } |
830 | } | |
2325eb66 PR |
831 | /* Gets called if a connection is established, or if a new minor gets created |
832 | in a connection */ | |
c141ebda | 833 | int drbd_connected(struct drbd_conf *mdev) |
907599e0 | 834 | { |
0829f5ed | 835 | int err; |
907599e0 PR |
836 | |
837 | atomic_set(&mdev->packet_seq, 0); | |
838 | mdev->peer_seq = 0; | |
839 | ||
8410da8f PR |
840 | mdev->state_mutex = mdev->tconn->agreed_pro_version < 100 ? |
841 | &mdev->tconn->cstate_mutex : | |
842 | &mdev->own_state_mutex; | |
843 | ||
0829f5ed AG |
844 | err = drbd_send_sync_param(mdev); |
845 | if (!err) | |
846 | err = drbd_send_sizes(mdev, 0, 0); | |
847 | if (!err) | |
848 | err = drbd_send_uuids(mdev); | |
849 | if (!err) | |
43de7c85 | 850 | err = drbd_send_current_state(mdev); |
907599e0 PR |
851 | clear_bit(USE_DEGR_WFC_T, &mdev->flags); |
852 | clear_bit(RESIZE_PENDING, &mdev->flags); | |
2d56a974 | 853 | atomic_set(&mdev->ap_in_flight, 0); |
8b924f1d | 854 | mod_timer(&mdev->request_timer, jiffies + HZ); /* just start it here. */ |
0829f5ed | 855 | return err; |
907599e0 | 856 | } |
b411b363 PR |
857 | |
858 | /* | |
859 | * return values: | |
860 | * 1 yes, we have a valid connection | |
861 | * 0 oops, did not work out, please try again | |
862 | * -1 peer talks different language, | |
863 | * no point in trying again, please go standalone. | |
864 | * -2 We do not have a network config... | |
865 | */ | |
81fa2e67 | 866 | static int conn_connect(struct drbd_tconn *tconn) |
b411b363 | 867 | { |
7da35862 | 868 | struct drbd_socket sock, msock; |
c141ebda | 869 | struct drbd_conf *mdev; |
44ed167d | 870 | struct net_conf *nc; |
92f14951 | 871 | int vnr, timeout, h, ok; |
08b165ba | 872 | bool discard_my_data; |
197296ff | 873 | enum drbd_state_rv rv; |
7a426fd8 PR |
874 | struct accept_wait_data ad = { |
875 | .tconn = tconn, | |
876 | .door_bell = COMPLETION_INITIALIZER_ONSTACK(ad.door_bell), | |
877 | }; | |
b411b363 | 878 | |
b66623e3 | 879 | clear_bit(DISCONNECT_SENT, &tconn->flags); |
bbeb641c | 880 | if (conn_request_state(tconn, NS(conn, C_WF_CONNECTION), CS_VERBOSE) < SS_SUCCESS) |
b411b363 PR |
881 | return -2; |
882 | ||
7da35862 PR |
883 | mutex_init(&sock.mutex); |
884 | sock.sbuf = tconn->data.sbuf; | |
885 | sock.rbuf = tconn->data.rbuf; | |
886 | sock.socket = NULL; | |
887 | mutex_init(&msock.mutex); | |
888 | msock.sbuf = tconn->meta.sbuf; | |
889 | msock.rbuf = tconn->meta.rbuf; | |
890 | msock.socket = NULL; | |
891 | ||
0916e0e3 AG |
892 | /* Assume that the peer only understands protocol 80 until we know better. */ |
893 | tconn->agreed_pro_version = 80; | |
b411b363 | 894 | |
7a426fd8 PR |
895 | if (prepare_listen_socket(tconn, &ad)) |
896 | return 0; | |
b411b363 PR |
897 | |
898 | do { | |
2bf89621 | 899 | struct socket *s; |
b411b363 | 900 | |
92f14951 | 901 | s = drbd_try_connect(tconn); |
b411b363 | 902 | if (s) { |
7da35862 PR |
903 | if (!sock.socket) { |
904 | sock.socket = s; | |
905 | send_first_packet(tconn, &sock, P_INITIAL_DATA); | |
906 | } else if (!msock.socket) { | |
427c0434 | 907 | clear_bit(RESOLVE_CONFLICTS, &tconn->flags); |
7da35862 PR |
908 | msock.socket = s; |
909 | send_first_packet(tconn, &msock, P_INITIAL_META); | |
b411b363 | 910 | } else { |
81fa2e67 | 911 | conn_err(tconn, "Logic error in conn_connect()\n"); |
b411b363 PR |
912 | goto out_release_sockets; |
913 | } | |
914 | } | |
915 | ||
7da35862 PR |
916 | if (sock.socket && msock.socket) { |
917 | rcu_read_lock(); | |
918 | nc = rcu_dereference(tconn->net_conf); | |
919 | timeout = nc->ping_timeo * HZ / 10; | |
920 | rcu_read_unlock(); | |
921 | schedule_timeout_interruptible(timeout); | |
922 | ok = drbd_socket_okay(&sock.socket); | |
923 | ok = drbd_socket_okay(&msock.socket) && ok; | |
b411b363 PR |
924 | if (ok) |
925 | break; | |
926 | } | |
927 | ||
928 | retry: | |
7a426fd8 | 929 | s = drbd_wait_for_connect(tconn, &ad); |
b411b363 | 930 | if (s) { |
92f14951 | 931 | int fp = receive_first_packet(tconn, s); |
7da35862 PR |
932 | drbd_socket_okay(&sock.socket); |
933 | drbd_socket_okay(&msock.socket); | |
92f14951 | 934 | switch (fp) { |
e5d6f33a | 935 | case P_INITIAL_DATA: |
7da35862 | 936 | if (sock.socket) { |
907599e0 | 937 | conn_warn(tconn, "initial packet S crossed\n"); |
7da35862 | 938 | sock_release(sock.socket); |
80c6eed4 PR |
939 | sock.socket = s; |
940 | goto randomize; | |
b411b363 | 941 | } |
7da35862 | 942 | sock.socket = s; |
b411b363 | 943 | break; |
e5d6f33a | 944 | case P_INITIAL_META: |
427c0434 | 945 | set_bit(RESOLVE_CONFLICTS, &tconn->flags); |
7da35862 | 946 | if (msock.socket) { |
907599e0 | 947 | conn_warn(tconn, "initial packet M crossed\n"); |
7da35862 | 948 | sock_release(msock.socket); |
80c6eed4 PR |
949 | msock.socket = s; |
950 | goto randomize; | |
b411b363 | 951 | } |
7da35862 | 952 | msock.socket = s; |
b411b363 PR |
953 | break; |
954 | default: | |
907599e0 | 955 | conn_warn(tconn, "Error receiving initial packet\n"); |
b411b363 | 956 | sock_release(s); |
80c6eed4 | 957 | randomize: |
38b682b2 | 958 | if (prandom_u32() & 1) |
b411b363 PR |
959 | goto retry; |
960 | } | |
961 | } | |
962 | ||
bbeb641c | 963 | if (tconn->cstate <= C_DISCONNECTING) |
b411b363 PR |
964 | goto out_release_sockets; |
965 | if (signal_pending(current)) { | |
966 | flush_signals(current); | |
967 | smp_rmb(); | |
907599e0 | 968 | if (get_t_state(&tconn->receiver) == EXITING) |
b411b363 PR |
969 | goto out_release_sockets; |
970 | } | |
971 | ||
b666dbf8 PR |
972 | ok = drbd_socket_okay(&sock.socket); |
973 | ok = drbd_socket_okay(&msock.socket) && ok; | |
974 | } while (!ok); | |
b411b363 | 975 | |
7a426fd8 PR |
976 | if (ad.s_listen) |
977 | sock_release(ad.s_listen); | |
b411b363 | 978 | |
98683650 PR |
979 | sock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ |
980 | msock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ | |
b411b363 | 981 | |
7da35862 PR |
982 | sock.socket->sk->sk_allocation = GFP_NOIO; |
983 | msock.socket->sk->sk_allocation = GFP_NOIO; | |
b411b363 | 984 | |
7da35862 PR |
985 | sock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; |
986 | msock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE; | |
b411b363 | 987 | |
b411b363 | 988 | /* NOT YET ... |
7da35862 PR |
989 | * sock.socket->sk->sk_sndtimeo = tconn->net_conf->timeout*HZ/10; |
990 | * sock.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; | |
6038178e | 991 | * first set it to the P_CONNECTION_FEATURES timeout, |
b411b363 | 992 | * which we set to 4x the configured ping_timeout. */ |
44ed167d PR |
993 | rcu_read_lock(); |
994 | nc = rcu_dereference(tconn->net_conf); | |
995 | ||
7da35862 PR |
996 | sock.socket->sk->sk_sndtimeo = |
997 | sock.socket->sk->sk_rcvtimeo = nc->ping_timeo*4*HZ/10; | |
b411b363 | 998 | |
7da35862 | 999 | msock.socket->sk->sk_rcvtimeo = nc->ping_int*HZ; |
44ed167d | 1000 | timeout = nc->timeout * HZ / 10; |
08b165ba | 1001 | discard_my_data = nc->discard_my_data; |
44ed167d | 1002 | rcu_read_unlock(); |
b411b363 | 1003 | |
7da35862 | 1004 | msock.socket->sk->sk_sndtimeo = timeout; |
b411b363 PR |
1005 | |
1006 | /* we don't want delays. | |
25985edc | 1007 | * we use TCP_CORK where appropriate, though */ |
7da35862 PR |
1008 | drbd_tcp_nodelay(sock.socket); |
1009 | drbd_tcp_nodelay(msock.socket); | |
b411b363 | 1010 | |
7da35862 PR |
1011 | tconn->data.socket = sock.socket; |
1012 | tconn->meta.socket = msock.socket; | |
907599e0 | 1013 | tconn->last_received = jiffies; |
b411b363 | 1014 | |
6038178e | 1015 | h = drbd_do_features(tconn); |
b411b363 PR |
1016 | if (h <= 0) |
1017 | return h; | |
1018 | ||
907599e0 | 1019 | if (tconn->cram_hmac_tfm) { |
b411b363 | 1020 | /* drbd_request_state(mdev, NS(conn, WFAuth)); */ |
907599e0 | 1021 | switch (drbd_do_auth(tconn)) { |
b10d96cb | 1022 | case -1: |
907599e0 | 1023 | conn_err(tconn, "Authentication of peer failed\n"); |
b411b363 | 1024 | return -1; |
b10d96cb | 1025 | case 0: |
907599e0 | 1026 | conn_err(tconn, "Authentication of peer failed, trying again.\n"); |
b10d96cb | 1027 | return 0; |
b411b363 PR |
1028 | } |
1029 | } | |
1030 | ||
7da35862 PR |
1031 | tconn->data.socket->sk->sk_sndtimeo = timeout; |
1032 | tconn->data.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; | |
b411b363 | 1033 | |
387eb308 | 1034 | if (drbd_send_protocol(tconn) == -EOPNOTSUPP) |
7e2455c1 | 1035 | return -1; |
b411b363 | 1036 | |
a1096a6e PR |
1037 | set_bit(STATE_SENT, &tconn->flags); |
1038 | ||
c141ebda PR |
1039 | rcu_read_lock(); |
1040 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { | |
1041 | kref_get(&mdev->kref); | |
26ea8f92 AG |
1042 | rcu_read_unlock(); |
1043 | ||
13c76aba PR |
1044 | /* Prevent a race between resync-handshake and |
1045 | * being promoted to Primary. | |
1046 | * | |
1047 | * Grab and release the state mutex, so we know that any current | |
1048 | * drbd_set_role() is finished, and any incoming drbd_set_role | |
1049 | * will see the STATE_SENT flag, and wait for it to be cleared. | |
1050 | */ | |
1051 | mutex_lock(mdev->state_mutex); | |
1052 | mutex_unlock(mdev->state_mutex); | |
1053 | ||
08b165ba PR |
1054 | if (discard_my_data) |
1055 | set_bit(DISCARD_MY_DATA, &mdev->flags); | |
1056 | else | |
1057 | clear_bit(DISCARD_MY_DATA, &mdev->flags); | |
1058 | ||
c141ebda PR |
1059 | drbd_connected(mdev); |
1060 | kref_put(&mdev->kref, &drbd_minor_destroy); | |
1061 | rcu_read_lock(); | |
1062 | } | |
1063 | rcu_read_unlock(); | |
1064 | ||
a1096a6e | 1065 | rv = conn_request_state(tconn, NS(conn, C_WF_REPORT_PARAMS), CS_VERBOSE); |
ed635cb0 | 1066 | if (rv < SS_SUCCESS || tconn->cstate != C_WF_REPORT_PARAMS) { |
a1096a6e | 1067 | clear_bit(STATE_SENT, &tconn->flags); |
1e86ac48 | 1068 | return 0; |
a1096a6e | 1069 | } |
1e86ac48 | 1070 | |
823bd832 | 1071 | drbd_thread_start(&tconn->asender); |
b411b363 | 1072 | |
08b165ba PR |
1073 | mutex_lock(&tconn->conf_update); |
1074 | /* The discard_my_data flag is a single-shot modifier to the next | |
1075 | * connection attempt, the handshake of which is now well underway. | |
1076 | * No need for rcu style copying of the whole struct | |
1077 | * just to clear a single value. */ | |
1078 | tconn->net_conf->discard_my_data = 0; | |
1079 | mutex_unlock(&tconn->conf_update); | |
1080 | ||
d3fcb490 | 1081 | return h; |
b411b363 PR |
1082 | |
1083 | out_release_sockets: | |
7a426fd8 PR |
1084 | if (ad.s_listen) |
1085 | sock_release(ad.s_listen); | |
7da35862 PR |
1086 | if (sock.socket) |
1087 | sock_release(sock.socket); | |
1088 | if (msock.socket) | |
1089 | sock_release(msock.socket); | |
b411b363 PR |
1090 | return -1; |
1091 | } | |
1092 | ||
e658983a | 1093 | static int decode_header(struct drbd_tconn *tconn, void *header, struct packet_info *pi) |
b411b363 | 1094 | { |
e658983a AG |
1095 | unsigned int header_size = drbd_header_size(tconn); |
1096 | ||
0c8e36d9 AG |
1097 | if (header_size == sizeof(struct p_header100) && |
1098 | *(__be32 *)header == cpu_to_be32(DRBD_MAGIC_100)) { | |
1099 | struct p_header100 *h = header; | |
1100 | if (h->pad != 0) { | |
1101 | conn_err(tconn, "Header padding is not zero\n"); | |
1102 | return -EINVAL; | |
1103 | } | |
1104 | pi->vnr = be16_to_cpu(h->volume); | |
1105 | pi->cmd = be16_to_cpu(h->command); | |
1106 | pi->size = be32_to_cpu(h->length); | |
1107 | } else if (header_size == sizeof(struct p_header95) && | |
1108 | *(__be16 *)header == cpu_to_be16(DRBD_MAGIC_BIG)) { | |
e658983a | 1109 | struct p_header95 *h = header; |
e658983a | 1110 | pi->cmd = be16_to_cpu(h->command); |
b55d84ba AG |
1111 | pi->size = be32_to_cpu(h->length); |
1112 | pi->vnr = 0; | |
e658983a AG |
1113 | } else if (header_size == sizeof(struct p_header80) && |
1114 | *(__be32 *)header == cpu_to_be32(DRBD_MAGIC)) { | |
1115 | struct p_header80 *h = header; | |
1116 | pi->cmd = be16_to_cpu(h->command); | |
1117 | pi->size = be16_to_cpu(h->length); | |
77351055 | 1118 | pi->vnr = 0; |
02918be2 | 1119 | } else { |
e658983a AG |
1120 | conn_err(tconn, "Wrong magic value 0x%08x in protocol version %d\n", |
1121 | be32_to_cpu(*(__be32 *)header), | |
1122 | tconn->agreed_pro_version); | |
8172f3e9 | 1123 | return -EINVAL; |
b411b363 | 1124 | } |
e658983a | 1125 | pi->data = header + header_size; |
8172f3e9 | 1126 | return 0; |
257d0af6 | 1127 | } |
b411b363 | 1128 | |
9ba7aa00 | 1129 | static int drbd_recv_header(struct drbd_tconn *tconn, struct packet_info *pi) |
257d0af6 | 1130 | { |
e658983a | 1131 | void *buffer = tconn->data.rbuf; |
69bc7bc3 | 1132 | int err; |
257d0af6 | 1133 | |
e658983a | 1134 | err = drbd_recv_all_warn(tconn, buffer, drbd_header_size(tconn)); |
a5c31904 | 1135 | if (err) |
69bc7bc3 | 1136 | return err; |
257d0af6 | 1137 | |
e658983a | 1138 | err = decode_header(tconn, buffer, pi); |
9ba7aa00 | 1139 | tconn->last_received = jiffies; |
b411b363 | 1140 | |
69bc7bc3 | 1141 | return err; |
b411b363 PR |
1142 | } |
1143 | ||
4b0007c0 | 1144 | static void drbd_flush(struct drbd_tconn *tconn) |
b411b363 PR |
1145 | { |
1146 | int rv; | |
4b0007c0 PR |
1147 | struct drbd_conf *mdev; |
1148 | int vnr; | |
1149 | ||
1150 | if (tconn->write_ordering >= WO_bdev_flush) { | |
615e087f | 1151 | rcu_read_lock(); |
4b0007c0 | 1152 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
615e087f LE |
1153 | if (!get_ldev(mdev)) |
1154 | continue; | |
1155 | kref_get(&mdev->kref); | |
1156 | rcu_read_unlock(); | |
1157 | ||
1158 | rv = blkdev_issue_flush(mdev->ldev->backing_bdev, | |
1159 | GFP_NOIO, NULL); | |
1160 | if (rv) { | |
1161 | dev_info(DEV, "local disk flush failed with status %d\n", rv); | |
1162 | /* would rather check on EOPNOTSUPP, but that is not reliable. | |
1163 | * don't try again for ANY return value != 0 | |
1164 | * if (rv == -EOPNOTSUPP) */ | |
1165 | drbd_bump_write_ordering(tconn, WO_drain_io); | |
4b0007c0 | 1166 | } |
615e087f LE |
1167 | put_ldev(mdev); |
1168 | kref_put(&mdev->kref, &drbd_minor_destroy); | |
b411b363 | 1169 | |
615e087f LE |
1170 | rcu_read_lock(); |
1171 | if (rv) | |
1172 | break; | |
b411b363 | 1173 | } |
615e087f | 1174 | rcu_read_unlock(); |
b411b363 | 1175 | } |
b411b363 PR |
1176 | } |
1177 | ||
1178 | /** | |
1179 | * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. | |
1180 | * @mdev: DRBD device. | |
1181 | * @epoch: Epoch object. | |
1182 | * @ev: Epoch event. | |
1183 | */ | |
1e9dd291 | 1184 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_tconn *tconn, |
b411b363 PR |
1185 | struct drbd_epoch *epoch, |
1186 | enum epoch_event ev) | |
1187 | { | |
2451fc3b | 1188 | int epoch_size; |
b411b363 | 1189 | struct drbd_epoch *next_epoch; |
b411b363 PR |
1190 | enum finish_epoch rv = FE_STILL_LIVE; |
1191 | ||
12038a3a | 1192 | spin_lock(&tconn->epoch_lock); |
b411b363 PR |
1193 | do { |
1194 | next_epoch = NULL; | |
b411b363 PR |
1195 | |
1196 | epoch_size = atomic_read(&epoch->epoch_size); | |
1197 | ||
1198 | switch (ev & ~EV_CLEANUP) { | |
1199 | case EV_PUT: | |
1200 | atomic_dec(&epoch->active); | |
1201 | break; | |
1202 | case EV_GOT_BARRIER_NR: | |
1203 | set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); | |
b411b363 PR |
1204 | break; |
1205 | case EV_BECAME_LAST: | |
1206 | /* nothing to do*/ | |
1207 | break; | |
1208 | } | |
1209 | ||
b411b363 PR |
1210 | if (epoch_size != 0 && |
1211 | atomic_read(&epoch->active) == 0 && | |
80f9fd55 | 1212 | (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) || ev & EV_CLEANUP)) { |
b411b363 | 1213 | if (!(ev & EV_CLEANUP)) { |
12038a3a | 1214 | spin_unlock(&tconn->epoch_lock); |
9ed57dcb | 1215 | drbd_send_b_ack(epoch->tconn, epoch->barrier_nr, epoch_size); |
12038a3a | 1216 | spin_lock(&tconn->epoch_lock); |
b411b363 | 1217 | } |
9ed57dcb LE |
1218 | #if 0 |
1219 | /* FIXME: dec unacked on connection, once we have | |
1220 | * something to count pending connection packets in. */ | |
80f9fd55 | 1221 | if (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) |
9ed57dcb LE |
1222 | dec_unacked(epoch->tconn); |
1223 | #endif | |
b411b363 | 1224 | |
12038a3a | 1225 | if (tconn->current_epoch != epoch) { |
b411b363 PR |
1226 | next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); |
1227 | list_del(&epoch->list); | |
1228 | ev = EV_BECAME_LAST | (ev & EV_CLEANUP); | |
12038a3a | 1229 | tconn->epochs--; |
b411b363 PR |
1230 | kfree(epoch); |
1231 | ||
1232 | if (rv == FE_STILL_LIVE) | |
1233 | rv = FE_DESTROYED; | |
1234 | } else { | |
1235 | epoch->flags = 0; | |
1236 | atomic_set(&epoch->epoch_size, 0); | |
698f9315 | 1237 | /* atomic_set(&epoch->active, 0); is already zero */ |
b411b363 PR |
1238 | if (rv == FE_STILL_LIVE) |
1239 | rv = FE_RECYCLED; | |
1240 | } | |
1241 | } | |
1242 | ||
1243 | if (!next_epoch) | |
1244 | break; | |
1245 | ||
1246 | epoch = next_epoch; | |
1247 | } while (1); | |
1248 | ||
12038a3a | 1249 | spin_unlock(&tconn->epoch_lock); |
b411b363 | 1250 | |
b411b363 PR |
1251 | return rv; |
1252 | } | |
1253 | ||
1254 | /** | |
1255 | * drbd_bump_write_ordering() - Fall back to an other write ordering method | |
4b0007c0 | 1256 | * @tconn: DRBD connection. |
b411b363 PR |
1257 | * @wo: Write ordering method to try. |
1258 | */ | |
4b0007c0 | 1259 | void drbd_bump_write_ordering(struct drbd_tconn *tconn, enum write_ordering_e wo) |
b411b363 | 1260 | { |
daeda1cc | 1261 | struct disk_conf *dc; |
4b0007c0 | 1262 | struct drbd_conf *mdev; |
b411b363 | 1263 | enum write_ordering_e pwo; |
4b0007c0 | 1264 | int vnr; |
b411b363 PR |
1265 | static char *write_ordering_str[] = { |
1266 | [WO_none] = "none", | |
1267 | [WO_drain_io] = "drain", | |
1268 | [WO_bdev_flush] = "flush", | |
b411b363 PR |
1269 | }; |
1270 | ||
4b0007c0 | 1271 | pwo = tconn->write_ordering; |
b411b363 | 1272 | wo = min(pwo, wo); |
daeda1cc | 1273 | rcu_read_lock(); |
4b0007c0 | 1274 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
27eb13e9 | 1275 | if (!get_ldev_if_state(mdev, D_ATTACHING)) |
4b0007c0 PR |
1276 | continue; |
1277 | dc = rcu_dereference(mdev->ldev->disk_conf); | |
1278 | ||
1279 | if (wo == WO_bdev_flush && !dc->disk_flushes) | |
1280 | wo = WO_drain_io; | |
1281 | if (wo == WO_drain_io && !dc->disk_drain) | |
1282 | wo = WO_none; | |
1283 | put_ldev(mdev); | |
1284 | } | |
daeda1cc | 1285 | rcu_read_unlock(); |
4b0007c0 PR |
1286 | tconn->write_ordering = wo; |
1287 | if (pwo != tconn->write_ordering || wo == WO_bdev_flush) | |
1288 | conn_info(tconn, "Method to ensure write ordering: %s\n", write_ordering_str[tconn->write_ordering]); | |
b411b363 PR |
1289 | } |
1290 | ||
45bb912b | 1291 | /** |
fbe29dec | 1292 | * drbd_submit_peer_request() |
45bb912b | 1293 | * @mdev: DRBD device. |
db830c46 | 1294 | * @peer_req: peer request |
45bb912b | 1295 | * @rw: flag field, see bio->bi_rw |
10f6d992 LE |
1296 | * |
1297 | * May spread the pages to multiple bios, | |
1298 | * depending on bio_add_page restrictions. | |
1299 | * | |
1300 | * Returns 0 if all bios have been submitted, | |
1301 | * -ENOMEM if we could not allocate enough bios, | |
1302 | * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a | |
1303 | * single page to an empty bio (which should never happen and likely indicates | |
1304 | * that the lower level IO stack is in some way broken). This has been observed | |
1305 | * on certain Xen deployments. | |
45bb912b LE |
1306 | */ |
1307 | /* TODO allocate from our own bio_set. */ | |
fbe29dec AG |
1308 | int drbd_submit_peer_request(struct drbd_conf *mdev, |
1309 | struct drbd_peer_request *peer_req, | |
1310 | const unsigned rw, const int fault_type) | |
45bb912b LE |
1311 | { |
1312 | struct bio *bios = NULL; | |
1313 | struct bio *bio; | |
db830c46 AG |
1314 | struct page *page = peer_req->pages; |
1315 | sector_t sector = peer_req->i.sector; | |
1316 | unsigned ds = peer_req->i.size; | |
45bb912b LE |
1317 | unsigned n_bios = 0; |
1318 | unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT; | |
10f6d992 | 1319 | int err = -ENOMEM; |
45bb912b LE |
1320 | |
1321 | /* In most cases, we will only need one bio. But in case the lower | |
1322 | * level restrictions happen to be different at this offset on this | |
1323 | * side than those of the sending peer, we may need to submit the | |
9476f39d LE |
1324 | * request in more than one bio. |
1325 | * | |
1326 | * Plain bio_alloc is good enough here, this is no DRBD internally | |
1327 | * generated bio, but a bio allocated on behalf of the peer. | |
1328 | */ | |
45bb912b LE |
1329 | next_bio: |
1330 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
1331 | if (!bio) { | |
1332 | dev_err(DEV, "submit_ee: Allocation of a bio failed\n"); | |
1333 | goto fail; | |
1334 | } | |
db830c46 | 1335 | /* > peer_req->i.sector, unless this is the first bio */ |
4f024f37 | 1336 | bio->bi_iter.bi_sector = sector; |
45bb912b | 1337 | bio->bi_bdev = mdev->ldev->backing_bdev; |
45bb912b | 1338 | bio->bi_rw = rw; |
db830c46 | 1339 | bio->bi_private = peer_req; |
fcefa62e | 1340 | bio->bi_end_io = drbd_peer_request_endio; |
45bb912b LE |
1341 | |
1342 | bio->bi_next = bios; | |
1343 | bios = bio; | |
1344 | ++n_bios; | |
1345 | ||
1346 | page_chain_for_each(page) { | |
1347 | unsigned len = min_t(unsigned, ds, PAGE_SIZE); | |
1348 | if (!bio_add_page(bio, page, len, 0)) { | |
10f6d992 LE |
1349 | /* A single page must always be possible! |
1350 | * But in case it fails anyways, | |
1351 | * we deal with it, and complain (below). */ | |
1352 | if (bio->bi_vcnt == 0) { | |
1353 | dev_err(DEV, | |
1354 | "bio_add_page failed for len=%u, " | |
1355 | "bi_vcnt=0 (bi_sector=%llu)\n", | |
4f024f37 | 1356 | len, (uint64_t)bio->bi_iter.bi_sector); |
10f6d992 LE |
1357 | err = -ENOSPC; |
1358 | goto fail; | |
1359 | } | |
45bb912b LE |
1360 | goto next_bio; |
1361 | } | |
1362 | ds -= len; | |
1363 | sector += len >> 9; | |
1364 | --nr_pages; | |
1365 | } | |
1366 | D_ASSERT(page == NULL); | |
1367 | D_ASSERT(ds == 0); | |
1368 | ||
db830c46 | 1369 | atomic_set(&peer_req->pending_bios, n_bios); |
45bb912b LE |
1370 | do { |
1371 | bio = bios; | |
1372 | bios = bios->bi_next; | |
1373 | bio->bi_next = NULL; | |
1374 | ||
45bb912b | 1375 | drbd_generic_make_request(mdev, fault_type, bio); |
45bb912b | 1376 | } while (bios); |
45bb912b LE |
1377 | return 0; |
1378 | ||
1379 | fail: | |
1380 | while (bios) { | |
1381 | bio = bios; | |
1382 | bios = bios->bi_next; | |
1383 | bio_put(bio); | |
1384 | } | |
10f6d992 | 1385 | return err; |
45bb912b LE |
1386 | } |
1387 | ||
53840641 | 1388 | static void drbd_remove_epoch_entry_interval(struct drbd_conf *mdev, |
db830c46 | 1389 | struct drbd_peer_request *peer_req) |
53840641 | 1390 | { |
db830c46 | 1391 | struct drbd_interval *i = &peer_req->i; |
53840641 AG |
1392 | |
1393 | drbd_remove_interval(&mdev->write_requests, i); | |
1394 | drbd_clear_interval(i); | |
1395 | ||
6c852bec | 1396 | /* Wake up any processes waiting for this peer request to complete. */ |
53840641 AG |
1397 | if (i->waiting) |
1398 | wake_up(&mdev->misc_wait); | |
1399 | } | |
1400 | ||
77fede51 PR |
1401 | void conn_wait_active_ee_empty(struct drbd_tconn *tconn) |
1402 | { | |
1403 | struct drbd_conf *mdev; | |
1404 | int vnr; | |
1405 | ||
1406 | rcu_read_lock(); | |
1407 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { | |
1408 | kref_get(&mdev->kref); | |
1409 | rcu_read_unlock(); | |
1410 | drbd_wait_ee_list_empty(mdev, &mdev->active_ee); | |
1411 | kref_put(&mdev->kref, &drbd_minor_destroy); | |
1412 | rcu_read_lock(); | |
1413 | } | |
1414 | rcu_read_unlock(); | |
1415 | } | |
1416 | ||
4a76b161 | 1417 | static int receive_Barrier(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 1418 | { |
2451fc3b | 1419 | int rv; |
e658983a | 1420 | struct p_barrier *p = pi->data; |
b411b363 PR |
1421 | struct drbd_epoch *epoch; |
1422 | ||
9ed57dcb LE |
1423 | /* FIXME these are unacked on connection, |
1424 | * not a specific (peer)device. | |
1425 | */ | |
12038a3a | 1426 | tconn->current_epoch->barrier_nr = p->barrier; |
9ed57dcb | 1427 | tconn->current_epoch->tconn = tconn; |
1e9dd291 | 1428 | rv = drbd_may_finish_epoch(tconn, tconn->current_epoch, EV_GOT_BARRIER_NR); |
b411b363 PR |
1429 | |
1430 | /* P_BARRIER_ACK may imply that the corresponding extent is dropped from | |
1431 | * the activity log, which means it would not be resynced in case the | |
1432 | * R_PRIMARY crashes now. | |
1433 | * Therefore we must send the barrier_ack after the barrier request was | |
1434 | * completed. */ | |
4b0007c0 | 1435 | switch (tconn->write_ordering) { |
b411b363 PR |
1436 | case WO_none: |
1437 | if (rv == FE_RECYCLED) | |
82bc0194 | 1438 | return 0; |
2451fc3b PR |
1439 | |
1440 | /* receiver context, in the writeout path of the other node. | |
1441 | * avoid potential distributed deadlock */ | |
1442 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1443 | if (epoch) | |
1444 | break; | |
1445 | else | |
9ed57dcb | 1446 | conn_warn(tconn, "Allocation of an epoch failed, slowing down\n"); |
2451fc3b | 1447 | /* Fall through */ |
b411b363 PR |
1448 | |
1449 | case WO_bdev_flush: | |
1450 | case WO_drain_io: | |
77fede51 | 1451 | conn_wait_active_ee_empty(tconn); |
4b0007c0 | 1452 | drbd_flush(tconn); |
2451fc3b | 1453 | |
12038a3a | 1454 | if (atomic_read(&tconn->current_epoch->epoch_size)) { |
2451fc3b PR |
1455 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); |
1456 | if (epoch) | |
1457 | break; | |
b411b363 PR |
1458 | } |
1459 | ||
82bc0194 | 1460 | return 0; |
2451fc3b | 1461 | default: |
9ed57dcb | 1462 | conn_err(tconn, "Strangeness in tconn->write_ordering %d\n", tconn->write_ordering); |
82bc0194 | 1463 | return -EIO; |
b411b363 PR |
1464 | } |
1465 | ||
1466 | epoch->flags = 0; | |
1467 | atomic_set(&epoch->epoch_size, 0); | |
1468 | atomic_set(&epoch->active, 0); | |
1469 | ||
12038a3a PR |
1470 | spin_lock(&tconn->epoch_lock); |
1471 | if (atomic_read(&tconn->current_epoch->epoch_size)) { | |
1472 | list_add(&epoch->list, &tconn->current_epoch->list); | |
1473 | tconn->current_epoch = epoch; | |
1474 | tconn->epochs++; | |
b411b363 PR |
1475 | } else { |
1476 | /* The current_epoch got recycled while we allocated this one... */ | |
1477 | kfree(epoch); | |
1478 | } | |
12038a3a | 1479 | spin_unlock(&tconn->epoch_lock); |
b411b363 | 1480 | |
82bc0194 | 1481 | return 0; |
b411b363 PR |
1482 | } |
1483 | ||
1484 | /* used from receive_RSDataReply (recv_resync_read) | |
1485 | * and from receive_Data */ | |
f6ffca9f AG |
1486 | static struct drbd_peer_request * |
1487 | read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, | |
1488 | int data_size) __must_hold(local) | |
b411b363 | 1489 | { |
6666032a | 1490 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); |
db830c46 | 1491 | struct drbd_peer_request *peer_req; |
b411b363 | 1492 | struct page *page; |
a5c31904 | 1493 | int dgs, ds, err; |
a0638456 PR |
1494 | void *dig_in = mdev->tconn->int_dig_in; |
1495 | void *dig_vv = mdev->tconn->int_dig_vv; | |
6b4388ac | 1496 | unsigned long *data; |
b411b363 | 1497 | |
88104ca4 AG |
1498 | dgs = 0; |
1499 | if (mdev->tconn->peer_integrity_tfm) { | |
1500 | dgs = crypto_hash_digestsize(mdev->tconn->peer_integrity_tfm); | |
9f5bdc33 AG |
1501 | /* |
1502 | * FIXME: Receive the incoming digest into the receive buffer | |
1503 | * here, together with its struct p_data? | |
1504 | */ | |
a5c31904 AG |
1505 | err = drbd_recv_all_warn(mdev->tconn, dig_in, dgs); |
1506 | if (err) | |
b411b363 | 1507 | return NULL; |
88104ca4 | 1508 | data_size -= dgs; |
b411b363 PR |
1509 | } |
1510 | ||
841ce241 AG |
1511 | if (!expect(IS_ALIGNED(data_size, 512))) |
1512 | return NULL; | |
1513 | if (!expect(data_size <= DRBD_MAX_BIO_SIZE)) | |
1514 | return NULL; | |
b411b363 | 1515 | |
6666032a LE |
1516 | /* even though we trust out peer, |
1517 | * we sometimes have to double check. */ | |
1518 | if (sector + (data_size>>9) > capacity) { | |
fdda6544 LE |
1519 | dev_err(DEV, "request from peer beyond end of local disk: " |
1520 | "capacity: %llus < sector: %llus + size: %u\n", | |
6666032a LE |
1521 | (unsigned long long)capacity, |
1522 | (unsigned long long)sector, data_size); | |
1523 | return NULL; | |
1524 | } | |
1525 | ||
b411b363 PR |
1526 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD |
1527 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
1528 | * which in turn might block on the other node at this very place. */ | |
0db55363 | 1529 | peer_req = drbd_alloc_peer_req(mdev, id, sector, data_size, GFP_NOIO); |
db830c46 | 1530 | if (!peer_req) |
b411b363 | 1531 | return NULL; |
45bb912b | 1532 | |
a73ff323 | 1533 | if (!data_size) |
81a3537a | 1534 | return peer_req; |
a73ff323 | 1535 | |
b411b363 | 1536 | ds = data_size; |
db830c46 | 1537 | page = peer_req->pages; |
45bb912b LE |
1538 | page_chain_for_each(page) { |
1539 | unsigned len = min_t(int, ds, PAGE_SIZE); | |
6b4388ac | 1540 | data = kmap(page); |
a5c31904 | 1541 | err = drbd_recv_all_warn(mdev->tconn, data, len); |
0cf9d27e | 1542 | if (drbd_insert_fault(mdev, DRBD_FAULT_RECEIVE)) { |
6b4388ac PR |
1543 | dev_err(DEV, "Fault injection: Corrupting data on receive\n"); |
1544 | data[0] = data[0] ^ (unsigned long)-1; | |
1545 | } | |
b411b363 | 1546 | kunmap(page); |
a5c31904 | 1547 | if (err) { |
3967deb1 | 1548 | drbd_free_peer_req(mdev, peer_req); |
b411b363 PR |
1549 | return NULL; |
1550 | } | |
a5c31904 | 1551 | ds -= len; |
b411b363 PR |
1552 | } |
1553 | ||
1554 | if (dgs) { | |
5b614abe | 1555 | drbd_csum_ee(mdev, mdev->tconn->peer_integrity_tfm, peer_req, dig_vv); |
b411b363 | 1556 | if (memcmp(dig_in, dig_vv, dgs)) { |
470be44a LE |
1557 | dev_err(DEV, "Digest integrity check FAILED: %llus +%u\n", |
1558 | (unsigned long long)sector, data_size); | |
3967deb1 | 1559 | drbd_free_peer_req(mdev, peer_req); |
b411b363 PR |
1560 | return NULL; |
1561 | } | |
1562 | } | |
1563 | mdev->recv_cnt += data_size>>9; | |
db830c46 | 1564 | return peer_req; |
b411b363 PR |
1565 | } |
1566 | ||
1567 | /* drbd_drain_block() just takes a data block | |
1568 | * out of the socket input buffer, and discards it. | |
1569 | */ | |
1570 | static int drbd_drain_block(struct drbd_conf *mdev, int data_size) | |
1571 | { | |
1572 | struct page *page; | |
a5c31904 | 1573 | int err = 0; |
b411b363 PR |
1574 | void *data; |
1575 | ||
c3470cde | 1576 | if (!data_size) |
fc5be839 | 1577 | return 0; |
c3470cde | 1578 | |
c37c8ecf | 1579 | page = drbd_alloc_pages(mdev, 1, 1); |
b411b363 PR |
1580 | |
1581 | data = kmap(page); | |
1582 | while (data_size) { | |
fc5be839 AG |
1583 | unsigned int len = min_t(int, data_size, PAGE_SIZE); |
1584 | ||
a5c31904 AG |
1585 | err = drbd_recv_all_warn(mdev->tconn, data, len); |
1586 | if (err) | |
b411b363 | 1587 | break; |
a5c31904 | 1588 | data_size -= len; |
b411b363 PR |
1589 | } |
1590 | kunmap(page); | |
5cc287e0 | 1591 | drbd_free_pages(mdev, page, 0); |
fc5be839 | 1592 | return err; |
b411b363 PR |
1593 | } |
1594 | ||
1595 | static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req, | |
1596 | sector_t sector, int data_size) | |
1597 | { | |
1598 | struct bio_vec *bvec; | |
1599 | struct bio *bio; | |
a5c31904 | 1600 | int dgs, err, i, expect; |
a0638456 PR |
1601 | void *dig_in = mdev->tconn->int_dig_in; |
1602 | void *dig_vv = mdev->tconn->int_dig_vv; | |
b411b363 | 1603 | |
88104ca4 AG |
1604 | dgs = 0; |
1605 | if (mdev->tconn->peer_integrity_tfm) { | |
1606 | dgs = crypto_hash_digestsize(mdev->tconn->peer_integrity_tfm); | |
a5c31904 AG |
1607 | err = drbd_recv_all_warn(mdev->tconn, dig_in, dgs); |
1608 | if (err) | |
1609 | return err; | |
88104ca4 | 1610 | data_size -= dgs; |
b411b363 PR |
1611 | } |
1612 | ||
b411b363 PR |
1613 | /* optimistically update recv_cnt. if receiving fails below, |
1614 | * we disconnect anyways, and counters will be reset. */ | |
1615 | mdev->recv_cnt += data_size>>9; | |
1616 | ||
1617 | bio = req->master_bio; | |
4f024f37 | 1618 | D_ASSERT(sector == bio->bi_iter.bi_sector); |
b411b363 PR |
1619 | |
1620 | bio_for_each_segment(bvec, bio, i) { | |
a5c31904 | 1621 | void *mapped = kmap(bvec->bv_page) + bvec->bv_offset; |
b411b363 | 1622 | expect = min_t(int, data_size, bvec->bv_len); |
a5c31904 | 1623 | err = drbd_recv_all_warn(mdev->tconn, mapped, expect); |
b411b363 | 1624 | kunmap(bvec->bv_page); |
a5c31904 AG |
1625 | if (err) |
1626 | return err; | |
1627 | data_size -= expect; | |
b411b363 PR |
1628 | } |
1629 | ||
1630 | if (dgs) { | |
5b614abe | 1631 | drbd_csum_bio(mdev, mdev->tconn->peer_integrity_tfm, bio, dig_vv); |
b411b363 PR |
1632 | if (memcmp(dig_in, dig_vv, dgs)) { |
1633 | dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n"); | |
28284cef | 1634 | return -EINVAL; |
b411b363 PR |
1635 | } |
1636 | } | |
1637 | ||
1638 | D_ASSERT(data_size == 0); | |
28284cef | 1639 | return 0; |
b411b363 PR |
1640 | } |
1641 | ||
a990be46 AG |
1642 | /* |
1643 | * e_end_resync_block() is called in asender context via | |
1644 | * drbd_finish_peer_reqs(). | |
1645 | */ | |
99920dc5 | 1646 | static int e_end_resync_block(struct drbd_work *w, int unused) |
b411b363 | 1647 | { |
8050e6d0 AG |
1648 | struct drbd_peer_request *peer_req = |
1649 | container_of(w, struct drbd_peer_request, w); | |
00d56944 | 1650 | struct drbd_conf *mdev = w->mdev; |
db830c46 | 1651 | sector_t sector = peer_req->i.sector; |
99920dc5 | 1652 | int err; |
b411b363 | 1653 | |
db830c46 | 1654 | D_ASSERT(drbd_interval_empty(&peer_req->i)); |
b411b363 | 1655 | |
db830c46 AG |
1656 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
1657 | drbd_set_in_sync(mdev, sector, peer_req->i.size); | |
99920dc5 | 1658 | err = drbd_send_ack(mdev, P_RS_WRITE_ACK, peer_req); |
b411b363 PR |
1659 | } else { |
1660 | /* Record failure to sync */ | |
db830c46 | 1661 | drbd_rs_failed_io(mdev, sector, peer_req->i.size); |
b411b363 | 1662 | |
99920dc5 | 1663 | err = drbd_send_ack(mdev, P_NEG_ACK, peer_req); |
b411b363 PR |
1664 | } |
1665 | dec_unacked(mdev); | |
1666 | ||
99920dc5 | 1667 | return err; |
b411b363 PR |
1668 | } |
1669 | ||
1670 | static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local) | |
1671 | { | |
db830c46 | 1672 | struct drbd_peer_request *peer_req; |
b411b363 | 1673 | |
db830c46 AG |
1674 | peer_req = read_in_block(mdev, ID_SYNCER, sector, data_size); |
1675 | if (!peer_req) | |
45bb912b | 1676 | goto fail; |
b411b363 PR |
1677 | |
1678 | dec_rs_pending(mdev); | |
1679 | ||
b411b363 PR |
1680 | inc_unacked(mdev); |
1681 | /* corresponding dec_unacked() in e_end_resync_block() | |
1682 | * respective _drbd_clear_done_ee */ | |
1683 | ||
db830c46 | 1684 | peer_req->w.cb = e_end_resync_block; |
45bb912b | 1685 | |
87eeee41 | 1686 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 1687 | list_add(&peer_req->w.list, &mdev->sync_ee); |
87eeee41 | 1688 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 1689 | |
0f0601f4 | 1690 | atomic_add(data_size >> 9, &mdev->rs_sect_ev); |
fbe29dec | 1691 | if (drbd_submit_peer_request(mdev, peer_req, WRITE, DRBD_FAULT_RS_WR) == 0) |
e1c1b0fc | 1692 | return 0; |
b411b363 | 1693 | |
10f6d992 LE |
1694 | /* don't care for the reason here */ |
1695 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 1696 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 1697 | list_del(&peer_req->w.list); |
87eeee41 | 1698 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 | 1699 | |
3967deb1 | 1700 | drbd_free_peer_req(mdev, peer_req); |
45bb912b LE |
1701 | fail: |
1702 | put_ldev(mdev); | |
e1c1b0fc | 1703 | return -EIO; |
b411b363 PR |
1704 | } |
1705 | ||
668eebc6 | 1706 | static struct drbd_request * |
bc9c5c41 AG |
1707 | find_request(struct drbd_conf *mdev, struct rb_root *root, u64 id, |
1708 | sector_t sector, bool missing_ok, const char *func) | |
51624585 | 1709 | { |
51624585 AG |
1710 | struct drbd_request *req; |
1711 | ||
bc9c5c41 AG |
1712 | /* Request object according to our peer */ |
1713 | req = (struct drbd_request *)(unsigned long)id; | |
5e472264 | 1714 | if (drbd_contains_interval(root, sector, &req->i) && req->i.local) |
668eebc6 | 1715 | return req; |
c3afd8f5 | 1716 | if (!missing_ok) { |
5af172ed | 1717 | dev_err(DEV, "%s: failed to find request 0x%lx, sector %llus\n", func, |
c3afd8f5 AG |
1718 | (unsigned long)id, (unsigned long long)sector); |
1719 | } | |
51624585 | 1720 | return NULL; |
b411b363 PR |
1721 | } |
1722 | ||
4a76b161 | 1723 | static int receive_DataReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 1724 | { |
4a76b161 | 1725 | struct drbd_conf *mdev; |
b411b363 PR |
1726 | struct drbd_request *req; |
1727 | sector_t sector; | |
82bc0194 | 1728 | int err; |
e658983a | 1729 | struct p_data *p = pi->data; |
4a76b161 AG |
1730 | |
1731 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
1732 | if (!mdev) | |
1733 | return -EIO; | |
b411b363 PR |
1734 | |
1735 | sector = be64_to_cpu(p->sector); | |
1736 | ||
87eeee41 | 1737 | spin_lock_irq(&mdev->tconn->req_lock); |
bc9c5c41 | 1738 | req = find_request(mdev, &mdev->read_requests, p->block_id, sector, false, __func__); |
87eeee41 | 1739 | spin_unlock_irq(&mdev->tconn->req_lock); |
c3afd8f5 | 1740 | if (unlikely(!req)) |
82bc0194 | 1741 | return -EIO; |
b411b363 | 1742 | |
24c4830c | 1743 | /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid |
b411b363 PR |
1744 | * special casing it there for the various failure cases. |
1745 | * still no race with drbd_fail_pending_reads */ | |
e2857216 | 1746 | err = recv_dless_read(mdev, req, sector, pi->size); |
82bc0194 | 1747 | if (!err) |
8554df1c | 1748 | req_mod(req, DATA_RECEIVED); |
b411b363 PR |
1749 | /* else: nothing. handled from drbd_disconnect... |
1750 | * I don't think we may complete this just yet | |
1751 | * in case we are "on-disconnect: freeze" */ | |
1752 | ||
82bc0194 | 1753 | return err; |
b411b363 PR |
1754 | } |
1755 | ||
4a76b161 | 1756 | static int receive_RSDataReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 1757 | { |
4a76b161 | 1758 | struct drbd_conf *mdev; |
b411b363 | 1759 | sector_t sector; |
82bc0194 | 1760 | int err; |
e658983a | 1761 | struct p_data *p = pi->data; |
4a76b161 AG |
1762 | |
1763 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
1764 | if (!mdev) | |
1765 | return -EIO; | |
b411b363 PR |
1766 | |
1767 | sector = be64_to_cpu(p->sector); | |
1768 | D_ASSERT(p->block_id == ID_SYNCER); | |
1769 | ||
1770 | if (get_ldev(mdev)) { | |
1771 | /* data is submitted to disk within recv_resync_read. | |
1772 | * corresponding put_ldev done below on error, | |
fcefa62e | 1773 | * or in drbd_peer_request_endio. */ |
e2857216 | 1774 | err = recv_resync_read(mdev, sector, pi->size); |
b411b363 PR |
1775 | } else { |
1776 | if (__ratelimit(&drbd_ratelimit_state)) | |
1777 | dev_err(DEV, "Can not write resync data to local disk.\n"); | |
1778 | ||
e2857216 | 1779 | err = drbd_drain_block(mdev, pi->size); |
b411b363 | 1780 | |
e2857216 | 1781 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, pi->size); |
b411b363 PR |
1782 | } |
1783 | ||
e2857216 | 1784 | atomic_add(pi->size >> 9, &mdev->rs_sect_in); |
778f271d | 1785 | |
82bc0194 | 1786 | return err; |
b411b363 PR |
1787 | } |
1788 | ||
7be8da07 AG |
1789 | static void restart_conflicting_writes(struct drbd_conf *mdev, |
1790 | sector_t sector, int size) | |
b411b363 | 1791 | { |
7be8da07 AG |
1792 | struct drbd_interval *i; |
1793 | struct drbd_request *req; | |
1794 | ||
1795 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
1796 | if (!i->local) | |
1797 | continue; | |
1798 | req = container_of(i, struct drbd_request, i); | |
1799 | if (req->rq_state & RQ_LOCAL_PENDING || | |
1800 | !(req->rq_state & RQ_POSTPONED)) | |
1801 | continue; | |
2312f0b3 LE |
1802 | /* as it is RQ_POSTPONED, this will cause it to |
1803 | * be queued on the retry workqueue. */ | |
d4dabbe2 | 1804 | __req_mod(req, CONFLICT_RESOLVED, NULL); |
7be8da07 AG |
1805 | } |
1806 | } | |
b411b363 | 1807 | |
a990be46 AG |
1808 | /* |
1809 | * e_end_block() is called in asender context via drbd_finish_peer_reqs(). | |
b411b363 | 1810 | */ |
99920dc5 | 1811 | static int e_end_block(struct drbd_work *w, int cancel) |
b411b363 | 1812 | { |
8050e6d0 AG |
1813 | struct drbd_peer_request *peer_req = |
1814 | container_of(w, struct drbd_peer_request, w); | |
00d56944 | 1815 | struct drbd_conf *mdev = w->mdev; |
db830c46 | 1816 | sector_t sector = peer_req->i.sector; |
99920dc5 | 1817 | int err = 0, pcmd; |
b411b363 | 1818 | |
303d1448 | 1819 | if (peer_req->flags & EE_SEND_WRITE_ACK) { |
db830c46 | 1820 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1821 | pcmd = (mdev->state.conn >= C_SYNC_SOURCE && |
1822 | mdev->state.conn <= C_PAUSED_SYNC_T && | |
db830c46 | 1823 | peer_req->flags & EE_MAY_SET_IN_SYNC) ? |
b411b363 | 1824 | P_RS_WRITE_ACK : P_WRITE_ACK; |
99920dc5 | 1825 | err = drbd_send_ack(mdev, pcmd, peer_req); |
b411b363 | 1826 | if (pcmd == P_RS_WRITE_ACK) |
db830c46 | 1827 | drbd_set_in_sync(mdev, sector, peer_req->i.size); |
b411b363 | 1828 | } else { |
99920dc5 | 1829 | err = drbd_send_ack(mdev, P_NEG_ACK, peer_req); |
b411b363 PR |
1830 | /* we expect it to be marked out of sync anyways... |
1831 | * maybe assert this? */ | |
1832 | } | |
1833 | dec_unacked(mdev); | |
1834 | } | |
1835 | /* we delete from the conflict detection hash _after_ we sent out the | |
1836 | * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ | |
302bdeae | 1837 | if (peer_req->flags & EE_IN_INTERVAL_TREE) { |
87eeee41 | 1838 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 AG |
1839 | D_ASSERT(!drbd_interval_empty(&peer_req->i)); |
1840 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
7be8da07 AG |
1841 | if (peer_req->flags & EE_RESTART_REQUESTS) |
1842 | restart_conflicting_writes(mdev, sector, peer_req->i.size); | |
87eeee41 | 1843 | spin_unlock_irq(&mdev->tconn->req_lock); |
bb3bfe96 | 1844 | } else |
db830c46 | 1845 | D_ASSERT(drbd_interval_empty(&peer_req->i)); |
b411b363 | 1846 | |
1e9dd291 | 1847 | drbd_may_finish_epoch(mdev->tconn, peer_req->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); |
b411b363 | 1848 | |
99920dc5 | 1849 | return err; |
b411b363 PR |
1850 | } |
1851 | ||
7be8da07 | 1852 | static int e_send_ack(struct drbd_work *w, enum drbd_packet ack) |
b411b363 | 1853 | { |
7be8da07 | 1854 | struct drbd_conf *mdev = w->mdev; |
8050e6d0 AG |
1855 | struct drbd_peer_request *peer_req = |
1856 | container_of(w, struct drbd_peer_request, w); | |
99920dc5 | 1857 | int err; |
b411b363 | 1858 | |
99920dc5 | 1859 | err = drbd_send_ack(mdev, ack, peer_req); |
b411b363 PR |
1860 | dec_unacked(mdev); |
1861 | ||
99920dc5 | 1862 | return err; |
b411b363 PR |
1863 | } |
1864 | ||
d4dabbe2 | 1865 | static int e_send_superseded(struct drbd_work *w, int unused) |
7be8da07 | 1866 | { |
d4dabbe2 | 1867 | return e_send_ack(w, P_SUPERSEDED); |
7be8da07 AG |
1868 | } |
1869 | ||
99920dc5 | 1870 | static int e_send_retry_write(struct drbd_work *w, int unused) |
7be8da07 AG |
1871 | { |
1872 | struct drbd_tconn *tconn = w->mdev->tconn; | |
1873 | ||
1874 | return e_send_ack(w, tconn->agreed_pro_version >= 100 ? | |
d4dabbe2 | 1875 | P_RETRY_WRITE : P_SUPERSEDED); |
7be8da07 | 1876 | } |
b411b363 | 1877 | |
3e394da1 AG |
1878 | static bool seq_greater(u32 a, u32 b) |
1879 | { | |
1880 | /* | |
1881 | * We assume 32-bit wrap-around here. | |
1882 | * For 24-bit wrap-around, we would have to shift: | |
1883 | * a <<= 8; b <<= 8; | |
1884 | */ | |
1885 | return (s32)a - (s32)b > 0; | |
1886 | } | |
b411b363 | 1887 | |
3e394da1 AG |
1888 | static u32 seq_max(u32 a, u32 b) |
1889 | { | |
1890 | return seq_greater(a, b) ? a : b; | |
b411b363 PR |
1891 | } |
1892 | ||
43ae077d | 1893 | static void update_peer_seq(struct drbd_conf *mdev, unsigned int peer_seq) |
3e394da1 | 1894 | { |
3c13b680 | 1895 | unsigned int newest_peer_seq; |
3e394da1 | 1896 | |
b874d231 | 1897 | if (test_bit(RESOLVE_CONFLICTS, &mdev->tconn->flags)) { |
7be8da07 | 1898 | spin_lock(&mdev->peer_seq_lock); |
3c13b680 LE |
1899 | newest_peer_seq = seq_max(mdev->peer_seq, peer_seq); |
1900 | mdev->peer_seq = newest_peer_seq; | |
7be8da07 | 1901 | spin_unlock(&mdev->peer_seq_lock); |
3c13b680 LE |
1902 | /* wake up only if we actually changed mdev->peer_seq */ |
1903 | if (peer_seq == newest_peer_seq) | |
7be8da07 AG |
1904 | wake_up(&mdev->seq_wait); |
1905 | } | |
b411b363 PR |
1906 | } |
1907 | ||
d93f6302 | 1908 | static inline int overlaps(sector_t s1, int l1, sector_t s2, int l2) |
b6a370ba | 1909 | { |
d93f6302 LE |
1910 | return !((s1 + (l1>>9) <= s2) || (s1 >= s2 + (l2>>9))); |
1911 | } | |
b6a370ba | 1912 | |
d93f6302 | 1913 | /* maybe change sync_ee into interval trees as well? */ |
3ea35df8 | 1914 | static bool overlapping_resync_write(struct drbd_conf *mdev, struct drbd_peer_request *peer_req) |
d93f6302 LE |
1915 | { |
1916 | struct drbd_peer_request *rs_req; | |
b6a370ba PR |
1917 | bool rv = 0; |
1918 | ||
d93f6302 LE |
1919 | spin_lock_irq(&mdev->tconn->req_lock); |
1920 | list_for_each_entry(rs_req, &mdev->sync_ee, w.list) { | |
1921 | if (overlaps(peer_req->i.sector, peer_req->i.size, | |
1922 | rs_req->i.sector, rs_req->i.size)) { | |
b6a370ba PR |
1923 | rv = 1; |
1924 | break; | |
1925 | } | |
1926 | } | |
d93f6302 | 1927 | spin_unlock_irq(&mdev->tconn->req_lock); |
b6a370ba PR |
1928 | |
1929 | return rv; | |
1930 | } | |
1931 | ||
b411b363 PR |
1932 | /* Called from receive_Data. |
1933 | * Synchronize packets on sock with packets on msock. | |
1934 | * | |
1935 | * This is here so even when a P_DATA packet traveling via sock overtook an Ack | |
1936 | * packet traveling on msock, they are still processed in the order they have | |
1937 | * been sent. | |
1938 | * | |
1939 | * Note: we don't care for Ack packets overtaking P_DATA packets. | |
1940 | * | |
1941 | * In case packet_seq is larger than mdev->peer_seq number, there are | |
1942 | * outstanding packets on the msock. We wait for them to arrive. | |
1943 | * In case we are the logically next packet, we update mdev->peer_seq | |
1944 | * ourselves. Correctly handles 32bit wrap around. | |
1945 | * | |
1946 | * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, | |
1947 | * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds | |
1948 | * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have | |
1949 | * 1<<9 == 512 seconds aka ages for the 32bit wrap around... | |
1950 | * | |
1951 | * returns 0 if we may process the packet, | |
1952 | * -ERESTARTSYS if we were interrupted (by disconnect signal). */ | |
7be8da07 | 1953 | static int wait_for_and_update_peer_seq(struct drbd_conf *mdev, const u32 peer_seq) |
b411b363 PR |
1954 | { |
1955 | DEFINE_WAIT(wait); | |
b411b363 | 1956 | long timeout; |
b874d231 | 1957 | int ret = 0, tp; |
7be8da07 | 1958 | |
b874d231 | 1959 | if (!test_bit(RESOLVE_CONFLICTS, &mdev->tconn->flags)) |
7be8da07 AG |
1960 | return 0; |
1961 | ||
b411b363 PR |
1962 | spin_lock(&mdev->peer_seq_lock); |
1963 | for (;;) { | |
7be8da07 AG |
1964 | if (!seq_greater(peer_seq - 1, mdev->peer_seq)) { |
1965 | mdev->peer_seq = seq_max(mdev->peer_seq, peer_seq); | |
b411b363 | 1966 | break; |
7be8da07 | 1967 | } |
b874d231 | 1968 | |
b411b363 PR |
1969 | if (signal_pending(current)) { |
1970 | ret = -ERESTARTSYS; | |
1971 | break; | |
1972 | } | |
b874d231 PR |
1973 | |
1974 | rcu_read_lock(); | |
1975 | tp = rcu_dereference(mdev->tconn->net_conf)->two_primaries; | |
1976 | rcu_read_unlock(); | |
1977 | ||
1978 | if (!tp) | |
1979 | break; | |
1980 | ||
1981 | /* Only need to wait if two_primaries is enabled */ | |
7be8da07 | 1982 | prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE); |
b411b363 | 1983 | spin_unlock(&mdev->peer_seq_lock); |
44ed167d PR |
1984 | rcu_read_lock(); |
1985 | timeout = rcu_dereference(mdev->tconn->net_conf)->ping_timeo*HZ/10; | |
1986 | rcu_read_unlock(); | |
71b1c1eb | 1987 | timeout = schedule_timeout(timeout); |
b411b363 | 1988 | spin_lock(&mdev->peer_seq_lock); |
7be8da07 | 1989 | if (!timeout) { |
b411b363 | 1990 | ret = -ETIMEDOUT; |
71b1c1eb | 1991 | dev_err(DEV, "Timed out waiting for missing ack packets; disconnecting\n"); |
b411b363 PR |
1992 | break; |
1993 | } | |
1994 | } | |
b411b363 | 1995 | spin_unlock(&mdev->peer_seq_lock); |
7be8da07 | 1996 | finish_wait(&mdev->seq_wait, &wait); |
b411b363 PR |
1997 | return ret; |
1998 | } | |
1999 | ||
688593c5 LE |
2000 | /* see also bio_flags_to_wire() |
2001 | * DRBD_REQ_*, because we need to semantically map the flags to data packet | |
2002 | * flags and back. We may replicate to other kernel versions. */ | |
2003 | static unsigned long wire_flags_to_bio(struct drbd_conf *mdev, u32 dpf) | |
76d2e7ec | 2004 | { |
688593c5 LE |
2005 | return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | |
2006 | (dpf & DP_FUA ? REQ_FUA : 0) | | |
2007 | (dpf & DP_FLUSH ? REQ_FLUSH : 0) | | |
2008 | (dpf & DP_DISCARD ? REQ_DISCARD : 0); | |
76d2e7ec PR |
2009 | } |
2010 | ||
7be8da07 AG |
2011 | static void fail_postponed_requests(struct drbd_conf *mdev, sector_t sector, |
2012 | unsigned int size) | |
2013 | { | |
2014 | struct drbd_interval *i; | |
2015 | ||
2016 | repeat: | |
2017 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
2018 | struct drbd_request *req; | |
2019 | struct bio_and_error m; | |
2020 | ||
2021 | if (!i->local) | |
2022 | continue; | |
2023 | req = container_of(i, struct drbd_request, i); | |
2024 | if (!(req->rq_state & RQ_POSTPONED)) | |
2025 | continue; | |
2026 | req->rq_state &= ~RQ_POSTPONED; | |
2027 | __req_mod(req, NEG_ACKED, &m); | |
2028 | spin_unlock_irq(&mdev->tconn->req_lock); | |
2029 | if (m.bio) | |
2030 | complete_master_bio(mdev, &m); | |
2031 | spin_lock_irq(&mdev->tconn->req_lock); | |
2032 | goto repeat; | |
2033 | } | |
2034 | } | |
2035 | ||
2036 | static int handle_write_conflicts(struct drbd_conf *mdev, | |
2037 | struct drbd_peer_request *peer_req) | |
2038 | { | |
2039 | struct drbd_tconn *tconn = mdev->tconn; | |
427c0434 | 2040 | bool resolve_conflicts = test_bit(RESOLVE_CONFLICTS, &tconn->flags); |
7be8da07 AG |
2041 | sector_t sector = peer_req->i.sector; |
2042 | const unsigned int size = peer_req->i.size; | |
2043 | struct drbd_interval *i; | |
2044 | bool equal; | |
2045 | int err; | |
2046 | ||
2047 | /* | |
2048 | * Inserting the peer request into the write_requests tree will prevent | |
2049 | * new conflicting local requests from being added. | |
2050 | */ | |
2051 | drbd_insert_interval(&mdev->write_requests, &peer_req->i); | |
2052 | ||
2053 | repeat: | |
2054 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
2055 | if (i == &peer_req->i) | |
2056 | continue; | |
2057 | ||
2058 | if (!i->local) { | |
2059 | /* | |
2060 | * Our peer has sent a conflicting remote request; this | |
2061 | * should not happen in a two-node setup. Wait for the | |
2062 | * earlier peer request to complete. | |
2063 | */ | |
2064 | err = drbd_wait_misc(mdev, i); | |
2065 | if (err) | |
2066 | goto out; | |
2067 | goto repeat; | |
2068 | } | |
2069 | ||
2070 | equal = i->sector == sector && i->size == size; | |
2071 | if (resolve_conflicts) { | |
2072 | /* | |
2073 | * If the peer request is fully contained within the | |
d4dabbe2 LE |
2074 | * overlapping request, it can be considered overwritten |
2075 | * and thus superseded; otherwise, it will be retried | |
2076 | * once all overlapping requests have completed. | |
7be8da07 | 2077 | */ |
d4dabbe2 | 2078 | bool superseded = i->sector <= sector && i->sector + |
7be8da07 AG |
2079 | (i->size >> 9) >= sector + (size >> 9); |
2080 | ||
2081 | if (!equal) | |
2082 | dev_alert(DEV, "Concurrent writes detected: " | |
2083 | "local=%llus +%u, remote=%llus +%u, " | |
2084 | "assuming %s came first\n", | |
2085 | (unsigned long long)i->sector, i->size, | |
2086 | (unsigned long long)sector, size, | |
d4dabbe2 | 2087 | superseded ? "local" : "remote"); |
7be8da07 AG |
2088 | |
2089 | inc_unacked(mdev); | |
d4dabbe2 | 2090 | peer_req->w.cb = superseded ? e_send_superseded : |
7be8da07 AG |
2091 | e_send_retry_write; |
2092 | list_add_tail(&peer_req->w.list, &mdev->done_ee); | |
2093 | wake_asender(mdev->tconn); | |
2094 | ||
2095 | err = -ENOENT; | |
2096 | goto out; | |
2097 | } else { | |
2098 | struct drbd_request *req = | |
2099 | container_of(i, struct drbd_request, i); | |
2100 | ||
2101 | if (!equal) | |
2102 | dev_alert(DEV, "Concurrent writes detected: " | |
2103 | "local=%llus +%u, remote=%llus +%u\n", | |
2104 | (unsigned long long)i->sector, i->size, | |
2105 | (unsigned long long)sector, size); | |
2106 | ||
2107 | if (req->rq_state & RQ_LOCAL_PENDING || | |
2108 | !(req->rq_state & RQ_POSTPONED)) { | |
2109 | /* | |
2110 | * Wait for the node with the discard flag to | |
d4dabbe2 LE |
2111 | * decide if this request has been superseded |
2112 | * or needs to be retried. | |
2113 | * Requests that have been superseded will | |
7be8da07 AG |
2114 | * disappear from the write_requests tree. |
2115 | * | |
2116 | * In addition, wait for the conflicting | |
2117 | * request to finish locally before submitting | |
2118 | * the conflicting peer request. | |
2119 | */ | |
2120 | err = drbd_wait_misc(mdev, &req->i); | |
2121 | if (err) { | |
2122 | _conn_request_state(mdev->tconn, | |
2123 | NS(conn, C_TIMEOUT), | |
2124 | CS_HARD); | |
2125 | fail_postponed_requests(mdev, sector, size); | |
2126 | goto out; | |
2127 | } | |
2128 | goto repeat; | |
2129 | } | |
2130 | /* | |
2131 | * Remember to restart the conflicting requests after | |
2132 | * the new peer request has completed. | |
2133 | */ | |
2134 | peer_req->flags |= EE_RESTART_REQUESTS; | |
2135 | } | |
2136 | } | |
2137 | err = 0; | |
2138 | ||
2139 | out: | |
2140 | if (err) | |
2141 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
2142 | return err; | |
2143 | } | |
2144 | ||
b411b363 | 2145 | /* mirrored write */ |
4a76b161 | 2146 | static int receive_Data(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 2147 | { |
4a76b161 | 2148 | struct drbd_conf *mdev; |
b411b363 | 2149 | sector_t sector; |
db830c46 | 2150 | struct drbd_peer_request *peer_req; |
e658983a | 2151 | struct p_data *p = pi->data; |
7be8da07 | 2152 | u32 peer_seq = be32_to_cpu(p->seq_num); |
b411b363 PR |
2153 | int rw = WRITE; |
2154 | u32 dp_flags; | |
302bdeae | 2155 | int err, tp; |
b411b363 | 2156 | |
4a76b161 AG |
2157 | mdev = vnr_to_mdev(tconn, pi->vnr); |
2158 | if (!mdev) | |
2159 | return -EIO; | |
b411b363 | 2160 | |
7be8da07 | 2161 | if (!get_ldev(mdev)) { |
82bc0194 AG |
2162 | int err2; |
2163 | ||
7be8da07 | 2164 | err = wait_for_and_update_peer_seq(mdev, peer_seq); |
e2857216 | 2165 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, pi->size); |
12038a3a | 2166 | atomic_inc(&tconn->current_epoch->epoch_size); |
e2857216 | 2167 | err2 = drbd_drain_block(mdev, pi->size); |
82bc0194 AG |
2168 | if (!err) |
2169 | err = err2; | |
2170 | return err; | |
b411b363 PR |
2171 | } |
2172 | ||
fcefa62e AG |
2173 | /* |
2174 | * Corresponding put_ldev done either below (on various errors), or in | |
2175 | * drbd_peer_request_endio, if we successfully submit the data at the | |
2176 | * end of this function. | |
2177 | */ | |
b411b363 PR |
2178 | |
2179 | sector = be64_to_cpu(p->sector); | |
e2857216 | 2180 | peer_req = read_in_block(mdev, p->block_id, sector, pi->size); |
db830c46 | 2181 | if (!peer_req) { |
b411b363 | 2182 | put_ldev(mdev); |
82bc0194 | 2183 | return -EIO; |
b411b363 PR |
2184 | } |
2185 | ||
db830c46 | 2186 | peer_req->w.cb = e_end_block; |
b411b363 | 2187 | |
688593c5 LE |
2188 | dp_flags = be32_to_cpu(p->dp_flags); |
2189 | rw |= wire_flags_to_bio(mdev, dp_flags); | |
81a3537a LE |
2190 | if (peer_req->pages == NULL) { |
2191 | D_ASSERT(peer_req->i.size == 0); | |
a73ff323 LE |
2192 | D_ASSERT(dp_flags & DP_FLUSH); |
2193 | } | |
688593c5 LE |
2194 | |
2195 | if (dp_flags & DP_MAY_SET_IN_SYNC) | |
db830c46 | 2196 | peer_req->flags |= EE_MAY_SET_IN_SYNC; |
688593c5 | 2197 | |
12038a3a PR |
2198 | spin_lock(&tconn->epoch_lock); |
2199 | peer_req->epoch = tconn->current_epoch; | |
db830c46 AG |
2200 | atomic_inc(&peer_req->epoch->epoch_size); |
2201 | atomic_inc(&peer_req->epoch->active); | |
12038a3a | 2202 | spin_unlock(&tconn->epoch_lock); |
b411b363 | 2203 | |
302bdeae PR |
2204 | rcu_read_lock(); |
2205 | tp = rcu_dereference(mdev->tconn->net_conf)->two_primaries; | |
2206 | rcu_read_unlock(); | |
2207 | if (tp) { | |
2208 | peer_req->flags |= EE_IN_INTERVAL_TREE; | |
7be8da07 AG |
2209 | err = wait_for_and_update_peer_seq(mdev, peer_seq); |
2210 | if (err) | |
b411b363 | 2211 | goto out_interrupted; |
87eeee41 | 2212 | spin_lock_irq(&mdev->tconn->req_lock); |
7be8da07 AG |
2213 | err = handle_write_conflicts(mdev, peer_req); |
2214 | if (err) { | |
2215 | spin_unlock_irq(&mdev->tconn->req_lock); | |
2216 | if (err == -ENOENT) { | |
b411b363 | 2217 | put_ldev(mdev); |
82bc0194 | 2218 | return 0; |
b411b363 | 2219 | } |
7be8da07 | 2220 | goto out_interrupted; |
b411b363 | 2221 | } |
b874d231 PR |
2222 | } else { |
2223 | update_peer_seq(mdev, peer_seq); | |
7be8da07 | 2224 | spin_lock_irq(&mdev->tconn->req_lock); |
b874d231 | 2225 | } |
db830c46 | 2226 | list_add(&peer_req->w.list, &mdev->active_ee); |
87eeee41 | 2227 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 2228 | |
d93f6302 | 2229 | if (mdev->state.conn == C_SYNC_TARGET) |
3ea35df8 | 2230 | wait_event(mdev->ee_wait, !overlapping_resync_write(mdev, peer_req)); |
b411b363 | 2231 | |
303d1448 | 2232 | if (mdev->tconn->agreed_pro_version < 100) { |
44ed167d PR |
2233 | rcu_read_lock(); |
2234 | switch (rcu_dereference(mdev->tconn->net_conf)->wire_protocol) { | |
303d1448 PR |
2235 | case DRBD_PROT_C: |
2236 | dp_flags |= DP_SEND_WRITE_ACK; | |
2237 | break; | |
2238 | case DRBD_PROT_B: | |
2239 | dp_flags |= DP_SEND_RECEIVE_ACK; | |
2240 | break; | |
b411b363 | 2241 | } |
44ed167d | 2242 | rcu_read_unlock(); |
b411b363 PR |
2243 | } |
2244 | ||
303d1448 PR |
2245 | if (dp_flags & DP_SEND_WRITE_ACK) { |
2246 | peer_req->flags |= EE_SEND_WRITE_ACK; | |
b411b363 PR |
2247 | inc_unacked(mdev); |
2248 | /* corresponding dec_unacked() in e_end_block() | |
2249 | * respective _drbd_clear_done_ee */ | |
303d1448 PR |
2250 | } |
2251 | ||
2252 | if (dp_flags & DP_SEND_RECEIVE_ACK) { | |
b411b363 PR |
2253 | /* I really don't like it that the receiver thread |
2254 | * sends on the msock, but anyways */ | |
db830c46 | 2255 | drbd_send_ack(mdev, P_RECV_ACK, peer_req); |
b411b363 PR |
2256 | } |
2257 | ||
6719fb03 | 2258 | if (mdev->state.pdsk < D_INCONSISTENT) { |
b411b363 | 2259 | /* In case we have the only disk of the cluster, */ |
db830c46 AG |
2260 | drbd_set_out_of_sync(mdev, peer_req->i.sector, peer_req->i.size); |
2261 | peer_req->flags |= EE_CALL_AL_COMPLETE_IO; | |
2262 | peer_req->flags &= ~EE_MAY_SET_IN_SYNC; | |
56392d2f | 2263 | drbd_al_begin_io(mdev, &peer_req->i, true); |
b411b363 PR |
2264 | } |
2265 | ||
82bc0194 AG |
2266 | err = drbd_submit_peer_request(mdev, peer_req, rw, DRBD_FAULT_DT_WR); |
2267 | if (!err) | |
2268 | return 0; | |
b411b363 | 2269 | |
10f6d992 LE |
2270 | /* don't care for the reason here */ |
2271 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 2272 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 AG |
2273 | list_del(&peer_req->w.list); |
2274 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
87eeee41 | 2275 | spin_unlock_irq(&mdev->tconn->req_lock); |
db830c46 | 2276 | if (peer_req->flags & EE_CALL_AL_COMPLETE_IO) |
181286ad | 2277 | drbd_al_complete_io(mdev, &peer_req->i); |
22cc37a9 | 2278 | |
b411b363 | 2279 | out_interrupted: |
1e9dd291 | 2280 | drbd_may_finish_epoch(tconn, peer_req->epoch, EV_PUT + EV_CLEANUP); |
b411b363 | 2281 | put_ldev(mdev); |
3967deb1 | 2282 | drbd_free_peer_req(mdev, peer_req); |
82bc0194 | 2283 | return err; |
b411b363 PR |
2284 | } |
2285 | ||
0f0601f4 LE |
2286 | /* We may throttle resync, if the lower device seems to be busy, |
2287 | * and current sync rate is above c_min_rate. | |
2288 | * | |
2289 | * To decide whether or not the lower device is busy, we use a scheme similar | |
2290 | * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" | |
2291 | * (more than 64 sectors) of activity we cannot account for with our own resync | |
2292 | * activity, it obviously is "busy". | |
2293 | * | |
2294 | * The current sync rate used here uses only the most recent two step marks, | |
2295 | * to have a short time average so we can react faster. | |
2296 | */ | |
e3555d85 | 2297 | int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector) |
0f0601f4 LE |
2298 | { |
2299 | struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk; | |
2300 | unsigned long db, dt, dbdt; | |
e3555d85 | 2301 | struct lc_element *tmp; |
0f0601f4 LE |
2302 | int curr_events; |
2303 | int throttle = 0; | |
daeda1cc PR |
2304 | unsigned int c_min_rate; |
2305 | ||
2306 | rcu_read_lock(); | |
2307 | c_min_rate = rcu_dereference(mdev->ldev->disk_conf)->c_min_rate; | |
2308 | rcu_read_unlock(); | |
0f0601f4 LE |
2309 | |
2310 | /* feature disabled? */ | |
daeda1cc | 2311 | if (c_min_rate == 0) |
0f0601f4 LE |
2312 | return 0; |
2313 | ||
e3555d85 PR |
2314 | spin_lock_irq(&mdev->al_lock); |
2315 | tmp = lc_find(mdev->resync, BM_SECT_TO_EXT(sector)); | |
2316 | if (tmp) { | |
2317 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
2318 | if (test_bit(BME_PRIORITY, &bm_ext->flags)) { | |
2319 | spin_unlock_irq(&mdev->al_lock); | |
2320 | return 0; | |
2321 | } | |
2322 | /* Do not slow down if app IO is already waiting for this extent */ | |
2323 | } | |
2324 | spin_unlock_irq(&mdev->al_lock); | |
2325 | ||
0f0601f4 LE |
2326 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
2327 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
2328 | atomic_read(&mdev->rs_sect_ev); | |
e3555d85 | 2329 | |
0f0601f4 LE |
2330 | if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) { |
2331 | unsigned long rs_left; | |
2332 | int i; | |
2333 | ||
2334 | mdev->rs_last_events = curr_events; | |
2335 | ||
2336 | /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, | |
2337 | * approx. */ | |
2649f080 LE |
2338 | i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS; |
2339 | ||
2340 | if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T) | |
2341 | rs_left = mdev->ov_left; | |
2342 | else | |
2343 | rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; | |
0f0601f4 LE |
2344 | |
2345 | dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ; | |
2346 | if (!dt) | |
2347 | dt++; | |
2348 | db = mdev->rs_mark_left[i] - rs_left; | |
2349 | dbdt = Bit2KB(db/dt); | |
2350 | ||
daeda1cc | 2351 | if (dbdt > c_min_rate) |
0f0601f4 LE |
2352 | throttle = 1; |
2353 | } | |
2354 | return throttle; | |
2355 | } | |
2356 | ||
2357 | ||
4a76b161 | 2358 | static int receive_DataRequest(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 2359 | { |
4a76b161 | 2360 | struct drbd_conf *mdev; |
b411b363 | 2361 | sector_t sector; |
4a76b161 | 2362 | sector_t capacity; |
db830c46 | 2363 | struct drbd_peer_request *peer_req; |
b411b363 | 2364 | struct digest_info *di = NULL; |
b18b37be | 2365 | int size, verb; |
b411b363 | 2366 | unsigned int fault_type; |
e658983a | 2367 | struct p_block_req *p = pi->data; |
4a76b161 AG |
2368 | |
2369 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
2370 | if (!mdev) | |
2371 | return -EIO; | |
2372 | capacity = drbd_get_capacity(mdev->this_bdev); | |
b411b363 PR |
2373 | |
2374 | sector = be64_to_cpu(p->sector); | |
2375 | size = be32_to_cpu(p->blksize); | |
2376 | ||
c670a398 | 2377 | if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { |
b411b363 PR |
2378 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, |
2379 | (unsigned long long)sector, size); | |
82bc0194 | 2380 | return -EINVAL; |
b411b363 PR |
2381 | } |
2382 | if (sector + (size>>9) > capacity) { | |
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 | ||
2388 | if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) { | |
b18b37be | 2389 | verb = 1; |
e2857216 | 2390 | switch (pi->cmd) { |
b18b37be PR |
2391 | case P_DATA_REQUEST: |
2392 | drbd_send_ack_rp(mdev, P_NEG_DREPLY, p); | |
2393 | break; | |
2394 | case P_RS_DATA_REQUEST: | |
2395 | case P_CSUM_RS_REQUEST: | |
2396 | case P_OV_REQUEST: | |
2397 | drbd_send_ack_rp(mdev, P_NEG_RS_DREPLY , p); | |
2398 | break; | |
2399 | case P_OV_REPLY: | |
2400 | verb = 0; | |
2401 | dec_rs_pending(mdev); | |
2402 | drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, ID_IN_SYNC); | |
2403 | break; | |
2404 | default: | |
49ba9b1b | 2405 | BUG(); |
b18b37be PR |
2406 | } |
2407 | if (verb && __ratelimit(&drbd_ratelimit_state)) | |
b411b363 PR |
2408 | dev_err(DEV, "Can not satisfy peer's read request, " |
2409 | "no local data.\n"); | |
b18b37be | 2410 | |
a821cc4a | 2411 | /* drain possibly payload */ |
e2857216 | 2412 | return drbd_drain_block(mdev, pi->size); |
b411b363 PR |
2413 | } |
2414 | ||
2415 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD | |
2416 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
2417 | * which in turn might block on the other node at this very place. */ | |
0db55363 | 2418 | peer_req = drbd_alloc_peer_req(mdev, p->block_id, sector, size, GFP_NOIO); |
db830c46 | 2419 | if (!peer_req) { |
b411b363 | 2420 | put_ldev(mdev); |
82bc0194 | 2421 | return -ENOMEM; |
b411b363 PR |
2422 | } |
2423 | ||
e2857216 | 2424 | switch (pi->cmd) { |
b411b363 | 2425 | case P_DATA_REQUEST: |
db830c46 | 2426 | peer_req->w.cb = w_e_end_data_req; |
b411b363 | 2427 | fault_type = DRBD_FAULT_DT_RD; |
80a40e43 LE |
2428 | /* application IO, don't drbd_rs_begin_io */ |
2429 | goto submit; | |
2430 | ||
b411b363 | 2431 | case P_RS_DATA_REQUEST: |
db830c46 | 2432 | peer_req->w.cb = w_e_end_rsdata_req; |
b411b363 | 2433 | fault_type = DRBD_FAULT_RS_RD; |
5f9915bb LE |
2434 | /* used in the sector offset progress display */ |
2435 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
b411b363 PR |
2436 | break; |
2437 | ||
2438 | case P_OV_REPLY: | |
2439 | case P_CSUM_RS_REQUEST: | |
2440 | fault_type = DRBD_FAULT_RS_RD; | |
e2857216 | 2441 | di = kmalloc(sizeof(*di) + pi->size, GFP_NOIO); |
b411b363 PR |
2442 | if (!di) |
2443 | goto out_free_e; | |
2444 | ||
e2857216 | 2445 | di->digest_size = pi->size; |
b411b363 PR |
2446 | di->digest = (((char *)di)+sizeof(struct digest_info)); |
2447 | ||
db830c46 AG |
2448 | peer_req->digest = di; |
2449 | peer_req->flags |= EE_HAS_DIGEST; | |
c36c3ced | 2450 | |
e2857216 | 2451 | if (drbd_recv_all(mdev->tconn, di->digest, pi->size)) |
b411b363 PR |
2452 | goto out_free_e; |
2453 | ||
e2857216 | 2454 | if (pi->cmd == P_CSUM_RS_REQUEST) { |
31890f4a | 2455 | D_ASSERT(mdev->tconn->agreed_pro_version >= 89); |
db830c46 | 2456 | peer_req->w.cb = w_e_end_csum_rs_req; |
5f9915bb LE |
2457 | /* used in the sector offset progress display */ |
2458 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
e2857216 | 2459 | } else if (pi->cmd == P_OV_REPLY) { |
2649f080 LE |
2460 | /* track progress, we may need to throttle */ |
2461 | atomic_add(size >> 9, &mdev->rs_sect_in); | |
db830c46 | 2462 | peer_req->w.cb = w_e_end_ov_reply; |
b411b363 | 2463 | dec_rs_pending(mdev); |
0f0601f4 LE |
2464 | /* drbd_rs_begin_io done when we sent this request, |
2465 | * but accounting still needs to be done. */ | |
2466 | goto submit_for_resync; | |
b411b363 PR |
2467 | } |
2468 | break; | |
2469 | ||
2470 | case P_OV_REQUEST: | |
b411b363 | 2471 | if (mdev->ov_start_sector == ~(sector_t)0 && |
31890f4a | 2472 | mdev->tconn->agreed_pro_version >= 90) { |
de228bba LE |
2473 | unsigned long now = jiffies; |
2474 | int i; | |
b411b363 PR |
2475 | mdev->ov_start_sector = sector; |
2476 | mdev->ov_position = sector; | |
30b743a2 LE |
2477 | mdev->ov_left = drbd_bm_bits(mdev) - BM_SECT_TO_BIT(sector); |
2478 | mdev->rs_total = mdev->ov_left; | |
de228bba LE |
2479 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
2480 | mdev->rs_mark_left[i] = mdev->ov_left; | |
2481 | mdev->rs_mark_time[i] = now; | |
2482 | } | |
b411b363 PR |
2483 | dev_info(DEV, "Online Verify start sector: %llu\n", |
2484 | (unsigned long long)sector); | |
2485 | } | |
db830c46 | 2486 | peer_req->w.cb = w_e_end_ov_req; |
b411b363 | 2487 | fault_type = DRBD_FAULT_RS_RD; |
b411b363 PR |
2488 | break; |
2489 | ||
b411b363 | 2490 | default: |
49ba9b1b | 2491 | BUG(); |
b411b363 PR |
2492 | } |
2493 | ||
0f0601f4 LE |
2494 | /* Throttle, drbd_rs_begin_io and submit should become asynchronous |
2495 | * wrt the receiver, but it is not as straightforward as it may seem. | |
2496 | * Various places in the resync start and stop logic assume resync | |
2497 | * requests are processed in order, requeuing this on the worker thread | |
2498 | * introduces a bunch of new code for synchronization between threads. | |
2499 | * | |
2500 | * Unlimited throttling before drbd_rs_begin_io may stall the resync | |
2501 | * "forever", throttling after drbd_rs_begin_io will lock that extent | |
2502 | * for application writes for the same time. For now, just throttle | |
2503 | * here, where the rest of the code expects the receiver to sleep for | |
2504 | * a while, anyways. | |
2505 | */ | |
2506 | ||
2507 | /* Throttle before drbd_rs_begin_io, as that locks out application IO; | |
2508 | * this defers syncer requests for some time, before letting at least | |
2509 | * on request through. The resync controller on the receiving side | |
2510 | * will adapt to the incoming rate accordingly. | |
2511 | * | |
2512 | * We cannot throttle here if remote is Primary/SyncTarget: | |
2513 | * we would also throttle its application reads. | |
2514 | * In that case, throttling is done on the SyncTarget only. | |
2515 | */ | |
e3555d85 PR |
2516 | if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev, sector)) |
2517 | schedule_timeout_uninterruptible(HZ/10); | |
2518 | if (drbd_rs_begin_io(mdev, sector)) | |
80a40e43 | 2519 | goto out_free_e; |
b411b363 | 2520 | |
0f0601f4 LE |
2521 | submit_for_resync: |
2522 | atomic_add(size >> 9, &mdev->rs_sect_ev); | |
2523 | ||
80a40e43 | 2524 | submit: |
b411b363 | 2525 | inc_unacked(mdev); |
87eeee41 | 2526 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 2527 | list_add_tail(&peer_req->w.list, &mdev->read_ee); |
87eeee41 | 2528 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 2529 | |
fbe29dec | 2530 | if (drbd_submit_peer_request(mdev, peer_req, READ, fault_type) == 0) |
82bc0194 | 2531 | return 0; |
b411b363 | 2532 | |
10f6d992 LE |
2533 | /* don't care for the reason here */ |
2534 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 2535 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 2536 | list_del(&peer_req->w.list); |
87eeee41 | 2537 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 LE |
2538 | /* no drbd_rs_complete_io(), we are dropping the connection anyways */ |
2539 | ||
b411b363 | 2540 | out_free_e: |
b411b363 | 2541 | put_ldev(mdev); |
3967deb1 | 2542 | drbd_free_peer_req(mdev, peer_req); |
82bc0194 | 2543 | return -EIO; |
b411b363 PR |
2544 | } |
2545 | ||
2546 | static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local) | |
2547 | { | |
2548 | int self, peer, rv = -100; | |
2549 | unsigned long ch_self, ch_peer; | |
44ed167d | 2550 | enum drbd_after_sb_p after_sb_0p; |
b411b363 PR |
2551 | |
2552 | self = mdev->ldev->md.uuid[UI_BITMAP] & 1; | |
2553 | peer = mdev->p_uuid[UI_BITMAP] & 1; | |
2554 | ||
2555 | ch_peer = mdev->p_uuid[UI_SIZE]; | |
2556 | ch_self = mdev->comm_bm_set; | |
2557 | ||
44ed167d PR |
2558 | rcu_read_lock(); |
2559 | after_sb_0p = rcu_dereference(mdev->tconn->net_conf)->after_sb_0p; | |
2560 | rcu_read_unlock(); | |
2561 | switch (after_sb_0p) { | |
b411b363 PR |
2562 | case ASB_CONSENSUS: |
2563 | case ASB_DISCARD_SECONDARY: | |
2564 | case ASB_CALL_HELPER: | |
44ed167d | 2565 | case ASB_VIOLENTLY: |
b411b363 PR |
2566 | dev_err(DEV, "Configuration error.\n"); |
2567 | break; | |
2568 | case ASB_DISCONNECT: | |
2569 | break; | |
2570 | case ASB_DISCARD_YOUNGER_PRI: | |
2571 | if (self == 0 && peer == 1) { | |
2572 | rv = -1; | |
2573 | break; | |
2574 | } | |
2575 | if (self == 1 && peer == 0) { | |
2576 | rv = 1; | |
2577 | break; | |
2578 | } | |
2579 | /* Else fall through to one of the other strategies... */ | |
2580 | case ASB_DISCARD_OLDER_PRI: | |
2581 | if (self == 0 && peer == 1) { | |
2582 | rv = 1; | |
2583 | break; | |
2584 | } | |
2585 | if (self == 1 && peer == 0) { | |
2586 | rv = -1; | |
2587 | break; | |
2588 | } | |
2589 | /* Else fall through to one of the other strategies... */ | |
ad19bf6e | 2590 | dev_warn(DEV, "Discard younger/older primary did not find a decision\n" |
b411b363 PR |
2591 | "Using discard-least-changes instead\n"); |
2592 | case ASB_DISCARD_ZERO_CHG: | |
2593 | if (ch_peer == 0 && ch_self == 0) { | |
427c0434 | 2594 | rv = test_bit(RESOLVE_CONFLICTS, &mdev->tconn->flags) |
b411b363 PR |
2595 | ? -1 : 1; |
2596 | break; | |
2597 | } else { | |
2598 | if (ch_peer == 0) { rv = 1; break; } | |
2599 | if (ch_self == 0) { rv = -1; break; } | |
2600 | } | |
44ed167d | 2601 | if (after_sb_0p == ASB_DISCARD_ZERO_CHG) |
b411b363 PR |
2602 | break; |
2603 | case ASB_DISCARD_LEAST_CHG: | |
2604 | if (ch_self < ch_peer) | |
2605 | rv = -1; | |
2606 | else if (ch_self > ch_peer) | |
2607 | rv = 1; | |
2608 | else /* ( ch_self == ch_peer ) */ | |
2609 | /* Well, then use something else. */ | |
427c0434 | 2610 | rv = test_bit(RESOLVE_CONFLICTS, &mdev->tconn->flags) |
b411b363 PR |
2611 | ? -1 : 1; |
2612 | break; | |
2613 | case ASB_DISCARD_LOCAL: | |
2614 | rv = -1; | |
2615 | break; | |
2616 | case ASB_DISCARD_REMOTE: | |
2617 | rv = 1; | |
2618 | } | |
2619 | ||
2620 | return rv; | |
2621 | } | |
2622 | ||
2623 | static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local) | |
2624 | { | |
6184ea21 | 2625 | int hg, rv = -100; |
44ed167d | 2626 | enum drbd_after_sb_p after_sb_1p; |
b411b363 | 2627 | |
44ed167d PR |
2628 | rcu_read_lock(); |
2629 | after_sb_1p = rcu_dereference(mdev->tconn->net_conf)->after_sb_1p; | |
2630 | rcu_read_unlock(); | |
2631 | switch (after_sb_1p) { | |
b411b363 PR |
2632 | case ASB_DISCARD_YOUNGER_PRI: |
2633 | case ASB_DISCARD_OLDER_PRI: | |
2634 | case ASB_DISCARD_LEAST_CHG: | |
2635 | case ASB_DISCARD_LOCAL: | |
2636 | case ASB_DISCARD_REMOTE: | |
44ed167d | 2637 | case ASB_DISCARD_ZERO_CHG: |
b411b363 PR |
2638 | dev_err(DEV, "Configuration error.\n"); |
2639 | break; | |
2640 | case ASB_DISCONNECT: | |
2641 | break; | |
2642 | case ASB_CONSENSUS: | |
2643 | hg = drbd_asb_recover_0p(mdev); | |
2644 | if (hg == -1 && mdev->state.role == R_SECONDARY) | |
2645 | rv = hg; | |
2646 | if (hg == 1 && mdev->state.role == R_PRIMARY) | |
2647 | rv = hg; | |
2648 | break; | |
2649 | case ASB_VIOLENTLY: | |
2650 | rv = drbd_asb_recover_0p(mdev); | |
2651 | break; | |
2652 | case ASB_DISCARD_SECONDARY: | |
2653 | return mdev->state.role == R_PRIMARY ? 1 : -1; | |
2654 | case ASB_CALL_HELPER: | |
2655 | hg = drbd_asb_recover_0p(mdev); | |
2656 | if (hg == -1 && mdev->state.role == R_PRIMARY) { | |
bb437946 AG |
2657 | enum drbd_state_rv rv2; |
2658 | ||
b411b363 PR |
2659 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2660 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2661 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2662 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2663 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2664 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2665 | } else { | |
2666 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2667 | rv = hg; | |
2668 | } | |
2669 | } else | |
2670 | rv = hg; | |
2671 | } | |
2672 | ||
2673 | return rv; | |
2674 | } | |
2675 | ||
2676 | static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local) | |
2677 | { | |
6184ea21 | 2678 | int hg, rv = -100; |
44ed167d | 2679 | enum drbd_after_sb_p after_sb_2p; |
b411b363 | 2680 | |
44ed167d PR |
2681 | rcu_read_lock(); |
2682 | after_sb_2p = rcu_dereference(mdev->tconn->net_conf)->after_sb_2p; | |
2683 | rcu_read_unlock(); | |
2684 | switch (after_sb_2p) { | |
b411b363 PR |
2685 | case ASB_DISCARD_YOUNGER_PRI: |
2686 | case ASB_DISCARD_OLDER_PRI: | |
2687 | case ASB_DISCARD_LEAST_CHG: | |
2688 | case ASB_DISCARD_LOCAL: | |
2689 | case ASB_DISCARD_REMOTE: | |
2690 | case ASB_CONSENSUS: | |
2691 | case ASB_DISCARD_SECONDARY: | |
44ed167d | 2692 | case ASB_DISCARD_ZERO_CHG: |
b411b363 PR |
2693 | dev_err(DEV, "Configuration error.\n"); |
2694 | break; | |
2695 | case ASB_VIOLENTLY: | |
2696 | rv = drbd_asb_recover_0p(mdev); | |
2697 | break; | |
2698 | case ASB_DISCONNECT: | |
2699 | break; | |
2700 | case ASB_CALL_HELPER: | |
2701 | hg = drbd_asb_recover_0p(mdev); | |
2702 | if (hg == -1) { | |
bb437946 AG |
2703 | enum drbd_state_rv rv2; |
2704 | ||
b411b363 PR |
2705 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2706 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2707 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2708 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2709 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2710 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2711 | } else { | |
2712 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2713 | rv = hg; | |
2714 | } | |
2715 | } else | |
2716 | rv = hg; | |
2717 | } | |
2718 | ||
2719 | return rv; | |
2720 | } | |
2721 | ||
2722 | static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid, | |
2723 | u64 bits, u64 flags) | |
2724 | { | |
2725 | if (!uuid) { | |
2726 | dev_info(DEV, "%s uuid info vanished while I was looking!\n", text); | |
2727 | return; | |
2728 | } | |
2729 | dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", | |
2730 | text, | |
2731 | (unsigned long long)uuid[UI_CURRENT], | |
2732 | (unsigned long long)uuid[UI_BITMAP], | |
2733 | (unsigned long long)uuid[UI_HISTORY_START], | |
2734 | (unsigned long long)uuid[UI_HISTORY_END], | |
2735 | (unsigned long long)bits, | |
2736 | (unsigned long long)flags); | |
2737 | } | |
2738 | ||
2739 | /* | |
2740 | 100 after split brain try auto recover | |
2741 | 2 C_SYNC_SOURCE set BitMap | |
2742 | 1 C_SYNC_SOURCE use BitMap | |
2743 | 0 no Sync | |
2744 | -1 C_SYNC_TARGET use BitMap | |
2745 | -2 C_SYNC_TARGET set BitMap | |
2746 | -100 after split brain, disconnect | |
2747 | -1000 unrelated data | |
4a23f264 PR |
2748 | -1091 requires proto 91 |
2749 | -1096 requires proto 96 | |
b411b363 PR |
2750 | */ |
2751 | static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local) | |
2752 | { | |
2753 | u64 self, peer; | |
2754 | int i, j; | |
2755 | ||
2756 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2757 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2758 | ||
2759 | *rule_nr = 10; | |
2760 | if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) | |
2761 | return 0; | |
2762 | ||
2763 | *rule_nr = 20; | |
2764 | if ((self == UUID_JUST_CREATED || self == (u64)0) && | |
2765 | peer != UUID_JUST_CREATED) | |
2766 | return -2; | |
2767 | ||
2768 | *rule_nr = 30; | |
2769 | if (self != UUID_JUST_CREATED && | |
2770 | (peer == UUID_JUST_CREATED || peer == (u64)0)) | |
2771 | return 2; | |
2772 | ||
2773 | if (self == peer) { | |
2774 | int rct, dc; /* roles at crash time */ | |
2775 | ||
2776 | if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) { | |
2777 | ||
31890f4a | 2778 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2779 | return -1091; |
b411b363 PR |
2780 | |
2781 | if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) && | |
2782 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { | |
2783 | dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n"); | |
9f2247bb PR |
2784 | drbd_uuid_move_history(mdev); |
2785 | mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[UI_BITMAP]; | |
2786 | mdev->ldev->md.uuid[UI_BITMAP] = 0; | |
b411b363 PR |
2787 | |
2788 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, | |
2789 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2790 | *rule_nr = 34; | |
2791 | } else { | |
2792 | dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n"); | |
2793 | *rule_nr = 36; | |
2794 | } | |
2795 | ||
2796 | return 1; | |
2797 | } | |
2798 | ||
2799 | if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) { | |
2800 | ||
31890f4a | 2801 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2802 | return -1091; |
b411b363 PR |
2803 | |
2804 | if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) && | |
2805 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) { | |
2806 | dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); | |
2807 | ||
2808 | mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START]; | |
2809 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP]; | |
2810 | mdev->p_uuid[UI_BITMAP] = 0UL; | |
2811 | ||
2812 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2813 | *rule_nr = 35; | |
2814 | } else { | |
2815 | dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n"); | |
2816 | *rule_nr = 37; | |
2817 | } | |
2818 | ||
2819 | return -1; | |
2820 | } | |
2821 | ||
2822 | /* Common power [off|failure] */ | |
2823 | rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) + | |
2824 | (mdev->p_uuid[UI_FLAGS] & 2); | |
2825 | /* lowest bit is set when we were primary, | |
2826 | * next bit (weight 2) is set when peer was primary */ | |
2827 | *rule_nr = 40; | |
2828 | ||
2829 | switch (rct) { | |
2830 | case 0: /* !self_pri && !peer_pri */ return 0; | |
2831 | case 1: /* self_pri && !peer_pri */ return 1; | |
2832 | case 2: /* !self_pri && peer_pri */ return -1; | |
2833 | case 3: /* self_pri && peer_pri */ | |
427c0434 | 2834 | dc = test_bit(RESOLVE_CONFLICTS, &mdev->tconn->flags); |
b411b363 PR |
2835 | return dc ? -1 : 1; |
2836 | } | |
2837 | } | |
2838 | ||
2839 | *rule_nr = 50; | |
2840 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2841 | if (self == peer) | |
2842 | return -1; | |
2843 | ||
2844 | *rule_nr = 51; | |
2845 | peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); | |
2846 | if (self == peer) { | |
31890f4a | 2847 | if (mdev->tconn->agreed_pro_version < 96 ? |
4a23f264 PR |
2848 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == |
2849 | (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) : | |
2850 | peer + UUID_NEW_BM_OFFSET == (mdev->p_uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2851 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2852 | resync as sync source modifications of the peer's UUIDs. */ | |
2853 | ||
31890f4a | 2854 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2855 | return -1091; |
b411b363 PR |
2856 | |
2857 | mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START]; | |
2858 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1]; | |
4a23f264 | 2859 | |
92b4ca29 | 2860 | dev_info(DEV, "Lost last syncUUID packet, corrected:\n"); |
4a23f264 PR |
2861 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); |
2862 | ||
b411b363 PR |
2863 | return -1; |
2864 | } | |
2865 | } | |
2866 | ||
2867 | *rule_nr = 60; | |
2868 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2869 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2870 | peer = mdev->p_uuid[i] & ~((u64)1); | |
2871 | if (self == peer) | |
2872 | return -2; | |
2873 | } | |
2874 | ||
2875 | *rule_nr = 70; | |
2876 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2877 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2878 | if (self == peer) | |
2879 | return 1; | |
2880 | ||
2881 | *rule_nr = 71; | |
2882 | self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); | |
2883 | if (self == peer) { | |
31890f4a | 2884 | if (mdev->tconn->agreed_pro_version < 96 ? |
4a23f264 PR |
2885 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == |
2886 | (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) : | |
2887 | self + UUID_NEW_BM_OFFSET == (mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2888 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2889 | resync as sync source modifications of our UUIDs. */ | |
2890 | ||
31890f4a | 2891 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2892 | return -1091; |
b411b363 | 2893 | |
9f2247bb PR |
2894 | __drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]); |
2895 | __drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]); | |
b411b363 | 2896 | |
4a23f264 | 2897 | dev_info(DEV, "Last syncUUID did not get through, corrected:\n"); |
b411b363 PR |
2898 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, |
2899 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2900 | ||
2901 | return 1; | |
2902 | } | |
2903 | } | |
2904 | ||
2905 | ||
2906 | *rule_nr = 80; | |
d8c2a36b | 2907 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); |
b411b363 PR |
2908 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
2909 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2910 | if (self == peer) | |
2911 | return 2; | |
2912 | } | |
2913 | ||
2914 | *rule_nr = 90; | |
2915 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2916 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2917 | if (self == peer && self != ((u64)0)) | |
2918 | return 100; | |
2919 | ||
2920 | *rule_nr = 100; | |
2921 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2922 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2923 | for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { | |
2924 | peer = mdev->p_uuid[j] & ~((u64)1); | |
2925 | if (self == peer) | |
2926 | return -100; | |
2927 | } | |
2928 | } | |
2929 | ||
2930 | return -1000; | |
2931 | } | |
2932 | ||
2933 | /* drbd_sync_handshake() returns the new conn state on success, or | |
2934 | CONN_MASK (-1) on failure. | |
2935 | */ | |
2936 | static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role, | |
2937 | enum drbd_disk_state peer_disk) __must_hold(local) | |
2938 | { | |
b411b363 PR |
2939 | enum drbd_conns rv = C_MASK; |
2940 | enum drbd_disk_state mydisk; | |
44ed167d | 2941 | struct net_conf *nc; |
6dff2902 | 2942 | int hg, rule_nr, rr_conflict, tentative; |
b411b363 PR |
2943 | |
2944 | mydisk = mdev->state.disk; | |
2945 | if (mydisk == D_NEGOTIATING) | |
2946 | mydisk = mdev->new_state_tmp.disk; | |
2947 | ||
2948 | dev_info(DEV, "drbd_sync_handshake:\n"); | |
9f2247bb PR |
2949 | |
2950 | spin_lock_irq(&mdev->ldev->md.uuid_lock); | |
b411b363 PR |
2951 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0); |
2952 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, | |
2953 | mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2954 | ||
2955 | hg = drbd_uuid_compare(mdev, &rule_nr); | |
9f2247bb | 2956 | spin_unlock_irq(&mdev->ldev->md.uuid_lock); |
b411b363 PR |
2957 | |
2958 | dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr); | |
2959 | ||
2960 | if (hg == -1000) { | |
2961 | dev_alert(DEV, "Unrelated data, aborting!\n"); | |
2962 | return C_MASK; | |
2963 | } | |
4a23f264 PR |
2964 | if (hg < -1000) { |
2965 | dev_alert(DEV, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000); | |
b411b363 PR |
2966 | return C_MASK; |
2967 | } | |
2968 | ||
2969 | if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || | |
2970 | (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { | |
2971 | int f = (hg == -100) || abs(hg) == 2; | |
2972 | hg = mydisk > D_INCONSISTENT ? 1 : -1; | |
2973 | if (f) | |
2974 | hg = hg*2; | |
2975 | dev_info(DEV, "Becoming sync %s due to disk states.\n", | |
2976 | hg > 0 ? "source" : "target"); | |
2977 | } | |
2978 | ||
3a11a487 AG |
2979 | if (abs(hg) == 100) |
2980 | drbd_khelper(mdev, "initial-split-brain"); | |
2981 | ||
44ed167d PR |
2982 | rcu_read_lock(); |
2983 | nc = rcu_dereference(mdev->tconn->net_conf); | |
2984 | ||
2985 | if (hg == 100 || (hg == -100 && nc->always_asbp)) { | |
b411b363 PR |
2986 | int pcount = (mdev->state.role == R_PRIMARY) |
2987 | + (peer_role == R_PRIMARY); | |
2988 | int forced = (hg == -100); | |
2989 | ||
2990 | switch (pcount) { | |
2991 | case 0: | |
2992 | hg = drbd_asb_recover_0p(mdev); | |
2993 | break; | |
2994 | case 1: | |
2995 | hg = drbd_asb_recover_1p(mdev); | |
2996 | break; | |
2997 | case 2: | |
2998 | hg = drbd_asb_recover_2p(mdev); | |
2999 | break; | |
3000 | } | |
3001 | if (abs(hg) < 100) { | |
3002 | dev_warn(DEV, "Split-Brain detected, %d primaries, " | |
3003 | "automatically solved. Sync from %s node\n", | |
3004 | pcount, (hg < 0) ? "peer" : "this"); | |
3005 | if (forced) { | |
3006 | dev_warn(DEV, "Doing a full sync, since" | |
3007 | " UUIDs where ambiguous.\n"); | |
3008 | hg = hg*2; | |
3009 | } | |
3010 | } | |
3011 | } | |
3012 | ||
3013 | if (hg == -100) { | |
08b165ba | 3014 | if (test_bit(DISCARD_MY_DATA, &mdev->flags) && !(mdev->p_uuid[UI_FLAGS]&1)) |
b411b363 | 3015 | hg = -1; |
08b165ba | 3016 | if (!test_bit(DISCARD_MY_DATA, &mdev->flags) && (mdev->p_uuid[UI_FLAGS]&1)) |
b411b363 PR |
3017 | hg = 1; |
3018 | ||
3019 | if (abs(hg) < 100) | |
3020 | dev_warn(DEV, "Split-Brain detected, manually solved. " | |
3021 | "Sync from %s node\n", | |
3022 | (hg < 0) ? "peer" : "this"); | |
3023 | } | |
44ed167d | 3024 | rr_conflict = nc->rr_conflict; |
6dff2902 | 3025 | tentative = nc->tentative; |
44ed167d | 3026 | rcu_read_unlock(); |
b411b363 PR |
3027 | |
3028 | if (hg == -100) { | |
580b9767 LE |
3029 | /* FIXME this log message is not correct if we end up here |
3030 | * after an attempted attach on a diskless node. | |
3031 | * We just refuse to attach -- well, we drop the "connection" | |
3032 | * to that disk, in a way... */ | |
3a11a487 | 3033 | dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n"); |
b411b363 PR |
3034 | drbd_khelper(mdev, "split-brain"); |
3035 | return C_MASK; | |
3036 | } | |
3037 | ||
3038 | if (hg > 0 && mydisk <= D_INCONSISTENT) { | |
3039 | dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n"); | |
3040 | return C_MASK; | |
3041 | } | |
3042 | ||
3043 | if (hg < 0 && /* by intention we do not use mydisk here. */ | |
3044 | mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) { | |
44ed167d | 3045 | switch (rr_conflict) { |
b411b363 PR |
3046 | case ASB_CALL_HELPER: |
3047 | drbd_khelper(mdev, "pri-lost"); | |
3048 | /* fall through */ | |
3049 | case ASB_DISCONNECT: | |
3050 | dev_err(DEV, "I shall become SyncTarget, but I am primary!\n"); | |
3051 | return C_MASK; | |
3052 | case ASB_VIOLENTLY: | |
3053 | dev_warn(DEV, "Becoming SyncTarget, violating the stable-data" | |
3054 | "assumption\n"); | |
3055 | } | |
3056 | } | |
3057 | ||
6dff2902 | 3058 | if (tentative || test_bit(CONN_DRY_RUN, &mdev->tconn->flags)) { |
cf14c2e9 PR |
3059 | if (hg == 0) |
3060 | dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n"); | |
3061 | else | |
3062 | dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.", | |
3063 | drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), | |
3064 | abs(hg) >= 2 ? "full" : "bit-map based"); | |
3065 | return C_MASK; | |
3066 | } | |
3067 | ||
b411b363 PR |
3068 | if (abs(hg) >= 2) { |
3069 | dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); | |
20ceb2b2 LE |
3070 | if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake", |
3071 | BM_LOCKED_SET_ALLOWED)) | |
b411b363 PR |
3072 | return C_MASK; |
3073 | } | |
3074 | ||
3075 | if (hg > 0) { /* become sync source. */ | |
3076 | rv = C_WF_BITMAP_S; | |
3077 | } else if (hg < 0) { /* become sync target */ | |
3078 | rv = C_WF_BITMAP_T; | |
3079 | } else { | |
3080 | rv = C_CONNECTED; | |
3081 | if (drbd_bm_total_weight(mdev)) { | |
3082 | dev_info(DEV, "No resync, but %lu bits in bitmap!\n", | |
3083 | drbd_bm_total_weight(mdev)); | |
3084 | } | |
3085 | } | |
3086 | ||
3087 | return rv; | |
3088 | } | |
3089 | ||
f179d76d | 3090 | static enum drbd_after_sb_p convert_after_sb(enum drbd_after_sb_p peer) |
b411b363 PR |
3091 | { |
3092 | /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ | |
f179d76d PR |
3093 | if (peer == ASB_DISCARD_REMOTE) |
3094 | return ASB_DISCARD_LOCAL; | |
b411b363 PR |
3095 | |
3096 | /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ | |
f179d76d PR |
3097 | if (peer == ASB_DISCARD_LOCAL) |
3098 | return ASB_DISCARD_REMOTE; | |
b411b363 PR |
3099 | |
3100 | /* everything else is valid if they are equal on both sides. */ | |
f179d76d | 3101 | return peer; |
b411b363 PR |
3102 | } |
3103 | ||
e2857216 | 3104 | static int receive_protocol(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3105 | { |
e658983a | 3106 | struct p_protocol *p = pi->data; |
036b17ea PR |
3107 | enum drbd_after_sb_p p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; |
3108 | int p_proto, p_discard_my_data, p_two_primaries, cf; | |
3109 | struct net_conf *nc, *old_net_conf, *new_net_conf = NULL; | |
3110 | char integrity_alg[SHARED_SECRET_MAX] = ""; | |
accdbcc5 | 3111 | struct crypto_hash *peer_integrity_tfm = NULL; |
7aca6c75 | 3112 | void *int_dig_in = NULL, *int_dig_vv = NULL; |
b411b363 | 3113 | |
b411b363 PR |
3114 | p_proto = be32_to_cpu(p->protocol); |
3115 | p_after_sb_0p = be32_to_cpu(p->after_sb_0p); | |
3116 | p_after_sb_1p = be32_to_cpu(p->after_sb_1p); | |
3117 | p_after_sb_2p = be32_to_cpu(p->after_sb_2p); | |
b411b363 | 3118 | p_two_primaries = be32_to_cpu(p->two_primaries); |
cf14c2e9 | 3119 | cf = be32_to_cpu(p->conn_flags); |
6139f60d | 3120 | p_discard_my_data = cf & CF_DISCARD_MY_DATA; |
cf14c2e9 | 3121 | |
86db0618 AG |
3122 | if (tconn->agreed_pro_version >= 87) { |
3123 | int err; | |
cf14c2e9 | 3124 | |
88104ca4 | 3125 | if (pi->size > sizeof(integrity_alg)) |
86db0618 | 3126 | return -EIO; |
88104ca4 | 3127 | err = drbd_recv_all(tconn, integrity_alg, pi->size); |
86db0618 AG |
3128 | if (err) |
3129 | return err; | |
036b17ea | 3130 | integrity_alg[SHARED_SECRET_MAX - 1] = 0; |
b411b363 PR |
3131 | } |
3132 | ||
7d4c782c | 3133 | if (pi->cmd != P_PROTOCOL_UPDATE) { |
fbc12f45 | 3134 | clear_bit(CONN_DRY_RUN, &tconn->flags); |
b411b363 | 3135 | |
fbc12f45 AG |
3136 | if (cf & CF_DRY_RUN) |
3137 | set_bit(CONN_DRY_RUN, &tconn->flags); | |
b411b363 | 3138 | |
fbc12f45 AG |
3139 | rcu_read_lock(); |
3140 | nc = rcu_dereference(tconn->net_conf); | |
b411b363 | 3141 | |
fbc12f45 | 3142 | if (p_proto != nc->wire_protocol) { |
d505d9be | 3143 | conn_err(tconn, "incompatible %s settings\n", "protocol"); |
fbc12f45 AG |
3144 | goto disconnect_rcu_unlock; |
3145 | } | |
b411b363 | 3146 | |
fbc12f45 | 3147 | if (convert_after_sb(p_after_sb_0p) != nc->after_sb_0p) { |
d505d9be | 3148 | conn_err(tconn, "incompatible %s settings\n", "after-sb-0pri"); |
fbc12f45 AG |
3149 | goto disconnect_rcu_unlock; |
3150 | } | |
b411b363 | 3151 | |
fbc12f45 | 3152 | if (convert_after_sb(p_after_sb_1p) != nc->after_sb_1p) { |
d505d9be | 3153 | conn_err(tconn, "incompatible %s settings\n", "after-sb-1pri"); |
fbc12f45 AG |
3154 | goto disconnect_rcu_unlock; |
3155 | } | |
b411b363 | 3156 | |
fbc12f45 | 3157 | if (convert_after_sb(p_after_sb_2p) != nc->after_sb_2p) { |
d505d9be | 3158 | conn_err(tconn, "incompatible %s settings\n", "after-sb-2pri"); |
fbc12f45 AG |
3159 | goto disconnect_rcu_unlock; |
3160 | } | |
b411b363 | 3161 | |
fbc12f45 | 3162 | if (p_discard_my_data && nc->discard_my_data) { |
d505d9be | 3163 | conn_err(tconn, "incompatible %s settings\n", "discard-my-data"); |
fbc12f45 AG |
3164 | goto disconnect_rcu_unlock; |
3165 | } | |
b411b363 | 3166 | |
fbc12f45 | 3167 | if (p_two_primaries != nc->two_primaries) { |
d505d9be | 3168 | conn_err(tconn, "incompatible %s settings\n", "allow-two-primaries"); |
fbc12f45 AG |
3169 | goto disconnect_rcu_unlock; |
3170 | } | |
b411b363 | 3171 | |
fbc12f45 | 3172 | if (strcmp(integrity_alg, nc->integrity_alg)) { |
d505d9be | 3173 | conn_err(tconn, "incompatible %s settings\n", "data-integrity-alg"); |
fbc12f45 AG |
3174 | goto disconnect_rcu_unlock; |
3175 | } | |
b411b363 | 3176 | |
fbc12f45 | 3177 | rcu_read_unlock(); |
b411b363 PR |
3178 | } |
3179 | ||
7d4c782c AG |
3180 | if (integrity_alg[0]) { |
3181 | int hash_size; | |
3182 | ||
3183 | /* | |
3184 | * We can only change the peer data integrity algorithm | |
3185 | * here. Changing our own data integrity algorithm | |
3186 | * requires that we send a P_PROTOCOL_UPDATE packet at | |
3187 | * the same time; otherwise, the peer has no way to | |
3188 | * tell between which packets the algorithm should | |
3189 | * change. | |
3190 | */ | |
b411b363 | 3191 | |
7d4c782c AG |
3192 | peer_integrity_tfm = crypto_alloc_hash(integrity_alg, 0, CRYPTO_ALG_ASYNC); |
3193 | if (!peer_integrity_tfm) { | |
3194 | conn_err(tconn, "peer data-integrity-alg %s not supported\n", | |
3195 | integrity_alg); | |
3196 | goto disconnect; | |
3197 | } | |
b411b363 | 3198 | |
7d4c782c AG |
3199 | hash_size = crypto_hash_digestsize(peer_integrity_tfm); |
3200 | int_dig_in = kmalloc(hash_size, GFP_KERNEL); | |
3201 | int_dig_vv = kmalloc(hash_size, GFP_KERNEL); | |
3202 | if (!(int_dig_in && int_dig_vv)) { | |
3203 | conn_err(tconn, "Allocation of buffers for data integrity checking failed\n"); | |
b411b363 PR |
3204 | goto disconnect; |
3205 | } | |
b411b363 PR |
3206 | } |
3207 | ||
7d4c782c AG |
3208 | new_net_conf = kmalloc(sizeof(struct net_conf), GFP_KERNEL); |
3209 | if (!new_net_conf) { | |
3210 | conn_err(tconn, "Allocation of new net_conf failed\n"); | |
3211 | goto disconnect; | |
3212 | } | |
3213 | ||
3214 | mutex_lock(&tconn->data.mutex); | |
3215 | mutex_lock(&tconn->conf_update); | |
3216 | old_net_conf = tconn->net_conf; | |
3217 | *new_net_conf = *old_net_conf; | |
3218 | ||
3219 | new_net_conf->wire_protocol = p_proto; | |
3220 | new_net_conf->after_sb_0p = convert_after_sb(p_after_sb_0p); | |
3221 | new_net_conf->after_sb_1p = convert_after_sb(p_after_sb_1p); | |
3222 | new_net_conf->after_sb_2p = convert_after_sb(p_after_sb_2p); | |
3223 | new_net_conf->two_primaries = p_two_primaries; | |
3224 | ||
3225 | rcu_assign_pointer(tconn->net_conf, new_net_conf); | |
3226 | mutex_unlock(&tconn->conf_update); | |
3227 | mutex_unlock(&tconn->data.mutex); | |
3228 | ||
3229 | crypto_free_hash(tconn->peer_integrity_tfm); | |
3230 | kfree(tconn->int_dig_in); | |
3231 | kfree(tconn->int_dig_vv); | |
3232 | tconn->peer_integrity_tfm = peer_integrity_tfm; | |
3233 | tconn->int_dig_in = int_dig_in; | |
3234 | tconn->int_dig_vv = int_dig_vv; | |
3235 | ||
3236 | if (strcmp(old_net_conf->integrity_alg, integrity_alg)) | |
3237 | conn_info(tconn, "peer data-integrity-alg: %s\n", | |
3238 | integrity_alg[0] ? integrity_alg : "(none)"); | |
3239 | ||
3240 | synchronize_rcu(); | |
3241 | kfree(old_net_conf); | |
82bc0194 | 3242 | return 0; |
b411b363 | 3243 | |
44ed167d PR |
3244 | disconnect_rcu_unlock: |
3245 | rcu_read_unlock(); | |
b411b363 | 3246 | disconnect: |
b792c35c | 3247 | crypto_free_hash(peer_integrity_tfm); |
036b17ea PR |
3248 | kfree(int_dig_in); |
3249 | kfree(int_dig_vv); | |
7204624c | 3250 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3251 | return -EIO; |
b411b363 PR |
3252 | } |
3253 | ||
3254 | /* helper function | |
3255 | * input: alg name, feature name | |
3256 | * return: NULL (alg name was "") | |
3257 | * ERR_PTR(error) if something goes wrong | |
3258 | * or the crypto hash ptr, if it worked out ok. */ | |
3259 | struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev, | |
3260 | const char *alg, const char *name) | |
3261 | { | |
3262 | struct crypto_hash *tfm; | |
3263 | ||
3264 | if (!alg[0]) | |
3265 | return NULL; | |
3266 | ||
3267 | tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); | |
3268 | if (IS_ERR(tfm)) { | |
3269 | dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n", | |
3270 | alg, name, PTR_ERR(tfm)); | |
3271 | return tfm; | |
3272 | } | |
b411b363 PR |
3273 | return tfm; |
3274 | } | |
3275 | ||
4a76b161 AG |
3276 | static int ignore_remaining_packet(struct drbd_tconn *tconn, struct packet_info *pi) |
3277 | { | |
3278 | void *buffer = tconn->data.rbuf; | |
3279 | int size = pi->size; | |
3280 | ||
3281 | while (size) { | |
3282 | int s = min_t(int, size, DRBD_SOCKET_BUFFER_SIZE); | |
3283 | s = drbd_recv(tconn, buffer, s); | |
3284 | if (s <= 0) { | |
3285 | if (s < 0) | |
3286 | return s; | |
3287 | break; | |
3288 | } | |
3289 | size -= s; | |
3290 | } | |
3291 | if (size) | |
3292 | return -EIO; | |
3293 | return 0; | |
3294 | } | |
3295 | ||
3296 | /* | |
3297 | * config_unknown_volume - device configuration command for unknown volume | |
3298 | * | |
3299 | * When a device is added to an existing connection, the node on which the | |
3300 | * device is added first will send configuration commands to its peer but the | |
3301 | * peer will not know about the device yet. It will warn and ignore these | |
3302 | * commands. Once the device is added on the second node, the second node will | |
3303 | * send the same device configuration commands, but in the other direction. | |
3304 | * | |
3305 | * (We can also end up here if drbd is misconfigured.) | |
3306 | */ | |
3307 | static int config_unknown_volume(struct drbd_tconn *tconn, struct packet_info *pi) | |
3308 | { | |
2fcb8f30 AG |
3309 | conn_warn(tconn, "%s packet received for volume %u, which is not configured locally\n", |
3310 | cmdname(pi->cmd), pi->vnr); | |
4a76b161 AG |
3311 | return ignore_remaining_packet(tconn, pi); |
3312 | } | |
3313 | ||
3314 | static int receive_SyncParam(struct drbd_tconn *tconn, struct packet_info *pi) | |
b411b363 | 3315 | { |
4a76b161 | 3316 | struct drbd_conf *mdev; |
e658983a | 3317 | struct p_rs_param_95 *p; |
b411b363 PR |
3318 | unsigned int header_size, data_size, exp_max_sz; |
3319 | struct crypto_hash *verify_tfm = NULL; | |
3320 | struct crypto_hash *csums_tfm = NULL; | |
2ec91e0e | 3321 | struct net_conf *old_net_conf, *new_net_conf = NULL; |
813472ce | 3322 | struct disk_conf *old_disk_conf = NULL, *new_disk_conf = NULL; |
4a76b161 | 3323 | const int apv = tconn->agreed_pro_version; |
813472ce | 3324 | struct fifo_buffer *old_plan = NULL, *new_plan = NULL; |
778f271d | 3325 | int fifo_size = 0; |
82bc0194 | 3326 | int err; |
b411b363 | 3327 | |
4a76b161 AG |
3328 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3329 | if (!mdev) | |
3330 | return config_unknown_volume(tconn, pi); | |
b411b363 PR |
3331 | |
3332 | exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) | |
3333 | : apv == 88 ? sizeof(struct p_rs_param) | |
3334 | + SHARED_SECRET_MAX | |
8e26f9cc PR |
3335 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
3336 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 3337 | |
e2857216 | 3338 | if (pi->size > exp_max_sz) { |
b411b363 | 3339 | dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n", |
e2857216 | 3340 | pi->size, exp_max_sz); |
82bc0194 | 3341 | return -EIO; |
b411b363 PR |
3342 | } |
3343 | ||
3344 | if (apv <= 88) { | |
e658983a | 3345 | header_size = sizeof(struct p_rs_param); |
e2857216 | 3346 | data_size = pi->size - header_size; |
8e26f9cc | 3347 | } else if (apv <= 94) { |
e658983a | 3348 | header_size = sizeof(struct p_rs_param_89); |
e2857216 | 3349 | data_size = pi->size - header_size; |
b411b363 | 3350 | D_ASSERT(data_size == 0); |
8e26f9cc | 3351 | } else { |
e658983a | 3352 | header_size = sizeof(struct p_rs_param_95); |
e2857216 | 3353 | data_size = pi->size - header_size; |
b411b363 PR |
3354 | D_ASSERT(data_size == 0); |
3355 | } | |
3356 | ||
3357 | /* initialize verify_alg and csums_alg */ | |
e658983a | 3358 | p = pi->data; |
b411b363 PR |
3359 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); |
3360 | ||
e658983a | 3361 | err = drbd_recv_all(mdev->tconn, p, header_size); |
82bc0194 AG |
3362 | if (err) |
3363 | return err; | |
b411b363 | 3364 | |
daeda1cc PR |
3365 | mutex_lock(&mdev->tconn->conf_update); |
3366 | old_net_conf = mdev->tconn->net_conf; | |
813472ce PR |
3367 | if (get_ldev(mdev)) { |
3368 | new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); | |
3369 | if (!new_disk_conf) { | |
3370 | put_ldev(mdev); | |
3371 | mutex_unlock(&mdev->tconn->conf_update); | |
3372 | dev_err(DEV, "Allocation of new disk_conf failed\n"); | |
3373 | return -ENOMEM; | |
3374 | } | |
daeda1cc | 3375 | |
813472ce PR |
3376 | old_disk_conf = mdev->ldev->disk_conf; |
3377 | *new_disk_conf = *old_disk_conf; | |
b411b363 | 3378 | |
6394b935 | 3379 | new_disk_conf->resync_rate = be32_to_cpu(p->resync_rate); |
813472ce | 3380 | } |
b411b363 PR |
3381 | |
3382 | if (apv >= 88) { | |
3383 | if (apv == 88) { | |
5de73827 PR |
3384 | if (data_size > SHARED_SECRET_MAX || data_size == 0) { |
3385 | dev_err(DEV, "verify-alg of wrong size, " | |
3386 | "peer wants %u, accepting only up to %u byte\n", | |
3387 | data_size, SHARED_SECRET_MAX); | |
813472ce PR |
3388 | err = -EIO; |
3389 | goto reconnect; | |
b411b363 PR |
3390 | } |
3391 | ||
82bc0194 | 3392 | err = drbd_recv_all(mdev->tconn, p->verify_alg, data_size); |
813472ce PR |
3393 | if (err) |
3394 | goto reconnect; | |
b411b363 PR |
3395 | /* we expect NUL terminated string */ |
3396 | /* but just in case someone tries to be evil */ | |
3397 | D_ASSERT(p->verify_alg[data_size-1] == 0); | |
3398 | p->verify_alg[data_size-1] = 0; | |
3399 | ||
3400 | } else /* apv >= 89 */ { | |
3401 | /* we still expect NUL terminated strings */ | |
3402 | /* but just in case someone tries to be evil */ | |
3403 | D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0); | |
3404 | D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0); | |
3405 | p->verify_alg[SHARED_SECRET_MAX-1] = 0; | |
3406 | p->csums_alg[SHARED_SECRET_MAX-1] = 0; | |
3407 | } | |
3408 | ||
2ec91e0e | 3409 | if (strcmp(old_net_conf->verify_alg, p->verify_alg)) { |
b411b363 PR |
3410 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { |
3411 | dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", | |
2ec91e0e | 3412 | old_net_conf->verify_alg, p->verify_alg); |
b411b363 PR |
3413 | goto disconnect; |
3414 | } | |
3415 | verify_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
3416 | p->verify_alg, "verify-alg"); | |
3417 | if (IS_ERR(verify_tfm)) { | |
3418 | verify_tfm = NULL; | |
3419 | goto disconnect; | |
3420 | } | |
3421 | } | |
3422 | ||
2ec91e0e | 3423 | if (apv >= 89 && strcmp(old_net_conf->csums_alg, p->csums_alg)) { |
b411b363 PR |
3424 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { |
3425 | dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", | |
2ec91e0e | 3426 | old_net_conf->csums_alg, p->csums_alg); |
b411b363 PR |
3427 | goto disconnect; |
3428 | } | |
3429 | csums_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
3430 | p->csums_alg, "csums-alg"); | |
3431 | if (IS_ERR(csums_tfm)) { | |
3432 | csums_tfm = NULL; | |
3433 | goto disconnect; | |
3434 | } | |
3435 | } | |
3436 | ||
813472ce | 3437 | if (apv > 94 && new_disk_conf) { |
daeda1cc PR |
3438 | new_disk_conf->c_plan_ahead = be32_to_cpu(p->c_plan_ahead); |
3439 | new_disk_conf->c_delay_target = be32_to_cpu(p->c_delay_target); | |
3440 | new_disk_conf->c_fill_target = be32_to_cpu(p->c_fill_target); | |
3441 | new_disk_conf->c_max_rate = be32_to_cpu(p->c_max_rate); | |
778f271d | 3442 | |
daeda1cc | 3443 | fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ; |
9958c857 | 3444 | if (fifo_size != mdev->rs_plan_s->size) { |
813472ce PR |
3445 | new_plan = fifo_alloc(fifo_size); |
3446 | if (!new_plan) { | |
778f271d | 3447 | dev_err(DEV, "kmalloc of fifo_buffer failed"); |
f399002e | 3448 | put_ldev(mdev); |
778f271d PR |
3449 | goto disconnect; |
3450 | } | |
3451 | } | |
8e26f9cc | 3452 | } |
b411b363 | 3453 | |
91fd4dad | 3454 | if (verify_tfm || csums_tfm) { |
2ec91e0e PR |
3455 | new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL); |
3456 | if (!new_net_conf) { | |
91fd4dad PR |
3457 | dev_err(DEV, "Allocation of new net_conf failed\n"); |
3458 | goto disconnect; | |
3459 | } | |
3460 | ||
2ec91e0e | 3461 | *new_net_conf = *old_net_conf; |
91fd4dad PR |
3462 | |
3463 | if (verify_tfm) { | |
2ec91e0e PR |
3464 | strcpy(new_net_conf->verify_alg, p->verify_alg); |
3465 | new_net_conf->verify_alg_len = strlen(p->verify_alg) + 1; | |
91fd4dad PR |
3466 | crypto_free_hash(mdev->tconn->verify_tfm); |
3467 | mdev->tconn->verify_tfm = verify_tfm; | |
3468 | dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg); | |
3469 | } | |
3470 | if (csums_tfm) { | |
2ec91e0e PR |
3471 | strcpy(new_net_conf->csums_alg, p->csums_alg); |
3472 | new_net_conf->csums_alg_len = strlen(p->csums_alg) + 1; | |
91fd4dad PR |
3473 | crypto_free_hash(mdev->tconn->csums_tfm); |
3474 | mdev->tconn->csums_tfm = csums_tfm; | |
3475 | dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg); | |
3476 | } | |
2ec91e0e | 3477 | rcu_assign_pointer(tconn->net_conf, new_net_conf); |
778f271d | 3478 | } |
b411b363 PR |
3479 | } |
3480 | ||
813472ce PR |
3481 | if (new_disk_conf) { |
3482 | rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf); | |
3483 | put_ldev(mdev); | |
3484 | } | |
3485 | ||
3486 | if (new_plan) { | |
3487 | old_plan = mdev->rs_plan_s; | |
3488 | rcu_assign_pointer(mdev->rs_plan_s, new_plan); | |
b411b363 | 3489 | } |
daeda1cc PR |
3490 | |
3491 | mutex_unlock(&mdev->tconn->conf_update); | |
3492 | synchronize_rcu(); | |
3493 | if (new_net_conf) | |
3494 | kfree(old_net_conf); | |
3495 | kfree(old_disk_conf); | |
813472ce | 3496 | kfree(old_plan); |
daeda1cc | 3497 | |
82bc0194 | 3498 | return 0; |
b411b363 | 3499 | |
813472ce PR |
3500 | reconnect: |
3501 | if (new_disk_conf) { | |
3502 | put_ldev(mdev); | |
3503 | kfree(new_disk_conf); | |
3504 | } | |
3505 | mutex_unlock(&mdev->tconn->conf_update); | |
3506 | return -EIO; | |
3507 | ||
b411b363 | 3508 | disconnect: |
813472ce PR |
3509 | kfree(new_plan); |
3510 | if (new_disk_conf) { | |
3511 | put_ldev(mdev); | |
3512 | kfree(new_disk_conf); | |
3513 | } | |
a0095508 | 3514 | mutex_unlock(&mdev->tconn->conf_update); |
b411b363 PR |
3515 | /* just for completeness: actually not needed, |
3516 | * as this is not reached if csums_tfm was ok. */ | |
3517 | crypto_free_hash(csums_tfm); | |
3518 | /* but free the verify_tfm again, if csums_tfm did not work out */ | |
3519 | crypto_free_hash(verify_tfm); | |
38fa9988 | 3520 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3521 | return -EIO; |
b411b363 PR |
3522 | } |
3523 | ||
b411b363 PR |
3524 | /* warn if the arguments differ by more than 12.5% */ |
3525 | static void warn_if_differ_considerably(struct drbd_conf *mdev, | |
3526 | const char *s, sector_t a, sector_t b) | |
3527 | { | |
3528 | sector_t d; | |
3529 | if (a == 0 || b == 0) | |
3530 | return; | |
3531 | d = (a > b) ? (a - b) : (b - a); | |
3532 | if (d > (a>>3) || d > (b>>3)) | |
3533 | dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s, | |
3534 | (unsigned long long)a, (unsigned long long)b); | |
3535 | } | |
3536 | ||
4a76b161 | 3537 | static int receive_sizes(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3538 | { |
4a76b161 | 3539 | struct drbd_conf *mdev; |
e658983a | 3540 | struct p_sizes *p = pi->data; |
e96c9633 | 3541 | enum determine_dev_size dd = DS_UNCHANGED; |
b411b363 PR |
3542 | sector_t p_size, p_usize, my_usize; |
3543 | int ldsc = 0; /* local disk size changed */ | |
e89b591c | 3544 | enum dds_flags ddsf; |
b411b363 | 3545 | |
4a76b161 AG |
3546 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3547 | if (!mdev) | |
3548 | return config_unknown_volume(tconn, pi); | |
3549 | ||
b411b363 PR |
3550 | p_size = be64_to_cpu(p->d_size); |
3551 | p_usize = be64_to_cpu(p->u_size); | |
3552 | ||
b411b363 PR |
3553 | /* just store the peer's disk size for now. |
3554 | * we still need to figure out whether we accept that. */ | |
3555 | mdev->p_size = p_size; | |
3556 | ||
b411b363 | 3557 | if (get_ldev(mdev)) { |
daeda1cc PR |
3558 | rcu_read_lock(); |
3559 | my_usize = rcu_dereference(mdev->ldev->disk_conf)->disk_size; | |
3560 | rcu_read_unlock(); | |
3561 | ||
b411b363 PR |
3562 | warn_if_differ_considerably(mdev, "lower level device sizes", |
3563 | p_size, drbd_get_max_capacity(mdev->ldev)); | |
3564 | warn_if_differ_considerably(mdev, "user requested size", | |
daeda1cc | 3565 | p_usize, my_usize); |
b411b363 PR |
3566 | |
3567 | /* if this is the first connect, or an otherwise expected | |
3568 | * param exchange, choose the minimum */ | |
3569 | if (mdev->state.conn == C_WF_REPORT_PARAMS) | |
daeda1cc | 3570 | p_usize = min_not_zero(my_usize, p_usize); |
b411b363 PR |
3571 | |
3572 | /* Never shrink a device with usable data during connect. | |
3573 | But allow online shrinking if we are connected. */ | |
ef5e44a6 | 3574 | if (drbd_new_dev_size(mdev, mdev->ldev, p_usize, 0) < |
daeda1cc PR |
3575 | drbd_get_capacity(mdev->this_bdev) && |
3576 | mdev->state.disk >= D_OUTDATED && | |
3577 | mdev->state.conn < C_CONNECTED) { | |
b411b363 | 3578 | dev_err(DEV, "The peer's disk size is too small!\n"); |
38fa9988 | 3579 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 3580 | put_ldev(mdev); |
82bc0194 | 3581 | return -EIO; |
b411b363 | 3582 | } |
daeda1cc PR |
3583 | |
3584 | if (my_usize != p_usize) { | |
3585 | struct disk_conf *old_disk_conf, *new_disk_conf = NULL; | |
3586 | ||
3587 | new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); | |
3588 | if (!new_disk_conf) { | |
3589 | dev_err(DEV, "Allocation of new disk_conf failed\n"); | |
3590 | put_ldev(mdev); | |
3591 | return -ENOMEM; | |
3592 | } | |
3593 | ||
3594 | mutex_lock(&mdev->tconn->conf_update); | |
3595 | old_disk_conf = mdev->ldev->disk_conf; | |
3596 | *new_disk_conf = *old_disk_conf; | |
3597 | new_disk_conf->disk_size = p_usize; | |
3598 | ||
3599 | rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf); | |
3600 | mutex_unlock(&mdev->tconn->conf_update); | |
3601 | synchronize_rcu(); | |
3602 | kfree(old_disk_conf); | |
3603 | ||
3604 | dev_info(DEV, "Peer sets u_size to %lu sectors\n", | |
3605 | (unsigned long)my_usize); | |
b411b363 | 3606 | } |
daeda1cc | 3607 | |
b411b363 PR |
3608 | put_ldev(mdev); |
3609 | } | |
b411b363 | 3610 | |
e89b591c | 3611 | ddsf = be16_to_cpu(p->dds_flags); |
b411b363 | 3612 | if (get_ldev(mdev)) { |
d752b269 | 3613 | dd = drbd_determine_dev_size(mdev, ddsf, NULL); |
b411b363 | 3614 | put_ldev(mdev); |
e96c9633 | 3615 | if (dd == DS_ERROR) |
82bc0194 | 3616 | return -EIO; |
b411b363 PR |
3617 | drbd_md_sync(mdev); |
3618 | } else { | |
3619 | /* I am diskless, need to accept the peer's size. */ | |
3620 | drbd_set_my_capacity(mdev, p_size); | |
3621 | } | |
3622 | ||
99432fcc PR |
3623 | mdev->peer_max_bio_size = be32_to_cpu(p->max_bio_size); |
3624 | drbd_reconsider_max_bio_size(mdev); | |
3625 | ||
b411b363 PR |
3626 | if (get_ldev(mdev)) { |
3627 | if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) { | |
3628 | mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); | |
3629 | ldsc = 1; | |
3630 | } | |
3631 | ||
b411b363 PR |
3632 | put_ldev(mdev); |
3633 | } | |
3634 | ||
3635 | if (mdev->state.conn > C_WF_REPORT_PARAMS) { | |
3636 | if (be64_to_cpu(p->c_size) != | |
3637 | drbd_get_capacity(mdev->this_bdev) || ldsc) { | |
3638 | /* we have different sizes, probably peer | |
3639 | * needs to know my new size... */ | |
e89b591c | 3640 | drbd_send_sizes(mdev, 0, ddsf); |
b411b363 PR |
3641 | } |
3642 | if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) || | |
e96c9633 | 3643 | (dd == DS_GREW && mdev->state.conn == C_CONNECTED)) { |
b411b363 | 3644 | if (mdev->state.pdsk >= D_INCONSISTENT && |
e89b591c PR |
3645 | mdev->state.disk >= D_INCONSISTENT) { |
3646 | if (ddsf & DDSF_NO_RESYNC) | |
3647 | dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n"); | |
3648 | else | |
3649 | resync_after_online_grow(mdev); | |
3650 | } else | |
b411b363 PR |
3651 | set_bit(RESYNC_AFTER_NEG, &mdev->flags); |
3652 | } | |
3653 | } | |
3654 | ||
82bc0194 | 3655 | return 0; |
b411b363 PR |
3656 | } |
3657 | ||
4a76b161 | 3658 | static int receive_uuids(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3659 | { |
4a76b161 | 3660 | struct drbd_conf *mdev; |
e658983a | 3661 | struct p_uuids *p = pi->data; |
b411b363 | 3662 | u64 *p_uuid; |
62b0da3a | 3663 | int i, updated_uuids = 0; |
b411b363 | 3664 | |
4a76b161 AG |
3665 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3666 | if (!mdev) | |
3667 | return config_unknown_volume(tconn, pi); | |
3668 | ||
b411b363 | 3669 | p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); |
063eacf8 JW |
3670 | if (!p_uuid) { |
3671 | dev_err(DEV, "kmalloc of p_uuid failed\n"); | |
3672 | return false; | |
3673 | } | |
b411b363 PR |
3674 | |
3675 | for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) | |
3676 | p_uuid[i] = be64_to_cpu(p->uuid[i]); | |
3677 | ||
3678 | kfree(mdev->p_uuid); | |
3679 | mdev->p_uuid = p_uuid; | |
3680 | ||
3681 | if (mdev->state.conn < C_CONNECTED && | |
3682 | mdev->state.disk < D_INCONSISTENT && | |
3683 | mdev->state.role == R_PRIMARY && | |
3684 | (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { | |
3685 | dev_err(DEV, "Can only connect to data with current UUID=%016llX\n", | |
3686 | (unsigned long long)mdev->ed_uuid); | |
38fa9988 | 3687 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3688 | return -EIO; |
b411b363 PR |
3689 | } |
3690 | ||
3691 | if (get_ldev(mdev)) { | |
3692 | int skip_initial_sync = | |
3693 | mdev->state.conn == C_CONNECTED && | |
31890f4a | 3694 | mdev->tconn->agreed_pro_version >= 90 && |
b411b363 PR |
3695 | mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && |
3696 | (p_uuid[UI_FLAGS] & 8); | |
3697 | if (skip_initial_sync) { | |
3698 | dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n"); | |
3699 | drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, | |
20ceb2b2 LE |
3700 | "clear_n_write from receive_uuids", |
3701 | BM_LOCKED_TEST_ALLOWED); | |
b411b363 PR |
3702 | _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]); |
3703 | _drbd_uuid_set(mdev, UI_BITMAP, 0); | |
3704 | _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), | |
3705 | CS_VERBOSE, NULL); | |
3706 | drbd_md_sync(mdev); | |
62b0da3a | 3707 | updated_uuids = 1; |
b411b363 PR |
3708 | } |
3709 | put_ldev(mdev); | |
18a50fa2 PR |
3710 | } else if (mdev->state.disk < D_INCONSISTENT && |
3711 | mdev->state.role == R_PRIMARY) { | |
3712 | /* I am a diskless primary, the peer just created a new current UUID | |
3713 | for me. */ | |
62b0da3a | 3714 | updated_uuids = drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
b411b363 PR |
3715 | } |
3716 | ||
3717 | /* Before we test for the disk state, we should wait until an eventually | |
3718 | ongoing cluster wide state change is finished. That is important if | |
3719 | we are primary and are detaching from our disk. We need to see the | |
3720 | new disk state... */ | |
8410da8f PR |
3721 | mutex_lock(mdev->state_mutex); |
3722 | mutex_unlock(mdev->state_mutex); | |
b411b363 | 3723 | if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT) |
62b0da3a LE |
3724 | updated_uuids |= drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
3725 | ||
3726 | if (updated_uuids) | |
3727 | drbd_print_uuids(mdev, "receiver updated UUIDs to"); | |
b411b363 | 3728 | |
82bc0194 | 3729 | return 0; |
b411b363 PR |
3730 | } |
3731 | ||
3732 | /** | |
3733 | * convert_state() - Converts the peer's view of the cluster state to our point of view | |
3734 | * @ps: The state as seen by the peer. | |
3735 | */ | |
3736 | static union drbd_state convert_state(union drbd_state ps) | |
3737 | { | |
3738 | union drbd_state ms; | |
3739 | ||
3740 | static enum drbd_conns c_tab[] = { | |
369bea63 | 3741 | [C_WF_REPORT_PARAMS] = C_WF_REPORT_PARAMS, |
b411b363 PR |
3742 | [C_CONNECTED] = C_CONNECTED, |
3743 | ||
3744 | [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, | |
3745 | [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, | |
3746 | [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ | |
3747 | [C_VERIFY_S] = C_VERIFY_T, | |
3748 | [C_MASK] = C_MASK, | |
3749 | }; | |
3750 | ||
3751 | ms.i = ps.i; | |
3752 | ||
3753 | ms.conn = c_tab[ps.conn]; | |
3754 | ms.peer = ps.role; | |
3755 | ms.role = ps.peer; | |
3756 | ms.pdsk = ps.disk; | |
3757 | ms.disk = ps.pdsk; | |
3758 | ms.peer_isp = (ps.aftr_isp | ps.user_isp); | |
3759 | ||
3760 | return ms; | |
3761 | } | |
3762 | ||
4a76b161 | 3763 | static int receive_req_state(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3764 | { |
4a76b161 | 3765 | struct drbd_conf *mdev; |
e658983a | 3766 | struct p_req_state *p = pi->data; |
b411b363 | 3767 | union drbd_state mask, val; |
bf885f8a | 3768 | enum drbd_state_rv rv; |
b411b363 | 3769 | |
4a76b161 AG |
3770 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3771 | if (!mdev) | |
3772 | return -EIO; | |
3773 | ||
b411b363 PR |
3774 | mask.i = be32_to_cpu(p->mask); |
3775 | val.i = be32_to_cpu(p->val); | |
3776 | ||
427c0434 | 3777 | if (test_bit(RESOLVE_CONFLICTS, &mdev->tconn->flags) && |
8410da8f | 3778 | mutex_is_locked(mdev->state_mutex)) { |
b411b363 | 3779 | drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG); |
82bc0194 | 3780 | return 0; |
b411b363 PR |
3781 | } |
3782 | ||
3783 | mask = convert_state(mask); | |
3784 | val = convert_state(val); | |
3785 | ||
dfafcc8a PR |
3786 | rv = drbd_change_state(mdev, CS_VERBOSE, mask, val); |
3787 | drbd_send_sr_reply(mdev, rv); | |
b411b363 | 3788 | |
b411b363 PR |
3789 | drbd_md_sync(mdev); |
3790 | ||
82bc0194 | 3791 | return 0; |
b411b363 PR |
3792 | } |
3793 | ||
e2857216 | 3794 | static int receive_req_conn_state(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3795 | { |
e658983a | 3796 | struct p_req_state *p = pi->data; |
b411b363 | 3797 | union drbd_state mask, val; |
bf885f8a | 3798 | enum drbd_state_rv rv; |
b411b363 | 3799 | |
b411b363 PR |
3800 | mask.i = be32_to_cpu(p->mask); |
3801 | val.i = be32_to_cpu(p->val); | |
3802 | ||
427c0434 | 3803 | if (test_bit(RESOLVE_CONFLICTS, &tconn->flags) && |
dfafcc8a PR |
3804 | mutex_is_locked(&tconn->cstate_mutex)) { |
3805 | conn_send_sr_reply(tconn, SS_CONCURRENT_ST_CHG); | |
82bc0194 | 3806 | return 0; |
b411b363 PR |
3807 | } |
3808 | ||
3809 | mask = convert_state(mask); | |
3810 | val = convert_state(val); | |
3811 | ||
778bcf2e | 3812 | rv = conn_request_state(tconn, mask, val, CS_VERBOSE | CS_LOCAL_ONLY | CS_IGN_OUTD_FAIL); |
dfafcc8a | 3813 | conn_send_sr_reply(tconn, rv); |
b411b363 | 3814 | |
82bc0194 | 3815 | return 0; |
b411b363 PR |
3816 | } |
3817 | ||
4a76b161 | 3818 | static int receive_state(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3819 | { |
4a76b161 | 3820 | struct drbd_conf *mdev; |
e658983a | 3821 | struct p_state *p = pi->data; |
4ac4aada | 3822 | union drbd_state os, ns, peer_state; |
b411b363 | 3823 | enum drbd_disk_state real_peer_disk; |
65d922c3 | 3824 | enum chg_state_flags cs_flags; |
b411b363 PR |
3825 | int rv; |
3826 | ||
4a76b161 AG |
3827 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3828 | if (!mdev) | |
3829 | return config_unknown_volume(tconn, pi); | |
3830 | ||
b411b363 PR |
3831 | peer_state.i = be32_to_cpu(p->state); |
3832 | ||
3833 | real_peer_disk = peer_state.disk; | |
3834 | if (peer_state.disk == D_NEGOTIATING) { | |
3835 | real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; | |
3836 | dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); | |
3837 | } | |
3838 | ||
87eeee41 | 3839 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 3840 | retry: |
78bae59b | 3841 | os = ns = drbd_read_state(mdev); |
87eeee41 | 3842 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 3843 | |
545752d5 LE |
3844 | /* If some other part of the code (asender thread, timeout) |
3845 | * already decided to close the connection again, | |
3846 | * we must not "re-establish" it here. */ | |
3847 | if (os.conn <= C_TEAR_DOWN) | |
58ffa580 | 3848 | return -ECONNRESET; |
545752d5 | 3849 | |
40424e4a LE |
3850 | /* If this is the "end of sync" confirmation, usually the peer disk |
3851 | * transitions from D_INCONSISTENT to D_UP_TO_DATE. For empty (0 bits | |
3852 | * set) resync started in PausedSyncT, or if the timing of pause-/ | |
3853 | * unpause-sync events has been "just right", the peer disk may | |
3854 | * transition from D_CONSISTENT to D_UP_TO_DATE as well. | |
3855 | */ | |
3856 | if ((os.pdsk == D_INCONSISTENT || os.pdsk == D_CONSISTENT) && | |
3857 | real_peer_disk == D_UP_TO_DATE && | |
e9ef7bb6 LE |
3858 | os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { |
3859 | /* If we are (becoming) SyncSource, but peer is still in sync | |
3860 | * preparation, ignore its uptodate-ness to avoid flapping, it | |
3861 | * will change to inconsistent once the peer reaches active | |
3862 | * syncing states. | |
3863 | * It may have changed syncer-paused flags, however, so we | |
3864 | * cannot ignore this completely. */ | |
3865 | if (peer_state.conn > C_CONNECTED && | |
3866 | peer_state.conn < C_SYNC_SOURCE) | |
3867 | real_peer_disk = D_INCONSISTENT; | |
3868 | ||
3869 | /* if peer_state changes to connected at the same time, | |
3870 | * it explicitly notifies us that it finished resync. | |
3871 | * Maybe we should finish it up, too? */ | |
3872 | else if (os.conn >= C_SYNC_SOURCE && | |
3873 | peer_state.conn == C_CONNECTED) { | |
3874 | if (drbd_bm_total_weight(mdev) <= mdev->rs_failed) | |
3875 | drbd_resync_finished(mdev); | |
82bc0194 | 3876 | return 0; |
e9ef7bb6 LE |
3877 | } |
3878 | } | |
3879 | ||
02b91b55 LE |
3880 | /* explicit verify finished notification, stop sector reached. */ |
3881 | if (os.conn == C_VERIFY_T && os.disk == D_UP_TO_DATE && | |
3882 | peer_state.conn == C_CONNECTED && real_peer_disk == D_UP_TO_DATE) { | |
58ffa580 | 3883 | ov_out_of_sync_print(mdev); |
02b91b55 | 3884 | drbd_resync_finished(mdev); |
58ffa580 | 3885 | return 0; |
02b91b55 LE |
3886 | } |
3887 | ||
e9ef7bb6 LE |
3888 | /* peer says his disk is inconsistent, while we think it is uptodate, |
3889 | * and this happens while the peer still thinks we have a sync going on, | |
3890 | * but we think we are already done with the sync. | |
3891 | * We ignore this to avoid flapping pdsk. | |
3892 | * This should not happen, if the peer is a recent version of drbd. */ | |
3893 | if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && | |
3894 | os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) | |
3895 | real_peer_disk = D_UP_TO_DATE; | |
3896 | ||
4ac4aada LE |
3897 | if (ns.conn == C_WF_REPORT_PARAMS) |
3898 | ns.conn = C_CONNECTED; | |
b411b363 | 3899 | |
67531718 PR |
3900 | if (peer_state.conn == C_AHEAD) |
3901 | ns.conn = C_BEHIND; | |
3902 | ||
b411b363 PR |
3903 | if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING && |
3904 | get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3905 | int cr; /* consider resync */ | |
3906 | ||
3907 | /* if we established a new connection */ | |
4ac4aada | 3908 | cr = (os.conn < C_CONNECTED); |
b411b363 PR |
3909 | /* if we had an established connection |
3910 | * and one of the nodes newly attaches a disk */ | |
4ac4aada | 3911 | cr |= (os.conn == C_CONNECTED && |
b411b363 | 3912 | (peer_state.disk == D_NEGOTIATING || |
4ac4aada | 3913 | os.disk == D_NEGOTIATING)); |
b411b363 PR |
3914 | /* if we have both been inconsistent, and the peer has been |
3915 | * forced to be UpToDate with --overwrite-data */ | |
3916 | cr |= test_bit(CONSIDER_RESYNC, &mdev->flags); | |
3917 | /* if we had been plain connected, and the admin requested to | |
3918 | * start a sync by "invalidate" or "invalidate-remote" */ | |
4ac4aada | 3919 | cr |= (os.conn == C_CONNECTED && |
b411b363 PR |
3920 | (peer_state.conn >= C_STARTING_SYNC_S && |
3921 | peer_state.conn <= C_WF_BITMAP_T)); | |
3922 | ||
3923 | if (cr) | |
4ac4aada | 3924 | ns.conn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk); |
b411b363 PR |
3925 | |
3926 | put_ldev(mdev); | |
4ac4aada LE |
3927 | if (ns.conn == C_MASK) { |
3928 | ns.conn = C_CONNECTED; | |
b411b363 | 3929 | if (mdev->state.disk == D_NEGOTIATING) { |
82f59cc6 | 3930 | drbd_force_state(mdev, NS(disk, D_FAILED)); |
b411b363 PR |
3931 | } else if (peer_state.disk == D_NEGOTIATING) { |
3932 | dev_err(DEV, "Disk attach process on the peer node was aborted.\n"); | |
3933 | peer_state.disk = D_DISKLESS; | |
580b9767 | 3934 | real_peer_disk = D_DISKLESS; |
b411b363 | 3935 | } else { |
8169e41b | 3936 | if (test_and_clear_bit(CONN_DRY_RUN, &mdev->tconn->flags)) |
82bc0194 | 3937 | return -EIO; |
4ac4aada | 3938 | D_ASSERT(os.conn == C_WF_REPORT_PARAMS); |
38fa9988 | 3939 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3940 | return -EIO; |
b411b363 PR |
3941 | } |
3942 | } | |
3943 | } | |
3944 | ||
87eeee41 | 3945 | spin_lock_irq(&mdev->tconn->req_lock); |
78bae59b | 3946 | if (os.i != drbd_read_state(mdev).i) |
b411b363 PR |
3947 | goto retry; |
3948 | clear_bit(CONSIDER_RESYNC, &mdev->flags); | |
b411b363 PR |
3949 | ns.peer = peer_state.role; |
3950 | ns.pdsk = real_peer_disk; | |
3951 | ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); | |
4ac4aada | 3952 | if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) |
b411b363 | 3953 | ns.disk = mdev->new_state_tmp.disk; |
4ac4aada | 3954 | cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); |
2aebfabb | 3955 | if (ns.pdsk == D_CONSISTENT && drbd_suspended(mdev) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && |
481c6f50 | 3956 | test_bit(NEW_CUR_UUID, &mdev->flags)) { |
8554df1c | 3957 | /* Do not allow tl_restart(RESEND) for a rebooted peer. We can only allow this |
481c6f50 | 3958 | for temporal network outages! */ |
87eeee41 | 3959 | spin_unlock_irq(&mdev->tconn->req_lock); |
481c6f50 | 3960 | dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); |
2f5cdd0b | 3961 | tl_clear(mdev->tconn); |
481c6f50 PR |
3962 | drbd_uuid_new_current(mdev); |
3963 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
38fa9988 | 3964 | conn_request_state(mdev->tconn, NS2(conn, C_PROTOCOL_ERROR, susp, 0), CS_HARD); |
82bc0194 | 3965 | return -EIO; |
481c6f50 | 3966 | } |
65d922c3 | 3967 | rv = _drbd_set_state(mdev, ns, cs_flags, NULL); |
78bae59b | 3968 | ns = drbd_read_state(mdev); |
87eeee41 | 3969 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
3970 | |
3971 | if (rv < SS_SUCCESS) { | |
38fa9988 | 3972 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3973 | return -EIO; |
b411b363 PR |
3974 | } |
3975 | ||
4ac4aada LE |
3976 | if (os.conn > C_WF_REPORT_PARAMS) { |
3977 | if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && | |
b411b363 PR |
3978 | peer_state.disk != D_NEGOTIATING ) { |
3979 | /* we want resync, peer has not yet decided to sync... */ | |
3980 | /* Nowadays only used when forcing a node into primary role and | |
3981 | setting its disk to UpToDate with that */ | |
3982 | drbd_send_uuids(mdev); | |
f479ea06 | 3983 | drbd_send_current_state(mdev); |
b411b363 PR |
3984 | } |
3985 | } | |
3986 | ||
08b165ba | 3987 | clear_bit(DISCARD_MY_DATA, &mdev->flags); |
b411b363 | 3988 | |
cccac985 | 3989 | drbd_md_sync(mdev); /* update connected indicator, la_size_sect, ... */ |
b411b363 | 3990 | |
82bc0194 | 3991 | return 0; |
b411b363 PR |
3992 | } |
3993 | ||
4a76b161 | 3994 | static int receive_sync_uuid(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3995 | { |
4a76b161 | 3996 | struct drbd_conf *mdev; |
e658983a | 3997 | struct p_rs_uuid *p = pi->data; |
4a76b161 AG |
3998 | |
3999 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
4000 | if (!mdev) | |
4001 | return -EIO; | |
b411b363 PR |
4002 | |
4003 | wait_event(mdev->misc_wait, | |
4004 | mdev->state.conn == C_WF_SYNC_UUID || | |
c4752ef1 | 4005 | mdev->state.conn == C_BEHIND || |
b411b363 PR |
4006 | mdev->state.conn < C_CONNECTED || |
4007 | mdev->state.disk < D_NEGOTIATING); | |
4008 | ||
4009 | /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */ | |
4010 | ||
b411b363 PR |
4011 | /* Here the _drbd_uuid_ functions are right, current should |
4012 | _not_ be rotated into the history */ | |
4013 | if (get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
4014 | _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid)); | |
4015 | _drbd_uuid_set(mdev, UI_BITMAP, 0UL); | |
4016 | ||
62b0da3a | 4017 | drbd_print_uuids(mdev, "updated sync uuid"); |
b411b363 PR |
4018 | drbd_start_resync(mdev, C_SYNC_TARGET); |
4019 | ||
4020 | put_ldev(mdev); | |
4021 | } else | |
4022 | dev_err(DEV, "Ignoring SyncUUID packet!\n"); | |
4023 | ||
82bc0194 | 4024 | return 0; |
b411b363 PR |
4025 | } |
4026 | ||
2c46407d AG |
4027 | /** |
4028 | * receive_bitmap_plain | |
4029 | * | |
4030 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4031 | * code upon failure. | |
4032 | */ | |
4033 | static int | |
50d0b1ad | 4034 | receive_bitmap_plain(struct drbd_conf *mdev, unsigned int size, |
e658983a | 4035 | unsigned long *p, struct bm_xfer_ctx *c) |
b411b363 | 4036 | { |
50d0b1ad AG |
4037 | unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - |
4038 | drbd_header_size(mdev->tconn); | |
e658983a | 4039 | unsigned int num_words = min_t(size_t, data_size / sizeof(*p), |
50d0b1ad | 4040 | c->bm_words - c->word_offset); |
e658983a | 4041 | unsigned int want = num_words * sizeof(*p); |
2c46407d | 4042 | int err; |
b411b363 | 4043 | |
50d0b1ad AG |
4044 | if (want != size) { |
4045 | dev_err(DEV, "%s:want (%u) != size (%u)\n", __func__, want, size); | |
2c46407d | 4046 | return -EIO; |
b411b363 PR |
4047 | } |
4048 | if (want == 0) | |
2c46407d | 4049 | return 0; |
e658983a | 4050 | err = drbd_recv_all(mdev->tconn, p, want); |
82bc0194 | 4051 | if (err) |
2c46407d | 4052 | return err; |
b411b363 | 4053 | |
e658983a | 4054 | drbd_bm_merge_lel(mdev, c->word_offset, num_words, p); |
b411b363 PR |
4055 | |
4056 | c->word_offset += num_words; | |
4057 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
4058 | if (c->bit_offset > c->bm_bits) | |
4059 | c->bit_offset = c->bm_bits; | |
4060 | ||
2c46407d | 4061 | return 1; |
b411b363 PR |
4062 | } |
4063 | ||
a02d1240 AG |
4064 | static enum drbd_bitmap_code dcbp_get_code(struct p_compressed_bm *p) |
4065 | { | |
4066 | return (enum drbd_bitmap_code)(p->encoding & 0x0f); | |
4067 | } | |
4068 | ||
4069 | static int dcbp_get_start(struct p_compressed_bm *p) | |
4070 | { | |
4071 | return (p->encoding & 0x80) != 0; | |
4072 | } | |
4073 | ||
4074 | static int dcbp_get_pad_bits(struct p_compressed_bm *p) | |
4075 | { | |
4076 | return (p->encoding >> 4) & 0x7; | |
4077 | } | |
4078 | ||
2c46407d AG |
4079 | /** |
4080 | * recv_bm_rle_bits | |
4081 | * | |
4082 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4083 | * code upon failure. | |
4084 | */ | |
4085 | static int | |
b411b363 PR |
4086 | recv_bm_rle_bits(struct drbd_conf *mdev, |
4087 | struct p_compressed_bm *p, | |
c6d25cfe PR |
4088 | struct bm_xfer_ctx *c, |
4089 | unsigned int len) | |
b411b363 PR |
4090 | { |
4091 | struct bitstream bs; | |
4092 | u64 look_ahead; | |
4093 | u64 rl; | |
4094 | u64 tmp; | |
4095 | unsigned long s = c->bit_offset; | |
4096 | unsigned long e; | |
a02d1240 | 4097 | int toggle = dcbp_get_start(p); |
b411b363 PR |
4098 | int have; |
4099 | int bits; | |
4100 | ||
a02d1240 | 4101 | bitstream_init(&bs, p->code, len, dcbp_get_pad_bits(p)); |
b411b363 PR |
4102 | |
4103 | bits = bitstream_get_bits(&bs, &look_ahead, 64); | |
4104 | if (bits < 0) | |
2c46407d | 4105 | return -EIO; |
b411b363 PR |
4106 | |
4107 | for (have = bits; have > 0; s += rl, toggle = !toggle) { | |
4108 | bits = vli_decode_bits(&rl, look_ahead); | |
4109 | if (bits <= 0) | |
2c46407d | 4110 | return -EIO; |
b411b363 PR |
4111 | |
4112 | if (toggle) { | |
4113 | e = s + rl -1; | |
4114 | if (e >= c->bm_bits) { | |
4115 | dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); | |
2c46407d | 4116 | return -EIO; |
b411b363 PR |
4117 | } |
4118 | _drbd_bm_set_bits(mdev, s, e); | |
4119 | } | |
4120 | ||
4121 | if (have < bits) { | |
4122 | dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", | |
4123 | have, bits, look_ahead, | |
4124 | (unsigned int)(bs.cur.b - p->code), | |
4125 | (unsigned int)bs.buf_len); | |
2c46407d | 4126 | return -EIO; |
b411b363 | 4127 | } |
d2da5b0c LE |
4128 | /* if we consumed all 64 bits, assign 0; >> 64 is "undefined"; */ |
4129 | if (likely(bits < 64)) | |
4130 | look_ahead >>= bits; | |
4131 | else | |
4132 | look_ahead = 0; | |
b411b363 PR |
4133 | have -= bits; |
4134 | ||
4135 | bits = bitstream_get_bits(&bs, &tmp, 64 - have); | |
4136 | if (bits < 0) | |
2c46407d | 4137 | return -EIO; |
b411b363 PR |
4138 | look_ahead |= tmp << have; |
4139 | have += bits; | |
4140 | } | |
4141 | ||
4142 | c->bit_offset = s; | |
4143 | bm_xfer_ctx_bit_to_word_offset(c); | |
4144 | ||
2c46407d | 4145 | return (s != c->bm_bits); |
b411b363 PR |
4146 | } |
4147 | ||
2c46407d AG |
4148 | /** |
4149 | * decode_bitmap_c | |
4150 | * | |
4151 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4152 | * code upon failure. | |
4153 | */ | |
4154 | static int | |
b411b363 PR |
4155 | decode_bitmap_c(struct drbd_conf *mdev, |
4156 | struct p_compressed_bm *p, | |
c6d25cfe PR |
4157 | struct bm_xfer_ctx *c, |
4158 | unsigned int len) | |
b411b363 | 4159 | { |
a02d1240 | 4160 | if (dcbp_get_code(p) == RLE_VLI_Bits) |
e658983a | 4161 | return recv_bm_rle_bits(mdev, p, c, len - sizeof(*p)); |
b411b363 PR |
4162 | |
4163 | /* other variants had been implemented for evaluation, | |
4164 | * but have been dropped as this one turned out to be "best" | |
4165 | * during all our tests. */ | |
4166 | ||
4167 | dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding); | |
38fa9988 | 4168 | conn_request_state(mdev->tconn, NS(conn, C_PROTOCOL_ERROR), CS_HARD); |
2c46407d | 4169 | return -EIO; |
b411b363 PR |
4170 | } |
4171 | ||
4172 | void INFO_bm_xfer_stats(struct drbd_conf *mdev, | |
4173 | const char *direction, struct bm_xfer_ctx *c) | |
4174 | { | |
4175 | /* what would it take to transfer it "plaintext" */ | |
50d0b1ad AG |
4176 | unsigned int header_size = drbd_header_size(mdev->tconn); |
4177 | unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; | |
4178 | unsigned int plain = | |
4179 | header_size * (DIV_ROUND_UP(c->bm_words, data_size) + 1) + | |
4180 | c->bm_words * sizeof(unsigned long); | |
4181 | unsigned int total = c->bytes[0] + c->bytes[1]; | |
4182 | unsigned int r; | |
b411b363 PR |
4183 | |
4184 | /* total can not be zero. but just in case: */ | |
4185 | if (total == 0) | |
4186 | return; | |
4187 | ||
4188 | /* don't report if not compressed */ | |
4189 | if (total >= plain) | |
4190 | return; | |
4191 | ||
4192 | /* total < plain. check for overflow, still */ | |
4193 | r = (total > UINT_MAX/1000) ? (total / (plain/1000)) | |
4194 | : (1000 * total / plain); | |
4195 | ||
4196 | if (r > 1000) | |
4197 | r = 1000; | |
4198 | ||
4199 | r = 1000 - r; | |
4200 | dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " | |
4201 | "total %u; compression: %u.%u%%\n", | |
4202 | direction, | |
4203 | c->bytes[1], c->packets[1], | |
4204 | c->bytes[0], c->packets[0], | |
4205 | total, r/10, r % 10); | |
4206 | } | |
4207 | ||
4208 | /* Since we are processing the bitfield from lower addresses to higher, | |
4209 | it does not matter if the process it in 32 bit chunks or 64 bit | |
4210 | chunks as long as it is little endian. (Understand it as byte stream, | |
4211 | beginning with the lowest byte...) If we would use big endian | |
4212 | we would need to process it from the highest address to the lowest, | |
4213 | in order to be agnostic to the 32 vs 64 bits issue. | |
4214 | ||
4215 | returns 0 on failure, 1 if we successfully received it. */ | |
4a76b161 | 4216 | static int receive_bitmap(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4217 | { |
4a76b161 | 4218 | struct drbd_conf *mdev; |
b411b363 | 4219 | struct bm_xfer_ctx c; |
2c46407d | 4220 | int err; |
4a76b161 AG |
4221 | |
4222 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
4223 | if (!mdev) | |
4224 | return -EIO; | |
b411b363 | 4225 | |
20ceb2b2 LE |
4226 | drbd_bm_lock(mdev, "receive bitmap", BM_LOCKED_SET_ALLOWED); |
4227 | /* you are supposed to send additional out-of-sync information | |
4228 | * if you actually set bits during this phase */ | |
b411b363 | 4229 | |
b411b363 PR |
4230 | c = (struct bm_xfer_ctx) { |
4231 | .bm_bits = drbd_bm_bits(mdev), | |
4232 | .bm_words = drbd_bm_words(mdev), | |
4233 | }; | |
4234 | ||
2c46407d | 4235 | for(;;) { |
e658983a AG |
4236 | if (pi->cmd == P_BITMAP) |
4237 | err = receive_bitmap_plain(mdev, pi->size, pi->data, &c); | |
4238 | else if (pi->cmd == P_COMPRESSED_BITMAP) { | |
b411b363 PR |
4239 | /* MAYBE: sanity check that we speak proto >= 90, |
4240 | * and the feature is enabled! */ | |
e658983a | 4241 | struct p_compressed_bm *p = pi->data; |
b411b363 | 4242 | |
50d0b1ad | 4243 | if (pi->size > DRBD_SOCKET_BUFFER_SIZE - drbd_header_size(tconn)) { |
b411b363 | 4244 | dev_err(DEV, "ReportCBitmap packet too large\n"); |
82bc0194 | 4245 | err = -EIO; |
b411b363 PR |
4246 | goto out; |
4247 | } | |
e658983a | 4248 | if (pi->size <= sizeof(*p)) { |
e2857216 | 4249 | dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", pi->size); |
82bc0194 | 4250 | err = -EIO; |
78fcbdae | 4251 | goto out; |
b411b363 | 4252 | } |
e658983a AG |
4253 | err = drbd_recv_all(mdev->tconn, p, pi->size); |
4254 | if (err) | |
4255 | goto out; | |
e2857216 | 4256 | err = decode_bitmap_c(mdev, p, &c, pi->size); |
b411b363 | 4257 | } else { |
e2857216 | 4258 | dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", pi->cmd); |
82bc0194 | 4259 | err = -EIO; |
b411b363 PR |
4260 | goto out; |
4261 | } | |
4262 | ||
e2857216 | 4263 | c.packets[pi->cmd == P_BITMAP]++; |
50d0b1ad | 4264 | c.bytes[pi->cmd == P_BITMAP] += drbd_header_size(tconn) + pi->size; |
b411b363 | 4265 | |
2c46407d AG |
4266 | if (err <= 0) { |
4267 | if (err < 0) | |
4268 | goto out; | |
b411b363 | 4269 | break; |
2c46407d | 4270 | } |
e2857216 | 4271 | err = drbd_recv_header(mdev->tconn, pi); |
82bc0194 | 4272 | if (err) |
b411b363 | 4273 | goto out; |
2c46407d | 4274 | } |
b411b363 PR |
4275 | |
4276 | INFO_bm_xfer_stats(mdev, "receive", &c); | |
4277 | ||
4278 | if (mdev->state.conn == C_WF_BITMAP_T) { | |
de1f8e4a AG |
4279 | enum drbd_state_rv rv; |
4280 | ||
82bc0194 AG |
4281 | err = drbd_send_bitmap(mdev); |
4282 | if (err) | |
b411b363 PR |
4283 | goto out; |
4284 | /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ | |
de1f8e4a AG |
4285 | rv = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); |
4286 | D_ASSERT(rv == SS_SUCCESS); | |
b411b363 PR |
4287 | } else if (mdev->state.conn != C_WF_BITMAP_S) { |
4288 | /* admin may have requested C_DISCONNECTING, | |
4289 | * other threads may have noticed network errors */ | |
4290 | dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n", | |
4291 | drbd_conn_str(mdev->state.conn)); | |
4292 | } | |
82bc0194 | 4293 | err = 0; |
b411b363 | 4294 | |
b411b363 | 4295 | out: |
20ceb2b2 | 4296 | drbd_bm_unlock(mdev); |
82bc0194 | 4297 | if (!err && mdev->state.conn == C_WF_BITMAP_S) |
b411b363 | 4298 | drbd_start_resync(mdev, C_SYNC_SOURCE); |
82bc0194 | 4299 | return err; |
b411b363 PR |
4300 | } |
4301 | ||
4a76b161 | 4302 | static int receive_skip(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4303 | { |
4a76b161 | 4304 | conn_warn(tconn, "skipping unknown optional packet type %d, l: %d!\n", |
e2857216 | 4305 | pi->cmd, pi->size); |
b411b363 | 4306 | |
4a76b161 | 4307 | return ignore_remaining_packet(tconn, pi); |
b411b363 PR |
4308 | } |
4309 | ||
4a76b161 | 4310 | static int receive_UnplugRemote(struct drbd_tconn *tconn, struct packet_info *pi) |
0ced55a3 | 4311 | { |
e7f52dfb LE |
4312 | /* Make sure we've acked all the TCP data associated |
4313 | * with the data requests being unplugged */ | |
4a76b161 | 4314 | drbd_tcp_quickack(tconn->data.socket); |
0ced55a3 | 4315 | |
82bc0194 | 4316 | return 0; |
0ced55a3 PR |
4317 | } |
4318 | ||
4a76b161 | 4319 | static int receive_out_of_sync(struct drbd_tconn *tconn, struct packet_info *pi) |
73a01a18 | 4320 | { |
4a76b161 | 4321 | struct drbd_conf *mdev; |
e658983a | 4322 | struct p_block_desc *p = pi->data; |
4a76b161 AG |
4323 | |
4324 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
4325 | if (!mdev) | |
4326 | return -EIO; | |
73a01a18 | 4327 | |
f735e363 LE |
4328 | switch (mdev->state.conn) { |
4329 | case C_WF_SYNC_UUID: | |
4330 | case C_WF_BITMAP_T: | |
4331 | case C_BEHIND: | |
4332 | break; | |
4333 | default: | |
4334 | dev_err(DEV, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n", | |
4335 | drbd_conn_str(mdev->state.conn)); | |
4336 | } | |
4337 | ||
73a01a18 PR |
4338 | drbd_set_out_of_sync(mdev, be64_to_cpu(p->sector), be32_to_cpu(p->blksize)); |
4339 | ||
82bc0194 | 4340 | return 0; |
73a01a18 PR |
4341 | } |
4342 | ||
02918be2 PR |
4343 | struct data_cmd { |
4344 | int expect_payload; | |
4345 | size_t pkt_size; | |
4a76b161 | 4346 | int (*fn)(struct drbd_tconn *, struct packet_info *); |
02918be2 PR |
4347 | }; |
4348 | ||
4349 | static struct data_cmd drbd_cmd_handler[] = { | |
4350 | [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, | |
4351 | [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, | |
4352 | [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , | |
4353 | [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , | |
e658983a AG |
4354 | [P_BITMAP] = { 1, 0, receive_bitmap } , |
4355 | [P_COMPRESSED_BITMAP] = { 1, 0, receive_bitmap } , | |
4356 | [P_UNPLUG_REMOTE] = { 0, 0, receive_UnplugRemote }, | |
02918be2 PR |
4357 | [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, |
4358 | [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
e658983a AG |
4359 | [P_SYNC_PARAM] = { 1, 0, receive_SyncParam }, |
4360 | [P_SYNC_PARAM89] = { 1, 0, receive_SyncParam }, | |
02918be2 PR |
4361 | [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, |
4362 | [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, | |
4363 | [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, | |
4364 | [P_STATE] = { 0, sizeof(struct p_state), receive_state }, | |
4365 | [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, | |
4366 | [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, | |
4367 | [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
4368 | [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
4369 | [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
4370 | [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, | |
73a01a18 | 4371 | [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync }, |
4a76b161 | 4372 | [P_CONN_ST_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_conn_state }, |
036b17ea | 4373 | [P_PROTOCOL_UPDATE] = { 1, sizeof(struct p_protocol), receive_protocol }, |
b411b363 PR |
4374 | }; |
4375 | ||
eefc2f7d | 4376 | static void drbdd(struct drbd_tconn *tconn) |
b411b363 | 4377 | { |
77351055 | 4378 | struct packet_info pi; |
02918be2 | 4379 | size_t shs; /* sub header size */ |
82bc0194 | 4380 | int err; |
b411b363 | 4381 | |
eefc2f7d | 4382 | while (get_t_state(&tconn->receiver) == RUNNING) { |
deebe195 | 4383 | struct data_cmd *cmd; |
b411b363 | 4384 | |
eefc2f7d | 4385 | drbd_thread_current_set_cpu(&tconn->receiver); |
69bc7bc3 | 4386 | if (drbd_recv_header(tconn, &pi)) |
02918be2 | 4387 | goto err_out; |
b411b363 | 4388 | |
deebe195 | 4389 | cmd = &drbd_cmd_handler[pi.cmd]; |
4a76b161 | 4390 | if (unlikely(pi.cmd >= ARRAY_SIZE(drbd_cmd_handler) || !cmd->fn)) { |
2fcb8f30 AG |
4391 | conn_err(tconn, "Unexpected data packet %s (0x%04x)", |
4392 | cmdname(pi.cmd), pi.cmd); | |
02918be2 | 4393 | goto err_out; |
0b33a916 | 4394 | } |
b411b363 | 4395 | |
e658983a AG |
4396 | shs = cmd->pkt_size; |
4397 | if (pi.size > shs && !cmd->expect_payload) { | |
2fcb8f30 AG |
4398 | conn_err(tconn, "No payload expected %s l:%d\n", |
4399 | cmdname(pi.cmd), pi.size); | |
02918be2 | 4400 | goto err_out; |
b411b363 | 4401 | } |
b411b363 | 4402 | |
c13f7e1a | 4403 | if (shs) { |
e658983a | 4404 | err = drbd_recv_all_warn(tconn, pi.data, shs); |
a5c31904 | 4405 | if (err) |
c13f7e1a | 4406 | goto err_out; |
e2857216 | 4407 | pi.size -= shs; |
c13f7e1a LE |
4408 | } |
4409 | ||
4a76b161 AG |
4410 | err = cmd->fn(tconn, &pi); |
4411 | if (err) { | |
9f5bdc33 AG |
4412 | conn_err(tconn, "error receiving %s, e: %d l: %d!\n", |
4413 | cmdname(pi.cmd), err, pi.size); | |
02918be2 | 4414 | goto err_out; |
b411b363 PR |
4415 | } |
4416 | } | |
82bc0194 | 4417 | return; |
b411b363 | 4418 | |
82bc0194 AG |
4419 | err_out: |
4420 | conn_request_state(tconn, NS(conn, C_PROTOCOL_ERROR), CS_HARD); | |
b411b363 PR |
4421 | } |
4422 | ||
0e29d163 | 4423 | void conn_flush_workqueue(struct drbd_tconn *tconn) |
b411b363 PR |
4424 | { |
4425 | struct drbd_wq_barrier barr; | |
4426 | ||
4427 | barr.w.cb = w_prev_work_done; | |
0e29d163 | 4428 | barr.w.tconn = tconn; |
b411b363 | 4429 | init_completion(&barr.done); |
d5b27b01 | 4430 | drbd_queue_work(&tconn->sender_work, &barr.w); |
b411b363 PR |
4431 | wait_for_completion(&barr.done); |
4432 | } | |
4433 | ||
81fa2e67 | 4434 | static void conn_disconnect(struct drbd_tconn *tconn) |
b411b363 | 4435 | { |
c141ebda | 4436 | struct drbd_conf *mdev; |
bbeb641c | 4437 | enum drbd_conns oc; |
376694a0 | 4438 | int vnr; |
b411b363 | 4439 | |
bbeb641c | 4440 | if (tconn->cstate == C_STANDALONE) |
b411b363 | 4441 | return; |
b411b363 | 4442 | |
545752d5 LE |
4443 | /* We are about to start the cleanup after connection loss. |
4444 | * Make sure drbd_make_request knows about that. | |
4445 | * Usually we should be in some network failure state already, | |
4446 | * but just in case we are not, we fix it up here. | |
4447 | */ | |
b8853dbd | 4448 | conn_request_state(tconn, NS(conn, C_NETWORK_FAILURE), CS_HARD); |
545752d5 | 4449 | |
b411b363 | 4450 | /* asender does not clean up anything. it must not interfere, either */ |
360cc740 PR |
4451 | drbd_thread_stop(&tconn->asender); |
4452 | drbd_free_sock(tconn); | |
4453 | ||
c141ebda PR |
4454 | rcu_read_lock(); |
4455 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { | |
4456 | kref_get(&mdev->kref); | |
4457 | rcu_read_unlock(); | |
4458 | drbd_disconnected(mdev); | |
4459 | kref_put(&mdev->kref, &drbd_minor_destroy); | |
4460 | rcu_read_lock(); | |
4461 | } | |
4462 | rcu_read_unlock(); | |
4463 | ||
12038a3a PR |
4464 | if (!list_empty(&tconn->current_epoch->list)) |
4465 | conn_err(tconn, "ASSERTION FAILED: tconn->current_epoch->list not empty\n"); | |
4466 | /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ | |
4467 | atomic_set(&tconn->current_epoch->epoch_size, 0); | |
b6dd1a89 | 4468 | tconn->send.seen_any_write_yet = false; |
12038a3a | 4469 | |
360cc740 PR |
4470 | conn_info(tconn, "Connection closed\n"); |
4471 | ||
cb703454 PR |
4472 | if (conn_highest_role(tconn) == R_PRIMARY && conn_highest_pdsk(tconn) >= D_UNKNOWN) |
4473 | conn_try_outdate_peer_async(tconn); | |
4474 | ||
360cc740 | 4475 | spin_lock_irq(&tconn->req_lock); |
bbeb641c PR |
4476 | oc = tconn->cstate; |
4477 | if (oc >= C_UNCONNECTED) | |
376694a0 | 4478 | _conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE); |
bbeb641c | 4479 | |
360cc740 PR |
4480 | spin_unlock_irq(&tconn->req_lock); |
4481 | ||
f3dfa40a | 4482 | if (oc == C_DISCONNECTING) |
d9cc6e23 | 4483 | conn_request_state(tconn, NS(conn, C_STANDALONE), CS_VERBOSE | CS_HARD); |
360cc740 PR |
4484 | } |
4485 | ||
c141ebda | 4486 | static int drbd_disconnected(struct drbd_conf *mdev) |
360cc740 | 4487 | { |
360cc740 | 4488 | unsigned int i; |
b411b363 | 4489 | |
85719573 | 4490 | /* wait for current activity to cease. */ |
87eeee41 | 4491 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4492 | _drbd_wait_ee_list_empty(mdev, &mdev->active_ee); |
4493 | _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee); | |
4494 | _drbd_wait_ee_list_empty(mdev, &mdev->read_ee); | |
87eeee41 | 4495 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4496 | |
4497 | /* We do not have data structures that would allow us to | |
4498 | * get the rs_pending_cnt down to 0 again. | |
4499 | * * On C_SYNC_TARGET we do not have any data structures describing | |
4500 | * the pending RSDataRequest's we have sent. | |
4501 | * * On C_SYNC_SOURCE there is no data structure that tracks | |
4502 | * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. | |
4503 | * And no, it is not the sum of the reference counts in the | |
4504 | * resync_LRU. The resync_LRU tracks the whole operation including | |
4505 | * the disk-IO, while the rs_pending_cnt only tracks the blocks | |
4506 | * on the fly. */ | |
4507 | drbd_rs_cancel_all(mdev); | |
4508 | mdev->rs_total = 0; | |
4509 | mdev->rs_failed = 0; | |
4510 | atomic_set(&mdev->rs_pending_cnt, 0); | |
4511 | wake_up(&mdev->misc_wait); | |
4512 | ||
b411b363 | 4513 | del_timer_sync(&mdev->resync_timer); |
b411b363 PR |
4514 | resync_timer_fn((unsigned long)mdev); |
4515 | ||
b411b363 PR |
4516 | /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, |
4517 | * w_make_resync_request etc. which may still be on the worker queue | |
4518 | * to be "canceled" */ | |
4519 | drbd_flush_workqueue(mdev); | |
4520 | ||
a990be46 | 4521 | drbd_finish_peer_reqs(mdev); |
b411b363 | 4522 | |
d10b4ea3 PR |
4523 | /* This second workqueue flush is necessary, since drbd_finish_peer_reqs() |
4524 | might have issued a work again. The one before drbd_finish_peer_reqs() is | |
4525 | necessary to reclain net_ee in drbd_finish_peer_reqs(). */ | |
4526 | drbd_flush_workqueue(mdev); | |
4527 | ||
08332d73 LE |
4528 | /* need to do it again, drbd_finish_peer_reqs() may have populated it |
4529 | * again via drbd_try_clear_on_disk_bm(). */ | |
4530 | drbd_rs_cancel_all(mdev); | |
b411b363 PR |
4531 | |
4532 | kfree(mdev->p_uuid); | |
4533 | mdev->p_uuid = NULL; | |
4534 | ||
2aebfabb | 4535 | if (!drbd_suspended(mdev)) |
2f5cdd0b | 4536 | tl_clear(mdev->tconn); |
b411b363 PR |
4537 | |
4538 | drbd_md_sync(mdev); | |
4539 | ||
20ceb2b2 LE |
4540 | /* serialize with bitmap writeout triggered by the state change, |
4541 | * if any. */ | |
4542 | wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags)); | |
4543 | ||
b411b363 PR |
4544 | /* tcp_close and release of sendpage pages can be deferred. I don't |
4545 | * want to use SO_LINGER, because apparently it can be deferred for | |
4546 | * more than 20 seconds (longest time I checked). | |
4547 | * | |
4548 | * Actually we don't care for exactly when the network stack does its | |
4549 | * put_page(), but release our reference on these pages right here. | |
4550 | */ | |
7721f567 | 4551 | i = drbd_free_peer_reqs(mdev, &mdev->net_ee); |
b411b363 PR |
4552 | if (i) |
4553 | dev_info(DEV, "net_ee not empty, killed %u entries\n", i); | |
435f0740 LE |
4554 | i = atomic_read(&mdev->pp_in_use_by_net); |
4555 | if (i) | |
4556 | dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i); | |
b411b363 PR |
4557 | i = atomic_read(&mdev->pp_in_use); |
4558 | if (i) | |
45bb912b | 4559 | dev_info(DEV, "pp_in_use = %d, expected 0\n", i); |
b411b363 PR |
4560 | |
4561 | D_ASSERT(list_empty(&mdev->read_ee)); | |
4562 | D_ASSERT(list_empty(&mdev->active_ee)); | |
4563 | D_ASSERT(list_empty(&mdev->sync_ee)); | |
4564 | D_ASSERT(list_empty(&mdev->done_ee)); | |
4565 | ||
360cc740 | 4566 | return 0; |
b411b363 PR |
4567 | } |
4568 | ||
4569 | /* | |
4570 | * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version | |
4571 | * we can agree on is stored in agreed_pro_version. | |
4572 | * | |
4573 | * feature flags and the reserved array should be enough room for future | |
4574 | * enhancements of the handshake protocol, and possible plugins... | |
4575 | * | |
4576 | * for now, they are expected to be zero, but ignored. | |
4577 | */ | |
6038178e | 4578 | static int drbd_send_features(struct drbd_tconn *tconn) |
b411b363 | 4579 | { |
9f5bdc33 AG |
4580 | struct drbd_socket *sock; |
4581 | struct p_connection_features *p; | |
b411b363 | 4582 | |
9f5bdc33 AG |
4583 | sock = &tconn->data; |
4584 | p = conn_prepare_command(tconn, sock); | |
4585 | if (!p) | |
e8d17b01 | 4586 | return -EIO; |
b411b363 PR |
4587 | memset(p, 0, sizeof(*p)); |
4588 | p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); | |
4589 | p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); | |
9f5bdc33 | 4590 | return conn_send_command(tconn, sock, P_CONNECTION_FEATURES, sizeof(*p), NULL, 0); |
b411b363 PR |
4591 | } |
4592 | ||
4593 | /* | |
4594 | * return values: | |
4595 | * 1 yes, we have a valid connection | |
4596 | * 0 oops, did not work out, please try again | |
4597 | * -1 peer talks different language, | |
4598 | * no point in trying again, please go standalone. | |
4599 | */ | |
6038178e | 4600 | static int drbd_do_features(struct drbd_tconn *tconn) |
b411b363 | 4601 | { |
65d11ed6 | 4602 | /* ASSERT current == tconn->receiver ... */ |
e658983a AG |
4603 | struct p_connection_features *p; |
4604 | const int expect = sizeof(struct p_connection_features); | |
77351055 | 4605 | struct packet_info pi; |
a5c31904 | 4606 | int err; |
b411b363 | 4607 | |
6038178e | 4608 | err = drbd_send_features(tconn); |
e8d17b01 | 4609 | if (err) |
b411b363 PR |
4610 | return 0; |
4611 | ||
69bc7bc3 AG |
4612 | err = drbd_recv_header(tconn, &pi); |
4613 | if (err) | |
b411b363 PR |
4614 | return 0; |
4615 | ||
6038178e AG |
4616 | if (pi.cmd != P_CONNECTION_FEATURES) { |
4617 | conn_err(tconn, "expected ConnectionFeatures packet, received: %s (0x%04x)\n", | |
2fcb8f30 | 4618 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4619 | return -1; |
4620 | } | |
4621 | ||
77351055 | 4622 | if (pi.size != expect) { |
6038178e | 4623 | conn_err(tconn, "expected ConnectionFeatures length: %u, received: %u\n", |
77351055 | 4624 | expect, pi.size); |
b411b363 PR |
4625 | return -1; |
4626 | } | |
4627 | ||
e658983a AG |
4628 | p = pi.data; |
4629 | err = drbd_recv_all_warn(tconn, p, expect); | |
a5c31904 | 4630 | if (err) |
b411b363 | 4631 | return 0; |
b411b363 | 4632 | |
b411b363 PR |
4633 | p->protocol_min = be32_to_cpu(p->protocol_min); |
4634 | p->protocol_max = be32_to_cpu(p->protocol_max); | |
4635 | if (p->protocol_max == 0) | |
4636 | p->protocol_max = p->protocol_min; | |
4637 | ||
4638 | if (PRO_VERSION_MAX < p->protocol_min || | |
4639 | PRO_VERSION_MIN > p->protocol_max) | |
4640 | goto incompat; | |
4641 | ||
65d11ed6 | 4642 | tconn->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); |
b411b363 | 4643 | |
65d11ed6 PR |
4644 | conn_info(tconn, "Handshake successful: " |
4645 | "Agreed network protocol version %d\n", tconn->agreed_pro_version); | |
b411b363 PR |
4646 | |
4647 | return 1; | |
4648 | ||
4649 | incompat: | |
65d11ed6 | 4650 | conn_err(tconn, "incompatible DRBD dialects: " |
b411b363 PR |
4651 | "I support %d-%d, peer supports %d-%d\n", |
4652 | PRO_VERSION_MIN, PRO_VERSION_MAX, | |
4653 | p->protocol_min, p->protocol_max); | |
4654 | return -1; | |
4655 | } | |
4656 | ||
4657 | #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) | |
13e6037d | 4658 | static int drbd_do_auth(struct drbd_tconn *tconn) |
b411b363 | 4659 | { |
ef57f9e6 PR |
4660 | conn_err(tconn, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); |
4661 | conn_err(tconn, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); | |
b10d96cb | 4662 | return -1; |
b411b363 PR |
4663 | } |
4664 | #else | |
4665 | #define CHALLENGE_LEN 64 | |
b10d96cb JT |
4666 | |
4667 | /* Return value: | |
4668 | 1 - auth succeeded, | |
4669 | 0 - failed, try again (network error), | |
4670 | -1 - auth failed, don't try again. | |
4671 | */ | |
4672 | ||
13e6037d | 4673 | static int drbd_do_auth(struct drbd_tconn *tconn) |
b411b363 | 4674 | { |
9f5bdc33 | 4675 | struct drbd_socket *sock; |
b411b363 PR |
4676 | char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ |
4677 | struct scatterlist sg; | |
4678 | char *response = NULL; | |
4679 | char *right_response = NULL; | |
4680 | char *peers_ch = NULL; | |
44ed167d PR |
4681 | unsigned int key_len; |
4682 | char secret[SHARED_SECRET_MAX]; /* 64 byte */ | |
b411b363 PR |
4683 | unsigned int resp_size; |
4684 | struct hash_desc desc; | |
77351055 | 4685 | struct packet_info pi; |
44ed167d | 4686 | struct net_conf *nc; |
69bc7bc3 | 4687 | int err, rv; |
b411b363 | 4688 | |
9f5bdc33 | 4689 | /* FIXME: Put the challenge/response into the preallocated socket buffer. */ |
b411b363 | 4690 | |
44ed167d PR |
4691 | rcu_read_lock(); |
4692 | nc = rcu_dereference(tconn->net_conf); | |
4693 | key_len = strlen(nc->shared_secret); | |
4694 | memcpy(secret, nc->shared_secret, key_len); | |
4695 | rcu_read_unlock(); | |
4696 | ||
13e6037d | 4697 | desc.tfm = tconn->cram_hmac_tfm; |
b411b363 PR |
4698 | desc.flags = 0; |
4699 | ||
44ed167d | 4700 | rv = crypto_hash_setkey(tconn->cram_hmac_tfm, (u8 *)secret, key_len); |
b411b363 | 4701 | if (rv) { |
13e6037d | 4702 | conn_err(tconn, "crypto_hash_setkey() failed with %d\n", rv); |
b10d96cb | 4703 | rv = -1; |
b411b363 PR |
4704 | goto fail; |
4705 | } | |
4706 | ||
4707 | get_random_bytes(my_challenge, CHALLENGE_LEN); | |
4708 | ||
9f5bdc33 AG |
4709 | sock = &tconn->data; |
4710 | if (!conn_prepare_command(tconn, sock)) { | |
4711 | rv = 0; | |
4712 | goto fail; | |
4713 | } | |
e658983a | 4714 | rv = !conn_send_command(tconn, sock, P_AUTH_CHALLENGE, 0, |
9f5bdc33 | 4715 | my_challenge, CHALLENGE_LEN); |
b411b363 PR |
4716 | if (!rv) |
4717 | goto fail; | |
4718 | ||
69bc7bc3 AG |
4719 | err = drbd_recv_header(tconn, &pi); |
4720 | if (err) { | |
4721 | rv = 0; | |
b411b363 | 4722 | goto fail; |
69bc7bc3 | 4723 | } |
b411b363 | 4724 | |
77351055 | 4725 | if (pi.cmd != P_AUTH_CHALLENGE) { |
13e6037d | 4726 | conn_err(tconn, "expected AuthChallenge packet, received: %s (0x%04x)\n", |
2fcb8f30 | 4727 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4728 | rv = 0; |
4729 | goto fail; | |
4730 | } | |
4731 | ||
77351055 | 4732 | if (pi.size > CHALLENGE_LEN * 2) { |
13e6037d | 4733 | conn_err(tconn, "expected AuthChallenge payload too big.\n"); |
b10d96cb | 4734 | rv = -1; |
b411b363 PR |
4735 | goto fail; |
4736 | } | |
4737 | ||
77351055 | 4738 | peers_ch = kmalloc(pi.size, GFP_NOIO); |
b411b363 | 4739 | if (peers_ch == NULL) { |
13e6037d | 4740 | conn_err(tconn, "kmalloc of peers_ch failed\n"); |
b10d96cb | 4741 | rv = -1; |
b411b363 PR |
4742 | goto fail; |
4743 | } | |
4744 | ||
a5c31904 AG |
4745 | err = drbd_recv_all_warn(tconn, peers_ch, pi.size); |
4746 | if (err) { | |
b411b363 PR |
4747 | rv = 0; |
4748 | goto fail; | |
4749 | } | |
4750 | ||
13e6037d | 4751 | resp_size = crypto_hash_digestsize(tconn->cram_hmac_tfm); |
b411b363 PR |
4752 | response = kmalloc(resp_size, GFP_NOIO); |
4753 | if (response == NULL) { | |
13e6037d | 4754 | conn_err(tconn, "kmalloc of response failed\n"); |
b10d96cb | 4755 | rv = -1; |
b411b363 PR |
4756 | goto fail; |
4757 | } | |
4758 | ||
4759 | sg_init_table(&sg, 1); | |
77351055 | 4760 | sg_set_buf(&sg, peers_ch, pi.size); |
b411b363 PR |
4761 | |
4762 | rv = crypto_hash_digest(&desc, &sg, sg.length, response); | |
4763 | if (rv) { | |
13e6037d | 4764 | conn_err(tconn, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 4765 | rv = -1; |
b411b363 PR |
4766 | goto fail; |
4767 | } | |
4768 | ||
9f5bdc33 AG |
4769 | if (!conn_prepare_command(tconn, sock)) { |
4770 | rv = 0; | |
b411b363 | 4771 | goto fail; |
9f5bdc33 | 4772 | } |
e658983a | 4773 | rv = !conn_send_command(tconn, sock, P_AUTH_RESPONSE, 0, |
9f5bdc33 | 4774 | response, resp_size); |
b411b363 PR |
4775 | if (!rv) |
4776 | goto fail; | |
4777 | ||
69bc7bc3 AG |
4778 | err = drbd_recv_header(tconn, &pi); |
4779 | if (err) { | |
b411b363 PR |
4780 | rv = 0; |
4781 | goto fail; | |
4782 | } | |
4783 | ||
77351055 | 4784 | if (pi.cmd != P_AUTH_RESPONSE) { |
13e6037d | 4785 | conn_err(tconn, "expected AuthResponse packet, received: %s (0x%04x)\n", |
2fcb8f30 | 4786 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4787 | rv = 0; |
4788 | goto fail; | |
4789 | } | |
4790 | ||
77351055 | 4791 | if (pi.size != resp_size) { |
13e6037d | 4792 | conn_err(tconn, "expected AuthResponse payload of wrong size\n"); |
b411b363 PR |
4793 | rv = 0; |
4794 | goto fail; | |
4795 | } | |
b411b363 | 4796 | |
a5c31904 AG |
4797 | err = drbd_recv_all_warn(tconn, response , resp_size); |
4798 | if (err) { | |
b411b363 PR |
4799 | rv = 0; |
4800 | goto fail; | |
4801 | } | |
4802 | ||
4803 | right_response = kmalloc(resp_size, GFP_NOIO); | |
2d1ee87d | 4804 | if (right_response == NULL) { |
13e6037d | 4805 | conn_err(tconn, "kmalloc of right_response failed\n"); |
b10d96cb | 4806 | rv = -1; |
b411b363 PR |
4807 | goto fail; |
4808 | } | |
4809 | ||
4810 | sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); | |
4811 | ||
4812 | rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); | |
4813 | if (rv) { | |
13e6037d | 4814 | conn_err(tconn, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 4815 | rv = -1; |
b411b363 PR |
4816 | goto fail; |
4817 | } | |
4818 | ||
4819 | rv = !memcmp(response, right_response, resp_size); | |
4820 | ||
4821 | if (rv) | |
44ed167d PR |
4822 | conn_info(tconn, "Peer authenticated using %d bytes HMAC\n", |
4823 | resp_size); | |
b10d96cb JT |
4824 | else |
4825 | rv = -1; | |
b411b363 PR |
4826 | |
4827 | fail: | |
4828 | kfree(peers_ch); | |
4829 | kfree(response); | |
4830 | kfree(right_response); | |
4831 | ||
4832 | return rv; | |
4833 | } | |
4834 | #endif | |
4835 | ||
4836 | int drbdd_init(struct drbd_thread *thi) | |
4837 | { | |
392c8801 | 4838 | struct drbd_tconn *tconn = thi->tconn; |
b411b363 PR |
4839 | int h; |
4840 | ||
4d641dd7 | 4841 | conn_info(tconn, "receiver (re)started\n"); |
b411b363 PR |
4842 | |
4843 | do { | |
81fa2e67 | 4844 | h = conn_connect(tconn); |
b411b363 | 4845 | if (h == 0) { |
81fa2e67 | 4846 | conn_disconnect(tconn); |
20ee6390 | 4847 | schedule_timeout_interruptible(HZ); |
b411b363 PR |
4848 | } |
4849 | if (h == -1) { | |
4d641dd7 | 4850 | conn_warn(tconn, "Discarding network configuration.\n"); |
bbeb641c | 4851 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
4852 | } |
4853 | } while (h == 0); | |
4854 | ||
91fd4dad PR |
4855 | if (h > 0) |
4856 | drbdd(tconn); | |
b411b363 | 4857 | |
81fa2e67 | 4858 | conn_disconnect(tconn); |
b411b363 | 4859 | |
4d641dd7 | 4860 | conn_info(tconn, "receiver terminated\n"); |
b411b363 PR |
4861 | return 0; |
4862 | } | |
4863 | ||
4864 | /* ********* acknowledge sender ******** */ | |
4865 | ||
e05e1e59 | 4866 | static int got_conn_RqSReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4867 | { |
e658983a | 4868 | struct p_req_state_reply *p = pi->data; |
e4f78ede PR |
4869 | int retcode = be32_to_cpu(p->retcode); |
4870 | ||
4871 | if (retcode >= SS_SUCCESS) { | |
4872 | set_bit(CONN_WD_ST_CHG_OKAY, &tconn->flags); | |
4873 | } else { | |
4874 | set_bit(CONN_WD_ST_CHG_FAIL, &tconn->flags); | |
4875 | conn_err(tconn, "Requested state change failed by peer: %s (%d)\n", | |
4876 | drbd_set_st_err_str(retcode), retcode); | |
4877 | } | |
4878 | wake_up(&tconn->ping_wait); | |
4879 | ||
2735a594 | 4880 | return 0; |
e4f78ede | 4881 | } |
b411b363 | 4882 | |
1952e916 | 4883 | static int got_RqSReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4884 | { |
1952e916 | 4885 | struct drbd_conf *mdev; |
e658983a | 4886 | struct p_req_state_reply *p = pi->data; |
b411b363 PR |
4887 | int retcode = be32_to_cpu(p->retcode); |
4888 | ||
1952e916 AG |
4889 | mdev = vnr_to_mdev(tconn, pi->vnr); |
4890 | if (!mdev) | |
2735a594 | 4891 | return -EIO; |
1952e916 | 4892 | |
4d0fc3fd PR |
4893 | if (test_bit(CONN_WD_ST_CHG_REQ, &tconn->flags)) { |
4894 | D_ASSERT(tconn->agreed_pro_version < 100); | |
4895 | return got_conn_RqSReply(tconn, pi); | |
4896 | } | |
4897 | ||
b411b363 | 4898 | if (retcode >= SS_SUCCESS) { |
e4f78ede | 4899 | set_bit(CL_ST_CHG_SUCCESS, &mdev->flags); |
b411b363 | 4900 | } else { |
e4f78ede | 4901 | set_bit(CL_ST_CHG_FAIL, &mdev->flags); |
b411b363 | 4902 | dev_err(DEV, "Requested state change failed by peer: %s (%d)\n", |
e4f78ede | 4903 | drbd_set_st_err_str(retcode), retcode); |
b411b363 PR |
4904 | } |
4905 | wake_up(&mdev->state_wait); | |
4906 | ||
2735a594 | 4907 | return 0; |
b411b363 PR |
4908 | } |
4909 | ||
e05e1e59 | 4910 | static int got_Ping(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4911 | { |
2735a594 | 4912 | return drbd_send_ping_ack(tconn); |
b411b363 PR |
4913 | |
4914 | } | |
4915 | ||
e05e1e59 | 4916 | static int got_PingAck(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 PR |
4917 | { |
4918 | /* restore idle timeout */ | |
2a67d8b9 PR |
4919 | tconn->meta.socket->sk->sk_rcvtimeo = tconn->net_conf->ping_int*HZ; |
4920 | if (!test_and_set_bit(GOT_PING_ACK, &tconn->flags)) | |
4921 | wake_up(&tconn->ping_wait); | |
b411b363 | 4922 | |
2735a594 | 4923 | return 0; |
b411b363 PR |
4924 | } |
4925 | ||
1952e916 | 4926 | static int got_IsInSync(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4927 | { |
1952e916 | 4928 | struct drbd_conf *mdev; |
e658983a | 4929 | struct p_block_ack *p = pi->data; |
b411b363 PR |
4930 | sector_t sector = be64_to_cpu(p->sector); |
4931 | int blksize = be32_to_cpu(p->blksize); | |
4932 | ||
1952e916 AG |
4933 | mdev = vnr_to_mdev(tconn, pi->vnr); |
4934 | if (!mdev) | |
2735a594 | 4935 | return -EIO; |
1952e916 | 4936 | |
31890f4a | 4937 | D_ASSERT(mdev->tconn->agreed_pro_version >= 89); |
b411b363 PR |
4938 | |
4939 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4940 | ||
1d53f09e LE |
4941 | if (get_ldev(mdev)) { |
4942 | drbd_rs_complete_io(mdev, sector); | |
4943 | drbd_set_in_sync(mdev, sector, blksize); | |
4944 | /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ | |
4945 | mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); | |
4946 | put_ldev(mdev); | |
4947 | } | |
b411b363 | 4948 | dec_rs_pending(mdev); |
778f271d | 4949 | atomic_add(blksize >> 9, &mdev->rs_sect_in); |
b411b363 | 4950 | |
2735a594 | 4951 | return 0; |
b411b363 PR |
4952 | } |
4953 | ||
bc9c5c41 AG |
4954 | static int |
4955 | validate_req_change_req_state(struct drbd_conf *mdev, u64 id, sector_t sector, | |
4956 | struct rb_root *root, const char *func, | |
4957 | enum drbd_req_event what, bool missing_ok) | |
b411b363 PR |
4958 | { |
4959 | struct drbd_request *req; | |
4960 | struct bio_and_error m; | |
4961 | ||
87eeee41 | 4962 | spin_lock_irq(&mdev->tconn->req_lock); |
bc9c5c41 | 4963 | req = find_request(mdev, root, id, sector, missing_ok, func); |
b411b363 | 4964 | if (unlikely(!req)) { |
87eeee41 | 4965 | spin_unlock_irq(&mdev->tconn->req_lock); |
85997675 | 4966 | return -EIO; |
b411b363 PR |
4967 | } |
4968 | __req_mod(req, what, &m); | |
87eeee41 | 4969 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4970 | |
4971 | if (m.bio) | |
4972 | complete_master_bio(mdev, &m); | |
85997675 | 4973 | return 0; |
b411b363 PR |
4974 | } |
4975 | ||
1952e916 | 4976 | static int got_BlockAck(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4977 | { |
1952e916 | 4978 | struct drbd_conf *mdev; |
e658983a | 4979 | struct p_block_ack *p = pi->data; |
b411b363 PR |
4980 | sector_t sector = be64_to_cpu(p->sector); |
4981 | int blksize = be32_to_cpu(p->blksize); | |
4982 | enum drbd_req_event what; | |
4983 | ||
1952e916 AG |
4984 | mdev = vnr_to_mdev(tconn, pi->vnr); |
4985 | if (!mdev) | |
2735a594 | 4986 | return -EIO; |
1952e916 | 4987 | |
b411b363 PR |
4988 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); |
4989 | ||
579b57ed | 4990 | if (p->block_id == ID_SYNCER) { |
b411b363 PR |
4991 | drbd_set_in_sync(mdev, sector, blksize); |
4992 | dec_rs_pending(mdev); | |
2735a594 | 4993 | return 0; |
b411b363 | 4994 | } |
e05e1e59 | 4995 | switch (pi->cmd) { |
b411b363 | 4996 | case P_RS_WRITE_ACK: |
8554df1c | 4997 | what = WRITE_ACKED_BY_PEER_AND_SIS; |
b411b363 PR |
4998 | break; |
4999 | case P_WRITE_ACK: | |
8554df1c | 5000 | what = WRITE_ACKED_BY_PEER; |
b411b363 PR |
5001 | break; |
5002 | case P_RECV_ACK: | |
8554df1c | 5003 | what = RECV_ACKED_BY_PEER; |
b411b363 | 5004 | break; |
d4dabbe2 LE |
5005 | case P_SUPERSEDED: |
5006 | what = CONFLICT_RESOLVED; | |
b411b363 | 5007 | break; |
7be8da07 | 5008 | case P_RETRY_WRITE: |
7be8da07 | 5009 | what = POSTPONE_WRITE; |
b411b363 PR |
5010 | break; |
5011 | default: | |
2735a594 | 5012 | BUG(); |
b411b363 PR |
5013 | } |
5014 | ||
5015 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
2735a594 AG |
5016 | &mdev->write_requests, __func__, |
5017 | what, false); | |
b411b363 PR |
5018 | } |
5019 | ||
1952e916 | 5020 | static int got_NegAck(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 5021 | { |
1952e916 | 5022 | struct drbd_conf *mdev; |
e658983a | 5023 | struct p_block_ack *p = pi->data; |
b411b363 | 5024 | sector_t sector = be64_to_cpu(p->sector); |
2deb8336 | 5025 | int size = be32_to_cpu(p->blksize); |
85997675 | 5026 | int err; |
b411b363 | 5027 | |
1952e916 AG |
5028 | mdev = vnr_to_mdev(tconn, pi->vnr); |
5029 | if (!mdev) | |
2735a594 | 5030 | return -EIO; |
b411b363 PR |
5031 | |
5032 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
5033 | ||
579b57ed | 5034 | if (p->block_id == ID_SYNCER) { |
b411b363 PR |
5035 | dec_rs_pending(mdev); |
5036 | drbd_rs_failed_io(mdev, sector, size); | |
2735a594 | 5037 | return 0; |
b411b363 | 5038 | } |
2deb8336 | 5039 | |
85997675 AG |
5040 | err = validate_req_change_req_state(mdev, p->block_id, sector, |
5041 | &mdev->write_requests, __func__, | |
303d1448 | 5042 | NEG_ACKED, true); |
85997675 | 5043 | if (err) { |
c3afd8f5 AG |
5044 | /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs. |
5045 | The master bio might already be completed, therefore the | |
5046 | request is no longer in the collision hash. */ | |
5047 | /* In Protocol B we might already have got a P_RECV_ACK | |
5048 | but then get a P_NEG_ACK afterwards. */ | |
c3afd8f5 | 5049 | drbd_set_out_of_sync(mdev, sector, size); |
2deb8336 | 5050 | } |
2735a594 | 5051 | return 0; |
b411b363 PR |
5052 | } |
5053 | ||
1952e916 | 5054 | static int got_NegDReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 5055 | { |
1952e916 | 5056 | struct drbd_conf *mdev; |
e658983a | 5057 | struct p_block_ack *p = pi->data; |
b411b363 PR |
5058 | sector_t sector = be64_to_cpu(p->sector); |
5059 | ||
1952e916 AG |
5060 | mdev = vnr_to_mdev(tconn, pi->vnr); |
5061 | if (!mdev) | |
2735a594 | 5062 | return -EIO; |
1952e916 | 5063 | |
b411b363 | 5064 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); |
7be8da07 | 5065 | |
380207d0 | 5066 | dev_err(DEV, "Got NegDReply; Sector %llus, len %u.\n", |
b411b363 PR |
5067 | (unsigned long long)sector, be32_to_cpu(p->blksize)); |
5068 | ||
5069 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
2735a594 AG |
5070 | &mdev->read_requests, __func__, |
5071 | NEG_ACKED, false); | |
b411b363 PR |
5072 | } |
5073 | ||
1952e916 | 5074 | static int got_NegRSDReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 5075 | { |
1952e916 | 5076 | struct drbd_conf *mdev; |
b411b363 PR |
5077 | sector_t sector; |
5078 | int size; | |
e658983a | 5079 | struct p_block_ack *p = pi->data; |
1952e916 AG |
5080 | |
5081 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
5082 | if (!mdev) | |
2735a594 | 5083 | return -EIO; |
b411b363 PR |
5084 | |
5085 | sector = be64_to_cpu(p->sector); | |
5086 | size = be32_to_cpu(p->blksize); | |
b411b363 PR |
5087 | |
5088 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
5089 | ||
5090 | dec_rs_pending(mdev); | |
5091 | ||
5092 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
5093 | drbd_rs_complete_io(mdev, sector); | |
e05e1e59 | 5094 | switch (pi->cmd) { |
d612d309 PR |
5095 | case P_NEG_RS_DREPLY: |
5096 | drbd_rs_failed_io(mdev, sector, size); | |
5097 | case P_RS_CANCEL: | |
5098 | break; | |
5099 | default: | |
2735a594 | 5100 | BUG(); |
d612d309 | 5101 | } |
b411b363 PR |
5102 | put_ldev(mdev); |
5103 | } | |
5104 | ||
2735a594 | 5105 | return 0; |
b411b363 PR |
5106 | } |
5107 | ||
1952e916 | 5108 | static int got_BarrierAck(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 5109 | { |
e658983a | 5110 | struct p_barrier_ack *p = pi->data; |
9ed57dcb LE |
5111 | struct drbd_conf *mdev; |
5112 | int vnr; | |
1952e916 | 5113 | |
9ed57dcb | 5114 | tl_release(tconn, p->barrier, be32_to_cpu(p->set_size)); |
b411b363 | 5115 | |
9ed57dcb LE |
5116 | rcu_read_lock(); |
5117 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { | |
5118 | if (mdev->state.conn == C_AHEAD && | |
5119 | atomic_read(&mdev->ap_in_flight) == 0 && | |
5120 | !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &mdev->flags)) { | |
5121 | mdev->start_resync_timer.expires = jiffies + HZ; | |
5122 | add_timer(&mdev->start_resync_timer); | |
5123 | } | |
c4752ef1 | 5124 | } |
9ed57dcb | 5125 | rcu_read_unlock(); |
c4752ef1 | 5126 | |
2735a594 | 5127 | return 0; |
b411b363 PR |
5128 | } |
5129 | ||
1952e916 | 5130 | static int got_OVResult(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 5131 | { |
1952e916 | 5132 | struct drbd_conf *mdev; |
e658983a | 5133 | struct p_block_ack *p = pi->data; |
b411b363 PR |
5134 | struct drbd_work *w; |
5135 | sector_t sector; | |
5136 | int size; | |
5137 | ||
1952e916 AG |
5138 | mdev = vnr_to_mdev(tconn, pi->vnr); |
5139 | if (!mdev) | |
2735a594 | 5140 | return -EIO; |
1952e916 | 5141 | |
b411b363 PR |
5142 | sector = be64_to_cpu(p->sector); |
5143 | size = be32_to_cpu(p->blksize); | |
5144 | ||
5145 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
5146 | ||
5147 | if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) | |
8f7bed77 | 5148 | drbd_ov_out_of_sync_found(mdev, sector, size); |
b411b363 | 5149 | else |
8f7bed77 | 5150 | ov_out_of_sync_print(mdev); |
b411b363 | 5151 | |
1d53f09e | 5152 | if (!get_ldev(mdev)) |
2735a594 | 5153 | return 0; |
1d53f09e | 5154 | |
b411b363 PR |
5155 | drbd_rs_complete_io(mdev, sector); |
5156 | dec_rs_pending(mdev); | |
5157 | ||
ea5442af LE |
5158 | --mdev->ov_left; |
5159 | ||
5160 | /* let's advance progress step marks only for every other megabyte */ | |
5161 | if ((mdev->ov_left & 0x200) == 0x200) | |
5162 | drbd_advance_rs_marks(mdev, mdev->ov_left); | |
5163 | ||
5164 | if (mdev->ov_left == 0) { | |
b411b363 PR |
5165 | w = kmalloc(sizeof(*w), GFP_NOIO); |
5166 | if (w) { | |
5167 | w->cb = w_ov_finished; | |
a21e9298 | 5168 | w->mdev = mdev; |
d5b27b01 | 5169 | drbd_queue_work(&mdev->tconn->sender_work, w); |
b411b363 PR |
5170 | } else { |
5171 | dev_err(DEV, "kmalloc(w) failed."); | |
8f7bed77 | 5172 | ov_out_of_sync_print(mdev); |
b411b363 PR |
5173 | drbd_resync_finished(mdev); |
5174 | } | |
5175 | } | |
1d53f09e | 5176 | put_ldev(mdev); |
2735a594 | 5177 | return 0; |
b411b363 PR |
5178 | } |
5179 | ||
1952e916 | 5180 | static int got_skip(struct drbd_tconn *tconn, struct packet_info *pi) |
0ced55a3 | 5181 | { |
2735a594 | 5182 | return 0; |
b411b363 PR |
5183 | } |
5184 | ||
a990be46 | 5185 | static int tconn_finish_peer_reqs(struct drbd_tconn *tconn) |
0ced55a3 | 5186 | { |
082a3439 | 5187 | struct drbd_conf *mdev; |
c141ebda | 5188 | int vnr, not_empty = 0; |
32862ec7 PR |
5189 | |
5190 | do { | |
5191 | clear_bit(SIGNAL_ASENDER, &tconn->flags); | |
5192 | flush_signals(current); | |
c141ebda PR |
5193 | |
5194 | rcu_read_lock(); | |
5195 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { | |
5196 | kref_get(&mdev->kref); | |
5197 | rcu_read_unlock(); | |
d3fcb490 | 5198 | if (drbd_finish_peer_reqs(mdev)) { |
c141ebda PR |
5199 | kref_put(&mdev->kref, &drbd_minor_destroy); |
5200 | return 1; | |
d3fcb490 | 5201 | } |
c141ebda PR |
5202 | kref_put(&mdev->kref, &drbd_minor_destroy); |
5203 | rcu_read_lock(); | |
082a3439 | 5204 | } |
32862ec7 | 5205 | set_bit(SIGNAL_ASENDER, &tconn->flags); |
082a3439 PR |
5206 | |
5207 | spin_lock_irq(&tconn->req_lock); | |
c141ebda | 5208 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
082a3439 PR |
5209 | not_empty = !list_empty(&mdev->done_ee); |
5210 | if (not_empty) | |
5211 | break; | |
5212 | } | |
5213 | spin_unlock_irq(&tconn->req_lock); | |
c141ebda | 5214 | rcu_read_unlock(); |
32862ec7 PR |
5215 | } while (not_empty); |
5216 | ||
5217 | return 0; | |
0ced55a3 PR |
5218 | } |
5219 | ||
b411b363 PR |
5220 | struct asender_cmd { |
5221 | size_t pkt_size; | |
1952e916 | 5222 | int (*fn)(struct drbd_tconn *tconn, struct packet_info *); |
b411b363 PR |
5223 | }; |
5224 | ||
7201b972 | 5225 | static struct asender_cmd asender_tbl[] = { |
e658983a AG |
5226 | [P_PING] = { 0, got_Ping }, |
5227 | [P_PING_ACK] = { 0, got_PingAck }, | |
b411b363 PR |
5228 | [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, |
5229 | [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
5230 | [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
d4dabbe2 | 5231 | [P_SUPERSEDED] = { sizeof(struct p_block_ack), got_BlockAck }, |
b411b363 PR |
5232 | [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, |
5233 | [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, | |
1952e916 | 5234 | [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply }, |
b411b363 PR |
5235 | [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, |
5236 | [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, | |
5237 | [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, | |
5238 | [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, | |
02918be2 | 5239 | [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, |
1952e916 AG |
5240 | [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply }, |
5241 | [P_CONN_ST_CHG_REPLY]={ sizeof(struct p_req_state_reply), got_conn_RqSReply }, | |
5242 | [P_RETRY_WRITE] = { sizeof(struct p_block_ack), got_BlockAck }, | |
7201b972 | 5243 | }; |
b411b363 PR |
5244 | |
5245 | int drbd_asender(struct drbd_thread *thi) | |
5246 | { | |
392c8801 | 5247 | struct drbd_tconn *tconn = thi->tconn; |
b411b363 | 5248 | struct asender_cmd *cmd = NULL; |
77351055 | 5249 | struct packet_info pi; |
257d0af6 | 5250 | int rv; |
e658983a | 5251 | void *buf = tconn->meta.rbuf; |
b411b363 | 5252 | int received = 0; |
52b061a4 AG |
5253 | unsigned int header_size = drbd_header_size(tconn); |
5254 | int expect = header_size; | |
44ed167d PR |
5255 | bool ping_timeout_active = false; |
5256 | struct net_conf *nc; | |
bb77d34e | 5257 | int ping_timeo, tcp_cork, ping_int; |
3990e04d | 5258 | struct sched_param param = { .sched_priority = 2 }; |
b411b363 | 5259 | |
3990e04d PR |
5260 | rv = sched_setscheduler(current, SCHED_RR, ¶m); |
5261 | if (rv < 0) | |
5262 | conn_err(tconn, "drbd_asender: ERROR set priority, ret=%d\n", rv); | |
b411b363 | 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 | } |