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 | ||
7e5fec31 | 28 | #include <linux/uaccess.h> |
b411b363 PR |
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> | |
ae7e81c0 | 39 | #include <uapi/linux/sched/types.h> |
174cd4b1 | 40 | #include <linux/sched/signal.h> |
b411b363 PR |
41 | #include <linux/pkt_sched.h> |
42 | #define __KERNEL_SYSCALLS__ | |
43 | #include <linux/unistd.h> | |
44 | #include <linux/vmalloc.h> | |
45 | #include <linux/random.h> | |
b411b363 PR |
46 | #include <linux/string.h> |
47 | #include <linux/scatterlist.h> | |
48 | #include "drbd_int.h" | |
a3603a6e | 49 | #include "drbd_protocol.h" |
b411b363 | 50 | #include "drbd_req.h" |
b411b363 PR |
51 | #include "drbd_vli.h" |
52 | ||
9104d31a | 53 | #define PRO_FEATURES (DRBD_FF_TRIM|DRBD_FF_THIN_RESYNC|DRBD_FF_WSAME) |
20c68fde | 54 | |
77351055 PR |
55 | struct packet_info { |
56 | enum drbd_packet cmd; | |
e2857216 AG |
57 | unsigned int size; |
58 | unsigned int vnr; | |
e658983a | 59 | void *data; |
77351055 PR |
60 | }; |
61 | ||
b411b363 PR |
62 | enum finish_epoch { |
63 | FE_STILL_LIVE, | |
64 | FE_DESTROYED, | |
65 | FE_RECYCLED, | |
66 | }; | |
67 | ||
bde89a9e AG |
68 | static int drbd_do_features(struct drbd_connection *connection); |
69 | static int drbd_do_auth(struct drbd_connection *connection); | |
69a22773 | 70 | static int drbd_disconnected(struct drbd_peer_device *); |
a0fb3c47 | 71 | static void conn_wait_active_ee_empty(struct drbd_connection *connection); |
bde89a9e | 72 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_connection *, struct drbd_epoch *, enum epoch_event); |
99920dc5 | 73 | static int e_end_block(struct drbd_work *, int); |
b411b363 | 74 | |
b411b363 PR |
75 | |
76 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
77 | ||
45bb912b LE |
78 | /* |
79 | * some helper functions to deal with single linked page lists, | |
80 | * page->private being our "next" pointer. | |
81 | */ | |
82 | ||
83 | /* If at least n pages are linked at head, get n pages off. | |
84 | * Otherwise, don't modify head, and return NULL. | |
85 | * Locking is the responsibility of the caller. | |
86 | */ | |
87 | static struct page *page_chain_del(struct page **head, int n) | |
88 | { | |
89 | struct page *page; | |
90 | struct page *tmp; | |
91 | ||
92 | BUG_ON(!n); | |
93 | BUG_ON(!head); | |
94 | ||
95 | page = *head; | |
23ce4227 PR |
96 | |
97 | if (!page) | |
98 | return NULL; | |
99 | ||
45bb912b LE |
100 | while (page) { |
101 | tmp = page_chain_next(page); | |
102 | if (--n == 0) | |
103 | break; /* found sufficient pages */ | |
104 | if (tmp == NULL) | |
105 | /* insufficient pages, don't use any of them. */ | |
106 | return NULL; | |
107 | page = tmp; | |
108 | } | |
109 | ||
110 | /* add end of list marker for the returned list */ | |
111 | set_page_private(page, 0); | |
112 | /* actual return value, and adjustment of head */ | |
113 | page = *head; | |
114 | *head = tmp; | |
115 | return page; | |
116 | } | |
117 | ||
118 | /* may be used outside of locks to find the tail of a (usually short) | |
119 | * "private" page chain, before adding it back to a global chain head | |
120 | * with page_chain_add() under a spinlock. */ | |
121 | static struct page *page_chain_tail(struct page *page, int *len) | |
122 | { | |
123 | struct page *tmp; | |
124 | int i = 1; | |
125 | while ((tmp = page_chain_next(page))) | |
126 | ++i, page = tmp; | |
127 | if (len) | |
128 | *len = i; | |
129 | return page; | |
130 | } | |
131 | ||
132 | static int page_chain_free(struct page *page) | |
133 | { | |
134 | struct page *tmp; | |
135 | int i = 0; | |
136 | page_chain_for_each_safe(page, tmp) { | |
137 | put_page(page); | |
138 | ++i; | |
139 | } | |
140 | return i; | |
141 | } | |
142 | ||
143 | static void page_chain_add(struct page **head, | |
144 | struct page *chain_first, struct page *chain_last) | |
145 | { | |
146 | #if 1 | |
147 | struct page *tmp; | |
148 | tmp = page_chain_tail(chain_first, NULL); | |
149 | BUG_ON(tmp != chain_last); | |
150 | #endif | |
151 | ||
152 | /* add chain to head */ | |
153 | set_page_private(chain_last, (unsigned long)*head); | |
154 | *head = chain_first; | |
155 | } | |
156 | ||
b30ab791 | 157 | static struct page *__drbd_alloc_pages(struct drbd_device *device, |
18c2d522 | 158 | unsigned int number) |
b411b363 PR |
159 | { |
160 | struct page *page = NULL; | |
45bb912b | 161 | struct page *tmp = NULL; |
18c2d522 | 162 | unsigned int i = 0; |
b411b363 PR |
163 | |
164 | /* Yes, testing drbd_pp_vacant outside the lock is racy. | |
165 | * So what. It saves a spin_lock. */ | |
45bb912b | 166 | if (drbd_pp_vacant >= number) { |
b411b363 | 167 | spin_lock(&drbd_pp_lock); |
45bb912b LE |
168 | page = page_chain_del(&drbd_pp_pool, number); |
169 | if (page) | |
170 | drbd_pp_vacant -= number; | |
b411b363 | 171 | spin_unlock(&drbd_pp_lock); |
45bb912b LE |
172 | if (page) |
173 | return page; | |
b411b363 | 174 | } |
45bb912b | 175 | |
b411b363 PR |
176 | /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD |
177 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
178 | * which in turn might block on the other node at this very place. */ | |
45bb912b LE |
179 | for (i = 0; i < number; i++) { |
180 | tmp = alloc_page(GFP_TRY); | |
181 | if (!tmp) | |
182 | break; | |
183 | set_page_private(tmp, (unsigned long)page); | |
184 | page = tmp; | |
185 | } | |
186 | ||
187 | if (i == number) | |
188 | return page; | |
189 | ||
190 | /* Not enough pages immediately available this time. | |
c37c8ecf | 191 | * No need to jump around here, drbd_alloc_pages will retry this |
45bb912b LE |
192 | * function "soon". */ |
193 | if (page) { | |
194 | tmp = page_chain_tail(page, NULL); | |
195 | spin_lock(&drbd_pp_lock); | |
196 | page_chain_add(&drbd_pp_pool, page, tmp); | |
197 | drbd_pp_vacant += i; | |
198 | spin_unlock(&drbd_pp_lock); | |
199 | } | |
200 | return NULL; | |
b411b363 PR |
201 | } |
202 | ||
b30ab791 | 203 | static void reclaim_finished_net_peer_reqs(struct drbd_device *device, |
a990be46 | 204 | struct list_head *to_be_freed) |
b411b363 | 205 | { |
a8cd15ba | 206 | struct drbd_peer_request *peer_req, *tmp; |
b411b363 PR |
207 | |
208 | /* The EEs are always appended to the end of the list. Since | |
209 | they are sent in order over the wire, they have to finish | |
210 | in order. As soon as we see the first not finished we can | |
211 | stop to examine the list... */ | |
212 | ||
a8cd15ba | 213 | list_for_each_entry_safe(peer_req, tmp, &device->net_ee, w.list) { |
045417f7 | 214 | if (drbd_peer_req_has_active_page(peer_req)) |
b411b363 | 215 | break; |
a8cd15ba | 216 | list_move(&peer_req->w.list, to_be_freed); |
b411b363 PR |
217 | } |
218 | } | |
219 | ||
668700b4 | 220 | static void drbd_reclaim_net_peer_reqs(struct drbd_device *device) |
b411b363 PR |
221 | { |
222 | LIST_HEAD(reclaimed); | |
db830c46 | 223 | struct drbd_peer_request *peer_req, *t; |
b411b363 | 224 | |
0500813f | 225 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 226 | reclaim_finished_net_peer_reqs(device, &reclaimed); |
0500813f | 227 | spin_unlock_irq(&device->resource->req_lock); |
a8cd15ba | 228 | list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) |
b30ab791 | 229 | drbd_free_net_peer_req(device, peer_req); |
b411b363 PR |
230 | } |
231 | ||
668700b4 PR |
232 | static void conn_reclaim_net_peer_reqs(struct drbd_connection *connection) |
233 | { | |
234 | struct drbd_peer_device *peer_device; | |
235 | int vnr; | |
236 | ||
237 | rcu_read_lock(); | |
238 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
239 | struct drbd_device *device = peer_device->device; | |
240 | if (!atomic_read(&device->pp_in_use_by_net)) | |
241 | continue; | |
242 | ||
243 | kref_get(&device->kref); | |
244 | rcu_read_unlock(); | |
245 | drbd_reclaim_net_peer_reqs(device); | |
246 | kref_put(&device->kref, drbd_destroy_device); | |
247 | rcu_read_lock(); | |
248 | } | |
249 | rcu_read_unlock(); | |
250 | } | |
251 | ||
b411b363 | 252 | /** |
c37c8ecf | 253 | * drbd_alloc_pages() - Returns @number pages, retries forever (or until signalled) |
b30ab791 | 254 | * @device: DRBD device. |
45bb912b LE |
255 | * @number: number of pages requested |
256 | * @retry: whether to retry, if not enough pages are available right now | |
257 | * | |
258 | * Tries to allocate number pages, first from our own page pool, then from | |
0e49d7b0 | 259 | * the kernel. |
45bb912b | 260 | * Possibly retry until DRBD frees sufficient pages somewhere else. |
b411b363 | 261 | * |
0e49d7b0 LE |
262 | * If this allocation would exceed the max_buffers setting, we throttle |
263 | * allocation (schedule_timeout) to give the system some room to breathe. | |
264 | * | |
265 | * We do not use max-buffers as hard limit, because it could lead to | |
266 | * congestion and further to a distributed deadlock during online-verify or | |
267 | * (checksum based) resync, if the max-buffers, socket buffer sizes and | |
268 | * resync-rate settings are mis-configured. | |
269 | * | |
45bb912b | 270 | * Returns a page chain linked via page->private. |
b411b363 | 271 | */ |
69a22773 | 272 | struct page *drbd_alloc_pages(struct drbd_peer_device *peer_device, unsigned int number, |
c37c8ecf | 273 | bool retry) |
b411b363 | 274 | { |
69a22773 | 275 | struct drbd_device *device = peer_device->device; |
b411b363 | 276 | struct page *page = NULL; |
44ed167d | 277 | struct net_conf *nc; |
b411b363 | 278 | DEFINE_WAIT(wait); |
0e49d7b0 | 279 | unsigned int mxb; |
b411b363 | 280 | |
44ed167d | 281 | rcu_read_lock(); |
69a22773 | 282 | nc = rcu_dereference(peer_device->connection->net_conf); |
44ed167d PR |
283 | mxb = nc ? nc->max_buffers : 1000000; |
284 | rcu_read_unlock(); | |
285 | ||
b30ab791 AG |
286 | if (atomic_read(&device->pp_in_use) < mxb) |
287 | page = __drbd_alloc_pages(device, number); | |
b411b363 | 288 | |
668700b4 PR |
289 | /* Try to keep the fast path fast, but occasionally we need |
290 | * to reclaim the pages we lended to the network stack. */ | |
291 | if (page && atomic_read(&device->pp_in_use_by_net) > 512) | |
292 | drbd_reclaim_net_peer_reqs(device); | |
293 | ||
45bb912b | 294 | while (page == NULL) { |
b411b363 PR |
295 | prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE); |
296 | ||
668700b4 | 297 | drbd_reclaim_net_peer_reqs(device); |
b411b363 | 298 | |
b30ab791 AG |
299 | if (atomic_read(&device->pp_in_use) < mxb) { |
300 | page = __drbd_alloc_pages(device, number); | |
b411b363 PR |
301 | if (page) |
302 | break; | |
303 | } | |
304 | ||
305 | if (!retry) | |
306 | break; | |
307 | ||
308 | if (signal_pending(current)) { | |
d0180171 | 309 | drbd_warn(device, "drbd_alloc_pages interrupted!\n"); |
b411b363 PR |
310 | break; |
311 | } | |
312 | ||
0e49d7b0 LE |
313 | if (schedule_timeout(HZ/10) == 0) |
314 | mxb = UINT_MAX; | |
b411b363 PR |
315 | } |
316 | finish_wait(&drbd_pp_wait, &wait); | |
317 | ||
45bb912b | 318 | if (page) |
b30ab791 | 319 | atomic_add(number, &device->pp_in_use); |
b411b363 PR |
320 | return page; |
321 | } | |
322 | ||
c37c8ecf | 323 | /* Must not be used from irq, as that may deadlock: see drbd_alloc_pages. |
0500813f | 324 | * Is also used from inside an other spin_lock_irq(&resource->req_lock); |
45bb912b LE |
325 | * Either links the page chain back to the global pool, |
326 | * or returns all pages to the system. */ | |
b30ab791 | 327 | static void drbd_free_pages(struct drbd_device *device, struct page *page, int is_net) |
b411b363 | 328 | { |
b30ab791 | 329 | atomic_t *a = is_net ? &device->pp_in_use_by_net : &device->pp_in_use; |
b411b363 | 330 | int i; |
435f0740 | 331 | |
a73ff323 LE |
332 | if (page == NULL) |
333 | return; | |
334 | ||
183ece30 | 335 | if (drbd_pp_vacant > (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count) |
45bb912b LE |
336 | i = page_chain_free(page); |
337 | else { | |
338 | struct page *tmp; | |
339 | tmp = page_chain_tail(page, &i); | |
340 | spin_lock(&drbd_pp_lock); | |
341 | page_chain_add(&drbd_pp_pool, page, tmp); | |
342 | drbd_pp_vacant += i; | |
343 | spin_unlock(&drbd_pp_lock); | |
b411b363 | 344 | } |
435f0740 | 345 | i = atomic_sub_return(i, a); |
45bb912b | 346 | if (i < 0) |
d0180171 | 347 | drbd_warn(device, "ASSERTION FAILED: %s: %d < 0\n", |
435f0740 | 348 | is_net ? "pp_in_use_by_net" : "pp_in_use", i); |
b411b363 PR |
349 | wake_up(&drbd_pp_wait); |
350 | } | |
351 | ||
352 | /* | |
353 | You need to hold the req_lock: | |
354 | _drbd_wait_ee_list_empty() | |
355 | ||
356 | You must not have the req_lock: | |
3967deb1 | 357 | drbd_free_peer_req() |
0db55363 | 358 | drbd_alloc_peer_req() |
7721f567 | 359 | drbd_free_peer_reqs() |
b411b363 | 360 | drbd_ee_fix_bhs() |
a990be46 | 361 | drbd_finish_peer_reqs() |
b411b363 PR |
362 | drbd_clear_done_ee() |
363 | drbd_wait_ee_list_empty() | |
364 | */ | |
365 | ||
9104d31a LE |
366 | /* normal: payload_size == request size (bi_size) |
367 | * w_same: payload_size == logical_block_size | |
368 | * trim: payload_size == 0 */ | |
f6ffca9f | 369 | struct drbd_peer_request * |
69a22773 | 370 | drbd_alloc_peer_req(struct drbd_peer_device *peer_device, u64 id, sector_t sector, |
9104d31a | 371 | unsigned int request_size, unsigned int payload_size, gfp_t gfp_mask) __must_hold(local) |
b411b363 | 372 | { |
69a22773 | 373 | struct drbd_device *device = peer_device->device; |
db830c46 | 374 | struct drbd_peer_request *peer_req; |
a73ff323 | 375 | struct page *page = NULL; |
9104d31a | 376 | unsigned nr_pages = (payload_size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b411b363 | 377 | |
b30ab791 | 378 | if (drbd_insert_fault(device, DRBD_FAULT_AL_EE)) |
b411b363 PR |
379 | return NULL; |
380 | ||
0892fac8 | 381 | peer_req = mempool_alloc(&drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM); |
db830c46 | 382 | if (!peer_req) { |
b411b363 | 383 | if (!(gfp_mask & __GFP_NOWARN)) |
d0180171 | 384 | drbd_err(device, "%s: allocation failed\n", __func__); |
b411b363 PR |
385 | return NULL; |
386 | } | |
387 | ||
9104d31a | 388 | if (nr_pages) { |
d0164adc MG |
389 | page = drbd_alloc_pages(peer_device, nr_pages, |
390 | gfpflags_allow_blocking(gfp_mask)); | |
a73ff323 LE |
391 | if (!page) |
392 | goto fail; | |
393 | } | |
b411b363 | 394 | |
c5a2c150 LE |
395 | memset(peer_req, 0, sizeof(*peer_req)); |
396 | INIT_LIST_HEAD(&peer_req->w.list); | |
db830c46 | 397 | drbd_clear_interval(&peer_req->i); |
9104d31a | 398 | peer_req->i.size = request_size; |
db830c46 | 399 | peer_req->i.sector = sector; |
c5a2c150 | 400 | peer_req->submit_jif = jiffies; |
a8cd15ba | 401 | peer_req->peer_device = peer_device; |
db830c46 | 402 | peer_req->pages = page; |
9a8e7753 AG |
403 | /* |
404 | * The block_id is opaque to the receiver. It is not endianness | |
405 | * converted, and sent back to the sender unchanged. | |
406 | */ | |
db830c46 | 407 | peer_req->block_id = id; |
b411b363 | 408 | |
db830c46 | 409 | return peer_req; |
b411b363 | 410 | |
45bb912b | 411 | fail: |
0892fac8 | 412 | mempool_free(peer_req, &drbd_ee_mempool); |
b411b363 PR |
413 | return NULL; |
414 | } | |
415 | ||
b30ab791 | 416 | void __drbd_free_peer_req(struct drbd_device *device, struct drbd_peer_request *peer_req, |
f6ffca9f | 417 | int is_net) |
b411b363 | 418 | { |
21ae5d7f | 419 | might_sleep(); |
db830c46 AG |
420 | if (peer_req->flags & EE_HAS_DIGEST) |
421 | kfree(peer_req->digest); | |
b30ab791 | 422 | drbd_free_pages(device, peer_req->pages, is_net); |
0b0ba1ef AG |
423 | D_ASSERT(device, atomic_read(&peer_req->pending_bios) == 0); |
424 | D_ASSERT(device, drbd_interval_empty(&peer_req->i)); | |
21ae5d7f LE |
425 | if (!expect(!(peer_req->flags & EE_CALL_AL_COMPLETE_IO))) { |
426 | peer_req->flags &= ~EE_CALL_AL_COMPLETE_IO; | |
427 | drbd_al_complete_io(device, &peer_req->i); | |
428 | } | |
0892fac8 | 429 | mempool_free(peer_req, &drbd_ee_mempool); |
b411b363 PR |
430 | } |
431 | ||
b30ab791 | 432 | int drbd_free_peer_reqs(struct drbd_device *device, struct list_head *list) |
b411b363 PR |
433 | { |
434 | LIST_HEAD(work_list); | |
db830c46 | 435 | struct drbd_peer_request *peer_req, *t; |
b411b363 | 436 | int count = 0; |
b30ab791 | 437 | int is_net = list == &device->net_ee; |
b411b363 | 438 | |
0500813f | 439 | spin_lock_irq(&device->resource->req_lock); |
b411b363 | 440 | list_splice_init(list, &work_list); |
0500813f | 441 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 442 | |
a8cd15ba | 443 | list_for_each_entry_safe(peer_req, t, &work_list, w.list) { |
b30ab791 | 444 | __drbd_free_peer_req(device, peer_req, is_net); |
b411b363 PR |
445 | count++; |
446 | } | |
447 | return count; | |
448 | } | |
449 | ||
b411b363 | 450 | /* |
a990be46 | 451 | * See also comments in _req_mod(,BARRIER_ACKED) and receive_Barrier. |
b411b363 | 452 | */ |
b30ab791 | 453 | static int drbd_finish_peer_reqs(struct drbd_device *device) |
b411b363 PR |
454 | { |
455 | LIST_HEAD(work_list); | |
456 | LIST_HEAD(reclaimed); | |
db830c46 | 457 | struct drbd_peer_request *peer_req, *t; |
e2b3032b | 458 | int err = 0; |
b411b363 | 459 | |
0500813f | 460 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 AG |
461 | reclaim_finished_net_peer_reqs(device, &reclaimed); |
462 | list_splice_init(&device->done_ee, &work_list); | |
0500813f | 463 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 464 | |
a8cd15ba | 465 | list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) |
b30ab791 | 466 | drbd_free_net_peer_req(device, peer_req); |
b411b363 PR |
467 | |
468 | /* possible callbacks here: | |
d4dabbe2 | 469 | * e_end_block, and e_end_resync_block, e_send_superseded. |
b411b363 PR |
470 | * all ignore the last argument. |
471 | */ | |
a8cd15ba | 472 | list_for_each_entry_safe(peer_req, t, &work_list, w.list) { |
e2b3032b AG |
473 | int err2; |
474 | ||
b411b363 | 475 | /* list_del not necessary, next/prev members not touched */ |
a8cd15ba | 476 | err2 = peer_req->w.cb(&peer_req->w, !!err); |
e2b3032b AG |
477 | if (!err) |
478 | err = err2; | |
b30ab791 | 479 | drbd_free_peer_req(device, peer_req); |
b411b363 | 480 | } |
b30ab791 | 481 | wake_up(&device->ee_wait); |
b411b363 | 482 | |
e2b3032b | 483 | return err; |
b411b363 PR |
484 | } |
485 | ||
b30ab791 | 486 | static void _drbd_wait_ee_list_empty(struct drbd_device *device, |
d4da1537 | 487 | struct list_head *head) |
b411b363 PR |
488 | { |
489 | DEFINE_WAIT(wait); | |
490 | ||
491 | /* avoids spin_lock/unlock | |
492 | * and calling prepare_to_wait in the fast path */ | |
493 | while (!list_empty(head)) { | |
b30ab791 | 494 | prepare_to_wait(&device->ee_wait, &wait, TASK_UNINTERRUPTIBLE); |
0500813f | 495 | spin_unlock_irq(&device->resource->req_lock); |
7eaceacc | 496 | io_schedule(); |
b30ab791 | 497 | finish_wait(&device->ee_wait, &wait); |
0500813f | 498 | spin_lock_irq(&device->resource->req_lock); |
b411b363 PR |
499 | } |
500 | } | |
501 | ||
b30ab791 | 502 | static void drbd_wait_ee_list_empty(struct drbd_device *device, |
d4da1537 | 503 | struct list_head *head) |
b411b363 | 504 | { |
0500813f | 505 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 506 | _drbd_wait_ee_list_empty(device, head); |
0500813f | 507 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
508 | } |
509 | ||
dbd9eea0 | 510 | static int drbd_recv_short(struct socket *sock, void *buf, size_t size, int flags) |
b411b363 | 511 | { |
b411b363 PR |
512 | struct kvec iov = { |
513 | .iov_base = buf, | |
514 | .iov_len = size, | |
515 | }; | |
516 | struct msghdr msg = { | |
b411b363 PR |
517 | .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) |
518 | }; | |
aa563d7b | 519 | iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, size); |
f7765c36 | 520 | return sock_recvmsg(sock, &msg, msg.msg_flags); |
b411b363 PR |
521 | } |
522 | ||
bde89a9e | 523 | static int drbd_recv(struct drbd_connection *connection, void *buf, size_t size) |
b411b363 | 524 | { |
b411b363 PR |
525 | int rv; |
526 | ||
bde89a9e | 527 | rv = drbd_recv_short(connection->data.socket, buf, size, 0); |
b411b363 | 528 | |
dbd0820c PR |
529 | if (rv < 0) { |
530 | if (rv == -ECONNRESET) | |
1ec861eb | 531 | drbd_info(connection, "sock was reset by peer\n"); |
dbd0820c | 532 | else if (rv != -ERESTARTSYS) |
1ec861eb | 533 | drbd_err(connection, "sock_recvmsg returned %d\n", rv); |
dbd0820c | 534 | } else if (rv == 0) { |
bde89a9e | 535 | if (test_bit(DISCONNECT_SENT, &connection->flags)) { |
b66623e3 PR |
536 | long t; |
537 | rcu_read_lock(); | |
bde89a9e | 538 | t = rcu_dereference(connection->net_conf)->ping_timeo * HZ/10; |
b66623e3 PR |
539 | rcu_read_unlock(); |
540 | ||
bde89a9e | 541 | t = wait_event_timeout(connection->ping_wait, connection->cstate < C_WF_REPORT_PARAMS, t); |
b66623e3 | 542 | |
599377ac PR |
543 | if (t) |
544 | goto out; | |
545 | } | |
1ec861eb | 546 | drbd_info(connection, "sock was shut down by peer\n"); |
599377ac PR |
547 | } |
548 | ||
b411b363 | 549 | if (rv != size) |
bde89a9e | 550 | conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 | 551 | |
599377ac | 552 | out: |
b411b363 PR |
553 | return rv; |
554 | } | |
555 | ||
bde89a9e | 556 | static int drbd_recv_all(struct drbd_connection *connection, void *buf, size_t size) |
c6967746 AG |
557 | { |
558 | int err; | |
559 | ||
bde89a9e | 560 | err = drbd_recv(connection, buf, size); |
c6967746 AG |
561 | if (err != size) { |
562 | if (err >= 0) | |
563 | err = -EIO; | |
564 | } else | |
565 | err = 0; | |
566 | return err; | |
567 | } | |
568 | ||
bde89a9e | 569 | static int drbd_recv_all_warn(struct drbd_connection *connection, void *buf, size_t size) |
a5c31904 AG |
570 | { |
571 | int err; | |
572 | ||
bde89a9e | 573 | err = drbd_recv_all(connection, buf, size); |
a5c31904 | 574 | if (err && !signal_pending(current)) |
1ec861eb | 575 | drbd_warn(connection, "short read (expected size %d)\n", (int)size); |
a5c31904 AG |
576 | return err; |
577 | } | |
578 | ||
5dbf1673 LE |
579 | /* quoting tcp(7): |
580 | * On individual connections, the socket buffer size must be set prior to the | |
581 | * listen(2) or connect(2) calls in order to have it take effect. | |
582 | * This is our wrapper to do so. | |
583 | */ | |
584 | static void drbd_setbufsize(struct socket *sock, unsigned int snd, | |
585 | unsigned int rcv) | |
586 | { | |
587 | /* open coded SO_SNDBUF, SO_RCVBUF */ | |
588 | if (snd) { | |
589 | sock->sk->sk_sndbuf = snd; | |
590 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
591 | } | |
592 | if (rcv) { | |
593 | sock->sk->sk_rcvbuf = rcv; | |
594 | sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
595 | } | |
596 | } | |
597 | ||
bde89a9e | 598 | static struct socket *drbd_try_connect(struct drbd_connection *connection) |
b411b363 PR |
599 | { |
600 | const char *what; | |
601 | struct socket *sock; | |
602 | struct sockaddr_in6 src_in6; | |
44ed167d PR |
603 | struct sockaddr_in6 peer_in6; |
604 | struct net_conf *nc; | |
605 | int err, peer_addr_len, my_addr_len; | |
69ef82de | 606 | int sndbuf_size, rcvbuf_size, connect_int; |
b411b363 PR |
607 | int disconnect_on_error = 1; |
608 | ||
44ed167d | 609 | rcu_read_lock(); |
bde89a9e | 610 | nc = rcu_dereference(connection->net_conf); |
44ed167d PR |
611 | if (!nc) { |
612 | rcu_read_unlock(); | |
b411b363 | 613 | return NULL; |
44ed167d | 614 | } |
44ed167d PR |
615 | sndbuf_size = nc->sndbuf_size; |
616 | rcvbuf_size = nc->rcvbuf_size; | |
69ef82de | 617 | connect_int = nc->connect_int; |
089c075d | 618 | rcu_read_unlock(); |
44ed167d | 619 | |
bde89a9e AG |
620 | my_addr_len = min_t(int, connection->my_addr_len, sizeof(src_in6)); |
621 | memcpy(&src_in6, &connection->my_addr, my_addr_len); | |
44ed167d | 622 | |
bde89a9e | 623 | if (((struct sockaddr *)&connection->my_addr)->sa_family == AF_INET6) |
44ed167d PR |
624 | src_in6.sin6_port = 0; |
625 | else | |
626 | ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ | |
627 | ||
bde89a9e AG |
628 | peer_addr_len = min_t(int, connection->peer_addr_len, sizeof(src_in6)); |
629 | memcpy(&peer_in6, &connection->peer_addr, peer_addr_len); | |
b411b363 PR |
630 | |
631 | what = "sock_create_kern"; | |
eeb1bd5c | 632 | err = sock_create_kern(&init_net, ((struct sockaddr *)&src_in6)->sa_family, |
44ed167d | 633 | SOCK_STREAM, IPPROTO_TCP, &sock); |
b411b363 PR |
634 | if (err < 0) { |
635 | sock = NULL; | |
636 | goto out; | |
637 | } | |
638 | ||
639 | sock->sk->sk_rcvtimeo = | |
69ef82de | 640 | sock->sk->sk_sndtimeo = connect_int * HZ; |
44ed167d | 641 | drbd_setbufsize(sock, sndbuf_size, rcvbuf_size); |
b411b363 PR |
642 | |
643 | /* explicitly bind to the configured IP as source IP | |
644 | * for the outgoing connections. | |
645 | * This is needed for multihomed hosts and to be | |
646 | * able to use lo: interfaces for drbd. | |
647 | * Make sure to use 0 as port number, so linux selects | |
648 | * a free one dynamically. | |
649 | */ | |
b411b363 | 650 | what = "bind before connect"; |
44ed167d | 651 | err = sock->ops->bind(sock, (struct sockaddr *) &src_in6, my_addr_len); |
b411b363 PR |
652 | if (err < 0) |
653 | goto out; | |
654 | ||
655 | /* connect may fail, peer not yet available. | |
656 | * stay C_WF_CONNECTION, don't go Disconnecting! */ | |
657 | disconnect_on_error = 0; | |
658 | what = "connect"; | |
44ed167d | 659 | err = sock->ops->connect(sock, (struct sockaddr *) &peer_in6, peer_addr_len, 0); |
b411b363 PR |
660 | |
661 | out: | |
662 | if (err < 0) { | |
663 | if (sock) { | |
664 | sock_release(sock); | |
665 | sock = NULL; | |
666 | } | |
667 | switch (-err) { | |
668 | /* timeout, busy, signal pending */ | |
669 | case ETIMEDOUT: case EAGAIN: case EINPROGRESS: | |
670 | case EINTR: case ERESTARTSYS: | |
671 | /* peer not (yet) available, network problem */ | |
672 | case ECONNREFUSED: case ENETUNREACH: | |
673 | case EHOSTDOWN: case EHOSTUNREACH: | |
674 | disconnect_on_error = 0; | |
675 | break; | |
676 | default: | |
1ec861eb | 677 | drbd_err(connection, "%s failed, err = %d\n", what, err); |
b411b363 PR |
678 | } |
679 | if (disconnect_on_error) | |
bde89a9e | 680 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 681 | } |
44ed167d | 682 | |
b411b363 PR |
683 | return sock; |
684 | } | |
685 | ||
7a426fd8 | 686 | struct accept_wait_data { |
bde89a9e | 687 | struct drbd_connection *connection; |
7a426fd8 PR |
688 | struct socket *s_listen; |
689 | struct completion door_bell; | |
690 | void (*original_sk_state_change)(struct sock *sk); | |
691 | ||
692 | }; | |
693 | ||
715306f6 | 694 | static void drbd_incoming_connection(struct sock *sk) |
7a426fd8 PR |
695 | { |
696 | struct accept_wait_data *ad = sk->sk_user_data; | |
715306f6 | 697 | void (*state_change)(struct sock *sk); |
7a426fd8 | 698 | |
715306f6 AG |
699 | state_change = ad->original_sk_state_change; |
700 | if (sk->sk_state == TCP_ESTABLISHED) | |
701 | complete(&ad->door_bell); | |
702 | state_change(sk); | |
7a426fd8 PR |
703 | } |
704 | ||
bde89a9e | 705 | static int prepare_listen_socket(struct drbd_connection *connection, struct accept_wait_data *ad) |
b411b363 | 706 | { |
1f3e509b | 707 | int err, sndbuf_size, rcvbuf_size, my_addr_len; |
44ed167d | 708 | struct sockaddr_in6 my_addr; |
1f3e509b | 709 | struct socket *s_listen; |
44ed167d | 710 | struct net_conf *nc; |
b411b363 PR |
711 | const char *what; |
712 | ||
44ed167d | 713 | rcu_read_lock(); |
bde89a9e | 714 | nc = rcu_dereference(connection->net_conf); |
44ed167d PR |
715 | if (!nc) { |
716 | rcu_read_unlock(); | |
7a426fd8 | 717 | return -EIO; |
44ed167d | 718 | } |
44ed167d PR |
719 | sndbuf_size = nc->sndbuf_size; |
720 | rcvbuf_size = nc->rcvbuf_size; | |
44ed167d | 721 | rcu_read_unlock(); |
b411b363 | 722 | |
bde89a9e AG |
723 | my_addr_len = min_t(int, connection->my_addr_len, sizeof(struct sockaddr_in6)); |
724 | memcpy(&my_addr, &connection->my_addr, my_addr_len); | |
b411b363 PR |
725 | |
726 | what = "sock_create_kern"; | |
eeb1bd5c | 727 | err = sock_create_kern(&init_net, ((struct sockaddr *)&my_addr)->sa_family, |
1f3e509b | 728 | SOCK_STREAM, IPPROTO_TCP, &s_listen); |
b411b363 PR |
729 | if (err) { |
730 | s_listen = NULL; | |
731 | goto out; | |
732 | } | |
733 | ||
98683650 | 734 | s_listen->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ |
44ed167d | 735 | drbd_setbufsize(s_listen, sndbuf_size, rcvbuf_size); |
b411b363 PR |
736 | |
737 | what = "bind before listen"; | |
44ed167d | 738 | err = s_listen->ops->bind(s_listen, (struct sockaddr *)&my_addr, my_addr_len); |
b411b363 PR |
739 | if (err < 0) |
740 | goto out; | |
741 | ||
7a426fd8 PR |
742 | ad->s_listen = s_listen; |
743 | write_lock_bh(&s_listen->sk->sk_callback_lock); | |
744 | ad->original_sk_state_change = s_listen->sk->sk_state_change; | |
715306f6 | 745 | s_listen->sk->sk_state_change = drbd_incoming_connection; |
7a426fd8 PR |
746 | s_listen->sk->sk_user_data = ad; |
747 | write_unlock_bh(&s_listen->sk->sk_callback_lock); | |
b411b363 | 748 | |
2820fd39 PR |
749 | what = "listen"; |
750 | err = s_listen->ops->listen(s_listen, 5); | |
751 | if (err < 0) | |
752 | goto out; | |
753 | ||
7a426fd8 | 754 | return 0; |
b411b363 PR |
755 | out: |
756 | if (s_listen) | |
757 | sock_release(s_listen); | |
758 | if (err < 0) { | |
759 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
1ec861eb | 760 | drbd_err(connection, "%s failed, err = %d\n", what, err); |
bde89a9e | 761 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
762 | } |
763 | } | |
b411b363 | 764 | |
7a426fd8 | 765 | return -EIO; |
b411b363 PR |
766 | } |
767 | ||
715306f6 | 768 | static void unregister_state_change(struct sock *sk, struct accept_wait_data *ad) |
b411b363 | 769 | { |
715306f6 AG |
770 | write_lock_bh(&sk->sk_callback_lock); |
771 | sk->sk_state_change = ad->original_sk_state_change; | |
772 | sk->sk_user_data = NULL; | |
773 | write_unlock_bh(&sk->sk_callback_lock); | |
b411b363 PR |
774 | } |
775 | ||
bde89a9e | 776 | static struct socket *drbd_wait_for_connect(struct drbd_connection *connection, struct accept_wait_data *ad) |
b411b363 | 777 | { |
1f3e509b PR |
778 | int timeo, connect_int, err = 0; |
779 | struct socket *s_estab = NULL; | |
1f3e509b PR |
780 | struct net_conf *nc; |
781 | ||
782 | rcu_read_lock(); | |
bde89a9e | 783 | nc = rcu_dereference(connection->net_conf); |
1f3e509b PR |
784 | if (!nc) { |
785 | rcu_read_unlock(); | |
786 | return NULL; | |
787 | } | |
788 | connect_int = nc->connect_int; | |
789 | rcu_read_unlock(); | |
790 | ||
791 | timeo = connect_int * HZ; | |
38b682b2 AM |
792 | /* 28.5% random jitter */ |
793 | timeo += (prandom_u32() & 1) ? timeo / 7 : -timeo / 7; | |
1f3e509b | 794 | |
7a426fd8 PR |
795 | err = wait_for_completion_interruptible_timeout(&ad->door_bell, timeo); |
796 | if (err <= 0) | |
797 | return NULL; | |
b411b363 | 798 | |
7a426fd8 | 799 | err = kernel_accept(ad->s_listen, &s_estab, 0); |
b411b363 PR |
800 | if (err < 0) { |
801 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
1ec861eb | 802 | drbd_err(connection, "accept failed, err = %d\n", err); |
bde89a9e | 803 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
804 | } |
805 | } | |
b411b363 | 806 | |
715306f6 AG |
807 | if (s_estab) |
808 | unregister_state_change(s_estab->sk, ad); | |
b411b363 | 809 | |
b411b363 PR |
810 | return s_estab; |
811 | } | |
b411b363 | 812 | |
bde89a9e | 813 | static int decode_header(struct drbd_connection *, void *, struct packet_info *); |
b411b363 | 814 | |
bde89a9e | 815 | static int send_first_packet(struct drbd_connection *connection, struct drbd_socket *sock, |
9f5bdc33 AG |
816 | enum drbd_packet cmd) |
817 | { | |
bde89a9e | 818 | if (!conn_prepare_command(connection, sock)) |
9f5bdc33 | 819 | return -EIO; |
bde89a9e | 820 | return conn_send_command(connection, sock, cmd, 0, NULL, 0); |
b411b363 PR |
821 | } |
822 | ||
bde89a9e | 823 | static int receive_first_packet(struct drbd_connection *connection, struct socket *sock) |
b411b363 | 824 | { |
bde89a9e | 825 | unsigned int header_size = drbd_header_size(connection); |
9f5bdc33 | 826 | struct packet_info pi; |
4920e37a | 827 | struct net_conf *nc; |
9f5bdc33 | 828 | int err; |
b411b363 | 829 | |
4920e37a PR |
830 | rcu_read_lock(); |
831 | nc = rcu_dereference(connection->net_conf); | |
832 | if (!nc) { | |
833 | rcu_read_unlock(); | |
834 | return -EIO; | |
835 | } | |
836 | sock->sk->sk_rcvtimeo = nc->ping_timeo * 4 * HZ / 10; | |
837 | rcu_read_unlock(); | |
838 | ||
bde89a9e | 839 | err = drbd_recv_short(sock, connection->data.rbuf, header_size, 0); |
9f5bdc33 AG |
840 | if (err != header_size) { |
841 | if (err >= 0) | |
842 | err = -EIO; | |
843 | return err; | |
844 | } | |
bde89a9e | 845 | err = decode_header(connection, connection->data.rbuf, &pi); |
9f5bdc33 AG |
846 | if (err) |
847 | return err; | |
848 | return pi.cmd; | |
b411b363 PR |
849 | } |
850 | ||
851 | /** | |
852 | * drbd_socket_okay() - Free the socket if its connection is not okay | |
b411b363 PR |
853 | * @sock: pointer to the pointer to the socket. |
854 | */ | |
5d0b17f1 | 855 | static bool drbd_socket_okay(struct socket **sock) |
b411b363 PR |
856 | { |
857 | int rr; | |
858 | char tb[4]; | |
859 | ||
860 | if (!*sock) | |
81e84650 | 861 | return false; |
b411b363 | 862 | |
dbd9eea0 | 863 | rr = drbd_recv_short(*sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); |
b411b363 PR |
864 | |
865 | if (rr > 0 || rr == -EAGAIN) { | |
81e84650 | 866 | return true; |
b411b363 PR |
867 | } else { |
868 | sock_release(*sock); | |
869 | *sock = NULL; | |
81e84650 | 870 | return false; |
b411b363 PR |
871 | } |
872 | } | |
5d0b17f1 PR |
873 | |
874 | static bool connection_established(struct drbd_connection *connection, | |
875 | struct socket **sock1, | |
876 | struct socket **sock2) | |
877 | { | |
878 | struct net_conf *nc; | |
879 | int timeout; | |
880 | bool ok; | |
881 | ||
882 | if (!*sock1 || !*sock2) | |
883 | return false; | |
884 | ||
885 | rcu_read_lock(); | |
886 | nc = rcu_dereference(connection->net_conf); | |
887 | timeout = (nc->sock_check_timeo ?: nc->ping_timeo) * HZ / 10; | |
888 | rcu_read_unlock(); | |
889 | schedule_timeout_interruptible(timeout); | |
890 | ||
891 | ok = drbd_socket_okay(sock1); | |
892 | ok = drbd_socket_okay(sock2) && ok; | |
893 | ||
894 | return ok; | |
895 | } | |
896 | ||
2325eb66 PR |
897 | /* Gets called if a connection is established, or if a new minor gets created |
898 | in a connection */ | |
69a22773 | 899 | int drbd_connected(struct drbd_peer_device *peer_device) |
907599e0 | 900 | { |
69a22773 | 901 | struct drbd_device *device = peer_device->device; |
0829f5ed | 902 | int err; |
907599e0 | 903 | |
b30ab791 AG |
904 | atomic_set(&device->packet_seq, 0); |
905 | device->peer_seq = 0; | |
907599e0 | 906 | |
69a22773 AG |
907 | device->state_mutex = peer_device->connection->agreed_pro_version < 100 ? |
908 | &peer_device->connection->cstate_mutex : | |
b30ab791 | 909 | &device->own_state_mutex; |
8410da8f | 910 | |
69a22773 | 911 | err = drbd_send_sync_param(peer_device); |
0829f5ed | 912 | if (!err) |
69a22773 | 913 | err = drbd_send_sizes(peer_device, 0, 0); |
0829f5ed | 914 | if (!err) |
69a22773 | 915 | err = drbd_send_uuids(peer_device); |
0829f5ed | 916 | if (!err) |
69a22773 | 917 | err = drbd_send_current_state(peer_device); |
b30ab791 AG |
918 | clear_bit(USE_DEGR_WFC_T, &device->flags); |
919 | clear_bit(RESIZE_PENDING, &device->flags); | |
920 | atomic_set(&device->ap_in_flight, 0); | |
921 | mod_timer(&device->request_timer, jiffies + HZ); /* just start it here. */ | |
0829f5ed | 922 | return err; |
907599e0 | 923 | } |
b411b363 PR |
924 | |
925 | /* | |
926 | * return values: | |
927 | * 1 yes, we have a valid connection | |
928 | * 0 oops, did not work out, please try again | |
929 | * -1 peer talks different language, | |
930 | * no point in trying again, please go standalone. | |
931 | * -2 We do not have a network config... | |
932 | */ | |
bde89a9e | 933 | static int conn_connect(struct drbd_connection *connection) |
b411b363 | 934 | { |
7da35862 | 935 | struct drbd_socket sock, msock; |
c06ece6b | 936 | struct drbd_peer_device *peer_device; |
44ed167d | 937 | struct net_conf *nc; |
5d0b17f1 PR |
938 | int vnr, timeout, h; |
939 | bool discard_my_data, ok; | |
197296ff | 940 | enum drbd_state_rv rv; |
7a426fd8 | 941 | struct accept_wait_data ad = { |
bde89a9e | 942 | .connection = connection, |
7a426fd8 PR |
943 | .door_bell = COMPLETION_INITIALIZER_ONSTACK(ad.door_bell), |
944 | }; | |
b411b363 | 945 | |
bde89a9e AG |
946 | clear_bit(DISCONNECT_SENT, &connection->flags); |
947 | if (conn_request_state(connection, NS(conn, C_WF_CONNECTION), CS_VERBOSE) < SS_SUCCESS) | |
b411b363 PR |
948 | return -2; |
949 | ||
7da35862 | 950 | mutex_init(&sock.mutex); |
bde89a9e AG |
951 | sock.sbuf = connection->data.sbuf; |
952 | sock.rbuf = connection->data.rbuf; | |
7da35862 PR |
953 | sock.socket = NULL; |
954 | mutex_init(&msock.mutex); | |
bde89a9e AG |
955 | msock.sbuf = connection->meta.sbuf; |
956 | msock.rbuf = connection->meta.rbuf; | |
7da35862 PR |
957 | msock.socket = NULL; |
958 | ||
0916e0e3 | 959 | /* Assume that the peer only understands protocol 80 until we know better. */ |
bde89a9e | 960 | connection->agreed_pro_version = 80; |
b411b363 | 961 | |
bde89a9e | 962 | if (prepare_listen_socket(connection, &ad)) |
7a426fd8 | 963 | return 0; |
b411b363 PR |
964 | |
965 | do { | |
2bf89621 | 966 | struct socket *s; |
b411b363 | 967 | |
bde89a9e | 968 | s = drbd_try_connect(connection); |
b411b363 | 969 | if (s) { |
7da35862 PR |
970 | if (!sock.socket) { |
971 | sock.socket = s; | |
bde89a9e | 972 | send_first_packet(connection, &sock, P_INITIAL_DATA); |
7da35862 | 973 | } else if (!msock.socket) { |
bde89a9e | 974 | clear_bit(RESOLVE_CONFLICTS, &connection->flags); |
7da35862 | 975 | msock.socket = s; |
bde89a9e | 976 | send_first_packet(connection, &msock, P_INITIAL_META); |
b411b363 | 977 | } else { |
1ec861eb | 978 | drbd_err(connection, "Logic error in conn_connect()\n"); |
b411b363 PR |
979 | goto out_release_sockets; |
980 | } | |
981 | } | |
982 | ||
5d0b17f1 PR |
983 | if (connection_established(connection, &sock.socket, &msock.socket)) |
984 | break; | |
b411b363 PR |
985 | |
986 | retry: | |
bde89a9e | 987 | s = drbd_wait_for_connect(connection, &ad); |
b411b363 | 988 | if (s) { |
bde89a9e | 989 | int fp = receive_first_packet(connection, s); |
7da35862 PR |
990 | drbd_socket_okay(&sock.socket); |
991 | drbd_socket_okay(&msock.socket); | |
92f14951 | 992 | switch (fp) { |
e5d6f33a | 993 | case P_INITIAL_DATA: |
7da35862 | 994 | if (sock.socket) { |
1ec861eb | 995 | drbd_warn(connection, "initial packet S crossed\n"); |
7da35862 | 996 | sock_release(sock.socket); |
80c6eed4 PR |
997 | sock.socket = s; |
998 | goto randomize; | |
b411b363 | 999 | } |
7da35862 | 1000 | sock.socket = s; |
b411b363 | 1001 | break; |
e5d6f33a | 1002 | case P_INITIAL_META: |
bde89a9e | 1003 | set_bit(RESOLVE_CONFLICTS, &connection->flags); |
7da35862 | 1004 | if (msock.socket) { |
1ec861eb | 1005 | drbd_warn(connection, "initial packet M crossed\n"); |
7da35862 | 1006 | sock_release(msock.socket); |
80c6eed4 PR |
1007 | msock.socket = s; |
1008 | goto randomize; | |
b411b363 | 1009 | } |
7da35862 | 1010 | msock.socket = s; |
b411b363 PR |
1011 | break; |
1012 | default: | |
1ec861eb | 1013 | drbd_warn(connection, "Error receiving initial packet\n"); |
b411b363 | 1014 | sock_release(s); |
80c6eed4 | 1015 | randomize: |
38b682b2 | 1016 | if (prandom_u32() & 1) |
b411b363 PR |
1017 | goto retry; |
1018 | } | |
1019 | } | |
1020 | ||
bde89a9e | 1021 | if (connection->cstate <= C_DISCONNECTING) |
b411b363 PR |
1022 | goto out_release_sockets; |
1023 | if (signal_pending(current)) { | |
1024 | flush_signals(current); | |
1025 | smp_rmb(); | |
bde89a9e | 1026 | if (get_t_state(&connection->receiver) == EXITING) |
b411b363 PR |
1027 | goto out_release_sockets; |
1028 | } | |
1029 | ||
5d0b17f1 | 1030 | ok = connection_established(connection, &sock.socket, &msock.socket); |
b666dbf8 | 1031 | } while (!ok); |
b411b363 | 1032 | |
7a426fd8 PR |
1033 | if (ad.s_listen) |
1034 | sock_release(ad.s_listen); | |
b411b363 | 1035 | |
98683650 PR |
1036 | sock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ |
1037 | msock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ | |
b411b363 | 1038 | |
7da35862 PR |
1039 | sock.socket->sk->sk_allocation = GFP_NOIO; |
1040 | msock.socket->sk->sk_allocation = GFP_NOIO; | |
b411b363 | 1041 | |
7da35862 PR |
1042 | sock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; |
1043 | msock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE; | |
b411b363 | 1044 | |
b411b363 | 1045 | /* NOT YET ... |
bde89a9e | 1046 | * sock.socket->sk->sk_sndtimeo = connection->net_conf->timeout*HZ/10; |
7da35862 | 1047 | * sock.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; |
6038178e | 1048 | * first set it to the P_CONNECTION_FEATURES timeout, |
b411b363 | 1049 | * which we set to 4x the configured ping_timeout. */ |
44ed167d | 1050 | rcu_read_lock(); |
bde89a9e | 1051 | nc = rcu_dereference(connection->net_conf); |
44ed167d | 1052 | |
7da35862 PR |
1053 | sock.socket->sk->sk_sndtimeo = |
1054 | sock.socket->sk->sk_rcvtimeo = nc->ping_timeo*4*HZ/10; | |
b411b363 | 1055 | |
7da35862 | 1056 | msock.socket->sk->sk_rcvtimeo = nc->ping_int*HZ; |
44ed167d | 1057 | timeout = nc->timeout * HZ / 10; |
08b165ba | 1058 | discard_my_data = nc->discard_my_data; |
44ed167d | 1059 | rcu_read_unlock(); |
b411b363 | 1060 | |
7da35862 | 1061 | msock.socket->sk->sk_sndtimeo = timeout; |
b411b363 PR |
1062 | |
1063 | /* we don't want delays. | |
25985edc | 1064 | * we use TCP_CORK where appropriate, though */ |
7da35862 PR |
1065 | drbd_tcp_nodelay(sock.socket); |
1066 | drbd_tcp_nodelay(msock.socket); | |
b411b363 | 1067 | |
bde89a9e AG |
1068 | connection->data.socket = sock.socket; |
1069 | connection->meta.socket = msock.socket; | |
1070 | connection->last_received = jiffies; | |
b411b363 | 1071 | |
bde89a9e | 1072 | h = drbd_do_features(connection); |
b411b363 PR |
1073 | if (h <= 0) |
1074 | return h; | |
1075 | ||
bde89a9e | 1076 | if (connection->cram_hmac_tfm) { |
b30ab791 | 1077 | /* drbd_request_state(device, NS(conn, WFAuth)); */ |
bde89a9e | 1078 | switch (drbd_do_auth(connection)) { |
b10d96cb | 1079 | case -1: |
1ec861eb | 1080 | drbd_err(connection, "Authentication of peer failed\n"); |
b411b363 | 1081 | return -1; |
b10d96cb | 1082 | case 0: |
1ec861eb | 1083 | drbd_err(connection, "Authentication of peer failed, trying again.\n"); |
b10d96cb | 1084 | return 0; |
b411b363 PR |
1085 | } |
1086 | } | |
1087 | ||
bde89a9e AG |
1088 | connection->data.socket->sk->sk_sndtimeo = timeout; |
1089 | connection->data.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; | |
b411b363 | 1090 | |
bde89a9e | 1091 | if (drbd_send_protocol(connection) == -EOPNOTSUPP) |
7e2455c1 | 1092 | return -1; |
b411b363 | 1093 | |
31007745 PR |
1094 | /* Prevent a race between resync-handshake and |
1095 | * being promoted to Primary. | |
1096 | * | |
1097 | * Grab and release the state mutex, so we know that any current | |
1098 | * drbd_set_role() is finished, and any incoming drbd_set_role | |
1099 | * will see the STATE_SENT flag, and wait for it to be cleared. | |
1100 | */ | |
1101 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) | |
1102 | mutex_lock(peer_device->device->state_mutex); | |
1103 | ||
cde81d99 LE |
1104 | /* avoid a race with conn_request_state( C_DISCONNECTING ) */ |
1105 | spin_lock_irq(&connection->resource->req_lock); | |
bde89a9e | 1106 | set_bit(STATE_SENT, &connection->flags); |
cde81d99 | 1107 | spin_unlock_irq(&connection->resource->req_lock); |
a1096a6e | 1108 | |
31007745 PR |
1109 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) |
1110 | mutex_unlock(peer_device->device->state_mutex); | |
1111 | ||
c141ebda | 1112 | rcu_read_lock(); |
c06ece6b AG |
1113 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
1114 | struct drbd_device *device = peer_device->device; | |
b30ab791 | 1115 | kref_get(&device->kref); |
26ea8f92 AG |
1116 | rcu_read_unlock(); |
1117 | ||
08b165ba | 1118 | if (discard_my_data) |
b30ab791 | 1119 | set_bit(DISCARD_MY_DATA, &device->flags); |
08b165ba | 1120 | else |
b30ab791 | 1121 | clear_bit(DISCARD_MY_DATA, &device->flags); |
08b165ba | 1122 | |
69a22773 | 1123 | drbd_connected(peer_device); |
05a10ec7 | 1124 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda PR |
1125 | rcu_read_lock(); |
1126 | } | |
1127 | rcu_read_unlock(); | |
1128 | ||
bde89a9e AG |
1129 | rv = conn_request_state(connection, NS(conn, C_WF_REPORT_PARAMS), CS_VERBOSE); |
1130 | if (rv < SS_SUCCESS || connection->cstate != C_WF_REPORT_PARAMS) { | |
1131 | clear_bit(STATE_SENT, &connection->flags); | |
1e86ac48 | 1132 | return 0; |
a1096a6e | 1133 | } |
1e86ac48 | 1134 | |
1c03e520 | 1135 | drbd_thread_start(&connection->ack_receiver); |
39e91a60 LE |
1136 | /* opencoded create_singlethread_workqueue(), |
1137 | * to be able to use format string arguments */ | |
1138 | connection->ack_sender = | |
1139 | alloc_ordered_workqueue("drbd_as_%s", WQ_MEM_RECLAIM, connection->resource->name); | |
668700b4 PR |
1140 | if (!connection->ack_sender) { |
1141 | drbd_err(connection, "Failed to create workqueue ack_sender\n"); | |
1142 | return 0; | |
1143 | } | |
b411b363 | 1144 | |
0500813f | 1145 | mutex_lock(&connection->resource->conf_update); |
08b165ba PR |
1146 | /* The discard_my_data flag is a single-shot modifier to the next |
1147 | * connection attempt, the handshake of which is now well underway. | |
1148 | * No need for rcu style copying of the whole struct | |
1149 | * just to clear a single value. */ | |
bde89a9e | 1150 | connection->net_conf->discard_my_data = 0; |
0500813f | 1151 | mutex_unlock(&connection->resource->conf_update); |
08b165ba | 1152 | |
d3fcb490 | 1153 | return h; |
b411b363 PR |
1154 | |
1155 | out_release_sockets: | |
7a426fd8 PR |
1156 | if (ad.s_listen) |
1157 | sock_release(ad.s_listen); | |
7da35862 PR |
1158 | if (sock.socket) |
1159 | sock_release(sock.socket); | |
1160 | if (msock.socket) | |
1161 | sock_release(msock.socket); | |
b411b363 PR |
1162 | return -1; |
1163 | } | |
1164 | ||
bde89a9e | 1165 | static int decode_header(struct drbd_connection *connection, void *header, struct packet_info *pi) |
b411b363 | 1166 | { |
bde89a9e | 1167 | unsigned int header_size = drbd_header_size(connection); |
e658983a | 1168 | |
0c8e36d9 AG |
1169 | if (header_size == sizeof(struct p_header100) && |
1170 | *(__be32 *)header == cpu_to_be32(DRBD_MAGIC_100)) { | |
1171 | struct p_header100 *h = header; | |
1172 | if (h->pad != 0) { | |
1ec861eb | 1173 | drbd_err(connection, "Header padding is not zero\n"); |
0c8e36d9 AG |
1174 | return -EINVAL; |
1175 | } | |
1176 | pi->vnr = be16_to_cpu(h->volume); | |
1177 | pi->cmd = be16_to_cpu(h->command); | |
1178 | pi->size = be32_to_cpu(h->length); | |
1179 | } else if (header_size == sizeof(struct p_header95) && | |
1180 | *(__be16 *)header == cpu_to_be16(DRBD_MAGIC_BIG)) { | |
e658983a | 1181 | struct p_header95 *h = header; |
e658983a | 1182 | pi->cmd = be16_to_cpu(h->command); |
b55d84ba AG |
1183 | pi->size = be32_to_cpu(h->length); |
1184 | pi->vnr = 0; | |
e658983a AG |
1185 | } else if (header_size == sizeof(struct p_header80) && |
1186 | *(__be32 *)header == cpu_to_be32(DRBD_MAGIC)) { | |
1187 | struct p_header80 *h = header; | |
1188 | pi->cmd = be16_to_cpu(h->command); | |
1189 | pi->size = be16_to_cpu(h->length); | |
77351055 | 1190 | pi->vnr = 0; |
02918be2 | 1191 | } else { |
1ec861eb | 1192 | drbd_err(connection, "Wrong magic value 0x%08x in protocol version %d\n", |
e658983a | 1193 | be32_to_cpu(*(__be32 *)header), |
bde89a9e | 1194 | connection->agreed_pro_version); |
8172f3e9 | 1195 | return -EINVAL; |
b411b363 | 1196 | } |
e658983a | 1197 | pi->data = header + header_size; |
8172f3e9 | 1198 | return 0; |
257d0af6 | 1199 | } |
b411b363 | 1200 | |
c51a0ef3 LE |
1201 | static void drbd_unplug_all_devices(struct drbd_connection *connection) |
1202 | { | |
1203 | if (current->plug == &connection->receiver_plug) { | |
1204 | blk_finish_plug(&connection->receiver_plug); | |
1205 | blk_start_plug(&connection->receiver_plug); | |
1206 | } /* else: maybe just schedule() ?? */ | |
1207 | } | |
1208 | ||
bde89a9e | 1209 | static int drbd_recv_header(struct drbd_connection *connection, struct packet_info *pi) |
257d0af6 | 1210 | { |
bde89a9e | 1211 | void *buffer = connection->data.rbuf; |
69bc7bc3 | 1212 | int err; |
257d0af6 | 1213 | |
bde89a9e | 1214 | err = drbd_recv_all_warn(connection, buffer, drbd_header_size(connection)); |
a5c31904 | 1215 | if (err) |
69bc7bc3 | 1216 | return err; |
257d0af6 | 1217 | |
bde89a9e AG |
1218 | err = decode_header(connection, buffer, pi); |
1219 | connection->last_received = jiffies; | |
b411b363 | 1220 | |
69bc7bc3 | 1221 | return err; |
b411b363 PR |
1222 | } |
1223 | ||
c51a0ef3 LE |
1224 | static int drbd_recv_header_maybe_unplug(struct drbd_connection *connection, struct packet_info *pi) |
1225 | { | |
1226 | void *buffer = connection->data.rbuf; | |
1227 | unsigned int size = drbd_header_size(connection); | |
1228 | int err; | |
1229 | ||
1230 | err = drbd_recv_short(connection->data.socket, buffer, size, MSG_NOSIGNAL|MSG_DONTWAIT); | |
1231 | if (err != size) { | |
1232 | /* If we have nothing in the receive buffer now, to reduce | |
1233 | * application latency, try to drain the backend queues as | |
1234 | * quickly as possible, and let remote TCP know what we have | |
1235 | * received so far. */ | |
1236 | if (err == -EAGAIN) { | |
1237 | drbd_tcp_quickack(connection->data.socket); | |
1238 | drbd_unplug_all_devices(connection); | |
1239 | } | |
1240 | if (err > 0) { | |
1241 | buffer += err; | |
1242 | size -= err; | |
1243 | } | |
1244 | err = drbd_recv_all_warn(connection, buffer, size); | |
1245 | if (err) | |
1246 | return err; | |
1247 | } | |
1248 | ||
1249 | err = decode_header(connection, connection->data.rbuf, pi); | |
1250 | connection->last_received = jiffies; | |
1251 | ||
1252 | return err; | |
1253 | } | |
f9ff0da5 LE |
1254 | /* This is blkdev_issue_flush, but asynchronous. |
1255 | * We want to submit to all component volumes in parallel, | |
1256 | * then wait for all completions. | |
1257 | */ | |
1258 | struct issue_flush_context { | |
1259 | atomic_t pending; | |
1260 | int error; | |
1261 | struct completion done; | |
1262 | }; | |
1263 | struct one_flush_context { | |
1264 | struct drbd_device *device; | |
1265 | struct issue_flush_context *ctx; | |
1266 | }; | |
1267 | ||
1ffa7bfa | 1268 | static void one_flush_endio(struct bio *bio) |
b411b363 | 1269 | { |
f9ff0da5 LE |
1270 | struct one_flush_context *octx = bio->bi_private; |
1271 | struct drbd_device *device = octx->device; | |
1272 | struct issue_flush_context *ctx = octx->ctx; | |
1273 | ||
4e4cbee9 CH |
1274 | if (bio->bi_status) { |
1275 | ctx->error = blk_status_to_errno(bio->bi_status); | |
1276 | drbd_info(device, "local disk FLUSH FAILED with status %d\n", bio->bi_status); | |
f9ff0da5 LE |
1277 | } |
1278 | kfree(octx); | |
1279 | bio_put(bio); | |
1280 | ||
1281 | clear_bit(FLUSH_PENDING, &device->flags); | |
1282 | put_ldev(device); | |
1283 | kref_put(&device->kref, drbd_destroy_device); | |
1284 | ||
1285 | if (atomic_dec_and_test(&ctx->pending)) | |
1286 | complete(&ctx->done); | |
1287 | } | |
1288 | ||
1289 | static void submit_one_flush(struct drbd_device *device, struct issue_flush_context *ctx) | |
1290 | { | |
1291 | struct bio *bio = bio_alloc(GFP_NOIO, 0); | |
1292 | struct one_flush_context *octx = kmalloc(sizeof(*octx), GFP_NOIO); | |
1293 | if (!bio || !octx) { | |
1294 | drbd_warn(device, "Could not allocate a bio, CANNOT ISSUE FLUSH\n"); | |
1295 | /* FIXME: what else can I do now? disconnecting or detaching | |
1296 | * really does not help to improve the state of the world, either. | |
1297 | */ | |
1298 | kfree(octx); | |
1299 | if (bio) | |
1300 | bio_put(bio); | |
1301 | ||
1302 | ctx->error = -ENOMEM; | |
1303 | put_ldev(device); | |
1304 | kref_put(&device->kref, drbd_destroy_device); | |
1305 | return; | |
1306 | } | |
4b0007c0 | 1307 | |
f9ff0da5 LE |
1308 | octx->device = device; |
1309 | octx->ctx = ctx; | |
74d46992 | 1310 | bio_set_dev(bio, device->ldev->backing_bdev); |
f9ff0da5 LE |
1311 | bio->bi_private = octx; |
1312 | bio->bi_end_io = one_flush_endio; | |
70fd7614 | 1313 | bio->bi_opf = REQ_OP_FLUSH | REQ_PREFLUSH; |
f9ff0da5 LE |
1314 | |
1315 | device->flush_jif = jiffies; | |
1316 | set_bit(FLUSH_PENDING, &device->flags); | |
1317 | atomic_inc(&ctx->pending); | |
1318 | submit_bio(bio); | |
1319 | } | |
1320 | ||
1321 | static void drbd_flush(struct drbd_connection *connection) | |
1322 | { | |
f6ba8636 | 1323 | if (connection->resource->write_ordering >= WO_BDEV_FLUSH) { |
f9ff0da5 LE |
1324 | struct drbd_peer_device *peer_device; |
1325 | struct issue_flush_context ctx; | |
1326 | int vnr; | |
1327 | ||
1328 | atomic_set(&ctx.pending, 1); | |
1329 | ctx.error = 0; | |
1330 | init_completion(&ctx.done); | |
1331 | ||
615e087f | 1332 | rcu_read_lock(); |
c06ece6b AG |
1333 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
1334 | struct drbd_device *device = peer_device->device; | |
1335 | ||
b30ab791 | 1336 | if (!get_ldev(device)) |
615e087f | 1337 | continue; |
b30ab791 | 1338 | kref_get(&device->kref); |
615e087f LE |
1339 | rcu_read_unlock(); |
1340 | ||
f9ff0da5 | 1341 | submit_one_flush(device, &ctx); |
b411b363 | 1342 | |
615e087f | 1343 | rcu_read_lock(); |
b411b363 | 1344 | } |
615e087f | 1345 | rcu_read_unlock(); |
f9ff0da5 LE |
1346 | |
1347 | /* Do we want to add a timeout, | |
1348 | * if disk-timeout is set? */ | |
1349 | if (!atomic_dec_and_test(&ctx.pending)) | |
1350 | wait_for_completion(&ctx.done); | |
1351 | ||
1352 | if (ctx.error) { | |
1353 | /* would rather check on EOPNOTSUPP, but that is not reliable. | |
1354 | * don't try again for ANY return value != 0 | |
1355 | * if (rv == -EOPNOTSUPP) */ | |
1356 | /* Any error is already reported by bio_endio callback. */ | |
1357 | drbd_bump_write_ordering(connection->resource, NULL, WO_DRAIN_IO); | |
1358 | } | |
b411b363 | 1359 | } |
b411b363 PR |
1360 | } |
1361 | ||
1362 | /** | |
1363 | * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. | |
b30ab791 | 1364 | * @device: DRBD device. |
b411b363 PR |
1365 | * @epoch: Epoch object. |
1366 | * @ev: Epoch event. | |
1367 | */ | |
bde89a9e | 1368 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_connection *connection, |
b411b363 PR |
1369 | struct drbd_epoch *epoch, |
1370 | enum epoch_event ev) | |
1371 | { | |
2451fc3b | 1372 | int epoch_size; |
b411b363 | 1373 | struct drbd_epoch *next_epoch; |
b411b363 PR |
1374 | enum finish_epoch rv = FE_STILL_LIVE; |
1375 | ||
bde89a9e | 1376 | spin_lock(&connection->epoch_lock); |
b411b363 PR |
1377 | do { |
1378 | next_epoch = NULL; | |
b411b363 PR |
1379 | |
1380 | epoch_size = atomic_read(&epoch->epoch_size); | |
1381 | ||
1382 | switch (ev & ~EV_CLEANUP) { | |
1383 | case EV_PUT: | |
1384 | atomic_dec(&epoch->active); | |
1385 | break; | |
1386 | case EV_GOT_BARRIER_NR: | |
1387 | set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); | |
b411b363 PR |
1388 | break; |
1389 | case EV_BECAME_LAST: | |
1390 | /* nothing to do*/ | |
1391 | break; | |
1392 | } | |
1393 | ||
b411b363 PR |
1394 | if (epoch_size != 0 && |
1395 | atomic_read(&epoch->active) == 0 && | |
80f9fd55 | 1396 | (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) || ev & EV_CLEANUP)) { |
b411b363 | 1397 | if (!(ev & EV_CLEANUP)) { |
bde89a9e AG |
1398 | spin_unlock(&connection->epoch_lock); |
1399 | drbd_send_b_ack(epoch->connection, epoch->barrier_nr, epoch_size); | |
1400 | spin_lock(&connection->epoch_lock); | |
b411b363 | 1401 | } |
9ed57dcb LE |
1402 | #if 0 |
1403 | /* FIXME: dec unacked on connection, once we have | |
1404 | * something to count pending connection packets in. */ | |
80f9fd55 | 1405 | if (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) |
bde89a9e | 1406 | dec_unacked(epoch->connection); |
9ed57dcb | 1407 | #endif |
b411b363 | 1408 | |
bde89a9e | 1409 | if (connection->current_epoch != epoch) { |
b411b363 PR |
1410 | next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); |
1411 | list_del(&epoch->list); | |
1412 | ev = EV_BECAME_LAST | (ev & EV_CLEANUP); | |
bde89a9e | 1413 | connection->epochs--; |
b411b363 PR |
1414 | kfree(epoch); |
1415 | ||
1416 | if (rv == FE_STILL_LIVE) | |
1417 | rv = FE_DESTROYED; | |
1418 | } else { | |
1419 | epoch->flags = 0; | |
1420 | atomic_set(&epoch->epoch_size, 0); | |
698f9315 | 1421 | /* atomic_set(&epoch->active, 0); is already zero */ |
b411b363 PR |
1422 | if (rv == FE_STILL_LIVE) |
1423 | rv = FE_RECYCLED; | |
1424 | } | |
1425 | } | |
1426 | ||
1427 | if (!next_epoch) | |
1428 | break; | |
1429 | ||
1430 | epoch = next_epoch; | |
1431 | } while (1); | |
1432 | ||
bde89a9e | 1433 | spin_unlock(&connection->epoch_lock); |
b411b363 | 1434 | |
b411b363 PR |
1435 | return rv; |
1436 | } | |
1437 | ||
8fe39aac PR |
1438 | static enum write_ordering_e |
1439 | max_allowed_wo(struct drbd_backing_dev *bdev, enum write_ordering_e wo) | |
1440 | { | |
1441 | struct disk_conf *dc; | |
1442 | ||
1443 | dc = rcu_dereference(bdev->disk_conf); | |
1444 | ||
f6ba8636 AG |
1445 | if (wo == WO_BDEV_FLUSH && !dc->disk_flushes) |
1446 | wo = WO_DRAIN_IO; | |
1447 | if (wo == WO_DRAIN_IO && !dc->disk_drain) | |
1448 | wo = WO_NONE; | |
8fe39aac PR |
1449 | |
1450 | return wo; | |
1451 | } | |
1452 | ||
b411b363 PR |
1453 | /** |
1454 | * drbd_bump_write_ordering() - Fall back to an other write ordering method | |
bde89a9e | 1455 | * @connection: DRBD connection. |
b411b363 PR |
1456 | * @wo: Write ordering method to try. |
1457 | */ | |
8fe39aac PR |
1458 | void drbd_bump_write_ordering(struct drbd_resource *resource, struct drbd_backing_dev *bdev, |
1459 | enum write_ordering_e wo) | |
b411b363 | 1460 | { |
e9526580 | 1461 | struct drbd_device *device; |
b411b363 | 1462 | enum write_ordering_e pwo; |
4b0007c0 | 1463 | int vnr; |
b411b363 | 1464 | static char *write_ordering_str[] = { |
f6ba8636 AG |
1465 | [WO_NONE] = "none", |
1466 | [WO_DRAIN_IO] = "drain", | |
1467 | [WO_BDEV_FLUSH] = "flush", | |
b411b363 PR |
1468 | }; |
1469 | ||
e9526580 | 1470 | pwo = resource->write_ordering; |
f6ba8636 | 1471 | if (wo != WO_BDEV_FLUSH) |
70df7092 | 1472 | wo = min(pwo, wo); |
daeda1cc | 1473 | rcu_read_lock(); |
e9526580 | 1474 | idr_for_each_entry(&resource->devices, device, vnr) { |
8fe39aac PR |
1475 | if (get_ldev(device)) { |
1476 | wo = max_allowed_wo(device->ldev, wo); | |
1477 | if (device->ldev == bdev) | |
1478 | bdev = NULL; | |
1479 | put_ldev(device); | |
1480 | } | |
4b0007c0 | 1481 | } |
8fe39aac PR |
1482 | |
1483 | if (bdev) | |
1484 | wo = max_allowed_wo(bdev, wo); | |
1485 | ||
70df7092 LE |
1486 | rcu_read_unlock(); |
1487 | ||
e9526580 | 1488 | resource->write_ordering = wo; |
f6ba8636 | 1489 | if (pwo != resource->write_ordering || wo == WO_BDEV_FLUSH) |
e9526580 | 1490 | drbd_info(resource, "Method to ensure write ordering: %s\n", write_ordering_str[resource->write_ordering]); |
b411b363 PR |
1491 | } |
1492 | ||
0dbed96a | 1493 | static void drbd_issue_peer_discard(struct drbd_device *device, struct drbd_peer_request *peer_req) |
dd4f699d LE |
1494 | { |
1495 | struct block_device *bdev = device->ldev->backing_bdev; | |
dd4f699d | 1496 | |
0dbed96a CH |
1497 | if (blkdev_issue_zeroout(bdev, peer_req->i.sector, peer_req->i.size >> 9, |
1498 | GFP_NOIO, 0)) | |
dd4f699d | 1499 | peer_req->flags |= EE_WAS_ERROR; |
0dbed96a | 1500 | |
dd4f699d LE |
1501 | drbd_endio_write_sec_final(peer_req); |
1502 | } | |
1503 | ||
9104d31a LE |
1504 | static void drbd_issue_peer_wsame(struct drbd_device *device, |
1505 | struct drbd_peer_request *peer_req) | |
1506 | { | |
1507 | struct block_device *bdev = device->ldev->backing_bdev; | |
1508 | sector_t s = peer_req->i.sector; | |
1509 | sector_t nr = peer_req->i.size >> 9; | |
1510 | if (blkdev_issue_write_same(bdev, s, nr, GFP_NOIO, peer_req->pages)) | |
1511 | peer_req->flags |= EE_WAS_ERROR; | |
1512 | drbd_endio_write_sec_final(peer_req); | |
1513 | } | |
1514 | ||
1515 | ||
45bb912b | 1516 | /** |
fbe29dec | 1517 | * drbd_submit_peer_request() |
b30ab791 | 1518 | * @device: DRBD device. |
db830c46 | 1519 | * @peer_req: peer request |
1eff9d32 | 1520 | * @rw: flag field, see bio->bi_opf |
10f6d992 LE |
1521 | * |
1522 | * May spread the pages to multiple bios, | |
1523 | * depending on bio_add_page restrictions. | |
1524 | * | |
1525 | * Returns 0 if all bios have been submitted, | |
1526 | * -ENOMEM if we could not allocate enough bios, | |
1527 | * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a | |
1528 | * single page to an empty bio (which should never happen and likely indicates | |
1529 | * that the lower level IO stack is in some way broken). This has been observed | |
1530 | * on certain Xen deployments. | |
45bb912b LE |
1531 | */ |
1532 | /* TODO allocate from our own bio_set. */ | |
b30ab791 | 1533 | int drbd_submit_peer_request(struct drbd_device *device, |
fbe29dec | 1534 | struct drbd_peer_request *peer_req, |
bb3cc85e MC |
1535 | const unsigned op, const unsigned op_flags, |
1536 | const int fault_type) | |
45bb912b LE |
1537 | { |
1538 | struct bio *bios = NULL; | |
1539 | struct bio *bio; | |
db830c46 AG |
1540 | struct page *page = peer_req->pages; |
1541 | sector_t sector = peer_req->i.sector; | |
11f8b2b6 | 1542 | unsigned data_size = peer_req->i.size; |
45bb912b | 1543 | unsigned n_bios = 0; |
11f8b2b6 | 1544 | unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT; |
10f6d992 | 1545 | int err = -ENOMEM; |
45bb912b | 1546 | |
dd4f699d LE |
1547 | /* TRIM/DISCARD: for now, always use the helper function |
1548 | * blkdev_issue_zeroout(..., discard=true). | |
1549 | * It's synchronous, but it does the right thing wrt. bio splitting. | |
1550 | * Correctness first, performance later. Next step is to code an | |
1551 | * asynchronous variant of the same. | |
1552 | */ | |
9104d31a | 1553 | if (peer_req->flags & (EE_IS_TRIM|EE_WRITE_SAME)) { |
a0fb3c47 LE |
1554 | /* wait for all pending IO completions, before we start |
1555 | * zeroing things out. */ | |
5dd2ca19 | 1556 | conn_wait_active_ee_empty(peer_req->peer_device->connection); |
45d2933c LE |
1557 | /* add it to the active list now, |
1558 | * so we can find it to present it in debugfs */ | |
21ae5d7f LE |
1559 | peer_req->submit_jif = jiffies; |
1560 | peer_req->flags |= EE_SUBMITTED; | |
700ca8c0 PR |
1561 | |
1562 | /* If this was a resync request from receive_rs_deallocated(), | |
1563 | * it is already on the sync_ee list */ | |
1564 | if (list_empty(&peer_req->w.list)) { | |
1565 | spin_lock_irq(&device->resource->req_lock); | |
1566 | list_add_tail(&peer_req->w.list, &device->active_ee); | |
1567 | spin_unlock_irq(&device->resource->req_lock); | |
1568 | } | |
1569 | ||
9104d31a LE |
1570 | if (peer_req->flags & EE_IS_TRIM) |
1571 | drbd_issue_peer_discard(device, peer_req); | |
1572 | else /* EE_WRITE_SAME */ | |
1573 | drbd_issue_peer_wsame(device, peer_req); | |
a0fb3c47 LE |
1574 | return 0; |
1575 | } | |
1576 | ||
45bb912b LE |
1577 | /* In most cases, we will only need one bio. But in case the lower |
1578 | * level restrictions happen to be different at this offset on this | |
1579 | * side than those of the sending peer, we may need to submit the | |
9476f39d LE |
1580 | * request in more than one bio. |
1581 | * | |
1582 | * Plain bio_alloc is good enough here, this is no DRBD internally | |
1583 | * generated bio, but a bio allocated on behalf of the peer. | |
1584 | */ | |
45bb912b LE |
1585 | next_bio: |
1586 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
1587 | if (!bio) { | |
a0fb3c47 | 1588 | drbd_err(device, "submit_ee: Allocation of a bio failed (nr_pages=%u)\n", nr_pages); |
45bb912b LE |
1589 | goto fail; |
1590 | } | |
db830c46 | 1591 | /* > peer_req->i.sector, unless this is the first bio */ |
4f024f37 | 1592 | bio->bi_iter.bi_sector = sector; |
74d46992 | 1593 | bio_set_dev(bio, device->ldev->backing_bdev); |
bb3cc85e | 1594 | bio_set_op_attrs(bio, op, op_flags); |
db830c46 | 1595 | bio->bi_private = peer_req; |
fcefa62e | 1596 | bio->bi_end_io = drbd_peer_request_endio; |
45bb912b LE |
1597 | |
1598 | bio->bi_next = bios; | |
1599 | bios = bio; | |
1600 | ++n_bios; | |
1601 | ||
1602 | page_chain_for_each(page) { | |
11f8b2b6 | 1603 | unsigned len = min_t(unsigned, data_size, PAGE_SIZE); |
06efffda | 1604 | if (!bio_add_page(bio, page, len, 0)) |
45bb912b | 1605 | goto next_bio; |
11f8b2b6 | 1606 | data_size -= len; |
45bb912b LE |
1607 | sector += len >> 9; |
1608 | --nr_pages; | |
1609 | } | |
11f8b2b6 | 1610 | D_ASSERT(device, data_size == 0); |
a0fb3c47 | 1611 | D_ASSERT(device, page == NULL); |
45bb912b | 1612 | |
db830c46 | 1613 | atomic_set(&peer_req->pending_bios, n_bios); |
21ae5d7f LE |
1614 | /* for debugfs: update timestamp, mark as submitted */ |
1615 | peer_req->submit_jif = jiffies; | |
1616 | peer_req->flags |= EE_SUBMITTED; | |
45bb912b LE |
1617 | do { |
1618 | bio = bios; | |
1619 | bios = bios->bi_next; | |
1620 | bio->bi_next = NULL; | |
1621 | ||
b30ab791 | 1622 | drbd_generic_make_request(device, fault_type, bio); |
45bb912b | 1623 | } while (bios); |
45bb912b LE |
1624 | return 0; |
1625 | ||
1626 | fail: | |
1627 | while (bios) { | |
1628 | bio = bios; | |
1629 | bios = bios->bi_next; | |
1630 | bio_put(bio); | |
1631 | } | |
10f6d992 | 1632 | return err; |
45bb912b LE |
1633 | } |
1634 | ||
b30ab791 | 1635 | static void drbd_remove_epoch_entry_interval(struct drbd_device *device, |
db830c46 | 1636 | struct drbd_peer_request *peer_req) |
53840641 | 1637 | { |
db830c46 | 1638 | struct drbd_interval *i = &peer_req->i; |
53840641 | 1639 | |
b30ab791 | 1640 | drbd_remove_interval(&device->write_requests, i); |
53840641 AG |
1641 | drbd_clear_interval(i); |
1642 | ||
6c852bec | 1643 | /* Wake up any processes waiting for this peer request to complete. */ |
53840641 | 1644 | if (i->waiting) |
b30ab791 | 1645 | wake_up(&device->misc_wait); |
53840641 AG |
1646 | } |
1647 | ||
bde89a9e | 1648 | static void conn_wait_active_ee_empty(struct drbd_connection *connection) |
77fede51 | 1649 | { |
c06ece6b | 1650 | struct drbd_peer_device *peer_device; |
77fede51 PR |
1651 | int vnr; |
1652 | ||
1653 | rcu_read_lock(); | |
c06ece6b AG |
1654 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
1655 | struct drbd_device *device = peer_device->device; | |
1656 | ||
b30ab791 | 1657 | kref_get(&device->kref); |
77fede51 | 1658 | rcu_read_unlock(); |
b30ab791 | 1659 | drbd_wait_ee_list_empty(device, &device->active_ee); |
05a10ec7 | 1660 | kref_put(&device->kref, drbd_destroy_device); |
77fede51 PR |
1661 | rcu_read_lock(); |
1662 | } | |
1663 | rcu_read_unlock(); | |
1664 | } | |
1665 | ||
bde89a9e | 1666 | static int receive_Barrier(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 1667 | { |
2451fc3b | 1668 | int rv; |
e658983a | 1669 | struct p_barrier *p = pi->data; |
b411b363 PR |
1670 | struct drbd_epoch *epoch; |
1671 | ||
9ed57dcb LE |
1672 | /* FIXME these are unacked on connection, |
1673 | * not a specific (peer)device. | |
1674 | */ | |
bde89a9e AG |
1675 | connection->current_epoch->barrier_nr = p->barrier; |
1676 | connection->current_epoch->connection = connection; | |
1677 | rv = drbd_may_finish_epoch(connection, connection->current_epoch, EV_GOT_BARRIER_NR); | |
b411b363 PR |
1678 | |
1679 | /* P_BARRIER_ACK may imply that the corresponding extent is dropped from | |
1680 | * the activity log, which means it would not be resynced in case the | |
1681 | * R_PRIMARY crashes now. | |
1682 | * Therefore we must send the barrier_ack after the barrier request was | |
1683 | * completed. */ | |
e9526580 | 1684 | switch (connection->resource->write_ordering) { |
f6ba8636 | 1685 | case WO_NONE: |
b411b363 | 1686 | if (rv == FE_RECYCLED) |
82bc0194 | 1687 | return 0; |
2451fc3b PR |
1688 | |
1689 | /* receiver context, in the writeout path of the other node. | |
1690 | * avoid potential distributed deadlock */ | |
1691 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1692 | if (epoch) | |
1693 | break; | |
1694 | else | |
1ec861eb | 1695 | drbd_warn(connection, "Allocation of an epoch failed, slowing down\n"); |
2451fc3b | 1696 | /* Fall through */ |
b411b363 | 1697 | |
f6ba8636 AG |
1698 | case WO_BDEV_FLUSH: |
1699 | case WO_DRAIN_IO: | |
bde89a9e AG |
1700 | conn_wait_active_ee_empty(connection); |
1701 | drbd_flush(connection); | |
2451fc3b | 1702 | |
bde89a9e | 1703 | if (atomic_read(&connection->current_epoch->epoch_size)) { |
2451fc3b PR |
1704 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); |
1705 | if (epoch) | |
1706 | break; | |
b411b363 PR |
1707 | } |
1708 | ||
82bc0194 | 1709 | return 0; |
2451fc3b | 1710 | default: |
e9526580 PR |
1711 | drbd_err(connection, "Strangeness in connection->write_ordering %d\n", |
1712 | connection->resource->write_ordering); | |
82bc0194 | 1713 | return -EIO; |
b411b363 PR |
1714 | } |
1715 | ||
1716 | epoch->flags = 0; | |
1717 | atomic_set(&epoch->epoch_size, 0); | |
1718 | atomic_set(&epoch->active, 0); | |
1719 | ||
bde89a9e AG |
1720 | spin_lock(&connection->epoch_lock); |
1721 | if (atomic_read(&connection->current_epoch->epoch_size)) { | |
1722 | list_add(&epoch->list, &connection->current_epoch->list); | |
1723 | connection->current_epoch = epoch; | |
1724 | connection->epochs++; | |
b411b363 PR |
1725 | } else { |
1726 | /* The current_epoch got recycled while we allocated this one... */ | |
1727 | kfree(epoch); | |
1728 | } | |
bde89a9e | 1729 | spin_unlock(&connection->epoch_lock); |
b411b363 | 1730 | |
82bc0194 | 1731 | return 0; |
b411b363 PR |
1732 | } |
1733 | ||
9104d31a | 1734 | /* quick wrapper in case payload size != request_size (write same) */ |
3d0e6375 | 1735 | static void drbd_csum_ee_size(struct crypto_shash *h, |
9104d31a LE |
1736 | struct drbd_peer_request *r, void *d, |
1737 | unsigned int payload_size) | |
1738 | { | |
1739 | unsigned int tmp = r->i.size; | |
1740 | r->i.size = payload_size; | |
1741 | drbd_csum_ee(h, r, d); | |
1742 | r->i.size = tmp; | |
1743 | } | |
1744 | ||
b411b363 | 1745 | /* used from receive_RSDataReply (recv_resync_read) |
9104d31a LE |
1746 | * and from receive_Data. |
1747 | * data_size: actual payload ("data in") | |
1748 | * for normal writes that is bi_size. | |
1749 | * for discards, that is zero. | |
1750 | * for write same, it is logical_block_size. | |
1751 | * both trim and write same have the bi_size ("data len to be affected") | |
1752 | * as extra argument in the packet header. | |
1753 | */ | |
f6ffca9f | 1754 | static struct drbd_peer_request * |
69a22773 | 1755 | read_in_block(struct drbd_peer_device *peer_device, u64 id, sector_t sector, |
a0fb3c47 | 1756 | struct packet_info *pi) __must_hold(local) |
b411b363 | 1757 | { |
69a22773 | 1758 | struct drbd_device *device = peer_device->device; |
b30ab791 | 1759 | const sector_t capacity = drbd_get_capacity(device->this_bdev); |
db830c46 | 1760 | struct drbd_peer_request *peer_req; |
b411b363 | 1761 | struct page *page; |
11f8b2b6 AG |
1762 | int digest_size, err; |
1763 | unsigned int data_size = pi->size, ds; | |
69a22773 AG |
1764 | void *dig_in = peer_device->connection->int_dig_in; |
1765 | void *dig_vv = peer_device->connection->int_dig_vv; | |
6b4388ac | 1766 | unsigned long *data; |
a0fb3c47 | 1767 | struct p_trim *trim = (pi->cmd == P_TRIM) ? pi->data : NULL; |
9104d31a | 1768 | struct p_trim *wsame = (pi->cmd == P_WSAME) ? pi->data : NULL; |
b411b363 | 1769 | |
11f8b2b6 | 1770 | digest_size = 0; |
a0fb3c47 | 1771 | if (!trim && peer_device->connection->peer_integrity_tfm) { |
3d0e6375 | 1772 | digest_size = crypto_shash_digestsize(peer_device->connection->peer_integrity_tfm); |
9f5bdc33 AG |
1773 | /* |
1774 | * FIXME: Receive the incoming digest into the receive buffer | |
1775 | * here, together with its struct p_data? | |
1776 | */ | |
11f8b2b6 | 1777 | err = drbd_recv_all_warn(peer_device->connection, dig_in, digest_size); |
a5c31904 | 1778 | if (err) |
b411b363 | 1779 | return NULL; |
11f8b2b6 | 1780 | data_size -= digest_size; |
b411b363 PR |
1781 | } |
1782 | ||
9104d31a LE |
1783 | /* assume request_size == data_size, but special case trim and wsame. */ |
1784 | ds = data_size; | |
a0fb3c47 | 1785 | if (trim) { |
9104d31a LE |
1786 | if (!expect(data_size == 0)) |
1787 | return NULL; | |
1788 | ds = be32_to_cpu(trim->size); | |
1789 | } else if (wsame) { | |
1790 | if (data_size != queue_logical_block_size(device->rq_queue)) { | |
1791 | drbd_err(peer_device, "data size (%u) != drbd logical block size (%u)\n", | |
1792 | data_size, queue_logical_block_size(device->rq_queue)); | |
1793 | return NULL; | |
1794 | } | |
1795 | if (data_size != bdev_logical_block_size(device->ldev->backing_bdev)) { | |
1796 | drbd_err(peer_device, "data size (%u) != backend logical block size (%u)\n", | |
1797 | data_size, bdev_logical_block_size(device->ldev->backing_bdev)); | |
1798 | return NULL; | |
1799 | } | |
1800 | ds = be32_to_cpu(wsame->size); | |
a0fb3c47 LE |
1801 | } |
1802 | ||
9104d31a | 1803 | if (!expect(IS_ALIGNED(ds, 512))) |
841ce241 | 1804 | return NULL; |
9104d31a LE |
1805 | if (trim || wsame) { |
1806 | if (!expect(ds <= (DRBD_MAX_BBIO_SECTORS << 9))) | |
1807 | return NULL; | |
1808 | } else if (!expect(ds <= DRBD_MAX_BIO_SIZE)) | |
841ce241 | 1809 | return NULL; |
b411b363 | 1810 | |
6666032a LE |
1811 | /* even though we trust out peer, |
1812 | * we sometimes have to double check. */ | |
9104d31a | 1813 | if (sector + (ds>>9) > capacity) { |
d0180171 | 1814 | drbd_err(device, "request from peer beyond end of local disk: " |
fdda6544 | 1815 | "capacity: %llus < sector: %llus + size: %u\n", |
6666032a | 1816 | (unsigned long long)capacity, |
9104d31a | 1817 | (unsigned long long)sector, ds); |
6666032a LE |
1818 | return NULL; |
1819 | } | |
1820 | ||
b411b363 PR |
1821 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD |
1822 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
1823 | * which in turn might block on the other node at this very place. */ | |
9104d31a | 1824 | peer_req = drbd_alloc_peer_req(peer_device, id, sector, ds, data_size, GFP_NOIO); |
db830c46 | 1825 | if (!peer_req) |
b411b363 | 1826 | return NULL; |
45bb912b | 1827 | |
21ae5d7f | 1828 | peer_req->flags |= EE_WRITE; |
9104d31a LE |
1829 | if (trim) { |
1830 | peer_req->flags |= EE_IS_TRIM; | |
81a3537a | 1831 | return peer_req; |
9104d31a LE |
1832 | } |
1833 | if (wsame) | |
1834 | peer_req->flags |= EE_WRITE_SAME; | |
a73ff323 | 1835 | |
9104d31a | 1836 | /* receive payload size bytes into page chain */ |
b411b363 | 1837 | ds = data_size; |
db830c46 | 1838 | page = peer_req->pages; |
45bb912b LE |
1839 | page_chain_for_each(page) { |
1840 | unsigned len = min_t(int, ds, PAGE_SIZE); | |
6b4388ac | 1841 | data = kmap(page); |
69a22773 | 1842 | err = drbd_recv_all_warn(peer_device->connection, data, len); |
b30ab791 | 1843 | if (drbd_insert_fault(device, DRBD_FAULT_RECEIVE)) { |
d0180171 | 1844 | drbd_err(device, "Fault injection: Corrupting data on receive\n"); |
6b4388ac PR |
1845 | data[0] = data[0] ^ (unsigned long)-1; |
1846 | } | |
b411b363 | 1847 | kunmap(page); |
a5c31904 | 1848 | if (err) { |
b30ab791 | 1849 | drbd_free_peer_req(device, peer_req); |
b411b363 PR |
1850 | return NULL; |
1851 | } | |
a5c31904 | 1852 | ds -= len; |
b411b363 PR |
1853 | } |
1854 | ||
11f8b2b6 | 1855 | if (digest_size) { |
9104d31a | 1856 | drbd_csum_ee_size(peer_device->connection->peer_integrity_tfm, peer_req, dig_vv, data_size); |
11f8b2b6 | 1857 | if (memcmp(dig_in, dig_vv, digest_size)) { |
d0180171 | 1858 | drbd_err(device, "Digest integrity check FAILED: %llus +%u\n", |
470be44a | 1859 | (unsigned long long)sector, data_size); |
b30ab791 | 1860 | drbd_free_peer_req(device, peer_req); |
b411b363 PR |
1861 | return NULL; |
1862 | } | |
1863 | } | |
11f8b2b6 | 1864 | device->recv_cnt += data_size >> 9; |
db830c46 | 1865 | return peer_req; |
b411b363 PR |
1866 | } |
1867 | ||
1868 | /* drbd_drain_block() just takes a data block | |
1869 | * out of the socket input buffer, and discards it. | |
1870 | */ | |
69a22773 | 1871 | static int drbd_drain_block(struct drbd_peer_device *peer_device, int data_size) |
b411b363 PR |
1872 | { |
1873 | struct page *page; | |
a5c31904 | 1874 | int err = 0; |
b411b363 PR |
1875 | void *data; |
1876 | ||
c3470cde | 1877 | if (!data_size) |
fc5be839 | 1878 | return 0; |
c3470cde | 1879 | |
69a22773 | 1880 | page = drbd_alloc_pages(peer_device, 1, 1); |
b411b363 PR |
1881 | |
1882 | data = kmap(page); | |
1883 | while (data_size) { | |
fc5be839 AG |
1884 | unsigned int len = min_t(int, data_size, PAGE_SIZE); |
1885 | ||
69a22773 | 1886 | err = drbd_recv_all_warn(peer_device->connection, data, len); |
a5c31904 | 1887 | if (err) |
b411b363 | 1888 | break; |
a5c31904 | 1889 | data_size -= len; |
b411b363 PR |
1890 | } |
1891 | kunmap(page); | |
69a22773 | 1892 | drbd_free_pages(peer_device->device, page, 0); |
fc5be839 | 1893 | return err; |
b411b363 PR |
1894 | } |
1895 | ||
69a22773 | 1896 | static int recv_dless_read(struct drbd_peer_device *peer_device, struct drbd_request *req, |
b411b363 PR |
1897 | sector_t sector, int data_size) |
1898 | { | |
7988613b KO |
1899 | struct bio_vec bvec; |
1900 | struct bvec_iter iter; | |
b411b363 | 1901 | struct bio *bio; |
11f8b2b6 | 1902 | int digest_size, err, expect; |
69a22773 AG |
1903 | void *dig_in = peer_device->connection->int_dig_in; |
1904 | void *dig_vv = peer_device->connection->int_dig_vv; | |
b411b363 | 1905 | |
11f8b2b6 | 1906 | digest_size = 0; |
69a22773 | 1907 | if (peer_device->connection->peer_integrity_tfm) { |
3d0e6375 | 1908 | digest_size = crypto_shash_digestsize(peer_device->connection->peer_integrity_tfm); |
11f8b2b6 | 1909 | err = drbd_recv_all_warn(peer_device->connection, dig_in, digest_size); |
a5c31904 AG |
1910 | if (err) |
1911 | return err; | |
11f8b2b6 | 1912 | data_size -= digest_size; |
b411b363 PR |
1913 | } |
1914 | ||
b411b363 PR |
1915 | /* optimistically update recv_cnt. if receiving fails below, |
1916 | * we disconnect anyways, and counters will be reset. */ | |
69a22773 | 1917 | peer_device->device->recv_cnt += data_size>>9; |
b411b363 PR |
1918 | |
1919 | bio = req->master_bio; | |
69a22773 | 1920 | D_ASSERT(peer_device->device, sector == bio->bi_iter.bi_sector); |
b411b363 | 1921 | |
7988613b KO |
1922 | bio_for_each_segment(bvec, bio, iter) { |
1923 | void *mapped = kmap(bvec.bv_page) + bvec.bv_offset; | |
1924 | expect = min_t(int, data_size, bvec.bv_len); | |
69a22773 | 1925 | err = drbd_recv_all_warn(peer_device->connection, mapped, expect); |
7988613b | 1926 | kunmap(bvec.bv_page); |
a5c31904 AG |
1927 | if (err) |
1928 | return err; | |
1929 | data_size -= expect; | |
b411b363 PR |
1930 | } |
1931 | ||
11f8b2b6 | 1932 | if (digest_size) { |
69a22773 | 1933 | drbd_csum_bio(peer_device->connection->peer_integrity_tfm, bio, dig_vv); |
11f8b2b6 | 1934 | if (memcmp(dig_in, dig_vv, digest_size)) { |
69a22773 | 1935 | drbd_err(peer_device, "Digest integrity check FAILED. Broken NICs?\n"); |
28284cef | 1936 | return -EINVAL; |
b411b363 PR |
1937 | } |
1938 | } | |
1939 | ||
69a22773 | 1940 | D_ASSERT(peer_device->device, data_size == 0); |
28284cef | 1941 | return 0; |
b411b363 PR |
1942 | } |
1943 | ||
a990be46 | 1944 | /* |
668700b4 | 1945 | * e_end_resync_block() is called in ack_sender context via |
a990be46 AG |
1946 | * drbd_finish_peer_reqs(). |
1947 | */ | |
99920dc5 | 1948 | static int e_end_resync_block(struct drbd_work *w, int unused) |
b411b363 | 1949 | { |
8050e6d0 | 1950 | struct drbd_peer_request *peer_req = |
a8cd15ba AG |
1951 | container_of(w, struct drbd_peer_request, w); |
1952 | struct drbd_peer_device *peer_device = peer_req->peer_device; | |
1953 | struct drbd_device *device = peer_device->device; | |
db830c46 | 1954 | sector_t sector = peer_req->i.sector; |
99920dc5 | 1955 | int err; |
b411b363 | 1956 | |
0b0ba1ef | 1957 | D_ASSERT(device, drbd_interval_empty(&peer_req->i)); |
b411b363 | 1958 | |
db830c46 | 1959 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b30ab791 | 1960 | drbd_set_in_sync(device, sector, peer_req->i.size); |
a8cd15ba | 1961 | err = drbd_send_ack(peer_device, P_RS_WRITE_ACK, peer_req); |
b411b363 PR |
1962 | } else { |
1963 | /* Record failure to sync */ | |
b30ab791 | 1964 | drbd_rs_failed_io(device, sector, peer_req->i.size); |
b411b363 | 1965 | |
a8cd15ba | 1966 | err = drbd_send_ack(peer_device, P_NEG_ACK, peer_req); |
b411b363 | 1967 | } |
b30ab791 | 1968 | dec_unacked(device); |
b411b363 | 1969 | |
99920dc5 | 1970 | return err; |
b411b363 PR |
1971 | } |
1972 | ||
69a22773 | 1973 | static int recv_resync_read(struct drbd_peer_device *peer_device, sector_t sector, |
a0fb3c47 | 1974 | struct packet_info *pi) __releases(local) |
b411b363 | 1975 | { |
69a22773 | 1976 | struct drbd_device *device = peer_device->device; |
db830c46 | 1977 | struct drbd_peer_request *peer_req; |
b411b363 | 1978 | |
a0fb3c47 | 1979 | peer_req = read_in_block(peer_device, ID_SYNCER, sector, pi); |
db830c46 | 1980 | if (!peer_req) |
45bb912b | 1981 | goto fail; |
b411b363 | 1982 | |
b30ab791 | 1983 | dec_rs_pending(device); |
b411b363 | 1984 | |
b30ab791 | 1985 | inc_unacked(device); |
b411b363 PR |
1986 | /* corresponding dec_unacked() in e_end_resync_block() |
1987 | * respective _drbd_clear_done_ee */ | |
1988 | ||
a8cd15ba | 1989 | peer_req->w.cb = e_end_resync_block; |
21ae5d7f | 1990 | peer_req->submit_jif = jiffies; |
45bb912b | 1991 | |
0500813f | 1992 | spin_lock_irq(&device->resource->req_lock); |
b9ed7080 | 1993 | list_add_tail(&peer_req->w.list, &device->sync_ee); |
0500813f | 1994 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 1995 | |
a0fb3c47 | 1996 | atomic_add(pi->size >> 9, &device->rs_sect_ev); |
bb3cc85e MC |
1997 | if (drbd_submit_peer_request(device, peer_req, REQ_OP_WRITE, 0, |
1998 | DRBD_FAULT_RS_WR) == 0) | |
e1c1b0fc | 1999 | return 0; |
b411b363 | 2000 | |
10f6d992 | 2001 | /* don't care for the reason here */ |
d0180171 | 2002 | drbd_err(device, "submit failed, triggering re-connect\n"); |
0500813f | 2003 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 2004 | list_del(&peer_req->w.list); |
0500813f | 2005 | spin_unlock_irq(&device->resource->req_lock); |
22cc37a9 | 2006 | |
b30ab791 | 2007 | drbd_free_peer_req(device, peer_req); |
45bb912b | 2008 | fail: |
b30ab791 | 2009 | put_ldev(device); |
e1c1b0fc | 2010 | return -EIO; |
b411b363 PR |
2011 | } |
2012 | ||
668eebc6 | 2013 | static struct drbd_request * |
b30ab791 | 2014 | find_request(struct drbd_device *device, struct rb_root *root, u64 id, |
bc9c5c41 | 2015 | sector_t sector, bool missing_ok, const char *func) |
51624585 | 2016 | { |
51624585 AG |
2017 | struct drbd_request *req; |
2018 | ||
bc9c5c41 AG |
2019 | /* Request object according to our peer */ |
2020 | req = (struct drbd_request *)(unsigned long)id; | |
5e472264 | 2021 | if (drbd_contains_interval(root, sector, &req->i) && req->i.local) |
668eebc6 | 2022 | return req; |
c3afd8f5 | 2023 | if (!missing_ok) { |
d0180171 | 2024 | drbd_err(device, "%s: failed to find request 0x%lx, sector %llus\n", func, |
c3afd8f5 AG |
2025 | (unsigned long)id, (unsigned long long)sector); |
2026 | } | |
51624585 | 2027 | return NULL; |
b411b363 PR |
2028 | } |
2029 | ||
bde89a9e | 2030 | static int receive_DataReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 2031 | { |
9f4fe9ad | 2032 | struct drbd_peer_device *peer_device; |
b30ab791 | 2033 | struct drbd_device *device; |
b411b363 PR |
2034 | struct drbd_request *req; |
2035 | sector_t sector; | |
82bc0194 | 2036 | int err; |
e658983a | 2037 | struct p_data *p = pi->data; |
4a76b161 | 2038 | |
9f4fe9ad AG |
2039 | peer_device = conn_peer_device(connection, pi->vnr); |
2040 | if (!peer_device) | |
4a76b161 | 2041 | return -EIO; |
9f4fe9ad | 2042 | device = peer_device->device; |
b411b363 PR |
2043 | |
2044 | sector = be64_to_cpu(p->sector); | |
2045 | ||
0500813f | 2046 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 2047 | req = find_request(device, &device->read_requests, p->block_id, sector, false, __func__); |
0500813f | 2048 | spin_unlock_irq(&device->resource->req_lock); |
c3afd8f5 | 2049 | if (unlikely(!req)) |
82bc0194 | 2050 | return -EIO; |
b411b363 | 2051 | |
24c4830c | 2052 | /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid |
b411b363 PR |
2053 | * special casing it there for the various failure cases. |
2054 | * still no race with drbd_fail_pending_reads */ | |
69a22773 | 2055 | err = recv_dless_read(peer_device, req, sector, pi->size); |
82bc0194 | 2056 | if (!err) |
8554df1c | 2057 | req_mod(req, DATA_RECEIVED); |
b411b363 PR |
2058 | /* else: nothing. handled from drbd_disconnect... |
2059 | * I don't think we may complete this just yet | |
2060 | * in case we are "on-disconnect: freeze" */ | |
2061 | ||
82bc0194 | 2062 | return err; |
b411b363 PR |
2063 | } |
2064 | ||
bde89a9e | 2065 | static int receive_RSDataReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 2066 | { |
9f4fe9ad | 2067 | struct drbd_peer_device *peer_device; |
b30ab791 | 2068 | struct drbd_device *device; |
b411b363 | 2069 | sector_t sector; |
82bc0194 | 2070 | int err; |
e658983a | 2071 | struct p_data *p = pi->data; |
4a76b161 | 2072 | |
9f4fe9ad AG |
2073 | peer_device = conn_peer_device(connection, pi->vnr); |
2074 | if (!peer_device) | |
4a76b161 | 2075 | return -EIO; |
9f4fe9ad | 2076 | device = peer_device->device; |
b411b363 PR |
2077 | |
2078 | sector = be64_to_cpu(p->sector); | |
0b0ba1ef | 2079 | D_ASSERT(device, p->block_id == ID_SYNCER); |
b411b363 | 2080 | |
b30ab791 | 2081 | if (get_ldev(device)) { |
b411b363 PR |
2082 | /* data is submitted to disk within recv_resync_read. |
2083 | * corresponding put_ldev done below on error, | |
fcefa62e | 2084 | * or in drbd_peer_request_endio. */ |
a0fb3c47 | 2085 | err = recv_resync_read(peer_device, sector, pi); |
b411b363 PR |
2086 | } else { |
2087 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 2088 | drbd_err(device, "Can not write resync data to local disk.\n"); |
b411b363 | 2089 | |
69a22773 | 2090 | err = drbd_drain_block(peer_device, pi->size); |
b411b363 | 2091 | |
69a22773 | 2092 | drbd_send_ack_dp(peer_device, P_NEG_ACK, p, pi->size); |
b411b363 PR |
2093 | } |
2094 | ||
b30ab791 | 2095 | atomic_add(pi->size >> 9, &device->rs_sect_in); |
778f271d | 2096 | |
82bc0194 | 2097 | return err; |
b411b363 PR |
2098 | } |
2099 | ||
b30ab791 | 2100 | static void restart_conflicting_writes(struct drbd_device *device, |
7be8da07 | 2101 | sector_t sector, int size) |
b411b363 | 2102 | { |
7be8da07 AG |
2103 | struct drbd_interval *i; |
2104 | struct drbd_request *req; | |
2105 | ||
b30ab791 | 2106 | drbd_for_each_overlap(i, &device->write_requests, sector, size) { |
7be8da07 AG |
2107 | if (!i->local) |
2108 | continue; | |
2109 | req = container_of(i, struct drbd_request, i); | |
2110 | if (req->rq_state & RQ_LOCAL_PENDING || | |
2111 | !(req->rq_state & RQ_POSTPONED)) | |
2112 | continue; | |
2312f0b3 LE |
2113 | /* as it is RQ_POSTPONED, this will cause it to |
2114 | * be queued on the retry workqueue. */ | |
d4dabbe2 | 2115 | __req_mod(req, CONFLICT_RESOLVED, NULL); |
7be8da07 AG |
2116 | } |
2117 | } | |
b411b363 | 2118 | |
a990be46 | 2119 | /* |
668700b4 | 2120 | * e_end_block() is called in ack_sender context via drbd_finish_peer_reqs(). |
b411b363 | 2121 | */ |
99920dc5 | 2122 | static int e_end_block(struct drbd_work *w, int cancel) |
b411b363 | 2123 | { |
8050e6d0 | 2124 | struct drbd_peer_request *peer_req = |
a8cd15ba AG |
2125 | container_of(w, struct drbd_peer_request, w); |
2126 | struct drbd_peer_device *peer_device = peer_req->peer_device; | |
2127 | struct drbd_device *device = peer_device->device; | |
db830c46 | 2128 | sector_t sector = peer_req->i.sector; |
99920dc5 | 2129 | int err = 0, pcmd; |
b411b363 | 2130 | |
303d1448 | 2131 | if (peer_req->flags & EE_SEND_WRITE_ACK) { |
db830c46 | 2132 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b30ab791 AG |
2133 | pcmd = (device->state.conn >= C_SYNC_SOURCE && |
2134 | device->state.conn <= C_PAUSED_SYNC_T && | |
db830c46 | 2135 | peer_req->flags & EE_MAY_SET_IN_SYNC) ? |
b411b363 | 2136 | P_RS_WRITE_ACK : P_WRITE_ACK; |
a8cd15ba | 2137 | err = drbd_send_ack(peer_device, pcmd, peer_req); |
b411b363 | 2138 | if (pcmd == P_RS_WRITE_ACK) |
b30ab791 | 2139 | drbd_set_in_sync(device, sector, peer_req->i.size); |
b411b363 | 2140 | } else { |
a8cd15ba | 2141 | err = drbd_send_ack(peer_device, P_NEG_ACK, peer_req); |
b411b363 PR |
2142 | /* we expect it to be marked out of sync anyways... |
2143 | * maybe assert this? */ | |
2144 | } | |
b30ab791 | 2145 | dec_unacked(device); |
b411b363 | 2146 | } |
08d0dabf | 2147 | |
b411b363 PR |
2148 | /* we delete from the conflict detection hash _after_ we sent out the |
2149 | * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ | |
302bdeae | 2150 | if (peer_req->flags & EE_IN_INTERVAL_TREE) { |
0500813f | 2151 | spin_lock_irq(&device->resource->req_lock); |
0b0ba1ef | 2152 | D_ASSERT(device, !drbd_interval_empty(&peer_req->i)); |
b30ab791 | 2153 | drbd_remove_epoch_entry_interval(device, peer_req); |
7be8da07 | 2154 | if (peer_req->flags & EE_RESTART_REQUESTS) |
b30ab791 | 2155 | restart_conflicting_writes(device, sector, peer_req->i.size); |
0500813f | 2156 | spin_unlock_irq(&device->resource->req_lock); |
bb3bfe96 | 2157 | } else |
0b0ba1ef | 2158 | D_ASSERT(device, drbd_interval_empty(&peer_req->i)); |
b411b363 | 2159 | |
5dd2ca19 | 2160 | drbd_may_finish_epoch(peer_device->connection, peer_req->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); |
b411b363 | 2161 | |
99920dc5 | 2162 | return err; |
b411b363 PR |
2163 | } |
2164 | ||
a8cd15ba | 2165 | static int e_send_ack(struct drbd_work *w, enum drbd_packet ack) |
b411b363 | 2166 | { |
8050e6d0 | 2167 | struct drbd_peer_request *peer_req = |
a8cd15ba AG |
2168 | container_of(w, struct drbd_peer_request, w); |
2169 | struct drbd_peer_device *peer_device = peer_req->peer_device; | |
99920dc5 | 2170 | int err; |
b411b363 | 2171 | |
a8cd15ba AG |
2172 | err = drbd_send_ack(peer_device, ack, peer_req); |
2173 | dec_unacked(peer_device->device); | |
b411b363 | 2174 | |
99920dc5 | 2175 | return err; |
b411b363 PR |
2176 | } |
2177 | ||
d4dabbe2 | 2178 | static int e_send_superseded(struct drbd_work *w, int unused) |
7be8da07 | 2179 | { |
a8cd15ba | 2180 | return e_send_ack(w, P_SUPERSEDED); |
7be8da07 AG |
2181 | } |
2182 | ||
99920dc5 | 2183 | static int e_send_retry_write(struct drbd_work *w, int unused) |
7be8da07 | 2184 | { |
a8cd15ba AG |
2185 | struct drbd_peer_request *peer_req = |
2186 | container_of(w, struct drbd_peer_request, w); | |
2187 | struct drbd_connection *connection = peer_req->peer_device->connection; | |
7be8da07 | 2188 | |
a8cd15ba | 2189 | return e_send_ack(w, connection->agreed_pro_version >= 100 ? |
d4dabbe2 | 2190 | P_RETRY_WRITE : P_SUPERSEDED); |
7be8da07 | 2191 | } |
b411b363 | 2192 | |
3e394da1 AG |
2193 | static bool seq_greater(u32 a, u32 b) |
2194 | { | |
2195 | /* | |
2196 | * We assume 32-bit wrap-around here. | |
2197 | * For 24-bit wrap-around, we would have to shift: | |
2198 | * a <<= 8; b <<= 8; | |
2199 | */ | |
2200 | return (s32)a - (s32)b > 0; | |
2201 | } | |
b411b363 | 2202 | |
3e394da1 AG |
2203 | static u32 seq_max(u32 a, u32 b) |
2204 | { | |
2205 | return seq_greater(a, b) ? a : b; | |
b411b363 PR |
2206 | } |
2207 | ||
69a22773 | 2208 | static void update_peer_seq(struct drbd_peer_device *peer_device, unsigned int peer_seq) |
3e394da1 | 2209 | { |
69a22773 | 2210 | struct drbd_device *device = peer_device->device; |
3c13b680 | 2211 | unsigned int newest_peer_seq; |
3e394da1 | 2212 | |
69a22773 | 2213 | if (test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags)) { |
b30ab791 AG |
2214 | spin_lock(&device->peer_seq_lock); |
2215 | newest_peer_seq = seq_max(device->peer_seq, peer_seq); | |
2216 | device->peer_seq = newest_peer_seq; | |
2217 | spin_unlock(&device->peer_seq_lock); | |
2218 | /* wake up only if we actually changed device->peer_seq */ | |
3c13b680 | 2219 | if (peer_seq == newest_peer_seq) |
b30ab791 | 2220 | wake_up(&device->seq_wait); |
7be8da07 | 2221 | } |
b411b363 PR |
2222 | } |
2223 | ||
d93f6302 | 2224 | static inline int overlaps(sector_t s1, int l1, sector_t s2, int l2) |
b6a370ba | 2225 | { |
d93f6302 LE |
2226 | return !((s1 + (l1>>9) <= s2) || (s1 >= s2 + (l2>>9))); |
2227 | } | |
b6a370ba | 2228 | |
d93f6302 | 2229 | /* maybe change sync_ee into interval trees as well? */ |
b30ab791 | 2230 | static bool overlapping_resync_write(struct drbd_device *device, struct drbd_peer_request *peer_req) |
d93f6302 LE |
2231 | { |
2232 | struct drbd_peer_request *rs_req; | |
7e5fec31 | 2233 | bool rv = false; |
b6a370ba | 2234 | |
0500813f | 2235 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 2236 | list_for_each_entry(rs_req, &device->sync_ee, w.list) { |
d93f6302 LE |
2237 | if (overlaps(peer_req->i.sector, peer_req->i.size, |
2238 | rs_req->i.sector, rs_req->i.size)) { | |
7e5fec31 | 2239 | rv = true; |
b6a370ba PR |
2240 | break; |
2241 | } | |
2242 | } | |
0500813f | 2243 | spin_unlock_irq(&device->resource->req_lock); |
b6a370ba PR |
2244 | |
2245 | return rv; | |
2246 | } | |
2247 | ||
b411b363 PR |
2248 | /* Called from receive_Data. |
2249 | * Synchronize packets on sock with packets on msock. | |
2250 | * | |
2251 | * This is here so even when a P_DATA packet traveling via sock overtook an Ack | |
2252 | * packet traveling on msock, they are still processed in the order they have | |
2253 | * been sent. | |
2254 | * | |
2255 | * Note: we don't care for Ack packets overtaking P_DATA packets. | |
2256 | * | |
b30ab791 | 2257 | * In case packet_seq is larger than device->peer_seq number, there are |
b411b363 | 2258 | * outstanding packets on the msock. We wait for them to arrive. |
b30ab791 | 2259 | * In case we are the logically next packet, we update device->peer_seq |
b411b363 PR |
2260 | * ourselves. Correctly handles 32bit wrap around. |
2261 | * | |
2262 | * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, | |
2263 | * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds | |
2264 | * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have | |
2265 | * 1<<9 == 512 seconds aka ages for the 32bit wrap around... | |
2266 | * | |
2267 | * returns 0 if we may process the packet, | |
2268 | * -ERESTARTSYS if we were interrupted (by disconnect signal). */ | |
69a22773 | 2269 | static int wait_for_and_update_peer_seq(struct drbd_peer_device *peer_device, const u32 peer_seq) |
b411b363 | 2270 | { |
69a22773 | 2271 | struct drbd_device *device = peer_device->device; |
b411b363 | 2272 | DEFINE_WAIT(wait); |
b411b363 | 2273 | long timeout; |
b874d231 | 2274 | int ret = 0, tp; |
7be8da07 | 2275 | |
69a22773 | 2276 | if (!test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags)) |
7be8da07 AG |
2277 | return 0; |
2278 | ||
b30ab791 | 2279 | spin_lock(&device->peer_seq_lock); |
b411b363 | 2280 | for (;;) { |
b30ab791 AG |
2281 | if (!seq_greater(peer_seq - 1, device->peer_seq)) { |
2282 | device->peer_seq = seq_max(device->peer_seq, peer_seq); | |
b411b363 | 2283 | break; |
7be8da07 | 2284 | } |
b874d231 | 2285 | |
b411b363 PR |
2286 | if (signal_pending(current)) { |
2287 | ret = -ERESTARTSYS; | |
2288 | break; | |
2289 | } | |
b874d231 PR |
2290 | |
2291 | rcu_read_lock(); | |
5dd2ca19 | 2292 | tp = rcu_dereference(peer_device->connection->net_conf)->two_primaries; |
b874d231 PR |
2293 | rcu_read_unlock(); |
2294 | ||
2295 | if (!tp) | |
2296 | break; | |
2297 | ||
2298 | /* Only need to wait if two_primaries is enabled */ | |
b30ab791 AG |
2299 | prepare_to_wait(&device->seq_wait, &wait, TASK_INTERRUPTIBLE); |
2300 | spin_unlock(&device->peer_seq_lock); | |
44ed167d | 2301 | rcu_read_lock(); |
69a22773 | 2302 | timeout = rcu_dereference(peer_device->connection->net_conf)->ping_timeo*HZ/10; |
44ed167d | 2303 | rcu_read_unlock(); |
71b1c1eb | 2304 | timeout = schedule_timeout(timeout); |
b30ab791 | 2305 | spin_lock(&device->peer_seq_lock); |
7be8da07 | 2306 | if (!timeout) { |
b411b363 | 2307 | ret = -ETIMEDOUT; |
d0180171 | 2308 | drbd_err(device, "Timed out waiting for missing ack packets; disconnecting\n"); |
b411b363 PR |
2309 | break; |
2310 | } | |
2311 | } | |
b30ab791 AG |
2312 | spin_unlock(&device->peer_seq_lock); |
2313 | finish_wait(&device->seq_wait, &wait); | |
b411b363 PR |
2314 | return ret; |
2315 | } | |
2316 | ||
688593c5 LE |
2317 | /* see also bio_flags_to_wire() |
2318 | * DRBD_REQ_*, because we need to semantically map the flags to data packet | |
2319 | * flags and back. We may replicate to other kernel versions. */ | |
bb3cc85e | 2320 | static unsigned long wire_flags_to_bio_flags(u32 dpf) |
76d2e7ec | 2321 | { |
688593c5 LE |
2322 | return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | |
2323 | (dpf & DP_FUA ? REQ_FUA : 0) | | |
28a8f0d3 | 2324 | (dpf & DP_FLUSH ? REQ_PREFLUSH : 0); |
bb3cc85e MC |
2325 | } |
2326 | ||
2327 | static unsigned long wire_flags_to_bio_op(u32 dpf) | |
2328 | { | |
2329 | if (dpf & DP_DISCARD) | |
45c21793 | 2330 | return REQ_OP_WRITE_ZEROES; |
bb3cc85e MC |
2331 | else |
2332 | return REQ_OP_WRITE; | |
76d2e7ec PR |
2333 | } |
2334 | ||
b30ab791 | 2335 | static void fail_postponed_requests(struct drbd_device *device, sector_t sector, |
7be8da07 AG |
2336 | unsigned int size) |
2337 | { | |
2338 | struct drbd_interval *i; | |
2339 | ||
2340 | repeat: | |
b30ab791 | 2341 | drbd_for_each_overlap(i, &device->write_requests, sector, size) { |
7be8da07 AG |
2342 | struct drbd_request *req; |
2343 | struct bio_and_error m; | |
2344 | ||
2345 | if (!i->local) | |
2346 | continue; | |
2347 | req = container_of(i, struct drbd_request, i); | |
2348 | if (!(req->rq_state & RQ_POSTPONED)) | |
2349 | continue; | |
2350 | req->rq_state &= ~RQ_POSTPONED; | |
2351 | __req_mod(req, NEG_ACKED, &m); | |
0500813f | 2352 | spin_unlock_irq(&device->resource->req_lock); |
7be8da07 | 2353 | if (m.bio) |
b30ab791 | 2354 | complete_master_bio(device, &m); |
0500813f | 2355 | spin_lock_irq(&device->resource->req_lock); |
7be8da07 AG |
2356 | goto repeat; |
2357 | } | |
2358 | } | |
2359 | ||
b30ab791 | 2360 | static int handle_write_conflicts(struct drbd_device *device, |
7be8da07 AG |
2361 | struct drbd_peer_request *peer_req) |
2362 | { | |
e33b32de | 2363 | struct drbd_connection *connection = peer_req->peer_device->connection; |
bde89a9e | 2364 | bool resolve_conflicts = test_bit(RESOLVE_CONFLICTS, &connection->flags); |
7be8da07 AG |
2365 | sector_t sector = peer_req->i.sector; |
2366 | const unsigned int size = peer_req->i.size; | |
2367 | struct drbd_interval *i; | |
2368 | bool equal; | |
2369 | int err; | |
2370 | ||
2371 | /* | |
2372 | * Inserting the peer request into the write_requests tree will prevent | |
2373 | * new conflicting local requests from being added. | |
2374 | */ | |
b30ab791 | 2375 | drbd_insert_interval(&device->write_requests, &peer_req->i); |
7be8da07 AG |
2376 | |
2377 | repeat: | |
b30ab791 | 2378 | drbd_for_each_overlap(i, &device->write_requests, sector, size) { |
7be8da07 AG |
2379 | if (i == &peer_req->i) |
2380 | continue; | |
08d0dabf LE |
2381 | if (i->completed) |
2382 | continue; | |
7be8da07 AG |
2383 | |
2384 | if (!i->local) { | |
2385 | /* | |
2386 | * Our peer has sent a conflicting remote request; this | |
2387 | * should not happen in a two-node setup. Wait for the | |
2388 | * earlier peer request to complete. | |
2389 | */ | |
b30ab791 | 2390 | err = drbd_wait_misc(device, i); |
7be8da07 AG |
2391 | if (err) |
2392 | goto out; | |
2393 | goto repeat; | |
2394 | } | |
2395 | ||
2396 | equal = i->sector == sector && i->size == size; | |
2397 | if (resolve_conflicts) { | |
2398 | /* | |
2399 | * If the peer request is fully contained within the | |
d4dabbe2 LE |
2400 | * overlapping request, it can be considered overwritten |
2401 | * and thus superseded; otherwise, it will be retried | |
2402 | * once all overlapping requests have completed. | |
7be8da07 | 2403 | */ |
d4dabbe2 | 2404 | bool superseded = i->sector <= sector && i->sector + |
7be8da07 AG |
2405 | (i->size >> 9) >= sector + (size >> 9); |
2406 | ||
2407 | if (!equal) | |
d0180171 | 2408 | drbd_alert(device, "Concurrent writes detected: " |
7be8da07 AG |
2409 | "local=%llus +%u, remote=%llus +%u, " |
2410 | "assuming %s came first\n", | |
2411 | (unsigned long long)i->sector, i->size, | |
2412 | (unsigned long long)sector, size, | |
d4dabbe2 | 2413 | superseded ? "local" : "remote"); |
7be8da07 | 2414 | |
a8cd15ba | 2415 | peer_req->w.cb = superseded ? e_send_superseded : |
7be8da07 | 2416 | e_send_retry_write; |
a8cd15ba | 2417 | list_add_tail(&peer_req->w.list, &device->done_ee); |
668700b4 | 2418 | queue_work(connection->ack_sender, &peer_req->peer_device->send_acks_work); |
7be8da07 AG |
2419 | |
2420 | err = -ENOENT; | |
2421 | goto out; | |
2422 | } else { | |
2423 | struct drbd_request *req = | |
2424 | container_of(i, struct drbd_request, i); | |
2425 | ||
2426 | if (!equal) | |
d0180171 | 2427 | drbd_alert(device, "Concurrent writes detected: " |
7be8da07 AG |
2428 | "local=%llus +%u, remote=%llus +%u\n", |
2429 | (unsigned long long)i->sector, i->size, | |
2430 | (unsigned long long)sector, size); | |
2431 | ||
2432 | if (req->rq_state & RQ_LOCAL_PENDING || | |
2433 | !(req->rq_state & RQ_POSTPONED)) { | |
2434 | /* | |
2435 | * Wait for the node with the discard flag to | |
d4dabbe2 LE |
2436 | * decide if this request has been superseded |
2437 | * or needs to be retried. | |
2438 | * Requests that have been superseded will | |
7be8da07 AG |
2439 | * disappear from the write_requests tree. |
2440 | * | |
2441 | * In addition, wait for the conflicting | |
2442 | * request to finish locally before submitting | |
2443 | * the conflicting peer request. | |
2444 | */ | |
b30ab791 | 2445 | err = drbd_wait_misc(device, &req->i); |
7be8da07 | 2446 | if (err) { |
e33b32de | 2447 | _conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD); |
b30ab791 | 2448 | fail_postponed_requests(device, sector, size); |
7be8da07 AG |
2449 | goto out; |
2450 | } | |
2451 | goto repeat; | |
2452 | } | |
2453 | /* | |
2454 | * Remember to restart the conflicting requests after | |
2455 | * the new peer request has completed. | |
2456 | */ | |
2457 | peer_req->flags |= EE_RESTART_REQUESTS; | |
2458 | } | |
2459 | } | |
2460 | err = 0; | |
2461 | ||
2462 | out: | |
2463 | if (err) | |
b30ab791 | 2464 | drbd_remove_epoch_entry_interval(device, peer_req); |
7be8da07 AG |
2465 | return err; |
2466 | } | |
2467 | ||
b411b363 | 2468 | /* mirrored write */ |
bde89a9e | 2469 | static int receive_Data(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 2470 | { |
9f4fe9ad | 2471 | struct drbd_peer_device *peer_device; |
b30ab791 | 2472 | struct drbd_device *device; |
21ae5d7f | 2473 | struct net_conf *nc; |
b411b363 | 2474 | sector_t sector; |
db830c46 | 2475 | struct drbd_peer_request *peer_req; |
e658983a | 2476 | struct p_data *p = pi->data; |
7be8da07 | 2477 | u32 peer_seq = be32_to_cpu(p->seq_num); |
bb3cc85e | 2478 | int op, op_flags; |
b411b363 | 2479 | u32 dp_flags; |
302bdeae | 2480 | int err, tp; |
b411b363 | 2481 | |
9f4fe9ad AG |
2482 | peer_device = conn_peer_device(connection, pi->vnr); |
2483 | if (!peer_device) | |
4a76b161 | 2484 | return -EIO; |
9f4fe9ad | 2485 | device = peer_device->device; |
b411b363 | 2486 | |
b30ab791 | 2487 | if (!get_ldev(device)) { |
82bc0194 AG |
2488 | int err2; |
2489 | ||
69a22773 AG |
2490 | err = wait_for_and_update_peer_seq(peer_device, peer_seq); |
2491 | drbd_send_ack_dp(peer_device, P_NEG_ACK, p, pi->size); | |
bde89a9e | 2492 | atomic_inc(&connection->current_epoch->epoch_size); |
69a22773 | 2493 | err2 = drbd_drain_block(peer_device, pi->size); |
82bc0194 AG |
2494 | if (!err) |
2495 | err = err2; | |
2496 | return err; | |
b411b363 PR |
2497 | } |
2498 | ||
fcefa62e AG |
2499 | /* |
2500 | * Corresponding put_ldev done either below (on various errors), or in | |
2501 | * drbd_peer_request_endio, if we successfully submit the data at the | |
2502 | * end of this function. | |
2503 | */ | |
b411b363 PR |
2504 | |
2505 | sector = be64_to_cpu(p->sector); | |
a0fb3c47 | 2506 | peer_req = read_in_block(peer_device, p->block_id, sector, pi); |
db830c46 | 2507 | if (!peer_req) { |
b30ab791 | 2508 | put_ldev(device); |
82bc0194 | 2509 | return -EIO; |
b411b363 PR |
2510 | } |
2511 | ||
a8cd15ba | 2512 | peer_req->w.cb = e_end_block; |
21ae5d7f LE |
2513 | peer_req->submit_jif = jiffies; |
2514 | peer_req->flags |= EE_APPLICATION; | |
b411b363 | 2515 | |
688593c5 | 2516 | dp_flags = be32_to_cpu(p->dp_flags); |
bb3cc85e MC |
2517 | op = wire_flags_to_bio_op(dp_flags); |
2518 | op_flags = wire_flags_to_bio_flags(dp_flags); | |
a0fb3c47 | 2519 | if (pi->cmd == P_TRIM) { |
a0fb3c47 | 2520 | D_ASSERT(peer_device, peer_req->i.size > 0); |
45c21793 | 2521 | D_ASSERT(peer_device, op == REQ_OP_WRITE_ZEROES); |
a0fb3c47 LE |
2522 | D_ASSERT(peer_device, peer_req->pages == NULL); |
2523 | } else if (peer_req->pages == NULL) { | |
0b0ba1ef AG |
2524 | D_ASSERT(device, peer_req->i.size == 0); |
2525 | D_ASSERT(device, dp_flags & DP_FLUSH); | |
a73ff323 | 2526 | } |
688593c5 LE |
2527 | |
2528 | if (dp_flags & DP_MAY_SET_IN_SYNC) | |
db830c46 | 2529 | peer_req->flags |= EE_MAY_SET_IN_SYNC; |
688593c5 | 2530 | |
bde89a9e AG |
2531 | spin_lock(&connection->epoch_lock); |
2532 | peer_req->epoch = connection->current_epoch; | |
db830c46 AG |
2533 | atomic_inc(&peer_req->epoch->epoch_size); |
2534 | atomic_inc(&peer_req->epoch->active); | |
bde89a9e | 2535 | spin_unlock(&connection->epoch_lock); |
b411b363 | 2536 | |
302bdeae | 2537 | rcu_read_lock(); |
21ae5d7f LE |
2538 | nc = rcu_dereference(peer_device->connection->net_conf); |
2539 | tp = nc->two_primaries; | |
2540 | if (peer_device->connection->agreed_pro_version < 100) { | |
2541 | switch (nc->wire_protocol) { | |
2542 | case DRBD_PROT_C: | |
2543 | dp_flags |= DP_SEND_WRITE_ACK; | |
2544 | break; | |
2545 | case DRBD_PROT_B: | |
2546 | dp_flags |= DP_SEND_RECEIVE_ACK; | |
2547 | break; | |
2548 | } | |
2549 | } | |
302bdeae | 2550 | rcu_read_unlock(); |
21ae5d7f LE |
2551 | |
2552 | if (dp_flags & DP_SEND_WRITE_ACK) { | |
2553 | peer_req->flags |= EE_SEND_WRITE_ACK; | |
2554 | inc_unacked(device); | |
2555 | /* corresponding dec_unacked() in e_end_block() | |
2556 | * respective _drbd_clear_done_ee */ | |
2557 | } | |
2558 | ||
2559 | if (dp_flags & DP_SEND_RECEIVE_ACK) { | |
2560 | /* I really don't like it that the receiver thread | |
2561 | * sends on the msock, but anyways */ | |
5dd2ca19 | 2562 | drbd_send_ack(peer_device, P_RECV_ACK, peer_req); |
21ae5d7f LE |
2563 | } |
2564 | ||
302bdeae | 2565 | if (tp) { |
21ae5d7f LE |
2566 | /* two primaries implies protocol C */ |
2567 | D_ASSERT(device, dp_flags & DP_SEND_WRITE_ACK); | |
302bdeae | 2568 | peer_req->flags |= EE_IN_INTERVAL_TREE; |
69a22773 | 2569 | err = wait_for_and_update_peer_seq(peer_device, peer_seq); |
7be8da07 | 2570 | if (err) |
b411b363 | 2571 | goto out_interrupted; |
0500813f | 2572 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 2573 | err = handle_write_conflicts(device, peer_req); |
7be8da07 | 2574 | if (err) { |
0500813f | 2575 | spin_unlock_irq(&device->resource->req_lock); |
7be8da07 | 2576 | if (err == -ENOENT) { |
b30ab791 | 2577 | put_ldev(device); |
82bc0194 | 2578 | return 0; |
b411b363 | 2579 | } |
7be8da07 | 2580 | goto out_interrupted; |
b411b363 | 2581 | } |
b874d231 | 2582 | } else { |
69a22773 | 2583 | update_peer_seq(peer_device, peer_seq); |
0500813f | 2584 | spin_lock_irq(&device->resource->req_lock); |
b874d231 | 2585 | } |
9104d31a LE |
2586 | /* TRIM and WRITE_SAME are processed synchronously, |
2587 | * we wait for all pending requests, respectively wait for | |
a0fb3c47 LE |
2588 | * active_ee to become empty in drbd_submit_peer_request(); |
2589 | * better not add ourselves here. */ | |
9104d31a | 2590 | if ((peer_req->flags & (EE_IS_TRIM|EE_WRITE_SAME)) == 0) |
b9ed7080 | 2591 | list_add_tail(&peer_req->w.list, &device->active_ee); |
0500813f | 2592 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 2593 | |
b30ab791 AG |
2594 | if (device->state.conn == C_SYNC_TARGET) |
2595 | wait_event(device->ee_wait, !overlapping_resync_write(device, peer_req)); | |
b411b363 | 2596 | |
b30ab791 | 2597 | if (device->state.pdsk < D_INCONSISTENT) { |
b411b363 | 2598 | /* In case we have the only disk of the cluster, */ |
b30ab791 | 2599 | drbd_set_out_of_sync(device, peer_req->i.sector, peer_req->i.size); |
db830c46 | 2600 | peer_req->flags &= ~EE_MAY_SET_IN_SYNC; |
4dd726f0 | 2601 | drbd_al_begin_io(device, &peer_req->i); |
21ae5d7f | 2602 | peer_req->flags |= EE_CALL_AL_COMPLETE_IO; |
b411b363 PR |
2603 | } |
2604 | ||
bb3cc85e MC |
2605 | err = drbd_submit_peer_request(device, peer_req, op, op_flags, |
2606 | DRBD_FAULT_DT_WR); | |
82bc0194 AG |
2607 | if (!err) |
2608 | return 0; | |
b411b363 | 2609 | |
10f6d992 | 2610 | /* don't care for the reason here */ |
d0180171 | 2611 | drbd_err(device, "submit failed, triggering re-connect\n"); |
0500813f | 2612 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 2613 | list_del(&peer_req->w.list); |
b30ab791 | 2614 | drbd_remove_epoch_entry_interval(device, peer_req); |
0500813f | 2615 | spin_unlock_irq(&device->resource->req_lock); |
21ae5d7f LE |
2616 | if (peer_req->flags & EE_CALL_AL_COMPLETE_IO) { |
2617 | peer_req->flags &= ~EE_CALL_AL_COMPLETE_IO; | |
b30ab791 | 2618 | drbd_al_complete_io(device, &peer_req->i); |
21ae5d7f | 2619 | } |
22cc37a9 | 2620 | |
b411b363 | 2621 | out_interrupted: |
7e5fec31 | 2622 | drbd_may_finish_epoch(connection, peer_req->epoch, EV_PUT | EV_CLEANUP); |
b30ab791 AG |
2623 | put_ldev(device); |
2624 | drbd_free_peer_req(device, peer_req); | |
82bc0194 | 2625 | return err; |
b411b363 PR |
2626 | } |
2627 | ||
0f0601f4 LE |
2628 | /* We may throttle resync, if the lower device seems to be busy, |
2629 | * and current sync rate is above c_min_rate. | |
2630 | * | |
2631 | * To decide whether or not the lower device is busy, we use a scheme similar | |
2632 | * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" | |
2633 | * (more than 64 sectors) of activity we cannot account for with our own resync | |
2634 | * activity, it obviously is "busy". | |
2635 | * | |
2636 | * The current sync rate used here uses only the most recent two step marks, | |
2637 | * to have a short time average so we can react faster. | |
2638 | */ | |
ad3fee79 LE |
2639 | bool drbd_rs_should_slow_down(struct drbd_device *device, sector_t sector, |
2640 | bool throttle_if_app_is_waiting) | |
0f0601f4 | 2641 | { |
e3555d85 | 2642 | struct lc_element *tmp; |
ad3fee79 | 2643 | bool throttle = drbd_rs_c_min_rate_throttle(device); |
daeda1cc | 2644 | |
ad3fee79 LE |
2645 | if (!throttle || throttle_if_app_is_waiting) |
2646 | return throttle; | |
0f0601f4 | 2647 | |
b30ab791 AG |
2648 | spin_lock_irq(&device->al_lock); |
2649 | tmp = lc_find(device->resync, BM_SECT_TO_EXT(sector)); | |
e3555d85 PR |
2650 | if (tmp) { |
2651 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
e8299874 LE |
2652 | if (test_bit(BME_PRIORITY, &bm_ext->flags)) |
2653 | throttle = false; | |
ad3fee79 LE |
2654 | /* Do not slow down if app IO is already waiting for this extent, |
2655 | * and our progress is necessary for application IO to complete. */ | |
e3555d85 | 2656 | } |
b30ab791 | 2657 | spin_unlock_irq(&device->al_lock); |
e3555d85 | 2658 | |
e8299874 LE |
2659 | return throttle; |
2660 | } | |
2661 | ||
2662 | bool drbd_rs_c_min_rate_throttle(struct drbd_device *device) | |
2663 | { | |
2664 | struct gendisk *disk = device->ldev->backing_bdev->bd_contains->bd_disk; | |
2665 | unsigned long db, dt, dbdt; | |
2666 | unsigned int c_min_rate; | |
2667 | int curr_events; | |
2668 | ||
2669 | rcu_read_lock(); | |
2670 | c_min_rate = rcu_dereference(device->ldev->disk_conf)->c_min_rate; | |
2671 | rcu_read_unlock(); | |
2672 | ||
2673 | /* feature disabled? */ | |
2674 | if (c_min_rate == 0) | |
2675 | return false; | |
2676 | ||
59767fbd | 2677 | curr_events = (int)part_stat_read_accum(&disk->part0, sectors) - |
b30ab791 | 2678 | atomic_read(&device->rs_sect_ev); |
ad3fee79 LE |
2679 | |
2680 | if (atomic_read(&device->ap_actlog_cnt) | |
ff8bd88b | 2681 | || curr_events - device->rs_last_events > 64) { |
0f0601f4 LE |
2682 | unsigned long rs_left; |
2683 | int i; | |
2684 | ||
b30ab791 | 2685 | device->rs_last_events = curr_events; |
0f0601f4 LE |
2686 | |
2687 | /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, | |
2688 | * approx. */ | |
b30ab791 | 2689 | i = (device->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS; |
2649f080 | 2690 | |
b30ab791 AG |
2691 | if (device->state.conn == C_VERIFY_S || device->state.conn == C_VERIFY_T) |
2692 | rs_left = device->ov_left; | |
2649f080 | 2693 | else |
b30ab791 | 2694 | rs_left = drbd_bm_total_weight(device) - device->rs_failed; |
0f0601f4 | 2695 | |
b30ab791 | 2696 | dt = ((long)jiffies - (long)device->rs_mark_time[i]) / HZ; |
0f0601f4 LE |
2697 | if (!dt) |
2698 | dt++; | |
b30ab791 | 2699 | db = device->rs_mark_left[i] - rs_left; |
0f0601f4 LE |
2700 | dbdt = Bit2KB(db/dt); |
2701 | ||
daeda1cc | 2702 | if (dbdt > c_min_rate) |
e8299874 | 2703 | return true; |
0f0601f4 | 2704 | } |
e8299874 | 2705 | return false; |
0f0601f4 LE |
2706 | } |
2707 | ||
bde89a9e | 2708 | static int receive_DataRequest(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 2709 | { |
9f4fe9ad | 2710 | struct drbd_peer_device *peer_device; |
b30ab791 | 2711 | struct drbd_device *device; |
b411b363 | 2712 | sector_t sector; |
4a76b161 | 2713 | sector_t capacity; |
db830c46 | 2714 | struct drbd_peer_request *peer_req; |
b411b363 | 2715 | struct digest_info *di = NULL; |
b18b37be | 2716 | int size, verb; |
b411b363 | 2717 | unsigned int fault_type; |
e658983a | 2718 | struct p_block_req *p = pi->data; |
4a76b161 | 2719 | |
9f4fe9ad AG |
2720 | peer_device = conn_peer_device(connection, pi->vnr); |
2721 | if (!peer_device) | |
4a76b161 | 2722 | return -EIO; |
9f4fe9ad | 2723 | device = peer_device->device; |
b30ab791 | 2724 | capacity = drbd_get_capacity(device->this_bdev); |
b411b363 PR |
2725 | |
2726 | sector = be64_to_cpu(p->sector); | |
2727 | size = be32_to_cpu(p->blksize); | |
2728 | ||
c670a398 | 2729 | if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { |
d0180171 | 2730 | drbd_err(device, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, |
b411b363 | 2731 | (unsigned long long)sector, size); |
82bc0194 | 2732 | return -EINVAL; |
b411b363 PR |
2733 | } |
2734 | if (sector + (size>>9) > capacity) { | |
d0180171 | 2735 | drbd_err(device, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, |
b411b363 | 2736 | (unsigned long long)sector, size); |
82bc0194 | 2737 | return -EINVAL; |
b411b363 PR |
2738 | } |
2739 | ||
b30ab791 | 2740 | if (!get_ldev_if_state(device, D_UP_TO_DATE)) { |
b18b37be | 2741 | verb = 1; |
e2857216 | 2742 | switch (pi->cmd) { |
b18b37be | 2743 | case P_DATA_REQUEST: |
69a22773 | 2744 | drbd_send_ack_rp(peer_device, P_NEG_DREPLY, p); |
b18b37be | 2745 | break; |
700ca8c0 | 2746 | case P_RS_THIN_REQ: |
b18b37be PR |
2747 | case P_RS_DATA_REQUEST: |
2748 | case P_CSUM_RS_REQUEST: | |
2749 | case P_OV_REQUEST: | |
69a22773 | 2750 | drbd_send_ack_rp(peer_device, P_NEG_RS_DREPLY , p); |
b18b37be PR |
2751 | break; |
2752 | case P_OV_REPLY: | |
2753 | verb = 0; | |
b30ab791 | 2754 | dec_rs_pending(device); |
69a22773 | 2755 | drbd_send_ack_ex(peer_device, P_OV_RESULT, sector, size, ID_IN_SYNC); |
b18b37be PR |
2756 | break; |
2757 | default: | |
49ba9b1b | 2758 | BUG(); |
b18b37be PR |
2759 | } |
2760 | if (verb && __ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 2761 | drbd_err(device, "Can not satisfy peer's read request, " |
b411b363 | 2762 | "no local data.\n"); |
b18b37be | 2763 | |
a821cc4a | 2764 | /* drain possibly payload */ |
69a22773 | 2765 | return drbd_drain_block(peer_device, pi->size); |
b411b363 PR |
2766 | } |
2767 | ||
2768 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD | |
2769 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
2770 | * which in turn might block on the other node at this very place. */ | |
a0fb3c47 | 2771 | peer_req = drbd_alloc_peer_req(peer_device, p->block_id, sector, size, |
9104d31a | 2772 | size, GFP_NOIO); |
db830c46 | 2773 | if (!peer_req) { |
b30ab791 | 2774 | put_ldev(device); |
82bc0194 | 2775 | return -ENOMEM; |
b411b363 PR |
2776 | } |
2777 | ||
e2857216 | 2778 | switch (pi->cmd) { |
b411b363 | 2779 | case P_DATA_REQUEST: |
a8cd15ba | 2780 | peer_req->w.cb = w_e_end_data_req; |
b411b363 | 2781 | fault_type = DRBD_FAULT_DT_RD; |
80a40e43 | 2782 | /* application IO, don't drbd_rs_begin_io */ |
21ae5d7f | 2783 | peer_req->flags |= EE_APPLICATION; |
80a40e43 LE |
2784 | goto submit; |
2785 | ||
700ca8c0 PR |
2786 | case P_RS_THIN_REQ: |
2787 | /* If at some point in the future we have a smart way to | |
2788 | find out if this data block is completely deallocated, | |
2789 | then we would do something smarter here than reading | |
2790 | the block... */ | |
2791 | peer_req->flags |= EE_RS_THIN_REQ; | |
d769a992 | 2792 | /* fall through */ |
b411b363 | 2793 | case P_RS_DATA_REQUEST: |
a8cd15ba | 2794 | peer_req->w.cb = w_e_end_rsdata_req; |
b411b363 | 2795 | fault_type = DRBD_FAULT_RS_RD; |
5f9915bb | 2796 | /* used in the sector offset progress display */ |
b30ab791 | 2797 | device->bm_resync_fo = BM_SECT_TO_BIT(sector); |
b411b363 PR |
2798 | break; |
2799 | ||
2800 | case P_OV_REPLY: | |
2801 | case P_CSUM_RS_REQUEST: | |
2802 | fault_type = DRBD_FAULT_RS_RD; | |
e2857216 | 2803 | di = kmalloc(sizeof(*di) + pi->size, GFP_NOIO); |
b411b363 PR |
2804 | if (!di) |
2805 | goto out_free_e; | |
2806 | ||
e2857216 | 2807 | di->digest_size = pi->size; |
b411b363 PR |
2808 | di->digest = (((char *)di)+sizeof(struct digest_info)); |
2809 | ||
db830c46 AG |
2810 | peer_req->digest = di; |
2811 | peer_req->flags |= EE_HAS_DIGEST; | |
c36c3ced | 2812 | |
9f4fe9ad | 2813 | if (drbd_recv_all(peer_device->connection, di->digest, pi->size)) |
b411b363 PR |
2814 | goto out_free_e; |
2815 | ||
e2857216 | 2816 | if (pi->cmd == P_CSUM_RS_REQUEST) { |
9f4fe9ad | 2817 | D_ASSERT(device, peer_device->connection->agreed_pro_version >= 89); |
a8cd15ba | 2818 | peer_req->w.cb = w_e_end_csum_rs_req; |
5f9915bb | 2819 | /* used in the sector offset progress display */ |
b30ab791 | 2820 | device->bm_resync_fo = BM_SECT_TO_BIT(sector); |
aaaba345 LE |
2821 | /* remember to report stats in drbd_resync_finished */ |
2822 | device->use_csums = true; | |
e2857216 | 2823 | } else if (pi->cmd == P_OV_REPLY) { |
2649f080 | 2824 | /* track progress, we may need to throttle */ |
b30ab791 | 2825 | atomic_add(size >> 9, &device->rs_sect_in); |
a8cd15ba | 2826 | peer_req->w.cb = w_e_end_ov_reply; |
b30ab791 | 2827 | dec_rs_pending(device); |
0f0601f4 LE |
2828 | /* drbd_rs_begin_io done when we sent this request, |
2829 | * but accounting still needs to be done. */ | |
2830 | goto submit_for_resync; | |
b411b363 PR |
2831 | } |
2832 | break; | |
2833 | ||
2834 | case P_OV_REQUEST: | |
b30ab791 | 2835 | if (device->ov_start_sector == ~(sector_t)0 && |
9f4fe9ad | 2836 | peer_device->connection->agreed_pro_version >= 90) { |
de228bba LE |
2837 | unsigned long now = jiffies; |
2838 | int i; | |
b30ab791 AG |
2839 | device->ov_start_sector = sector; |
2840 | device->ov_position = sector; | |
2841 | device->ov_left = drbd_bm_bits(device) - BM_SECT_TO_BIT(sector); | |
2842 | device->rs_total = device->ov_left; | |
de228bba | 2843 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
2844 | device->rs_mark_left[i] = device->ov_left; |
2845 | device->rs_mark_time[i] = now; | |
de228bba | 2846 | } |
d0180171 | 2847 | drbd_info(device, "Online Verify start sector: %llu\n", |
b411b363 PR |
2848 | (unsigned long long)sector); |
2849 | } | |
a8cd15ba | 2850 | peer_req->w.cb = w_e_end_ov_req; |
b411b363 | 2851 | fault_type = DRBD_FAULT_RS_RD; |
b411b363 PR |
2852 | break; |
2853 | ||
b411b363 | 2854 | default: |
49ba9b1b | 2855 | BUG(); |
b411b363 PR |
2856 | } |
2857 | ||
0f0601f4 LE |
2858 | /* Throttle, drbd_rs_begin_io and submit should become asynchronous |
2859 | * wrt the receiver, but it is not as straightforward as it may seem. | |
2860 | * Various places in the resync start and stop logic assume resync | |
2861 | * requests are processed in order, requeuing this on the worker thread | |
2862 | * introduces a bunch of new code for synchronization between threads. | |
2863 | * | |
2864 | * Unlimited throttling before drbd_rs_begin_io may stall the resync | |
2865 | * "forever", throttling after drbd_rs_begin_io will lock that extent | |
2866 | * for application writes for the same time. For now, just throttle | |
2867 | * here, where the rest of the code expects the receiver to sleep for | |
2868 | * a while, anyways. | |
2869 | */ | |
2870 | ||
2871 | /* Throttle before drbd_rs_begin_io, as that locks out application IO; | |
2872 | * this defers syncer requests for some time, before letting at least | |
2873 | * on request through. The resync controller on the receiving side | |
2874 | * will adapt to the incoming rate accordingly. | |
2875 | * | |
2876 | * We cannot throttle here if remote is Primary/SyncTarget: | |
2877 | * we would also throttle its application reads. | |
2878 | * In that case, throttling is done on the SyncTarget only. | |
2879 | */ | |
c5a2c150 LE |
2880 | |
2881 | /* Even though this may be a resync request, we do add to "read_ee"; | |
2882 | * "sync_ee" is only used for resync WRITEs. | |
2883 | * Add to list early, so debugfs can find this request | |
2884 | * even if we have to sleep below. */ | |
2885 | spin_lock_irq(&device->resource->req_lock); | |
2886 | list_add_tail(&peer_req->w.list, &device->read_ee); | |
2887 | spin_unlock_irq(&device->resource->req_lock); | |
2888 | ||
944410e9 | 2889 | update_receiver_timing_details(connection, drbd_rs_should_slow_down); |
ad3fee79 LE |
2890 | if (device->state.peer != R_PRIMARY |
2891 | && drbd_rs_should_slow_down(device, sector, false)) | |
e3555d85 | 2892 | schedule_timeout_uninterruptible(HZ/10); |
944410e9 | 2893 | update_receiver_timing_details(connection, drbd_rs_begin_io); |
b30ab791 | 2894 | if (drbd_rs_begin_io(device, sector)) |
80a40e43 | 2895 | goto out_free_e; |
b411b363 | 2896 | |
0f0601f4 | 2897 | submit_for_resync: |
b30ab791 | 2898 | atomic_add(size >> 9, &device->rs_sect_ev); |
0f0601f4 | 2899 | |
80a40e43 | 2900 | submit: |
944410e9 | 2901 | update_receiver_timing_details(connection, drbd_submit_peer_request); |
b30ab791 | 2902 | inc_unacked(device); |
bb3cc85e MC |
2903 | if (drbd_submit_peer_request(device, peer_req, REQ_OP_READ, 0, |
2904 | fault_type) == 0) | |
82bc0194 | 2905 | return 0; |
b411b363 | 2906 | |
10f6d992 | 2907 | /* don't care for the reason here */ |
d0180171 | 2908 | drbd_err(device, "submit failed, triggering re-connect\n"); |
c5a2c150 LE |
2909 | |
2910 | out_free_e: | |
0500813f | 2911 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 2912 | list_del(&peer_req->w.list); |
0500813f | 2913 | spin_unlock_irq(&device->resource->req_lock); |
22cc37a9 LE |
2914 | /* no drbd_rs_complete_io(), we are dropping the connection anyways */ |
2915 | ||
b30ab791 AG |
2916 | put_ldev(device); |
2917 | drbd_free_peer_req(device, peer_req); | |
82bc0194 | 2918 | return -EIO; |
b411b363 PR |
2919 | } |
2920 | ||
69a22773 AG |
2921 | /** |
2922 | * drbd_asb_recover_0p - Recover after split-brain with no remaining primaries | |
2923 | */ | |
2924 | static int drbd_asb_recover_0p(struct drbd_peer_device *peer_device) __must_hold(local) | |
b411b363 | 2925 | { |
69a22773 | 2926 | struct drbd_device *device = peer_device->device; |
b411b363 PR |
2927 | int self, peer, rv = -100; |
2928 | unsigned long ch_self, ch_peer; | |
44ed167d | 2929 | enum drbd_after_sb_p after_sb_0p; |
b411b363 | 2930 | |
b30ab791 AG |
2931 | self = device->ldev->md.uuid[UI_BITMAP] & 1; |
2932 | peer = device->p_uuid[UI_BITMAP] & 1; | |
b411b363 | 2933 | |
b30ab791 AG |
2934 | ch_peer = device->p_uuid[UI_SIZE]; |
2935 | ch_self = device->comm_bm_set; | |
b411b363 | 2936 | |
44ed167d | 2937 | rcu_read_lock(); |
69a22773 | 2938 | after_sb_0p = rcu_dereference(peer_device->connection->net_conf)->after_sb_0p; |
44ed167d PR |
2939 | rcu_read_unlock(); |
2940 | switch (after_sb_0p) { | |
b411b363 PR |
2941 | case ASB_CONSENSUS: |
2942 | case ASB_DISCARD_SECONDARY: | |
2943 | case ASB_CALL_HELPER: | |
44ed167d | 2944 | case ASB_VIOLENTLY: |
d0180171 | 2945 | drbd_err(device, "Configuration error.\n"); |
b411b363 PR |
2946 | break; |
2947 | case ASB_DISCONNECT: | |
2948 | break; | |
2949 | case ASB_DISCARD_YOUNGER_PRI: | |
2950 | if (self == 0 && peer == 1) { | |
2951 | rv = -1; | |
2952 | break; | |
2953 | } | |
2954 | if (self == 1 && peer == 0) { | |
2955 | rv = 1; | |
2956 | break; | |
2957 | } | |
2958 | /* Else fall through to one of the other strategies... */ | |
2959 | case ASB_DISCARD_OLDER_PRI: | |
2960 | if (self == 0 && peer == 1) { | |
2961 | rv = 1; | |
2962 | break; | |
2963 | } | |
2964 | if (self == 1 && peer == 0) { | |
2965 | rv = -1; | |
2966 | break; | |
2967 | } | |
2968 | /* Else fall through to one of the other strategies... */ | |
d0180171 | 2969 | drbd_warn(device, "Discard younger/older primary did not find a decision\n" |
b411b363 | 2970 | "Using discard-least-changes instead\n"); |
d769a992 | 2971 | /* fall through */ |
b411b363 PR |
2972 | case ASB_DISCARD_ZERO_CHG: |
2973 | if (ch_peer == 0 && ch_self == 0) { | |
69a22773 | 2974 | rv = test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags) |
b411b363 PR |
2975 | ? -1 : 1; |
2976 | break; | |
2977 | } else { | |
2978 | if (ch_peer == 0) { rv = 1; break; } | |
2979 | if (ch_self == 0) { rv = -1; break; } | |
2980 | } | |
44ed167d | 2981 | if (after_sb_0p == ASB_DISCARD_ZERO_CHG) |
b411b363 | 2982 | break; |
d769a992 | 2983 | /* else: fall through */ |
b411b363 PR |
2984 | case ASB_DISCARD_LEAST_CHG: |
2985 | if (ch_self < ch_peer) | |
2986 | rv = -1; | |
2987 | else if (ch_self > ch_peer) | |
2988 | rv = 1; | |
2989 | else /* ( ch_self == ch_peer ) */ | |
2990 | /* Well, then use something else. */ | |
69a22773 | 2991 | rv = test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags) |
b411b363 PR |
2992 | ? -1 : 1; |
2993 | break; | |
2994 | case ASB_DISCARD_LOCAL: | |
2995 | rv = -1; | |
2996 | break; | |
2997 | case ASB_DISCARD_REMOTE: | |
2998 | rv = 1; | |
2999 | } | |
3000 | ||
3001 | return rv; | |
3002 | } | |
3003 | ||
69a22773 AG |
3004 | /** |
3005 | * drbd_asb_recover_1p - Recover after split-brain with one remaining primary | |
3006 | */ | |
3007 | static int drbd_asb_recover_1p(struct drbd_peer_device *peer_device) __must_hold(local) | |
b411b363 | 3008 | { |
69a22773 | 3009 | struct drbd_device *device = peer_device->device; |
6184ea21 | 3010 | int hg, rv = -100; |
44ed167d | 3011 | enum drbd_after_sb_p after_sb_1p; |
b411b363 | 3012 | |
44ed167d | 3013 | rcu_read_lock(); |
69a22773 | 3014 | after_sb_1p = rcu_dereference(peer_device->connection->net_conf)->after_sb_1p; |
44ed167d PR |
3015 | rcu_read_unlock(); |
3016 | switch (after_sb_1p) { | |
b411b363 PR |
3017 | case ASB_DISCARD_YOUNGER_PRI: |
3018 | case ASB_DISCARD_OLDER_PRI: | |
3019 | case ASB_DISCARD_LEAST_CHG: | |
3020 | case ASB_DISCARD_LOCAL: | |
3021 | case ASB_DISCARD_REMOTE: | |
44ed167d | 3022 | case ASB_DISCARD_ZERO_CHG: |
d0180171 | 3023 | drbd_err(device, "Configuration error.\n"); |
b411b363 PR |
3024 | break; |
3025 | case ASB_DISCONNECT: | |
3026 | break; | |
3027 | case ASB_CONSENSUS: | |
69a22773 | 3028 | hg = drbd_asb_recover_0p(peer_device); |
b30ab791 | 3029 | if (hg == -1 && device->state.role == R_SECONDARY) |
b411b363 | 3030 | rv = hg; |
b30ab791 | 3031 | if (hg == 1 && device->state.role == R_PRIMARY) |
b411b363 PR |
3032 | rv = hg; |
3033 | break; | |
3034 | case ASB_VIOLENTLY: | |
69a22773 | 3035 | rv = drbd_asb_recover_0p(peer_device); |
b411b363 PR |
3036 | break; |
3037 | case ASB_DISCARD_SECONDARY: | |
b30ab791 | 3038 | return device->state.role == R_PRIMARY ? 1 : -1; |
b411b363 | 3039 | case ASB_CALL_HELPER: |
69a22773 | 3040 | hg = drbd_asb_recover_0p(peer_device); |
b30ab791 | 3041 | if (hg == -1 && device->state.role == R_PRIMARY) { |
bb437946 AG |
3042 | enum drbd_state_rv rv2; |
3043 | ||
b411b363 PR |
3044 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
3045 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
3046 | * we do not need to wait for the after state change work either. */ | |
b30ab791 | 3047 | rv2 = drbd_change_state(device, CS_VERBOSE, NS(role, R_SECONDARY)); |
bb437946 | 3048 | if (rv2 != SS_SUCCESS) { |
b30ab791 | 3049 | drbd_khelper(device, "pri-lost-after-sb"); |
b411b363 | 3050 | } else { |
d0180171 | 3051 | drbd_warn(device, "Successfully gave up primary role.\n"); |
b411b363 PR |
3052 | rv = hg; |
3053 | } | |
3054 | } else | |
3055 | rv = hg; | |
3056 | } | |
3057 | ||
3058 | return rv; | |
3059 | } | |
3060 | ||
69a22773 AG |
3061 | /** |
3062 | * drbd_asb_recover_2p - Recover after split-brain with two remaining primaries | |
3063 | */ | |
3064 | static int drbd_asb_recover_2p(struct drbd_peer_device *peer_device) __must_hold(local) | |
b411b363 | 3065 | { |
69a22773 | 3066 | struct drbd_device *device = peer_device->device; |
6184ea21 | 3067 | int hg, rv = -100; |
44ed167d | 3068 | enum drbd_after_sb_p after_sb_2p; |
b411b363 | 3069 | |
44ed167d | 3070 | rcu_read_lock(); |
69a22773 | 3071 | after_sb_2p = rcu_dereference(peer_device->connection->net_conf)->after_sb_2p; |
44ed167d PR |
3072 | rcu_read_unlock(); |
3073 | switch (after_sb_2p) { | |
b411b363 PR |
3074 | case ASB_DISCARD_YOUNGER_PRI: |
3075 | case ASB_DISCARD_OLDER_PRI: | |
3076 | case ASB_DISCARD_LEAST_CHG: | |
3077 | case ASB_DISCARD_LOCAL: | |
3078 | case ASB_DISCARD_REMOTE: | |
3079 | case ASB_CONSENSUS: | |
3080 | case ASB_DISCARD_SECONDARY: | |
44ed167d | 3081 | case ASB_DISCARD_ZERO_CHG: |
d0180171 | 3082 | drbd_err(device, "Configuration error.\n"); |
b411b363 PR |
3083 | break; |
3084 | case ASB_VIOLENTLY: | |
69a22773 | 3085 | rv = drbd_asb_recover_0p(peer_device); |
b411b363 PR |
3086 | break; |
3087 | case ASB_DISCONNECT: | |
3088 | break; | |
3089 | case ASB_CALL_HELPER: | |
69a22773 | 3090 | hg = drbd_asb_recover_0p(peer_device); |
b411b363 | 3091 | if (hg == -1) { |
bb437946 AG |
3092 | enum drbd_state_rv rv2; |
3093 | ||
b411b363 PR |
3094 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
3095 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
3096 | * we do not need to wait for the after state change work either. */ | |
b30ab791 | 3097 | rv2 = drbd_change_state(device, CS_VERBOSE, NS(role, R_SECONDARY)); |
bb437946 | 3098 | if (rv2 != SS_SUCCESS) { |
b30ab791 | 3099 | drbd_khelper(device, "pri-lost-after-sb"); |
b411b363 | 3100 | } else { |
d0180171 | 3101 | drbd_warn(device, "Successfully gave up primary role.\n"); |
b411b363 PR |
3102 | rv = hg; |
3103 | } | |
3104 | } else | |
3105 | rv = hg; | |
3106 | } | |
3107 | ||
3108 | return rv; | |
3109 | } | |
3110 | ||
b30ab791 | 3111 | static void drbd_uuid_dump(struct drbd_device *device, char *text, u64 *uuid, |
b411b363 PR |
3112 | u64 bits, u64 flags) |
3113 | { | |
3114 | if (!uuid) { | |
d0180171 | 3115 | drbd_info(device, "%s uuid info vanished while I was looking!\n", text); |
b411b363 PR |
3116 | return; |
3117 | } | |
d0180171 | 3118 | drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", |
b411b363 PR |
3119 | text, |
3120 | (unsigned long long)uuid[UI_CURRENT], | |
3121 | (unsigned long long)uuid[UI_BITMAP], | |
3122 | (unsigned long long)uuid[UI_HISTORY_START], | |
3123 | (unsigned long long)uuid[UI_HISTORY_END], | |
3124 | (unsigned long long)bits, | |
3125 | (unsigned long long)flags); | |
3126 | } | |
3127 | ||
3128 | /* | |
3129 | 100 after split brain try auto recover | |
3130 | 2 C_SYNC_SOURCE set BitMap | |
3131 | 1 C_SYNC_SOURCE use BitMap | |
3132 | 0 no Sync | |
3133 | -1 C_SYNC_TARGET use BitMap | |
3134 | -2 C_SYNC_TARGET set BitMap | |
3135 | -100 after split brain, disconnect | |
3136 | -1000 unrelated data | |
4a23f264 PR |
3137 | -1091 requires proto 91 |
3138 | -1096 requires proto 96 | |
b411b363 | 3139 | */ |
f2d3d75b LE |
3140 | |
3141 | static int drbd_uuid_compare(struct drbd_device *const device, enum drbd_role const peer_role, int *rule_nr) __must_hold(local) | |
b411b363 | 3142 | { |
44a4d551 LE |
3143 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
3144 | struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL; | |
b411b363 PR |
3145 | u64 self, peer; |
3146 | int i, j; | |
3147 | ||
b30ab791 AG |
3148 | self = device->ldev->md.uuid[UI_CURRENT] & ~((u64)1); |
3149 | peer = device->p_uuid[UI_CURRENT] & ~((u64)1); | |
b411b363 PR |
3150 | |
3151 | *rule_nr = 10; | |
3152 | if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) | |
3153 | return 0; | |
3154 | ||
3155 | *rule_nr = 20; | |
3156 | if ((self == UUID_JUST_CREATED || self == (u64)0) && | |
3157 | peer != UUID_JUST_CREATED) | |
3158 | return -2; | |
3159 | ||
3160 | *rule_nr = 30; | |
3161 | if (self != UUID_JUST_CREATED && | |
3162 | (peer == UUID_JUST_CREATED || peer == (u64)0)) | |
3163 | return 2; | |
3164 | ||
3165 | if (self == peer) { | |
3166 | int rct, dc; /* roles at crash time */ | |
3167 | ||
b30ab791 | 3168 | if (device->p_uuid[UI_BITMAP] == (u64)0 && device->ldev->md.uuid[UI_BITMAP] != (u64)0) { |
b411b363 | 3169 | |
44a4d551 | 3170 | if (connection->agreed_pro_version < 91) |
4a23f264 | 3171 | return -1091; |
b411b363 | 3172 | |
b30ab791 AG |
3173 | if ((device->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START] & ~((u64)1)) && |
3174 | (device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { | |
d0180171 | 3175 | drbd_info(device, "was SyncSource, missed the resync finished event, corrected myself:\n"); |
b30ab791 AG |
3176 | drbd_uuid_move_history(device); |
3177 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP]; | |
3178 | device->ldev->md.uuid[UI_BITMAP] = 0; | |
b411b363 | 3179 | |
b30ab791 AG |
3180 | drbd_uuid_dump(device, "self", device->ldev->md.uuid, |
3181 | device->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(device) : 0, 0); | |
b411b363 PR |
3182 | *rule_nr = 34; |
3183 | } else { | |
d0180171 | 3184 | drbd_info(device, "was SyncSource (peer failed to write sync_uuid)\n"); |
b411b363 PR |
3185 | *rule_nr = 36; |
3186 | } | |
3187 | ||
3188 | return 1; | |
3189 | } | |
3190 | ||
b30ab791 | 3191 | if (device->ldev->md.uuid[UI_BITMAP] == (u64)0 && device->p_uuid[UI_BITMAP] != (u64)0) { |
b411b363 | 3192 | |
44a4d551 | 3193 | if (connection->agreed_pro_version < 91) |
4a23f264 | 3194 | return -1091; |
b411b363 | 3195 | |
b30ab791 AG |
3196 | if ((device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (device->p_uuid[UI_BITMAP] & ~((u64)1)) && |
3197 | (device->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START] & ~((u64)1))) { | |
d0180171 | 3198 | drbd_info(device, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); |
b411b363 | 3199 | |
b30ab791 AG |
3200 | device->p_uuid[UI_HISTORY_START + 1] = device->p_uuid[UI_HISTORY_START]; |
3201 | device->p_uuid[UI_HISTORY_START] = device->p_uuid[UI_BITMAP]; | |
3202 | device->p_uuid[UI_BITMAP] = 0UL; | |
b411b363 | 3203 | |
b30ab791 | 3204 | drbd_uuid_dump(device, "peer", device->p_uuid, device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); |
b411b363 PR |
3205 | *rule_nr = 35; |
3206 | } else { | |
d0180171 | 3207 | drbd_info(device, "was SyncTarget (failed to write sync_uuid)\n"); |
b411b363 PR |
3208 | *rule_nr = 37; |
3209 | } | |
3210 | ||
3211 | return -1; | |
3212 | } | |
3213 | ||
3214 | /* Common power [off|failure] */ | |
b30ab791 AG |
3215 | rct = (test_bit(CRASHED_PRIMARY, &device->flags) ? 1 : 0) + |
3216 | (device->p_uuid[UI_FLAGS] & 2); | |
b411b363 PR |
3217 | /* lowest bit is set when we were primary, |
3218 | * next bit (weight 2) is set when peer was primary */ | |
3219 | *rule_nr = 40; | |
3220 | ||
f2d3d75b LE |
3221 | /* Neither has the "crashed primary" flag set, |
3222 | * only a replication link hickup. */ | |
3223 | if (rct == 0) | |
3224 | return 0; | |
3225 | ||
3226 | /* Current UUID equal and no bitmap uuid; does not necessarily | |
3227 | * mean this was a "simultaneous hard crash", maybe IO was | |
3228 | * frozen, so no UUID-bump happened. | |
3229 | * This is a protocol change, overload DRBD_FF_WSAME as flag | |
3230 | * for "new-enough" peer DRBD version. */ | |
3231 | if (device->state.role == R_PRIMARY || peer_role == R_PRIMARY) { | |
3232 | *rule_nr = 41; | |
3233 | if (!(connection->agreed_features & DRBD_FF_WSAME)) { | |
3234 | drbd_warn(peer_device, "Equivalent unrotated UUIDs, but current primary present.\n"); | |
3235 | return -(0x10000 | PRO_VERSION_MAX | (DRBD_FF_WSAME << 8)); | |
3236 | } | |
3237 | if (device->state.role == R_PRIMARY && peer_role == R_PRIMARY) { | |
3238 | /* At least one has the "crashed primary" bit set, | |
3239 | * both are primary now, but neither has rotated its UUIDs? | |
3240 | * "Can not happen." */ | |
3241 | drbd_err(peer_device, "Equivalent unrotated UUIDs, but both are primary. Can not resolve this.\n"); | |
3242 | return -100; | |
3243 | } | |
3244 | if (device->state.role == R_PRIMARY) | |
3245 | return 1; | |
3246 | return -1; | |
3247 | } | |
3248 | ||
3249 | /* Both are secondary. | |
3250 | * Really looks like recovery from simultaneous hard crash. | |
3251 | * Check which had been primary before, and arbitrate. */ | |
b411b363 | 3252 | switch (rct) { |
f2d3d75b | 3253 | case 0: /* !self_pri && !peer_pri */ return 0; /* already handled */ |
b411b363 PR |
3254 | case 1: /* self_pri && !peer_pri */ return 1; |
3255 | case 2: /* !self_pri && peer_pri */ return -1; | |
3256 | case 3: /* self_pri && peer_pri */ | |
44a4d551 | 3257 | dc = test_bit(RESOLVE_CONFLICTS, &connection->flags); |
b411b363 PR |
3258 | return dc ? -1 : 1; |
3259 | } | |
3260 | } | |
3261 | ||
3262 | *rule_nr = 50; | |
b30ab791 | 3263 | peer = device->p_uuid[UI_BITMAP] & ~((u64)1); |
b411b363 PR |
3264 | if (self == peer) |
3265 | return -1; | |
3266 | ||
3267 | *rule_nr = 51; | |
b30ab791 | 3268 | peer = device->p_uuid[UI_HISTORY_START] & ~((u64)1); |
b411b363 | 3269 | if (self == peer) { |
44a4d551 | 3270 | if (connection->agreed_pro_version < 96 ? |
b30ab791 AG |
3271 | (device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == |
3272 | (device->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) : | |
3273 | peer + UUID_NEW_BM_OFFSET == (device->p_uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
3274 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
3275 | resync as sync source modifications of the peer's UUIDs. */ | |
3276 | ||
44a4d551 | 3277 | if (connection->agreed_pro_version < 91) |
4a23f264 | 3278 | return -1091; |
b411b363 | 3279 | |
b30ab791 AG |
3280 | device->p_uuid[UI_BITMAP] = device->p_uuid[UI_HISTORY_START]; |
3281 | device->p_uuid[UI_HISTORY_START] = device->p_uuid[UI_HISTORY_START + 1]; | |
4a23f264 | 3282 | |
d0180171 | 3283 | drbd_info(device, "Lost last syncUUID packet, corrected:\n"); |
b30ab791 | 3284 | drbd_uuid_dump(device, "peer", device->p_uuid, device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); |
4a23f264 | 3285 | |
b411b363 PR |
3286 | return -1; |
3287 | } | |
3288 | } | |
3289 | ||
3290 | *rule_nr = 60; | |
b30ab791 | 3291 | self = device->ldev->md.uuid[UI_CURRENT] & ~((u64)1); |
b411b363 | 3292 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
b30ab791 | 3293 | peer = device->p_uuid[i] & ~((u64)1); |
b411b363 PR |
3294 | if (self == peer) |
3295 | return -2; | |
3296 | } | |
3297 | ||
3298 | *rule_nr = 70; | |
b30ab791 AG |
3299 | self = device->ldev->md.uuid[UI_BITMAP] & ~((u64)1); |
3300 | peer = device->p_uuid[UI_CURRENT] & ~((u64)1); | |
b411b363 PR |
3301 | if (self == peer) |
3302 | return 1; | |
3303 | ||
3304 | *rule_nr = 71; | |
b30ab791 | 3305 | self = device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); |
b411b363 | 3306 | if (self == peer) { |
44a4d551 | 3307 | if (connection->agreed_pro_version < 96 ? |
b30ab791 AG |
3308 | (device->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == |
3309 | (device->p_uuid[UI_HISTORY_START] & ~((u64)1)) : | |
3310 | self + UUID_NEW_BM_OFFSET == (device->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
3311 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
3312 | resync as sync source modifications of our UUIDs. */ | |
3313 | ||
44a4d551 | 3314 | if (connection->agreed_pro_version < 91) |
4a23f264 | 3315 | return -1091; |
b411b363 | 3316 | |
b30ab791 AG |
3317 | __drbd_uuid_set(device, UI_BITMAP, device->ldev->md.uuid[UI_HISTORY_START]); |
3318 | __drbd_uuid_set(device, UI_HISTORY_START, device->ldev->md.uuid[UI_HISTORY_START + 1]); | |
b411b363 | 3319 | |
d0180171 | 3320 | drbd_info(device, "Last syncUUID did not get through, corrected:\n"); |
b30ab791 AG |
3321 | drbd_uuid_dump(device, "self", device->ldev->md.uuid, |
3322 | device->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(device) : 0, 0); | |
b411b363 PR |
3323 | |
3324 | return 1; | |
3325 | } | |
3326 | } | |
3327 | ||
3328 | ||
3329 | *rule_nr = 80; | |
b30ab791 | 3330 | peer = device->p_uuid[UI_CURRENT] & ~((u64)1); |
b411b363 | 3331 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
b30ab791 | 3332 | self = device->ldev->md.uuid[i] & ~((u64)1); |
b411b363 PR |
3333 | if (self == peer) |
3334 | return 2; | |
3335 | } | |
3336 | ||
3337 | *rule_nr = 90; | |
b30ab791 AG |
3338 | self = device->ldev->md.uuid[UI_BITMAP] & ~((u64)1); |
3339 | peer = device->p_uuid[UI_BITMAP] & ~((u64)1); | |
b411b363 PR |
3340 | if (self == peer && self != ((u64)0)) |
3341 | return 100; | |
3342 | ||
3343 | *rule_nr = 100; | |
3344 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
b30ab791 | 3345 | self = device->ldev->md.uuid[i] & ~((u64)1); |
b411b363 | 3346 | for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { |
b30ab791 | 3347 | peer = device->p_uuid[j] & ~((u64)1); |
b411b363 PR |
3348 | if (self == peer) |
3349 | return -100; | |
3350 | } | |
3351 | } | |
3352 | ||
3353 | return -1000; | |
3354 | } | |
3355 | ||
3356 | /* drbd_sync_handshake() returns the new conn state on success, or | |
3357 | CONN_MASK (-1) on failure. | |
3358 | */ | |
69a22773 AG |
3359 | static enum drbd_conns drbd_sync_handshake(struct drbd_peer_device *peer_device, |
3360 | enum drbd_role peer_role, | |
b411b363 PR |
3361 | enum drbd_disk_state peer_disk) __must_hold(local) |
3362 | { | |
69a22773 | 3363 | struct drbd_device *device = peer_device->device; |
b411b363 PR |
3364 | enum drbd_conns rv = C_MASK; |
3365 | enum drbd_disk_state mydisk; | |
44ed167d | 3366 | struct net_conf *nc; |
d29e89e3 | 3367 | int hg, rule_nr, rr_conflict, tentative, always_asbp; |
b411b363 | 3368 | |
b30ab791 | 3369 | mydisk = device->state.disk; |
b411b363 | 3370 | if (mydisk == D_NEGOTIATING) |
b30ab791 | 3371 | mydisk = device->new_state_tmp.disk; |
b411b363 | 3372 | |
d0180171 | 3373 | drbd_info(device, "drbd_sync_handshake:\n"); |
9f2247bb | 3374 | |
b30ab791 AG |
3375 | spin_lock_irq(&device->ldev->md.uuid_lock); |
3376 | drbd_uuid_dump(device, "self", device->ldev->md.uuid, device->comm_bm_set, 0); | |
3377 | drbd_uuid_dump(device, "peer", device->p_uuid, | |
3378 | device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); | |
b411b363 | 3379 | |
f2d3d75b | 3380 | hg = drbd_uuid_compare(device, peer_role, &rule_nr); |
b30ab791 | 3381 | spin_unlock_irq(&device->ldev->md.uuid_lock); |
b411b363 | 3382 | |
d0180171 | 3383 | drbd_info(device, "uuid_compare()=%d by rule %d\n", hg, rule_nr); |
b411b363 PR |
3384 | |
3385 | if (hg == -1000) { | |
d0180171 | 3386 | drbd_alert(device, "Unrelated data, aborting!\n"); |
b411b363 PR |
3387 | return C_MASK; |
3388 | } | |
f2d3d75b LE |
3389 | if (hg < -0x10000) { |
3390 | int proto, fflags; | |
3391 | hg = -hg; | |
3392 | proto = hg & 0xff; | |
3393 | fflags = (hg >> 8) & 0xff; | |
3394 | drbd_alert(device, "To resolve this both sides have to support at least protocol %d and feature flags 0x%x\n", | |
3395 | proto, fflags); | |
3396 | return C_MASK; | |
3397 | } | |
4a23f264 | 3398 | if (hg < -1000) { |
d0180171 | 3399 | drbd_alert(device, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000); |
b411b363 PR |
3400 | return C_MASK; |
3401 | } | |
3402 | ||
3403 | if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || | |
3404 | (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { | |
3405 | int f = (hg == -100) || abs(hg) == 2; | |
3406 | hg = mydisk > D_INCONSISTENT ? 1 : -1; | |
3407 | if (f) | |
3408 | hg = hg*2; | |
d0180171 | 3409 | drbd_info(device, "Becoming sync %s due to disk states.\n", |
b411b363 PR |
3410 | hg > 0 ? "source" : "target"); |
3411 | } | |
3412 | ||
3a11a487 | 3413 | if (abs(hg) == 100) |
b30ab791 | 3414 | drbd_khelper(device, "initial-split-brain"); |
3a11a487 | 3415 | |
44ed167d | 3416 | rcu_read_lock(); |
69a22773 | 3417 | nc = rcu_dereference(peer_device->connection->net_conf); |
d29e89e3 RK |
3418 | always_asbp = nc->always_asbp; |
3419 | rr_conflict = nc->rr_conflict; | |
3420 | tentative = nc->tentative; | |
3421 | rcu_read_unlock(); | |
44ed167d | 3422 | |
d29e89e3 | 3423 | if (hg == 100 || (hg == -100 && always_asbp)) { |
b30ab791 | 3424 | int pcount = (device->state.role == R_PRIMARY) |
b411b363 PR |
3425 | + (peer_role == R_PRIMARY); |
3426 | int forced = (hg == -100); | |
3427 | ||
3428 | switch (pcount) { | |
3429 | case 0: | |
69a22773 | 3430 | hg = drbd_asb_recover_0p(peer_device); |
b411b363 PR |
3431 | break; |
3432 | case 1: | |
69a22773 | 3433 | hg = drbd_asb_recover_1p(peer_device); |
b411b363 PR |
3434 | break; |
3435 | case 2: | |
69a22773 | 3436 | hg = drbd_asb_recover_2p(peer_device); |
b411b363 PR |
3437 | break; |
3438 | } | |
3439 | if (abs(hg) < 100) { | |
d0180171 | 3440 | drbd_warn(device, "Split-Brain detected, %d primaries, " |
b411b363 PR |
3441 | "automatically solved. Sync from %s node\n", |
3442 | pcount, (hg < 0) ? "peer" : "this"); | |
3443 | if (forced) { | |
d0180171 | 3444 | drbd_warn(device, "Doing a full sync, since" |
b411b363 PR |
3445 | " UUIDs where ambiguous.\n"); |
3446 | hg = hg*2; | |
3447 | } | |
3448 | } | |
3449 | } | |
3450 | ||
3451 | if (hg == -100) { | |
b30ab791 | 3452 | if (test_bit(DISCARD_MY_DATA, &device->flags) && !(device->p_uuid[UI_FLAGS]&1)) |
b411b363 | 3453 | hg = -1; |
b30ab791 | 3454 | if (!test_bit(DISCARD_MY_DATA, &device->flags) && (device->p_uuid[UI_FLAGS]&1)) |
b411b363 PR |
3455 | hg = 1; |
3456 | ||
3457 | if (abs(hg) < 100) | |
d0180171 | 3458 | drbd_warn(device, "Split-Brain detected, manually solved. " |
b411b363 PR |
3459 | "Sync from %s node\n", |
3460 | (hg < 0) ? "peer" : "this"); | |
3461 | } | |
3462 | ||
3463 | if (hg == -100) { | |
580b9767 LE |
3464 | /* FIXME this log message is not correct if we end up here |
3465 | * after an attempted attach on a diskless node. | |
3466 | * We just refuse to attach -- well, we drop the "connection" | |
3467 | * to that disk, in a way... */ | |
d0180171 | 3468 | drbd_alert(device, "Split-Brain detected but unresolved, dropping connection!\n"); |
b30ab791 | 3469 | drbd_khelper(device, "split-brain"); |
b411b363 PR |
3470 | return C_MASK; |
3471 | } | |
3472 | ||
3473 | if (hg > 0 && mydisk <= D_INCONSISTENT) { | |
d0180171 | 3474 | drbd_err(device, "I shall become SyncSource, but I am inconsistent!\n"); |
b411b363 PR |
3475 | return C_MASK; |
3476 | } | |
3477 | ||
3478 | if (hg < 0 && /* by intention we do not use mydisk here. */ | |
b30ab791 | 3479 | device->state.role == R_PRIMARY && device->state.disk >= D_CONSISTENT) { |
44ed167d | 3480 | switch (rr_conflict) { |
b411b363 | 3481 | case ASB_CALL_HELPER: |
b30ab791 | 3482 | drbd_khelper(device, "pri-lost"); |
b411b363 PR |
3483 | /* fall through */ |
3484 | case ASB_DISCONNECT: | |
d0180171 | 3485 | drbd_err(device, "I shall become SyncTarget, but I am primary!\n"); |
b411b363 PR |
3486 | return C_MASK; |
3487 | case ASB_VIOLENTLY: | |
d0180171 | 3488 | drbd_warn(device, "Becoming SyncTarget, violating the stable-data" |
b411b363 PR |
3489 | "assumption\n"); |
3490 | } | |
3491 | } | |
3492 | ||
69a22773 | 3493 | if (tentative || test_bit(CONN_DRY_RUN, &peer_device->connection->flags)) { |
cf14c2e9 | 3494 | if (hg == 0) |
d0180171 | 3495 | drbd_info(device, "dry-run connect: No resync, would become Connected immediately.\n"); |
cf14c2e9 | 3496 | else |
d0180171 | 3497 | drbd_info(device, "dry-run connect: Would become %s, doing a %s resync.", |
cf14c2e9 PR |
3498 | drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), |
3499 | abs(hg) >= 2 ? "full" : "bit-map based"); | |
3500 | return C_MASK; | |
3501 | } | |
3502 | ||
b411b363 | 3503 | if (abs(hg) >= 2) { |
d0180171 | 3504 | drbd_info(device, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); |
b30ab791 | 3505 | if (drbd_bitmap_io(device, &drbd_bmio_set_n_write, "set_n_write from sync_handshake", |
20ceb2b2 | 3506 | BM_LOCKED_SET_ALLOWED)) |
b411b363 PR |
3507 | return C_MASK; |
3508 | } | |
3509 | ||
3510 | if (hg > 0) { /* become sync source. */ | |
3511 | rv = C_WF_BITMAP_S; | |
3512 | } else if (hg < 0) { /* become sync target */ | |
3513 | rv = C_WF_BITMAP_T; | |
3514 | } else { | |
3515 | rv = C_CONNECTED; | |
b30ab791 | 3516 | if (drbd_bm_total_weight(device)) { |
d0180171 | 3517 | drbd_info(device, "No resync, but %lu bits in bitmap!\n", |
b30ab791 | 3518 | drbd_bm_total_weight(device)); |
b411b363 PR |
3519 | } |
3520 | } | |
3521 | ||
3522 | return rv; | |
3523 | } | |
3524 | ||
f179d76d | 3525 | static enum drbd_after_sb_p convert_after_sb(enum drbd_after_sb_p peer) |
b411b363 PR |
3526 | { |
3527 | /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ | |
f179d76d PR |
3528 | if (peer == ASB_DISCARD_REMOTE) |
3529 | return ASB_DISCARD_LOCAL; | |
b411b363 PR |
3530 | |
3531 | /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ | |
f179d76d PR |
3532 | if (peer == ASB_DISCARD_LOCAL) |
3533 | return ASB_DISCARD_REMOTE; | |
b411b363 PR |
3534 | |
3535 | /* everything else is valid if they are equal on both sides. */ | |
f179d76d | 3536 | return peer; |
b411b363 PR |
3537 | } |
3538 | ||
bde89a9e | 3539 | static int receive_protocol(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3540 | { |
e658983a | 3541 | struct p_protocol *p = pi->data; |
036b17ea PR |
3542 | enum drbd_after_sb_p p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; |
3543 | int p_proto, p_discard_my_data, p_two_primaries, cf; | |
3544 | struct net_conf *nc, *old_net_conf, *new_net_conf = NULL; | |
3545 | char integrity_alg[SHARED_SECRET_MAX] = ""; | |
3d0e6375 | 3546 | struct crypto_shash *peer_integrity_tfm = NULL; |
7aca6c75 | 3547 | void *int_dig_in = NULL, *int_dig_vv = NULL; |
b411b363 | 3548 | |
b411b363 PR |
3549 | p_proto = be32_to_cpu(p->protocol); |
3550 | p_after_sb_0p = be32_to_cpu(p->after_sb_0p); | |
3551 | p_after_sb_1p = be32_to_cpu(p->after_sb_1p); | |
3552 | p_after_sb_2p = be32_to_cpu(p->after_sb_2p); | |
b411b363 | 3553 | p_two_primaries = be32_to_cpu(p->two_primaries); |
cf14c2e9 | 3554 | cf = be32_to_cpu(p->conn_flags); |
6139f60d | 3555 | p_discard_my_data = cf & CF_DISCARD_MY_DATA; |
cf14c2e9 | 3556 | |
bde89a9e | 3557 | if (connection->agreed_pro_version >= 87) { |
86db0618 | 3558 | int err; |
cf14c2e9 | 3559 | |
88104ca4 | 3560 | if (pi->size > sizeof(integrity_alg)) |
86db0618 | 3561 | return -EIO; |
bde89a9e | 3562 | err = drbd_recv_all(connection, integrity_alg, pi->size); |
86db0618 AG |
3563 | if (err) |
3564 | return err; | |
036b17ea | 3565 | integrity_alg[SHARED_SECRET_MAX - 1] = 0; |
b411b363 PR |
3566 | } |
3567 | ||
7d4c782c | 3568 | if (pi->cmd != P_PROTOCOL_UPDATE) { |
bde89a9e | 3569 | clear_bit(CONN_DRY_RUN, &connection->flags); |
b411b363 | 3570 | |
fbc12f45 | 3571 | if (cf & CF_DRY_RUN) |
bde89a9e | 3572 | set_bit(CONN_DRY_RUN, &connection->flags); |
b411b363 | 3573 | |
fbc12f45 | 3574 | rcu_read_lock(); |
bde89a9e | 3575 | nc = rcu_dereference(connection->net_conf); |
b411b363 | 3576 | |
fbc12f45 | 3577 | if (p_proto != nc->wire_protocol) { |
1ec861eb | 3578 | drbd_err(connection, "incompatible %s settings\n", "protocol"); |
fbc12f45 AG |
3579 | goto disconnect_rcu_unlock; |
3580 | } | |
b411b363 | 3581 | |
fbc12f45 | 3582 | if (convert_after_sb(p_after_sb_0p) != nc->after_sb_0p) { |
1ec861eb | 3583 | drbd_err(connection, "incompatible %s settings\n", "after-sb-0pri"); |
fbc12f45 AG |
3584 | goto disconnect_rcu_unlock; |
3585 | } | |
b411b363 | 3586 | |
fbc12f45 | 3587 | if (convert_after_sb(p_after_sb_1p) != nc->after_sb_1p) { |
1ec861eb | 3588 | drbd_err(connection, "incompatible %s settings\n", "after-sb-1pri"); |
fbc12f45 AG |
3589 | goto disconnect_rcu_unlock; |
3590 | } | |
b411b363 | 3591 | |
fbc12f45 | 3592 | if (convert_after_sb(p_after_sb_2p) != nc->after_sb_2p) { |
1ec861eb | 3593 | drbd_err(connection, "incompatible %s settings\n", "after-sb-2pri"); |
fbc12f45 AG |
3594 | goto disconnect_rcu_unlock; |
3595 | } | |
b411b363 | 3596 | |
fbc12f45 | 3597 | if (p_discard_my_data && nc->discard_my_data) { |
1ec861eb | 3598 | drbd_err(connection, "incompatible %s settings\n", "discard-my-data"); |
fbc12f45 AG |
3599 | goto disconnect_rcu_unlock; |
3600 | } | |
b411b363 | 3601 | |
fbc12f45 | 3602 | if (p_two_primaries != nc->two_primaries) { |
1ec861eb | 3603 | drbd_err(connection, "incompatible %s settings\n", "allow-two-primaries"); |
fbc12f45 AG |
3604 | goto disconnect_rcu_unlock; |
3605 | } | |
b411b363 | 3606 | |
fbc12f45 | 3607 | if (strcmp(integrity_alg, nc->integrity_alg)) { |
1ec861eb | 3608 | drbd_err(connection, "incompatible %s settings\n", "data-integrity-alg"); |
fbc12f45 AG |
3609 | goto disconnect_rcu_unlock; |
3610 | } | |
b411b363 | 3611 | |
fbc12f45 | 3612 | rcu_read_unlock(); |
b411b363 PR |
3613 | } |
3614 | ||
7d4c782c AG |
3615 | if (integrity_alg[0]) { |
3616 | int hash_size; | |
3617 | ||
3618 | /* | |
3619 | * We can only change the peer data integrity algorithm | |
3620 | * here. Changing our own data integrity algorithm | |
3621 | * requires that we send a P_PROTOCOL_UPDATE packet at | |
3622 | * the same time; otherwise, the peer has no way to | |
3623 | * tell between which packets the algorithm should | |
3624 | * change. | |
3625 | */ | |
b411b363 | 3626 | |
3d0e6375 | 3627 | peer_integrity_tfm = crypto_alloc_shash(integrity_alg, 0, CRYPTO_ALG_ASYNC); |
1b57e663 LE |
3628 | if (IS_ERR(peer_integrity_tfm)) { |
3629 | peer_integrity_tfm = NULL; | |
1ec861eb | 3630 | drbd_err(connection, "peer data-integrity-alg %s not supported\n", |
7d4c782c AG |
3631 | integrity_alg); |
3632 | goto disconnect; | |
3633 | } | |
b411b363 | 3634 | |
3d0e6375 | 3635 | hash_size = crypto_shash_digestsize(peer_integrity_tfm); |
7d4c782c AG |
3636 | int_dig_in = kmalloc(hash_size, GFP_KERNEL); |
3637 | int_dig_vv = kmalloc(hash_size, GFP_KERNEL); | |
3638 | if (!(int_dig_in && int_dig_vv)) { | |
1ec861eb | 3639 | drbd_err(connection, "Allocation of buffers for data integrity checking failed\n"); |
b411b363 PR |
3640 | goto disconnect; |
3641 | } | |
b411b363 PR |
3642 | } |
3643 | ||
7d4c782c AG |
3644 | new_net_conf = kmalloc(sizeof(struct net_conf), GFP_KERNEL); |
3645 | if (!new_net_conf) { | |
1ec861eb | 3646 | drbd_err(connection, "Allocation of new net_conf failed\n"); |
7d4c782c AG |
3647 | goto disconnect; |
3648 | } | |
3649 | ||
bde89a9e | 3650 | mutex_lock(&connection->data.mutex); |
0500813f | 3651 | mutex_lock(&connection->resource->conf_update); |
bde89a9e | 3652 | old_net_conf = connection->net_conf; |
7d4c782c AG |
3653 | *new_net_conf = *old_net_conf; |
3654 | ||
3655 | new_net_conf->wire_protocol = p_proto; | |
3656 | new_net_conf->after_sb_0p = convert_after_sb(p_after_sb_0p); | |
3657 | new_net_conf->after_sb_1p = convert_after_sb(p_after_sb_1p); | |
3658 | new_net_conf->after_sb_2p = convert_after_sb(p_after_sb_2p); | |
3659 | new_net_conf->two_primaries = p_two_primaries; | |
3660 | ||
bde89a9e | 3661 | rcu_assign_pointer(connection->net_conf, new_net_conf); |
0500813f | 3662 | mutex_unlock(&connection->resource->conf_update); |
bde89a9e | 3663 | mutex_unlock(&connection->data.mutex); |
7d4c782c | 3664 | |
3d0e6375 | 3665 | crypto_free_shash(connection->peer_integrity_tfm); |
bde89a9e AG |
3666 | kfree(connection->int_dig_in); |
3667 | kfree(connection->int_dig_vv); | |
3668 | connection->peer_integrity_tfm = peer_integrity_tfm; | |
3669 | connection->int_dig_in = int_dig_in; | |
3670 | connection->int_dig_vv = int_dig_vv; | |
7d4c782c AG |
3671 | |
3672 | if (strcmp(old_net_conf->integrity_alg, integrity_alg)) | |
1ec861eb | 3673 | drbd_info(connection, "peer data-integrity-alg: %s\n", |
7d4c782c AG |
3674 | integrity_alg[0] ? integrity_alg : "(none)"); |
3675 | ||
3676 | synchronize_rcu(); | |
3677 | kfree(old_net_conf); | |
82bc0194 | 3678 | return 0; |
b411b363 | 3679 | |
44ed167d PR |
3680 | disconnect_rcu_unlock: |
3681 | rcu_read_unlock(); | |
b411b363 | 3682 | disconnect: |
3d0e6375 | 3683 | crypto_free_shash(peer_integrity_tfm); |
036b17ea PR |
3684 | kfree(int_dig_in); |
3685 | kfree(int_dig_vv); | |
bde89a9e | 3686 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3687 | return -EIO; |
b411b363 PR |
3688 | } |
3689 | ||
3690 | /* helper function | |
3691 | * input: alg name, feature name | |
3692 | * return: NULL (alg name was "") | |
3693 | * ERR_PTR(error) if something goes wrong | |
3694 | * or the crypto hash ptr, if it worked out ok. */ | |
3d0e6375 KC |
3695 | static struct crypto_shash *drbd_crypto_alloc_digest_safe( |
3696 | const struct drbd_device *device, | |
b411b363 PR |
3697 | const char *alg, const char *name) |
3698 | { | |
3d0e6375 | 3699 | struct crypto_shash *tfm; |
b411b363 PR |
3700 | |
3701 | if (!alg[0]) | |
3702 | return NULL; | |
3703 | ||
3d0e6375 | 3704 | tfm = crypto_alloc_shash(alg, 0, 0); |
b411b363 | 3705 | if (IS_ERR(tfm)) { |
d0180171 | 3706 | drbd_err(device, "Can not allocate \"%s\" as %s (reason: %ld)\n", |
b411b363 PR |
3707 | alg, name, PTR_ERR(tfm)); |
3708 | return tfm; | |
3709 | } | |
b411b363 PR |
3710 | return tfm; |
3711 | } | |
3712 | ||
bde89a9e | 3713 | static int ignore_remaining_packet(struct drbd_connection *connection, struct packet_info *pi) |
4a76b161 | 3714 | { |
bde89a9e | 3715 | void *buffer = connection->data.rbuf; |
4a76b161 AG |
3716 | int size = pi->size; |
3717 | ||
3718 | while (size) { | |
3719 | int s = min_t(int, size, DRBD_SOCKET_BUFFER_SIZE); | |
bde89a9e | 3720 | s = drbd_recv(connection, buffer, s); |
4a76b161 AG |
3721 | if (s <= 0) { |
3722 | if (s < 0) | |
3723 | return s; | |
3724 | break; | |
3725 | } | |
3726 | size -= s; | |
3727 | } | |
3728 | if (size) | |
3729 | return -EIO; | |
3730 | return 0; | |
3731 | } | |
3732 | ||
3733 | /* | |
3734 | * config_unknown_volume - device configuration command for unknown volume | |
3735 | * | |
3736 | * When a device is added to an existing connection, the node on which the | |
3737 | * device is added first will send configuration commands to its peer but the | |
3738 | * peer will not know about the device yet. It will warn and ignore these | |
3739 | * commands. Once the device is added on the second node, the second node will | |
3740 | * send the same device configuration commands, but in the other direction. | |
3741 | * | |
3742 | * (We can also end up here if drbd is misconfigured.) | |
3743 | */ | |
bde89a9e | 3744 | static int config_unknown_volume(struct drbd_connection *connection, struct packet_info *pi) |
4a76b161 | 3745 | { |
1ec861eb | 3746 | drbd_warn(connection, "%s packet received for volume %u, which is not configured locally\n", |
2fcb8f30 | 3747 | cmdname(pi->cmd), pi->vnr); |
bde89a9e | 3748 | return ignore_remaining_packet(connection, pi); |
4a76b161 AG |
3749 | } |
3750 | ||
bde89a9e | 3751 | static int receive_SyncParam(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3752 | { |
9f4fe9ad | 3753 | struct drbd_peer_device *peer_device; |
b30ab791 | 3754 | struct drbd_device *device; |
e658983a | 3755 | struct p_rs_param_95 *p; |
b411b363 | 3756 | unsigned int header_size, data_size, exp_max_sz; |
3d0e6375 KC |
3757 | struct crypto_shash *verify_tfm = NULL; |
3758 | struct crypto_shash *csums_tfm = NULL; | |
2ec91e0e | 3759 | struct net_conf *old_net_conf, *new_net_conf = NULL; |
813472ce | 3760 | struct disk_conf *old_disk_conf = NULL, *new_disk_conf = NULL; |
bde89a9e | 3761 | const int apv = connection->agreed_pro_version; |
813472ce | 3762 | struct fifo_buffer *old_plan = NULL, *new_plan = NULL; |
778f271d | 3763 | int fifo_size = 0; |
82bc0194 | 3764 | int err; |
b411b363 | 3765 | |
9f4fe9ad AG |
3766 | peer_device = conn_peer_device(connection, pi->vnr); |
3767 | if (!peer_device) | |
bde89a9e | 3768 | return config_unknown_volume(connection, pi); |
9f4fe9ad | 3769 | device = peer_device->device; |
b411b363 PR |
3770 | |
3771 | exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) | |
3772 | : apv == 88 ? sizeof(struct p_rs_param) | |
3773 | + SHARED_SECRET_MAX | |
8e26f9cc PR |
3774 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
3775 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 3776 | |
e2857216 | 3777 | if (pi->size > exp_max_sz) { |
d0180171 | 3778 | drbd_err(device, "SyncParam packet too long: received %u, expected <= %u bytes\n", |
e2857216 | 3779 | pi->size, exp_max_sz); |
82bc0194 | 3780 | return -EIO; |
b411b363 PR |
3781 | } |
3782 | ||
3783 | if (apv <= 88) { | |
e658983a | 3784 | header_size = sizeof(struct p_rs_param); |
e2857216 | 3785 | data_size = pi->size - header_size; |
8e26f9cc | 3786 | } else if (apv <= 94) { |
e658983a | 3787 | header_size = sizeof(struct p_rs_param_89); |
e2857216 | 3788 | data_size = pi->size - header_size; |
0b0ba1ef | 3789 | D_ASSERT(device, data_size == 0); |
8e26f9cc | 3790 | } else { |
e658983a | 3791 | header_size = sizeof(struct p_rs_param_95); |
e2857216 | 3792 | data_size = pi->size - header_size; |
0b0ba1ef | 3793 | D_ASSERT(device, data_size == 0); |
b411b363 PR |
3794 | } |
3795 | ||
3796 | /* initialize verify_alg and csums_alg */ | |
e658983a | 3797 | p = pi->data; |
b411b363 PR |
3798 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); |
3799 | ||
9f4fe9ad | 3800 | err = drbd_recv_all(peer_device->connection, p, header_size); |
82bc0194 AG |
3801 | if (err) |
3802 | return err; | |
b411b363 | 3803 | |
0500813f | 3804 | mutex_lock(&connection->resource->conf_update); |
9f4fe9ad | 3805 | old_net_conf = peer_device->connection->net_conf; |
b30ab791 | 3806 | if (get_ldev(device)) { |
813472ce PR |
3807 | new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); |
3808 | if (!new_disk_conf) { | |
b30ab791 | 3809 | put_ldev(device); |
0500813f | 3810 | mutex_unlock(&connection->resource->conf_update); |
d0180171 | 3811 | drbd_err(device, "Allocation of new disk_conf failed\n"); |
813472ce PR |
3812 | return -ENOMEM; |
3813 | } | |
daeda1cc | 3814 | |
b30ab791 | 3815 | old_disk_conf = device->ldev->disk_conf; |
813472ce | 3816 | *new_disk_conf = *old_disk_conf; |
b411b363 | 3817 | |
6394b935 | 3818 | new_disk_conf->resync_rate = be32_to_cpu(p->resync_rate); |
813472ce | 3819 | } |
b411b363 PR |
3820 | |
3821 | if (apv >= 88) { | |
3822 | if (apv == 88) { | |
5de73827 | 3823 | if (data_size > SHARED_SECRET_MAX || data_size == 0) { |
d0180171 | 3824 | drbd_err(device, "verify-alg of wrong size, " |
5de73827 PR |
3825 | "peer wants %u, accepting only up to %u byte\n", |
3826 | data_size, SHARED_SECRET_MAX); | |
813472ce PR |
3827 | err = -EIO; |
3828 | goto reconnect; | |
b411b363 PR |
3829 | } |
3830 | ||
9f4fe9ad | 3831 | err = drbd_recv_all(peer_device->connection, p->verify_alg, data_size); |
813472ce PR |
3832 | if (err) |
3833 | goto reconnect; | |
b411b363 PR |
3834 | /* we expect NUL terminated string */ |
3835 | /* but just in case someone tries to be evil */ | |
0b0ba1ef | 3836 | D_ASSERT(device, p->verify_alg[data_size-1] == 0); |
b411b363 PR |
3837 | p->verify_alg[data_size-1] = 0; |
3838 | ||
3839 | } else /* apv >= 89 */ { | |
3840 | /* we still expect NUL terminated strings */ | |
3841 | /* but just in case someone tries to be evil */ | |
0b0ba1ef AG |
3842 | D_ASSERT(device, p->verify_alg[SHARED_SECRET_MAX-1] == 0); |
3843 | D_ASSERT(device, p->csums_alg[SHARED_SECRET_MAX-1] == 0); | |
b411b363 PR |
3844 | p->verify_alg[SHARED_SECRET_MAX-1] = 0; |
3845 | p->csums_alg[SHARED_SECRET_MAX-1] = 0; | |
3846 | } | |
3847 | ||
2ec91e0e | 3848 | if (strcmp(old_net_conf->verify_alg, p->verify_alg)) { |
b30ab791 | 3849 | if (device->state.conn == C_WF_REPORT_PARAMS) { |
d0180171 | 3850 | drbd_err(device, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", |
2ec91e0e | 3851 | old_net_conf->verify_alg, p->verify_alg); |
b411b363 PR |
3852 | goto disconnect; |
3853 | } | |
b30ab791 | 3854 | verify_tfm = drbd_crypto_alloc_digest_safe(device, |
b411b363 PR |
3855 | p->verify_alg, "verify-alg"); |
3856 | if (IS_ERR(verify_tfm)) { | |
3857 | verify_tfm = NULL; | |
3858 | goto disconnect; | |
3859 | } | |
3860 | } | |
3861 | ||
2ec91e0e | 3862 | if (apv >= 89 && strcmp(old_net_conf->csums_alg, p->csums_alg)) { |
b30ab791 | 3863 | if (device->state.conn == C_WF_REPORT_PARAMS) { |
d0180171 | 3864 | drbd_err(device, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", |
2ec91e0e | 3865 | old_net_conf->csums_alg, p->csums_alg); |
b411b363 PR |
3866 | goto disconnect; |
3867 | } | |
b30ab791 | 3868 | csums_tfm = drbd_crypto_alloc_digest_safe(device, |
b411b363 PR |
3869 | p->csums_alg, "csums-alg"); |
3870 | if (IS_ERR(csums_tfm)) { | |
3871 | csums_tfm = NULL; | |
3872 | goto disconnect; | |
3873 | } | |
3874 | } | |
3875 | ||
813472ce | 3876 | if (apv > 94 && new_disk_conf) { |
daeda1cc PR |
3877 | new_disk_conf->c_plan_ahead = be32_to_cpu(p->c_plan_ahead); |
3878 | new_disk_conf->c_delay_target = be32_to_cpu(p->c_delay_target); | |
3879 | new_disk_conf->c_fill_target = be32_to_cpu(p->c_fill_target); | |
3880 | new_disk_conf->c_max_rate = be32_to_cpu(p->c_max_rate); | |
778f271d | 3881 | |
daeda1cc | 3882 | fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ; |
b30ab791 | 3883 | if (fifo_size != device->rs_plan_s->size) { |
813472ce PR |
3884 | new_plan = fifo_alloc(fifo_size); |
3885 | if (!new_plan) { | |
d0180171 | 3886 | drbd_err(device, "kmalloc of fifo_buffer failed"); |
b30ab791 | 3887 | put_ldev(device); |
778f271d PR |
3888 | goto disconnect; |
3889 | } | |
3890 | } | |
8e26f9cc | 3891 | } |
b411b363 | 3892 | |
91fd4dad | 3893 | if (verify_tfm || csums_tfm) { |
2ec91e0e PR |
3894 | new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL); |
3895 | if (!new_net_conf) { | |
d0180171 | 3896 | drbd_err(device, "Allocation of new net_conf failed\n"); |
91fd4dad PR |
3897 | goto disconnect; |
3898 | } | |
3899 | ||
2ec91e0e | 3900 | *new_net_conf = *old_net_conf; |
91fd4dad PR |
3901 | |
3902 | if (verify_tfm) { | |
2ec91e0e PR |
3903 | strcpy(new_net_conf->verify_alg, p->verify_alg); |
3904 | new_net_conf->verify_alg_len = strlen(p->verify_alg) + 1; | |
3d0e6375 | 3905 | crypto_free_shash(peer_device->connection->verify_tfm); |
9f4fe9ad | 3906 | peer_device->connection->verify_tfm = verify_tfm; |
d0180171 | 3907 | drbd_info(device, "using verify-alg: \"%s\"\n", p->verify_alg); |
91fd4dad PR |
3908 | } |
3909 | if (csums_tfm) { | |
2ec91e0e PR |
3910 | strcpy(new_net_conf->csums_alg, p->csums_alg); |
3911 | new_net_conf->csums_alg_len = strlen(p->csums_alg) + 1; | |
3d0e6375 | 3912 | crypto_free_shash(peer_device->connection->csums_tfm); |
9f4fe9ad | 3913 | peer_device->connection->csums_tfm = csums_tfm; |
d0180171 | 3914 | drbd_info(device, "using csums-alg: \"%s\"\n", p->csums_alg); |
91fd4dad | 3915 | } |
bde89a9e | 3916 | rcu_assign_pointer(connection->net_conf, new_net_conf); |
778f271d | 3917 | } |
b411b363 PR |
3918 | } |
3919 | ||
813472ce | 3920 | if (new_disk_conf) { |
b30ab791 AG |
3921 | rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf); |
3922 | put_ldev(device); | |
813472ce PR |
3923 | } |
3924 | ||
3925 | if (new_plan) { | |
b30ab791 AG |
3926 | old_plan = device->rs_plan_s; |
3927 | rcu_assign_pointer(device->rs_plan_s, new_plan); | |
b411b363 | 3928 | } |
daeda1cc | 3929 | |
0500813f | 3930 | mutex_unlock(&connection->resource->conf_update); |
daeda1cc PR |
3931 | synchronize_rcu(); |
3932 | if (new_net_conf) | |
3933 | kfree(old_net_conf); | |
3934 | kfree(old_disk_conf); | |
813472ce | 3935 | kfree(old_plan); |
daeda1cc | 3936 | |
82bc0194 | 3937 | return 0; |
b411b363 | 3938 | |
813472ce PR |
3939 | reconnect: |
3940 | if (new_disk_conf) { | |
b30ab791 | 3941 | put_ldev(device); |
813472ce PR |
3942 | kfree(new_disk_conf); |
3943 | } | |
0500813f | 3944 | mutex_unlock(&connection->resource->conf_update); |
813472ce PR |
3945 | return -EIO; |
3946 | ||
b411b363 | 3947 | disconnect: |
813472ce PR |
3948 | kfree(new_plan); |
3949 | if (new_disk_conf) { | |
b30ab791 | 3950 | put_ldev(device); |
813472ce PR |
3951 | kfree(new_disk_conf); |
3952 | } | |
0500813f | 3953 | mutex_unlock(&connection->resource->conf_update); |
b411b363 PR |
3954 | /* just for completeness: actually not needed, |
3955 | * as this is not reached if csums_tfm was ok. */ | |
3d0e6375 | 3956 | crypto_free_shash(csums_tfm); |
b411b363 | 3957 | /* but free the verify_tfm again, if csums_tfm did not work out */ |
3d0e6375 | 3958 | crypto_free_shash(verify_tfm); |
9f4fe9ad | 3959 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3960 | return -EIO; |
b411b363 PR |
3961 | } |
3962 | ||
b411b363 | 3963 | /* warn if the arguments differ by more than 12.5% */ |
b30ab791 | 3964 | static void warn_if_differ_considerably(struct drbd_device *device, |
b411b363 PR |
3965 | const char *s, sector_t a, sector_t b) |
3966 | { | |
3967 | sector_t d; | |
3968 | if (a == 0 || b == 0) | |
3969 | return; | |
3970 | d = (a > b) ? (a - b) : (b - a); | |
3971 | if (d > (a>>3) || d > (b>>3)) | |
d0180171 | 3972 | drbd_warn(device, "Considerable difference in %s: %llus vs. %llus\n", s, |
b411b363 PR |
3973 | (unsigned long long)a, (unsigned long long)b); |
3974 | } | |
3975 | ||
bde89a9e | 3976 | static int receive_sizes(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3977 | { |
9f4fe9ad | 3978 | struct drbd_peer_device *peer_device; |
b30ab791 | 3979 | struct drbd_device *device; |
e658983a | 3980 | struct p_sizes *p = pi->data; |
9104d31a | 3981 | struct o_qlim *o = (connection->agreed_features & DRBD_FF_WSAME) ? p->qlim : NULL; |
e96c9633 | 3982 | enum determine_dev_size dd = DS_UNCHANGED; |
6a8d68b1 | 3983 | sector_t p_size, p_usize, p_csize, my_usize; |
94c43a13 | 3984 | sector_t new_size, cur_size; |
b411b363 | 3985 | int ldsc = 0; /* local disk size changed */ |
e89b591c | 3986 | enum dds_flags ddsf; |
b411b363 | 3987 | |
9f4fe9ad AG |
3988 | peer_device = conn_peer_device(connection, pi->vnr); |
3989 | if (!peer_device) | |
bde89a9e | 3990 | return config_unknown_volume(connection, pi); |
9f4fe9ad | 3991 | device = peer_device->device; |
94c43a13 | 3992 | cur_size = drbd_get_capacity(device->this_bdev); |
4a76b161 | 3993 | |
b411b363 PR |
3994 | p_size = be64_to_cpu(p->d_size); |
3995 | p_usize = be64_to_cpu(p->u_size); | |
6a8d68b1 | 3996 | p_csize = be64_to_cpu(p->c_size); |
b411b363 | 3997 | |
b411b363 PR |
3998 | /* just store the peer's disk size for now. |
3999 | * we still need to figure out whether we accept that. */ | |
b30ab791 | 4000 | device->p_size = p_size; |
b411b363 | 4001 | |
b30ab791 | 4002 | if (get_ldev(device)) { |
daeda1cc | 4003 | rcu_read_lock(); |
b30ab791 | 4004 | my_usize = rcu_dereference(device->ldev->disk_conf)->disk_size; |
daeda1cc PR |
4005 | rcu_read_unlock(); |
4006 | ||
b30ab791 AG |
4007 | warn_if_differ_considerably(device, "lower level device sizes", |
4008 | p_size, drbd_get_max_capacity(device->ldev)); | |
4009 | warn_if_differ_considerably(device, "user requested size", | |
daeda1cc | 4010 | p_usize, my_usize); |
b411b363 PR |
4011 | |
4012 | /* if this is the first connect, or an otherwise expected | |
4013 | * param exchange, choose the minimum */ | |
b30ab791 | 4014 | if (device->state.conn == C_WF_REPORT_PARAMS) |
daeda1cc | 4015 | p_usize = min_not_zero(my_usize, p_usize); |
b411b363 | 4016 | |
ad6e8979 LE |
4017 | /* Never shrink a device with usable data during connect, |
4018 | * or "attach" on the peer. | |
4019 | * But allow online shrinking if we are connected. */ | |
60bac040 | 4020 | new_size = drbd_new_dev_size(device, device->ldev, p_usize, 0); |
60bac040 | 4021 | if (new_size < cur_size && |
b30ab791 | 4022 | device->state.disk >= D_OUTDATED && |
ad6e8979 | 4023 | (device->state.conn < C_CONNECTED || device->state.pdsk == D_DISKLESS)) { |
60bac040 LE |
4024 | drbd_err(device, "The peer's disk size is too small! (%llu < %llu sectors)\n", |
4025 | (unsigned long long)new_size, (unsigned long long)cur_size); | |
9f4fe9ad | 4026 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b30ab791 | 4027 | put_ldev(device); |
82bc0194 | 4028 | return -EIO; |
b411b363 | 4029 | } |
daeda1cc PR |
4030 | |
4031 | if (my_usize != p_usize) { | |
4032 | struct disk_conf *old_disk_conf, *new_disk_conf = NULL; | |
4033 | ||
4034 | new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); | |
4035 | if (!new_disk_conf) { | |
d0180171 | 4036 | drbd_err(device, "Allocation of new disk_conf failed\n"); |
b30ab791 | 4037 | put_ldev(device); |
daeda1cc PR |
4038 | return -ENOMEM; |
4039 | } | |
4040 | ||
0500813f | 4041 | mutex_lock(&connection->resource->conf_update); |
b30ab791 | 4042 | old_disk_conf = device->ldev->disk_conf; |
daeda1cc PR |
4043 | *new_disk_conf = *old_disk_conf; |
4044 | new_disk_conf->disk_size = p_usize; | |
4045 | ||
b30ab791 | 4046 | rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf); |
0500813f | 4047 | mutex_unlock(&connection->resource->conf_update); |
daeda1cc PR |
4048 | synchronize_rcu(); |
4049 | kfree(old_disk_conf); | |
4050 | ||
ad6e8979 LE |
4051 | drbd_info(device, "Peer sets u_size to %lu sectors (old: %lu)\n", |
4052 | (unsigned long)p_usize, (unsigned long)my_usize); | |
b411b363 | 4053 | } |
daeda1cc | 4054 | |
b30ab791 | 4055 | put_ldev(device); |
b411b363 | 4056 | } |
b411b363 | 4057 | |
20c68fde | 4058 | device->peer_max_bio_size = be32_to_cpu(p->max_bio_size); |
dd4f699d | 4059 | /* Leave drbd_reconsider_queue_parameters() before drbd_determine_dev_size(). |
20c68fde | 4060 | In case we cleared the QUEUE_FLAG_DISCARD from our queue in |
dd4f699d | 4061 | drbd_reconsider_queue_parameters(), we can be sure that after |
20c68fde LE |
4062 | drbd_determine_dev_size() no REQ_DISCARDs are in the queue. */ |
4063 | ||
e89b591c | 4064 | ddsf = be16_to_cpu(p->dds_flags); |
b30ab791 | 4065 | if (get_ldev(device)) { |
9104d31a | 4066 | drbd_reconsider_queue_parameters(device, device->ldev, o); |
b30ab791 AG |
4067 | dd = drbd_determine_dev_size(device, ddsf, NULL); |
4068 | put_ldev(device); | |
e96c9633 | 4069 | if (dd == DS_ERROR) |
82bc0194 | 4070 | return -EIO; |
b30ab791 | 4071 | drbd_md_sync(device); |
b411b363 | 4072 | } else { |
6a8d68b1 LE |
4073 | /* |
4074 | * I am diskless, need to accept the peer's *current* size. | |
4075 | * I must NOT accept the peers backing disk size, | |
4076 | * it may have been larger than mine all along... | |
4077 | * | |
4078 | * At this point, the peer knows more about my disk, or at | |
4079 | * least about what we last agreed upon, than myself. | |
4080 | * So if his c_size is less than his d_size, the most likely | |
4081 | * reason is that *my* d_size was smaller last time we checked. | |
4082 | * | |
4083 | * However, if he sends a zero current size, | |
4084 | * take his (user-capped or) backing disk size anyways. | |
94c43a13 LE |
4085 | * |
4086 | * Unless of course he does not have a disk himself. | |
4087 | * In which case we ignore this completely. | |
6a8d68b1 | 4088 | */ |
94c43a13 | 4089 | sector_t new_size = p_csize ?: p_usize ?: p_size; |
9104d31a | 4090 | drbd_reconsider_queue_parameters(device, NULL, o); |
94c43a13 LE |
4091 | if (new_size == 0) { |
4092 | /* Ignore, peer does not know nothing. */ | |
4093 | } else if (new_size == cur_size) { | |
4094 | /* nothing to do */ | |
4095 | } else if (cur_size != 0 && p_size == 0) { | |
4096 | drbd_warn(device, "Ignored diskless peer device size (peer:%llu != me:%llu sectors)!\n", | |
4097 | (unsigned long long)new_size, (unsigned long long)cur_size); | |
4098 | } else if (new_size < cur_size && device->state.role == R_PRIMARY) { | |
4099 | drbd_err(device, "The peer's device size is too small! (%llu < %llu sectors); demote me first!\n", | |
4100 | (unsigned long long)new_size, (unsigned long long)cur_size); | |
4101 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); | |
4102 | return -EIO; | |
4103 | } else { | |
4104 | /* I believe the peer, if | |
4105 | * - I don't have a current size myself | |
4106 | * - we agree on the size anyways | |
4107 | * - I do have a current size, am Secondary, | |
4108 | * and he has the only disk | |
4109 | * - I do have a current size, am Primary, | |
4110 | * and he has the only disk, | |
4111 | * which is larger than my current size | |
4112 | */ | |
4113 | drbd_set_my_capacity(device, new_size); | |
4114 | } | |
b411b363 PR |
4115 | } |
4116 | ||
b30ab791 AG |
4117 | if (get_ldev(device)) { |
4118 | if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev)) { | |
4119 | device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev); | |
b411b363 PR |
4120 | ldsc = 1; |
4121 | } | |
4122 | ||
b30ab791 | 4123 | put_ldev(device); |
b411b363 PR |
4124 | } |
4125 | ||
b30ab791 | 4126 | if (device->state.conn > C_WF_REPORT_PARAMS) { |
b411b363 | 4127 | if (be64_to_cpu(p->c_size) != |
b30ab791 | 4128 | drbd_get_capacity(device->this_bdev) || ldsc) { |
b411b363 PR |
4129 | /* we have different sizes, probably peer |
4130 | * needs to know my new size... */ | |
69a22773 | 4131 | drbd_send_sizes(peer_device, 0, ddsf); |
b411b363 | 4132 | } |
b30ab791 AG |
4133 | if (test_and_clear_bit(RESIZE_PENDING, &device->flags) || |
4134 | (dd == DS_GREW && device->state.conn == C_CONNECTED)) { | |
4135 | if (device->state.pdsk >= D_INCONSISTENT && | |
4136 | device->state.disk >= D_INCONSISTENT) { | |
e89b591c | 4137 | if (ddsf & DDSF_NO_RESYNC) |
d0180171 | 4138 | drbd_info(device, "Resync of new storage suppressed with --assume-clean\n"); |
e89b591c | 4139 | else |
b30ab791 | 4140 | resync_after_online_grow(device); |
e89b591c | 4141 | } else |
b30ab791 | 4142 | set_bit(RESYNC_AFTER_NEG, &device->flags); |
b411b363 PR |
4143 | } |
4144 | } | |
4145 | ||
82bc0194 | 4146 | return 0; |
b411b363 PR |
4147 | } |
4148 | ||
bde89a9e | 4149 | static int receive_uuids(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4150 | { |
9f4fe9ad | 4151 | struct drbd_peer_device *peer_device; |
b30ab791 | 4152 | struct drbd_device *device; |
e658983a | 4153 | struct p_uuids *p = pi->data; |
b411b363 | 4154 | u64 *p_uuid; |
62b0da3a | 4155 | int i, updated_uuids = 0; |
b411b363 | 4156 | |
9f4fe9ad AG |
4157 | peer_device = conn_peer_device(connection, pi->vnr); |
4158 | if (!peer_device) | |
bde89a9e | 4159 | return config_unknown_volume(connection, pi); |
9f4fe9ad | 4160 | device = peer_device->device; |
4a76b161 | 4161 | |
365cf663 | 4162 | p_uuid = kmalloc_array(UI_EXTENDED_SIZE, sizeof(*p_uuid), GFP_NOIO); |
063eacf8 | 4163 | if (!p_uuid) { |
d0180171 | 4164 | drbd_err(device, "kmalloc of p_uuid failed\n"); |
063eacf8 JW |
4165 | return false; |
4166 | } | |
b411b363 PR |
4167 | |
4168 | for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) | |
4169 | p_uuid[i] = be64_to_cpu(p->uuid[i]); | |
4170 | ||
b30ab791 AG |
4171 | kfree(device->p_uuid); |
4172 | device->p_uuid = p_uuid; | |
b411b363 | 4173 | |
b17b5960 | 4174 | if ((device->state.conn < C_CONNECTED || device->state.pdsk == D_DISKLESS) && |
b30ab791 AG |
4175 | device->state.disk < D_INCONSISTENT && |
4176 | device->state.role == R_PRIMARY && | |
4177 | (device->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { | |
d0180171 | 4178 | drbd_err(device, "Can only connect to data with current UUID=%016llX\n", |
b30ab791 | 4179 | (unsigned long long)device->ed_uuid); |
9f4fe9ad | 4180 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 4181 | return -EIO; |
b411b363 PR |
4182 | } |
4183 | ||
b30ab791 | 4184 | if (get_ldev(device)) { |
b411b363 | 4185 | int skip_initial_sync = |
b30ab791 | 4186 | device->state.conn == C_CONNECTED && |
9f4fe9ad | 4187 | peer_device->connection->agreed_pro_version >= 90 && |
b30ab791 | 4188 | device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && |
b411b363 PR |
4189 | (p_uuid[UI_FLAGS] & 8); |
4190 | if (skip_initial_sync) { | |
d0180171 | 4191 | drbd_info(device, "Accepted new current UUID, preparing to skip initial sync\n"); |
b30ab791 | 4192 | drbd_bitmap_io(device, &drbd_bmio_clear_n_write, |
20ceb2b2 LE |
4193 | "clear_n_write from receive_uuids", |
4194 | BM_LOCKED_TEST_ALLOWED); | |
b30ab791 AG |
4195 | _drbd_uuid_set(device, UI_CURRENT, p_uuid[UI_CURRENT]); |
4196 | _drbd_uuid_set(device, UI_BITMAP, 0); | |
4197 | _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), | |
b411b363 | 4198 | CS_VERBOSE, NULL); |
b30ab791 | 4199 | drbd_md_sync(device); |
62b0da3a | 4200 | updated_uuids = 1; |
b411b363 | 4201 | } |
b30ab791 AG |
4202 | put_ldev(device); |
4203 | } else if (device->state.disk < D_INCONSISTENT && | |
4204 | device->state.role == R_PRIMARY) { | |
18a50fa2 PR |
4205 | /* I am a diskless primary, the peer just created a new current UUID |
4206 | for me. */ | |
b30ab791 | 4207 | updated_uuids = drbd_set_ed_uuid(device, p_uuid[UI_CURRENT]); |
b411b363 PR |
4208 | } |
4209 | ||
4210 | /* Before we test for the disk state, we should wait until an eventually | |
4211 | ongoing cluster wide state change is finished. That is important if | |
4212 | we are primary and are detaching from our disk. We need to see the | |
4213 | new disk state... */ | |
b30ab791 AG |
4214 | mutex_lock(device->state_mutex); |
4215 | mutex_unlock(device->state_mutex); | |
4216 | if (device->state.conn >= C_CONNECTED && device->state.disk < D_INCONSISTENT) | |
4217 | updated_uuids |= drbd_set_ed_uuid(device, p_uuid[UI_CURRENT]); | |
62b0da3a LE |
4218 | |
4219 | if (updated_uuids) | |
b30ab791 | 4220 | drbd_print_uuids(device, "receiver updated UUIDs to"); |
b411b363 | 4221 | |
82bc0194 | 4222 | return 0; |
b411b363 PR |
4223 | } |
4224 | ||
4225 | /** | |
4226 | * convert_state() - Converts the peer's view of the cluster state to our point of view | |
4227 | * @ps: The state as seen by the peer. | |
4228 | */ | |
4229 | static union drbd_state convert_state(union drbd_state ps) | |
4230 | { | |
4231 | union drbd_state ms; | |
4232 | ||
4233 | static enum drbd_conns c_tab[] = { | |
369bea63 | 4234 | [C_WF_REPORT_PARAMS] = C_WF_REPORT_PARAMS, |
b411b363 PR |
4235 | [C_CONNECTED] = C_CONNECTED, |
4236 | ||
4237 | [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, | |
4238 | [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, | |
4239 | [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ | |
4240 | [C_VERIFY_S] = C_VERIFY_T, | |
4241 | [C_MASK] = C_MASK, | |
4242 | }; | |
4243 | ||
4244 | ms.i = ps.i; | |
4245 | ||
4246 | ms.conn = c_tab[ps.conn]; | |
4247 | ms.peer = ps.role; | |
4248 | ms.role = ps.peer; | |
4249 | ms.pdsk = ps.disk; | |
4250 | ms.disk = ps.pdsk; | |
4251 | ms.peer_isp = (ps.aftr_isp | ps.user_isp); | |
4252 | ||
4253 | return ms; | |
4254 | } | |
4255 | ||
bde89a9e | 4256 | static int receive_req_state(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4257 | { |
9f4fe9ad | 4258 | struct drbd_peer_device *peer_device; |
b30ab791 | 4259 | struct drbd_device *device; |
e658983a | 4260 | struct p_req_state *p = pi->data; |
b411b363 | 4261 | union drbd_state mask, val; |
bf885f8a | 4262 | enum drbd_state_rv rv; |
b411b363 | 4263 | |
9f4fe9ad AG |
4264 | peer_device = conn_peer_device(connection, pi->vnr); |
4265 | if (!peer_device) | |
4a76b161 | 4266 | return -EIO; |
9f4fe9ad | 4267 | device = peer_device->device; |
4a76b161 | 4268 | |
b411b363 PR |
4269 | mask.i = be32_to_cpu(p->mask); |
4270 | val.i = be32_to_cpu(p->val); | |
4271 | ||
9f4fe9ad | 4272 | if (test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags) && |
b30ab791 | 4273 | mutex_is_locked(device->state_mutex)) { |
69a22773 | 4274 | drbd_send_sr_reply(peer_device, SS_CONCURRENT_ST_CHG); |
82bc0194 | 4275 | return 0; |
b411b363 PR |
4276 | } |
4277 | ||
4278 | mask = convert_state(mask); | |
4279 | val = convert_state(val); | |
4280 | ||
b30ab791 | 4281 | rv = drbd_change_state(device, CS_VERBOSE, mask, val); |
69a22773 | 4282 | drbd_send_sr_reply(peer_device, rv); |
b411b363 | 4283 | |
b30ab791 | 4284 | drbd_md_sync(device); |
b411b363 | 4285 | |
82bc0194 | 4286 | return 0; |
b411b363 PR |
4287 | } |
4288 | ||
bde89a9e | 4289 | static int receive_req_conn_state(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4290 | { |
e658983a | 4291 | struct p_req_state *p = pi->data; |
b411b363 | 4292 | union drbd_state mask, val; |
bf885f8a | 4293 | enum drbd_state_rv rv; |
b411b363 | 4294 | |
b411b363 PR |
4295 | mask.i = be32_to_cpu(p->mask); |
4296 | val.i = be32_to_cpu(p->val); | |
4297 | ||
bde89a9e AG |
4298 | if (test_bit(RESOLVE_CONFLICTS, &connection->flags) && |
4299 | mutex_is_locked(&connection->cstate_mutex)) { | |
4300 | conn_send_sr_reply(connection, SS_CONCURRENT_ST_CHG); | |
82bc0194 | 4301 | return 0; |
b411b363 PR |
4302 | } |
4303 | ||
4304 | mask = convert_state(mask); | |
4305 | val = convert_state(val); | |
4306 | ||
bde89a9e AG |
4307 | rv = conn_request_state(connection, mask, val, CS_VERBOSE | CS_LOCAL_ONLY | CS_IGN_OUTD_FAIL); |
4308 | conn_send_sr_reply(connection, rv); | |
b411b363 | 4309 | |
82bc0194 | 4310 | return 0; |
b411b363 PR |
4311 | } |
4312 | ||
bde89a9e | 4313 | static int receive_state(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4314 | { |
9f4fe9ad | 4315 | struct drbd_peer_device *peer_device; |
b30ab791 | 4316 | struct drbd_device *device; |
e658983a | 4317 | struct p_state *p = pi->data; |
4ac4aada | 4318 | union drbd_state os, ns, peer_state; |
b411b363 | 4319 | enum drbd_disk_state real_peer_disk; |
65d922c3 | 4320 | enum chg_state_flags cs_flags; |
b411b363 PR |
4321 | int rv; |
4322 | ||
9f4fe9ad AG |
4323 | peer_device = conn_peer_device(connection, pi->vnr); |
4324 | if (!peer_device) | |
bde89a9e | 4325 | return config_unknown_volume(connection, pi); |
9f4fe9ad | 4326 | device = peer_device->device; |
4a76b161 | 4327 | |
b411b363 PR |
4328 | peer_state.i = be32_to_cpu(p->state); |
4329 | ||
4330 | real_peer_disk = peer_state.disk; | |
4331 | if (peer_state.disk == D_NEGOTIATING) { | |
b30ab791 | 4332 | real_peer_disk = device->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; |
d0180171 | 4333 | drbd_info(device, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); |
b411b363 PR |
4334 | } |
4335 | ||
0500813f | 4336 | spin_lock_irq(&device->resource->req_lock); |
b411b363 | 4337 | retry: |
b30ab791 | 4338 | os = ns = drbd_read_state(device); |
0500813f | 4339 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 4340 | |
668700b4 | 4341 | /* If some other part of the code (ack_receiver thread, timeout) |
545752d5 LE |
4342 | * already decided to close the connection again, |
4343 | * we must not "re-establish" it here. */ | |
4344 | if (os.conn <= C_TEAR_DOWN) | |
58ffa580 | 4345 | return -ECONNRESET; |
545752d5 | 4346 | |
40424e4a LE |
4347 | /* If this is the "end of sync" confirmation, usually the peer disk |
4348 | * transitions from D_INCONSISTENT to D_UP_TO_DATE. For empty (0 bits | |
4349 | * set) resync started in PausedSyncT, or if the timing of pause-/ | |
4350 | * unpause-sync events has been "just right", the peer disk may | |
4351 | * transition from D_CONSISTENT to D_UP_TO_DATE as well. | |
4352 | */ | |
4353 | if ((os.pdsk == D_INCONSISTENT || os.pdsk == D_CONSISTENT) && | |
4354 | real_peer_disk == D_UP_TO_DATE && | |
e9ef7bb6 LE |
4355 | os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { |
4356 | /* If we are (becoming) SyncSource, but peer is still in sync | |
4357 | * preparation, ignore its uptodate-ness to avoid flapping, it | |
4358 | * will change to inconsistent once the peer reaches active | |
4359 | * syncing states. | |
4360 | * It may have changed syncer-paused flags, however, so we | |
4361 | * cannot ignore this completely. */ | |
4362 | if (peer_state.conn > C_CONNECTED && | |
4363 | peer_state.conn < C_SYNC_SOURCE) | |
4364 | real_peer_disk = D_INCONSISTENT; | |
4365 | ||
4366 | /* if peer_state changes to connected at the same time, | |
4367 | * it explicitly notifies us that it finished resync. | |
4368 | * Maybe we should finish it up, too? */ | |
4369 | else if (os.conn >= C_SYNC_SOURCE && | |
4370 | peer_state.conn == C_CONNECTED) { | |
b30ab791 AG |
4371 | if (drbd_bm_total_weight(device) <= device->rs_failed) |
4372 | drbd_resync_finished(device); | |
82bc0194 | 4373 | return 0; |
e9ef7bb6 LE |
4374 | } |
4375 | } | |
4376 | ||
02b91b55 LE |
4377 | /* explicit verify finished notification, stop sector reached. */ |
4378 | if (os.conn == C_VERIFY_T && os.disk == D_UP_TO_DATE && | |
4379 | peer_state.conn == C_CONNECTED && real_peer_disk == D_UP_TO_DATE) { | |
b30ab791 AG |
4380 | ov_out_of_sync_print(device); |
4381 | drbd_resync_finished(device); | |
58ffa580 | 4382 | return 0; |
02b91b55 LE |
4383 | } |
4384 | ||
e9ef7bb6 LE |
4385 | /* peer says his disk is inconsistent, while we think it is uptodate, |
4386 | * and this happens while the peer still thinks we have a sync going on, | |
4387 | * but we think we are already done with the sync. | |
4388 | * We ignore this to avoid flapping pdsk. | |
4389 | * This should not happen, if the peer is a recent version of drbd. */ | |
4390 | if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && | |
4391 | os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) | |
4392 | real_peer_disk = D_UP_TO_DATE; | |
4393 | ||
4ac4aada LE |
4394 | if (ns.conn == C_WF_REPORT_PARAMS) |
4395 | ns.conn = C_CONNECTED; | |
b411b363 | 4396 | |
67531718 PR |
4397 | if (peer_state.conn == C_AHEAD) |
4398 | ns.conn = C_BEHIND; | |
4399 | ||
fe43ed97 LE |
4400 | /* TODO: |
4401 | * if (primary and diskless and peer uuid != effective uuid) | |
4402 | * abort attach on peer; | |
4403 | * | |
4404 | * If this node does not have good data, was already connected, but | |
4405 | * the peer did a late attach only now, trying to "negotiate" with me, | |
4406 | * AND I am currently Primary, possibly frozen, with some specific | |
4407 | * "effective" uuid, this should never be reached, really, because | |
4408 | * we first send the uuids, then the current state. | |
4409 | * | |
4410 | * In this scenario, we already dropped the connection hard | |
4411 | * when we received the unsuitable uuids (receive_uuids(). | |
4412 | * | |
4413 | * Should we want to change this, that is: not drop the connection in | |
4414 | * receive_uuids() already, then we would need to add a branch here | |
4415 | * that aborts the attach of "unsuitable uuids" on the peer in case | |
4416 | * this node is currently Diskless Primary. | |
4417 | */ | |
4418 | ||
b30ab791 AG |
4419 | if (device->p_uuid && peer_state.disk >= D_NEGOTIATING && |
4420 | get_ldev_if_state(device, D_NEGOTIATING)) { | |
b411b363 PR |
4421 | int cr; /* consider resync */ |
4422 | ||
4423 | /* if we established a new connection */ | |
4ac4aada | 4424 | cr = (os.conn < C_CONNECTED); |
b411b363 PR |
4425 | /* if we had an established connection |
4426 | * and one of the nodes newly attaches a disk */ | |
4ac4aada | 4427 | cr |= (os.conn == C_CONNECTED && |
b411b363 | 4428 | (peer_state.disk == D_NEGOTIATING || |
4ac4aada | 4429 | os.disk == D_NEGOTIATING)); |
b411b363 PR |
4430 | /* if we have both been inconsistent, and the peer has been |
4431 | * forced to be UpToDate with --overwrite-data */ | |
b30ab791 | 4432 | cr |= test_bit(CONSIDER_RESYNC, &device->flags); |
b411b363 PR |
4433 | /* if we had been plain connected, and the admin requested to |
4434 | * start a sync by "invalidate" or "invalidate-remote" */ | |
4ac4aada | 4435 | cr |= (os.conn == C_CONNECTED && |
b411b363 PR |
4436 | (peer_state.conn >= C_STARTING_SYNC_S && |
4437 | peer_state.conn <= C_WF_BITMAP_T)); | |
4438 | ||
4439 | if (cr) | |
69a22773 | 4440 | ns.conn = drbd_sync_handshake(peer_device, peer_state.role, real_peer_disk); |
b411b363 | 4441 | |
b30ab791 | 4442 | put_ldev(device); |
4ac4aada LE |
4443 | if (ns.conn == C_MASK) { |
4444 | ns.conn = C_CONNECTED; | |
b30ab791 AG |
4445 | if (device->state.disk == D_NEGOTIATING) { |
4446 | drbd_force_state(device, NS(disk, D_FAILED)); | |
b411b363 | 4447 | } else if (peer_state.disk == D_NEGOTIATING) { |
d0180171 | 4448 | drbd_err(device, "Disk attach process on the peer node was aborted.\n"); |
b411b363 | 4449 | peer_state.disk = D_DISKLESS; |
580b9767 | 4450 | real_peer_disk = D_DISKLESS; |
b411b363 | 4451 | } else { |
9f4fe9ad | 4452 | if (test_and_clear_bit(CONN_DRY_RUN, &peer_device->connection->flags)) |
82bc0194 | 4453 | return -EIO; |
0b0ba1ef | 4454 | D_ASSERT(device, os.conn == C_WF_REPORT_PARAMS); |
9f4fe9ad | 4455 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 4456 | return -EIO; |
b411b363 PR |
4457 | } |
4458 | } | |
4459 | } | |
4460 | ||
0500813f | 4461 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 4462 | if (os.i != drbd_read_state(device).i) |
b411b363 | 4463 | goto retry; |
b30ab791 | 4464 | clear_bit(CONSIDER_RESYNC, &device->flags); |
b411b363 PR |
4465 | ns.peer = peer_state.role; |
4466 | ns.pdsk = real_peer_disk; | |
4467 | ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); | |
4ac4aada | 4468 | if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) |
b30ab791 | 4469 | ns.disk = device->new_state_tmp.disk; |
4ac4aada | 4470 | cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); |
b30ab791 AG |
4471 | if (ns.pdsk == D_CONSISTENT && drbd_suspended(device) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && |
4472 | test_bit(NEW_CUR_UUID, &device->flags)) { | |
8554df1c | 4473 | /* Do not allow tl_restart(RESEND) for a rebooted peer. We can only allow this |
481c6f50 | 4474 | for temporal network outages! */ |
0500813f | 4475 | spin_unlock_irq(&device->resource->req_lock); |
d0180171 | 4476 | drbd_err(device, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); |
9f4fe9ad | 4477 | tl_clear(peer_device->connection); |
b30ab791 AG |
4478 | drbd_uuid_new_current(device); |
4479 | clear_bit(NEW_CUR_UUID, &device->flags); | |
9f4fe9ad | 4480 | conn_request_state(peer_device->connection, NS2(conn, C_PROTOCOL_ERROR, susp, 0), CS_HARD); |
82bc0194 | 4481 | return -EIO; |
481c6f50 | 4482 | } |
b30ab791 AG |
4483 | rv = _drbd_set_state(device, ns, cs_flags, NULL); |
4484 | ns = drbd_read_state(device); | |
0500813f | 4485 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
4486 | |
4487 | if (rv < SS_SUCCESS) { | |
9f4fe9ad | 4488 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 4489 | return -EIO; |
b411b363 PR |
4490 | } |
4491 | ||
4ac4aada LE |
4492 | if (os.conn > C_WF_REPORT_PARAMS) { |
4493 | if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && | |
b411b363 PR |
4494 | peer_state.disk != D_NEGOTIATING ) { |
4495 | /* we want resync, peer has not yet decided to sync... */ | |
4496 | /* Nowadays only used when forcing a node into primary role and | |
4497 | setting its disk to UpToDate with that */ | |
69a22773 AG |
4498 | drbd_send_uuids(peer_device); |
4499 | drbd_send_current_state(peer_device); | |
b411b363 PR |
4500 | } |
4501 | } | |
4502 | ||
b30ab791 | 4503 | clear_bit(DISCARD_MY_DATA, &device->flags); |
b411b363 | 4504 | |
b30ab791 | 4505 | drbd_md_sync(device); /* update connected indicator, la_size_sect, ... */ |
b411b363 | 4506 | |
82bc0194 | 4507 | return 0; |
b411b363 PR |
4508 | } |
4509 | ||
bde89a9e | 4510 | static int receive_sync_uuid(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4511 | { |
9f4fe9ad | 4512 | struct drbd_peer_device *peer_device; |
b30ab791 | 4513 | struct drbd_device *device; |
e658983a | 4514 | struct p_rs_uuid *p = pi->data; |
4a76b161 | 4515 | |
9f4fe9ad AG |
4516 | peer_device = conn_peer_device(connection, pi->vnr); |
4517 | if (!peer_device) | |
4a76b161 | 4518 | return -EIO; |
9f4fe9ad | 4519 | device = peer_device->device; |
b411b363 | 4520 | |
b30ab791 AG |
4521 | wait_event(device->misc_wait, |
4522 | device->state.conn == C_WF_SYNC_UUID || | |
4523 | device->state.conn == C_BEHIND || | |
4524 | device->state.conn < C_CONNECTED || | |
4525 | device->state.disk < D_NEGOTIATING); | |
b411b363 | 4526 | |
0b0ba1ef | 4527 | /* D_ASSERT(device, device->state.conn == C_WF_SYNC_UUID ); */ |
b411b363 | 4528 | |
b411b363 PR |
4529 | /* Here the _drbd_uuid_ functions are right, current should |
4530 | _not_ be rotated into the history */ | |
b30ab791 AG |
4531 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
4532 | _drbd_uuid_set(device, UI_CURRENT, be64_to_cpu(p->uuid)); | |
4533 | _drbd_uuid_set(device, UI_BITMAP, 0UL); | |
b411b363 | 4534 | |
b30ab791 AG |
4535 | drbd_print_uuids(device, "updated sync uuid"); |
4536 | drbd_start_resync(device, C_SYNC_TARGET); | |
b411b363 | 4537 | |
b30ab791 | 4538 | put_ldev(device); |
b411b363 | 4539 | } else |
d0180171 | 4540 | drbd_err(device, "Ignoring SyncUUID packet!\n"); |
b411b363 | 4541 | |
82bc0194 | 4542 | return 0; |
b411b363 PR |
4543 | } |
4544 | ||
2c46407d AG |
4545 | /** |
4546 | * receive_bitmap_plain | |
4547 | * | |
4548 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4549 | * code upon failure. | |
4550 | */ | |
4551 | static int | |
69a22773 | 4552 | receive_bitmap_plain(struct drbd_peer_device *peer_device, unsigned int size, |
e658983a | 4553 | unsigned long *p, struct bm_xfer_ctx *c) |
b411b363 | 4554 | { |
50d0b1ad | 4555 | unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - |
69a22773 | 4556 | drbd_header_size(peer_device->connection); |
e658983a | 4557 | unsigned int num_words = min_t(size_t, data_size / sizeof(*p), |
50d0b1ad | 4558 | c->bm_words - c->word_offset); |
e658983a | 4559 | unsigned int want = num_words * sizeof(*p); |
2c46407d | 4560 | int err; |
b411b363 | 4561 | |
50d0b1ad | 4562 | if (want != size) { |
69a22773 | 4563 | drbd_err(peer_device, "%s:want (%u) != size (%u)\n", __func__, want, size); |
2c46407d | 4564 | return -EIO; |
b411b363 PR |
4565 | } |
4566 | if (want == 0) | |
2c46407d | 4567 | return 0; |
69a22773 | 4568 | err = drbd_recv_all(peer_device->connection, p, want); |
82bc0194 | 4569 | if (err) |
2c46407d | 4570 | return err; |
b411b363 | 4571 | |
69a22773 | 4572 | drbd_bm_merge_lel(peer_device->device, c->word_offset, num_words, p); |
b411b363 PR |
4573 | |
4574 | c->word_offset += num_words; | |
4575 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
4576 | if (c->bit_offset > c->bm_bits) | |
4577 | c->bit_offset = c->bm_bits; | |
4578 | ||
2c46407d | 4579 | return 1; |
b411b363 PR |
4580 | } |
4581 | ||
a02d1240 AG |
4582 | static enum drbd_bitmap_code dcbp_get_code(struct p_compressed_bm *p) |
4583 | { | |
4584 | return (enum drbd_bitmap_code)(p->encoding & 0x0f); | |
4585 | } | |
4586 | ||
4587 | static int dcbp_get_start(struct p_compressed_bm *p) | |
4588 | { | |
4589 | return (p->encoding & 0x80) != 0; | |
4590 | } | |
4591 | ||
4592 | static int dcbp_get_pad_bits(struct p_compressed_bm *p) | |
4593 | { | |
4594 | return (p->encoding >> 4) & 0x7; | |
4595 | } | |
4596 | ||
2c46407d AG |
4597 | /** |
4598 | * recv_bm_rle_bits | |
4599 | * | |
4600 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4601 | * code upon failure. | |
4602 | */ | |
4603 | static int | |
69a22773 | 4604 | recv_bm_rle_bits(struct drbd_peer_device *peer_device, |
b411b363 | 4605 | struct p_compressed_bm *p, |
c6d25cfe PR |
4606 | struct bm_xfer_ctx *c, |
4607 | unsigned int len) | |
b411b363 PR |
4608 | { |
4609 | struct bitstream bs; | |
4610 | u64 look_ahead; | |
4611 | u64 rl; | |
4612 | u64 tmp; | |
4613 | unsigned long s = c->bit_offset; | |
4614 | unsigned long e; | |
a02d1240 | 4615 | int toggle = dcbp_get_start(p); |
b411b363 PR |
4616 | int have; |
4617 | int bits; | |
4618 | ||
a02d1240 | 4619 | bitstream_init(&bs, p->code, len, dcbp_get_pad_bits(p)); |
b411b363 PR |
4620 | |
4621 | bits = bitstream_get_bits(&bs, &look_ahead, 64); | |
4622 | if (bits < 0) | |
2c46407d | 4623 | return -EIO; |
b411b363 PR |
4624 | |
4625 | for (have = bits; have > 0; s += rl, toggle = !toggle) { | |
4626 | bits = vli_decode_bits(&rl, look_ahead); | |
4627 | if (bits <= 0) | |
2c46407d | 4628 | return -EIO; |
b411b363 PR |
4629 | |
4630 | if (toggle) { | |
4631 | e = s + rl -1; | |
4632 | if (e >= c->bm_bits) { | |
69a22773 | 4633 | drbd_err(peer_device, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); |
2c46407d | 4634 | return -EIO; |
b411b363 | 4635 | } |
69a22773 | 4636 | _drbd_bm_set_bits(peer_device->device, s, e); |
b411b363 PR |
4637 | } |
4638 | ||
4639 | if (have < bits) { | |
69a22773 | 4640 | drbd_err(peer_device, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", |
b411b363 PR |
4641 | have, bits, look_ahead, |
4642 | (unsigned int)(bs.cur.b - p->code), | |
4643 | (unsigned int)bs.buf_len); | |
2c46407d | 4644 | return -EIO; |
b411b363 | 4645 | } |
d2da5b0c LE |
4646 | /* if we consumed all 64 bits, assign 0; >> 64 is "undefined"; */ |
4647 | if (likely(bits < 64)) | |
4648 | look_ahead >>= bits; | |
4649 | else | |
4650 | look_ahead = 0; | |
b411b363 PR |
4651 | have -= bits; |
4652 | ||
4653 | bits = bitstream_get_bits(&bs, &tmp, 64 - have); | |
4654 | if (bits < 0) | |
2c46407d | 4655 | return -EIO; |
b411b363 PR |
4656 | look_ahead |= tmp << have; |
4657 | have += bits; | |
4658 | } | |
4659 | ||
4660 | c->bit_offset = s; | |
4661 | bm_xfer_ctx_bit_to_word_offset(c); | |
4662 | ||
2c46407d | 4663 | return (s != c->bm_bits); |
b411b363 PR |
4664 | } |
4665 | ||
2c46407d AG |
4666 | /** |
4667 | * decode_bitmap_c | |
4668 | * | |
4669 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4670 | * code upon failure. | |
4671 | */ | |
4672 | static int | |
69a22773 | 4673 | decode_bitmap_c(struct drbd_peer_device *peer_device, |
b411b363 | 4674 | struct p_compressed_bm *p, |
c6d25cfe PR |
4675 | struct bm_xfer_ctx *c, |
4676 | unsigned int len) | |
b411b363 | 4677 | { |
a02d1240 | 4678 | if (dcbp_get_code(p) == RLE_VLI_Bits) |
69a22773 | 4679 | return recv_bm_rle_bits(peer_device, p, c, len - sizeof(*p)); |
b411b363 PR |
4680 | |
4681 | /* other variants had been implemented for evaluation, | |
4682 | * but have been dropped as this one turned out to be "best" | |
4683 | * during all our tests. */ | |
4684 | ||
69a22773 AG |
4685 | drbd_err(peer_device, "receive_bitmap_c: unknown encoding %u\n", p->encoding); |
4686 | conn_request_state(peer_device->connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); | |
2c46407d | 4687 | return -EIO; |
b411b363 PR |
4688 | } |
4689 | ||
b30ab791 | 4690 | void INFO_bm_xfer_stats(struct drbd_device *device, |
b411b363 PR |
4691 | const char *direction, struct bm_xfer_ctx *c) |
4692 | { | |
4693 | /* what would it take to transfer it "plaintext" */ | |
a6b32bc3 | 4694 | unsigned int header_size = drbd_header_size(first_peer_device(device)->connection); |
50d0b1ad AG |
4695 | unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; |
4696 | unsigned int plain = | |
4697 | header_size * (DIV_ROUND_UP(c->bm_words, data_size) + 1) + | |
4698 | c->bm_words * sizeof(unsigned long); | |
4699 | unsigned int total = c->bytes[0] + c->bytes[1]; | |
4700 | unsigned int r; | |
b411b363 PR |
4701 | |
4702 | /* total can not be zero. but just in case: */ | |
4703 | if (total == 0) | |
4704 | return; | |
4705 | ||
4706 | /* don't report if not compressed */ | |
4707 | if (total >= plain) | |
4708 | return; | |
4709 | ||
4710 | /* total < plain. check for overflow, still */ | |
4711 | r = (total > UINT_MAX/1000) ? (total / (plain/1000)) | |
4712 | : (1000 * total / plain); | |
4713 | ||
4714 | if (r > 1000) | |
4715 | r = 1000; | |
4716 | ||
4717 | r = 1000 - r; | |
d0180171 | 4718 | drbd_info(device, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " |
b411b363 PR |
4719 | "total %u; compression: %u.%u%%\n", |
4720 | direction, | |
4721 | c->bytes[1], c->packets[1], | |
4722 | c->bytes[0], c->packets[0], | |
4723 | total, r/10, r % 10); | |
4724 | } | |
4725 | ||
4726 | /* Since we are processing the bitfield from lower addresses to higher, | |
4727 | it does not matter if the process it in 32 bit chunks or 64 bit | |
4728 | chunks as long as it is little endian. (Understand it as byte stream, | |
4729 | beginning with the lowest byte...) If we would use big endian | |
4730 | we would need to process it from the highest address to the lowest, | |
4731 | in order to be agnostic to the 32 vs 64 bits issue. | |
4732 | ||
4733 | returns 0 on failure, 1 if we successfully received it. */ | |
bde89a9e | 4734 | static int receive_bitmap(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4735 | { |
9f4fe9ad | 4736 | struct drbd_peer_device *peer_device; |
b30ab791 | 4737 | struct drbd_device *device; |
b411b363 | 4738 | struct bm_xfer_ctx c; |
2c46407d | 4739 | int err; |
4a76b161 | 4740 | |
9f4fe9ad AG |
4741 | peer_device = conn_peer_device(connection, pi->vnr); |
4742 | if (!peer_device) | |
4a76b161 | 4743 | return -EIO; |
9f4fe9ad | 4744 | device = peer_device->device; |
b411b363 | 4745 | |
b30ab791 | 4746 | drbd_bm_lock(device, "receive bitmap", BM_LOCKED_SET_ALLOWED); |
20ceb2b2 LE |
4747 | /* you are supposed to send additional out-of-sync information |
4748 | * if you actually set bits during this phase */ | |
b411b363 | 4749 | |
b411b363 | 4750 | c = (struct bm_xfer_ctx) { |
b30ab791 AG |
4751 | .bm_bits = drbd_bm_bits(device), |
4752 | .bm_words = drbd_bm_words(device), | |
b411b363 PR |
4753 | }; |
4754 | ||
2c46407d | 4755 | for(;;) { |
e658983a | 4756 | if (pi->cmd == P_BITMAP) |
69a22773 | 4757 | err = receive_bitmap_plain(peer_device, pi->size, pi->data, &c); |
e658983a | 4758 | else if (pi->cmd == P_COMPRESSED_BITMAP) { |
b411b363 PR |
4759 | /* MAYBE: sanity check that we speak proto >= 90, |
4760 | * and the feature is enabled! */ | |
e658983a | 4761 | struct p_compressed_bm *p = pi->data; |
b411b363 | 4762 | |
bde89a9e | 4763 | if (pi->size > DRBD_SOCKET_BUFFER_SIZE - drbd_header_size(connection)) { |
d0180171 | 4764 | drbd_err(device, "ReportCBitmap packet too large\n"); |
82bc0194 | 4765 | err = -EIO; |
b411b363 PR |
4766 | goto out; |
4767 | } | |
e658983a | 4768 | if (pi->size <= sizeof(*p)) { |
d0180171 | 4769 | drbd_err(device, "ReportCBitmap packet too small (l:%u)\n", pi->size); |
82bc0194 | 4770 | err = -EIO; |
78fcbdae | 4771 | goto out; |
b411b363 | 4772 | } |
9f4fe9ad | 4773 | err = drbd_recv_all(peer_device->connection, p, pi->size); |
e658983a AG |
4774 | if (err) |
4775 | goto out; | |
69a22773 | 4776 | err = decode_bitmap_c(peer_device, p, &c, pi->size); |
b411b363 | 4777 | } else { |
d0180171 | 4778 | drbd_warn(device, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", pi->cmd); |
82bc0194 | 4779 | err = -EIO; |
b411b363 PR |
4780 | goto out; |
4781 | } | |
4782 | ||
e2857216 | 4783 | c.packets[pi->cmd == P_BITMAP]++; |
bde89a9e | 4784 | c.bytes[pi->cmd == P_BITMAP] += drbd_header_size(connection) + pi->size; |
b411b363 | 4785 | |
2c46407d AG |
4786 | if (err <= 0) { |
4787 | if (err < 0) | |
4788 | goto out; | |
b411b363 | 4789 | break; |
2c46407d | 4790 | } |
9f4fe9ad | 4791 | err = drbd_recv_header(peer_device->connection, pi); |
82bc0194 | 4792 | if (err) |
b411b363 | 4793 | goto out; |
2c46407d | 4794 | } |
b411b363 | 4795 | |
b30ab791 | 4796 | INFO_bm_xfer_stats(device, "receive", &c); |
b411b363 | 4797 | |
b30ab791 | 4798 | if (device->state.conn == C_WF_BITMAP_T) { |
de1f8e4a AG |
4799 | enum drbd_state_rv rv; |
4800 | ||
b30ab791 | 4801 | err = drbd_send_bitmap(device); |
82bc0194 | 4802 | if (err) |
b411b363 PR |
4803 | goto out; |
4804 | /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ | |
b30ab791 | 4805 | rv = _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); |
0b0ba1ef | 4806 | D_ASSERT(device, rv == SS_SUCCESS); |
b30ab791 | 4807 | } else if (device->state.conn != C_WF_BITMAP_S) { |
b411b363 PR |
4808 | /* admin may have requested C_DISCONNECTING, |
4809 | * other threads may have noticed network errors */ | |
d0180171 | 4810 | drbd_info(device, "unexpected cstate (%s) in receive_bitmap\n", |
b30ab791 | 4811 | drbd_conn_str(device->state.conn)); |
b411b363 | 4812 | } |
82bc0194 | 4813 | err = 0; |
b411b363 | 4814 | |
b411b363 | 4815 | out: |
b30ab791 AG |
4816 | drbd_bm_unlock(device); |
4817 | if (!err && device->state.conn == C_WF_BITMAP_S) | |
4818 | drbd_start_resync(device, C_SYNC_SOURCE); | |
82bc0194 | 4819 | return err; |
b411b363 PR |
4820 | } |
4821 | ||
bde89a9e | 4822 | static int receive_skip(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4823 | { |
1ec861eb | 4824 | drbd_warn(connection, "skipping unknown optional packet type %d, l: %d!\n", |
e2857216 | 4825 | pi->cmd, pi->size); |
b411b363 | 4826 | |
bde89a9e | 4827 | return ignore_remaining_packet(connection, pi); |
b411b363 PR |
4828 | } |
4829 | ||
bde89a9e | 4830 | static int receive_UnplugRemote(struct drbd_connection *connection, struct packet_info *pi) |
0ced55a3 | 4831 | { |
e7f52dfb LE |
4832 | /* Make sure we've acked all the TCP data associated |
4833 | * with the data requests being unplugged */ | |
bde89a9e | 4834 | drbd_tcp_quickack(connection->data.socket); |
0ced55a3 | 4835 | |
82bc0194 | 4836 | return 0; |
0ced55a3 PR |
4837 | } |
4838 | ||
bde89a9e | 4839 | static int receive_out_of_sync(struct drbd_connection *connection, struct packet_info *pi) |
73a01a18 | 4840 | { |
9f4fe9ad | 4841 | struct drbd_peer_device *peer_device; |
b30ab791 | 4842 | struct drbd_device *device; |
e658983a | 4843 | struct p_block_desc *p = pi->data; |
4a76b161 | 4844 | |
9f4fe9ad AG |
4845 | peer_device = conn_peer_device(connection, pi->vnr); |
4846 | if (!peer_device) | |
4a76b161 | 4847 | return -EIO; |
9f4fe9ad | 4848 | device = peer_device->device; |
73a01a18 | 4849 | |
b30ab791 | 4850 | switch (device->state.conn) { |
f735e363 LE |
4851 | case C_WF_SYNC_UUID: |
4852 | case C_WF_BITMAP_T: | |
4853 | case C_BEHIND: | |
4854 | break; | |
4855 | default: | |
d0180171 | 4856 | drbd_err(device, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n", |
b30ab791 | 4857 | drbd_conn_str(device->state.conn)); |
f735e363 LE |
4858 | } |
4859 | ||
b30ab791 | 4860 | drbd_set_out_of_sync(device, be64_to_cpu(p->sector), be32_to_cpu(p->blksize)); |
73a01a18 | 4861 | |
82bc0194 | 4862 | return 0; |
73a01a18 PR |
4863 | } |
4864 | ||
700ca8c0 PR |
4865 | static int receive_rs_deallocated(struct drbd_connection *connection, struct packet_info *pi) |
4866 | { | |
4867 | struct drbd_peer_device *peer_device; | |
4868 | struct p_block_desc *p = pi->data; | |
4869 | struct drbd_device *device; | |
4870 | sector_t sector; | |
4871 | int size, err = 0; | |
4872 | ||
4873 | peer_device = conn_peer_device(connection, pi->vnr); | |
4874 | if (!peer_device) | |
4875 | return -EIO; | |
4876 | device = peer_device->device; | |
4877 | ||
4878 | sector = be64_to_cpu(p->sector); | |
4879 | size = be32_to_cpu(p->blksize); | |
4880 | ||
4881 | dec_rs_pending(device); | |
4882 | ||
4883 | if (get_ldev(device)) { | |
4884 | struct drbd_peer_request *peer_req; | |
45c21793 | 4885 | const int op = REQ_OP_WRITE_ZEROES; |
700ca8c0 PR |
4886 | |
4887 | peer_req = drbd_alloc_peer_req(peer_device, ID_SYNCER, sector, | |
9104d31a | 4888 | size, 0, GFP_NOIO); |
700ca8c0 PR |
4889 | if (!peer_req) { |
4890 | put_ldev(device); | |
4891 | return -ENOMEM; | |
4892 | } | |
4893 | ||
4894 | peer_req->w.cb = e_end_resync_block; | |
4895 | peer_req->submit_jif = jiffies; | |
4896 | peer_req->flags |= EE_IS_TRIM; | |
4897 | ||
4898 | spin_lock_irq(&device->resource->req_lock); | |
4899 | list_add_tail(&peer_req->w.list, &device->sync_ee); | |
4900 | spin_unlock_irq(&device->resource->req_lock); | |
4901 | ||
4902 | atomic_add(pi->size >> 9, &device->rs_sect_ev); | |
4903 | err = drbd_submit_peer_request(device, peer_req, op, 0, DRBD_FAULT_RS_WR); | |
4904 | ||
4905 | if (err) { | |
4906 | spin_lock_irq(&device->resource->req_lock); | |
4907 | list_del(&peer_req->w.list); | |
4908 | spin_unlock_irq(&device->resource->req_lock); | |
4909 | ||
4910 | drbd_free_peer_req(device, peer_req); | |
4911 | put_ldev(device); | |
4912 | err = 0; | |
4913 | goto fail; | |
4914 | } | |
4915 | ||
4916 | inc_unacked(device); | |
4917 | ||
4918 | /* No put_ldev() here. Gets called in drbd_endio_write_sec_final(), | |
4919 | as well as drbd_rs_complete_io() */ | |
4920 | } else { | |
4921 | fail: | |
4922 | drbd_rs_complete_io(device, sector); | |
4923 | drbd_send_ack_ex(peer_device, P_NEG_ACK, sector, size, ID_SYNCER); | |
4924 | } | |
4925 | ||
4926 | atomic_add(size >> 9, &device->rs_sect_in); | |
4927 | ||
4928 | return err; | |
4929 | } | |
4930 | ||
02918be2 PR |
4931 | struct data_cmd { |
4932 | int expect_payload; | |
9104d31a | 4933 | unsigned int pkt_size; |
bde89a9e | 4934 | int (*fn)(struct drbd_connection *, struct packet_info *); |
02918be2 PR |
4935 | }; |
4936 | ||
4937 | static struct data_cmd drbd_cmd_handler[] = { | |
4938 | [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, | |
4939 | [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, | |
4940 | [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , | |
4941 | [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , | |
e658983a AG |
4942 | [P_BITMAP] = { 1, 0, receive_bitmap } , |
4943 | [P_COMPRESSED_BITMAP] = { 1, 0, receive_bitmap } , | |
4944 | [P_UNPLUG_REMOTE] = { 0, 0, receive_UnplugRemote }, | |
02918be2 PR |
4945 | [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, |
4946 | [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
e658983a AG |
4947 | [P_SYNC_PARAM] = { 1, 0, receive_SyncParam }, |
4948 | [P_SYNC_PARAM89] = { 1, 0, receive_SyncParam }, | |
02918be2 PR |
4949 | [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, |
4950 | [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, | |
4951 | [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, | |
4952 | [P_STATE] = { 0, sizeof(struct p_state), receive_state }, | |
4953 | [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, | |
4954 | [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, | |
4955 | [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
4956 | [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
4957 | [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
700ca8c0 | 4958 | [P_RS_THIN_REQ] = { 0, sizeof(struct p_block_req), receive_DataRequest }, |
02918be2 | 4959 | [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, |
73a01a18 | 4960 | [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync }, |
4a76b161 | 4961 | [P_CONN_ST_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_conn_state }, |
036b17ea | 4962 | [P_PROTOCOL_UPDATE] = { 1, sizeof(struct p_protocol), receive_protocol }, |
a0fb3c47 | 4963 | [P_TRIM] = { 0, sizeof(struct p_trim), receive_Data }, |
700ca8c0 | 4964 | [P_RS_DEALLOCATED] = { 0, sizeof(struct p_block_desc), receive_rs_deallocated }, |
9104d31a | 4965 | [P_WSAME] = { 1, sizeof(struct p_wsame), receive_Data }, |
b411b363 PR |
4966 | }; |
4967 | ||
bde89a9e | 4968 | static void drbdd(struct drbd_connection *connection) |
b411b363 | 4969 | { |
77351055 | 4970 | struct packet_info pi; |
02918be2 | 4971 | size_t shs; /* sub header size */ |
82bc0194 | 4972 | int err; |
b411b363 | 4973 | |
bde89a9e | 4974 | while (get_t_state(&connection->receiver) == RUNNING) { |
9104d31a | 4975 | struct data_cmd const *cmd; |
b411b363 | 4976 | |
bde89a9e | 4977 | drbd_thread_current_set_cpu(&connection->receiver); |
c51a0ef3 LE |
4978 | update_receiver_timing_details(connection, drbd_recv_header_maybe_unplug); |
4979 | if (drbd_recv_header_maybe_unplug(connection, &pi)) | |
02918be2 | 4980 | goto err_out; |
b411b363 | 4981 | |
deebe195 | 4982 | cmd = &drbd_cmd_handler[pi.cmd]; |
4a76b161 | 4983 | if (unlikely(pi.cmd >= ARRAY_SIZE(drbd_cmd_handler) || !cmd->fn)) { |
1ec861eb | 4984 | drbd_err(connection, "Unexpected data packet %s (0x%04x)", |
2fcb8f30 | 4985 | cmdname(pi.cmd), pi.cmd); |
02918be2 | 4986 | goto err_out; |
0b33a916 | 4987 | } |
b411b363 | 4988 | |
e658983a | 4989 | shs = cmd->pkt_size; |
9104d31a LE |
4990 | if (pi.cmd == P_SIZES && connection->agreed_features & DRBD_FF_WSAME) |
4991 | shs += sizeof(struct o_qlim); | |
e658983a | 4992 | if (pi.size > shs && !cmd->expect_payload) { |
1ec861eb | 4993 | drbd_err(connection, "No payload expected %s l:%d\n", |
2fcb8f30 | 4994 | cmdname(pi.cmd), pi.size); |
02918be2 | 4995 | goto err_out; |
b411b363 | 4996 | } |
9104d31a LE |
4997 | if (pi.size < shs) { |
4998 | drbd_err(connection, "%s: unexpected packet size, expected:%d received:%d\n", | |
4999 | cmdname(pi.cmd), (int)shs, pi.size); | |
5000 | goto err_out; | |
5001 | } | |
b411b363 | 5002 | |
c13f7e1a | 5003 | if (shs) { |
944410e9 | 5004 | update_receiver_timing_details(connection, drbd_recv_all_warn); |
bde89a9e | 5005 | err = drbd_recv_all_warn(connection, pi.data, shs); |
a5c31904 | 5006 | if (err) |
c13f7e1a | 5007 | goto err_out; |
e2857216 | 5008 | pi.size -= shs; |
c13f7e1a LE |
5009 | } |
5010 | ||
944410e9 | 5011 | update_receiver_timing_details(connection, cmd->fn); |
bde89a9e | 5012 | err = cmd->fn(connection, &pi); |
4a76b161 | 5013 | if (err) { |
1ec861eb | 5014 | drbd_err(connection, "error receiving %s, e: %d l: %d!\n", |
9f5bdc33 | 5015 | cmdname(pi.cmd), err, pi.size); |
02918be2 | 5016 | goto err_out; |
b411b363 PR |
5017 | } |
5018 | } | |
82bc0194 | 5019 | return; |
b411b363 | 5020 | |
82bc0194 | 5021 | err_out: |
bde89a9e | 5022 | conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); |
b411b363 PR |
5023 | } |
5024 | ||
bde89a9e | 5025 | static void conn_disconnect(struct drbd_connection *connection) |
b411b363 | 5026 | { |
c06ece6b | 5027 | struct drbd_peer_device *peer_device; |
bbeb641c | 5028 | enum drbd_conns oc; |
376694a0 | 5029 | int vnr; |
b411b363 | 5030 | |
bde89a9e | 5031 | if (connection->cstate == C_STANDALONE) |
b411b363 | 5032 | return; |
b411b363 | 5033 | |
545752d5 LE |
5034 | /* We are about to start the cleanup after connection loss. |
5035 | * Make sure drbd_make_request knows about that. | |
5036 | * Usually we should be in some network failure state already, | |
5037 | * but just in case we are not, we fix it up here. | |
5038 | */ | |
bde89a9e | 5039 | conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); |
545752d5 | 5040 | |
668700b4 | 5041 | /* ack_receiver does not clean up anything. it must not interfere, either */ |
1c03e520 | 5042 | drbd_thread_stop(&connection->ack_receiver); |
668700b4 PR |
5043 | if (connection->ack_sender) { |
5044 | destroy_workqueue(connection->ack_sender); | |
5045 | connection->ack_sender = NULL; | |
5046 | } | |
bde89a9e | 5047 | drbd_free_sock(connection); |
360cc740 | 5048 | |
c141ebda | 5049 | rcu_read_lock(); |
c06ece6b AG |
5050 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
5051 | struct drbd_device *device = peer_device->device; | |
b30ab791 | 5052 | kref_get(&device->kref); |
c141ebda | 5053 | rcu_read_unlock(); |
69a22773 | 5054 | drbd_disconnected(peer_device); |
c06ece6b | 5055 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda PR |
5056 | rcu_read_lock(); |
5057 | } | |
5058 | rcu_read_unlock(); | |
5059 | ||
bde89a9e | 5060 | if (!list_empty(&connection->current_epoch->list)) |
1ec861eb | 5061 | drbd_err(connection, "ASSERTION FAILED: connection->current_epoch->list not empty\n"); |
12038a3a | 5062 | /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ |
bde89a9e AG |
5063 | atomic_set(&connection->current_epoch->epoch_size, 0); |
5064 | connection->send.seen_any_write_yet = false; | |
12038a3a | 5065 | |
1ec861eb | 5066 | drbd_info(connection, "Connection closed\n"); |
360cc740 | 5067 | |
bde89a9e AG |
5068 | if (conn_highest_role(connection) == R_PRIMARY && conn_highest_pdsk(connection) >= D_UNKNOWN) |
5069 | conn_try_outdate_peer_async(connection); | |
cb703454 | 5070 | |
0500813f | 5071 | spin_lock_irq(&connection->resource->req_lock); |
bde89a9e | 5072 | oc = connection->cstate; |
bbeb641c | 5073 | if (oc >= C_UNCONNECTED) |
bde89a9e | 5074 | _conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE); |
bbeb641c | 5075 | |
0500813f | 5076 | spin_unlock_irq(&connection->resource->req_lock); |
360cc740 | 5077 | |
f3dfa40a | 5078 | if (oc == C_DISCONNECTING) |
bde89a9e | 5079 | conn_request_state(connection, NS(conn, C_STANDALONE), CS_VERBOSE | CS_HARD); |
360cc740 PR |
5080 | } |
5081 | ||
69a22773 | 5082 | static int drbd_disconnected(struct drbd_peer_device *peer_device) |
360cc740 | 5083 | { |
69a22773 | 5084 | struct drbd_device *device = peer_device->device; |
360cc740 | 5085 | unsigned int i; |
b411b363 | 5086 | |
85719573 | 5087 | /* wait for current activity to cease. */ |
0500813f | 5088 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 AG |
5089 | _drbd_wait_ee_list_empty(device, &device->active_ee); |
5090 | _drbd_wait_ee_list_empty(device, &device->sync_ee); | |
5091 | _drbd_wait_ee_list_empty(device, &device->read_ee); | |
0500813f | 5092 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
5093 | |
5094 | /* We do not have data structures that would allow us to | |
5095 | * get the rs_pending_cnt down to 0 again. | |
5096 | * * On C_SYNC_TARGET we do not have any data structures describing | |
5097 | * the pending RSDataRequest's we have sent. | |
5098 | * * On C_SYNC_SOURCE there is no data structure that tracks | |
5099 | * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. | |
5100 | * And no, it is not the sum of the reference counts in the | |
5101 | * resync_LRU. The resync_LRU tracks the whole operation including | |
5102 | * the disk-IO, while the rs_pending_cnt only tracks the blocks | |
5103 | * on the fly. */ | |
b30ab791 AG |
5104 | drbd_rs_cancel_all(device); |
5105 | device->rs_total = 0; | |
5106 | device->rs_failed = 0; | |
5107 | atomic_set(&device->rs_pending_cnt, 0); | |
5108 | wake_up(&device->misc_wait); | |
b411b363 | 5109 | |
b30ab791 | 5110 | del_timer_sync(&device->resync_timer); |
2bccef39 | 5111 | resync_timer_fn(&device->resync_timer); |
b411b363 | 5112 | |
b411b363 PR |
5113 | /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, |
5114 | * w_make_resync_request etc. which may still be on the worker queue | |
5115 | * to be "canceled" */ | |
b5043c5e | 5116 | drbd_flush_workqueue(&peer_device->connection->sender_work); |
b411b363 | 5117 | |
b30ab791 | 5118 | drbd_finish_peer_reqs(device); |
b411b363 | 5119 | |
d10b4ea3 PR |
5120 | /* This second workqueue flush is necessary, since drbd_finish_peer_reqs() |
5121 | might have issued a work again. The one before drbd_finish_peer_reqs() is | |
5122 | necessary to reclain net_ee in drbd_finish_peer_reqs(). */ | |
b5043c5e | 5123 | drbd_flush_workqueue(&peer_device->connection->sender_work); |
d10b4ea3 | 5124 | |
08332d73 LE |
5125 | /* need to do it again, drbd_finish_peer_reqs() may have populated it |
5126 | * again via drbd_try_clear_on_disk_bm(). */ | |
b30ab791 | 5127 | drbd_rs_cancel_all(device); |
b411b363 | 5128 | |
b30ab791 AG |
5129 | kfree(device->p_uuid); |
5130 | device->p_uuid = NULL; | |
b411b363 | 5131 | |
b30ab791 | 5132 | if (!drbd_suspended(device)) |
69a22773 | 5133 | tl_clear(peer_device->connection); |
b411b363 | 5134 | |
b30ab791 | 5135 | drbd_md_sync(device); |
b411b363 | 5136 | |
be115b69 LE |
5137 | if (get_ldev(device)) { |
5138 | drbd_bitmap_io(device, &drbd_bm_write_copy_pages, | |
5139 | "write from disconnected", BM_LOCKED_CHANGE_ALLOWED); | |
5140 | put_ldev(device); | |
5141 | } | |
20ceb2b2 | 5142 | |
b411b363 PR |
5143 | /* tcp_close and release of sendpage pages can be deferred. I don't |
5144 | * want to use SO_LINGER, because apparently it can be deferred for | |
5145 | * more than 20 seconds (longest time I checked). | |
5146 | * | |
5147 | * Actually we don't care for exactly when the network stack does its | |
5148 | * put_page(), but release our reference on these pages right here. | |
5149 | */ | |
b30ab791 | 5150 | i = drbd_free_peer_reqs(device, &device->net_ee); |
b411b363 | 5151 | if (i) |
d0180171 | 5152 | drbd_info(device, "net_ee not empty, killed %u entries\n", i); |
b30ab791 | 5153 | i = atomic_read(&device->pp_in_use_by_net); |
435f0740 | 5154 | if (i) |
d0180171 | 5155 | drbd_info(device, "pp_in_use_by_net = %d, expected 0\n", i); |
b30ab791 | 5156 | i = atomic_read(&device->pp_in_use); |
b411b363 | 5157 | if (i) |
d0180171 | 5158 | drbd_info(device, "pp_in_use = %d, expected 0\n", i); |
b411b363 | 5159 | |
0b0ba1ef AG |
5160 | D_ASSERT(device, list_empty(&device->read_ee)); |
5161 | D_ASSERT(device, list_empty(&device->active_ee)); | |
5162 | D_ASSERT(device, list_empty(&device->sync_ee)); | |
5163 | D_ASSERT(device, list_empty(&device->done_ee)); | |
b411b363 | 5164 | |
360cc740 | 5165 | return 0; |
b411b363 PR |
5166 | } |
5167 | ||
5168 | /* | |
5169 | * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version | |
5170 | * we can agree on is stored in agreed_pro_version. | |
5171 | * | |
5172 | * feature flags and the reserved array should be enough room for future | |
5173 | * enhancements of the handshake protocol, and possible plugins... | |
5174 | * | |
5175 | * for now, they are expected to be zero, but ignored. | |
5176 | */ | |
bde89a9e | 5177 | static int drbd_send_features(struct drbd_connection *connection) |
b411b363 | 5178 | { |
9f5bdc33 AG |
5179 | struct drbd_socket *sock; |
5180 | struct p_connection_features *p; | |
b411b363 | 5181 | |
bde89a9e AG |
5182 | sock = &connection->data; |
5183 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 | 5184 | if (!p) |
e8d17b01 | 5185 | return -EIO; |
b411b363 PR |
5186 | memset(p, 0, sizeof(*p)); |
5187 | p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); | |
5188 | p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); | |
20c68fde | 5189 | p->feature_flags = cpu_to_be32(PRO_FEATURES); |
bde89a9e | 5190 | return conn_send_command(connection, sock, P_CONNECTION_FEATURES, sizeof(*p), NULL, 0); |
b411b363 PR |
5191 | } |
5192 | ||
5193 | /* | |
5194 | * return values: | |
5195 | * 1 yes, we have a valid connection | |
5196 | * 0 oops, did not work out, please try again | |
5197 | * -1 peer talks different language, | |
5198 | * no point in trying again, please go standalone. | |
5199 | */ | |
bde89a9e | 5200 | static int drbd_do_features(struct drbd_connection *connection) |
b411b363 | 5201 | { |
bde89a9e | 5202 | /* ASSERT current == connection->receiver ... */ |
e658983a AG |
5203 | struct p_connection_features *p; |
5204 | const int expect = sizeof(struct p_connection_features); | |
77351055 | 5205 | struct packet_info pi; |
a5c31904 | 5206 | int err; |
b411b363 | 5207 | |
bde89a9e | 5208 | err = drbd_send_features(connection); |
e8d17b01 | 5209 | if (err) |
b411b363 PR |
5210 | return 0; |
5211 | ||
bde89a9e | 5212 | err = drbd_recv_header(connection, &pi); |
69bc7bc3 | 5213 | if (err) |
b411b363 PR |
5214 | return 0; |
5215 | ||
6038178e | 5216 | if (pi.cmd != P_CONNECTION_FEATURES) { |
1ec861eb | 5217 | drbd_err(connection, "expected ConnectionFeatures packet, received: %s (0x%04x)\n", |
2fcb8f30 | 5218 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
5219 | return -1; |
5220 | } | |
5221 | ||
77351055 | 5222 | if (pi.size != expect) { |
1ec861eb | 5223 | drbd_err(connection, "expected ConnectionFeatures length: %u, received: %u\n", |
77351055 | 5224 | expect, pi.size); |
b411b363 PR |
5225 | return -1; |
5226 | } | |
5227 | ||
e658983a | 5228 | p = pi.data; |
bde89a9e | 5229 | err = drbd_recv_all_warn(connection, p, expect); |
a5c31904 | 5230 | if (err) |
b411b363 | 5231 | return 0; |
b411b363 | 5232 | |
b411b363 PR |
5233 | p->protocol_min = be32_to_cpu(p->protocol_min); |
5234 | p->protocol_max = be32_to_cpu(p->protocol_max); | |
5235 | if (p->protocol_max == 0) | |
5236 | p->protocol_max = p->protocol_min; | |
5237 | ||
5238 | if (PRO_VERSION_MAX < p->protocol_min || | |
5239 | PRO_VERSION_MIN > p->protocol_max) | |
5240 | goto incompat; | |
5241 | ||
bde89a9e | 5242 | connection->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); |
20c68fde | 5243 | connection->agreed_features = PRO_FEATURES & be32_to_cpu(p->feature_flags); |
b411b363 | 5244 | |
1ec861eb | 5245 | drbd_info(connection, "Handshake successful: " |
bde89a9e | 5246 | "Agreed network protocol version %d\n", connection->agreed_pro_version); |
b411b363 | 5247 | |
9104d31a LE |
5248 | drbd_info(connection, "Feature flags enabled on protocol level: 0x%x%s%s%s.\n", |
5249 | connection->agreed_features, | |
5250 | connection->agreed_features & DRBD_FF_TRIM ? " TRIM" : "", | |
5251 | connection->agreed_features & DRBD_FF_THIN_RESYNC ? " THIN_RESYNC" : "", | |
5252 | connection->agreed_features & DRBD_FF_WSAME ? " WRITE_SAME" : | |
5253 | connection->agreed_features ? "" : " none"); | |
92d94ae6 | 5254 | |
b411b363 PR |
5255 | return 1; |
5256 | ||
5257 | incompat: | |
1ec861eb | 5258 | drbd_err(connection, "incompatible DRBD dialects: " |
b411b363 PR |
5259 | "I support %d-%d, peer supports %d-%d\n", |
5260 | PRO_VERSION_MIN, PRO_VERSION_MAX, | |
5261 | p->protocol_min, p->protocol_max); | |
5262 | return -1; | |
5263 | } | |
5264 | ||
5265 | #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) | |
bde89a9e | 5266 | static int drbd_do_auth(struct drbd_connection *connection) |
b411b363 | 5267 | { |
1ec861eb AG |
5268 | drbd_err(connection, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); |
5269 | drbd_err(connection, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); | |
b10d96cb | 5270 | return -1; |
b411b363 PR |
5271 | } |
5272 | #else | |
5273 | #define CHALLENGE_LEN 64 | |
b10d96cb JT |
5274 | |
5275 | /* Return value: | |
5276 | 1 - auth succeeded, | |
5277 | 0 - failed, try again (network error), | |
5278 | -1 - auth failed, don't try again. | |
5279 | */ | |
5280 | ||
bde89a9e | 5281 | static int drbd_do_auth(struct drbd_connection *connection) |
b411b363 | 5282 | { |
9f5bdc33 | 5283 | struct drbd_socket *sock; |
b411b363 | 5284 | char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ |
b411b363 PR |
5285 | char *response = NULL; |
5286 | char *right_response = NULL; | |
5287 | char *peers_ch = NULL; | |
44ed167d PR |
5288 | unsigned int key_len; |
5289 | char secret[SHARED_SECRET_MAX]; /* 64 byte */ | |
b411b363 | 5290 | unsigned int resp_size; |
9534d671 | 5291 | SHASH_DESC_ON_STACK(desc, connection->cram_hmac_tfm); |
77351055 | 5292 | struct packet_info pi; |
44ed167d | 5293 | struct net_conf *nc; |
69bc7bc3 | 5294 | int err, rv; |
b411b363 | 5295 | |
9f5bdc33 | 5296 | /* FIXME: Put the challenge/response into the preallocated socket buffer. */ |
b411b363 | 5297 | |
44ed167d | 5298 | rcu_read_lock(); |
bde89a9e | 5299 | nc = rcu_dereference(connection->net_conf); |
44ed167d PR |
5300 | key_len = strlen(nc->shared_secret); |
5301 | memcpy(secret, nc->shared_secret, key_len); | |
5302 | rcu_read_unlock(); | |
5303 | ||
9534d671 HX |
5304 | desc->tfm = connection->cram_hmac_tfm; |
5305 | desc->flags = 0; | |
b411b363 | 5306 | |
9534d671 | 5307 | rv = crypto_shash_setkey(connection->cram_hmac_tfm, (u8 *)secret, key_len); |
b411b363 | 5308 | if (rv) { |
9534d671 | 5309 | drbd_err(connection, "crypto_shash_setkey() failed with %d\n", rv); |
b10d96cb | 5310 | rv = -1; |
b411b363 PR |
5311 | goto fail; |
5312 | } | |
5313 | ||
5314 | get_random_bytes(my_challenge, CHALLENGE_LEN); | |
5315 | ||
bde89a9e AG |
5316 | sock = &connection->data; |
5317 | if (!conn_prepare_command(connection, sock)) { | |
9f5bdc33 AG |
5318 | rv = 0; |
5319 | goto fail; | |
5320 | } | |
bde89a9e | 5321 | rv = !conn_send_command(connection, sock, P_AUTH_CHALLENGE, 0, |
9f5bdc33 | 5322 | my_challenge, CHALLENGE_LEN); |
b411b363 PR |
5323 | if (!rv) |
5324 | goto fail; | |
5325 | ||
bde89a9e | 5326 | err = drbd_recv_header(connection, &pi); |
69bc7bc3 AG |
5327 | if (err) { |
5328 | rv = 0; | |
b411b363 | 5329 | goto fail; |
69bc7bc3 | 5330 | } |
b411b363 | 5331 | |
77351055 | 5332 | if (pi.cmd != P_AUTH_CHALLENGE) { |
1ec861eb | 5333 | drbd_err(connection, "expected AuthChallenge packet, received: %s (0x%04x)\n", |
2fcb8f30 | 5334 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
5335 | rv = 0; |
5336 | goto fail; | |
5337 | } | |
5338 | ||
77351055 | 5339 | if (pi.size > CHALLENGE_LEN * 2) { |
1ec861eb | 5340 | drbd_err(connection, "expected AuthChallenge payload too big.\n"); |
b10d96cb | 5341 | rv = -1; |
b411b363 PR |
5342 | goto fail; |
5343 | } | |
5344 | ||
67cca286 PR |
5345 | if (pi.size < CHALLENGE_LEN) { |
5346 | drbd_err(connection, "AuthChallenge payload too small.\n"); | |
5347 | rv = -1; | |
5348 | goto fail; | |
5349 | } | |
5350 | ||
77351055 | 5351 | peers_ch = kmalloc(pi.size, GFP_NOIO); |
b411b363 | 5352 | if (peers_ch == NULL) { |
1ec861eb | 5353 | drbd_err(connection, "kmalloc of peers_ch failed\n"); |
b10d96cb | 5354 | rv = -1; |
b411b363 PR |
5355 | goto fail; |
5356 | } | |
5357 | ||
bde89a9e | 5358 | err = drbd_recv_all_warn(connection, peers_ch, pi.size); |
a5c31904 | 5359 | if (err) { |
b411b363 PR |
5360 | rv = 0; |
5361 | goto fail; | |
5362 | } | |
5363 | ||
67cca286 PR |
5364 | if (!memcmp(my_challenge, peers_ch, CHALLENGE_LEN)) { |
5365 | drbd_err(connection, "Peer presented the same challenge!\n"); | |
5366 | rv = -1; | |
5367 | goto fail; | |
5368 | } | |
5369 | ||
9534d671 | 5370 | resp_size = crypto_shash_digestsize(connection->cram_hmac_tfm); |
b411b363 PR |
5371 | response = kmalloc(resp_size, GFP_NOIO); |
5372 | if (response == NULL) { | |
1ec861eb | 5373 | drbd_err(connection, "kmalloc of response failed\n"); |
b10d96cb | 5374 | rv = -1; |
b411b363 PR |
5375 | goto fail; |
5376 | } | |
5377 | ||
9534d671 | 5378 | rv = crypto_shash_digest(desc, peers_ch, pi.size, response); |
b411b363 | 5379 | if (rv) { |
1ec861eb | 5380 | drbd_err(connection, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 5381 | rv = -1; |
b411b363 PR |
5382 | goto fail; |
5383 | } | |
5384 | ||
bde89a9e | 5385 | if (!conn_prepare_command(connection, sock)) { |
9f5bdc33 | 5386 | rv = 0; |
b411b363 | 5387 | goto fail; |
9f5bdc33 | 5388 | } |
bde89a9e | 5389 | rv = !conn_send_command(connection, sock, P_AUTH_RESPONSE, 0, |
9f5bdc33 | 5390 | response, resp_size); |
b411b363 PR |
5391 | if (!rv) |
5392 | goto fail; | |
5393 | ||
bde89a9e | 5394 | err = drbd_recv_header(connection, &pi); |
69bc7bc3 | 5395 | if (err) { |
b411b363 PR |
5396 | rv = 0; |
5397 | goto fail; | |
5398 | } | |
5399 | ||
77351055 | 5400 | if (pi.cmd != P_AUTH_RESPONSE) { |
1ec861eb | 5401 | drbd_err(connection, "expected AuthResponse packet, received: %s (0x%04x)\n", |
2fcb8f30 | 5402 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
5403 | rv = 0; |
5404 | goto fail; | |
5405 | } | |
5406 | ||
77351055 | 5407 | if (pi.size != resp_size) { |
1ec861eb | 5408 | drbd_err(connection, "expected AuthResponse payload of wrong size\n"); |
b411b363 PR |
5409 | rv = 0; |
5410 | goto fail; | |
5411 | } | |
b411b363 | 5412 | |
bde89a9e | 5413 | err = drbd_recv_all_warn(connection, response , resp_size); |
a5c31904 | 5414 | if (err) { |
b411b363 PR |
5415 | rv = 0; |
5416 | goto fail; | |
5417 | } | |
5418 | ||
5419 | right_response = kmalloc(resp_size, GFP_NOIO); | |
2d1ee87d | 5420 | if (right_response == NULL) { |
1ec861eb | 5421 | drbd_err(connection, "kmalloc of right_response failed\n"); |
b10d96cb | 5422 | rv = -1; |
b411b363 PR |
5423 | goto fail; |
5424 | } | |
5425 | ||
9534d671 HX |
5426 | rv = crypto_shash_digest(desc, my_challenge, CHALLENGE_LEN, |
5427 | right_response); | |
b411b363 | 5428 | if (rv) { |
1ec861eb | 5429 | drbd_err(connection, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 5430 | rv = -1; |
b411b363 PR |
5431 | goto fail; |
5432 | } | |
5433 | ||
5434 | rv = !memcmp(response, right_response, resp_size); | |
5435 | ||
5436 | if (rv) | |
1ec861eb | 5437 | drbd_info(connection, "Peer authenticated using %d bytes HMAC\n", |
44ed167d | 5438 | resp_size); |
b10d96cb JT |
5439 | else |
5440 | rv = -1; | |
b411b363 PR |
5441 | |
5442 | fail: | |
5443 | kfree(peers_ch); | |
5444 | kfree(response); | |
5445 | kfree(right_response); | |
9534d671 | 5446 | shash_desc_zero(desc); |
b411b363 PR |
5447 | |
5448 | return rv; | |
5449 | } | |
5450 | #endif | |
5451 | ||
8fe60551 | 5452 | int drbd_receiver(struct drbd_thread *thi) |
b411b363 | 5453 | { |
bde89a9e | 5454 | struct drbd_connection *connection = thi->connection; |
b411b363 PR |
5455 | int h; |
5456 | ||
1ec861eb | 5457 | drbd_info(connection, "receiver (re)started\n"); |
b411b363 PR |
5458 | |
5459 | do { | |
bde89a9e | 5460 | h = conn_connect(connection); |
b411b363 | 5461 | if (h == 0) { |
bde89a9e | 5462 | conn_disconnect(connection); |
20ee6390 | 5463 | schedule_timeout_interruptible(HZ); |
b411b363 PR |
5464 | } |
5465 | if (h == -1) { | |
1ec861eb | 5466 | drbd_warn(connection, "Discarding network configuration.\n"); |
bde89a9e | 5467 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
5468 | } |
5469 | } while (h == 0); | |
5470 | ||
c51a0ef3 LE |
5471 | if (h > 0) { |
5472 | blk_start_plug(&connection->receiver_plug); | |
bde89a9e | 5473 | drbdd(connection); |
c51a0ef3 LE |
5474 | blk_finish_plug(&connection->receiver_plug); |
5475 | } | |
b411b363 | 5476 | |
bde89a9e | 5477 | conn_disconnect(connection); |
b411b363 | 5478 | |
1ec861eb | 5479 | drbd_info(connection, "receiver terminated\n"); |
b411b363 PR |
5480 | return 0; |
5481 | } | |
5482 | ||
5483 | /* ********* acknowledge sender ******** */ | |
5484 | ||
bde89a9e | 5485 | static int got_conn_RqSReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5486 | { |
e658983a | 5487 | struct p_req_state_reply *p = pi->data; |
e4f78ede PR |
5488 | int retcode = be32_to_cpu(p->retcode); |
5489 | ||
5490 | if (retcode >= SS_SUCCESS) { | |
bde89a9e | 5491 | set_bit(CONN_WD_ST_CHG_OKAY, &connection->flags); |
e4f78ede | 5492 | } else { |
bde89a9e | 5493 | set_bit(CONN_WD_ST_CHG_FAIL, &connection->flags); |
1ec861eb | 5494 | drbd_err(connection, "Requested state change failed by peer: %s (%d)\n", |
e4f78ede PR |
5495 | drbd_set_st_err_str(retcode), retcode); |
5496 | } | |
bde89a9e | 5497 | wake_up(&connection->ping_wait); |
e4f78ede | 5498 | |
2735a594 | 5499 | return 0; |
e4f78ede | 5500 | } |
b411b363 | 5501 | |
bde89a9e | 5502 | static int got_RqSReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5503 | { |
9f4fe9ad | 5504 | struct drbd_peer_device *peer_device; |
b30ab791 | 5505 | struct drbd_device *device; |
e658983a | 5506 | struct p_req_state_reply *p = pi->data; |
b411b363 PR |
5507 | int retcode = be32_to_cpu(p->retcode); |
5508 | ||
9f4fe9ad AG |
5509 | peer_device = conn_peer_device(connection, pi->vnr); |
5510 | if (!peer_device) | |
2735a594 | 5511 | return -EIO; |
9f4fe9ad | 5512 | device = peer_device->device; |
1952e916 | 5513 | |
bde89a9e | 5514 | if (test_bit(CONN_WD_ST_CHG_REQ, &connection->flags)) { |
0b0ba1ef | 5515 | D_ASSERT(device, connection->agreed_pro_version < 100); |
bde89a9e | 5516 | return got_conn_RqSReply(connection, pi); |
4d0fc3fd PR |
5517 | } |
5518 | ||
b411b363 | 5519 | if (retcode >= SS_SUCCESS) { |
b30ab791 | 5520 | set_bit(CL_ST_CHG_SUCCESS, &device->flags); |
b411b363 | 5521 | } else { |
b30ab791 | 5522 | set_bit(CL_ST_CHG_FAIL, &device->flags); |
d0180171 | 5523 | drbd_err(device, "Requested state change failed by peer: %s (%d)\n", |
e4f78ede | 5524 | drbd_set_st_err_str(retcode), retcode); |
b411b363 | 5525 | } |
b30ab791 | 5526 | wake_up(&device->state_wait); |
b411b363 | 5527 | |
2735a594 | 5528 | return 0; |
b411b363 PR |
5529 | } |
5530 | ||
bde89a9e | 5531 | static int got_Ping(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5532 | { |
bde89a9e | 5533 | return drbd_send_ping_ack(connection); |
b411b363 PR |
5534 | |
5535 | } | |
5536 | ||
bde89a9e | 5537 | static int got_PingAck(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 PR |
5538 | { |
5539 | /* restore idle timeout */ | |
bde89a9e AG |
5540 | connection->meta.socket->sk->sk_rcvtimeo = connection->net_conf->ping_int*HZ; |
5541 | if (!test_and_set_bit(GOT_PING_ACK, &connection->flags)) | |
5542 | wake_up(&connection->ping_wait); | |
b411b363 | 5543 | |
2735a594 | 5544 | return 0; |
b411b363 PR |
5545 | } |
5546 | ||
bde89a9e | 5547 | static int got_IsInSync(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5548 | { |
9f4fe9ad | 5549 | struct drbd_peer_device *peer_device; |
b30ab791 | 5550 | struct drbd_device *device; |
e658983a | 5551 | struct p_block_ack *p = pi->data; |
b411b363 PR |
5552 | sector_t sector = be64_to_cpu(p->sector); |
5553 | int blksize = be32_to_cpu(p->blksize); | |
5554 | ||
9f4fe9ad AG |
5555 | peer_device = conn_peer_device(connection, pi->vnr); |
5556 | if (!peer_device) | |
2735a594 | 5557 | return -EIO; |
9f4fe9ad | 5558 | device = peer_device->device; |
1952e916 | 5559 | |
9f4fe9ad | 5560 | D_ASSERT(device, peer_device->connection->agreed_pro_version >= 89); |
b411b363 | 5561 | |
69a22773 | 5562 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 | 5563 | |
b30ab791 AG |
5564 | if (get_ldev(device)) { |
5565 | drbd_rs_complete_io(device, sector); | |
5566 | drbd_set_in_sync(device, sector, blksize); | |
1d53f09e | 5567 | /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ |
b30ab791 AG |
5568 | device->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); |
5569 | put_ldev(device); | |
1d53f09e | 5570 | } |
b30ab791 AG |
5571 | dec_rs_pending(device); |
5572 | atomic_add(blksize >> 9, &device->rs_sect_in); | |
b411b363 | 5573 | |
2735a594 | 5574 | return 0; |
b411b363 PR |
5575 | } |
5576 | ||
bc9c5c41 | 5577 | static int |
b30ab791 | 5578 | validate_req_change_req_state(struct drbd_device *device, u64 id, sector_t sector, |
bc9c5c41 AG |
5579 | struct rb_root *root, const char *func, |
5580 | enum drbd_req_event what, bool missing_ok) | |
b411b363 PR |
5581 | { |
5582 | struct drbd_request *req; | |
5583 | struct bio_and_error m; | |
5584 | ||
0500813f | 5585 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 5586 | req = find_request(device, root, id, sector, missing_ok, func); |
b411b363 | 5587 | if (unlikely(!req)) { |
0500813f | 5588 | spin_unlock_irq(&device->resource->req_lock); |
85997675 | 5589 | return -EIO; |
b411b363 PR |
5590 | } |
5591 | __req_mod(req, what, &m); | |
0500813f | 5592 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
5593 | |
5594 | if (m.bio) | |
b30ab791 | 5595 | complete_master_bio(device, &m); |
85997675 | 5596 | return 0; |
b411b363 PR |
5597 | } |
5598 | ||
bde89a9e | 5599 | static int got_BlockAck(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5600 | { |
9f4fe9ad | 5601 | struct drbd_peer_device *peer_device; |
b30ab791 | 5602 | struct drbd_device *device; |
e658983a | 5603 | struct p_block_ack *p = pi->data; |
b411b363 PR |
5604 | sector_t sector = be64_to_cpu(p->sector); |
5605 | int blksize = be32_to_cpu(p->blksize); | |
5606 | enum drbd_req_event what; | |
5607 | ||
9f4fe9ad AG |
5608 | peer_device = conn_peer_device(connection, pi->vnr); |
5609 | if (!peer_device) | |
2735a594 | 5610 | return -EIO; |
9f4fe9ad | 5611 | device = peer_device->device; |
1952e916 | 5612 | |
69a22773 | 5613 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 | 5614 | |
579b57ed | 5615 | if (p->block_id == ID_SYNCER) { |
b30ab791 AG |
5616 | drbd_set_in_sync(device, sector, blksize); |
5617 | dec_rs_pending(device); | |
2735a594 | 5618 | return 0; |
b411b363 | 5619 | } |
e05e1e59 | 5620 | switch (pi->cmd) { |
b411b363 | 5621 | case P_RS_WRITE_ACK: |
8554df1c | 5622 | what = WRITE_ACKED_BY_PEER_AND_SIS; |
b411b363 PR |
5623 | break; |
5624 | case P_WRITE_ACK: | |
8554df1c | 5625 | what = WRITE_ACKED_BY_PEER; |
b411b363 PR |
5626 | break; |
5627 | case P_RECV_ACK: | |
8554df1c | 5628 | what = RECV_ACKED_BY_PEER; |
b411b363 | 5629 | break; |
d4dabbe2 LE |
5630 | case P_SUPERSEDED: |
5631 | what = CONFLICT_RESOLVED; | |
b411b363 | 5632 | break; |
7be8da07 | 5633 | case P_RETRY_WRITE: |
7be8da07 | 5634 | what = POSTPONE_WRITE; |
b411b363 PR |
5635 | break; |
5636 | default: | |
2735a594 | 5637 | BUG(); |
b411b363 PR |
5638 | } |
5639 | ||
b30ab791 AG |
5640 | return validate_req_change_req_state(device, p->block_id, sector, |
5641 | &device->write_requests, __func__, | |
2735a594 | 5642 | what, false); |
b411b363 PR |
5643 | } |
5644 | ||
bde89a9e | 5645 | static int got_NegAck(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5646 | { |
9f4fe9ad | 5647 | struct drbd_peer_device *peer_device; |
b30ab791 | 5648 | struct drbd_device *device; |
e658983a | 5649 | struct p_block_ack *p = pi->data; |
b411b363 | 5650 | sector_t sector = be64_to_cpu(p->sector); |
2deb8336 | 5651 | int size = be32_to_cpu(p->blksize); |
85997675 | 5652 | int err; |
b411b363 | 5653 | |
9f4fe9ad AG |
5654 | peer_device = conn_peer_device(connection, pi->vnr); |
5655 | if (!peer_device) | |
2735a594 | 5656 | return -EIO; |
9f4fe9ad | 5657 | device = peer_device->device; |
b411b363 | 5658 | |
69a22773 | 5659 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 | 5660 | |
579b57ed | 5661 | if (p->block_id == ID_SYNCER) { |
b30ab791 AG |
5662 | dec_rs_pending(device); |
5663 | drbd_rs_failed_io(device, sector, size); | |
2735a594 | 5664 | return 0; |
b411b363 | 5665 | } |
2deb8336 | 5666 | |
b30ab791 AG |
5667 | err = validate_req_change_req_state(device, p->block_id, sector, |
5668 | &device->write_requests, __func__, | |
303d1448 | 5669 | NEG_ACKED, true); |
85997675 | 5670 | if (err) { |
c3afd8f5 AG |
5671 | /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs. |
5672 | The master bio might already be completed, therefore the | |
5673 | request is no longer in the collision hash. */ | |
5674 | /* In Protocol B we might already have got a P_RECV_ACK | |
5675 | but then get a P_NEG_ACK afterwards. */ | |
b30ab791 | 5676 | drbd_set_out_of_sync(device, sector, size); |
2deb8336 | 5677 | } |
2735a594 | 5678 | return 0; |
b411b363 PR |
5679 | } |
5680 | ||
bde89a9e | 5681 | static int got_NegDReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5682 | { |
9f4fe9ad | 5683 | struct drbd_peer_device *peer_device; |
b30ab791 | 5684 | struct drbd_device *device; |
e658983a | 5685 | struct p_block_ack *p = pi->data; |
b411b363 PR |
5686 | sector_t sector = be64_to_cpu(p->sector); |
5687 | ||
9f4fe9ad AG |
5688 | peer_device = conn_peer_device(connection, pi->vnr); |
5689 | if (!peer_device) | |
2735a594 | 5690 | return -EIO; |
9f4fe9ad | 5691 | device = peer_device->device; |
1952e916 | 5692 | |
69a22773 | 5693 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
7be8da07 | 5694 | |
d0180171 | 5695 | drbd_err(device, "Got NegDReply; Sector %llus, len %u.\n", |
b411b363 PR |
5696 | (unsigned long long)sector, be32_to_cpu(p->blksize)); |
5697 | ||
b30ab791 AG |
5698 | return validate_req_change_req_state(device, p->block_id, sector, |
5699 | &device->read_requests, __func__, | |
2735a594 | 5700 | NEG_ACKED, false); |
b411b363 PR |
5701 | } |
5702 | ||
bde89a9e | 5703 | static int got_NegRSDReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5704 | { |
9f4fe9ad | 5705 | struct drbd_peer_device *peer_device; |
b30ab791 | 5706 | struct drbd_device *device; |
b411b363 PR |
5707 | sector_t sector; |
5708 | int size; | |
e658983a | 5709 | struct p_block_ack *p = pi->data; |
1952e916 | 5710 | |
9f4fe9ad AG |
5711 | peer_device = conn_peer_device(connection, pi->vnr); |
5712 | if (!peer_device) | |
2735a594 | 5713 | return -EIO; |
9f4fe9ad | 5714 | device = peer_device->device; |
b411b363 PR |
5715 | |
5716 | sector = be64_to_cpu(p->sector); | |
5717 | size = be32_to_cpu(p->blksize); | |
b411b363 | 5718 | |
69a22773 | 5719 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 | 5720 | |
b30ab791 | 5721 | dec_rs_pending(device); |
b411b363 | 5722 | |
b30ab791 AG |
5723 | if (get_ldev_if_state(device, D_FAILED)) { |
5724 | drbd_rs_complete_io(device, sector); | |
e05e1e59 | 5725 | switch (pi->cmd) { |
d612d309 | 5726 | case P_NEG_RS_DREPLY: |
b30ab791 | 5727 | drbd_rs_failed_io(device, sector, size); |
d612d309 PR |
5728 | case P_RS_CANCEL: |
5729 | break; | |
5730 | default: | |
2735a594 | 5731 | BUG(); |
d612d309 | 5732 | } |
b30ab791 | 5733 | put_ldev(device); |
b411b363 PR |
5734 | } |
5735 | ||
2735a594 | 5736 | return 0; |
b411b363 PR |
5737 | } |
5738 | ||
bde89a9e | 5739 | static int got_BarrierAck(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5740 | { |
e658983a | 5741 | struct p_barrier_ack *p = pi->data; |
c06ece6b | 5742 | struct drbd_peer_device *peer_device; |
9ed57dcb | 5743 | int vnr; |
1952e916 | 5744 | |
bde89a9e | 5745 | tl_release(connection, p->barrier, be32_to_cpu(p->set_size)); |
b411b363 | 5746 | |
9ed57dcb | 5747 | rcu_read_lock(); |
c06ece6b AG |
5748 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
5749 | struct drbd_device *device = peer_device->device; | |
5750 | ||
b30ab791 AG |
5751 | if (device->state.conn == C_AHEAD && |
5752 | atomic_read(&device->ap_in_flight) == 0 && | |
5753 | !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &device->flags)) { | |
5754 | device->start_resync_timer.expires = jiffies + HZ; | |
5755 | add_timer(&device->start_resync_timer); | |
9ed57dcb | 5756 | } |
c4752ef1 | 5757 | } |
9ed57dcb | 5758 | rcu_read_unlock(); |
c4752ef1 | 5759 | |
2735a594 | 5760 | return 0; |
b411b363 PR |
5761 | } |
5762 | ||
bde89a9e | 5763 | static int got_OVResult(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5764 | { |
9f4fe9ad | 5765 | struct drbd_peer_device *peer_device; |
b30ab791 | 5766 | struct drbd_device *device; |
e658983a | 5767 | struct p_block_ack *p = pi->data; |
84b8c06b | 5768 | struct drbd_device_work *dw; |
b411b363 PR |
5769 | sector_t sector; |
5770 | int size; | |
5771 | ||
9f4fe9ad AG |
5772 | peer_device = conn_peer_device(connection, pi->vnr); |
5773 | if (!peer_device) | |
2735a594 | 5774 | return -EIO; |
9f4fe9ad | 5775 | device = peer_device->device; |
1952e916 | 5776 | |
b411b363 PR |
5777 | sector = be64_to_cpu(p->sector); |
5778 | size = be32_to_cpu(p->blksize); | |
5779 | ||
69a22773 | 5780 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 PR |
5781 | |
5782 | if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) | |
b30ab791 | 5783 | drbd_ov_out_of_sync_found(device, sector, size); |
b411b363 | 5784 | else |
b30ab791 | 5785 | ov_out_of_sync_print(device); |
b411b363 | 5786 | |
b30ab791 | 5787 | if (!get_ldev(device)) |
2735a594 | 5788 | return 0; |
1d53f09e | 5789 | |
b30ab791 AG |
5790 | drbd_rs_complete_io(device, sector); |
5791 | dec_rs_pending(device); | |
b411b363 | 5792 | |
b30ab791 | 5793 | --device->ov_left; |
ea5442af LE |
5794 | |
5795 | /* let's advance progress step marks only for every other megabyte */ | |
b30ab791 AG |
5796 | if ((device->ov_left & 0x200) == 0x200) |
5797 | drbd_advance_rs_marks(device, device->ov_left); | |
ea5442af | 5798 | |
b30ab791 | 5799 | if (device->ov_left == 0) { |
84b8c06b AG |
5800 | dw = kmalloc(sizeof(*dw), GFP_NOIO); |
5801 | if (dw) { | |
5802 | dw->w.cb = w_ov_finished; | |
5803 | dw->device = device; | |
5804 | drbd_queue_work(&peer_device->connection->sender_work, &dw->w); | |
b411b363 | 5805 | } else { |
84b8c06b | 5806 | drbd_err(device, "kmalloc(dw) failed."); |
b30ab791 AG |
5807 | ov_out_of_sync_print(device); |
5808 | drbd_resync_finished(device); | |
b411b363 PR |
5809 | } |
5810 | } | |
b30ab791 | 5811 | put_ldev(device); |
2735a594 | 5812 | return 0; |
b411b363 PR |
5813 | } |
5814 | ||
bde89a9e | 5815 | static int got_skip(struct drbd_connection *connection, struct packet_info *pi) |
0ced55a3 | 5816 | { |
2735a594 | 5817 | return 0; |
b411b363 PR |
5818 | } |
5819 | ||
668700b4 PR |
5820 | struct meta_sock_cmd { |
5821 | size_t pkt_size; | |
5822 | int (*fn)(struct drbd_connection *connection, struct packet_info *); | |
5823 | }; | |
5824 | ||
5825 | static void set_rcvtimeo(struct drbd_connection *connection, bool ping_timeout) | |
0ced55a3 | 5826 | { |
668700b4 PR |
5827 | long t; |
5828 | struct net_conf *nc; | |
32862ec7 | 5829 | |
668700b4 PR |
5830 | rcu_read_lock(); |
5831 | nc = rcu_dereference(connection->net_conf); | |
5832 | t = ping_timeout ? nc->ping_timeo : nc->ping_int; | |
5833 | rcu_read_unlock(); | |
c141ebda | 5834 | |
668700b4 PR |
5835 | t *= HZ; |
5836 | if (ping_timeout) | |
5837 | t /= 10; | |
082a3439 | 5838 | |
668700b4 PR |
5839 | connection->meta.socket->sk->sk_rcvtimeo = t; |
5840 | } | |
32862ec7 | 5841 | |
668700b4 PR |
5842 | static void set_ping_timeout(struct drbd_connection *connection) |
5843 | { | |
5844 | set_rcvtimeo(connection, 1); | |
0ced55a3 PR |
5845 | } |
5846 | ||
668700b4 PR |
5847 | static void set_idle_timeout(struct drbd_connection *connection) |
5848 | { | |
5849 | set_rcvtimeo(connection, 0); | |
5850 | } | |
b411b363 | 5851 | |
668700b4 | 5852 | static struct meta_sock_cmd ack_receiver_tbl[] = { |
e658983a AG |
5853 | [P_PING] = { 0, got_Ping }, |
5854 | [P_PING_ACK] = { 0, got_PingAck }, | |
b411b363 PR |
5855 | [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, |
5856 | [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
5857 | [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
d4dabbe2 | 5858 | [P_SUPERSEDED] = { sizeof(struct p_block_ack), got_BlockAck }, |
b411b363 PR |
5859 | [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, |
5860 | [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, | |
1952e916 | 5861 | [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply }, |
b411b363 PR |
5862 | [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, |
5863 | [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, | |
5864 | [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, | |
5865 | [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, | |
02918be2 | 5866 | [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, |
1952e916 AG |
5867 | [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply }, |
5868 | [P_CONN_ST_CHG_REPLY]={ sizeof(struct p_req_state_reply), got_conn_RqSReply }, | |
5869 | [P_RETRY_WRITE] = { sizeof(struct p_block_ack), got_BlockAck }, | |
7201b972 | 5870 | }; |
b411b363 | 5871 | |
1c03e520 | 5872 | int drbd_ack_receiver(struct drbd_thread *thi) |
b411b363 | 5873 | { |
bde89a9e | 5874 | struct drbd_connection *connection = thi->connection; |
668700b4 | 5875 | struct meta_sock_cmd *cmd = NULL; |
77351055 | 5876 | struct packet_info pi; |
668700b4 | 5877 | unsigned long pre_recv_jif; |
257d0af6 | 5878 | int rv; |
bde89a9e | 5879 | void *buf = connection->meta.rbuf; |
b411b363 | 5880 | int received = 0; |
bde89a9e | 5881 | unsigned int header_size = drbd_header_size(connection); |
52b061a4 | 5882 | int expect = header_size; |
44ed167d | 5883 | bool ping_timeout_active = false; |
3990e04d | 5884 | struct sched_param param = { .sched_priority = 2 }; |
b411b363 | 5885 | |
3990e04d PR |
5886 | rv = sched_setscheduler(current, SCHED_RR, ¶m); |
5887 | if (rv < 0) | |
668700b4 | 5888 | drbd_err(connection, "drbd_ack_receiver: ERROR set priority, ret=%d\n", rv); |
b411b363 | 5889 | |
e77a0a5c | 5890 | while (get_t_state(thi) == RUNNING) { |
80822284 | 5891 | drbd_thread_current_set_cpu(thi); |
b411b363 | 5892 | |
668700b4 | 5893 | conn_reclaim_net_peer_reqs(connection); |
44ed167d | 5894 | |
bde89a9e AG |
5895 | if (test_and_clear_bit(SEND_PING, &connection->flags)) { |
5896 | if (drbd_send_ping(connection)) { | |
1ec861eb | 5897 | drbd_err(connection, "drbd_send_ping has failed\n"); |
b411b363 | 5898 | goto reconnect; |
841ce241 | 5899 | } |
668700b4 | 5900 | set_ping_timeout(connection); |
44ed167d | 5901 | ping_timeout_active = true; |
b411b363 PR |
5902 | } |
5903 | ||
668700b4 | 5904 | pre_recv_jif = jiffies; |
bde89a9e | 5905 | rv = drbd_recv_short(connection->meta.socket, buf, expect-received, 0); |
b411b363 PR |
5906 | |
5907 | /* Note: | |
5908 | * -EINTR (on meta) we got a signal | |
5909 | * -EAGAIN (on meta) rcvtimeo expired | |
5910 | * -ECONNRESET other side closed the connection | |
5911 | * -ERESTARTSYS (on data) we got a signal | |
5912 | * rv < 0 other than above: unexpected error! | |
5913 | * rv == expected: full header or command | |
5914 | * rv < expected: "woken" by signal during receive | |
5915 | * rv == 0 : "connection shut down by peer" | |
5916 | */ | |
5917 | if (likely(rv > 0)) { | |
5918 | received += rv; | |
5919 | buf += rv; | |
5920 | } else if (rv == 0) { | |
bde89a9e | 5921 | if (test_bit(DISCONNECT_SENT, &connection->flags)) { |
b66623e3 PR |
5922 | long t; |
5923 | rcu_read_lock(); | |
bde89a9e | 5924 | t = rcu_dereference(connection->net_conf)->ping_timeo * HZ/10; |
b66623e3 PR |
5925 | rcu_read_unlock(); |
5926 | ||
bde89a9e AG |
5927 | t = wait_event_timeout(connection->ping_wait, |
5928 | connection->cstate < C_WF_REPORT_PARAMS, | |
b66623e3 | 5929 | t); |
599377ac PR |
5930 | if (t) |
5931 | break; | |
5932 | } | |
1ec861eb | 5933 | drbd_err(connection, "meta connection shut down by peer.\n"); |
b411b363 PR |
5934 | goto reconnect; |
5935 | } else if (rv == -EAGAIN) { | |
cb6518cb LE |
5936 | /* If the data socket received something meanwhile, |
5937 | * that is good enough: peer is still alive. */ | |
668700b4 | 5938 | if (time_after(connection->last_received, pre_recv_jif)) |
cb6518cb | 5939 | continue; |
f36af18c | 5940 | if (ping_timeout_active) { |
1ec861eb | 5941 | drbd_err(connection, "PingAck did not arrive in time.\n"); |
b411b363 PR |
5942 | goto reconnect; |
5943 | } | |
bde89a9e | 5944 | set_bit(SEND_PING, &connection->flags); |
b411b363 PR |
5945 | continue; |
5946 | } else if (rv == -EINTR) { | |
668700b4 PR |
5947 | /* maybe drbd_thread_stop(): the while condition will notice. |
5948 | * maybe woken for send_ping: we'll send a ping above, | |
5949 | * and change the rcvtimeo */ | |
5950 | flush_signals(current); | |
b411b363 PR |
5951 | continue; |
5952 | } else { | |
1ec861eb | 5953 | drbd_err(connection, "sock_recvmsg returned %d\n", rv); |
b411b363 PR |
5954 | goto reconnect; |
5955 | } | |
5956 | ||
5957 | if (received == expect && cmd == NULL) { | |
bde89a9e | 5958 | if (decode_header(connection, connection->meta.rbuf, &pi)) |
b411b363 | 5959 | goto reconnect; |
668700b4 PR |
5960 | cmd = &ack_receiver_tbl[pi.cmd]; |
5961 | if (pi.cmd >= ARRAY_SIZE(ack_receiver_tbl) || !cmd->fn) { | |
1ec861eb | 5962 | drbd_err(connection, "Unexpected meta packet %s (0x%04x)\n", |
2fcb8f30 | 5963 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
5964 | goto disconnect; |
5965 | } | |
e658983a | 5966 | expect = header_size + cmd->pkt_size; |
52b061a4 | 5967 | if (pi.size != expect - header_size) { |
1ec861eb | 5968 | drbd_err(connection, "Wrong packet size on meta (c: %d, l: %d)\n", |
77351055 | 5969 | pi.cmd, pi.size); |
b411b363 | 5970 | goto reconnect; |
257d0af6 | 5971 | } |
b411b363 PR |
5972 | } |
5973 | if (received == expect) { | |
2735a594 | 5974 | bool err; |
a4fbda8e | 5975 | |
bde89a9e | 5976 | err = cmd->fn(connection, &pi); |
2735a594 | 5977 | if (err) { |
1ec861eb | 5978 | drbd_err(connection, "%pf failed\n", cmd->fn); |
b411b363 | 5979 | goto reconnect; |
1952e916 | 5980 | } |
b411b363 | 5981 | |
bde89a9e | 5982 | connection->last_received = jiffies; |
f36af18c | 5983 | |
668700b4 PR |
5984 | if (cmd == &ack_receiver_tbl[P_PING_ACK]) { |
5985 | set_idle_timeout(connection); | |
44ed167d PR |
5986 | ping_timeout_active = false; |
5987 | } | |
f36af18c | 5988 | |
bde89a9e | 5989 | buf = connection->meta.rbuf; |
b411b363 | 5990 | received = 0; |
52b061a4 | 5991 | expect = header_size; |
b411b363 PR |
5992 | cmd = NULL; |
5993 | } | |
5994 | } | |
5995 | ||
5996 | if (0) { | |
5997 | reconnect: | |
bde89a9e AG |
5998 | conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); |
5999 | conn_md_sync(connection); | |
b411b363 PR |
6000 | } |
6001 | if (0) { | |
6002 | disconnect: | |
bde89a9e | 6003 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 6004 | } |
b411b363 | 6005 | |
668700b4 | 6006 | drbd_info(connection, "ack_receiver terminated\n"); |
b411b363 PR |
6007 | |
6008 | return 0; | |
6009 | } | |
668700b4 PR |
6010 | |
6011 | void drbd_send_acks_wf(struct work_struct *ws) | |
6012 | { | |
6013 | struct drbd_peer_device *peer_device = | |
6014 | container_of(ws, struct drbd_peer_device, send_acks_work); | |
6015 | struct drbd_connection *connection = peer_device->connection; | |
6016 | struct drbd_device *device = peer_device->device; | |
6017 | struct net_conf *nc; | |
6018 | int tcp_cork, err; | |
6019 | ||
6020 | rcu_read_lock(); | |
6021 | nc = rcu_dereference(connection->net_conf); | |
6022 | tcp_cork = nc->tcp_cork; | |
6023 | rcu_read_unlock(); | |
6024 | ||
6025 | if (tcp_cork) | |
6026 | drbd_tcp_cork(connection->meta.socket); | |
6027 | ||
6028 | err = drbd_finish_peer_reqs(device); | |
6029 | kref_put(&device->kref, drbd_destroy_device); | |
6030 | /* get is in drbd_endio_write_sec_final(). That is necessary to keep the | |
6031 | struct work_struct send_acks_work alive, which is in the peer_device object */ | |
6032 | ||
6033 | if (err) { | |
6034 | conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); | |
6035 | return; | |
6036 | } | |
6037 | ||
6038 | if (tcp_cork) | |
6039 | drbd_tcp_uncork(connection->meta.socket); | |
6040 | ||
6041 | return; | |
6042 | } |