Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_receiver.c | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | drbd is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation; either version 2, or (at your option) | |
13 | any later version. | |
14 | ||
15 | drbd is distributed in the hope that it will be useful, | |
16 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | GNU General Public License for more details. | |
19 | ||
20 | You should have received a copy of the GNU General Public License | |
21 | along with drbd; see the file COPYING. If not, write to | |
22 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | */ | |
24 | ||
25 | ||
b411b363 PR |
26 | #include <linux/module.h> |
27 | ||
28 | #include <asm/uaccess.h> | |
29 | #include <net/sock.h> | |
30 | ||
b411b363 PR |
31 | #include <linux/drbd.h> |
32 | #include <linux/fs.h> | |
33 | #include <linux/file.h> | |
34 | #include <linux/in.h> | |
35 | #include <linux/mm.h> | |
36 | #include <linux/memcontrol.h> | |
37 | #include <linux/mm_inline.h> | |
38 | #include <linux/slab.h> | |
b411b363 PR |
39 | #include <linux/pkt_sched.h> |
40 | #define __KERNEL_SYSCALLS__ | |
41 | #include <linux/unistd.h> | |
42 | #include <linux/vmalloc.h> | |
43 | #include <linux/random.h> | |
b411b363 PR |
44 | #include <linux/string.h> |
45 | #include <linux/scatterlist.h> | |
46 | #include "drbd_int.h" | |
a3603a6e | 47 | #include "drbd_protocol.h" |
b411b363 | 48 | #include "drbd_req.h" |
b411b363 PR |
49 | #include "drbd_vli.h" |
50 | ||
20c68fde LE |
51 | #define PRO_FEATURES (FF_TRIM) |
52 | ||
77351055 PR |
53 | struct packet_info { |
54 | enum drbd_packet cmd; | |
e2857216 AG |
55 | unsigned int size; |
56 | unsigned int vnr; | |
e658983a | 57 | void *data; |
77351055 PR |
58 | }; |
59 | ||
b411b363 PR |
60 | enum finish_epoch { |
61 | FE_STILL_LIVE, | |
62 | FE_DESTROYED, | |
63 | FE_RECYCLED, | |
64 | }; | |
65 | ||
bde89a9e AG |
66 | static int drbd_do_features(struct drbd_connection *connection); |
67 | static int drbd_do_auth(struct drbd_connection *connection); | |
69a22773 | 68 | static int drbd_disconnected(struct drbd_peer_device *); |
a0fb3c47 | 69 | static void conn_wait_active_ee_empty(struct drbd_connection *connection); |
bde89a9e | 70 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_connection *, struct drbd_epoch *, enum epoch_event); |
99920dc5 | 71 | static int e_end_block(struct drbd_work *, int); |
b411b363 | 72 | |
b411b363 PR |
73 | |
74 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
75 | ||
45bb912b LE |
76 | /* |
77 | * some helper functions to deal with single linked page lists, | |
78 | * page->private being our "next" pointer. | |
79 | */ | |
80 | ||
81 | /* If at least n pages are linked at head, get n pages off. | |
82 | * Otherwise, don't modify head, and return NULL. | |
83 | * Locking is the responsibility of the caller. | |
84 | */ | |
85 | static struct page *page_chain_del(struct page **head, int n) | |
86 | { | |
87 | struct page *page; | |
88 | struct page *tmp; | |
89 | ||
90 | BUG_ON(!n); | |
91 | BUG_ON(!head); | |
92 | ||
93 | page = *head; | |
23ce4227 PR |
94 | |
95 | if (!page) | |
96 | return NULL; | |
97 | ||
45bb912b LE |
98 | while (page) { |
99 | tmp = page_chain_next(page); | |
100 | if (--n == 0) | |
101 | break; /* found sufficient pages */ | |
102 | if (tmp == NULL) | |
103 | /* insufficient pages, don't use any of them. */ | |
104 | return NULL; | |
105 | page = tmp; | |
106 | } | |
107 | ||
108 | /* add end of list marker for the returned list */ | |
109 | set_page_private(page, 0); | |
110 | /* actual return value, and adjustment of head */ | |
111 | page = *head; | |
112 | *head = tmp; | |
113 | return page; | |
114 | } | |
115 | ||
116 | /* may be used outside of locks to find the tail of a (usually short) | |
117 | * "private" page chain, before adding it back to a global chain head | |
118 | * with page_chain_add() under a spinlock. */ | |
119 | static struct page *page_chain_tail(struct page *page, int *len) | |
120 | { | |
121 | struct page *tmp; | |
122 | int i = 1; | |
123 | while ((tmp = page_chain_next(page))) | |
124 | ++i, page = tmp; | |
125 | if (len) | |
126 | *len = i; | |
127 | return page; | |
128 | } | |
129 | ||
130 | static int page_chain_free(struct page *page) | |
131 | { | |
132 | struct page *tmp; | |
133 | int i = 0; | |
134 | page_chain_for_each_safe(page, tmp) { | |
135 | put_page(page); | |
136 | ++i; | |
137 | } | |
138 | return i; | |
139 | } | |
140 | ||
141 | static void page_chain_add(struct page **head, | |
142 | struct page *chain_first, struct page *chain_last) | |
143 | { | |
144 | #if 1 | |
145 | struct page *tmp; | |
146 | tmp = page_chain_tail(chain_first, NULL); | |
147 | BUG_ON(tmp != chain_last); | |
148 | #endif | |
149 | ||
150 | /* add chain to head */ | |
151 | set_page_private(chain_last, (unsigned long)*head); | |
152 | *head = chain_first; | |
153 | } | |
154 | ||
b30ab791 | 155 | static struct page *__drbd_alloc_pages(struct drbd_device *device, |
18c2d522 | 156 | unsigned int number) |
b411b363 PR |
157 | { |
158 | struct page *page = NULL; | |
45bb912b | 159 | struct page *tmp = NULL; |
18c2d522 | 160 | unsigned int i = 0; |
b411b363 PR |
161 | |
162 | /* Yes, testing drbd_pp_vacant outside the lock is racy. | |
163 | * So what. It saves a spin_lock. */ | |
45bb912b | 164 | if (drbd_pp_vacant >= number) { |
b411b363 | 165 | spin_lock(&drbd_pp_lock); |
45bb912b LE |
166 | page = page_chain_del(&drbd_pp_pool, number); |
167 | if (page) | |
168 | drbd_pp_vacant -= number; | |
b411b363 | 169 | spin_unlock(&drbd_pp_lock); |
45bb912b LE |
170 | if (page) |
171 | return page; | |
b411b363 | 172 | } |
45bb912b | 173 | |
b411b363 PR |
174 | /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD |
175 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
176 | * which in turn might block on the other node at this very place. */ | |
45bb912b LE |
177 | for (i = 0; i < number; i++) { |
178 | tmp = alloc_page(GFP_TRY); | |
179 | if (!tmp) | |
180 | break; | |
181 | set_page_private(tmp, (unsigned long)page); | |
182 | page = tmp; | |
183 | } | |
184 | ||
185 | if (i == number) | |
186 | return page; | |
187 | ||
188 | /* Not enough pages immediately available this time. | |
c37c8ecf | 189 | * No need to jump around here, drbd_alloc_pages will retry this |
45bb912b LE |
190 | * function "soon". */ |
191 | if (page) { | |
192 | tmp = page_chain_tail(page, NULL); | |
193 | spin_lock(&drbd_pp_lock); | |
194 | page_chain_add(&drbd_pp_pool, page, tmp); | |
195 | drbd_pp_vacant += i; | |
196 | spin_unlock(&drbd_pp_lock); | |
197 | } | |
198 | return NULL; | |
b411b363 PR |
199 | } |
200 | ||
b30ab791 | 201 | static void reclaim_finished_net_peer_reqs(struct drbd_device *device, |
a990be46 | 202 | struct list_head *to_be_freed) |
b411b363 | 203 | { |
a8cd15ba | 204 | struct drbd_peer_request *peer_req, *tmp; |
b411b363 PR |
205 | |
206 | /* The EEs are always appended to the end of the list. Since | |
207 | they are sent in order over the wire, they have to finish | |
208 | in order. As soon as we see the first not finished we can | |
209 | stop to examine the list... */ | |
210 | ||
a8cd15ba | 211 | list_for_each_entry_safe(peer_req, tmp, &device->net_ee, w.list) { |
045417f7 | 212 | if (drbd_peer_req_has_active_page(peer_req)) |
b411b363 | 213 | break; |
a8cd15ba | 214 | list_move(&peer_req->w.list, to_be_freed); |
b411b363 PR |
215 | } |
216 | } | |
217 | ||
b30ab791 | 218 | static void drbd_kick_lo_and_reclaim_net(struct drbd_device *device) |
b411b363 PR |
219 | { |
220 | LIST_HEAD(reclaimed); | |
db830c46 | 221 | struct drbd_peer_request *peer_req, *t; |
b411b363 | 222 | |
0500813f | 223 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 224 | reclaim_finished_net_peer_reqs(device, &reclaimed); |
0500813f | 225 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 226 | |
a8cd15ba | 227 | list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) |
b30ab791 | 228 | drbd_free_net_peer_req(device, peer_req); |
b411b363 PR |
229 | } |
230 | ||
231 | /** | |
c37c8ecf | 232 | * drbd_alloc_pages() - Returns @number pages, retries forever (or until signalled) |
b30ab791 | 233 | * @device: DRBD device. |
45bb912b LE |
234 | * @number: number of pages requested |
235 | * @retry: whether to retry, if not enough pages are available right now | |
236 | * | |
237 | * Tries to allocate number pages, first from our own page pool, then from | |
0e49d7b0 | 238 | * the kernel. |
45bb912b | 239 | * Possibly retry until DRBD frees sufficient pages somewhere else. |
b411b363 | 240 | * |
0e49d7b0 LE |
241 | * If this allocation would exceed the max_buffers setting, we throttle |
242 | * allocation (schedule_timeout) to give the system some room to breathe. | |
243 | * | |
244 | * We do not use max-buffers as hard limit, because it could lead to | |
245 | * congestion and further to a distributed deadlock during online-verify or | |
246 | * (checksum based) resync, if the max-buffers, socket buffer sizes and | |
247 | * resync-rate settings are mis-configured. | |
248 | * | |
45bb912b | 249 | * Returns a page chain linked via page->private. |
b411b363 | 250 | */ |
69a22773 | 251 | struct page *drbd_alloc_pages(struct drbd_peer_device *peer_device, unsigned int number, |
c37c8ecf | 252 | bool retry) |
b411b363 | 253 | { |
69a22773 | 254 | struct drbd_device *device = peer_device->device; |
b411b363 | 255 | struct page *page = NULL; |
44ed167d | 256 | struct net_conf *nc; |
b411b363 | 257 | DEFINE_WAIT(wait); |
0e49d7b0 | 258 | unsigned int mxb; |
b411b363 | 259 | |
44ed167d | 260 | rcu_read_lock(); |
69a22773 | 261 | nc = rcu_dereference(peer_device->connection->net_conf); |
44ed167d PR |
262 | mxb = nc ? nc->max_buffers : 1000000; |
263 | rcu_read_unlock(); | |
264 | ||
b30ab791 AG |
265 | if (atomic_read(&device->pp_in_use) < mxb) |
266 | page = __drbd_alloc_pages(device, number); | |
b411b363 | 267 | |
45bb912b | 268 | while (page == NULL) { |
b411b363 PR |
269 | prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE); |
270 | ||
b30ab791 | 271 | drbd_kick_lo_and_reclaim_net(device); |
b411b363 | 272 | |
b30ab791 AG |
273 | if (atomic_read(&device->pp_in_use) < mxb) { |
274 | page = __drbd_alloc_pages(device, number); | |
b411b363 PR |
275 | if (page) |
276 | break; | |
277 | } | |
278 | ||
279 | if (!retry) | |
280 | break; | |
281 | ||
282 | if (signal_pending(current)) { | |
d0180171 | 283 | drbd_warn(device, "drbd_alloc_pages interrupted!\n"); |
b411b363 PR |
284 | break; |
285 | } | |
286 | ||
0e49d7b0 LE |
287 | if (schedule_timeout(HZ/10) == 0) |
288 | mxb = UINT_MAX; | |
b411b363 PR |
289 | } |
290 | finish_wait(&drbd_pp_wait, &wait); | |
291 | ||
45bb912b | 292 | if (page) |
b30ab791 | 293 | atomic_add(number, &device->pp_in_use); |
b411b363 PR |
294 | return page; |
295 | } | |
296 | ||
c37c8ecf | 297 | /* Must not be used from irq, as that may deadlock: see drbd_alloc_pages. |
0500813f | 298 | * Is also used from inside an other spin_lock_irq(&resource->req_lock); |
45bb912b LE |
299 | * Either links the page chain back to the global pool, |
300 | * or returns all pages to the system. */ | |
b30ab791 | 301 | static void drbd_free_pages(struct drbd_device *device, struct page *page, int is_net) |
b411b363 | 302 | { |
b30ab791 | 303 | atomic_t *a = is_net ? &device->pp_in_use_by_net : &device->pp_in_use; |
b411b363 | 304 | int i; |
435f0740 | 305 | |
a73ff323 LE |
306 | if (page == NULL) |
307 | return; | |
308 | ||
81a5d60e | 309 | if (drbd_pp_vacant > (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count) |
45bb912b LE |
310 | i = page_chain_free(page); |
311 | else { | |
312 | struct page *tmp; | |
313 | tmp = page_chain_tail(page, &i); | |
314 | spin_lock(&drbd_pp_lock); | |
315 | page_chain_add(&drbd_pp_pool, page, tmp); | |
316 | drbd_pp_vacant += i; | |
317 | spin_unlock(&drbd_pp_lock); | |
b411b363 | 318 | } |
435f0740 | 319 | i = atomic_sub_return(i, a); |
45bb912b | 320 | if (i < 0) |
d0180171 | 321 | drbd_warn(device, "ASSERTION FAILED: %s: %d < 0\n", |
435f0740 | 322 | is_net ? "pp_in_use_by_net" : "pp_in_use", i); |
b411b363 PR |
323 | wake_up(&drbd_pp_wait); |
324 | } | |
325 | ||
326 | /* | |
327 | You need to hold the req_lock: | |
328 | _drbd_wait_ee_list_empty() | |
329 | ||
330 | You must not have the req_lock: | |
3967deb1 | 331 | drbd_free_peer_req() |
0db55363 | 332 | drbd_alloc_peer_req() |
7721f567 | 333 | drbd_free_peer_reqs() |
b411b363 | 334 | drbd_ee_fix_bhs() |
a990be46 | 335 | drbd_finish_peer_reqs() |
b411b363 PR |
336 | drbd_clear_done_ee() |
337 | drbd_wait_ee_list_empty() | |
338 | */ | |
339 | ||
f6ffca9f | 340 | struct drbd_peer_request * |
69a22773 | 341 | drbd_alloc_peer_req(struct drbd_peer_device *peer_device, u64 id, sector_t sector, |
a0fb3c47 | 342 | unsigned int data_size, bool has_payload, gfp_t gfp_mask) __must_hold(local) |
b411b363 | 343 | { |
69a22773 | 344 | struct drbd_device *device = peer_device->device; |
db830c46 | 345 | struct drbd_peer_request *peer_req; |
a73ff323 | 346 | struct page *page = NULL; |
45bb912b | 347 | unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b411b363 | 348 | |
b30ab791 | 349 | if (drbd_insert_fault(device, DRBD_FAULT_AL_EE)) |
b411b363 PR |
350 | return NULL; |
351 | ||
db830c46 AG |
352 | peer_req = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM); |
353 | if (!peer_req) { | |
b411b363 | 354 | if (!(gfp_mask & __GFP_NOWARN)) |
d0180171 | 355 | drbd_err(device, "%s: allocation failed\n", __func__); |
b411b363 PR |
356 | return NULL; |
357 | } | |
358 | ||
a0fb3c47 | 359 | if (has_payload && data_size) { |
69a22773 | 360 | page = drbd_alloc_pages(peer_device, nr_pages, (gfp_mask & __GFP_WAIT)); |
a73ff323 LE |
361 | if (!page) |
362 | goto fail; | |
363 | } | |
b411b363 | 364 | |
db830c46 AG |
365 | drbd_clear_interval(&peer_req->i); |
366 | peer_req->i.size = data_size; | |
367 | peer_req->i.sector = sector; | |
368 | peer_req->i.local = false; | |
369 | peer_req->i.waiting = false; | |
370 | ||
371 | peer_req->epoch = NULL; | |
a8cd15ba | 372 | peer_req->peer_device = peer_device; |
db830c46 AG |
373 | peer_req->pages = page; |
374 | atomic_set(&peer_req->pending_bios, 0); | |
375 | peer_req->flags = 0; | |
9a8e7753 AG |
376 | /* |
377 | * The block_id is opaque to the receiver. It is not endianness | |
378 | * converted, and sent back to the sender unchanged. | |
379 | */ | |
db830c46 | 380 | peer_req->block_id = id; |
b411b363 | 381 | |
db830c46 | 382 | return peer_req; |
b411b363 | 383 | |
45bb912b | 384 | fail: |
db830c46 | 385 | mempool_free(peer_req, drbd_ee_mempool); |
b411b363 PR |
386 | return NULL; |
387 | } | |
388 | ||
b30ab791 | 389 | void __drbd_free_peer_req(struct drbd_device *device, struct drbd_peer_request *peer_req, |
f6ffca9f | 390 | int is_net) |
b411b363 | 391 | { |
db830c46 AG |
392 | if (peer_req->flags & EE_HAS_DIGEST) |
393 | kfree(peer_req->digest); | |
b30ab791 | 394 | drbd_free_pages(device, peer_req->pages, is_net); |
0b0ba1ef AG |
395 | D_ASSERT(device, atomic_read(&peer_req->pending_bios) == 0); |
396 | D_ASSERT(device, drbd_interval_empty(&peer_req->i)); | |
db830c46 | 397 | mempool_free(peer_req, drbd_ee_mempool); |
b411b363 PR |
398 | } |
399 | ||
b30ab791 | 400 | int drbd_free_peer_reqs(struct drbd_device *device, struct list_head *list) |
b411b363 PR |
401 | { |
402 | LIST_HEAD(work_list); | |
db830c46 | 403 | struct drbd_peer_request *peer_req, *t; |
b411b363 | 404 | int count = 0; |
b30ab791 | 405 | int is_net = list == &device->net_ee; |
b411b363 | 406 | |
0500813f | 407 | spin_lock_irq(&device->resource->req_lock); |
b411b363 | 408 | list_splice_init(list, &work_list); |
0500813f | 409 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 410 | |
a8cd15ba | 411 | list_for_each_entry_safe(peer_req, t, &work_list, w.list) { |
b30ab791 | 412 | __drbd_free_peer_req(device, peer_req, is_net); |
b411b363 PR |
413 | count++; |
414 | } | |
415 | return count; | |
416 | } | |
417 | ||
b411b363 | 418 | /* |
a990be46 | 419 | * See also comments in _req_mod(,BARRIER_ACKED) and receive_Barrier. |
b411b363 | 420 | */ |
b30ab791 | 421 | static int drbd_finish_peer_reqs(struct drbd_device *device) |
b411b363 PR |
422 | { |
423 | LIST_HEAD(work_list); | |
424 | LIST_HEAD(reclaimed); | |
db830c46 | 425 | struct drbd_peer_request *peer_req, *t; |
e2b3032b | 426 | int err = 0; |
b411b363 | 427 | |
0500813f | 428 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 AG |
429 | reclaim_finished_net_peer_reqs(device, &reclaimed); |
430 | list_splice_init(&device->done_ee, &work_list); | |
0500813f | 431 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 432 | |
a8cd15ba | 433 | list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) |
b30ab791 | 434 | drbd_free_net_peer_req(device, peer_req); |
b411b363 PR |
435 | |
436 | /* possible callbacks here: | |
d4dabbe2 | 437 | * e_end_block, and e_end_resync_block, e_send_superseded. |
b411b363 PR |
438 | * all ignore the last argument. |
439 | */ | |
a8cd15ba | 440 | list_for_each_entry_safe(peer_req, t, &work_list, w.list) { |
e2b3032b AG |
441 | int err2; |
442 | ||
b411b363 | 443 | /* list_del not necessary, next/prev members not touched */ |
a8cd15ba | 444 | err2 = peer_req->w.cb(&peer_req->w, !!err); |
e2b3032b AG |
445 | if (!err) |
446 | err = err2; | |
b30ab791 | 447 | drbd_free_peer_req(device, peer_req); |
b411b363 | 448 | } |
b30ab791 | 449 | wake_up(&device->ee_wait); |
b411b363 | 450 | |
e2b3032b | 451 | return err; |
b411b363 PR |
452 | } |
453 | ||
b30ab791 | 454 | static void _drbd_wait_ee_list_empty(struct drbd_device *device, |
d4da1537 | 455 | struct list_head *head) |
b411b363 PR |
456 | { |
457 | DEFINE_WAIT(wait); | |
458 | ||
459 | /* avoids spin_lock/unlock | |
460 | * and calling prepare_to_wait in the fast path */ | |
461 | while (!list_empty(head)) { | |
b30ab791 | 462 | prepare_to_wait(&device->ee_wait, &wait, TASK_UNINTERRUPTIBLE); |
0500813f | 463 | spin_unlock_irq(&device->resource->req_lock); |
7eaceacc | 464 | io_schedule(); |
b30ab791 | 465 | finish_wait(&device->ee_wait, &wait); |
0500813f | 466 | spin_lock_irq(&device->resource->req_lock); |
b411b363 PR |
467 | } |
468 | } | |
469 | ||
b30ab791 | 470 | static void drbd_wait_ee_list_empty(struct drbd_device *device, |
d4da1537 | 471 | struct list_head *head) |
b411b363 | 472 | { |
0500813f | 473 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 474 | _drbd_wait_ee_list_empty(device, head); |
0500813f | 475 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
476 | } |
477 | ||
dbd9eea0 | 478 | static int drbd_recv_short(struct socket *sock, void *buf, size_t size, int flags) |
b411b363 | 479 | { |
b411b363 PR |
480 | struct kvec iov = { |
481 | .iov_base = buf, | |
482 | .iov_len = size, | |
483 | }; | |
484 | struct msghdr msg = { | |
b411b363 PR |
485 | .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) |
486 | }; | |
f730c848 | 487 | return kernel_recvmsg(sock, &msg, &iov, 1, size, msg.msg_flags); |
b411b363 PR |
488 | } |
489 | ||
bde89a9e | 490 | static int drbd_recv(struct drbd_connection *connection, void *buf, size_t size) |
b411b363 | 491 | { |
b411b363 PR |
492 | int rv; |
493 | ||
bde89a9e | 494 | rv = drbd_recv_short(connection->data.socket, buf, size, 0); |
b411b363 | 495 | |
dbd0820c PR |
496 | if (rv < 0) { |
497 | if (rv == -ECONNRESET) | |
1ec861eb | 498 | drbd_info(connection, "sock was reset by peer\n"); |
dbd0820c | 499 | else if (rv != -ERESTARTSYS) |
1ec861eb | 500 | drbd_err(connection, "sock_recvmsg returned %d\n", rv); |
dbd0820c | 501 | } else if (rv == 0) { |
bde89a9e | 502 | if (test_bit(DISCONNECT_SENT, &connection->flags)) { |
b66623e3 PR |
503 | long t; |
504 | rcu_read_lock(); | |
bde89a9e | 505 | t = rcu_dereference(connection->net_conf)->ping_timeo * HZ/10; |
b66623e3 PR |
506 | rcu_read_unlock(); |
507 | ||
bde89a9e | 508 | t = wait_event_timeout(connection->ping_wait, connection->cstate < C_WF_REPORT_PARAMS, t); |
b66623e3 | 509 | |
599377ac PR |
510 | if (t) |
511 | goto out; | |
512 | } | |
1ec861eb | 513 | drbd_info(connection, "sock was shut down by peer\n"); |
599377ac PR |
514 | } |
515 | ||
b411b363 | 516 | if (rv != size) |
bde89a9e | 517 | conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 | 518 | |
599377ac | 519 | out: |
b411b363 PR |
520 | return rv; |
521 | } | |
522 | ||
bde89a9e | 523 | static int drbd_recv_all(struct drbd_connection *connection, void *buf, size_t size) |
c6967746 AG |
524 | { |
525 | int err; | |
526 | ||
bde89a9e | 527 | err = drbd_recv(connection, buf, size); |
c6967746 AG |
528 | if (err != size) { |
529 | if (err >= 0) | |
530 | err = -EIO; | |
531 | } else | |
532 | err = 0; | |
533 | return err; | |
534 | } | |
535 | ||
bde89a9e | 536 | static int drbd_recv_all_warn(struct drbd_connection *connection, void *buf, size_t size) |
a5c31904 AG |
537 | { |
538 | int err; | |
539 | ||
bde89a9e | 540 | err = drbd_recv_all(connection, buf, size); |
a5c31904 | 541 | if (err && !signal_pending(current)) |
1ec861eb | 542 | drbd_warn(connection, "short read (expected size %d)\n", (int)size); |
a5c31904 AG |
543 | return err; |
544 | } | |
545 | ||
5dbf1673 LE |
546 | /* quoting tcp(7): |
547 | * On individual connections, the socket buffer size must be set prior to the | |
548 | * listen(2) or connect(2) calls in order to have it take effect. | |
549 | * This is our wrapper to do so. | |
550 | */ | |
551 | static void drbd_setbufsize(struct socket *sock, unsigned int snd, | |
552 | unsigned int rcv) | |
553 | { | |
554 | /* open coded SO_SNDBUF, SO_RCVBUF */ | |
555 | if (snd) { | |
556 | sock->sk->sk_sndbuf = snd; | |
557 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
558 | } | |
559 | if (rcv) { | |
560 | sock->sk->sk_rcvbuf = rcv; | |
561 | sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
562 | } | |
563 | } | |
564 | ||
bde89a9e | 565 | static struct socket *drbd_try_connect(struct drbd_connection *connection) |
b411b363 PR |
566 | { |
567 | const char *what; | |
568 | struct socket *sock; | |
569 | struct sockaddr_in6 src_in6; | |
44ed167d PR |
570 | struct sockaddr_in6 peer_in6; |
571 | struct net_conf *nc; | |
572 | int err, peer_addr_len, my_addr_len; | |
69ef82de | 573 | int sndbuf_size, rcvbuf_size, connect_int; |
b411b363 PR |
574 | int disconnect_on_error = 1; |
575 | ||
44ed167d | 576 | rcu_read_lock(); |
bde89a9e | 577 | nc = rcu_dereference(connection->net_conf); |
44ed167d PR |
578 | if (!nc) { |
579 | rcu_read_unlock(); | |
b411b363 | 580 | return NULL; |
44ed167d | 581 | } |
44ed167d PR |
582 | sndbuf_size = nc->sndbuf_size; |
583 | rcvbuf_size = nc->rcvbuf_size; | |
69ef82de | 584 | connect_int = nc->connect_int; |
089c075d | 585 | rcu_read_unlock(); |
44ed167d | 586 | |
bde89a9e AG |
587 | my_addr_len = min_t(int, connection->my_addr_len, sizeof(src_in6)); |
588 | memcpy(&src_in6, &connection->my_addr, my_addr_len); | |
44ed167d | 589 | |
bde89a9e | 590 | if (((struct sockaddr *)&connection->my_addr)->sa_family == AF_INET6) |
44ed167d PR |
591 | src_in6.sin6_port = 0; |
592 | else | |
593 | ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ | |
594 | ||
bde89a9e AG |
595 | peer_addr_len = min_t(int, connection->peer_addr_len, sizeof(src_in6)); |
596 | memcpy(&peer_in6, &connection->peer_addr, peer_addr_len); | |
b411b363 PR |
597 | |
598 | what = "sock_create_kern"; | |
44ed167d PR |
599 | err = sock_create_kern(((struct sockaddr *)&src_in6)->sa_family, |
600 | SOCK_STREAM, IPPROTO_TCP, &sock); | |
b411b363 PR |
601 | if (err < 0) { |
602 | sock = NULL; | |
603 | goto out; | |
604 | } | |
605 | ||
606 | sock->sk->sk_rcvtimeo = | |
69ef82de | 607 | sock->sk->sk_sndtimeo = connect_int * HZ; |
44ed167d | 608 | drbd_setbufsize(sock, sndbuf_size, rcvbuf_size); |
b411b363 PR |
609 | |
610 | /* explicitly bind to the configured IP as source IP | |
611 | * for the outgoing connections. | |
612 | * This is needed for multihomed hosts and to be | |
613 | * able to use lo: interfaces for drbd. | |
614 | * Make sure to use 0 as port number, so linux selects | |
615 | * a free one dynamically. | |
616 | */ | |
b411b363 | 617 | what = "bind before connect"; |
44ed167d | 618 | err = sock->ops->bind(sock, (struct sockaddr *) &src_in6, my_addr_len); |
b411b363 PR |
619 | if (err < 0) |
620 | goto out; | |
621 | ||
622 | /* connect may fail, peer not yet available. | |
623 | * stay C_WF_CONNECTION, don't go Disconnecting! */ | |
624 | disconnect_on_error = 0; | |
625 | what = "connect"; | |
44ed167d | 626 | err = sock->ops->connect(sock, (struct sockaddr *) &peer_in6, peer_addr_len, 0); |
b411b363 PR |
627 | |
628 | out: | |
629 | if (err < 0) { | |
630 | if (sock) { | |
631 | sock_release(sock); | |
632 | sock = NULL; | |
633 | } | |
634 | switch (-err) { | |
635 | /* timeout, busy, signal pending */ | |
636 | case ETIMEDOUT: case EAGAIN: case EINPROGRESS: | |
637 | case EINTR: case ERESTARTSYS: | |
638 | /* peer not (yet) available, network problem */ | |
639 | case ECONNREFUSED: case ENETUNREACH: | |
640 | case EHOSTDOWN: case EHOSTUNREACH: | |
641 | disconnect_on_error = 0; | |
642 | break; | |
643 | default: | |
1ec861eb | 644 | drbd_err(connection, "%s failed, err = %d\n", what, err); |
b411b363 PR |
645 | } |
646 | if (disconnect_on_error) | |
bde89a9e | 647 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 648 | } |
44ed167d | 649 | |
b411b363 PR |
650 | return sock; |
651 | } | |
652 | ||
7a426fd8 | 653 | struct accept_wait_data { |
bde89a9e | 654 | struct drbd_connection *connection; |
7a426fd8 PR |
655 | struct socket *s_listen; |
656 | struct completion door_bell; | |
657 | void (*original_sk_state_change)(struct sock *sk); | |
658 | ||
659 | }; | |
660 | ||
715306f6 | 661 | static void drbd_incoming_connection(struct sock *sk) |
7a426fd8 PR |
662 | { |
663 | struct accept_wait_data *ad = sk->sk_user_data; | |
715306f6 | 664 | void (*state_change)(struct sock *sk); |
7a426fd8 | 665 | |
715306f6 AG |
666 | state_change = ad->original_sk_state_change; |
667 | if (sk->sk_state == TCP_ESTABLISHED) | |
668 | complete(&ad->door_bell); | |
669 | state_change(sk); | |
7a426fd8 PR |
670 | } |
671 | ||
bde89a9e | 672 | static int prepare_listen_socket(struct drbd_connection *connection, struct accept_wait_data *ad) |
b411b363 | 673 | { |
1f3e509b | 674 | int err, sndbuf_size, rcvbuf_size, my_addr_len; |
44ed167d | 675 | struct sockaddr_in6 my_addr; |
1f3e509b | 676 | struct socket *s_listen; |
44ed167d | 677 | struct net_conf *nc; |
b411b363 PR |
678 | const char *what; |
679 | ||
44ed167d | 680 | rcu_read_lock(); |
bde89a9e | 681 | nc = rcu_dereference(connection->net_conf); |
44ed167d PR |
682 | if (!nc) { |
683 | rcu_read_unlock(); | |
7a426fd8 | 684 | return -EIO; |
44ed167d | 685 | } |
44ed167d PR |
686 | sndbuf_size = nc->sndbuf_size; |
687 | rcvbuf_size = nc->rcvbuf_size; | |
44ed167d | 688 | rcu_read_unlock(); |
b411b363 | 689 | |
bde89a9e AG |
690 | my_addr_len = min_t(int, connection->my_addr_len, sizeof(struct sockaddr_in6)); |
691 | memcpy(&my_addr, &connection->my_addr, my_addr_len); | |
b411b363 PR |
692 | |
693 | what = "sock_create_kern"; | |
44ed167d | 694 | err = sock_create_kern(((struct sockaddr *)&my_addr)->sa_family, |
1f3e509b | 695 | SOCK_STREAM, IPPROTO_TCP, &s_listen); |
b411b363 PR |
696 | if (err) { |
697 | s_listen = NULL; | |
698 | goto out; | |
699 | } | |
700 | ||
98683650 | 701 | s_listen->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ |
44ed167d | 702 | drbd_setbufsize(s_listen, sndbuf_size, rcvbuf_size); |
b411b363 PR |
703 | |
704 | what = "bind before listen"; | |
44ed167d | 705 | err = s_listen->ops->bind(s_listen, (struct sockaddr *)&my_addr, my_addr_len); |
b411b363 PR |
706 | if (err < 0) |
707 | goto out; | |
708 | ||
7a426fd8 PR |
709 | ad->s_listen = s_listen; |
710 | write_lock_bh(&s_listen->sk->sk_callback_lock); | |
711 | ad->original_sk_state_change = s_listen->sk->sk_state_change; | |
715306f6 | 712 | s_listen->sk->sk_state_change = drbd_incoming_connection; |
7a426fd8 PR |
713 | s_listen->sk->sk_user_data = ad; |
714 | write_unlock_bh(&s_listen->sk->sk_callback_lock); | |
b411b363 | 715 | |
2820fd39 PR |
716 | what = "listen"; |
717 | err = s_listen->ops->listen(s_listen, 5); | |
718 | if (err < 0) | |
719 | goto out; | |
720 | ||
7a426fd8 | 721 | return 0; |
b411b363 PR |
722 | out: |
723 | if (s_listen) | |
724 | sock_release(s_listen); | |
725 | if (err < 0) { | |
726 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
1ec861eb | 727 | drbd_err(connection, "%s failed, err = %d\n", what, err); |
bde89a9e | 728 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
729 | } |
730 | } | |
b411b363 | 731 | |
7a426fd8 | 732 | return -EIO; |
b411b363 PR |
733 | } |
734 | ||
715306f6 | 735 | static void unregister_state_change(struct sock *sk, struct accept_wait_data *ad) |
b411b363 | 736 | { |
715306f6 AG |
737 | write_lock_bh(&sk->sk_callback_lock); |
738 | sk->sk_state_change = ad->original_sk_state_change; | |
739 | sk->sk_user_data = NULL; | |
740 | write_unlock_bh(&sk->sk_callback_lock); | |
b411b363 PR |
741 | } |
742 | ||
bde89a9e | 743 | static struct socket *drbd_wait_for_connect(struct drbd_connection *connection, struct accept_wait_data *ad) |
b411b363 | 744 | { |
1f3e509b PR |
745 | int timeo, connect_int, err = 0; |
746 | struct socket *s_estab = NULL; | |
1f3e509b PR |
747 | struct net_conf *nc; |
748 | ||
749 | rcu_read_lock(); | |
bde89a9e | 750 | nc = rcu_dereference(connection->net_conf); |
1f3e509b PR |
751 | if (!nc) { |
752 | rcu_read_unlock(); | |
753 | return NULL; | |
754 | } | |
755 | connect_int = nc->connect_int; | |
756 | rcu_read_unlock(); | |
757 | ||
758 | timeo = connect_int * HZ; | |
38b682b2 AM |
759 | /* 28.5% random jitter */ |
760 | timeo += (prandom_u32() & 1) ? timeo / 7 : -timeo / 7; | |
1f3e509b | 761 | |
7a426fd8 PR |
762 | err = wait_for_completion_interruptible_timeout(&ad->door_bell, timeo); |
763 | if (err <= 0) | |
764 | return NULL; | |
b411b363 | 765 | |
7a426fd8 | 766 | err = kernel_accept(ad->s_listen, &s_estab, 0); |
b411b363 PR |
767 | if (err < 0) { |
768 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
1ec861eb | 769 | drbd_err(connection, "accept failed, err = %d\n", err); |
bde89a9e | 770 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
771 | } |
772 | } | |
b411b363 | 773 | |
715306f6 AG |
774 | if (s_estab) |
775 | unregister_state_change(s_estab->sk, ad); | |
b411b363 | 776 | |
b411b363 PR |
777 | return s_estab; |
778 | } | |
b411b363 | 779 | |
bde89a9e | 780 | static int decode_header(struct drbd_connection *, void *, struct packet_info *); |
b411b363 | 781 | |
bde89a9e | 782 | static int send_first_packet(struct drbd_connection *connection, struct drbd_socket *sock, |
9f5bdc33 AG |
783 | enum drbd_packet cmd) |
784 | { | |
bde89a9e | 785 | if (!conn_prepare_command(connection, sock)) |
9f5bdc33 | 786 | return -EIO; |
bde89a9e | 787 | return conn_send_command(connection, sock, cmd, 0, NULL, 0); |
b411b363 PR |
788 | } |
789 | ||
bde89a9e | 790 | static int receive_first_packet(struct drbd_connection *connection, struct socket *sock) |
b411b363 | 791 | { |
bde89a9e | 792 | unsigned int header_size = drbd_header_size(connection); |
9f5bdc33 AG |
793 | struct packet_info pi; |
794 | int err; | |
b411b363 | 795 | |
bde89a9e | 796 | err = drbd_recv_short(sock, connection->data.rbuf, header_size, 0); |
9f5bdc33 AG |
797 | if (err != header_size) { |
798 | if (err >= 0) | |
799 | err = -EIO; | |
800 | return err; | |
801 | } | |
bde89a9e | 802 | err = decode_header(connection, connection->data.rbuf, &pi); |
9f5bdc33 AG |
803 | if (err) |
804 | return err; | |
805 | return pi.cmd; | |
b411b363 PR |
806 | } |
807 | ||
808 | /** | |
809 | * drbd_socket_okay() - Free the socket if its connection is not okay | |
b411b363 PR |
810 | * @sock: pointer to the pointer to the socket. |
811 | */ | |
dbd9eea0 | 812 | static int drbd_socket_okay(struct socket **sock) |
b411b363 PR |
813 | { |
814 | int rr; | |
815 | char tb[4]; | |
816 | ||
817 | if (!*sock) | |
81e84650 | 818 | return false; |
b411b363 | 819 | |
dbd9eea0 | 820 | rr = drbd_recv_short(*sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); |
b411b363 PR |
821 | |
822 | if (rr > 0 || rr == -EAGAIN) { | |
81e84650 | 823 | return true; |
b411b363 PR |
824 | } else { |
825 | sock_release(*sock); | |
826 | *sock = NULL; | |
81e84650 | 827 | return false; |
b411b363 PR |
828 | } |
829 | } | |
2325eb66 PR |
830 | /* Gets called if a connection is established, or if a new minor gets created |
831 | in a connection */ | |
69a22773 | 832 | int drbd_connected(struct drbd_peer_device *peer_device) |
907599e0 | 833 | { |
69a22773 | 834 | struct drbd_device *device = peer_device->device; |
0829f5ed | 835 | int err; |
907599e0 | 836 | |
b30ab791 AG |
837 | atomic_set(&device->packet_seq, 0); |
838 | device->peer_seq = 0; | |
907599e0 | 839 | |
69a22773 AG |
840 | device->state_mutex = peer_device->connection->agreed_pro_version < 100 ? |
841 | &peer_device->connection->cstate_mutex : | |
b30ab791 | 842 | &device->own_state_mutex; |
8410da8f | 843 | |
69a22773 | 844 | err = drbd_send_sync_param(peer_device); |
0829f5ed | 845 | if (!err) |
69a22773 | 846 | err = drbd_send_sizes(peer_device, 0, 0); |
0829f5ed | 847 | if (!err) |
69a22773 | 848 | err = drbd_send_uuids(peer_device); |
0829f5ed | 849 | if (!err) |
69a22773 | 850 | err = drbd_send_current_state(peer_device); |
b30ab791 AG |
851 | clear_bit(USE_DEGR_WFC_T, &device->flags); |
852 | clear_bit(RESIZE_PENDING, &device->flags); | |
853 | atomic_set(&device->ap_in_flight, 0); | |
854 | mod_timer(&device->request_timer, jiffies + HZ); /* just start it here. */ | |
0829f5ed | 855 | return err; |
907599e0 | 856 | } |
b411b363 PR |
857 | |
858 | /* | |
859 | * return values: | |
860 | * 1 yes, we have a valid connection | |
861 | * 0 oops, did not work out, please try again | |
862 | * -1 peer talks different language, | |
863 | * no point in trying again, please go standalone. | |
864 | * -2 We do not have a network config... | |
865 | */ | |
bde89a9e | 866 | static int conn_connect(struct drbd_connection *connection) |
b411b363 | 867 | { |
7da35862 | 868 | struct drbd_socket sock, msock; |
c06ece6b | 869 | struct drbd_peer_device *peer_device; |
44ed167d | 870 | struct net_conf *nc; |
92f14951 | 871 | int vnr, timeout, h, ok; |
08b165ba | 872 | bool discard_my_data; |
197296ff | 873 | enum drbd_state_rv rv; |
7a426fd8 | 874 | struct accept_wait_data ad = { |
bde89a9e | 875 | .connection = connection, |
7a426fd8 PR |
876 | .door_bell = COMPLETION_INITIALIZER_ONSTACK(ad.door_bell), |
877 | }; | |
b411b363 | 878 | |
bde89a9e AG |
879 | clear_bit(DISCONNECT_SENT, &connection->flags); |
880 | if (conn_request_state(connection, NS(conn, C_WF_CONNECTION), CS_VERBOSE) < SS_SUCCESS) | |
b411b363 PR |
881 | return -2; |
882 | ||
7da35862 | 883 | mutex_init(&sock.mutex); |
bde89a9e AG |
884 | sock.sbuf = connection->data.sbuf; |
885 | sock.rbuf = connection->data.rbuf; | |
7da35862 PR |
886 | sock.socket = NULL; |
887 | mutex_init(&msock.mutex); | |
bde89a9e AG |
888 | msock.sbuf = connection->meta.sbuf; |
889 | msock.rbuf = connection->meta.rbuf; | |
7da35862 PR |
890 | msock.socket = NULL; |
891 | ||
0916e0e3 | 892 | /* Assume that the peer only understands protocol 80 until we know better. */ |
bde89a9e | 893 | connection->agreed_pro_version = 80; |
b411b363 | 894 | |
bde89a9e | 895 | if (prepare_listen_socket(connection, &ad)) |
7a426fd8 | 896 | return 0; |
b411b363 PR |
897 | |
898 | do { | |
2bf89621 | 899 | struct socket *s; |
b411b363 | 900 | |
bde89a9e | 901 | s = drbd_try_connect(connection); |
b411b363 | 902 | if (s) { |
7da35862 PR |
903 | if (!sock.socket) { |
904 | sock.socket = s; | |
bde89a9e | 905 | send_first_packet(connection, &sock, P_INITIAL_DATA); |
7da35862 | 906 | } else if (!msock.socket) { |
bde89a9e | 907 | clear_bit(RESOLVE_CONFLICTS, &connection->flags); |
7da35862 | 908 | msock.socket = s; |
bde89a9e | 909 | send_first_packet(connection, &msock, P_INITIAL_META); |
b411b363 | 910 | } else { |
1ec861eb | 911 | drbd_err(connection, "Logic error in conn_connect()\n"); |
b411b363 PR |
912 | goto out_release_sockets; |
913 | } | |
914 | } | |
915 | ||
7da35862 PR |
916 | if (sock.socket && msock.socket) { |
917 | rcu_read_lock(); | |
bde89a9e | 918 | nc = rcu_dereference(connection->net_conf); |
7da35862 PR |
919 | timeout = nc->ping_timeo * HZ / 10; |
920 | rcu_read_unlock(); | |
921 | schedule_timeout_interruptible(timeout); | |
922 | ok = drbd_socket_okay(&sock.socket); | |
923 | ok = drbd_socket_okay(&msock.socket) && ok; | |
b411b363 PR |
924 | if (ok) |
925 | break; | |
926 | } | |
927 | ||
928 | retry: | |
bde89a9e | 929 | s = drbd_wait_for_connect(connection, &ad); |
b411b363 | 930 | if (s) { |
bde89a9e | 931 | int fp = receive_first_packet(connection, s); |
7da35862 PR |
932 | drbd_socket_okay(&sock.socket); |
933 | drbd_socket_okay(&msock.socket); | |
92f14951 | 934 | switch (fp) { |
e5d6f33a | 935 | case P_INITIAL_DATA: |
7da35862 | 936 | if (sock.socket) { |
1ec861eb | 937 | drbd_warn(connection, "initial packet S crossed\n"); |
7da35862 | 938 | sock_release(sock.socket); |
80c6eed4 PR |
939 | sock.socket = s; |
940 | goto randomize; | |
b411b363 | 941 | } |
7da35862 | 942 | sock.socket = s; |
b411b363 | 943 | break; |
e5d6f33a | 944 | case P_INITIAL_META: |
bde89a9e | 945 | set_bit(RESOLVE_CONFLICTS, &connection->flags); |
7da35862 | 946 | if (msock.socket) { |
1ec861eb | 947 | drbd_warn(connection, "initial packet M crossed\n"); |
7da35862 | 948 | sock_release(msock.socket); |
80c6eed4 PR |
949 | msock.socket = s; |
950 | goto randomize; | |
b411b363 | 951 | } |
7da35862 | 952 | msock.socket = s; |
b411b363 PR |
953 | break; |
954 | default: | |
1ec861eb | 955 | drbd_warn(connection, "Error receiving initial packet\n"); |
b411b363 | 956 | sock_release(s); |
80c6eed4 | 957 | randomize: |
38b682b2 | 958 | if (prandom_u32() & 1) |
b411b363 PR |
959 | goto retry; |
960 | } | |
961 | } | |
962 | ||
bde89a9e | 963 | if (connection->cstate <= C_DISCONNECTING) |
b411b363 PR |
964 | goto out_release_sockets; |
965 | if (signal_pending(current)) { | |
966 | flush_signals(current); | |
967 | smp_rmb(); | |
bde89a9e | 968 | if (get_t_state(&connection->receiver) == EXITING) |
b411b363 PR |
969 | goto out_release_sockets; |
970 | } | |
971 | ||
b666dbf8 PR |
972 | ok = drbd_socket_okay(&sock.socket); |
973 | ok = drbd_socket_okay(&msock.socket) && ok; | |
974 | } while (!ok); | |
b411b363 | 975 | |
7a426fd8 PR |
976 | if (ad.s_listen) |
977 | sock_release(ad.s_listen); | |
b411b363 | 978 | |
98683650 PR |
979 | sock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ |
980 | msock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ | |
b411b363 | 981 | |
7da35862 PR |
982 | sock.socket->sk->sk_allocation = GFP_NOIO; |
983 | msock.socket->sk->sk_allocation = GFP_NOIO; | |
b411b363 | 984 | |
7da35862 PR |
985 | sock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; |
986 | msock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE; | |
b411b363 | 987 | |
b411b363 | 988 | /* NOT YET ... |
bde89a9e | 989 | * sock.socket->sk->sk_sndtimeo = connection->net_conf->timeout*HZ/10; |
7da35862 | 990 | * sock.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; |
6038178e | 991 | * first set it to the P_CONNECTION_FEATURES timeout, |
b411b363 | 992 | * which we set to 4x the configured ping_timeout. */ |
44ed167d | 993 | rcu_read_lock(); |
bde89a9e | 994 | nc = rcu_dereference(connection->net_conf); |
44ed167d | 995 | |
7da35862 PR |
996 | sock.socket->sk->sk_sndtimeo = |
997 | sock.socket->sk->sk_rcvtimeo = nc->ping_timeo*4*HZ/10; | |
b411b363 | 998 | |
7da35862 | 999 | msock.socket->sk->sk_rcvtimeo = nc->ping_int*HZ; |
44ed167d | 1000 | timeout = nc->timeout * HZ / 10; |
08b165ba | 1001 | discard_my_data = nc->discard_my_data; |
44ed167d | 1002 | rcu_read_unlock(); |
b411b363 | 1003 | |
7da35862 | 1004 | msock.socket->sk->sk_sndtimeo = timeout; |
b411b363 PR |
1005 | |
1006 | /* we don't want delays. | |
25985edc | 1007 | * we use TCP_CORK where appropriate, though */ |
7da35862 PR |
1008 | drbd_tcp_nodelay(sock.socket); |
1009 | drbd_tcp_nodelay(msock.socket); | |
b411b363 | 1010 | |
bde89a9e AG |
1011 | connection->data.socket = sock.socket; |
1012 | connection->meta.socket = msock.socket; | |
1013 | connection->last_received = jiffies; | |
b411b363 | 1014 | |
bde89a9e | 1015 | h = drbd_do_features(connection); |
b411b363 PR |
1016 | if (h <= 0) |
1017 | return h; | |
1018 | ||
bde89a9e | 1019 | if (connection->cram_hmac_tfm) { |
b30ab791 | 1020 | /* drbd_request_state(device, NS(conn, WFAuth)); */ |
bde89a9e | 1021 | switch (drbd_do_auth(connection)) { |
b10d96cb | 1022 | case -1: |
1ec861eb | 1023 | drbd_err(connection, "Authentication of peer failed\n"); |
b411b363 | 1024 | return -1; |
b10d96cb | 1025 | case 0: |
1ec861eb | 1026 | drbd_err(connection, "Authentication of peer failed, trying again.\n"); |
b10d96cb | 1027 | return 0; |
b411b363 PR |
1028 | } |
1029 | } | |
1030 | ||
bde89a9e AG |
1031 | connection->data.socket->sk->sk_sndtimeo = timeout; |
1032 | connection->data.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; | |
b411b363 | 1033 | |
bde89a9e | 1034 | if (drbd_send_protocol(connection) == -EOPNOTSUPP) |
7e2455c1 | 1035 | return -1; |
b411b363 | 1036 | |
31007745 PR |
1037 | /* Prevent a race between resync-handshake and |
1038 | * being promoted to Primary. | |
1039 | * | |
1040 | * Grab and release the state mutex, so we know that any current | |
1041 | * drbd_set_role() is finished, and any incoming drbd_set_role | |
1042 | * will see the STATE_SENT flag, and wait for it to be cleared. | |
1043 | */ | |
1044 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) | |
1045 | mutex_lock(peer_device->device->state_mutex); | |
1046 | ||
bde89a9e | 1047 | set_bit(STATE_SENT, &connection->flags); |
a1096a6e | 1048 | |
31007745 PR |
1049 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) |
1050 | mutex_unlock(peer_device->device->state_mutex); | |
1051 | ||
c141ebda | 1052 | rcu_read_lock(); |
c06ece6b AG |
1053 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
1054 | struct drbd_device *device = peer_device->device; | |
b30ab791 | 1055 | kref_get(&device->kref); |
26ea8f92 AG |
1056 | rcu_read_unlock(); |
1057 | ||
08b165ba | 1058 | if (discard_my_data) |
b30ab791 | 1059 | set_bit(DISCARD_MY_DATA, &device->flags); |
08b165ba | 1060 | else |
b30ab791 | 1061 | clear_bit(DISCARD_MY_DATA, &device->flags); |
08b165ba | 1062 | |
69a22773 | 1063 | drbd_connected(peer_device); |
05a10ec7 | 1064 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda PR |
1065 | rcu_read_lock(); |
1066 | } | |
1067 | rcu_read_unlock(); | |
1068 | ||
bde89a9e AG |
1069 | rv = conn_request_state(connection, NS(conn, C_WF_REPORT_PARAMS), CS_VERBOSE); |
1070 | if (rv < SS_SUCCESS || connection->cstate != C_WF_REPORT_PARAMS) { | |
1071 | clear_bit(STATE_SENT, &connection->flags); | |
1e86ac48 | 1072 | return 0; |
a1096a6e | 1073 | } |
1e86ac48 | 1074 | |
bde89a9e | 1075 | drbd_thread_start(&connection->asender); |
b411b363 | 1076 | |
0500813f | 1077 | mutex_lock(&connection->resource->conf_update); |
08b165ba PR |
1078 | /* The discard_my_data flag is a single-shot modifier to the next |
1079 | * connection attempt, the handshake of which is now well underway. | |
1080 | * No need for rcu style copying of the whole struct | |
1081 | * just to clear a single value. */ | |
bde89a9e | 1082 | connection->net_conf->discard_my_data = 0; |
0500813f | 1083 | mutex_unlock(&connection->resource->conf_update); |
08b165ba | 1084 | |
d3fcb490 | 1085 | return h; |
b411b363 PR |
1086 | |
1087 | out_release_sockets: | |
7a426fd8 PR |
1088 | if (ad.s_listen) |
1089 | sock_release(ad.s_listen); | |
7da35862 PR |
1090 | if (sock.socket) |
1091 | sock_release(sock.socket); | |
1092 | if (msock.socket) | |
1093 | sock_release(msock.socket); | |
b411b363 PR |
1094 | return -1; |
1095 | } | |
1096 | ||
bde89a9e | 1097 | static int decode_header(struct drbd_connection *connection, void *header, struct packet_info *pi) |
b411b363 | 1098 | { |
bde89a9e | 1099 | unsigned int header_size = drbd_header_size(connection); |
e658983a | 1100 | |
0c8e36d9 AG |
1101 | if (header_size == sizeof(struct p_header100) && |
1102 | *(__be32 *)header == cpu_to_be32(DRBD_MAGIC_100)) { | |
1103 | struct p_header100 *h = header; | |
1104 | if (h->pad != 0) { | |
1ec861eb | 1105 | drbd_err(connection, "Header padding is not zero\n"); |
0c8e36d9 AG |
1106 | return -EINVAL; |
1107 | } | |
1108 | pi->vnr = be16_to_cpu(h->volume); | |
1109 | pi->cmd = be16_to_cpu(h->command); | |
1110 | pi->size = be32_to_cpu(h->length); | |
1111 | } else if (header_size == sizeof(struct p_header95) && | |
1112 | *(__be16 *)header == cpu_to_be16(DRBD_MAGIC_BIG)) { | |
e658983a | 1113 | struct p_header95 *h = header; |
e658983a | 1114 | pi->cmd = be16_to_cpu(h->command); |
b55d84ba AG |
1115 | pi->size = be32_to_cpu(h->length); |
1116 | pi->vnr = 0; | |
e658983a AG |
1117 | } else if (header_size == sizeof(struct p_header80) && |
1118 | *(__be32 *)header == cpu_to_be32(DRBD_MAGIC)) { | |
1119 | struct p_header80 *h = header; | |
1120 | pi->cmd = be16_to_cpu(h->command); | |
1121 | pi->size = be16_to_cpu(h->length); | |
77351055 | 1122 | pi->vnr = 0; |
02918be2 | 1123 | } else { |
1ec861eb | 1124 | drbd_err(connection, "Wrong magic value 0x%08x in protocol version %d\n", |
e658983a | 1125 | be32_to_cpu(*(__be32 *)header), |
bde89a9e | 1126 | connection->agreed_pro_version); |
8172f3e9 | 1127 | return -EINVAL; |
b411b363 | 1128 | } |
e658983a | 1129 | pi->data = header + header_size; |
8172f3e9 | 1130 | return 0; |
257d0af6 | 1131 | } |
b411b363 | 1132 | |
bde89a9e | 1133 | static int drbd_recv_header(struct drbd_connection *connection, struct packet_info *pi) |
257d0af6 | 1134 | { |
bde89a9e | 1135 | void *buffer = connection->data.rbuf; |
69bc7bc3 | 1136 | int err; |
257d0af6 | 1137 | |
bde89a9e | 1138 | err = drbd_recv_all_warn(connection, buffer, drbd_header_size(connection)); |
a5c31904 | 1139 | if (err) |
69bc7bc3 | 1140 | return err; |
257d0af6 | 1141 | |
bde89a9e AG |
1142 | err = decode_header(connection, buffer, pi); |
1143 | connection->last_received = jiffies; | |
b411b363 | 1144 | |
69bc7bc3 | 1145 | return err; |
b411b363 PR |
1146 | } |
1147 | ||
bde89a9e | 1148 | static void drbd_flush(struct drbd_connection *connection) |
b411b363 PR |
1149 | { |
1150 | int rv; | |
c06ece6b | 1151 | struct drbd_peer_device *peer_device; |
4b0007c0 PR |
1152 | int vnr; |
1153 | ||
e9526580 | 1154 | if (connection->resource->write_ordering >= WO_bdev_flush) { |
615e087f | 1155 | rcu_read_lock(); |
c06ece6b AG |
1156 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
1157 | struct drbd_device *device = peer_device->device; | |
1158 | ||
b30ab791 | 1159 | if (!get_ldev(device)) |
615e087f | 1160 | continue; |
b30ab791 | 1161 | kref_get(&device->kref); |
615e087f LE |
1162 | rcu_read_unlock(); |
1163 | ||
b30ab791 | 1164 | rv = blkdev_issue_flush(device->ldev->backing_bdev, |
615e087f LE |
1165 | GFP_NOIO, NULL); |
1166 | if (rv) { | |
d0180171 | 1167 | drbd_info(device, "local disk flush failed with status %d\n", rv); |
615e087f LE |
1168 | /* would rather check on EOPNOTSUPP, but that is not reliable. |
1169 | * don't try again for ANY return value != 0 | |
1170 | * if (rv == -EOPNOTSUPP) */ | |
8fe39aac | 1171 | drbd_bump_write_ordering(connection->resource, NULL, WO_drain_io); |
4b0007c0 | 1172 | } |
b30ab791 | 1173 | put_ldev(device); |
05a10ec7 | 1174 | kref_put(&device->kref, drbd_destroy_device); |
b411b363 | 1175 | |
615e087f LE |
1176 | rcu_read_lock(); |
1177 | if (rv) | |
1178 | break; | |
b411b363 | 1179 | } |
615e087f | 1180 | rcu_read_unlock(); |
b411b363 | 1181 | } |
b411b363 PR |
1182 | } |
1183 | ||
1184 | /** | |
1185 | * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. | |
b30ab791 | 1186 | * @device: DRBD device. |
b411b363 PR |
1187 | * @epoch: Epoch object. |
1188 | * @ev: Epoch event. | |
1189 | */ | |
bde89a9e | 1190 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_connection *connection, |
b411b363 PR |
1191 | struct drbd_epoch *epoch, |
1192 | enum epoch_event ev) | |
1193 | { | |
2451fc3b | 1194 | int epoch_size; |
b411b363 | 1195 | struct drbd_epoch *next_epoch; |
b411b363 PR |
1196 | enum finish_epoch rv = FE_STILL_LIVE; |
1197 | ||
bde89a9e | 1198 | spin_lock(&connection->epoch_lock); |
b411b363 PR |
1199 | do { |
1200 | next_epoch = NULL; | |
b411b363 PR |
1201 | |
1202 | epoch_size = atomic_read(&epoch->epoch_size); | |
1203 | ||
1204 | switch (ev & ~EV_CLEANUP) { | |
1205 | case EV_PUT: | |
1206 | atomic_dec(&epoch->active); | |
1207 | break; | |
1208 | case EV_GOT_BARRIER_NR: | |
1209 | set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); | |
b411b363 PR |
1210 | break; |
1211 | case EV_BECAME_LAST: | |
1212 | /* nothing to do*/ | |
1213 | break; | |
1214 | } | |
1215 | ||
b411b363 PR |
1216 | if (epoch_size != 0 && |
1217 | atomic_read(&epoch->active) == 0 && | |
80f9fd55 | 1218 | (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) || ev & EV_CLEANUP)) { |
b411b363 | 1219 | if (!(ev & EV_CLEANUP)) { |
bde89a9e AG |
1220 | spin_unlock(&connection->epoch_lock); |
1221 | drbd_send_b_ack(epoch->connection, epoch->barrier_nr, epoch_size); | |
1222 | spin_lock(&connection->epoch_lock); | |
b411b363 | 1223 | } |
9ed57dcb LE |
1224 | #if 0 |
1225 | /* FIXME: dec unacked on connection, once we have | |
1226 | * something to count pending connection packets in. */ | |
80f9fd55 | 1227 | if (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) |
bde89a9e | 1228 | dec_unacked(epoch->connection); |
9ed57dcb | 1229 | #endif |
b411b363 | 1230 | |
bde89a9e | 1231 | if (connection->current_epoch != epoch) { |
b411b363 PR |
1232 | next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); |
1233 | list_del(&epoch->list); | |
1234 | ev = EV_BECAME_LAST | (ev & EV_CLEANUP); | |
bde89a9e | 1235 | connection->epochs--; |
b411b363 PR |
1236 | kfree(epoch); |
1237 | ||
1238 | if (rv == FE_STILL_LIVE) | |
1239 | rv = FE_DESTROYED; | |
1240 | } else { | |
1241 | epoch->flags = 0; | |
1242 | atomic_set(&epoch->epoch_size, 0); | |
698f9315 | 1243 | /* atomic_set(&epoch->active, 0); is already zero */ |
b411b363 PR |
1244 | if (rv == FE_STILL_LIVE) |
1245 | rv = FE_RECYCLED; | |
1246 | } | |
1247 | } | |
1248 | ||
1249 | if (!next_epoch) | |
1250 | break; | |
1251 | ||
1252 | epoch = next_epoch; | |
1253 | } while (1); | |
1254 | ||
bde89a9e | 1255 | spin_unlock(&connection->epoch_lock); |
b411b363 | 1256 | |
b411b363 PR |
1257 | return rv; |
1258 | } | |
1259 | ||
8fe39aac PR |
1260 | static enum write_ordering_e |
1261 | max_allowed_wo(struct drbd_backing_dev *bdev, enum write_ordering_e wo) | |
1262 | { | |
1263 | struct disk_conf *dc; | |
1264 | ||
1265 | dc = rcu_dereference(bdev->disk_conf); | |
1266 | ||
1267 | if (wo == WO_bdev_flush && !dc->disk_flushes) | |
1268 | wo = WO_drain_io; | |
1269 | if (wo == WO_drain_io && !dc->disk_drain) | |
1270 | wo = WO_none; | |
1271 | ||
1272 | return wo; | |
1273 | } | |
1274 | ||
b411b363 PR |
1275 | /** |
1276 | * drbd_bump_write_ordering() - Fall back to an other write ordering method | |
bde89a9e | 1277 | * @connection: DRBD connection. |
b411b363 PR |
1278 | * @wo: Write ordering method to try. |
1279 | */ | |
8fe39aac PR |
1280 | void drbd_bump_write_ordering(struct drbd_resource *resource, struct drbd_backing_dev *bdev, |
1281 | enum write_ordering_e wo) | |
b411b363 | 1282 | { |
e9526580 | 1283 | struct drbd_device *device; |
b411b363 | 1284 | enum write_ordering_e pwo; |
4b0007c0 | 1285 | int vnr; |
b411b363 PR |
1286 | static char *write_ordering_str[] = { |
1287 | [WO_none] = "none", | |
1288 | [WO_drain_io] = "drain", | |
1289 | [WO_bdev_flush] = "flush", | |
b411b363 PR |
1290 | }; |
1291 | ||
e9526580 | 1292 | pwo = resource->write_ordering; |
70df7092 LE |
1293 | if (wo != WO_bdev_flush) |
1294 | wo = min(pwo, wo); | |
daeda1cc | 1295 | rcu_read_lock(); |
e9526580 | 1296 | idr_for_each_entry(&resource->devices, device, vnr) { |
8fe39aac PR |
1297 | if (get_ldev(device)) { |
1298 | wo = max_allowed_wo(device->ldev, wo); | |
1299 | if (device->ldev == bdev) | |
1300 | bdev = NULL; | |
1301 | put_ldev(device); | |
1302 | } | |
4b0007c0 | 1303 | } |
8fe39aac PR |
1304 | |
1305 | if (bdev) | |
1306 | wo = max_allowed_wo(bdev, wo); | |
1307 | ||
70df7092 LE |
1308 | rcu_read_unlock(); |
1309 | ||
e9526580 PR |
1310 | resource->write_ordering = wo; |
1311 | if (pwo != resource->write_ordering || wo == WO_bdev_flush) | |
1312 | drbd_info(resource, "Method to ensure write ordering: %s\n", write_ordering_str[resource->write_ordering]); | |
b411b363 PR |
1313 | } |
1314 | ||
45bb912b | 1315 | /** |
fbe29dec | 1316 | * drbd_submit_peer_request() |
b30ab791 | 1317 | * @device: DRBD device. |
db830c46 | 1318 | * @peer_req: peer request |
45bb912b | 1319 | * @rw: flag field, see bio->bi_rw |
10f6d992 LE |
1320 | * |
1321 | * May spread the pages to multiple bios, | |
1322 | * depending on bio_add_page restrictions. | |
1323 | * | |
1324 | * Returns 0 if all bios have been submitted, | |
1325 | * -ENOMEM if we could not allocate enough bios, | |
1326 | * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a | |
1327 | * single page to an empty bio (which should never happen and likely indicates | |
1328 | * that the lower level IO stack is in some way broken). This has been observed | |
1329 | * on certain Xen deployments. | |
45bb912b LE |
1330 | */ |
1331 | /* TODO allocate from our own bio_set. */ | |
b30ab791 | 1332 | int drbd_submit_peer_request(struct drbd_device *device, |
fbe29dec AG |
1333 | struct drbd_peer_request *peer_req, |
1334 | const unsigned rw, const int fault_type) | |
45bb912b LE |
1335 | { |
1336 | struct bio *bios = NULL; | |
1337 | struct bio *bio; | |
db830c46 AG |
1338 | struct page *page = peer_req->pages; |
1339 | sector_t sector = peer_req->i.sector; | |
1340 | unsigned ds = peer_req->i.size; | |
45bb912b LE |
1341 | unsigned n_bios = 0; |
1342 | unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT; | |
10f6d992 | 1343 | int err = -ENOMEM; |
45bb912b | 1344 | |
a0fb3c47 LE |
1345 | if (peer_req->flags & EE_IS_TRIM_USE_ZEROOUT) { |
1346 | /* wait for all pending IO completions, before we start | |
1347 | * zeroing things out. */ | |
1348 | conn_wait_active_ee_empty(first_peer_device(device)->connection); | |
1349 | if (blkdev_issue_zeroout(device->ldev->backing_bdev, | |
1350 | sector, ds >> 9, GFP_NOIO)) | |
1351 | peer_req->flags |= EE_WAS_ERROR; | |
1352 | drbd_endio_write_sec_final(peer_req); | |
1353 | return 0; | |
1354 | } | |
1355 | ||
54ed4ed8 LE |
1356 | /* Discards don't have any payload. |
1357 | * But the scsi layer still expects a bio_vec it can use internally, | |
1358 | * see sd_setup_discard_cmnd() and blk_add_request_payload(). */ | |
a0fb3c47 | 1359 | if (peer_req->flags & EE_IS_TRIM) |
54ed4ed8 | 1360 | nr_pages = 1; |
a0fb3c47 | 1361 | |
45bb912b LE |
1362 | /* In most cases, we will only need one bio. But in case the lower |
1363 | * level restrictions happen to be different at this offset on this | |
1364 | * side than those of the sending peer, we may need to submit the | |
9476f39d LE |
1365 | * request in more than one bio. |
1366 | * | |
1367 | * Plain bio_alloc is good enough here, this is no DRBD internally | |
1368 | * generated bio, but a bio allocated on behalf of the peer. | |
1369 | */ | |
45bb912b LE |
1370 | next_bio: |
1371 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
1372 | if (!bio) { | |
a0fb3c47 | 1373 | drbd_err(device, "submit_ee: Allocation of a bio failed (nr_pages=%u)\n", nr_pages); |
45bb912b LE |
1374 | goto fail; |
1375 | } | |
db830c46 | 1376 | /* > peer_req->i.sector, unless this is the first bio */ |
4f024f37 | 1377 | bio->bi_iter.bi_sector = sector; |
b30ab791 | 1378 | bio->bi_bdev = device->ldev->backing_bdev; |
45bb912b | 1379 | bio->bi_rw = rw; |
db830c46 | 1380 | bio->bi_private = peer_req; |
fcefa62e | 1381 | bio->bi_end_io = drbd_peer_request_endio; |
45bb912b LE |
1382 | |
1383 | bio->bi_next = bios; | |
1384 | bios = bio; | |
1385 | ++n_bios; | |
1386 | ||
a0fb3c47 LE |
1387 | if (rw & REQ_DISCARD) { |
1388 | bio->bi_iter.bi_size = ds; | |
1389 | goto submit; | |
1390 | } | |
1391 | ||
45bb912b LE |
1392 | page_chain_for_each(page) { |
1393 | unsigned len = min_t(unsigned, ds, PAGE_SIZE); | |
1394 | if (!bio_add_page(bio, page, len, 0)) { | |
10f6d992 LE |
1395 | /* A single page must always be possible! |
1396 | * But in case it fails anyways, | |
1397 | * we deal with it, and complain (below). */ | |
1398 | if (bio->bi_vcnt == 0) { | |
d0180171 | 1399 | drbd_err(device, |
10f6d992 LE |
1400 | "bio_add_page failed for len=%u, " |
1401 | "bi_vcnt=0 (bi_sector=%llu)\n", | |
4f024f37 | 1402 | len, (uint64_t)bio->bi_iter.bi_sector); |
10f6d992 LE |
1403 | err = -ENOSPC; |
1404 | goto fail; | |
1405 | } | |
45bb912b LE |
1406 | goto next_bio; |
1407 | } | |
1408 | ds -= len; | |
1409 | sector += len >> 9; | |
1410 | --nr_pages; | |
1411 | } | |
0b0ba1ef | 1412 | D_ASSERT(device, ds == 0); |
a0fb3c47 LE |
1413 | submit: |
1414 | D_ASSERT(device, page == NULL); | |
45bb912b | 1415 | |
db830c46 | 1416 | atomic_set(&peer_req->pending_bios, n_bios); |
45bb912b LE |
1417 | do { |
1418 | bio = bios; | |
1419 | bios = bios->bi_next; | |
1420 | bio->bi_next = NULL; | |
1421 | ||
b30ab791 | 1422 | drbd_generic_make_request(device, fault_type, bio); |
45bb912b | 1423 | } while (bios); |
45bb912b LE |
1424 | return 0; |
1425 | ||
1426 | fail: | |
1427 | while (bios) { | |
1428 | bio = bios; | |
1429 | bios = bios->bi_next; | |
1430 | bio_put(bio); | |
1431 | } | |
10f6d992 | 1432 | return err; |
45bb912b LE |
1433 | } |
1434 | ||
b30ab791 | 1435 | static void drbd_remove_epoch_entry_interval(struct drbd_device *device, |
db830c46 | 1436 | struct drbd_peer_request *peer_req) |
53840641 | 1437 | { |
db830c46 | 1438 | struct drbd_interval *i = &peer_req->i; |
53840641 | 1439 | |
b30ab791 | 1440 | drbd_remove_interval(&device->write_requests, i); |
53840641 AG |
1441 | drbd_clear_interval(i); |
1442 | ||
6c852bec | 1443 | /* Wake up any processes waiting for this peer request to complete. */ |
53840641 | 1444 | if (i->waiting) |
b30ab791 | 1445 | wake_up(&device->misc_wait); |
53840641 AG |
1446 | } |
1447 | ||
bde89a9e | 1448 | static void conn_wait_active_ee_empty(struct drbd_connection *connection) |
77fede51 | 1449 | { |
c06ece6b | 1450 | struct drbd_peer_device *peer_device; |
77fede51 PR |
1451 | int vnr; |
1452 | ||
1453 | rcu_read_lock(); | |
c06ece6b AG |
1454 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
1455 | struct drbd_device *device = peer_device->device; | |
1456 | ||
b30ab791 | 1457 | kref_get(&device->kref); |
77fede51 | 1458 | rcu_read_unlock(); |
b30ab791 | 1459 | drbd_wait_ee_list_empty(device, &device->active_ee); |
05a10ec7 | 1460 | kref_put(&device->kref, drbd_destroy_device); |
77fede51 PR |
1461 | rcu_read_lock(); |
1462 | } | |
1463 | rcu_read_unlock(); | |
1464 | } | |
1465 | ||
9f4fe9ad AG |
1466 | static struct drbd_peer_device * |
1467 | conn_peer_device(struct drbd_connection *connection, int volume_number) | |
1468 | { | |
1469 | return idr_find(&connection->peer_devices, volume_number); | |
1470 | } | |
1471 | ||
bde89a9e | 1472 | static int receive_Barrier(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 1473 | { |
2451fc3b | 1474 | int rv; |
e658983a | 1475 | struct p_barrier *p = pi->data; |
b411b363 PR |
1476 | struct drbd_epoch *epoch; |
1477 | ||
9ed57dcb LE |
1478 | /* FIXME these are unacked on connection, |
1479 | * not a specific (peer)device. | |
1480 | */ | |
bde89a9e AG |
1481 | connection->current_epoch->barrier_nr = p->barrier; |
1482 | connection->current_epoch->connection = connection; | |
1483 | rv = drbd_may_finish_epoch(connection, connection->current_epoch, EV_GOT_BARRIER_NR); | |
b411b363 PR |
1484 | |
1485 | /* P_BARRIER_ACK may imply that the corresponding extent is dropped from | |
1486 | * the activity log, which means it would not be resynced in case the | |
1487 | * R_PRIMARY crashes now. | |
1488 | * Therefore we must send the barrier_ack after the barrier request was | |
1489 | * completed. */ | |
e9526580 | 1490 | switch (connection->resource->write_ordering) { |
b411b363 PR |
1491 | case WO_none: |
1492 | if (rv == FE_RECYCLED) | |
82bc0194 | 1493 | return 0; |
2451fc3b PR |
1494 | |
1495 | /* receiver context, in the writeout path of the other node. | |
1496 | * avoid potential distributed deadlock */ | |
1497 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1498 | if (epoch) | |
1499 | break; | |
1500 | else | |
1ec861eb | 1501 | drbd_warn(connection, "Allocation of an epoch failed, slowing down\n"); |
2451fc3b | 1502 | /* Fall through */ |
b411b363 PR |
1503 | |
1504 | case WO_bdev_flush: | |
1505 | case WO_drain_io: | |
bde89a9e AG |
1506 | conn_wait_active_ee_empty(connection); |
1507 | drbd_flush(connection); | |
2451fc3b | 1508 | |
bde89a9e | 1509 | if (atomic_read(&connection->current_epoch->epoch_size)) { |
2451fc3b PR |
1510 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); |
1511 | if (epoch) | |
1512 | break; | |
b411b363 PR |
1513 | } |
1514 | ||
82bc0194 | 1515 | return 0; |
2451fc3b | 1516 | default: |
e9526580 PR |
1517 | drbd_err(connection, "Strangeness in connection->write_ordering %d\n", |
1518 | connection->resource->write_ordering); | |
82bc0194 | 1519 | return -EIO; |
b411b363 PR |
1520 | } |
1521 | ||
1522 | epoch->flags = 0; | |
1523 | atomic_set(&epoch->epoch_size, 0); | |
1524 | atomic_set(&epoch->active, 0); | |
1525 | ||
bde89a9e AG |
1526 | spin_lock(&connection->epoch_lock); |
1527 | if (atomic_read(&connection->current_epoch->epoch_size)) { | |
1528 | list_add(&epoch->list, &connection->current_epoch->list); | |
1529 | connection->current_epoch = epoch; | |
1530 | connection->epochs++; | |
b411b363 PR |
1531 | } else { |
1532 | /* The current_epoch got recycled while we allocated this one... */ | |
1533 | kfree(epoch); | |
1534 | } | |
bde89a9e | 1535 | spin_unlock(&connection->epoch_lock); |
b411b363 | 1536 | |
82bc0194 | 1537 | return 0; |
b411b363 PR |
1538 | } |
1539 | ||
1540 | /* used from receive_RSDataReply (recv_resync_read) | |
1541 | * and from receive_Data */ | |
f6ffca9f | 1542 | static struct drbd_peer_request * |
69a22773 | 1543 | read_in_block(struct drbd_peer_device *peer_device, u64 id, sector_t sector, |
a0fb3c47 | 1544 | struct packet_info *pi) __must_hold(local) |
b411b363 | 1545 | { |
69a22773 | 1546 | struct drbd_device *device = peer_device->device; |
b30ab791 | 1547 | const sector_t capacity = drbd_get_capacity(device->this_bdev); |
db830c46 | 1548 | struct drbd_peer_request *peer_req; |
b411b363 | 1549 | struct page *page; |
a5c31904 | 1550 | int dgs, ds, err; |
a0fb3c47 | 1551 | int data_size = pi->size; |
69a22773 AG |
1552 | void *dig_in = peer_device->connection->int_dig_in; |
1553 | void *dig_vv = peer_device->connection->int_dig_vv; | |
6b4388ac | 1554 | unsigned long *data; |
a0fb3c47 | 1555 | struct p_trim *trim = (pi->cmd == P_TRIM) ? pi->data : NULL; |
b411b363 | 1556 | |
88104ca4 | 1557 | dgs = 0; |
a0fb3c47 | 1558 | if (!trim && peer_device->connection->peer_integrity_tfm) { |
69a22773 | 1559 | dgs = crypto_hash_digestsize(peer_device->connection->peer_integrity_tfm); |
9f5bdc33 AG |
1560 | /* |
1561 | * FIXME: Receive the incoming digest into the receive buffer | |
1562 | * here, together with its struct p_data? | |
1563 | */ | |
69a22773 | 1564 | err = drbd_recv_all_warn(peer_device->connection, dig_in, dgs); |
a5c31904 | 1565 | if (err) |
b411b363 | 1566 | return NULL; |
88104ca4 | 1567 | data_size -= dgs; |
b411b363 PR |
1568 | } |
1569 | ||
a0fb3c47 LE |
1570 | if (trim) { |
1571 | D_ASSERT(peer_device, data_size == 0); | |
1572 | data_size = be32_to_cpu(trim->size); | |
1573 | } | |
1574 | ||
841ce241 AG |
1575 | if (!expect(IS_ALIGNED(data_size, 512))) |
1576 | return NULL; | |
a0fb3c47 LE |
1577 | /* prepare for larger trim requests. */ |
1578 | if (!trim && !expect(data_size <= DRBD_MAX_BIO_SIZE)) | |
841ce241 | 1579 | return NULL; |
b411b363 | 1580 | |
6666032a LE |
1581 | /* even though we trust out peer, |
1582 | * we sometimes have to double check. */ | |
1583 | if (sector + (data_size>>9) > capacity) { | |
d0180171 | 1584 | drbd_err(device, "request from peer beyond end of local disk: " |
fdda6544 | 1585 | "capacity: %llus < sector: %llus + size: %u\n", |
6666032a LE |
1586 | (unsigned long long)capacity, |
1587 | (unsigned long long)sector, data_size); | |
1588 | return NULL; | |
1589 | } | |
1590 | ||
b411b363 PR |
1591 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD |
1592 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
1593 | * which in turn might block on the other node at this very place. */ | |
a0fb3c47 | 1594 | peer_req = drbd_alloc_peer_req(peer_device, id, sector, data_size, trim == NULL, GFP_NOIO); |
db830c46 | 1595 | if (!peer_req) |
b411b363 | 1596 | return NULL; |
45bb912b | 1597 | |
a0fb3c47 | 1598 | if (trim) |
81a3537a | 1599 | return peer_req; |
a73ff323 | 1600 | |
b411b363 | 1601 | ds = data_size; |
db830c46 | 1602 | page = peer_req->pages; |
45bb912b LE |
1603 | page_chain_for_each(page) { |
1604 | unsigned len = min_t(int, ds, PAGE_SIZE); | |
6b4388ac | 1605 | data = kmap(page); |
69a22773 | 1606 | err = drbd_recv_all_warn(peer_device->connection, data, len); |
b30ab791 | 1607 | if (drbd_insert_fault(device, DRBD_FAULT_RECEIVE)) { |
d0180171 | 1608 | drbd_err(device, "Fault injection: Corrupting data on receive\n"); |
6b4388ac PR |
1609 | data[0] = data[0] ^ (unsigned long)-1; |
1610 | } | |
b411b363 | 1611 | kunmap(page); |
a5c31904 | 1612 | if (err) { |
b30ab791 | 1613 | drbd_free_peer_req(device, peer_req); |
b411b363 PR |
1614 | return NULL; |
1615 | } | |
a5c31904 | 1616 | ds -= len; |
b411b363 PR |
1617 | } |
1618 | ||
1619 | if (dgs) { | |
69a22773 | 1620 | drbd_csum_ee(peer_device->connection->peer_integrity_tfm, peer_req, dig_vv); |
b411b363 | 1621 | if (memcmp(dig_in, dig_vv, dgs)) { |
d0180171 | 1622 | drbd_err(device, "Digest integrity check FAILED: %llus +%u\n", |
470be44a | 1623 | (unsigned long long)sector, data_size); |
b30ab791 | 1624 | drbd_free_peer_req(device, peer_req); |
b411b363 PR |
1625 | return NULL; |
1626 | } | |
1627 | } | |
b30ab791 | 1628 | device->recv_cnt += data_size>>9; |
db830c46 | 1629 | return peer_req; |
b411b363 PR |
1630 | } |
1631 | ||
1632 | /* drbd_drain_block() just takes a data block | |
1633 | * out of the socket input buffer, and discards it. | |
1634 | */ | |
69a22773 | 1635 | static int drbd_drain_block(struct drbd_peer_device *peer_device, int data_size) |
b411b363 PR |
1636 | { |
1637 | struct page *page; | |
a5c31904 | 1638 | int err = 0; |
b411b363 PR |
1639 | void *data; |
1640 | ||
c3470cde | 1641 | if (!data_size) |
fc5be839 | 1642 | return 0; |
c3470cde | 1643 | |
69a22773 | 1644 | page = drbd_alloc_pages(peer_device, 1, 1); |
b411b363 PR |
1645 | |
1646 | data = kmap(page); | |
1647 | while (data_size) { | |
fc5be839 AG |
1648 | unsigned int len = min_t(int, data_size, PAGE_SIZE); |
1649 | ||
69a22773 | 1650 | err = drbd_recv_all_warn(peer_device->connection, data, len); |
a5c31904 | 1651 | if (err) |
b411b363 | 1652 | break; |
a5c31904 | 1653 | data_size -= len; |
b411b363 PR |
1654 | } |
1655 | kunmap(page); | |
69a22773 | 1656 | drbd_free_pages(peer_device->device, page, 0); |
fc5be839 | 1657 | return err; |
b411b363 PR |
1658 | } |
1659 | ||
69a22773 | 1660 | static int recv_dless_read(struct drbd_peer_device *peer_device, struct drbd_request *req, |
b411b363 PR |
1661 | sector_t sector, int data_size) |
1662 | { | |
7988613b KO |
1663 | struct bio_vec bvec; |
1664 | struct bvec_iter iter; | |
b411b363 | 1665 | struct bio *bio; |
7988613b | 1666 | int dgs, err, expect; |
69a22773 AG |
1667 | void *dig_in = peer_device->connection->int_dig_in; |
1668 | void *dig_vv = peer_device->connection->int_dig_vv; | |
b411b363 | 1669 | |
88104ca4 | 1670 | dgs = 0; |
69a22773 AG |
1671 | if (peer_device->connection->peer_integrity_tfm) { |
1672 | dgs = crypto_hash_digestsize(peer_device->connection->peer_integrity_tfm); | |
1673 | err = drbd_recv_all_warn(peer_device->connection, dig_in, dgs); | |
a5c31904 AG |
1674 | if (err) |
1675 | return err; | |
88104ca4 | 1676 | data_size -= dgs; |
b411b363 PR |
1677 | } |
1678 | ||
b411b363 PR |
1679 | /* optimistically update recv_cnt. if receiving fails below, |
1680 | * we disconnect anyways, and counters will be reset. */ | |
69a22773 | 1681 | peer_device->device->recv_cnt += data_size>>9; |
b411b363 PR |
1682 | |
1683 | bio = req->master_bio; | |
69a22773 | 1684 | D_ASSERT(peer_device->device, sector == bio->bi_iter.bi_sector); |
b411b363 | 1685 | |
7988613b KO |
1686 | bio_for_each_segment(bvec, bio, iter) { |
1687 | void *mapped = kmap(bvec.bv_page) + bvec.bv_offset; | |
1688 | expect = min_t(int, data_size, bvec.bv_len); | |
69a22773 | 1689 | err = drbd_recv_all_warn(peer_device->connection, mapped, expect); |
7988613b | 1690 | kunmap(bvec.bv_page); |
a5c31904 AG |
1691 | if (err) |
1692 | return err; | |
1693 | data_size -= expect; | |
b411b363 PR |
1694 | } |
1695 | ||
1696 | if (dgs) { | |
69a22773 | 1697 | drbd_csum_bio(peer_device->connection->peer_integrity_tfm, bio, dig_vv); |
b411b363 | 1698 | if (memcmp(dig_in, dig_vv, dgs)) { |
69a22773 | 1699 | drbd_err(peer_device, "Digest integrity check FAILED. Broken NICs?\n"); |
28284cef | 1700 | return -EINVAL; |
b411b363 PR |
1701 | } |
1702 | } | |
1703 | ||
69a22773 | 1704 | D_ASSERT(peer_device->device, data_size == 0); |
28284cef | 1705 | return 0; |
b411b363 PR |
1706 | } |
1707 | ||
a990be46 AG |
1708 | /* |
1709 | * e_end_resync_block() is called in asender context via | |
1710 | * drbd_finish_peer_reqs(). | |
1711 | */ | |
99920dc5 | 1712 | static int e_end_resync_block(struct drbd_work *w, int unused) |
b411b363 | 1713 | { |
8050e6d0 | 1714 | struct drbd_peer_request *peer_req = |
a8cd15ba AG |
1715 | container_of(w, struct drbd_peer_request, w); |
1716 | struct drbd_peer_device *peer_device = peer_req->peer_device; | |
1717 | struct drbd_device *device = peer_device->device; | |
db830c46 | 1718 | sector_t sector = peer_req->i.sector; |
99920dc5 | 1719 | int err; |
b411b363 | 1720 | |
0b0ba1ef | 1721 | D_ASSERT(device, drbd_interval_empty(&peer_req->i)); |
b411b363 | 1722 | |
db830c46 | 1723 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b30ab791 | 1724 | drbd_set_in_sync(device, sector, peer_req->i.size); |
a8cd15ba | 1725 | err = drbd_send_ack(peer_device, P_RS_WRITE_ACK, peer_req); |
b411b363 PR |
1726 | } else { |
1727 | /* Record failure to sync */ | |
b30ab791 | 1728 | drbd_rs_failed_io(device, sector, peer_req->i.size); |
b411b363 | 1729 | |
a8cd15ba | 1730 | err = drbd_send_ack(peer_device, P_NEG_ACK, peer_req); |
b411b363 | 1731 | } |
b30ab791 | 1732 | dec_unacked(device); |
b411b363 | 1733 | |
99920dc5 | 1734 | return err; |
b411b363 PR |
1735 | } |
1736 | ||
69a22773 | 1737 | static int recv_resync_read(struct drbd_peer_device *peer_device, sector_t sector, |
a0fb3c47 | 1738 | struct packet_info *pi) __releases(local) |
b411b363 | 1739 | { |
69a22773 | 1740 | struct drbd_device *device = peer_device->device; |
db830c46 | 1741 | struct drbd_peer_request *peer_req; |
b411b363 | 1742 | |
a0fb3c47 | 1743 | peer_req = read_in_block(peer_device, ID_SYNCER, sector, pi); |
db830c46 | 1744 | if (!peer_req) |
45bb912b | 1745 | goto fail; |
b411b363 | 1746 | |
b30ab791 | 1747 | dec_rs_pending(device); |
b411b363 | 1748 | |
b30ab791 | 1749 | inc_unacked(device); |
b411b363 PR |
1750 | /* corresponding dec_unacked() in e_end_resync_block() |
1751 | * respective _drbd_clear_done_ee */ | |
1752 | ||
a8cd15ba | 1753 | peer_req->w.cb = e_end_resync_block; |
45bb912b | 1754 | |
0500813f | 1755 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 1756 | list_add(&peer_req->w.list, &device->sync_ee); |
0500813f | 1757 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 1758 | |
a0fb3c47 | 1759 | atomic_add(pi->size >> 9, &device->rs_sect_ev); |
b30ab791 | 1760 | if (drbd_submit_peer_request(device, peer_req, WRITE, DRBD_FAULT_RS_WR) == 0) |
e1c1b0fc | 1761 | return 0; |
b411b363 | 1762 | |
10f6d992 | 1763 | /* don't care for the reason here */ |
d0180171 | 1764 | drbd_err(device, "submit failed, triggering re-connect\n"); |
0500813f | 1765 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 1766 | list_del(&peer_req->w.list); |
0500813f | 1767 | spin_unlock_irq(&device->resource->req_lock); |
22cc37a9 | 1768 | |
b30ab791 | 1769 | drbd_free_peer_req(device, peer_req); |
45bb912b | 1770 | fail: |
b30ab791 | 1771 | put_ldev(device); |
e1c1b0fc | 1772 | return -EIO; |
b411b363 PR |
1773 | } |
1774 | ||
668eebc6 | 1775 | static struct drbd_request * |
b30ab791 | 1776 | find_request(struct drbd_device *device, struct rb_root *root, u64 id, |
bc9c5c41 | 1777 | sector_t sector, bool missing_ok, const char *func) |
51624585 | 1778 | { |
51624585 AG |
1779 | struct drbd_request *req; |
1780 | ||
bc9c5c41 AG |
1781 | /* Request object according to our peer */ |
1782 | req = (struct drbd_request *)(unsigned long)id; | |
5e472264 | 1783 | if (drbd_contains_interval(root, sector, &req->i) && req->i.local) |
668eebc6 | 1784 | return req; |
c3afd8f5 | 1785 | if (!missing_ok) { |
d0180171 | 1786 | drbd_err(device, "%s: failed to find request 0x%lx, sector %llus\n", func, |
c3afd8f5 AG |
1787 | (unsigned long)id, (unsigned long long)sector); |
1788 | } | |
51624585 | 1789 | return NULL; |
b411b363 PR |
1790 | } |
1791 | ||
bde89a9e | 1792 | static int receive_DataReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 1793 | { |
9f4fe9ad | 1794 | struct drbd_peer_device *peer_device; |
b30ab791 | 1795 | struct drbd_device *device; |
b411b363 PR |
1796 | struct drbd_request *req; |
1797 | sector_t sector; | |
82bc0194 | 1798 | int err; |
e658983a | 1799 | struct p_data *p = pi->data; |
4a76b161 | 1800 | |
9f4fe9ad AG |
1801 | peer_device = conn_peer_device(connection, pi->vnr); |
1802 | if (!peer_device) | |
4a76b161 | 1803 | return -EIO; |
9f4fe9ad | 1804 | device = peer_device->device; |
b411b363 PR |
1805 | |
1806 | sector = be64_to_cpu(p->sector); | |
1807 | ||
0500813f | 1808 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 1809 | req = find_request(device, &device->read_requests, p->block_id, sector, false, __func__); |
0500813f | 1810 | spin_unlock_irq(&device->resource->req_lock); |
c3afd8f5 | 1811 | if (unlikely(!req)) |
82bc0194 | 1812 | return -EIO; |
b411b363 | 1813 | |
24c4830c | 1814 | /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid |
b411b363 PR |
1815 | * special casing it there for the various failure cases. |
1816 | * still no race with drbd_fail_pending_reads */ | |
69a22773 | 1817 | err = recv_dless_read(peer_device, req, sector, pi->size); |
82bc0194 | 1818 | if (!err) |
8554df1c | 1819 | req_mod(req, DATA_RECEIVED); |
b411b363 PR |
1820 | /* else: nothing. handled from drbd_disconnect... |
1821 | * I don't think we may complete this just yet | |
1822 | * in case we are "on-disconnect: freeze" */ | |
1823 | ||
82bc0194 | 1824 | return err; |
b411b363 PR |
1825 | } |
1826 | ||
bde89a9e | 1827 | static int receive_RSDataReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 1828 | { |
9f4fe9ad | 1829 | struct drbd_peer_device *peer_device; |
b30ab791 | 1830 | struct drbd_device *device; |
b411b363 | 1831 | sector_t sector; |
82bc0194 | 1832 | int err; |
e658983a | 1833 | struct p_data *p = pi->data; |
4a76b161 | 1834 | |
9f4fe9ad AG |
1835 | peer_device = conn_peer_device(connection, pi->vnr); |
1836 | if (!peer_device) | |
4a76b161 | 1837 | return -EIO; |
9f4fe9ad | 1838 | device = peer_device->device; |
b411b363 PR |
1839 | |
1840 | sector = be64_to_cpu(p->sector); | |
0b0ba1ef | 1841 | D_ASSERT(device, p->block_id == ID_SYNCER); |
b411b363 | 1842 | |
b30ab791 | 1843 | if (get_ldev(device)) { |
b411b363 PR |
1844 | /* data is submitted to disk within recv_resync_read. |
1845 | * corresponding put_ldev done below on error, | |
fcefa62e | 1846 | * or in drbd_peer_request_endio. */ |
a0fb3c47 | 1847 | err = recv_resync_read(peer_device, sector, pi); |
b411b363 PR |
1848 | } else { |
1849 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1850 | drbd_err(device, "Can not write resync data to local disk.\n"); |
b411b363 | 1851 | |
69a22773 | 1852 | err = drbd_drain_block(peer_device, pi->size); |
b411b363 | 1853 | |
69a22773 | 1854 | drbd_send_ack_dp(peer_device, P_NEG_ACK, p, pi->size); |
b411b363 PR |
1855 | } |
1856 | ||
b30ab791 | 1857 | atomic_add(pi->size >> 9, &device->rs_sect_in); |
778f271d | 1858 | |
82bc0194 | 1859 | return err; |
b411b363 PR |
1860 | } |
1861 | ||
b30ab791 | 1862 | static void restart_conflicting_writes(struct drbd_device *device, |
7be8da07 | 1863 | sector_t sector, int size) |
b411b363 | 1864 | { |
7be8da07 AG |
1865 | struct drbd_interval *i; |
1866 | struct drbd_request *req; | |
1867 | ||
b30ab791 | 1868 | drbd_for_each_overlap(i, &device->write_requests, sector, size) { |
7be8da07 AG |
1869 | if (!i->local) |
1870 | continue; | |
1871 | req = container_of(i, struct drbd_request, i); | |
1872 | if (req->rq_state & RQ_LOCAL_PENDING || | |
1873 | !(req->rq_state & RQ_POSTPONED)) | |
1874 | continue; | |
2312f0b3 LE |
1875 | /* as it is RQ_POSTPONED, this will cause it to |
1876 | * be queued on the retry workqueue. */ | |
d4dabbe2 | 1877 | __req_mod(req, CONFLICT_RESOLVED, NULL); |
7be8da07 AG |
1878 | } |
1879 | } | |
b411b363 | 1880 | |
a990be46 AG |
1881 | /* |
1882 | * e_end_block() is called in asender context via drbd_finish_peer_reqs(). | |
b411b363 | 1883 | */ |
99920dc5 | 1884 | static int e_end_block(struct drbd_work *w, int cancel) |
b411b363 | 1885 | { |
8050e6d0 | 1886 | struct drbd_peer_request *peer_req = |
a8cd15ba AG |
1887 | container_of(w, struct drbd_peer_request, w); |
1888 | struct drbd_peer_device *peer_device = peer_req->peer_device; | |
1889 | struct drbd_device *device = peer_device->device; | |
db830c46 | 1890 | sector_t sector = peer_req->i.sector; |
99920dc5 | 1891 | int err = 0, pcmd; |
b411b363 | 1892 | |
303d1448 | 1893 | if (peer_req->flags & EE_SEND_WRITE_ACK) { |
db830c46 | 1894 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b30ab791 AG |
1895 | pcmd = (device->state.conn >= C_SYNC_SOURCE && |
1896 | device->state.conn <= C_PAUSED_SYNC_T && | |
db830c46 | 1897 | peer_req->flags & EE_MAY_SET_IN_SYNC) ? |
b411b363 | 1898 | P_RS_WRITE_ACK : P_WRITE_ACK; |
a8cd15ba | 1899 | err = drbd_send_ack(peer_device, pcmd, peer_req); |
b411b363 | 1900 | if (pcmd == P_RS_WRITE_ACK) |
b30ab791 | 1901 | drbd_set_in_sync(device, sector, peer_req->i.size); |
b411b363 | 1902 | } else { |
a8cd15ba | 1903 | err = drbd_send_ack(peer_device, P_NEG_ACK, peer_req); |
b411b363 PR |
1904 | /* we expect it to be marked out of sync anyways... |
1905 | * maybe assert this? */ | |
1906 | } | |
b30ab791 | 1907 | dec_unacked(device); |
b411b363 PR |
1908 | } |
1909 | /* we delete from the conflict detection hash _after_ we sent out the | |
1910 | * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ | |
302bdeae | 1911 | if (peer_req->flags & EE_IN_INTERVAL_TREE) { |
0500813f | 1912 | spin_lock_irq(&device->resource->req_lock); |
0b0ba1ef | 1913 | D_ASSERT(device, !drbd_interval_empty(&peer_req->i)); |
b30ab791 | 1914 | drbd_remove_epoch_entry_interval(device, peer_req); |
7be8da07 | 1915 | if (peer_req->flags & EE_RESTART_REQUESTS) |
b30ab791 | 1916 | restart_conflicting_writes(device, sector, peer_req->i.size); |
0500813f | 1917 | spin_unlock_irq(&device->resource->req_lock); |
bb3bfe96 | 1918 | } else |
0b0ba1ef | 1919 | D_ASSERT(device, drbd_interval_empty(&peer_req->i)); |
b411b363 | 1920 | |
a6b32bc3 | 1921 | drbd_may_finish_epoch(first_peer_device(device)->connection, peer_req->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); |
b411b363 | 1922 | |
99920dc5 | 1923 | return err; |
b411b363 PR |
1924 | } |
1925 | ||
a8cd15ba | 1926 | static int e_send_ack(struct drbd_work *w, enum drbd_packet ack) |
b411b363 | 1927 | { |
8050e6d0 | 1928 | struct drbd_peer_request *peer_req = |
a8cd15ba AG |
1929 | container_of(w, struct drbd_peer_request, w); |
1930 | struct drbd_peer_device *peer_device = peer_req->peer_device; | |
99920dc5 | 1931 | int err; |
b411b363 | 1932 | |
a8cd15ba AG |
1933 | err = drbd_send_ack(peer_device, ack, peer_req); |
1934 | dec_unacked(peer_device->device); | |
b411b363 | 1935 | |
99920dc5 | 1936 | return err; |
b411b363 PR |
1937 | } |
1938 | ||
d4dabbe2 | 1939 | static int e_send_superseded(struct drbd_work *w, int unused) |
7be8da07 | 1940 | { |
a8cd15ba | 1941 | return e_send_ack(w, P_SUPERSEDED); |
7be8da07 AG |
1942 | } |
1943 | ||
99920dc5 | 1944 | static int e_send_retry_write(struct drbd_work *w, int unused) |
7be8da07 | 1945 | { |
a8cd15ba AG |
1946 | struct drbd_peer_request *peer_req = |
1947 | container_of(w, struct drbd_peer_request, w); | |
1948 | struct drbd_connection *connection = peer_req->peer_device->connection; | |
7be8da07 | 1949 | |
a8cd15ba | 1950 | return e_send_ack(w, connection->agreed_pro_version >= 100 ? |
d4dabbe2 | 1951 | P_RETRY_WRITE : P_SUPERSEDED); |
7be8da07 | 1952 | } |
b411b363 | 1953 | |
3e394da1 AG |
1954 | static bool seq_greater(u32 a, u32 b) |
1955 | { | |
1956 | /* | |
1957 | * We assume 32-bit wrap-around here. | |
1958 | * For 24-bit wrap-around, we would have to shift: | |
1959 | * a <<= 8; b <<= 8; | |
1960 | */ | |
1961 | return (s32)a - (s32)b > 0; | |
1962 | } | |
b411b363 | 1963 | |
3e394da1 AG |
1964 | static u32 seq_max(u32 a, u32 b) |
1965 | { | |
1966 | return seq_greater(a, b) ? a : b; | |
b411b363 PR |
1967 | } |
1968 | ||
69a22773 | 1969 | static void update_peer_seq(struct drbd_peer_device *peer_device, unsigned int peer_seq) |
3e394da1 | 1970 | { |
69a22773 | 1971 | struct drbd_device *device = peer_device->device; |
3c13b680 | 1972 | unsigned int newest_peer_seq; |
3e394da1 | 1973 | |
69a22773 | 1974 | if (test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags)) { |
b30ab791 AG |
1975 | spin_lock(&device->peer_seq_lock); |
1976 | newest_peer_seq = seq_max(device->peer_seq, peer_seq); | |
1977 | device->peer_seq = newest_peer_seq; | |
1978 | spin_unlock(&device->peer_seq_lock); | |
1979 | /* wake up only if we actually changed device->peer_seq */ | |
3c13b680 | 1980 | if (peer_seq == newest_peer_seq) |
b30ab791 | 1981 | wake_up(&device->seq_wait); |
7be8da07 | 1982 | } |
b411b363 PR |
1983 | } |
1984 | ||
d93f6302 | 1985 | static inline int overlaps(sector_t s1, int l1, sector_t s2, int l2) |
b6a370ba | 1986 | { |
d93f6302 LE |
1987 | return !((s1 + (l1>>9) <= s2) || (s1 >= s2 + (l2>>9))); |
1988 | } | |
b6a370ba | 1989 | |
d93f6302 | 1990 | /* maybe change sync_ee into interval trees as well? */ |
b30ab791 | 1991 | static bool overlapping_resync_write(struct drbd_device *device, struct drbd_peer_request *peer_req) |
d93f6302 LE |
1992 | { |
1993 | struct drbd_peer_request *rs_req; | |
b6a370ba PR |
1994 | bool rv = 0; |
1995 | ||
0500813f | 1996 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 1997 | list_for_each_entry(rs_req, &device->sync_ee, w.list) { |
d93f6302 LE |
1998 | if (overlaps(peer_req->i.sector, peer_req->i.size, |
1999 | rs_req->i.sector, rs_req->i.size)) { | |
b6a370ba PR |
2000 | rv = 1; |
2001 | break; | |
2002 | } | |
2003 | } | |
0500813f | 2004 | spin_unlock_irq(&device->resource->req_lock); |
b6a370ba PR |
2005 | |
2006 | return rv; | |
2007 | } | |
2008 | ||
b411b363 PR |
2009 | /* Called from receive_Data. |
2010 | * Synchronize packets on sock with packets on msock. | |
2011 | * | |
2012 | * This is here so even when a P_DATA packet traveling via sock overtook an Ack | |
2013 | * packet traveling on msock, they are still processed in the order they have | |
2014 | * been sent. | |
2015 | * | |
2016 | * Note: we don't care for Ack packets overtaking P_DATA packets. | |
2017 | * | |
b30ab791 | 2018 | * In case packet_seq is larger than device->peer_seq number, there are |
b411b363 | 2019 | * outstanding packets on the msock. We wait for them to arrive. |
b30ab791 | 2020 | * In case we are the logically next packet, we update device->peer_seq |
b411b363 PR |
2021 | * ourselves. Correctly handles 32bit wrap around. |
2022 | * | |
2023 | * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, | |
2024 | * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds | |
2025 | * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have | |
2026 | * 1<<9 == 512 seconds aka ages for the 32bit wrap around... | |
2027 | * | |
2028 | * returns 0 if we may process the packet, | |
2029 | * -ERESTARTSYS if we were interrupted (by disconnect signal). */ | |
69a22773 | 2030 | static int wait_for_and_update_peer_seq(struct drbd_peer_device *peer_device, const u32 peer_seq) |
b411b363 | 2031 | { |
69a22773 | 2032 | struct drbd_device *device = peer_device->device; |
b411b363 | 2033 | DEFINE_WAIT(wait); |
b411b363 | 2034 | long timeout; |
b874d231 | 2035 | int ret = 0, tp; |
7be8da07 | 2036 | |
69a22773 | 2037 | if (!test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags)) |
7be8da07 AG |
2038 | return 0; |
2039 | ||
b30ab791 | 2040 | spin_lock(&device->peer_seq_lock); |
b411b363 | 2041 | for (;;) { |
b30ab791 AG |
2042 | if (!seq_greater(peer_seq - 1, device->peer_seq)) { |
2043 | device->peer_seq = seq_max(device->peer_seq, peer_seq); | |
b411b363 | 2044 | break; |
7be8da07 | 2045 | } |
b874d231 | 2046 | |
b411b363 PR |
2047 | if (signal_pending(current)) { |
2048 | ret = -ERESTARTSYS; | |
2049 | break; | |
2050 | } | |
b874d231 PR |
2051 | |
2052 | rcu_read_lock(); | |
a6b32bc3 | 2053 | tp = rcu_dereference(first_peer_device(device)->connection->net_conf)->two_primaries; |
b874d231 PR |
2054 | rcu_read_unlock(); |
2055 | ||
2056 | if (!tp) | |
2057 | break; | |
2058 | ||
2059 | /* Only need to wait if two_primaries is enabled */ | |
b30ab791 AG |
2060 | prepare_to_wait(&device->seq_wait, &wait, TASK_INTERRUPTIBLE); |
2061 | spin_unlock(&device->peer_seq_lock); | |
44ed167d | 2062 | rcu_read_lock(); |
69a22773 | 2063 | timeout = rcu_dereference(peer_device->connection->net_conf)->ping_timeo*HZ/10; |
44ed167d | 2064 | rcu_read_unlock(); |
71b1c1eb | 2065 | timeout = schedule_timeout(timeout); |
b30ab791 | 2066 | spin_lock(&device->peer_seq_lock); |
7be8da07 | 2067 | if (!timeout) { |
b411b363 | 2068 | ret = -ETIMEDOUT; |
d0180171 | 2069 | drbd_err(device, "Timed out waiting for missing ack packets; disconnecting\n"); |
b411b363 PR |
2070 | break; |
2071 | } | |
2072 | } | |
b30ab791 AG |
2073 | spin_unlock(&device->peer_seq_lock); |
2074 | finish_wait(&device->seq_wait, &wait); | |
b411b363 PR |
2075 | return ret; |
2076 | } | |
2077 | ||
688593c5 LE |
2078 | /* see also bio_flags_to_wire() |
2079 | * DRBD_REQ_*, because we need to semantically map the flags to data packet | |
2080 | * flags and back. We may replicate to other kernel versions. */ | |
81f0ffd2 | 2081 | static unsigned long wire_flags_to_bio(u32 dpf) |
76d2e7ec | 2082 | { |
688593c5 LE |
2083 | return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | |
2084 | (dpf & DP_FUA ? REQ_FUA : 0) | | |
2085 | (dpf & DP_FLUSH ? REQ_FLUSH : 0) | | |
2086 | (dpf & DP_DISCARD ? REQ_DISCARD : 0); | |
76d2e7ec PR |
2087 | } |
2088 | ||
b30ab791 | 2089 | static void fail_postponed_requests(struct drbd_device *device, sector_t sector, |
7be8da07 AG |
2090 | unsigned int size) |
2091 | { | |
2092 | struct drbd_interval *i; | |
2093 | ||
2094 | repeat: | |
b30ab791 | 2095 | drbd_for_each_overlap(i, &device->write_requests, sector, size) { |
7be8da07 AG |
2096 | struct drbd_request *req; |
2097 | struct bio_and_error m; | |
2098 | ||
2099 | if (!i->local) | |
2100 | continue; | |
2101 | req = container_of(i, struct drbd_request, i); | |
2102 | if (!(req->rq_state & RQ_POSTPONED)) | |
2103 | continue; | |
2104 | req->rq_state &= ~RQ_POSTPONED; | |
2105 | __req_mod(req, NEG_ACKED, &m); | |
0500813f | 2106 | spin_unlock_irq(&device->resource->req_lock); |
7be8da07 | 2107 | if (m.bio) |
b30ab791 | 2108 | complete_master_bio(device, &m); |
0500813f | 2109 | spin_lock_irq(&device->resource->req_lock); |
7be8da07 AG |
2110 | goto repeat; |
2111 | } | |
2112 | } | |
2113 | ||
b30ab791 | 2114 | static int handle_write_conflicts(struct drbd_device *device, |
7be8da07 AG |
2115 | struct drbd_peer_request *peer_req) |
2116 | { | |
e33b32de | 2117 | struct drbd_connection *connection = peer_req->peer_device->connection; |
bde89a9e | 2118 | bool resolve_conflicts = test_bit(RESOLVE_CONFLICTS, &connection->flags); |
7be8da07 AG |
2119 | sector_t sector = peer_req->i.sector; |
2120 | const unsigned int size = peer_req->i.size; | |
2121 | struct drbd_interval *i; | |
2122 | bool equal; | |
2123 | int err; | |
2124 | ||
2125 | /* | |
2126 | * Inserting the peer request into the write_requests tree will prevent | |
2127 | * new conflicting local requests from being added. | |
2128 | */ | |
b30ab791 | 2129 | drbd_insert_interval(&device->write_requests, &peer_req->i); |
7be8da07 AG |
2130 | |
2131 | repeat: | |
b30ab791 | 2132 | drbd_for_each_overlap(i, &device->write_requests, sector, size) { |
7be8da07 AG |
2133 | if (i == &peer_req->i) |
2134 | continue; | |
2135 | ||
2136 | if (!i->local) { | |
2137 | /* | |
2138 | * Our peer has sent a conflicting remote request; this | |
2139 | * should not happen in a two-node setup. Wait for the | |
2140 | * earlier peer request to complete. | |
2141 | */ | |
b30ab791 | 2142 | err = drbd_wait_misc(device, i); |
7be8da07 AG |
2143 | if (err) |
2144 | goto out; | |
2145 | goto repeat; | |
2146 | } | |
2147 | ||
2148 | equal = i->sector == sector && i->size == size; | |
2149 | if (resolve_conflicts) { | |
2150 | /* | |
2151 | * If the peer request is fully contained within the | |
d4dabbe2 LE |
2152 | * overlapping request, it can be considered overwritten |
2153 | * and thus superseded; otherwise, it will be retried | |
2154 | * once all overlapping requests have completed. | |
7be8da07 | 2155 | */ |
d4dabbe2 | 2156 | bool superseded = i->sector <= sector && i->sector + |
7be8da07 AG |
2157 | (i->size >> 9) >= sector + (size >> 9); |
2158 | ||
2159 | if (!equal) | |
d0180171 | 2160 | drbd_alert(device, "Concurrent writes detected: " |
7be8da07 AG |
2161 | "local=%llus +%u, remote=%llus +%u, " |
2162 | "assuming %s came first\n", | |
2163 | (unsigned long long)i->sector, i->size, | |
2164 | (unsigned long long)sector, size, | |
d4dabbe2 | 2165 | superseded ? "local" : "remote"); |
7be8da07 | 2166 | |
b30ab791 | 2167 | inc_unacked(device); |
a8cd15ba | 2168 | peer_req->w.cb = superseded ? e_send_superseded : |
7be8da07 | 2169 | e_send_retry_write; |
a8cd15ba | 2170 | list_add_tail(&peer_req->w.list, &device->done_ee); |
e33b32de | 2171 | wake_asender(connection); |
7be8da07 AG |
2172 | |
2173 | err = -ENOENT; | |
2174 | goto out; | |
2175 | } else { | |
2176 | struct drbd_request *req = | |
2177 | container_of(i, struct drbd_request, i); | |
2178 | ||
2179 | if (!equal) | |
d0180171 | 2180 | drbd_alert(device, "Concurrent writes detected: " |
7be8da07 AG |
2181 | "local=%llus +%u, remote=%llus +%u\n", |
2182 | (unsigned long long)i->sector, i->size, | |
2183 | (unsigned long long)sector, size); | |
2184 | ||
2185 | if (req->rq_state & RQ_LOCAL_PENDING || | |
2186 | !(req->rq_state & RQ_POSTPONED)) { | |
2187 | /* | |
2188 | * Wait for the node with the discard flag to | |
d4dabbe2 LE |
2189 | * decide if this request has been superseded |
2190 | * or needs to be retried. | |
2191 | * Requests that have been superseded will | |
7be8da07 AG |
2192 | * disappear from the write_requests tree. |
2193 | * | |
2194 | * In addition, wait for the conflicting | |
2195 | * request to finish locally before submitting | |
2196 | * the conflicting peer request. | |
2197 | */ | |
b30ab791 | 2198 | err = drbd_wait_misc(device, &req->i); |
7be8da07 | 2199 | if (err) { |
e33b32de | 2200 | _conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD); |
b30ab791 | 2201 | fail_postponed_requests(device, sector, size); |
7be8da07 AG |
2202 | goto out; |
2203 | } | |
2204 | goto repeat; | |
2205 | } | |
2206 | /* | |
2207 | * Remember to restart the conflicting requests after | |
2208 | * the new peer request has completed. | |
2209 | */ | |
2210 | peer_req->flags |= EE_RESTART_REQUESTS; | |
2211 | } | |
2212 | } | |
2213 | err = 0; | |
2214 | ||
2215 | out: | |
2216 | if (err) | |
b30ab791 | 2217 | drbd_remove_epoch_entry_interval(device, peer_req); |
7be8da07 AG |
2218 | return err; |
2219 | } | |
2220 | ||
b411b363 | 2221 | /* mirrored write */ |
bde89a9e | 2222 | static int receive_Data(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 2223 | { |
9f4fe9ad | 2224 | struct drbd_peer_device *peer_device; |
b30ab791 | 2225 | struct drbd_device *device; |
b411b363 | 2226 | sector_t sector; |
db830c46 | 2227 | struct drbd_peer_request *peer_req; |
e658983a | 2228 | struct p_data *p = pi->data; |
7be8da07 | 2229 | u32 peer_seq = be32_to_cpu(p->seq_num); |
b411b363 PR |
2230 | int rw = WRITE; |
2231 | u32 dp_flags; | |
302bdeae | 2232 | int err, tp; |
b411b363 | 2233 | |
9f4fe9ad AG |
2234 | peer_device = conn_peer_device(connection, pi->vnr); |
2235 | if (!peer_device) | |
4a76b161 | 2236 | return -EIO; |
9f4fe9ad | 2237 | device = peer_device->device; |
b411b363 | 2238 | |
b30ab791 | 2239 | if (!get_ldev(device)) { |
82bc0194 AG |
2240 | int err2; |
2241 | ||
69a22773 AG |
2242 | err = wait_for_and_update_peer_seq(peer_device, peer_seq); |
2243 | drbd_send_ack_dp(peer_device, P_NEG_ACK, p, pi->size); | |
bde89a9e | 2244 | atomic_inc(&connection->current_epoch->epoch_size); |
69a22773 | 2245 | err2 = drbd_drain_block(peer_device, pi->size); |
82bc0194 AG |
2246 | if (!err) |
2247 | err = err2; | |
2248 | return err; | |
b411b363 PR |
2249 | } |
2250 | ||
fcefa62e AG |
2251 | /* |
2252 | * Corresponding put_ldev done either below (on various errors), or in | |
2253 | * drbd_peer_request_endio, if we successfully submit the data at the | |
2254 | * end of this function. | |
2255 | */ | |
b411b363 PR |
2256 | |
2257 | sector = be64_to_cpu(p->sector); | |
a0fb3c47 | 2258 | peer_req = read_in_block(peer_device, p->block_id, sector, pi); |
db830c46 | 2259 | if (!peer_req) { |
b30ab791 | 2260 | put_ldev(device); |
82bc0194 | 2261 | return -EIO; |
b411b363 PR |
2262 | } |
2263 | ||
a8cd15ba | 2264 | peer_req->w.cb = e_end_block; |
b411b363 | 2265 | |
688593c5 | 2266 | dp_flags = be32_to_cpu(p->dp_flags); |
81f0ffd2 | 2267 | rw |= wire_flags_to_bio(dp_flags); |
a0fb3c47 LE |
2268 | if (pi->cmd == P_TRIM) { |
2269 | struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev); | |
2270 | peer_req->flags |= EE_IS_TRIM; | |
2271 | if (!blk_queue_discard(q)) | |
2272 | peer_req->flags |= EE_IS_TRIM_USE_ZEROOUT; | |
2273 | D_ASSERT(peer_device, peer_req->i.size > 0); | |
2274 | D_ASSERT(peer_device, rw & REQ_DISCARD); | |
2275 | D_ASSERT(peer_device, peer_req->pages == NULL); | |
2276 | } else if (peer_req->pages == NULL) { | |
0b0ba1ef AG |
2277 | D_ASSERT(device, peer_req->i.size == 0); |
2278 | D_ASSERT(device, dp_flags & DP_FLUSH); | |
a73ff323 | 2279 | } |
688593c5 LE |
2280 | |
2281 | if (dp_flags & DP_MAY_SET_IN_SYNC) | |
db830c46 | 2282 | peer_req->flags |= EE_MAY_SET_IN_SYNC; |
688593c5 | 2283 | |
bde89a9e AG |
2284 | spin_lock(&connection->epoch_lock); |
2285 | peer_req->epoch = connection->current_epoch; | |
db830c46 AG |
2286 | atomic_inc(&peer_req->epoch->epoch_size); |
2287 | atomic_inc(&peer_req->epoch->active); | |
bde89a9e | 2288 | spin_unlock(&connection->epoch_lock); |
b411b363 | 2289 | |
302bdeae | 2290 | rcu_read_lock(); |
9f4fe9ad | 2291 | tp = rcu_dereference(peer_device->connection->net_conf)->two_primaries; |
302bdeae PR |
2292 | rcu_read_unlock(); |
2293 | if (tp) { | |
2294 | peer_req->flags |= EE_IN_INTERVAL_TREE; | |
69a22773 | 2295 | err = wait_for_and_update_peer_seq(peer_device, peer_seq); |
7be8da07 | 2296 | if (err) |
b411b363 | 2297 | goto out_interrupted; |
0500813f | 2298 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 2299 | err = handle_write_conflicts(device, peer_req); |
7be8da07 | 2300 | if (err) { |
0500813f | 2301 | spin_unlock_irq(&device->resource->req_lock); |
7be8da07 | 2302 | if (err == -ENOENT) { |
b30ab791 | 2303 | put_ldev(device); |
82bc0194 | 2304 | return 0; |
b411b363 | 2305 | } |
7be8da07 | 2306 | goto out_interrupted; |
b411b363 | 2307 | } |
b874d231 | 2308 | } else { |
69a22773 | 2309 | update_peer_seq(peer_device, peer_seq); |
0500813f | 2310 | spin_lock_irq(&device->resource->req_lock); |
b874d231 | 2311 | } |
a0fb3c47 LE |
2312 | /* if we use the zeroout fallback code, we process synchronously |
2313 | * and we wait for all pending requests, respectively wait for | |
2314 | * active_ee to become empty in drbd_submit_peer_request(); | |
2315 | * better not add ourselves here. */ | |
2316 | if ((peer_req->flags & EE_IS_TRIM_USE_ZEROOUT) == 0) | |
2317 | list_add(&peer_req->w.list, &device->active_ee); | |
0500813f | 2318 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 2319 | |
b30ab791 AG |
2320 | if (device->state.conn == C_SYNC_TARGET) |
2321 | wait_event(device->ee_wait, !overlapping_resync_write(device, peer_req)); | |
b411b363 | 2322 | |
9f4fe9ad | 2323 | if (peer_device->connection->agreed_pro_version < 100) { |
44ed167d | 2324 | rcu_read_lock(); |
9f4fe9ad | 2325 | switch (rcu_dereference(peer_device->connection->net_conf)->wire_protocol) { |
303d1448 PR |
2326 | case DRBD_PROT_C: |
2327 | dp_flags |= DP_SEND_WRITE_ACK; | |
2328 | break; | |
2329 | case DRBD_PROT_B: | |
2330 | dp_flags |= DP_SEND_RECEIVE_ACK; | |
2331 | break; | |
b411b363 | 2332 | } |
44ed167d | 2333 | rcu_read_unlock(); |
b411b363 PR |
2334 | } |
2335 | ||
303d1448 PR |
2336 | if (dp_flags & DP_SEND_WRITE_ACK) { |
2337 | peer_req->flags |= EE_SEND_WRITE_ACK; | |
b30ab791 | 2338 | inc_unacked(device); |
b411b363 PR |
2339 | /* corresponding dec_unacked() in e_end_block() |
2340 | * respective _drbd_clear_done_ee */ | |
303d1448 PR |
2341 | } |
2342 | ||
2343 | if (dp_flags & DP_SEND_RECEIVE_ACK) { | |
b411b363 PR |
2344 | /* I really don't like it that the receiver thread |
2345 | * sends on the msock, but anyways */ | |
69a22773 | 2346 | drbd_send_ack(first_peer_device(device), P_RECV_ACK, peer_req); |
b411b363 PR |
2347 | } |
2348 | ||
b30ab791 | 2349 | if (device->state.pdsk < D_INCONSISTENT) { |
b411b363 | 2350 | /* In case we have the only disk of the cluster, */ |
b30ab791 | 2351 | drbd_set_out_of_sync(device, peer_req->i.sector, peer_req->i.size); |
db830c46 AG |
2352 | peer_req->flags |= EE_CALL_AL_COMPLETE_IO; |
2353 | peer_req->flags &= ~EE_MAY_SET_IN_SYNC; | |
4dd726f0 | 2354 | drbd_al_begin_io(device, &peer_req->i); |
b411b363 PR |
2355 | } |
2356 | ||
b30ab791 | 2357 | err = drbd_submit_peer_request(device, peer_req, rw, DRBD_FAULT_DT_WR); |
82bc0194 AG |
2358 | if (!err) |
2359 | return 0; | |
b411b363 | 2360 | |
10f6d992 | 2361 | /* don't care for the reason here */ |
d0180171 | 2362 | drbd_err(device, "submit failed, triggering re-connect\n"); |
0500813f | 2363 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 2364 | list_del(&peer_req->w.list); |
b30ab791 | 2365 | drbd_remove_epoch_entry_interval(device, peer_req); |
0500813f | 2366 | spin_unlock_irq(&device->resource->req_lock); |
db830c46 | 2367 | if (peer_req->flags & EE_CALL_AL_COMPLETE_IO) |
b30ab791 | 2368 | drbd_al_complete_io(device, &peer_req->i); |
22cc37a9 | 2369 | |
b411b363 | 2370 | out_interrupted: |
bde89a9e | 2371 | drbd_may_finish_epoch(connection, peer_req->epoch, EV_PUT + EV_CLEANUP); |
b30ab791 AG |
2372 | put_ldev(device); |
2373 | drbd_free_peer_req(device, peer_req); | |
82bc0194 | 2374 | return err; |
b411b363 PR |
2375 | } |
2376 | ||
0f0601f4 LE |
2377 | /* We may throttle resync, if the lower device seems to be busy, |
2378 | * and current sync rate is above c_min_rate. | |
2379 | * | |
2380 | * To decide whether or not the lower device is busy, we use a scheme similar | |
2381 | * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" | |
2382 | * (more than 64 sectors) of activity we cannot account for with our own resync | |
2383 | * activity, it obviously is "busy". | |
2384 | * | |
2385 | * The current sync rate used here uses only the most recent two step marks, | |
2386 | * to have a short time average so we can react faster. | |
2387 | */ | |
e8299874 | 2388 | bool drbd_rs_should_slow_down(struct drbd_device *device, sector_t sector) |
0f0601f4 | 2389 | { |
e3555d85 | 2390 | struct lc_element *tmp; |
e8299874 | 2391 | bool throttle = true; |
daeda1cc | 2392 | |
e8299874 LE |
2393 | if (!drbd_rs_c_min_rate_throttle(device)) |
2394 | return false; | |
0f0601f4 | 2395 | |
b30ab791 AG |
2396 | spin_lock_irq(&device->al_lock); |
2397 | tmp = lc_find(device->resync, BM_SECT_TO_EXT(sector)); | |
e3555d85 PR |
2398 | if (tmp) { |
2399 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
e8299874 LE |
2400 | if (test_bit(BME_PRIORITY, &bm_ext->flags)) |
2401 | throttle = false; | |
e3555d85 PR |
2402 | /* Do not slow down if app IO is already waiting for this extent */ |
2403 | } | |
b30ab791 | 2404 | spin_unlock_irq(&device->al_lock); |
e3555d85 | 2405 | |
e8299874 LE |
2406 | return throttle; |
2407 | } | |
2408 | ||
2409 | bool drbd_rs_c_min_rate_throttle(struct drbd_device *device) | |
2410 | { | |
2411 | struct gendisk *disk = device->ldev->backing_bdev->bd_contains->bd_disk; | |
2412 | unsigned long db, dt, dbdt; | |
2413 | unsigned int c_min_rate; | |
2414 | int curr_events; | |
2415 | ||
2416 | rcu_read_lock(); | |
2417 | c_min_rate = rcu_dereference(device->ldev->disk_conf)->c_min_rate; | |
2418 | rcu_read_unlock(); | |
2419 | ||
2420 | /* feature disabled? */ | |
2421 | if (c_min_rate == 0) | |
2422 | return false; | |
2423 | ||
0f0601f4 LE |
2424 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
2425 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
b30ab791 | 2426 | atomic_read(&device->rs_sect_ev); |
b30ab791 | 2427 | if (!device->rs_last_events || curr_events - device->rs_last_events > 64) { |
0f0601f4 LE |
2428 | unsigned long rs_left; |
2429 | int i; | |
2430 | ||
b30ab791 | 2431 | device->rs_last_events = curr_events; |
0f0601f4 LE |
2432 | |
2433 | /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, | |
2434 | * approx. */ | |
b30ab791 | 2435 | i = (device->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS; |
2649f080 | 2436 | |
b30ab791 AG |
2437 | if (device->state.conn == C_VERIFY_S || device->state.conn == C_VERIFY_T) |
2438 | rs_left = device->ov_left; | |
2649f080 | 2439 | else |
b30ab791 | 2440 | rs_left = drbd_bm_total_weight(device) - device->rs_failed; |
0f0601f4 | 2441 | |
b30ab791 | 2442 | dt = ((long)jiffies - (long)device->rs_mark_time[i]) / HZ; |
0f0601f4 LE |
2443 | if (!dt) |
2444 | dt++; | |
b30ab791 | 2445 | db = device->rs_mark_left[i] - rs_left; |
0f0601f4 LE |
2446 | dbdt = Bit2KB(db/dt); |
2447 | ||
daeda1cc | 2448 | if (dbdt > c_min_rate) |
e8299874 | 2449 | return true; |
0f0601f4 | 2450 | } |
e8299874 | 2451 | return false; |
0f0601f4 LE |
2452 | } |
2453 | ||
bde89a9e | 2454 | static int receive_DataRequest(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 2455 | { |
9f4fe9ad | 2456 | struct drbd_peer_device *peer_device; |
b30ab791 | 2457 | struct drbd_device *device; |
b411b363 | 2458 | sector_t sector; |
4a76b161 | 2459 | sector_t capacity; |
db830c46 | 2460 | struct drbd_peer_request *peer_req; |
b411b363 | 2461 | struct digest_info *di = NULL; |
b18b37be | 2462 | int size, verb; |
b411b363 | 2463 | unsigned int fault_type; |
e658983a | 2464 | struct p_block_req *p = pi->data; |
4a76b161 | 2465 | |
9f4fe9ad AG |
2466 | peer_device = conn_peer_device(connection, pi->vnr); |
2467 | if (!peer_device) | |
4a76b161 | 2468 | return -EIO; |
9f4fe9ad | 2469 | device = peer_device->device; |
b30ab791 | 2470 | capacity = drbd_get_capacity(device->this_bdev); |
b411b363 PR |
2471 | |
2472 | sector = be64_to_cpu(p->sector); | |
2473 | size = be32_to_cpu(p->blksize); | |
2474 | ||
c670a398 | 2475 | if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { |
d0180171 | 2476 | drbd_err(device, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, |
b411b363 | 2477 | (unsigned long long)sector, size); |
82bc0194 | 2478 | return -EINVAL; |
b411b363 PR |
2479 | } |
2480 | if (sector + (size>>9) > capacity) { | |
d0180171 | 2481 | drbd_err(device, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, |
b411b363 | 2482 | (unsigned long long)sector, size); |
82bc0194 | 2483 | return -EINVAL; |
b411b363 PR |
2484 | } |
2485 | ||
b30ab791 | 2486 | if (!get_ldev_if_state(device, D_UP_TO_DATE)) { |
b18b37be | 2487 | verb = 1; |
e2857216 | 2488 | switch (pi->cmd) { |
b18b37be | 2489 | case P_DATA_REQUEST: |
69a22773 | 2490 | drbd_send_ack_rp(peer_device, P_NEG_DREPLY, p); |
b18b37be PR |
2491 | break; |
2492 | case P_RS_DATA_REQUEST: | |
2493 | case P_CSUM_RS_REQUEST: | |
2494 | case P_OV_REQUEST: | |
69a22773 | 2495 | drbd_send_ack_rp(peer_device, P_NEG_RS_DREPLY , p); |
b18b37be PR |
2496 | break; |
2497 | case P_OV_REPLY: | |
2498 | verb = 0; | |
b30ab791 | 2499 | dec_rs_pending(device); |
69a22773 | 2500 | drbd_send_ack_ex(peer_device, P_OV_RESULT, sector, size, ID_IN_SYNC); |
b18b37be PR |
2501 | break; |
2502 | default: | |
49ba9b1b | 2503 | BUG(); |
b18b37be PR |
2504 | } |
2505 | if (verb && __ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 2506 | drbd_err(device, "Can not satisfy peer's read request, " |
b411b363 | 2507 | "no local data.\n"); |
b18b37be | 2508 | |
a821cc4a | 2509 | /* drain possibly payload */ |
69a22773 | 2510 | return drbd_drain_block(peer_device, pi->size); |
b411b363 PR |
2511 | } |
2512 | ||
2513 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD | |
2514 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
2515 | * which in turn might block on the other node at this very place. */ | |
a0fb3c47 LE |
2516 | peer_req = drbd_alloc_peer_req(peer_device, p->block_id, sector, size, |
2517 | true /* has real payload */, GFP_NOIO); | |
db830c46 | 2518 | if (!peer_req) { |
b30ab791 | 2519 | put_ldev(device); |
82bc0194 | 2520 | return -ENOMEM; |
b411b363 PR |
2521 | } |
2522 | ||
e2857216 | 2523 | switch (pi->cmd) { |
b411b363 | 2524 | case P_DATA_REQUEST: |
a8cd15ba | 2525 | peer_req->w.cb = w_e_end_data_req; |
b411b363 | 2526 | fault_type = DRBD_FAULT_DT_RD; |
80a40e43 LE |
2527 | /* application IO, don't drbd_rs_begin_io */ |
2528 | goto submit; | |
2529 | ||
b411b363 | 2530 | case P_RS_DATA_REQUEST: |
a8cd15ba | 2531 | peer_req->w.cb = w_e_end_rsdata_req; |
b411b363 | 2532 | fault_type = DRBD_FAULT_RS_RD; |
5f9915bb | 2533 | /* used in the sector offset progress display */ |
b30ab791 | 2534 | device->bm_resync_fo = BM_SECT_TO_BIT(sector); |
b411b363 PR |
2535 | break; |
2536 | ||
2537 | case P_OV_REPLY: | |
2538 | case P_CSUM_RS_REQUEST: | |
2539 | fault_type = DRBD_FAULT_RS_RD; | |
e2857216 | 2540 | di = kmalloc(sizeof(*di) + pi->size, GFP_NOIO); |
b411b363 PR |
2541 | if (!di) |
2542 | goto out_free_e; | |
2543 | ||
e2857216 | 2544 | di->digest_size = pi->size; |
b411b363 PR |
2545 | di->digest = (((char *)di)+sizeof(struct digest_info)); |
2546 | ||
db830c46 AG |
2547 | peer_req->digest = di; |
2548 | peer_req->flags |= EE_HAS_DIGEST; | |
c36c3ced | 2549 | |
9f4fe9ad | 2550 | if (drbd_recv_all(peer_device->connection, di->digest, pi->size)) |
b411b363 PR |
2551 | goto out_free_e; |
2552 | ||
e2857216 | 2553 | if (pi->cmd == P_CSUM_RS_REQUEST) { |
9f4fe9ad | 2554 | D_ASSERT(device, peer_device->connection->agreed_pro_version >= 89); |
a8cd15ba | 2555 | peer_req->w.cb = w_e_end_csum_rs_req; |
5f9915bb | 2556 | /* used in the sector offset progress display */ |
b30ab791 | 2557 | device->bm_resync_fo = BM_SECT_TO_BIT(sector); |
e2857216 | 2558 | } else if (pi->cmd == P_OV_REPLY) { |
2649f080 | 2559 | /* track progress, we may need to throttle */ |
b30ab791 | 2560 | atomic_add(size >> 9, &device->rs_sect_in); |
a8cd15ba | 2561 | peer_req->w.cb = w_e_end_ov_reply; |
b30ab791 | 2562 | dec_rs_pending(device); |
0f0601f4 LE |
2563 | /* drbd_rs_begin_io done when we sent this request, |
2564 | * but accounting still needs to be done. */ | |
2565 | goto submit_for_resync; | |
b411b363 PR |
2566 | } |
2567 | break; | |
2568 | ||
2569 | case P_OV_REQUEST: | |
b30ab791 | 2570 | if (device->ov_start_sector == ~(sector_t)0 && |
9f4fe9ad | 2571 | peer_device->connection->agreed_pro_version >= 90) { |
de228bba LE |
2572 | unsigned long now = jiffies; |
2573 | int i; | |
b30ab791 AG |
2574 | device->ov_start_sector = sector; |
2575 | device->ov_position = sector; | |
2576 | device->ov_left = drbd_bm_bits(device) - BM_SECT_TO_BIT(sector); | |
2577 | device->rs_total = device->ov_left; | |
de228bba | 2578 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
2579 | device->rs_mark_left[i] = device->ov_left; |
2580 | device->rs_mark_time[i] = now; | |
de228bba | 2581 | } |
d0180171 | 2582 | drbd_info(device, "Online Verify start sector: %llu\n", |
b411b363 PR |
2583 | (unsigned long long)sector); |
2584 | } | |
a8cd15ba | 2585 | peer_req->w.cb = w_e_end_ov_req; |
b411b363 | 2586 | fault_type = DRBD_FAULT_RS_RD; |
b411b363 PR |
2587 | break; |
2588 | ||
b411b363 | 2589 | default: |
49ba9b1b | 2590 | BUG(); |
b411b363 PR |
2591 | } |
2592 | ||
0f0601f4 LE |
2593 | /* Throttle, drbd_rs_begin_io and submit should become asynchronous |
2594 | * wrt the receiver, but it is not as straightforward as it may seem. | |
2595 | * Various places in the resync start and stop logic assume resync | |
2596 | * requests are processed in order, requeuing this on the worker thread | |
2597 | * introduces a bunch of new code for synchronization between threads. | |
2598 | * | |
2599 | * Unlimited throttling before drbd_rs_begin_io may stall the resync | |
2600 | * "forever", throttling after drbd_rs_begin_io will lock that extent | |
2601 | * for application writes for the same time. For now, just throttle | |
2602 | * here, where the rest of the code expects the receiver to sleep for | |
2603 | * a while, anyways. | |
2604 | */ | |
2605 | ||
2606 | /* Throttle before drbd_rs_begin_io, as that locks out application IO; | |
2607 | * this defers syncer requests for some time, before letting at least | |
2608 | * on request through. The resync controller on the receiving side | |
2609 | * will adapt to the incoming rate accordingly. | |
2610 | * | |
2611 | * We cannot throttle here if remote is Primary/SyncTarget: | |
2612 | * we would also throttle its application reads. | |
2613 | * In that case, throttling is done on the SyncTarget only. | |
2614 | */ | |
b30ab791 | 2615 | if (device->state.peer != R_PRIMARY && drbd_rs_should_slow_down(device, sector)) |
e3555d85 | 2616 | schedule_timeout_uninterruptible(HZ/10); |
b30ab791 | 2617 | if (drbd_rs_begin_io(device, sector)) |
80a40e43 | 2618 | goto out_free_e; |
b411b363 | 2619 | |
0f0601f4 | 2620 | submit_for_resync: |
b30ab791 | 2621 | atomic_add(size >> 9, &device->rs_sect_ev); |
0f0601f4 | 2622 | |
80a40e43 | 2623 | submit: |
b30ab791 | 2624 | inc_unacked(device); |
0500813f | 2625 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 2626 | list_add_tail(&peer_req->w.list, &device->read_ee); |
0500813f | 2627 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 2628 | |
b30ab791 | 2629 | if (drbd_submit_peer_request(device, peer_req, READ, fault_type) == 0) |
82bc0194 | 2630 | return 0; |
b411b363 | 2631 | |
10f6d992 | 2632 | /* don't care for the reason here */ |
d0180171 | 2633 | drbd_err(device, "submit failed, triggering re-connect\n"); |
0500813f | 2634 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 2635 | list_del(&peer_req->w.list); |
0500813f | 2636 | spin_unlock_irq(&device->resource->req_lock); |
22cc37a9 LE |
2637 | /* no drbd_rs_complete_io(), we are dropping the connection anyways */ |
2638 | ||
b411b363 | 2639 | out_free_e: |
b30ab791 AG |
2640 | put_ldev(device); |
2641 | drbd_free_peer_req(device, peer_req); | |
82bc0194 | 2642 | return -EIO; |
b411b363 PR |
2643 | } |
2644 | ||
69a22773 AG |
2645 | /** |
2646 | * drbd_asb_recover_0p - Recover after split-brain with no remaining primaries | |
2647 | */ | |
2648 | static int drbd_asb_recover_0p(struct drbd_peer_device *peer_device) __must_hold(local) | |
b411b363 | 2649 | { |
69a22773 | 2650 | struct drbd_device *device = peer_device->device; |
b411b363 PR |
2651 | int self, peer, rv = -100; |
2652 | unsigned long ch_self, ch_peer; | |
44ed167d | 2653 | enum drbd_after_sb_p after_sb_0p; |
b411b363 | 2654 | |
b30ab791 AG |
2655 | self = device->ldev->md.uuid[UI_BITMAP] & 1; |
2656 | peer = device->p_uuid[UI_BITMAP] & 1; | |
b411b363 | 2657 | |
b30ab791 AG |
2658 | ch_peer = device->p_uuid[UI_SIZE]; |
2659 | ch_self = device->comm_bm_set; | |
b411b363 | 2660 | |
44ed167d | 2661 | rcu_read_lock(); |
69a22773 | 2662 | after_sb_0p = rcu_dereference(peer_device->connection->net_conf)->after_sb_0p; |
44ed167d PR |
2663 | rcu_read_unlock(); |
2664 | switch (after_sb_0p) { | |
b411b363 PR |
2665 | case ASB_CONSENSUS: |
2666 | case ASB_DISCARD_SECONDARY: | |
2667 | case ASB_CALL_HELPER: | |
44ed167d | 2668 | case ASB_VIOLENTLY: |
d0180171 | 2669 | drbd_err(device, "Configuration error.\n"); |
b411b363 PR |
2670 | break; |
2671 | case ASB_DISCONNECT: | |
2672 | break; | |
2673 | case ASB_DISCARD_YOUNGER_PRI: | |
2674 | if (self == 0 && peer == 1) { | |
2675 | rv = -1; | |
2676 | break; | |
2677 | } | |
2678 | if (self == 1 && peer == 0) { | |
2679 | rv = 1; | |
2680 | break; | |
2681 | } | |
2682 | /* Else fall through to one of the other strategies... */ | |
2683 | case ASB_DISCARD_OLDER_PRI: | |
2684 | if (self == 0 && peer == 1) { | |
2685 | rv = 1; | |
2686 | break; | |
2687 | } | |
2688 | if (self == 1 && peer == 0) { | |
2689 | rv = -1; | |
2690 | break; | |
2691 | } | |
2692 | /* Else fall through to one of the other strategies... */ | |
d0180171 | 2693 | drbd_warn(device, "Discard younger/older primary did not find a decision\n" |
b411b363 PR |
2694 | "Using discard-least-changes instead\n"); |
2695 | case ASB_DISCARD_ZERO_CHG: | |
2696 | if (ch_peer == 0 && ch_self == 0) { | |
69a22773 | 2697 | rv = test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags) |
b411b363 PR |
2698 | ? -1 : 1; |
2699 | break; | |
2700 | } else { | |
2701 | if (ch_peer == 0) { rv = 1; break; } | |
2702 | if (ch_self == 0) { rv = -1; break; } | |
2703 | } | |
44ed167d | 2704 | if (after_sb_0p == ASB_DISCARD_ZERO_CHG) |
b411b363 PR |
2705 | break; |
2706 | case ASB_DISCARD_LEAST_CHG: | |
2707 | if (ch_self < ch_peer) | |
2708 | rv = -1; | |
2709 | else if (ch_self > ch_peer) | |
2710 | rv = 1; | |
2711 | else /* ( ch_self == ch_peer ) */ | |
2712 | /* Well, then use something else. */ | |
69a22773 | 2713 | rv = test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags) |
b411b363 PR |
2714 | ? -1 : 1; |
2715 | break; | |
2716 | case ASB_DISCARD_LOCAL: | |
2717 | rv = -1; | |
2718 | break; | |
2719 | case ASB_DISCARD_REMOTE: | |
2720 | rv = 1; | |
2721 | } | |
2722 | ||
2723 | return rv; | |
2724 | } | |
2725 | ||
69a22773 AG |
2726 | /** |
2727 | * drbd_asb_recover_1p - Recover after split-brain with one remaining primary | |
2728 | */ | |
2729 | static int drbd_asb_recover_1p(struct drbd_peer_device *peer_device) __must_hold(local) | |
b411b363 | 2730 | { |
69a22773 | 2731 | struct drbd_device *device = peer_device->device; |
6184ea21 | 2732 | int hg, rv = -100; |
44ed167d | 2733 | enum drbd_after_sb_p after_sb_1p; |
b411b363 | 2734 | |
44ed167d | 2735 | rcu_read_lock(); |
69a22773 | 2736 | after_sb_1p = rcu_dereference(peer_device->connection->net_conf)->after_sb_1p; |
44ed167d PR |
2737 | rcu_read_unlock(); |
2738 | switch (after_sb_1p) { | |
b411b363 PR |
2739 | case ASB_DISCARD_YOUNGER_PRI: |
2740 | case ASB_DISCARD_OLDER_PRI: | |
2741 | case ASB_DISCARD_LEAST_CHG: | |
2742 | case ASB_DISCARD_LOCAL: | |
2743 | case ASB_DISCARD_REMOTE: | |
44ed167d | 2744 | case ASB_DISCARD_ZERO_CHG: |
d0180171 | 2745 | drbd_err(device, "Configuration error.\n"); |
b411b363 PR |
2746 | break; |
2747 | case ASB_DISCONNECT: | |
2748 | break; | |
2749 | case ASB_CONSENSUS: | |
69a22773 | 2750 | hg = drbd_asb_recover_0p(peer_device); |
b30ab791 | 2751 | if (hg == -1 && device->state.role == R_SECONDARY) |
b411b363 | 2752 | rv = hg; |
b30ab791 | 2753 | if (hg == 1 && device->state.role == R_PRIMARY) |
b411b363 PR |
2754 | rv = hg; |
2755 | break; | |
2756 | case ASB_VIOLENTLY: | |
69a22773 | 2757 | rv = drbd_asb_recover_0p(peer_device); |
b411b363 PR |
2758 | break; |
2759 | case ASB_DISCARD_SECONDARY: | |
b30ab791 | 2760 | return device->state.role == R_PRIMARY ? 1 : -1; |
b411b363 | 2761 | case ASB_CALL_HELPER: |
69a22773 | 2762 | hg = drbd_asb_recover_0p(peer_device); |
b30ab791 | 2763 | if (hg == -1 && device->state.role == R_PRIMARY) { |
bb437946 AG |
2764 | enum drbd_state_rv rv2; |
2765 | ||
b411b363 PR |
2766 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2767 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2768 | * we do not need to wait for the after state change work either. */ | |
b30ab791 | 2769 | rv2 = drbd_change_state(device, CS_VERBOSE, NS(role, R_SECONDARY)); |
bb437946 | 2770 | if (rv2 != SS_SUCCESS) { |
b30ab791 | 2771 | drbd_khelper(device, "pri-lost-after-sb"); |
b411b363 | 2772 | } else { |
d0180171 | 2773 | drbd_warn(device, "Successfully gave up primary role.\n"); |
b411b363 PR |
2774 | rv = hg; |
2775 | } | |
2776 | } else | |
2777 | rv = hg; | |
2778 | } | |
2779 | ||
2780 | return rv; | |
2781 | } | |
2782 | ||
69a22773 AG |
2783 | /** |
2784 | * drbd_asb_recover_2p - Recover after split-brain with two remaining primaries | |
2785 | */ | |
2786 | static int drbd_asb_recover_2p(struct drbd_peer_device *peer_device) __must_hold(local) | |
b411b363 | 2787 | { |
69a22773 | 2788 | struct drbd_device *device = peer_device->device; |
6184ea21 | 2789 | int hg, rv = -100; |
44ed167d | 2790 | enum drbd_after_sb_p after_sb_2p; |
b411b363 | 2791 | |
44ed167d | 2792 | rcu_read_lock(); |
69a22773 | 2793 | after_sb_2p = rcu_dereference(peer_device->connection->net_conf)->after_sb_2p; |
44ed167d PR |
2794 | rcu_read_unlock(); |
2795 | switch (after_sb_2p) { | |
b411b363 PR |
2796 | case ASB_DISCARD_YOUNGER_PRI: |
2797 | case ASB_DISCARD_OLDER_PRI: | |
2798 | case ASB_DISCARD_LEAST_CHG: | |
2799 | case ASB_DISCARD_LOCAL: | |
2800 | case ASB_DISCARD_REMOTE: | |
2801 | case ASB_CONSENSUS: | |
2802 | case ASB_DISCARD_SECONDARY: | |
44ed167d | 2803 | case ASB_DISCARD_ZERO_CHG: |
d0180171 | 2804 | drbd_err(device, "Configuration error.\n"); |
b411b363 PR |
2805 | break; |
2806 | case ASB_VIOLENTLY: | |
69a22773 | 2807 | rv = drbd_asb_recover_0p(peer_device); |
b411b363 PR |
2808 | break; |
2809 | case ASB_DISCONNECT: | |
2810 | break; | |
2811 | case ASB_CALL_HELPER: | |
69a22773 | 2812 | hg = drbd_asb_recover_0p(peer_device); |
b411b363 | 2813 | if (hg == -1) { |
bb437946 AG |
2814 | enum drbd_state_rv rv2; |
2815 | ||
b411b363 PR |
2816 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2817 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2818 | * we do not need to wait for the after state change work either. */ | |
b30ab791 | 2819 | rv2 = drbd_change_state(device, CS_VERBOSE, NS(role, R_SECONDARY)); |
bb437946 | 2820 | if (rv2 != SS_SUCCESS) { |
b30ab791 | 2821 | drbd_khelper(device, "pri-lost-after-sb"); |
b411b363 | 2822 | } else { |
d0180171 | 2823 | drbd_warn(device, "Successfully gave up primary role.\n"); |
b411b363 PR |
2824 | rv = hg; |
2825 | } | |
2826 | } else | |
2827 | rv = hg; | |
2828 | } | |
2829 | ||
2830 | return rv; | |
2831 | } | |
2832 | ||
b30ab791 | 2833 | static void drbd_uuid_dump(struct drbd_device *device, char *text, u64 *uuid, |
b411b363 PR |
2834 | u64 bits, u64 flags) |
2835 | { | |
2836 | if (!uuid) { | |
d0180171 | 2837 | drbd_info(device, "%s uuid info vanished while I was looking!\n", text); |
b411b363 PR |
2838 | return; |
2839 | } | |
d0180171 | 2840 | drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", |
b411b363 PR |
2841 | text, |
2842 | (unsigned long long)uuid[UI_CURRENT], | |
2843 | (unsigned long long)uuid[UI_BITMAP], | |
2844 | (unsigned long long)uuid[UI_HISTORY_START], | |
2845 | (unsigned long long)uuid[UI_HISTORY_END], | |
2846 | (unsigned long long)bits, | |
2847 | (unsigned long long)flags); | |
2848 | } | |
2849 | ||
2850 | /* | |
2851 | 100 after split brain try auto recover | |
2852 | 2 C_SYNC_SOURCE set BitMap | |
2853 | 1 C_SYNC_SOURCE use BitMap | |
2854 | 0 no Sync | |
2855 | -1 C_SYNC_TARGET use BitMap | |
2856 | -2 C_SYNC_TARGET set BitMap | |
2857 | -100 after split brain, disconnect | |
2858 | -1000 unrelated data | |
4a23f264 PR |
2859 | -1091 requires proto 91 |
2860 | -1096 requires proto 96 | |
b411b363 | 2861 | */ |
44a4d551 | 2862 | static int drbd_uuid_compare(struct drbd_device *const device, int *rule_nr) __must_hold(local) |
b411b363 | 2863 | { |
44a4d551 LE |
2864 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
2865 | struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL; | |
b411b363 PR |
2866 | u64 self, peer; |
2867 | int i, j; | |
2868 | ||
b30ab791 AG |
2869 | self = device->ldev->md.uuid[UI_CURRENT] & ~((u64)1); |
2870 | peer = device->p_uuid[UI_CURRENT] & ~((u64)1); | |
b411b363 PR |
2871 | |
2872 | *rule_nr = 10; | |
2873 | if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) | |
2874 | return 0; | |
2875 | ||
2876 | *rule_nr = 20; | |
2877 | if ((self == UUID_JUST_CREATED || self == (u64)0) && | |
2878 | peer != UUID_JUST_CREATED) | |
2879 | return -2; | |
2880 | ||
2881 | *rule_nr = 30; | |
2882 | if (self != UUID_JUST_CREATED && | |
2883 | (peer == UUID_JUST_CREATED || peer == (u64)0)) | |
2884 | return 2; | |
2885 | ||
2886 | if (self == peer) { | |
2887 | int rct, dc; /* roles at crash time */ | |
2888 | ||
b30ab791 | 2889 | if (device->p_uuid[UI_BITMAP] == (u64)0 && device->ldev->md.uuid[UI_BITMAP] != (u64)0) { |
b411b363 | 2890 | |
44a4d551 | 2891 | if (connection->agreed_pro_version < 91) |
4a23f264 | 2892 | return -1091; |
b411b363 | 2893 | |
b30ab791 AG |
2894 | if ((device->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START] & ~((u64)1)) && |
2895 | (device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { | |
d0180171 | 2896 | drbd_info(device, "was SyncSource, missed the resync finished event, corrected myself:\n"); |
b30ab791 AG |
2897 | drbd_uuid_move_history(device); |
2898 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP]; | |
2899 | device->ldev->md.uuid[UI_BITMAP] = 0; | |
b411b363 | 2900 | |
b30ab791 AG |
2901 | drbd_uuid_dump(device, "self", device->ldev->md.uuid, |
2902 | device->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(device) : 0, 0); | |
b411b363 PR |
2903 | *rule_nr = 34; |
2904 | } else { | |
d0180171 | 2905 | drbd_info(device, "was SyncSource (peer failed to write sync_uuid)\n"); |
b411b363 PR |
2906 | *rule_nr = 36; |
2907 | } | |
2908 | ||
2909 | return 1; | |
2910 | } | |
2911 | ||
b30ab791 | 2912 | if (device->ldev->md.uuid[UI_BITMAP] == (u64)0 && device->p_uuid[UI_BITMAP] != (u64)0) { |
b411b363 | 2913 | |
44a4d551 | 2914 | if (connection->agreed_pro_version < 91) |
4a23f264 | 2915 | return -1091; |
b411b363 | 2916 | |
b30ab791 AG |
2917 | if ((device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (device->p_uuid[UI_BITMAP] & ~((u64)1)) && |
2918 | (device->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START] & ~((u64)1))) { | |
d0180171 | 2919 | drbd_info(device, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); |
b411b363 | 2920 | |
b30ab791 AG |
2921 | device->p_uuid[UI_HISTORY_START + 1] = device->p_uuid[UI_HISTORY_START]; |
2922 | device->p_uuid[UI_HISTORY_START] = device->p_uuid[UI_BITMAP]; | |
2923 | device->p_uuid[UI_BITMAP] = 0UL; | |
b411b363 | 2924 | |
b30ab791 | 2925 | drbd_uuid_dump(device, "peer", device->p_uuid, device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); |
b411b363 PR |
2926 | *rule_nr = 35; |
2927 | } else { | |
d0180171 | 2928 | drbd_info(device, "was SyncTarget (failed to write sync_uuid)\n"); |
b411b363 PR |
2929 | *rule_nr = 37; |
2930 | } | |
2931 | ||
2932 | return -1; | |
2933 | } | |
2934 | ||
2935 | /* Common power [off|failure] */ | |
b30ab791 AG |
2936 | rct = (test_bit(CRASHED_PRIMARY, &device->flags) ? 1 : 0) + |
2937 | (device->p_uuid[UI_FLAGS] & 2); | |
b411b363 PR |
2938 | /* lowest bit is set when we were primary, |
2939 | * next bit (weight 2) is set when peer was primary */ | |
2940 | *rule_nr = 40; | |
2941 | ||
2942 | switch (rct) { | |
2943 | case 0: /* !self_pri && !peer_pri */ return 0; | |
2944 | case 1: /* self_pri && !peer_pri */ return 1; | |
2945 | case 2: /* !self_pri && peer_pri */ return -1; | |
2946 | case 3: /* self_pri && peer_pri */ | |
44a4d551 | 2947 | dc = test_bit(RESOLVE_CONFLICTS, &connection->flags); |
b411b363 PR |
2948 | return dc ? -1 : 1; |
2949 | } | |
2950 | } | |
2951 | ||
2952 | *rule_nr = 50; | |
b30ab791 | 2953 | peer = device->p_uuid[UI_BITMAP] & ~((u64)1); |
b411b363 PR |
2954 | if (self == peer) |
2955 | return -1; | |
2956 | ||
2957 | *rule_nr = 51; | |
b30ab791 | 2958 | peer = device->p_uuid[UI_HISTORY_START] & ~((u64)1); |
b411b363 | 2959 | if (self == peer) { |
44a4d551 | 2960 | if (connection->agreed_pro_version < 96 ? |
b30ab791 AG |
2961 | (device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == |
2962 | (device->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) : | |
2963 | peer + UUID_NEW_BM_OFFSET == (device->p_uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2964 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2965 | resync as sync source modifications of the peer's UUIDs. */ | |
2966 | ||
44a4d551 | 2967 | if (connection->agreed_pro_version < 91) |
4a23f264 | 2968 | return -1091; |
b411b363 | 2969 | |
b30ab791 AG |
2970 | device->p_uuid[UI_BITMAP] = device->p_uuid[UI_HISTORY_START]; |
2971 | device->p_uuid[UI_HISTORY_START] = device->p_uuid[UI_HISTORY_START + 1]; | |
4a23f264 | 2972 | |
d0180171 | 2973 | drbd_info(device, "Lost last syncUUID packet, corrected:\n"); |
b30ab791 | 2974 | drbd_uuid_dump(device, "peer", device->p_uuid, device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); |
4a23f264 | 2975 | |
b411b363 PR |
2976 | return -1; |
2977 | } | |
2978 | } | |
2979 | ||
2980 | *rule_nr = 60; | |
b30ab791 | 2981 | self = device->ldev->md.uuid[UI_CURRENT] & ~((u64)1); |
b411b363 | 2982 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
b30ab791 | 2983 | peer = device->p_uuid[i] & ~((u64)1); |
b411b363 PR |
2984 | if (self == peer) |
2985 | return -2; | |
2986 | } | |
2987 | ||
2988 | *rule_nr = 70; | |
b30ab791 AG |
2989 | self = device->ldev->md.uuid[UI_BITMAP] & ~((u64)1); |
2990 | peer = device->p_uuid[UI_CURRENT] & ~((u64)1); | |
b411b363 PR |
2991 | if (self == peer) |
2992 | return 1; | |
2993 | ||
2994 | *rule_nr = 71; | |
b30ab791 | 2995 | self = device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); |
b411b363 | 2996 | if (self == peer) { |
44a4d551 | 2997 | if (connection->agreed_pro_version < 96 ? |
b30ab791 AG |
2998 | (device->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == |
2999 | (device->p_uuid[UI_HISTORY_START] & ~((u64)1)) : | |
3000 | self + UUID_NEW_BM_OFFSET == (device->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
3001 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
3002 | resync as sync source modifications of our UUIDs. */ | |
3003 | ||
44a4d551 | 3004 | if (connection->agreed_pro_version < 91) |
4a23f264 | 3005 | return -1091; |
b411b363 | 3006 | |
b30ab791 AG |
3007 | __drbd_uuid_set(device, UI_BITMAP, device->ldev->md.uuid[UI_HISTORY_START]); |
3008 | __drbd_uuid_set(device, UI_HISTORY_START, device->ldev->md.uuid[UI_HISTORY_START + 1]); | |
b411b363 | 3009 | |
d0180171 | 3010 | drbd_info(device, "Last syncUUID did not get through, corrected:\n"); |
b30ab791 AG |
3011 | drbd_uuid_dump(device, "self", device->ldev->md.uuid, |
3012 | device->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(device) : 0, 0); | |
b411b363 PR |
3013 | |
3014 | return 1; | |
3015 | } | |
3016 | } | |
3017 | ||
3018 | ||
3019 | *rule_nr = 80; | |
b30ab791 | 3020 | peer = device->p_uuid[UI_CURRENT] & ~((u64)1); |
b411b363 | 3021 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
b30ab791 | 3022 | self = device->ldev->md.uuid[i] & ~((u64)1); |
b411b363 PR |
3023 | if (self == peer) |
3024 | return 2; | |
3025 | } | |
3026 | ||
3027 | *rule_nr = 90; | |
b30ab791 AG |
3028 | self = device->ldev->md.uuid[UI_BITMAP] & ~((u64)1); |
3029 | peer = device->p_uuid[UI_BITMAP] & ~((u64)1); | |
b411b363 PR |
3030 | if (self == peer && self != ((u64)0)) |
3031 | return 100; | |
3032 | ||
3033 | *rule_nr = 100; | |
3034 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
b30ab791 | 3035 | self = device->ldev->md.uuid[i] & ~((u64)1); |
b411b363 | 3036 | for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { |
b30ab791 | 3037 | peer = device->p_uuid[j] & ~((u64)1); |
b411b363 PR |
3038 | if (self == peer) |
3039 | return -100; | |
3040 | } | |
3041 | } | |
3042 | ||
3043 | return -1000; | |
3044 | } | |
3045 | ||
3046 | /* drbd_sync_handshake() returns the new conn state on success, or | |
3047 | CONN_MASK (-1) on failure. | |
3048 | */ | |
69a22773 AG |
3049 | static enum drbd_conns drbd_sync_handshake(struct drbd_peer_device *peer_device, |
3050 | enum drbd_role peer_role, | |
b411b363 PR |
3051 | enum drbd_disk_state peer_disk) __must_hold(local) |
3052 | { | |
69a22773 | 3053 | struct drbd_device *device = peer_device->device; |
b411b363 PR |
3054 | enum drbd_conns rv = C_MASK; |
3055 | enum drbd_disk_state mydisk; | |
44ed167d | 3056 | struct net_conf *nc; |
6dff2902 | 3057 | int hg, rule_nr, rr_conflict, tentative; |
b411b363 | 3058 | |
b30ab791 | 3059 | mydisk = device->state.disk; |
b411b363 | 3060 | if (mydisk == D_NEGOTIATING) |
b30ab791 | 3061 | mydisk = device->new_state_tmp.disk; |
b411b363 | 3062 | |
d0180171 | 3063 | drbd_info(device, "drbd_sync_handshake:\n"); |
9f2247bb | 3064 | |
b30ab791 AG |
3065 | spin_lock_irq(&device->ldev->md.uuid_lock); |
3066 | drbd_uuid_dump(device, "self", device->ldev->md.uuid, device->comm_bm_set, 0); | |
3067 | drbd_uuid_dump(device, "peer", device->p_uuid, | |
3068 | device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); | |
b411b363 | 3069 | |
b30ab791 AG |
3070 | hg = drbd_uuid_compare(device, &rule_nr); |
3071 | spin_unlock_irq(&device->ldev->md.uuid_lock); | |
b411b363 | 3072 | |
d0180171 | 3073 | drbd_info(device, "uuid_compare()=%d by rule %d\n", hg, rule_nr); |
b411b363 PR |
3074 | |
3075 | if (hg == -1000) { | |
d0180171 | 3076 | drbd_alert(device, "Unrelated data, aborting!\n"); |
b411b363 PR |
3077 | return C_MASK; |
3078 | } | |
4a23f264 | 3079 | if (hg < -1000) { |
d0180171 | 3080 | drbd_alert(device, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000); |
b411b363 PR |
3081 | return C_MASK; |
3082 | } | |
3083 | ||
3084 | if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || | |
3085 | (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { | |
3086 | int f = (hg == -100) || abs(hg) == 2; | |
3087 | hg = mydisk > D_INCONSISTENT ? 1 : -1; | |
3088 | if (f) | |
3089 | hg = hg*2; | |
d0180171 | 3090 | drbd_info(device, "Becoming sync %s due to disk states.\n", |
b411b363 PR |
3091 | hg > 0 ? "source" : "target"); |
3092 | } | |
3093 | ||
3a11a487 | 3094 | if (abs(hg) == 100) |
b30ab791 | 3095 | drbd_khelper(device, "initial-split-brain"); |
3a11a487 | 3096 | |
44ed167d | 3097 | rcu_read_lock(); |
69a22773 | 3098 | nc = rcu_dereference(peer_device->connection->net_conf); |
44ed167d PR |
3099 | |
3100 | if (hg == 100 || (hg == -100 && nc->always_asbp)) { | |
b30ab791 | 3101 | int pcount = (device->state.role == R_PRIMARY) |
b411b363 PR |
3102 | + (peer_role == R_PRIMARY); |
3103 | int forced = (hg == -100); | |
3104 | ||
3105 | switch (pcount) { | |
3106 | case 0: | |
69a22773 | 3107 | hg = drbd_asb_recover_0p(peer_device); |
b411b363 PR |
3108 | break; |
3109 | case 1: | |
69a22773 | 3110 | hg = drbd_asb_recover_1p(peer_device); |
b411b363 PR |
3111 | break; |
3112 | case 2: | |
69a22773 | 3113 | hg = drbd_asb_recover_2p(peer_device); |
b411b363 PR |
3114 | break; |
3115 | } | |
3116 | if (abs(hg) < 100) { | |
d0180171 | 3117 | drbd_warn(device, "Split-Brain detected, %d primaries, " |
b411b363 PR |
3118 | "automatically solved. Sync from %s node\n", |
3119 | pcount, (hg < 0) ? "peer" : "this"); | |
3120 | if (forced) { | |
d0180171 | 3121 | drbd_warn(device, "Doing a full sync, since" |
b411b363 PR |
3122 | " UUIDs where ambiguous.\n"); |
3123 | hg = hg*2; | |
3124 | } | |
3125 | } | |
3126 | } | |
3127 | ||
3128 | if (hg == -100) { | |
b30ab791 | 3129 | if (test_bit(DISCARD_MY_DATA, &device->flags) && !(device->p_uuid[UI_FLAGS]&1)) |
b411b363 | 3130 | hg = -1; |
b30ab791 | 3131 | if (!test_bit(DISCARD_MY_DATA, &device->flags) && (device->p_uuid[UI_FLAGS]&1)) |
b411b363 PR |
3132 | hg = 1; |
3133 | ||
3134 | if (abs(hg) < 100) | |
d0180171 | 3135 | drbd_warn(device, "Split-Brain detected, manually solved. " |
b411b363 PR |
3136 | "Sync from %s node\n", |
3137 | (hg < 0) ? "peer" : "this"); | |
3138 | } | |
44ed167d | 3139 | rr_conflict = nc->rr_conflict; |
6dff2902 | 3140 | tentative = nc->tentative; |
44ed167d | 3141 | rcu_read_unlock(); |
b411b363 PR |
3142 | |
3143 | if (hg == -100) { | |
580b9767 LE |
3144 | /* FIXME this log message is not correct if we end up here |
3145 | * after an attempted attach on a diskless node. | |
3146 | * We just refuse to attach -- well, we drop the "connection" | |
3147 | * to that disk, in a way... */ | |
d0180171 | 3148 | drbd_alert(device, "Split-Brain detected but unresolved, dropping connection!\n"); |
b30ab791 | 3149 | drbd_khelper(device, "split-brain"); |
b411b363 PR |
3150 | return C_MASK; |
3151 | } | |
3152 | ||
3153 | if (hg > 0 && mydisk <= D_INCONSISTENT) { | |
d0180171 | 3154 | drbd_err(device, "I shall become SyncSource, but I am inconsistent!\n"); |
b411b363 PR |
3155 | return C_MASK; |
3156 | } | |
3157 | ||
3158 | if (hg < 0 && /* by intention we do not use mydisk here. */ | |
b30ab791 | 3159 | device->state.role == R_PRIMARY && device->state.disk >= D_CONSISTENT) { |
44ed167d | 3160 | switch (rr_conflict) { |
b411b363 | 3161 | case ASB_CALL_HELPER: |
b30ab791 | 3162 | drbd_khelper(device, "pri-lost"); |
b411b363 PR |
3163 | /* fall through */ |
3164 | case ASB_DISCONNECT: | |
d0180171 | 3165 | drbd_err(device, "I shall become SyncTarget, but I am primary!\n"); |
b411b363 PR |
3166 | return C_MASK; |
3167 | case ASB_VIOLENTLY: | |
d0180171 | 3168 | drbd_warn(device, "Becoming SyncTarget, violating the stable-data" |
b411b363 PR |
3169 | "assumption\n"); |
3170 | } | |
3171 | } | |
3172 | ||
69a22773 | 3173 | if (tentative || test_bit(CONN_DRY_RUN, &peer_device->connection->flags)) { |
cf14c2e9 | 3174 | if (hg == 0) |
d0180171 | 3175 | drbd_info(device, "dry-run connect: No resync, would become Connected immediately.\n"); |
cf14c2e9 | 3176 | else |
d0180171 | 3177 | drbd_info(device, "dry-run connect: Would become %s, doing a %s resync.", |
cf14c2e9 PR |
3178 | drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), |
3179 | abs(hg) >= 2 ? "full" : "bit-map based"); | |
3180 | return C_MASK; | |
3181 | } | |
3182 | ||
b411b363 | 3183 | if (abs(hg) >= 2) { |
d0180171 | 3184 | drbd_info(device, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); |
b30ab791 | 3185 | if (drbd_bitmap_io(device, &drbd_bmio_set_n_write, "set_n_write from sync_handshake", |
20ceb2b2 | 3186 | BM_LOCKED_SET_ALLOWED)) |
b411b363 PR |
3187 | return C_MASK; |
3188 | } | |
3189 | ||
3190 | if (hg > 0) { /* become sync source. */ | |
3191 | rv = C_WF_BITMAP_S; | |
3192 | } else if (hg < 0) { /* become sync target */ | |
3193 | rv = C_WF_BITMAP_T; | |
3194 | } else { | |
3195 | rv = C_CONNECTED; | |
b30ab791 | 3196 | if (drbd_bm_total_weight(device)) { |
d0180171 | 3197 | drbd_info(device, "No resync, but %lu bits in bitmap!\n", |
b30ab791 | 3198 | drbd_bm_total_weight(device)); |
b411b363 PR |
3199 | } |
3200 | } | |
3201 | ||
3202 | return rv; | |
3203 | } | |
3204 | ||
f179d76d | 3205 | static enum drbd_after_sb_p convert_after_sb(enum drbd_after_sb_p peer) |
b411b363 PR |
3206 | { |
3207 | /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ | |
f179d76d PR |
3208 | if (peer == ASB_DISCARD_REMOTE) |
3209 | return ASB_DISCARD_LOCAL; | |
b411b363 PR |
3210 | |
3211 | /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ | |
f179d76d PR |
3212 | if (peer == ASB_DISCARD_LOCAL) |
3213 | return ASB_DISCARD_REMOTE; | |
b411b363 PR |
3214 | |
3215 | /* everything else is valid if they are equal on both sides. */ | |
f179d76d | 3216 | return peer; |
b411b363 PR |
3217 | } |
3218 | ||
bde89a9e | 3219 | static int receive_protocol(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3220 | { |
e658983a | 3221 | struct p_protocol *p = pi->data; |
036b17ea PR |
3222 | enum drbd_after_sb_p p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; |
3223 | int p_proto, p_discard_my_data, p_two_primaries, cf; | |
3224 | struct net_conf *nc, *old_net_conf, *new_net_conf = NULL; | |
3225 | char integrity_alg[SHARED_SECRET_MAX] = ""; | |
accdbcc5 | 3226 | struct crypto_hash *peer_integrity_tfm = NULL; |
7aca6c75 | 3227 | void *int_dig_in = NULL, *int_dig_vv = NULL; |
b411b363 | 3228 | |
b411b363 PR |
3229 | p_proto = be32_to_cpu(p->protocol); |
3230 | p_after_sb_0p = be32_to_cpu(p->after_sb_0p); | |
3231 | p_after_sb_1p = be32_to_cpu(p->after_sb_1p); | |
3232 | p_after_sb_2p = be32_to_cpu(p->after_sb_2p); | |
b411b363 | 3233 | p_two_primaries = be32_to_cpu(p->two_primaries); |
cf14c2e9 | 3234 | cf = be32_to_cpu(p->conn_flags); |
6139f60d | 3235 | p_discard_my_data = cf & CF_DISCARD_MY_DATA; |
cf14c2e9 | 3236 | |
bde89a9e | 3237 | if (connection->agreed_pro_version >= 87) { |
86db0618 | 3238 | int err; |
cf14c2e9 | 3239 | |
88104ca4 | 3240 | if (pi->size > sizeof(integrity_alg)) |
86db0618 | 3241 | return -EIO; |
bde89a9e | 3242 | err = drbd_recv_all(connection, integrity_alg, pi->size); |
86db0618 AG |
3243 | if (err) |
3244 | return err; | |
036b17ea | 3245 | integrity_alg[SHARED_SECRET_MAX - 1] = 0; |
b411b363 PR |
3246 | } |
3247 | ||
7d4c782c | 3248 | if (pi->cmd != P_PROTOCOL_UPDATE) { |
bde89a9e | 3249 | clear_bit(CONN_DRY_RUN, &connection->flags); |
b411b363 | 3250 | |
fbc12f45 | 3251 | if (cf & CF_DRY_RUN) |
bde89a9e | 3252 | set_bit(CONN_DRY_RUN, &connection->flags); |
b411b363 | 3253 | |
fbc12f45 | 3254 | rcu_read_lock(); |
bde89a9e | 3255 | nc = rcu_dereference(connection->net_conf); |
b411b363 | 3256 | |
fbc12f45 | 3257 | if (p_proto != nc->wire_protocol) { |
1ec861eb | 3258 | drbd_err(connection, "incompatible %s settings\n", "protocol"); |
fbc12f45 AG |
3259 | goto disconnect_rcu_unlock; |
3260 | } | |
b411b363 | 3261 | |
fbc12f45 | 3262 | if (convert_after_sb(p_after_sb_0p) != nc->after_sb_0p) { |
1ec861eb | 3263 | drbd_err(connection, "incompatible %s settings\n", "after-sb-0pri"); |
fbc12f45 AG |
3264 | goto disconnect_rcu_unlock; |
3265 | } | |
b411b363 | 3266 | |
fbc12f45 | 3267 | if (convert_after_sb(p_after_sb_1p) != nc->after_sb_1p) { |
1ec861eb | 3268 | drbd_err(connection, "incompatible %s settings\n", "after-sb-1pri"); |
fbc12f45 AG |
3269 | goto disconnect_rcu_unlock; |
3270 | } | |
b411b363 | 3271 | |
fbc12f45 | 3272 | if (convert_after_sb(p_after_sb_2p) != nc->after_sb_2p) { |
1ec861eb | 3273 | drbd_err(connection, "incompatible %s settings\n", "after-sb-2pri"); |
fbc12f45 AG |
3274 | goto disconnect_rcu_unlock; |
3275 | } | |
b411b363 | 3276 | |
fbc12f45 | 3277 | if (p_discard_my_data && nc->discard_my_data) { |
1ec861eb | 3278 | drbd_err(connection, "incompatible %s settings\n", "discard-my-data"); |
fbc12f45 AG |
3279 | goto disconnect_rcu_unlock; |
3280 | } | |
b411b363 | 3281 | |
fbc12f45 | 3282 | if (p_two_primaries != nc->two_primaries) { |
1ec861eb | 3283 | drbd_err(connection, "incompatible %s settings\n", "allow-two-primaries"); |
fbc12f45 AG |
3284 | goto disconnect_rcu_unlock; |
3285 | } | |
b411b363 | 3286 | |
fbc12f45 | 3287 | if (strcmp(integrity_alg, nc->integrity_alg)) { |
1ec861eb | 3288 | drbd_err(connection, "incompatible %s settings\n", "data-integrity-alg"); |
fbc12f45 AG |
3289 | goto disconnect_rcu_unlock; |
3290 | } | |
b411b363 | 3291 | |
fbc12f45 | 3292 | rcu_read_unlock(); |
b411b363 PR |
3293 | } |
3294 | ||
7d4c782c AG |
3295 | if (integrity_alg[0]) { |
3296 | int hash_size; | |
3297 | ||
3298 | /* | |
3299 | * We can only change the peer data integrity algorithm | |
3300 | * here. Changing our own data integrity algorithm | |
3301 | * requires that we send a P_PROTOCOL_UPDATE packet at | |
3302 | * the same time; otherwise, the peer has no way to | |
3303 | * tell between which packets the algorithm should | |
3304 | * change. | |
3305 | */ | |
b411b363 | 3306 | |
7d4c782c AG |
3307 | peer_integrity_tfm = crypto_alloc_hash(integrity_alg, 0, CRYPTO_ALG_ASYNC); |
3308 | if (!peer_integrity_tfm) { | |
1ec861eb | 3309 | drbd_err(connection, "peer data-integrity-alg %s not supported\n", |
7d4c782c AG |
3310 | integrity_alg); |
3311 | goto disconnect; | |
3312 | } | |
b411b363 | 3313 | |
7d4c782c AG |
3314 | hash_size = crypto_hash_digestsize(peer_integrity_tfm); |
3315 | int_dig_in = kmalloc(hash_size, GFP_KERNEL); | |
3316 | int_dig_vv = kmalloc(hash_size, GFP_KERNEL); | |
3317 | if (!(int_dig_in && int_dig_vv)) { | |
1ec861eb | 3318 | drbd_err(connection, "Allocation of buffers for data integrity checking failed\n"); |
b411b363 PR |
3319 | goto disconnect; |
3320 | } | |
b411b363 PR |
3321 | } |
3322 | ||
7d4c782c AG |
3323 | new_net_conf = kmalloc(sizeof(struct net_conf), GFP_KERNEL); |
3324 | if (!new_net_conf) { | |
1ec861eb | 3325 | drbd_err(connection, "Allocation of new net_conf failed\n"); |
7d4c782c AG |
3326 | goto disconnect; |
3327 | } | |
3328 | ||
bde89a9e | 3329 | mutex_lock(&connection->data.mutex); |
0500813f | 3330 | mutex_lock(&connection->resource->conf_update); |
bde89a9e | 3331 | old_net_conf = connection->net_conf; |
7d4c782c AG |
3332 | *new_net_conf = *old_net_conf; |
3333 | ||
3334 | new_net_conf->wire_protocol = p_proto; | |
3335 | new_net_conf->after_sb_0p = convert_after_sb(p_after_sb_0p); | |
3336 | new_net_conf->after_sb_1p = convert_after_sb(p_after_sb_1p); | |
3337 | new_net_conf->after_sb_2p = convert_after_sb(p_after_sb_2p); | |
3338 | new_net_conf->two_primaries = p_two_primaries; | |
3339 | ||
bde89a9e | 3340 | rcu_assign_pointer(connection->net_conf, new_net_conf); |
0500813f | 3341 | mutex_unlock(&connection->resource->conf_update); |
bde89a9e | 3342 | mutex_unlock(&connection->data.mutex); |
7d4c782c | 3343 | |
bde89a9e AG |
3344 | crypto_free_hash(connection->peer_integrity_tfm); |
3345 | kfree(connection->int_dig_in); | |
3346 | kfree(connection->int_dig_vv); | |
3347 | connection->peer_integrity_tfm = peer_integrity_tfm; | |
3348 | connection->int_dig_in = int_dig_in; | |
3349 | connection->int_dig_vv = int_dig_vv; | |
7d4c782c AG |
3350 | |
3351 | if (strcmp(old_net_conf->integrity_alg, integrity_alg)) | |
1ec861eb | 3352 | drbd_info(connection, "peer data-integrity-alg: %s\n", |
7d4c782c AG |
3353 | integrity_alg[0] ? integrity_alg : "(none)"); |
3354 | ||
3355 | synchronize_rcu(); | |
3356 | kfree(old_net_conf); | |
82bc0194 | 3357 | return 0; |
b411b363 | 3358 | |
44ed167d PR |
3359 | disconnect_rcu_unlock: |
3360 | rcu_read_unlock(); | |
b411b363 | 3361 | disconnect: |
b792c35c | 3362 | crypto_free_hash(peer_integrity_tfm); |
036b17ea PR |
3363 | kfree(int_dig_in); |
3364 | kfree(int_dig_vv); | |
bde89a9e | 3365 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3366 | return -EIO; |
b411b363 PR |
3367 | } |
3368 | ||
3369 | /* helper function | |
3370 | * input: alg name, feature name | |
3371 | * return: NULL (alg name was "") | |
3372 | * ERR_PTR(error) if something goes wrong | |
3373 | * or the crypto hash ptr, if it worked out ok. */ | |
f63e631a | 3374 | static |
b30ab791 | 3375 | struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_device *device, |
b411b363 PR |
3376 | const char *alg, const char *name) |
3377 | { | |
3378 | struct crypto_hash *tfm; | |
3379 | ||
3380 | if (!alg[0]) | |
3381 | return NULL; | |
3382 | ||
3383 | tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); | |
3384 | if (IS_ERR(tfm)) { | |
d0180171 | 3385 | drbd_err(device, "Can not allocate \"%s\" as %s (reason: %ld)\n", |
b411b363 PR |
3386 | alg, name, PTR_ERR(tfm)); |
3387 | return tfm; | |
3388 | } | |
b411b363 PR |
3389 | return tfm; |
3390 | } | |
3391 | ||
bde89a9e | 3392 | static int ignore_remaining_packet(struct drbd_connection *connection, struct packet_info *pi) |
4a76b161 | 3393 | { |
bde89a9e | 3394 | void *buffer = connection->data.rbuf; |
4a76b161 AG |
3395 | int size = pi->size; |
3396 | ||
3397 | while (size) { | |
3398 | int s = min_t(int, size, DRBD_SOCKET_BUFFER_SIZE); | |
bde89a9e | 3399 | s = drbd_recv(connection, buffer, s); |
4a76b161 AG |
3400 | if (s <= 0) { |
3401 | if (s < 0) | |
3402 | return s; | |
3403 | break; | |
3404 | } | |
3405 | size -= s; | |
3406 | } | |
3407 | if (size) | |
3408 | return -EIO; | |
3409 | return 0; | |
3410 | } | |
3411 | ||
3412 | /* | |
3413 | * config_unknown_volume - device configuration command for unknown volume | |
3414 | * | |
3415 | * When a device is added to an existing connection, the node on which the | |
3416 | * device is added first will send configuration commands to its peer but the | |
3417 | * peer will not know about the device yet. It will warn and ignore these | |
3418 | * commands. Once the device is added on the second node, the second node will | |
3419 | * send the same device configuration commands, but in the other direction. | |
3420 | * | |
3421 | * (We can also end up here if drbd is misconfigured.) | |
3422 | */ | |
bde89a9e | 3423 | static int config_unknown_volume(struct drbd_connection *connection, struct packet_info *pi) |
4a76b161 | 3424 | { |
1ec861eb | 3425 | drbd_warn(connection, "%s packet received for volume %u, which is not configured locally\n", |
2fcb8f30 | 3426 | cmdname(pi->cmd), pi->vnr); |
bde89a9e | 3427 | return ignore_remaining_packet(connection, pi); |
4a76b161 AG |
3428 | } |
3429 | ||
bde89a9e | 3430 | static int receive_SyncParam(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3431 | { |
9f4fe9ad | 3432 | struct drbd_peer_device *peer_device; |
b30ab791 | 3433 | struct drbd_device *device; |
e658983a | 3434 | struct p_rs_param_95 *p; |
b411b363 PR |
3435 | unsigned int header_size, data_size, exp_max_sz; |
3436 | struct crypto_hash *verify_tfm = NULL; | |
3437 | struct crypto_hash *csums_tfm = NULL; | |
2ec91e0e | 3438 | struct net_conf *old_net_conf, *new_net_conf = NULL; |
813472ce | 3439 | struct disk_conf *old_disk_conf = NULL, *new_disk_conf = NULL; |
bde89a9e | 3440 | const int apv = connection->agreed_pro_version; |
813472ce | 3441 | struct fifo_buffer *old_plan = NULL, *new_plan = NULL; |
778f271d | 3442 | int fifo_size = 0; |
82bc0194 | 3443 | int err; |
b411b363 | 3444 | |
9f4fe9ad AG |
3445 | peer_device = conn_peer_device(connection, pi->vnr); |
3446 | if (!peer_device) | |
bde89a9e | 3447 | return config_unknown_volume(connection, pi); |
9f4fe9ad | 3448 | device = peer_device->device; |
b411b363 PR |
3449 | |
3450 | exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) | |
3451 | : apv == 88 ? sizeof(struct p_rs_param) | |
3452 | + SHARED_SECRET_MAX | |
8e26f9cc PR |
3453 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
3454 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 3455 | |
e2857216 | 3456 | if (pi->size > exp_max_sz) { |
d0180171 | 3457 | drbd_err(device, "SyncParam packet too long: received %u, expected <= %u bytes\n", |
e2857216 | 3458 | pi->size, exp_max_sz); |
82bc0194 | 3459 | return -EIO; |
b411b363 PR |
3460 | } |
3461 | ||
3462 | if (apv <= 88) { | |
e658983a | 3463 | header_size = sizeof(struct p_rs_param); |
e2857216 | 3464 | data_size = pi->size - header_size; |
8e26f9cc | 3465 | } else if (apv <= 94) { |
e658983a | 3466 | header_size = sizeof(struct p_rs_param_89); |
e2857216 | 3467 | data_size = pi->size - header_size; |
0b0ba1ef | 3468 | D_ASSERT(device, data_size == 0); |
8e26f9cc | 3469 | } else { |
e658983a | 3470 | header_size = sizeof(struct p_rs_param_95); |
e2857216 | 3471 | data_size = pi->size - header_size; |
0b0ba1ef | 3472 | D_ASSERT(device, data_size == 0); |
b411b363 PR |
3473 | } |
3474 | ||
3475 | /* initialize verify_alg and csums_alg */ | |
e658983a | 3476 | p = pi->data; |
b411b363 PR |
3477 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); |
3478 | ||
9f4fe9ad | 3479 | err = drbd_recv_all(peer_device->connection, p, header_size); |
82bc0194 AG |
3480 | if (err) |
3481 | return err; | |
b411b363 | 3482 | |
0500813f | 3483 | mutex_lock(&connection->resource->conf_update); |
9f4fe9ad | 3484 | old_net_conf = peer_device->connection->net_conf; |
b30ab791 | 3485 | if (get_ldev(device)) { |
813472ce PR |
3486 | new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); |
3487 | if (!new_disk_conf) { | |
b30ab791 | 3488 | put_ldev(device); |
0500813f | 3489 | mutex_unlock(&connection->resource->conf_update); |
d0180171 | 3490 | drbd_err(device, "Allocation of new disk_conf failed\n"); |
813472ce PR |
3491 | return -ENOMEM; |
3492 | } | |
daeda1cc | 3493 | |
b30ab791 | 3494 | old_disk_conf = device->ldev->disk_conf; |
813472ce | 3495 | *new_disk_conf = *old_disk_conf; |
b411b363 | 3496 | |
6394b935 | 3497 | new_disk_conf->resync_rate = be32_to_cpu(p->resync_rate); |
813472ce | 3498 | } |
b411b363 PR |
3499 | |
3500 | if (apv >= 88) { | |
3501 | if (apv == 88) { | |
5de73827 | 3502 | if (data_size > SHARED_SECRET_MAX || data_size == 0) { |
d0180171 | 3503 | drbd_err(device, "verify-alg of wrong size, " |
5de73827 PR |
3504 | "peer wants %u, accepting only up to %u byte\n", |
3505 | data_size, SHARED_SECRET_MAX); | |
813472ce PR |
3506 | err = -EIO; |
3507 | goto reconnect; | |
b411b363 PR |
3508 | } |
3509 | ||
9f4fe9ad | 3510 | err = drbd_recv_all(peer_device->connection, p->verify_alg, data_size); |
813472ce PR |
3511 | if (err) |
3512 | goto reconnect; | |
b411b363 PR |
3513 | /* we expect NUL terminated string */ |
3514 | /* but just in case someone tries to be evil */ | |
0b0ba1ef | 3515 | D_ASSERT(device, p->verify_alg[data_size-1] == 0); |
b411b363 PR |
3516 | p->verify_alg[data_size-1] = 0; |
3517 | ||
3518 | } else /* apv >= 89 */ { | |
3519 | /* we still expect NUL terminated strings */ | |
3520 | /* but just in case someone tries to be evil */ | |
0b0ba1ef AG |
3521 | D_ASSERT(device, p->verify_alg[SHARED_SECRET_MAX-1] == 0); |
3522 | D_ASSERT(device, p->csums_alg[SHARED_SECRET_MAX-1] == 0); | |
b411b363 PR |
3523 | p->verify_alg[SHARED_SECRET_MAX-1] = 0; |
3524 | p->csums_alg[SHARED_SECRET_MAX-1] = 0; | |
3525 | } | |
3526 | ||
2ec91e0e | 3527 | if (strcmp(old_net_conf->verify_alg, p->verify_alg)) { |
b30ab791 | 3528 | if (device->state.conn == C_WF_REPORT_PARAMS) { |
d0180171 | 3529 | drbd_err(device, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", |
2ec91e0e | 3530 | old_net_conf->verify_alg, p->verify_alg); |
b411b363 PR |
3531 | goto disconnect; |
3532 | } | |
b30ab791 | 3533 | verify_tfm = drbd_crypto_alloc_digest_safe(device, |
b411b363 PR |
3534 | p->verify_alg, "verify-alg"); |
3535 | if (IS_ERR(verify_tfm)) { | |
3536 | verify_tfm = NULL; | |
3537 | goto disconnect; | |
3538 | } | |
3539 | } | |
3540 | ||
2ec91e0e | 3541 | if (apv >= 89 && strcmp(old_net_conf->csums_alg, p->csums_alg)) { |
b30ab791 | 3542 | if (device->state.conn == C_WF_REPORT_PARAMS) { |
d0180171 | 3543 | drbd_err(device, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", |
2ec91e0e | 3544 | old_net_conf->csums_alg, p->csums_alg); |
b411b363 PR |
3545 | goto disconnect; |
3546 | } | |
b30ab791 | 3547 | csums_tfm = drbd_crypto_alloc_digest_safe(device, |
b411b363 PR |
3548 | p->csums_alg, "csums-alg"); |
3549 | if (IS_ERR(csums_tfm)) { | |
3550 | csums_tfm = NULL; | |
3551 | goto disconnect; | |
3552 | } | |
3553 | } | |
3554 | ||
813472ce | 3555 | if (apv > 94 && new_disk_conf) { |
daeda1cc PR |
3556 | new_disk_conf->c_plan_ahead = be32_to_cpu(p->c_plan_ahead); |
3557 | new_disk_conf->c_delay_target = be32_to_cpu(p->c_delay_target); | |
3558 | new_disk_conf->c_fill_target = be32_to_cpu(p->c_fill_target); | |
3559 | new_disk_conf->c_max_rate = be32_to_cpu(p->c_max_rate); | |
778f271d | 3560 | |
daeda1cc | 3561 | fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ; |
b30ab791 | 3562 | if (fifo_size != device->rs_plan_s->size) { |
813472ce PR |
3563 | new_plan = fifo_alloc(fifo_size); |
3564 | if (!new_plan) { | |
d0180171 | 3565 | drbd_err(device, "kmalloc of fifo_buffer failed"); |
b30ab791 | 3566 | put_ldev(device); |
778f271d PR |
3567 | goto disconnect; |
3568 | } | |
3569 | } | |
8e26f9cc | 3570 | } |
b411b363 | 3571 | |
91fd4dad | 3572 | if (verify_tfm || csums_tfm) { |
2ec91e0e PR |
3573 | new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL); |
3574 | if (!new_net_conf) { | |
d0180171 | 3575 | drbd_err(device, "Allocation of new net_conf failed\n"); |
91fd4dad PR |
3576 | goto disconnect; |
3577 | } | |
3578 | ||
2ec91e0e | 3579 | *new_net_conf = *old_net_conf; |
91fd4dad PR |
3580 | |
3581 | if (verify_tfm) { | |
2ec91e0e PR |
3582 | strcpy(new_net_conf->verify_alg, p->verify_alg); |
3583 | new_net_conf->verify_alg_len = strlen(p->verify_alg) + 1; | |
9f4fe9ad AG |
3584 | crypto_free_hash(peer_device->connection->verify_tfm); |
3585 | peer_device->connection->verify_tfm = verify_tfm; | |
d0180171 | 3586 | drbd_info(device, "using verify-alg: \"%s\"\n", p->verify_alg); |
91fd4dad PR |
3587 | } |
3588 | if (csums_tfm) { | |
2ec91e0e PR |
3589 | strcpy(new_net_conf->csums_alg, p->csums_alg); |
3590 | new_net_conf->csums_alg_len = strlen(p->csums_alg) + 1; | |
9f4fe9ad AG |
3591 | crypto_free_hash(peer_device->connection->csums_tfm); |
3592 | peer_device->connection->csums_tfm = csums_tfm; | |
d0180171 | 3593 | drbd_info(device, "using csums-alg: \"%s\"\n", p->csums_alg); |
91fd4dad | 3594 | } |
bde89a9e | 3595 | rcu_assign_pointer(connection->net_conf, new_net_conf); |
778f271d | 3596 | } |
b411b363 PR |
3597 | } |
3598 | ||
813472ce | 3599 | if (new_disk_conf) { |
b30ab791 AG |
3600 | rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf); |
3601 | put_ldev(device); | |
813472ce PR |
3602 | } |
3603 | ||
3604 | if (new_plan) { | |
b30ab791 AG |
3605 | old_plan = device->rs_plan_s; |
3606 | rcu_assign_pointer(device->rs_plan_s, new_plan); | |
b411b363 | 3607 | } |
daeda1cc | 3608 | |
0500813f | 3609 | mutex_unlock(&connection->resource->conf_update); |
daeda1cc PR |
3610 | synchronize_rcu(); |
3611 | if (new_net_conf) | |
3612 | kfree(old_net_conf); | |
3613 | kfree(old_disk_conf); | |
813472ce | 3614 | kfree(old_plan); |
daeda1cc | 3615 | |
82bc0194 | 3616 | return 0; |
b411b363 | 3617 | |
813472ce PR |
3618 | reconnect: |
3619 | if (new_disk_conf) { | |
b30ab791 | 3620 | put_ldev(device); |
813472ce PR |
3621 | kfree(new_disk_conf); |
3622 | } | |
0500813f | 3623 | mutex_unlock(&connection->resource->conf_update); |
813472ce PR |
3624 | return -EIO; |
3625 | ||
b411b363 | 3626 | disconnect: |
813472ce PR |
3627 | kfree(new_plan); |
3628 | if (new_disk_conf) { | |
b30ab791 | 3629 | put_ldev(device); |
813472ce PR |
3630 | kfree(new_disk_conf); |
3631 | } | |
0500813f | 3632 | mutex_unlock(&connection->resource->conf_update); |
b411b363 PR |
3633 | /* just for completeness: actually not needed, |
3634 | * as this is not reached if csums_tfm was ok. */ | |
3635 | crypto_free_hash(csums_tfm); | |
3636 | /* but free the verify_tfm again, if csums_tfm did not work out */ | |
3637 | crypto_free_hash(verify_tfm); | |
9f4fe9ad | 3638 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3639 | return -EIO; |
b411b363 PR |
3640 | } |
3641 | ||
b411b363 | 3642 | /* warn if the arguments differ by more than 12.5% */ |
b30ab791 | 3643 | static void warn_if_differ_considerably(struct drbd_device *device, |
b411b363 PR |
3644 | const char *s, sector_t a, sector_t b) |
3645 | { | |
3646 | sector_t d; | |
3647 | if (a == 0 || b == 0) | |
3648 | return; | |
3649 | d = (a > b) ? (a - b) : (b - a); | |
3650 | if (d > (a>>3) || d > (b>>3)) | |
d0180171 | 3651 | drbd_warn(device, "Considerable difference in %s: %llus vs. %llus\n", s, |
b411b363 PR |
3652 | (unsigned long long)a, (unsigned long long)b); |
3653 | } | |
3654 | ||
bde89a9e | 3655 | static int receive_sizes(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3656 | { |
9f4fe9ad | 3657 | struct drbd_peer_device *peer_device; |
b30ab791 | 3658 | struct drbd_device *device; |
e658983a | 3659 | struct p_sizes *p = pi->data; |
e96c9633 | 3660 | enum determine_dev_size dd = DS_UNCHANGED; |
b411b363 PR |
3661 | sector_t p_size, p_usize, my_usize; |
3662 | int ldsc = 0; /* local disk size changed */ | |
e89b591c | 3663 | enum dds_flags ddsf; |
b411b363 | 3664 | |
9f4fe9ad AG |
3665 | peer_device = conn_peer_device(connection, pi->vnr); |
3666 | if (!peer_device) | |
bde89a9e | 3667 | return config_unknown_volume(connection, pi); |
9f4fe9ad | 3668 | device = peer_device->device; |
4a76b161 | 3669 | |
b411b363 PR |
3670 | p_size = be64_to_cpu(p->d_size); |
3671 | p_usize = be64_to_cpu(p->u_size); | |
3672 | ||
b411b363 PR |
3673 | /* just store the peer's disk size for now. |
3674 | * we still need to figure out whether we accept that. */ | |
b30ab791 | 3675 | device->p_size = p_size; |
b411b363 | 3676 | |
b30ab791 | 3677 | if (get_ldev(device)) { |
daeda1cc | 3678 | rcu_read_lock(); |
b30ab791 | 3679 | my_usize = rcu_dereference(device->ldev->disk_conf)->disk_size; |
daeda1cc PR |
3680 | rcu_read_unlock(); |
3681 | ||
b30ab791 AG |
3682 | warn_if_differ_considerably(device, "lower level device sizes", |
3683 | p_size, drbd_get_max_capacity(device->ldev)); | |
3684 | warn_if_differ_considerably(device, "user requested size", | |
daeda1cc | 3685 | p_usize, my_usize); |
b411b363 PR |
3686 | |
3687 | /* if this is the first connect, or an otherwise expected | |
3688 | * param exchange, choose the minimum */ | |
b30ab791 | 3689 | if (device->state.conn == C_WF_REPORT_PARAMS) |
daeda1cc | 3690 | p_usize = min_not_zero(my_usize, p_usize); |
b411b363 PR |
3691 | |
3692 | /* Never shrink a device with usable data during connect. | |
3693 | But allow online shrinking if we are connected. */ | |
b30ab791 AG |
3694 | if (drbd_new_dev_size(device, device->ldev, p_usize, 0) < |
3695 | drbd_get_capacity(device->this_bdev) && | |
3696 | device->state.disk >= D_OUTDATED && | |
3697 | device->state.conn < C_CONNECTED) { | |
d0180171 | 3698 | drbd_err(device, "The peer's disk size is too small!\n"); |
9f4fe9ad | 3699 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b30ab791 | 3700 | put_ldev(device); |
82bc0194 | 3701 | return -EIO; |
b411b363 | 3702 | } |
daeda1cc PR |
3703 | |
3704 | if (my_usize != p_usize) { | |
3705 | struct disk_conf *old_disk_conf, *new_disk_conf = NULL; | |
3706 | ||
3707 | new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); | |
3708 | if (!new_disk_conf) { | |
d0180171 | 3709 | drbd_err(device, "Allocation of new disk_conf failed\n"); |
b30ab791 | 3710 | put_ldev(device); |
daeda1cc PR |
3711 | return -ENOMEM; |
3712 | } | |
3713 | ||
0500813f | 3714 | mutex_lock(&connection->resource->conf_update); |
b30ab791 | 3715 | old_disk_conf = device->ldev->disk_conf; |
daeda1cc PR |
3716 | *new_disk_conf = *old_disk_conf; |
3717 | new_disk_conf->disk_size = p_usize; | |
3718 | ||
b30ab791 | 3719 | rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf); |
0500813f | 3720 | mutex_unlock(&connection->resource->conf_update); |
daeda1cc PR |
3721 | synchronize_rcu(); |
3722 | kfree(old_disk_conf); | |
3723 | ||
d0180171 | 3724 | drbd_info(device, "Peer sets u_size to %lu sectors\n", |
daeda1cc | 3725 | (unsigned long)my_usize); |
b411b363 | 3726 | } |
daeda1cc | 3727 | |
b30ab791 | 3728 | put_ldev(device); |
b411b363 | 3729 | } |
b411b363 | 3730 | |
20c68fde | 3731 | device->peer_max_bio_size = be32_to_cpu(p->max_bio_size); |
20c68fde LE |
3732 | /* Leave drbd_reconsider_max_bio_size() before drbd_determine_dev_size(). |
3733 | In case we cleared the QUEUE_FLAG_DISCARD from our queue in | |
3734 | drbd_reconsider_max_bio_size(), we can be sure that after | |
3735 | drbd_determine_dev_size() no REQ_DISCARDs are in the queue. */ | |
3736 | ||
e89b591c | 3737 | ddsf = be16_to_cpu(p->dds_flags); |
b30ab791 | 3738 | if (get_ldev(device)) { |
8fe39aac | 3739 | drbd_reconsider_max_bio_size(device, device->ldev); |
b30ab791 AG |
3740 | dd = drbd_determine_dev_size(device, ddsf, NULL); |
3741 | put_ldev(device); | |
e96c9633 | 3742 | if (dd == DS_ERROR) |
82bc0194 | 3743 | return -EIO; |
b30ab791 | 3744 | drbd_md_sync(device); |
b411b363 PR |
3745 | } else { |
3746 | /* I am diskless, need to accept the peer's size. */ | |
8fe39aac | 3747 | drbd_reconsider_max_bio_size(device, NULL); |
b30ab791 | 3748 | drbd_set_my_capacity(device, p_size); |
b411b363 PR |
3749 | } |
3750 | ||
b30ab791 AG |
3751 | if (get_ldev(device)) { |
3752 | if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev)) { | |
3753 | device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev); | |
b411b363 PR |
3754 | ldsc = 1; |
3755 | } | |
3756 | ||
b30ab791 | 3757 | put_ldev(device); |
b411b363 PR |
3758 | } |
3759 | ||
b30ab791 | 3760 | if (device->state.conn > C_WF_REPORT_PARAMS) { |
b411b363 | 3761 | if (be64_to_cpu(p->c_size) != |
b30ab791 | 3762 | drbd_get_capacity(device->this_bdev) || ldsc) { |
b411b363 PR |
3763 | /* we have different sizes, probably peer |
3764 | * needs to know my new size... */ | |
69a22773 | 3765 | drbd_send_sizes(peer_device, 0, ddsf); |
b411b363 | 3766 | } |
b30ab791 AG |
3767 | if (test_and_clear_bit(RESIZE_PENDING, &device->flags) || |
3768 | (dd == DS_GREW && device->state.conn == C_CONNECTED)) { | |
3769 | if (device->state.pdsk >= D_INCONSISTENT && | |
3770 | device->state.disk >= D_INCONSISTENT) { | |
e89b591c | 3771 | if (ddsf & DDSF_NO_RESYNC) |
d0180171 | 3772 | drbd_info(device, "Resync of new storage suppressed with --assume-clean\n"); |
e89b591c | 3773 | else |
b30ab791 | 3774 | resync_after_online_grow(device); |
e89b591c | 3775 | } else |
b30ab791 | 3776 | set_bit(RESYNC_AFTER_NEG, &device->flags); |
b411b363 PR |
3777 | } |
3778 | } | |
3779 | ||
82bc0194 | 3780 | return 0; |
b411b363 PR |
3781 | } |
3782 | ||
bde89a9e | 3783 | static int receive_uuids(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3784 | { |
9f4fe9ad | 3785 | struct drbd_peer_device *peer_device; |
b30ab791 | 3786 | struct drbd_device *device; |
e658983a | 3787 | struct p_uuids *p = pi->data; |
b411b363 | 3788 | u64 *p_uuid; |
62b0da3a | 3789 | int i, updated_uuids = 0; |
b411b363 | 3790 | |
9f4fe9ad AG |
3791 | peer_device = conn_peer_device(connection, pi->vnr); |
3792 | if (!peer_device) | |
bde89a9e | 3793 | return config_unknown_volume(connection, pi); |
9f4fe9ad | 3794 | device = peer_device->device; |
4a76b161 | 3795 | |
b411b363 | 3796 | p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); |
063eacf8 | 3797 | if (!p_uuid) { |
d0180171 | 3798 | drbd_err(device, "kmalloc of p_uuid failed\n"); |
063eacf8 JW |
3799 | return false; |
3800 | } | |
b411b363 PR |
3801 | |
3802 | for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) | |
3803 | p_uuid[i] = be64_to_cpu(p->uuid[i]); | |
3804 | ||
b30ab791 AG |
3805 | kfree(device->p_uuid); |
3806 | device->p_uuid = p_uuid; | |
b411b363 | 3807 | |
b30ab791 AG |
3808 | if (device->state.conn < C_CONNECTED && |
3809 | device->state.disk < D_INCONSISTENT && | |
3810 | device->state.role == R_PRIMARY && | |
3811 | (device->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { | |
d0180171 | 3812 | drbd_err(device, "Can only connect to data with current UUID=%016llX\n", |
b30ab791 | 3813 | (unsigned long long)device->ed_uuid); |
9f4fe9ad | 3814 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3815 | return -EIO; |
b411b363 PR |
3816 | } |
3817 | ||
b30ab791 | 3818 | if (get_ldev(device)) { |
b411b363 | 3819 | int skip_initial_sync = |
b30ab791 | 3820 | device->state.conn == C_CONNECTED && |
9f4fe9ad | 3821 | peer_device->connection->agreed_pro_version >= 90 && |
b30ab791 | 3822 | device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && |
b411b363 PR |
3823 | (p_uuid[UI_FLAGS] & 8); |
3824 | if (skip_initial_sync) { | |
d0180171 | 3825 | drbd_info(device, "Accepted new current UUID, preparing to skip initial sync\n"); |
b30ab791 | 3826 | drbd_bitmap_io(device, &drbd_bmio_clear_n_write, |
20ceb2b2 LE |
3827 | "clear_n_write from receive_uuids", |
3828 | BM_LOCKED_TEST_ALLOWED); | |
b30ab791 AG |
3829 | _drbd_uuid_set(device, UI_CURRENT, p_uuid[UI_CURRENT]); |
3830 | _drbd_uuid_set(device, UI_BITMAP, 0); | |
3831 | _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), | |
b411b363 | 3832 | CS_VERBOSE, NULL); |
b30ab791 | 3833 | drbd_md_sync(device); |
62b0da3a | 3834 | updated_uuids = 1; |
b411b363 | 3835 | } |
b30ab791 AG |
3836 | put_ldev(device); |
3837 | } else if (device->state.disk < D_INCONSISTENT && | |
3838 | device->state.role == R_PRIMARY) { | |
18a50fa2 PR |
3839 | /* I am a diskless primary, the peer just created a new current UUID |
3840 | for me. */ | |
b30ab791 | 3841 | updated_uuids = drbd_set_ed_uuid(device, p_uuid[UI_CURRENT]); |
b411b363 PR |
3842 | } |
3843 | ||
3844 | /* Before we test for the disk state, we should wait until an eventually | |
3845 | ongoing cluster wide state change is finished. That is important if | |
3846 | we are primary and are detaching from our disk. We need to see the | |
3847 | new disk state... */ | |
b30ab791 AG |
3848 | mutex_lock(device->state_mutex); |
3849 | mutex_unlock(device->state_mutex); | |
3850 | if (device->state.conn >= C_CONNECTED && device->state.disk < D_INCONSISTENT) | |
3851 | updated_uuids |= drbd_set_ed_uuid(device, p_uuid[UI_CURRENT]); | |
62b0da3a LE |
3852 | |
3853 | if (updated_uuids) | |
b30ab791 | 3854 | drbd_print_uuids(device, "receiver updated UUIDs to"); |
b411b363 | 3855 | |
82bc0194 | 3856 | return 0; |
b411b363 PR |
3857 | } |
3858 | ||
3859 | /** | |
3860 | * convert_state() - Converts the peer's view of the cluster state to our point of view | |
3861 | * @ps: The state as seen by the peer. | |
3862 | */ | |
3863 | static union drbd_state convert_state(union drbd_state ps) | |
3864 | { | |
3865 | union drbd_state ms; | |
3866 | ||
3867 | static enum drbd_conns c_tab[] = { | |
369bea63 | 3868 | [C_WF_REPORT_PARAMS] = C_WF_REPORT_PARAMS, |
b411b363 PR |
3869 | [C_CONNECTED] = C_CONNECTED, |
3870 | ||
3871 | [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, | |
3872 | [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, | |
3873 | [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ | |
3874 | [C_VERIFY_S] = C_VERIFY_T, | |
3875 | [C_MASK] = C_MASK, | |
3876 | }; | |
3877 | ||
3878 | ms.i = ps.i; | |
3879 | ||
3880 | ms.conn = c_tab[ps.conn]; | |
3881 | ms.peer = ps.role; | |
3882 | ms.role = ps.peer; | |
3883 | ms.pdsk = ps.disk; | |
3884 | ms.disk = ps.pdsk; | |
3885 | ms.peer_isp = (ps.aftr_isp | ps.user_isp); | |
3886 | ||
3887 | return ms; | |
3888 | } | |
3889 | ||
bde89a9e | 3890 | static int receive_req_state(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3891 | { |
9f4fe9ad | 3892 | struct drbd_peer_device *peer_device; |
b30ab791 | 3893 | struct drbd_device *device; |
e658983a | 3894 | struct p_req_state *p = pi->data; |
b411b363 | 3895 | union drbd_state mask, val; |
bf885f8a | 3896 | enum drbd_state_rv rv; |
b411b363 | 3897 | |
9f4fe9ad AG |
3898 | peer_device = conn_peer_device(connection, pi->vnr); |
3899 | if (!peer_device) | |
4a76b161 | 3900 | return -EIO; |
9f4fe9ad | 3901 | device = peer_device->device; |
4a76b161 | 3902 | |
b411b363 PR |
3903 | mask.i = be32_to_cpu(p->mask); |
3904 | val.i = be32_to_cpu(p->val); | |
3905 | ||
9f4fe9ad | 3906 | if (test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags) && |
b30ab791 | 3907 | mutex_is_locked(device->state_mutex)) { |
69a22773 | 3908 | drbd_send_sr_reply(peer_device, SS_CONCURRENT_ST_CHG); |
82bc0194 | 3909 | return 0; |
b411b363 PR |
3910 | } |
3911 | ||
3912 | mask = convert_state(mask); | |
3913 | val = convert_state(val); | |
3914 | ||
b30ab791 | 3915 | rv = drbd_change_state(device, CS_VERBOSE, mask, val); |
69a22773 | 3916 | drbd_send_sr_reply(peer_device, rv); |
b411b363 | 3917 | |
b30ab791 | 3918 | drbd_md_sync(device); |
b411b363 | 3919 | |
82bc0194 | 3920 | return 0; |
b411b363 PR |
3921 | } |
3922 | ||
bde89a9e | 3923 | static int receive_req_conn_state(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3924 | { |
e658983a | 3925 | struct p_req_state *p = pi->data; |
b411b363 | 3926 | union drbd_state mask, val; |
bf885f8a | 3927 | enum drbd_state_rv rv; |
b411b363 | 3928 | |
b411b363 PR |
3929 | mask.i = be32_to_cpu(p->mask); |
3930 | val.i = be32_to_cpu(p->val); | |
3931 | ||
bde89a9e AG |
3932 | if (test_bit(RESOLVE_CONFLICTS, &connection->flags) && |
3933 | mutex_is_locked(&connection->cstate_mutex)) { | |
3934 | conn_send_sr_reply(connection, SS_CONCURRENT_ST_CHG); | |
82bc0194 | 3935 | return 0; |
b411b363 PR |
3936 | } |
3937 | ||
3938 | mask = convert_state(mask); | |
3939 | val = convert_state(val); | |
3940 | ||
bde89a9e AG |
3941 | rv = conn_request_state(connection, mask, val, CS_VERBOSE | CS_LOCAL_ONLY | CS_IGN_OUTD_FAIL); |
3942 | conn_send_sr_reply(connection, rv); | |
b411b363 | 3943 | |
82bc0194 | 3944 | return 0; |
b411b363 PR |
3945 | } |
3946 | ||
bde89a9e | 3947 | static int receive_state(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3948 | { |
9f4fe9ad | 3949 | struct drbd_peer_device *peer_device; |
b30ab791 | 3950 | struct drbd_device *device; |
e658983a | 3951 | struct p_state *p = pi->data; |
4ac4aada | 3952 | union drbd_state os, ns, peer_state; |
b411b363 | 3953 | enum drbd_disk_state real_peer_disk; |
65d922c3 | 3954 | enum chg_state_flags cs_flags; |
b411b363 PR |
3955 | int rv; |
3956 | ||
9f4fe9ad AG |
3957 | peer_device = conn_peer_device(connection, pi->vnr); |
3958 | if (!peer_device) | |
bde89a9e | 3959 | return config_unknown_volume(connection, pi); |
9f4fe9ad | 3960 | device = peer_device->device; |
4a76b161 | 3961 | |
b411b363 PR |
3962 | peer_state.i = be32_to_cpu(p->state); |
3963 | ||
3964 | real_peer_disk = peer_state.disk; | |
3965 | if (peer_state.disk == D_NEGOTIATING) { | |
b30ab791 | 3966 | real_peer_disk = device->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; |
d0180171 | 3967 | drbd_info(device, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); |
b411b363 PR |
3968 | } |
3969 | ||
0500813f | 3970 | spin_lock_irq(&device->resource->req_lock); |
b411b363 | 3971 | retry: |
b30ab791 | 3972 | os = ns = drbd_read_state(device); |
0500813f | 3973 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 3974 | |
545752d5 LE |
3975 | /* If some other part of the code (asender thread, timeout) |
3976 | * already decided to close the connection again, | |
3977 | * we must not "re-establish" it here. */ | |
3978 | if (os.conn <= C_TEAR_DOWN) | |
58ffa580 | 3979 | return -ECONNRESET; |
545752d5 | 3980 | |
40424e4a LE |
3981 | /* If this is the "end of sync" confirmation, usually the peer disk |
3982 | * transitions from D_INCONSISTENT to D_UP_TO_DATE. For empty (0 bits | |
3983 | * set) resync started in PausedSyncT, or if the timing of pause-/ | |
3984 | * unpause-sync events has been "just right", the peer disk may | |
3985 | * transition from D_CONSISTENT to D_UP_TO_DATE as well. | |
3986 | */ | |
3987 | if ((os.pdsk == D_INCONSISTENT || os.pdsk == D_CONSISTENT) && | |
3988 | real_peer_disk == D_UP_TO_DATE && | |
e9ef7bb6 LE |
3989 | os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { |
3990 | /* If we are (becoming) SyncSource, but peer is still in sync | |
3991 | * preparation, ignore its uptodate-ness to avoid flapping, it | |
3992 | * will change to inconsistent once the peer reaches active | |
3993 | * syncing states. | |
3994 | * It may have changed syncer-paused flags, however, so we | |
3995 | * cannot ignore this completely. */ | |
3996 | if (peer_state.conn > C_CONNECTED && | |
3997 | peer_state.conn < C_SYNC_SOURCE) | |
3998 | real_peer_disk = D_INCONSISTENT; | |
3999 | ||
4000 | /* if peer_state changes to connected at the same time, | |
4001 | * it explicitly notifies us that it finished resync. | |
4002 | * Maybe we should finish it up, too? */ | |
4003 | else if (os.conn >= C_SYNC_SOURCE && | |
4004 | peer_state.conn == C_CONNECTED) { | |
b30ab791 AG |
4005 | if (drbd_bm_total_weight(device) <= device->rs_failed) |
4006 | drbd_resync_finished(device); | |
82bc0194 | 4007 | return 0; |
e9ef7bb6 LE |
4008 | } |
4009 | } | |
4010 | ||
02b91b55 LE |
4011 | /* explicit verify finished notification, stop sector reached. */ |
4012 | if (os.conn == C_VERIFY_T && os.disk == D_UP_TO_DATE && | |
4013 | peer_state.conn == C_CONNECTED && real_peer_disk == D_UP_TO_DATE) { | |
b30ab791 AG |
4014 | ov_out_of_sync_print(device); |
4015 | drbd_resync_finished(device); | |
58ffa580 | 4016 | return 0; |
02b91b55 LE |
4017 | } |
4018 | ||
e9ef7bb6 LE |
4019 | /* peer says his disk is inconsistent, while we think it is uptodate, |
4020 | * and this happens while the peer still thinks we have a sync going on, | |
4021 | * but we think we are already done with the sync. | |
4022 | * We ignore this to avoid flapping pdsk. | |
4023 | * This should not happen, if the peer is a recent version of drbd. */ | |
4024 | if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && | |
4025 | os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) | |
4026 | real_peer_disk = D_UP_TO_DATE; | |
4027 | ||
4ac4aada LE |
4028 | if (ns.conn == C_WF_REPORT_PARAMS) |
4029 | ns.conn = C_CONNECTED; | |
b411b363 | 4030 | |
67531718 PR |
4031 | if (peer_state.conn == C_AHEAD) |
4032 | ns.conn = C_BEHIND; | |
4033 | ||
b30ab791 AG |
4034 | if (device->p_uuid && peer_state.disk >= D_NEGOTIATING && |
4035 | get_ldev_if_state(device, D_NEGOTIATING)) { | |
b411b363 PR |
4036 | int cr; /* consider resync */ |
4037 | ||
4038 | /* if we established a new connection */ | |
4ac4aada | 4039 | cr = (os.conn < C_CONNECTED); |
b411b363 PR |
4040 | /* if we had an established connection |
4041 | * and one of the nodes newly attaches a disk */ | |
4ac4aada | 4042 | cr |= (os.conn == C_CONNECTED && |
b411b363 | 4043 | (peer_state.disk == D_NEGOTIATING || |
4ac4aada | 4044 | os.disk == D_NEGOTIATING)); |
b411b363 PR |
4045 | /* if we have both been inconsistent, and the peer has been |
4046 | * forced to be UpToDate with --overwrite-data */ | |
b30ab791 | 4047 | cr |= test_bit(CONSIDER_RESYNC, &device->flags); |
b411b363 PR |
4048 | /* if we had been plain connected, and the admin requested to |
4049 | * start a sync by "invalidate" or "invalidate-remote" */ | |
4ac4aada | 4050 | cr |= (os.conn == C_CONNECTED && |
b411b363 PR |
4051 | (peer_state.conn >= C_STARTING_SYNC_S && |
4052 | peer_state.conn <= C_WF_BITMAP_T)); | |
4053 | ||
4054 | if (cr) | |
69a22773 | 4055 | ns.conn = drbd_sync_handshake(peer_device, peer_state.role, real_peer_disk); |
b411b363 | 4056 | |
b30ab791 | 4057 | put_ldev(device); |
4ac4aada LE |
4058 | if (ns.conn == C_MASK) { |
4059 | ns.conn = C_CONNECTED; | |
b30ab791 AG |
4060 | if (device->state.disk == D_NEGOTIATING) { |
4061 | drbd_force_state(device, NS(disk, D_FAILED)); | |
b411b363 | 4062 | } else if (peer_state.disk == D_NEGOTIATING) { |
d0180171 | 4063 | drbd_err(device, "Disk attach process on the peer node was aborted.\n"); |
b411b363 | 4064 | peer_state.disk = D_DISKLESS; |
580b9767 | 4065 | real_peer_disk = D_DISKLESS; |
b411b363 | 4066 | } else { |
9f4fe9ad | 4067 | if (test_and_clear_bit(CONN_DRY_RUN, &peer_device->connection->flags)) |
82bc0194 | 4068 | return -EIO; |
0b0ba1ef | 4069 | D_ASSERT(device, os.conn == C_WF_REPORT_PARAMS); |
9f4fe9ad | 4070 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 4071 | return -EIO; |
b411b363 PR |
4072 | } |
4073 | } | |
4074 | } | |
4075 | ||
0500813f | 4076 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 4077 | if (os.i != drbd_read_state(device).i) |
b411b363 | 4078 | goto retry; |
b30ab791 | 4079 | clear_bit(CONSIDER_RESYNC, &device->flags); |
b411b363 PR |
4080 | ns.peer = peer_state.role; |
4081 | ns.pdsk = real_peer_disk; | |
4082 | ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); | |
4ac4aada | 4083 | if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) |
b30ab791 | 4084 | ns.disk = device->new_state_tmp.disk; |
4ac4aada | 4085 | cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); |
b30ab791 AG |
4086 | if (ns.pdsk == D_CONSISTENT && drbd_suspended(device) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && |
4087 | test_bit(NEW_CUR_UUID, &device->flags)) { | |
8554df1c | 4088 | /* Do not allow tl_restart(RESEND) for a rebooted peer. We can only allow this |
481c6f50 | 4089 | for temporal network outages! */ |
0500813f | 4090 | spin_unlock_irq(&device->resource->req_lock); |
d0180171 | 4091 | drbd_err(device, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); |
9f4fe9ad | 4092 | tl_clear(peer_device->connection); |
b30ab791 AG |
4093 | drbd_uuid_new_current(device); |
4094 | clear_bit(NEW_CUR_UUID, &device->flags); | |
9f4fe9ad | 4095 | conn_request_state(peer_device->connection, NS2(conn, C_PROTOCOL_ERROR, susp, 0), CS_HARD); |
82bc0194 | 4096 | return -EIO; |
481c6f50 | 4097 | } |
b30ab791 AG |
4098 | rv = _drbd_set_state(device, ns, cs_flags, NULL); |
4099 | ns = drbd_read_state(device); | |
0500813f | 4100 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
4101 | |
4102 | if (rv < SS_SUCCESS) { | |
9f4fe9ad | 4103 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 4104 | return -EIO; |
b411b363 PR |
4105 | } |
4106 | ||
4ac4aada LE |
4107 | if (os.conn > C_WF_REPORT_PARAMS) { |
4108 | if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && | |
b411b363 PR |
4109 | peer_state.disk != D_NEGOTIATING ) { |
4110 | /* we want resync, peer has not yet decided to sync... */ | |
4111 | /* Nowadays only used when forcing a node into primary role and | |
4112 | setting its disk to UpToDate with that */ | |
69a22773 AG |
4113 | drbd_send_uuids(peer_device); |
4114 | drbd_send_current_state(peer_device); | |
b411b363 PR |
4115 | } |
4116 | } | |
4117 | ||
b30ab791 | 4118 | clear_bit(DISCARD_MY_DATA, &device->flags); |
b411b363 | 4119 | |
b30ab791 | 4120 | drbd_md_sync(device); /* update connected indicator, la_size_sect, ... */ |
b411b363 | 4121 | |
82bc0194 | 4122 | return 0; |
b411b363 PR |
4123 | } |
4124 | ||
bde89a9e | 4125 | static int receive_sync_uuid(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4126 | { |
9f4fe9ad | 4127 | struct drbd_peer_device *peer_device; |
b30ab791 | 4128 | struct drbd_device *device; |
e658983a | 4129 | struct p_rs_uuid *p = pi->data; |
4a76b161 | 4130 | |
9f4fe9ad AG |
4131 | peer_device = conn_peer_device(connection, pi->vnr); |
4132 | if (!peer_device) | |
4a76b161 | 4133 | return -EIO; |
9f4fe9ad | 4134 | device = peer_device->device; |
b411b363 | 4135 | |
b30ab791 AG |
4136 | wait_event(device->misc_wait, |
4137 | device->state.conn == C_WF_SYNC_UUID || | |
4138 | device->state.conn == C_BEHIND || | |
4139 | device->state.conn < C_CONNECTED || | |
4140 | device->state.disk < D_NEGOTIATING); | |
b411b363 | 4141 | |
0b0ba1ef | 4142 | /* D_ASSERT(device, device->state.conn == C_WF_SYNC_UUID ); */ |
b411b363 | 4143 | |
b411b363 PR |
4144 | /* Here the _drbd_uuid_ functions are right, current should |
4145 | _not_ be rotated into the history */ | |
b30ab791 AG |
4146 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
4147 | _drbd_uuid_set(device, UI_CURRENT, be64_to_cpu(p->uuid)); | |
4148 | _drbd_uuid_set(device, UI_BITMAP, 0UL); | |
b411b363 | 4149 | |
b30ab791 AG |
4150 | drbd_print_uuids(device, "updated sync uuid"); |
4151 | drbd_start_resync(device, C_SYNC_TARGET); | |
b411b363 | 4152 | |
b30ab791 | 4153 | put_ldev(device); |
b411b363 | 4154 | } else |
d0180171 | 4155 | drbd_err(device, "Ignoring SyncUUID packet!\n"); |
b411b363 | 4156 | |
82bc0194 | 4157 | return 0; |
b411b363 PR |
4158 | } |
4159 | ||
2c46407d AG |
4160 | /** |
4161 | * receive_bitmap_plain | |
4162 | * | |
4163 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4164 | * code upon failure. | |
4165 | */ | |
4166 | static int | |
69a22773 | 4167 | receive_bitmap_plain(struct drbd_peer_device *peer_device, unsigned int size, |
e658983a | 4168 | unsigned long *p, struct bm_xfer_ctx *c) |
b411b363 | 4169 | { |
50d0b1ad | 4170 | unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - |
69a22773 | 4171 | drbd_header_size(peer_device->connection); |
e658983a | 4172 | unsigned int num_words = min_t(size_t, data_size / sizeof(*p), |
50d0b1ad | 4173 | c->bm_words - c->word_offset); |
e658983a | 4174 | unsigned int want = num_words * sizeof(*p); |
2c46407d | 4175 | int err; |
b411b363 | 4176 | |
50d0b1ad | 4177 | if (want != size) { |
69a22773 | 4178 | drbd_err(peer_device, "%s:want (%u) != size (%u)\n", __func__, want, size); |
2c46407d | 4179 | return -EIO; |
b411b363 PR |
4180 | } |
4181 | if (want == 0) | |
2c46407d | 4182 | return 0; |
69a22773 | 4183 | err = drbd_recv_all(peer_device->connection, p, want); |
82bc0194 | 4184 | if (err) |
2c46407d | 4185 | return err; |
b411b363 | 4186 | |
69a22773 | 4187 | drbd_bm_merge_lel(peer_device->device, c->word_offset, num_words, p); |
b411b363 PR |
4188 | |
4189 | c->word_offset += num_words; | |
4190 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
4191 | if (c->bit_offset > c->bm_bits) | |
4192 | c->bit_offset = c->bm_bits; | |
4193 | ||
2c46407d | 4194 | return 1; |
b411b363 PR |
4195 | } |
4196 | ||
a02d1240 AG |
4197 | static enum drbd_bitmap_code dcbp_get_code(struct p_compressed_bm *p) |
4198 | { | |
4199 | return (enum drbd_bitmap_code)(p->encoding & 0x0f); | |
4200 | } | |
4201 | ||
4202 | static int dcbp_get_start(struct p_compressed_bm *p) | |
4203 | { | |
4204 | return (p->encoding & 0x80) != 0; | |
4205 | } | |
4206 | ||
4207 | static int dcbp_get_pad_bits(struct p_compressed_bm *p) | |
4208 | { | |
4209 | return (p->encoding >> 4) & 0x7; | |
4210 | } | |
4211 | ||
2c46407d AG |
4212 | /** |
4213 | * recv_bm_rle_bits | |
4214 | * | |
4215 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4216 | * code upon failure. | |
4217 | */ | |
4218 | static int | |
69a22773 | 4219 | recv_bm_rle_bits(struct drbd_peer_device *peer_device, |
b411b363 | 4220 | struct p_compressed_bm *p, |
c6d25cfe PR |
4221 | struct bm_xfer_ctx *c, |
4222 | unsigned int len) | |
b411b363 PR |
4223 | { |
4224 | struct bitstream bs; | |
4225 | u64 look_ahead; | |
4226 | u64 rl; | |
4227 | u64 tmp; | |
4228 | unsigned long s = c->bit_offset; | |
4229 | unsigned long e; | |
a02d1240 | 4230 | int toggle = dcbp_get_start(p); |
b411b363 PR |
4231 | int have; |
4232 | int bits; | |
4233 | ||
a02d1240 | 4234 | bitstream_init(&bs, p->code, len, dcbp_get_pad_bits(p)); |
b411b363 PR |
4235 | |
4236 | bits = bitstream_get_bits(&bs, &look_ahead, 64); | |
4237 | if (bits < 0) | |
2c46407d | 4238 | return -EIO; |
b411b363 PR |
4239 | |
4240 | for (have = bits; have > 0; s += rl, toggle = !toggle) { | |
4241 | bits = vli_decode_bits(&rl, look_ahead); | |
4242 | if (bits <= 0) | |
2c46407d | 4243 | return -EIO; |
b411b363 PR |
4244 | |
4245 | if (toggle) { | |
4246 | e = s + rl -1; | |
4247 | if (e >= c->bm_bits) { | |
69a22773 | 4248 | drbd_err(peer_device, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); |
2c46407d | 4249 | return -EIO; |
b411b363 | 4250 | } |
69a22773 | 4251 | _drbd_bm_set_bits(peer_device->device, s, e); |
b411b363 PR |
4252 | } |
4253 | ||
4254 | if (have < bits) { | |
69a22773 | 4255 | drbd_err(peer_device, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", |
b411b363 PR |
4256 | have, bits, look_ahead, |
4257 | (unsigned int)(bs.cur.b - p->code), | |
4258 | (unsigned int)bs.buf_len); | |
2c46407d | 4259 | return -EIO; |
b411b363 | 4260 | } |
d2da5b0c LE |
4261 | /* if we consumed all 64 bits, assign 0; >> 64 is "undefined"; */ |
4262 | if (likely(bits < 64)) | |
4263 | look_ahead >>= bits; | |
4264 | else | |
4265 | look_ahead = 0; | |
b411b363 PR |
4266 | have -= bits; |
4267 | ||
4268 | bits = bitstream_get_bits(&bs, &tmp, 64 - have); | |
4269 | if (bits < 0) | |
2c46407d | 4270 | return -EIO; |
b411b363 PR |
4271 | look_ahead |= tmp << have; |
4272 | have += bits; | |
4273 | } | |
4274 | ||
4275 | c->bit_offset = s; | |
4276 | bm_xfer_ctx_bit_to_word_offset(c); | |
4277 | ||
2c46407d | 4278 | return (s != c->bm_bits); |
b411b363 PR |
4279 | } |
4280 | ||
2c46407d AG |
4281 | /** |
4282 | * decode_bitmap_c | |
4283 | * | |
4284 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4285 | * code upon failure. | |
4286 | */ | |
4287 | static int | |
69a22773 | 4288 | decode_bitmap_c(struct drbd_peer_device *peer_device, |
b411b363 | 4289 | struct p_compressed_bm *p, |
c6d25cfe PR |
4290 | struct bm_xfer_ctx *c, |
4291 | unsigned int len) | |
b411b363 | 4292 | { |
a02d1240 | 4293 | if (dcbp_get_code(p) == RLE_VLI_Bits) |
69a22773 | 4294 | return recv_bm_rle_bits(peer_device, p, c, len - sizeof(*p)); |
b411b363 PR |
4295 | |
4296 | /* other variants had been implemented for evaluation, | |
4297 | * but have been dropped as this one turned out to be "best" | |
4298 | * during all our tests. */ | |
4299 | ||
69a22773 AG |
4300 | drbd_err(peer_device, "receive_bitmap_c: unknown encoding %u\n", p->encoding); |
4301 | conn_request_state(peer_device->connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); | |
2c46407d | 4302 | return -EIO; |
b411b363 PR |
4303 | } |
4304 | ||
b30ab791 | 4305 | void INFO_bm_xfer_stats(struct drbd_device *device, |
b411b363 PR |
4306 | const char *direction, struct bm_xfer_ctx *c) |
4307 | { | |
4308 | /* what would it take to transfer it "plaintext" */ | |
a6b32bc3 | 4309 | unsigned int header_size = drbd_header_size(first_peer_device(device)->connection); |
50d0b1ad AG |
4310 | unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; |
4311 | unsigned int plain = | |
4312 | header_size * (DIV_ROUND_UP(c->bm_words, data_size) + 1) + | |
4313 | c->bm_words * sizeof(unsigned long); | |
4314 | unsigned int total = c->bytes[0] + c->bytes[1]; | |
4315 | unsigned int r; | |
b411b363 PR |
4316 | |
4317 | /* total can not be zero. but just in case: */ | |
4318 | if (total == 0) | |
4319 | return; | |
4320 | ||
4321 | /* don't report if not compressed */ | |
4322 | if (total >= plain) | |
4323 | return; | |
4324 | ||
4325 | /* total < plain. check for overflow, still */ | |
4326 | r = (total > UINT_MAX/1000) ? (total / (plain/1000)) | |
4327 | : (1000 * total / plain); | |
4328 | ||
4329 | if (r > 1000) | |
4330 | r = 1000; | |
4331 | ||
4332 | r = 1000 - r; | |
d0180171 | 4333 | drbd_info(device, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " |
b411b363 PR |
4334 | "total %u; compression: %u.%u%%\n", |
4335 | direction, | |
4336 | c->bytes[1], c->packets[1], | |
4337 | c->bytes[0], c->packets[0], | |
4338 | total, r/10, r % 10); | |
4339 | } | |
4340 | ||
4341 | /* Since we are processing the bitfield from lower addresses to higher, | |
4342 | it does not matter if the process it in 32 bit chunks or 64 bit | |
4343 | chunks as long as it is little endian. (Understand it as byte stream, | |
4344 | beginning with the lowest byte...) If we would use big endian | |
4345 | we would need to process it from the highest address to the lowest, | |
4346 | in order to be agnostic to the 32 vs 64 bits issue. | |
4347 | ||
4348 | returns 0 on failure, 1 if we successfully received it. */ | |
bde89a9e | 4349 | static int receive_bitmap(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4350 | { |
9f4fe9ad | 4351 | struct drbd_peer_device *peer_device; |
b30ab791 | 4352 | struct drbd_device *device; |
b411b363 | 4353 | struct bm_xfer_ctx c; |
2c46407d | 4354 | int err; |
4a76b161 | 4355 | |
9f4fe9ad AG |
4356 | peer_device = conn_peer_device(connection, pi->vnr); |
4357 | if (!peer_device) | |
4a76b161 | 4358 | return -EIO; |
9f4fe9ad | 4359 | device = peer_device->device; |
b411b363 | 4360 | |
b30ab791 | 4361 | drbd_bm_lock(device, "receive bitmap", BM_LOCKED_SET_ALLOWED); |
20ceb2b2 LE |
4362 | /* you are supposed to send additional out-of-sync information |
4363 | * if you actually set bits during this phase */ | |
b411b363 | 4364 | |
b411b363 | 4365 | c = (struct bm_xfer_ctx) { |
b30ab791 AG |
4366 | .bm_bits = drbd_bm_bits(device), |
4367 | .bm_words = drbd_bm_words(device), | |
b411b363 PR |
4368 | }; |
4369 | ||
2c46407d | 4370 | for(;;) { |
e658983a | 4371 | if (pi->cmd == P_BITMAP) |
69a22773 | 4372 | err = receive_bitmap_plain(peer_device, pi->size, pi->data, &c); |
e658983a | 4373 | else if (pi->cmd == P_COMPRESSED_BITMAP) { |
b411b363 PR |
4374 | /* MAYBE: sanity check that we speak proto >= 90, |
4375 | * and the feature is enabled! */ | |
e658983a | 4376 | struct p_compressed_bm *p = pi->data; |
b411b363 | 4377 | |
bde89a9e | 4378 | if (pi->size > DRBD_SOCKET_BUFFER_SIZE - drbd_header_size(connection)) { |
d0180171 | 4379 | drbd_err(device, "ReportCBitmap packet too large\n"); |
82bc0194 | 4380 | err = -EIO; |
b411b363 PR |
4381 | goto out; |
4382 | } | |
e658983a | 4383 | if (pi->size <= sizeof(*p)) { |
d0180171 | 4384 | drbd_err(device, "ReportCBitmap packet too small (l:%u)\n", pi->size); |
82bc0194 | 4385 | err = -EIO; |
78fcbdae | 4386 | goto out; |
b411b363 | 4387 | } |
9f4fe9ad | 4388 | err = drbd_recv_all(peer_device->connection, p, pi->size); |
e658983a AG |
4389 | if (err) |
4390 | goto out; | |
69a22773 | 4391 | err = decode_bitmap_c(peer_device, p, &c, pi->size); |
b411b363 | 4392 | } else { |
d0180171 | 4393 | drbd_warn(device, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", pi->cmd); |
82bc0194 | 4394 | err = -EIO; |
b411b363 PR |
4395 | goto out; |
4396 | } | |
4397 | ||
e2857216 | 4398 | c.packets[pi->cmd == P_BITMAP]++; |
bde89a9e | 4399 | c.bytes[pi->cmd == P_BITMAP] += drbd_header_size(connection) + pi->size; |
b411b363 | 4400 | |
2c46407d AG |
4401 | if (err <= 0) { |
4402 | if (err < 0) | |
4403 | goto out; | |
b411b363 | 4404 | break; |
2c46407d | 4405 | } |
9f4fe9ad | 4406 | err = drbd_recv_header(peer_device->connection, pi); |
82bc0194 | 4407 | if (err) |
b411b363 | 4408 | goto out; |
2c46407d | 4409 | } |
b411b363 | 4410 | |
b30ab791 | 4411 | INFO_bm_xfer_stats(device, "receive", &c); |
b411b363 | 4412 | |
b30ab791 | 4413 | if (device->state.conn == C_WF_BITMAP_T) { |
de1f8e4a AG |
4414 | enum drbd_state_rv rv; |
4415 | ||
b30ab791 | 4416 | err = drbd_send_bitmap(device); |
82bc0194 | 4417 | if (err) |
b411b363 PR |
4418 | goto out; |
4419 | /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ | |
b30ab791 | 4420 | rv = _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); |
0b0ba1ef | 4421 | D_ASSERT(device, rv == SS_SUCCESS); |
b30ab791 | 4422 | } else if (device->state.conn != C_WF_BITMAP_S) { |
b411b363 PR |
4423 | /* admin may have requested C_DISCONNECTING, |
4424 | * other threads may have noticed network errors */ | |
d0180171 | 4425 | drbd_info(device, "unexpected cstate (%s) in receive_bitmap\n", |
b30ab791 | 4426 | drbd_conn_str(device->state.conn)); |
b411b363 | 4427 | } |
82bc0194 | 4428 | err = 0; |
b411b363 | 4429 | |
b411b363 | 4430 | out: |
b30ab791 AG |
4431 | drbd_bm_unlock(device); |
4432 | if (!err && device->state.conn == C_WF_BITMAP_S) | |
4433 | drbd_start_resync(device, C_SYNC_SOURCE); | |
82bc0194 | 4434 | return err; |
b411b363 PR |
4435 | } |
4436 | ||
bde89a9e | 4437 | static int receive_skip(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4438 | { |
1ec861eb | 4439 | drbd_warn(connection, "skipping unknown optional packet type %d, l: %d!\n", |
e2857216 | 4440 | pi->cmd, pi->size); |
b411b363 | 4441 | |
bde89a9e | 4442 | return ignore_remaining_packet(connection, pi); |
b411b363 PR |
4443 | } |
4444 | ||
bde89a9e | 4445 | static int receive_UnplugRemote(struct drbd_connection *connection, struct packet_info *pi) |
0ced55a3 | 4446 | { |
e7f52dfb LE |
4447 | /* Make sure we've acked all the TCP data associated |
4448 | * with the data requests being unplugged */ | |
bde89a9e | 4449 | drbd_tcp_quickack(connection->data.socket); |
0ced55a3 | 4450 | |
82bc0194 | 4451 | return 0; |
0ced55a3 PR |
4452 | } |
4453 | ||
bde89a9e | 4454 | static int receive_out_of_sync(struct drbd_connection *connection, struct packet_info *pi) |
73a01a18 | 4455 | { |
9f4fe9ad | 4456 | struct drbd_peer_device *peer_device; |
b30ab791 | 4457 | struct drbd_device *device; |
e658983a | 4458 | struct p_block_desc *p = pi->data; |
4a76b161 | 4459 | |
9f4fe9ad AG |
4460 | peer_device = conn_peer_device(connection, pi->vnr); |
4461 | if (!peer_device) | |
4a76b161 | 4462 | return -EIO; |
9f4fe9ad | 4463 | device = peer_device->device; |
73a01a18 | 4464 | |
b30ab791 | 4465 | switch (device->state.conn) { |
f735e363 LE |
4466 | case C_WF_SYNC_UUID: |
4467 | case C_WF_BITMAP_T: | |
4468 | case C_BEHIND: | |
4469 | break; | |
4470 | default: | |
d0180171 | 4471 | drbd_err(device, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n", |
b30ab791 | 4472 | drbd_conn_str(device->state.conn)); |
f735e363 LE |
4473 | } |
4474 | ||
b30ab791 | 4475 | drbd_set_out_of_sync(device, be64_to_cpu(p->sector), be32_to_cpu(p->blksize)); |
73a01a18 | 4476 | |
82bc0194 | 4477 | return 0; |
73a01a18 PR |
4478 | } |
4479 | ||
02918be2 PR |
4480 | struct data_cmd { |
4481 | int expect_payload; | |
4482 | size_t pkt_size; | |
bde89a9e | 4483 | int (*fn)(struct drbd_connection *, struct packet_info *); |
02918be2 PR |
4484 | }; |
4485 | ||
4486 | static struct data_cmd drbd_cmd_handler[] = { | |
4487 | [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, | |
4488 | [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, | |
4489 | [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , | |
4490 | [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , | |
e658983a AG |
4491 | [P_BITMAP] = { 1, 0, receive_bitmap } , |
4492 | [P_COMPRESSED_BITMAP] = { 1, 0, receive_bitmap } , | |
4493 | [P_UNPLUG_REMOTE] = { 0, 0, receive_UnplugRemote }, | |
02918be2 PR |
4494 | [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, |
4495 | [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
e658983a AG |
4496 | [P_SYNC_PARAM] = { 1, 0, receive_SyncParam }, |
4497 | [P_SYNC_PARAM89] = { 1, 0, receive_SyncParam }, | |
02918be2 PR |
4498 | [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, |
4499 | [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, | |
4500 | [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, | |
4501 | [P_STATE] = { 0, sizeof(struct p_state), receive_state }, | |
4502 | [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, | |
4503 | [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, | |
4504 | [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
4505 | [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
4506 | [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
4507 | [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, | |
73a01a18 | 4508 | [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync }, |
4a76b161 | 4509 | [P_CONN_ST_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_conn_state }, |
036b17ea | 4510 | [P_PROTOCOL_UPDATE] = { 1, sizeof(struct p_protocol), receive_protocol }, |
a0fb3c47 | 4511 | [P_TRIM] = { 0, sizeof(struct p_trim), receive_Data }, |
b411b363 PR |
4512 | }; |
4513 | ||
bde89a9e | 4514 | static void drbdd(struct drbd_connection *connection) |
b411b363 | 4515 | { |
77351055 | 4516 | struct packet_info pi; |
02918be2 | 4517 | size_t shs; /* sub header size */ |
82bc0194 | 4518 | int err; |
b411b363 | 4519 | |
bde89a9e | 4520 | while (get_t_state(&connection->receiver) == RUNNING) { |
deebe195 | 4521 | struct data_cmd *cmd; |
b411b363 | 4522 | |
bde89a9e AG |
4523 | drbd_thread_current_set_cpu(&connection->receiver); |
4524 | if (drbd_recv_header(connection, &pi)) | |
02918be2 | 4525 | goto err_out; |
b411b363 | 4526 | |
deebe195 | 4527 | cmd = &drbd_cmd_handler[pi.cmd]; |
4a76b161 | 4528 | if (unlikely(pi.cmd >= ARRAY_SIZE(drbd_cmd_handler) || !cmd->fn)) { |
1ec861eb | 4529 | drbd_err(connection, "Unexpected data packet %s (0x%04x)", |
2fcb8f30 | 4530 | cmdname(pi.cmd), pi.cmd); |
02918be2 | 4531 | goto err_out; |
0b33a916 | 4532 | } |
b411b363 | 4533 | |
e658983a AG |
4534 | shs = cmd->pkt_size; |
4535 | if (pi.size > shs && !cmd->expect_payload) { | |
1ec861eb | 4536 | drbd_err(connection, "No payload expected %s l:%d\n", |
2fcb8f30 | 4537 | cmdname(pi.cmd), pi.size); |
02918be2 | 4538 | goto err_out; |
b411b363 | 4539 | } |
b411b363 | 4540 | |
c13f7e1a | 4541 | if (shs) { |
bde89a9e | 4542 | err = drbd_recv_all_warn(connection, pi.data, shs); |
a5c31904 | 4543 | if (err) |
c13f7e1a | 4544 | goto err_out; |
e2857216 | 4545 | pi.size -= shs; |
c13f7e1a LE |
4546 | } |
4547 | ||
bde89a9e | 4548 | err = cmd->fn(connection, &pi); |
4a76b161 | 4549 | if (err) { |
1ec861eb | 4550 | drbd_err(connection, "error receiving %s, e: %d l: %d!\n", |
9f5bdc33 | 4551 | cmdname(pi.cmd), err, pi.size); |
02918be2 | 4552 | goto err_out; |
b411b363 PR |
4553 | } |
4554 | } | |
82bc0194 | 4555 | return; |
b411b363 | 4556 | |
82bc0194 | 4557 | err_out: |
bde89a9e | 4558 | conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); |
b411b363 PR |
4559 | } |
4560 | ||
bde89a9e | 4561 | static void conn_disconnect(struct drbd_connection *connection) |
b411b363 | 4562 | { |
c06ece6b | 4563 | struct drbd_peer_device *peer_device; |
bbeb641c | 4564 | enum drbd_conns oc; |
376694a0 | 4565 | int vnr; |
b411b363 | 4566 | |
bde89a9e | 4567 | if (connection->cstate == C_STANDALONE) |
b411b363 | 4568 | return; |
b411b363 | 4569 | |
545752d5 LE |
4570 | /* We are about to start the cleanup after connection loss. |
4571 | * Make sure drbd_make_request knows about that. | |
4572 | * Usually we should be in some network failure state already, | |
4573 | * but just in case we are not, we fix it up here. | |
4574 | */ | |
bde89a9e | 4575 | conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); |
545752d5 | 4576 | |
b411b363 | 4577 | /* asender does not clean up anything. it must not interfere, either */ |
bde89a9e AG |
4578 | drbd_thread_stop(&connection->asender); |
4579 | drbd_free_sock(connection); | |
360cc740 | 4580 | |
c141ebda | 4581 | rcu_read_lock(); |
c06ece6b AG |
4582 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
4583 | struct drbd_device *device = peer_device->device; | |
b30ab791 | 4584 | kref_get(&device->kref); |
c141ebda | 4585 | rcu_read_unlock(); |
69a22773 | 4586 | drbd_disconnected(peer_device); |
c06ece6b | 4587 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda PR |
4588 | rcu_read_lock(); |
4589 | } | |
4590 | rcu_read_unlock(); | |
4591 | ||
bde89a9e | 4592 | if (!list_empty(&connection->current_epoch->list)) |
1ec861eb | 4593 | drbd_err(connection, "ASSERTION FAILED: connection->current_epoch->list not empty\n"); |
12038a3a | 4594 | /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ |
bde89a9e AG |
4595 | atomic_set(&connection->current_epoch->epoch_size, 0); |
4596 | connection->send.seen_any_write_yet = false; | |
12038a3a | 4597 | |
1ec861eb | 4598 | drbd_info(connection, "Connection closed\n"); |
360cc740 | 4599 | |
bde89a9e AG |
4600 | if (conn_highest_role(connection) == R_PRIMARY && conn_highest_pdsk(connection) >= D_UNKNOWN) |
4601 | conn_try_outdate_peer_async(connection); | |
cb703454 | 4602 | |
0500813f | 4603 | spin_lock_irq(&connection->resource->req_lock); |
bde89a9e | 4604 | oc = connection->cstate; |
bbeb641c | 4605 | if (oc >= C_UNCONNECTED) |
bde89a9e | 4606 | _conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE); |
bbeb641c | 4607 | |
0500813f | 4608 | spin_unlock_irq(&connection->resource->req_lock); |
360cc740 | 4609 | |
f3dfa40a | 4610 | if (oc == C_DISCONNECTING) |
bde89a9e | 4611 | conn_request_state(connection, NS(conn, C_STANDALONE), CS_VERBOSE | CS_HARD); |
360cc740 PR |
4612 | } |
4613 | ||
69a22773 | 4614 | static int drbd_disconnected(struct drbd_peer_device *peer_device) |
360cc740 | 4615 | { |
69a22773 | 4616 | struct drbd_device *device = peer_device->device; |
360cc740 | 4617 | unsigned int i; |
b411b363 | 4618 | |
85719573 | 4619 | /* wait for current activity to cease. */ |
0500813f | 4620 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 AG |
4621 | _drbd_wait_ee_list_empty(device, &device->active_ee); |
4622 | _drbd_wait_ee_list_empty(device, &device->sync_ee); | |
4623 | _drbd_wait_ee_list_empty(device, &device->read_ee); | |
0500813f | 4624 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
4625 | |
4626 | /* We do not have data structures that would allow us to | |
4627 | * get the rs_pending_cnt down to 0 again. | |
4628 | * * On C_SYNC_TARGET we do not have any data structures describing | |
4629 | * the pending RSDataRequest's we have sent. | |
4630 | * * On C_SYNC_SOURCE there is no data structure that tracks | |
4631 | * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. | |
4632 | * And no, it is not the sum of the reference counts in the | |
4633 | * resync_LRU. The resync_LRU tracks the whole operation including | |
4634 | * the disk-IO, while the rs_pending_cnt only tracks the blocks | |
4635 | * on the fly. */ | |
b30ab791 AG |
4636 | drbd_rs_cancel_all(device); |
4637 | device->rs_total = 0; | |
4638 | device->rs_failed = 0; | |
4639 | atomic_set(&device->rs_pending_cnt, 0); | |
4640 | wake_up(&device->misc_wait); | |
b411b363 | 4641 | |
b30ab791 AG |
4642 | del_timer_sync(&device->resync_timer); |
4643 | resync_timer_fn((unsigned long)device); | |
b411b363 | 4644 | |
b411b363 PR |
4645 | /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, |
4646 | * w_make_resync_request etc. which may still be on the worker queue | |
4647 | * to be "canceled" */ | |
b5043c5e | 4648 | drbd_flush_workqueue(&peer_device->connection->sender_work); |
b411b363 | 4649 | |
b30ab791 | 4650 | drbd_finish_peer_reqs(device); |
b411b363 | 4651 | |
d10b4ea3 PR |
4652 | /* This second workqueue flush is necessary, since drbd_finish_peer_reqs() |
4653 | might have issued a work again. The one before drbd_finish_peer_reqs() is | |
4654 | necessary to reclain net_ee in drbd_finish_peer_reqs(). */ | |
b5043c5e | 4655 | drbd_flush_workqueue(&peer_device->connection->sender_work); |
d10b4ea3 | 4656 | |
08332d73 LE |
4657 | /* need to do it again, drbd_finish_peer_reqs() may have populated it |
4658 | * again via drbd_try_clear_on_disk_bm(). */ | |
b30ab791 | 4659 | drbd_rs_cancel_all(device); |
b411b363 | 4660 | |
b30ab791 AG |
4661 | kfree(device->p_uuid); |
4662 | device->p_uuid = NULL; | |
b411b363 | 4663 | |
b30ab791 | 4664 | if (!drbd_suspended(device)) |
69a22773 | 4665 | tl_clear(peer_device->connection); |
b411b363 | 4666 | |
b30ab791 | 4667 | drbd_md_sync(device); |
b411b363 | 4668 | |
20ceb2b2 LE |
4669 | /* serialize with bitmap writeout triggered by the state change, |
4670 | * if any. */ | |
b30ab791 | 4671 | wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags)); |
20ceb2b2 | 4672 | |
b411b363 PR |
4673 | /* tcp_close and release of sendpage pages can be deferred. I don't |
4674 | * want to use SO_LINGER, because apparently it can be deferred for | |
4675 | * more than 20 seconds (longest time I checked). | |
4676 | * | |
4677 | * Actually we don't care for exactly when the network stack does its | |
4678 | * put_page(), but release our reference on these pages right here. | |
4679 | */ | |
b30ab791 | 4680 | i = drbd_free_peer_reqs(device, &device->net_ee); |
b411b363 | 4681 | if (i) |
d0180171 | 4682 | drbd_info(device, "net_ee not empty, killed %u entries\n", i); |
b30ab791 | 4683 | i = atomic_read(&device->pp_in_use_by_net); |
435f0740 | 4684 | if (i) |
d0180171 | 4685 | drbd_info(device, "pp_in_use_by_net = %d, expected 0\n", i); |
b30ab791 | 4686 | i = atomic_read(&device->pp_in_use); |
b411b363 | 4687 | if (i) |
d0180171 | 4688 | drbd_info(device, "pp_in_use = %d, expected 0\n", i); |
b411b363 | 4689 | |
0b0ba1ef AG |
4690 | D_ASSERT(device, list_empty(&device->read_ee)); |
4691 | D_ASSERT(device, list_empty(&device->active_ee)); | |
4692 | D_ASSERT(device, list_empty(&device->sync_ee)); | |
4693 | D_ASSERT(device, list_empty(&device->done_ee)); | |
b411b363 | 4694 | |
360cc740 | 4695 | return 0; |
b411b363 PR |
4696 | } |
4697 | ||
4698 | /* | |
4699 | * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version | |
4700 | * we can agree on is stored in agreed_pro_version. | |
4701 | * | |
4702 | * feature flags and the reserved array should be enough room for future | |
4703 | * enhancements of the handshake protocol, and possible plugins... | |
4704 | * | |
4705 | * for now, they are expected to be zero, but ignored. | |
4706 | */ | |
bde89a9e | 4707 | static int drbd_send_features(struct drbd_connection *connection) |
b411b363 | 4708 | { |
9f5bdc33 AG |
4709 | struct drbd_socket *sock; |
4710 | struct p_connection_features *p; | |
b411b363 | 4711 | |
bde89a9e AG |
4712 | sock = &connection->data; |
4713 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 | 4714 | if (!p) |
e8d17b01 | 4715 | return -EIO; |
b411b363 PR |
4716 | memset(p, 0, sizeof(*p)); |
4717 | p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); | |
4718 | p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); | |
20c68fde | 4719 | p->feature_flags = cpu_to_be32(PRO_FEATURES); |
bde89a9e | 4720 | return conn_send_command(connection, sock, P_CONNECTION_FEATURES, sizeof(*p), NULL, 0); |
b411b363 PR |
4721 | } |
4722 | ||
4723 | /* | |
4724 | * return values: | |
4725 | * 1 yes, we have a valid connection | |
4726 | * 0 oops, did not work out, please try again | |
4727 | * -1 peer talks different language, | |
4728 | * no point in trying again, please go standalone. | |
4729 | */ | |
bde89a9e | 4730 | static int drbd_do_features(struct drbd_connection *connection) |
b411b363 | 4731 | { |
bde89a9e | 4732 | /* ASSERT current == connection->receiver ... */ |
e658983a AG |
4733 | struct p_connection_features *p; |
4734 | const int expect = sizeof(struct p_connection_features); | |
77351055 | 4735 | struct packet_info pi; |
a5c31904 | 4736 | int err; |
b411b363 | 4737 | |
bde89a9e | 4738 | err = drbd_send_features(connection); |
e8d17b01 | 4739 | if (err) |
b411b363 PR |
4740 | return 0; |
4741 | ||
bde89a9e | 4742 | err = drbd_recv_header(connection, &pi); |
69bc7bc3 | 4743 | if (err) |
b411b363 PR |
4744 | return 0; |
4745 | ||
6038178e | 4746 | if (pi.cmd != P_CONNECTION_FEATURES) { |
1ec861eb | 4747 | drbd_err(connection, "expected ConnectionFeatures packet, received: %s (0x%04x)\n", |
2fcb8f30 | 4748 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4749 | return -1; |
4750 | } | |
4751 | ||
77351055 | 4752 | if (pi.size != expect) { |
1ec861eb | 4753 | drbd_err(connection, "expected ConnectionFeatures length: %u, received: %u\n", |
77351055 | 4754 | expect, pi.size); |
b411b363 PR |
4755 | return -1; |
4756 | } | |
4757 | ||
e658983a | 4758 | p = pi.data; |
bde89a9e | 4759 | err = drbd_recv_all_warn(connection, p, expect); |
a5c31904 | 4760 | if (err) |
b411b363 | 4761 | return 0; |
b411b363 | 4762 | |
b411b363 PR |
4763 | p->protocol_min = be32_to_cpu(p->protocol_min); |
4764 | p->protocol_max = be32_to_cpu(p->protocol_max); | |
4765 | if (p->protocol_max == 0) | |
4766 | p->protocol_max = p->protocol_min; | |
4767 | ||
4768 | if (PRO_VERSION_MAX < p->protocol_min || | |
4769 | PRO_VERSION_MIN > p->protocol_max) | |
4770 | goto incompat; | |
4771 | ||
bde89a9e | 4772 | connection->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); |
20c68fde | 4773 | connection->agreed_features = PRO_FEATURES & be32_to_cpu(p->feature_flags); |
b411b363 | 4774 | |
1ec861eb | 4775 | drbd_info(connection, "Handshake successful: " |
bde89a9e | 4776 | "Agreed network protocol version %d\n", connection->agreed_pro_version); |
b411b363 | 4777 | |
20c68fde LE |
4778 | drbd_info(connection, "Agreed to%ssupport TRIM on protocol level\n", |
4779 | connection->agreed_features & FF_TRIM ? " " : " not "); | |
4780 | ||
b411b363 PR |
4781 | return 1; |
4782 | ||
4783 | incompat: | |
1ec861eb | 4784 | drbd_err(connection, "incompatible DRBD dialects: " |
b411b363 PR |
4785 | "I support %d-%d, peer supports %d-%d\n", |
4786 | PRO_VERSION_MIN, PRO_VERSION_MAX, | |
4787 | p->protocol_min, p->protocol_max); | |
4788 | return -1; | |
4789 | } | |
4790 | ||
4791 | #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) | |
bde89a9e | 4792 | static int drbd_do_auth(struct drbd_connection *connection) |
b411b363 | 4793 | { |
1ec861eb AG |
4794 | drbd_err(connection, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); |
4795 | drbd_err(connection, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); | |
b10d96cb | 4796 | return -1; |
b411b363 PR |
4797 | } |
4798 | #else | |
4799 | #define CHALLENGE_LEN 64 | |
b10d96cb JT |
4800 | |
4801 | /* Return value: | |
4802 | 1 - auth succeeded, | |
4803 | 0 - failed, try again (network error), | |
4804 | -1 - auth failed, don't try again. | |
4805 | */ | |
4806 | ||
bde89a9e | 4807 | static int drbd_do_auth(struct drbd_connection *connection) |
b411b363 | 4808 | { |
9f5bdc33 | 4809 | struct drbd_socket *sock; |
b411b363 PR |
4810 | char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ |
4811 | struct scatterlist sg; | |
4812 | char *response = NULL; | |
4813 | char *right_response = NULL; | |
4814 | char *peers_ch = NULL; | |
44ed167d PR |
4815 | unsigned int key_len; |
4816 | char secret[SHARED_SECRET_MAX]; /* 64 byte */ | |
b411b363 PR |
4817 | unsigned int resp_size; |
4818 | struct hash_desc desc; | |
77351055 | 4819 | struct packet_info pi; |
44ed167d | 4820 | struct net_conf *nc; |
69bc7bc3 | 4821 | int err, rv; |
b411b363 | 4822 | |
9f5bdc33 | 4823 | /* FIXME: Put the challenge/response into the preallocated socket buffer. */ |
b411b363 | 4824 | |
44ed167d | 4825 | rcu_read_lock(); |
bde89a9e | 4826 | nc = rcu_dereference(connection->net_conf); |
44ed167d PR |
4827 | key_len = strlen(nc->shared_secret); |
4828 | memcpy(secret, nc->shared_secret, key_len); | |
4829 | rcu_read_unlock(); | |
4830 | ||
bde89a9e | 4831 | desc.tfm = connection->cram_hmac_tfm; |
b411b363 PR |
4832 | desc.flags = 0; |
4833 | ||
bde89a9e | 4834 | rv = crypto_hash_setkey(connection->cram_hmac_tfm, (u8 *)secret, key_len); |
b411b363 | 4835 | if (rv) { |
1ec861eb | 4836 | drbd_err(connection, "crypto_hash_setkey() failed with %d\n", rv); |
b10d96cb | 4837 | rv = -1; |
b411b363 PR |
4838 | goto fail; |
4839 | } | |
4840 | ||
4841 | get_random_bytes(my_challenge, CHALLENGE_LEN); | |
4842 | ||
bde89a9e AG |
4843 | sock = &connection->data; |
4844 | if (!conn_prepare_command(connection, sock)) { | |
9f5bdc33 AG |
4845 | rv = 0; |
4846 | goto fail; | |
4847 | } | |
bde89a9e | 4848 | rv = !conn_send_command(connection, sock, P_AUTH_CHALLENGE, 0, |
9f5bdc33 | 4849 | my_challenge, CHALLENGE_LEN); |
b411b363 PR |
4850 | if (!rv) |
4851 | goto fail; | |
4852 | ||
bde89a9e | 4853 | err = drbd_recv_header(connection, &pi); |
69bc7bc3 AG |
4854 | if (err) { |
4855 | rv = 0; | |
b411b363 | 4856 | goto fail; |
69bc7bc3 | 4857 | } |
b411b363 | 4858 | |
77351055 | 4859 | if (pi.cmd != P_AUTH_CHALLENGE) { |
1ec861eb | 4860 | drbd_err(connection, "expected AuthChallenge packet, received: %s (0x%04x)\n", |
2fcb8f30 | 4861 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4862 | rv = 0; |
4863 | goto fail; | |
4864 | } | |
4865 | ||
77351055 | 4866 | if (pi.size > CHALLENGE_LEN * 2) { |
1ec861eb | 4867 | drbd_err(connection, "expected AuthChallenge payload too big.\n"); |
b10d96cb | 4868 | rv = -1; |
b411b363 PR |
4869 | goto fail; |
4870 | } | |
4871 | ||
67cca286 PR |
4872 | if (pi.size < CHALLENGE_LEN) { |
4873 | drbd_err(connection, "AuthChallenge payload too small.\n"); | |
4874 | rv = -1; | |
4875 | goto fail; | |
4876 | } | |
4877 | ||
77351055 | 4878 | peers_ch = kmalloc(pi.size, GFP_NOIO); |
b411b363 | 4879 | if (peers_ch == NULL) { |
1ec861eb | 4880 | drbd_err(connection, "kmalloc of peers_ch failed\n"); |
b10d96cb | 4881 | rv = -1; |
b411b363 PR |
4882 | goto fail; |
4883 | } | |
4884 | ||
bde89a9e | 4885 | err = drbd_recv_all_warn(connection, peers_ch, pi.size); |
a5c31904 | 4886 | if (err) { |
b411b363 PR |
4887 | rv = 0; |
4888 | goto fail; | |
4889 | } | |
4890 | ||
67cca286 PR |
4891 | if (!memcmp(my_challenge, peers_ch, CHALLENGE_LEN)) { |
4892 | drbd_err(connection, "Peer presented the same challenge!\n"); | |
4893 | rv = -1; | |
4894 | goto fail; | |
4895 | } | |
4896 | ||
bde89a9e | 4897 | resp_size = crypto_hash_digestsize(connection->cram_hmac_tfm); |
b411b363 PR |
4898 | response = kmalloc(resp_size, GFP_NOIO); |
4899 | if (response == NULL) { | |
1ec861eb | 4900 | drbd_err(connection, "kmalloc of response failed\n"); |
b10d96cb | 4901 | rv = -1; |
b411b363 PR |
4902 | goto fail; |
4903 | } | |
4904 | ||
4905 | sg_init_table(&sg, 1); | |
77351055 | 4906 | sg_set_buf(&sg, peers_ch, pi.size); |
b411b363 PR |
4907 | |
4908 | rv = crypto_hash_digest(&desc, &sg, sg.length, response); | |
4909 | if (rv) { | |
1ec861eb | 4910 | drbd_err(connection, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 4911 | rv = -1; |
b411b363 PR |
4912 | goto fail; |
4913 | } | |
4914 | ||
bde89a9e | 4915 | if (!conn_prepare_command(connection, sock)) { |
9f5bdc33 | 4916 | rv = 0; |
b411b363 | 4917 | goto fail; |
9f5bdc33 | 4918 | } |
bde89a9e | 4919 | rv = !conn_send_command(connection, sock, P_AUTH_RESPONSE, 0, |
9f5bdc33 | 4920 | response, resp_size); |
b411b363 PR |
4921 | if (!rv) |
4922 | goto fail; | |
4923 | ||
bde89a9e | 4924 | err = drbd_recv_header(connection, &pi); |
69bc7bc3 | 4925 | if (err) { |
b411b363 PR |
4926 | rv = 0; |
4927 | goto fail; | |
4928 | } | |
4929 | ||
77351055 | 4930 | if (pi.cmd != P_AUTH_RESPONSE) { |
1ec861eb | 4931 | drbd_err(connection, "expected AuthResponse packet, received: %s (0x%04x)\n", |
2fcb8f30 | 4932 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4933 | rv = 0; |
4934 | goto fail; | |
4935 | } | |
4936 | ||
77351055 | 4937 | if (pi.size != resp_size) { |
1ec861eb | 4938 | drbd_err(connection, "expected AuthResponse payload of wrong size\n"); |
b411b363 PR |
4939 | rv = 0; |
4940 | goto fail; | |
4941 | } | |
b411b363 | 4942 | |
bde89a9e | 4943 | err = drbd_recv_all_warn(connection, response , resp_size); |
a5c31904 | 4944 | if (err) { |
b411b363 PR |
4945 | rv = 0; |
4946 | goto fail; | |
4947 | } | |
4948 | ||
4949 | right_response = kmalloc(resp_size, GFP_NOIO); | |
2d1ee87d | 4950 | if (right_response == NULL) { |
1ec861eb | 4951 | drbd_err(connection, "kmalloc of right_response failed\n"); |
b10d96cb | 4952 | rv = -1; |
b411b363 PR |
4953 | goto fail; |
4954 | } | |
4955 | ||
4956 | sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); | |
4957 | ||
4958 | rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); | |
4959 | if (rv) { | |
1ec861eb | 4960 | drbd_err(connection, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 4961 | rv = -1; |
b411b363 PR |
4962 | goto fail; |
4963 | } | |
4964 | ||
4965 | rv = !memcmp(response, right_response, resp_size); | |
4966 | ||
4967 | if (rv) | |
1ec861eb | 4968 | drbd_info(connection, "Peer authenticated using %d bytes HMAC\n", |
44ed167d | 4969 | resp_size); |
b10d96cb JT |
4970 | else |
4971 | rv = -1; | |
b411b363 PR |
4972 | |
4973 | fail: | |
4974 | kfree(peers_ch); | |
4975 | kfree(response); | |
4976 | kfree(right_response); | |
4977 | ||
4978 | return rv; | |
4979 | } | |
4980 | #endif | |
4981 | ||
8fe60551 | 4982 | int drbd_receiver(struct drbd_thread *thi) |
b411b363 | 4983 | { |
bde89a9e | 4984 | struct drbd_connection *connection = thi->connection; |
b411b363 PR |
4985 | int h; |
4986 | ||
1ec861eb | 4987 | drbd_info(connection, "receiver (re)started\n"); |
b411b363 PR |
4988 | |
4989 | do { | |
bde89a9e | 4990 | h = conn_connect(connection); |
b411b363 | 4991 | if (h == 0) { |
bde89a9e | 4992 | conn_disconnect(connection); |
20ee6390 | 4993 | schedule_timeout_interruptible(HZ); |
b411b363 PR |
4994 | } |
4995 | if (h == -1) { | |
1ec861eb | 4996 | drbd_warn(connection, "Discarding network configuration.\n"); |
bde89a9e | 4997 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
4998 | } |
4999 | } while (h == 0); | |
5000 | ||
91fd4dad | 5001 | if (h > 0) |
bde89a9e | 5002 | drbdd(connection); |
b411b363 | 5003 | |
bde89a9e | 5004 | conn_disconnect(connection); |
b411b363 | 5005 | |
1ec861eb | 5006 | drbd_info(connection, "receiver terminated\n"); |
b411b363 PR |
5007 | return 0; |
5008 | } | |
5009 | ||
5010 | /* ********* acknowledge sender ******** */ | |
5011 | ||
bde89a9e | 5012 | static int got_conn_RqSReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5013 | { |
e658983a | 5014 | struct p_req_state_reply *p = pi->data; |
e4f78ede PR |
5015 | int retcode = be32_to_cpu(p->retcode); |
5016 | ||
5017 | if (retcode >= SS_SUCCESS) { | |
bde89a9e | 5018 | set_bit(CONN_WD_ST_CHG_OKAY, &connection->flags); |
e4f78ede | 5019 | } else { |
bde89a9e | 5020 | set_bit(CONN_WD_ST_CHG_FAIL, &connection->flags); |
1ec861eb | 5021 | drbd_err(connection, "Requested state change failed by peer: %s (%d)\n", |
e4f78ede PR |
5022 | drbd_set_st_err_str(retcode), retcode); |
5023 | } | |
bde89a9e | 5024 | wake_up(&connection->ping_wait); |
e4f78ede | 5025 | |
2735a594 | 5026 | return 0; |
e4f78ede | 5027 | } |
b411b363 | 5028 | |
bde89a9e | 5029 | static int got_RqSReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5030 | { |
9f4fe9ad | 5031 | struct drbd_peer_device *peer_device; |
b30ab791 | 5032 | struct drbd_device *device; |
e658983a | 5033 | struct p_req_state_reply *p = pi->data; |
b411b363 PR |
5034 | int retcode = be32_to_cpu(p->retcode); |
5035 | ||
9f4fe9ad AG |
5036 | peer_device = conn_peer_device(connection, pi->vnr); |
5037 | if (!peer_device) | |
2735a594 | 5038 | return -EIO; |
9f4fe9ad | 5039 | device = peer_device->device; |
1952e916 | 5040 | |
bde89a9e | 5041 | if (test_bit(CONN_WD_ST_CHG_REQ, &connection->flags)) { |
0b0ba1ef | 5042 | D_ASSERT(device, connection->agreed_pro_version < 100); |
bde89a9e | 5043 | return got_conn_RqSReply(connection, pi); |
4d0fc3fd PR |
5044 | } |
5045 | ||
b411b363 | 5046 | if (retcode >= SS_SUCCESS) { |
b30ab791 | 5047 | set_bit(CL_ST_CHG_SUCCESS, &device->flags); |
b411b363 | 5048 | } else { |
b30ab791 | 5049 | set_bit(CL_ST_CHG_FAIL, &device->flags); |
d0180171 | 5050 | drbd_err(device, "Requested state change failed by peer: %s (%d)\n", |
e4f78ede | 5051 | drbd_set_st_err_str(retcode), retcode); |
b411b363 | 5052 | } |
b30ab791 | 5053 | wake_up(&device->state_wait); |
b411b363 | 5054 | |
2735a594 | 5055 | return 0; |
b411b363 PR |
5056 | } |
5057 | ||
bde89a9e | 5058 | static int got_Ping(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5059 | { |
bde89a9e | 5060 | return drbd_send_ping_ack(connection); |
b411b363 PR |
5061 | |
5062 | } | |
5063 | ||
bde89a9e | 5064 | static int got_PingAck(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 PR |
5065 | { |
5066 | /* restore idle timeout */ | |
bde89a9e AG |
5067 | connection->meta.socket->sk->sk_rcvtimeo = connection->net_conf->ping_int*HZ; |
5068 | if (!test_and_set_bit(GOT_PING_ACK, &connection->flags)) | |
5069 | wake_up(&connection->ping_wait); | |
b411b363 | 5070 | |
2735a594 | 5071 | return 0; |
b411b363 PR |
5072 | } |
5073 | ||
bde89a9e | 5074 | static int got_IsInSync(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5075 | { |
9f4fe9ad | 5076 | struct drbd_peer_device *peer_device; |
b30ab791 | 5077 | struct drbd_device *device; |
e658983a | 5078 | struct p_block_ack *p = pi->data; |
b411b363 PR |
5079 | sector_t sector = be64_to_cpu(p->sector); |
5080 | int blksize = be32_to_cpu(p->blksize); | |
5081 | ||
9f4fe9ad AG |
5082 | peer_device = conn_peer_device(connection, pi->vnr); |
5083 | if (!peer_device) | |
2735a594 | 5084 | return -EIO; |
9f4fe9ad | 5085 | device = peer_device->device; |
1952e916 | 5086 | |
9f4fe9ad | 5087 | D_ASSERT(device, peer_device->connection->agreed_pro_version >= 89); |
b411b363 | 5088 | |
69a22773 | 5089 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 | 5090 | |
b30ab791 AG |
5091 | if (get_ldev(device)) { |
5092 | drbd_rs_complete_io(device, sector); | |
5093 | drbd_set_in_sync(device, sector, blksize); | |
1d53f09e | 5094 | /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ |
b30ab791 AG |
5095 | device->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); |
5096 | put_ldev(device); | |
1d53f09e | 5097 | } |
b30ab791 AG |
5098 | dec_rs_pending(device); |
5099 | atomic_add(blksize >> 9, &device->rs_sect_in); | |
b411b363 | 5100 | |
2735a594 | 5101 | return 0; |
b411b363 PR |
5102 | } |
5103 | ||
bc9c5c41 | 5104 | static int |
b30ab791 | 5105 | validate_req_change_req_state(struct drbd_device *device, u64 id, sector_t sector, |
bc9c5c41 AG |
5106 | struct rb_root *root, const char *func, |
5107 | enum drbd_req_event what, bool missing_ok) | |
b411b363 PR |
5108 | { |
5109 | struct drbd_request *req; | |
5110 | struct bio_and_error m; | |
5111 | ||
0500813f | 5112 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 5113 | req = find_request(device, root, id, sector, missing_ok, func); |
b411b363 | 5114 | if (unlikely(!req)) { |
0500813f | 5115 | spin_unlock_irq(&device->resource->req_lock); |
85997675 | 5116 | return -EIO; |
b411b363 PR |
5117 | } |
5118 | __req_mod(req, what, &m); | |
0500813f | 5119 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
5120 | |
5121 | if (m.bio) | |
b30ab791 | 5122 | complete_master_bio(device, &m); |
85997675 | 5123 | return 0; |
b411b363 PR |
5124 | } |
5125 | ||
bde89a9e | 5126 | static int got_BlockAck(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5127 | { |
9f4fe9ad | 5128 | struct drbd_peer_device *peer_device; |
b30ab791 | 5129 | struct drbd_device *device; |
e658983a | 5130 | struct p_block_ack *p = pi->data; |
b411b363 PR |
5131 | sector_t sector = be64_to_cpu(p->sector); |
5132 | int blksize = be32_to_cpu(p->blksize); | |
5133 | enum drbd_req_event what; | |
5134 | ||
9f4fe9ad AG |
5135 | peer_device = conn_peer_device(connection, pi->vnr); |
5136 | if (!peer_device) | |
2735a594 | 5137 | return -EIO; |
9f4fe9ad | 5138 | device = peer_device->device; |
1952e916 | 5139 | |
69a22773 | 5140 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 | 5141 | |
579b57ed | 5142 | if (p->block_id == ID_SYNCER) { |
b30ab791 AG |
5143 | drbd_set_in_sync(device, sector, blksize); |
5144 | dec_rs_pending(device); | |
2735a594 | 5145 | return 0; |
b411b363 | 5146 | } |
e05e1e59 | 5147 | switch (pi->cmd) { |
b411b363 | 5148 | case P_RS_WRITE_ACK: |
8554df1c | 5149 | what = WRITE_ACKED_BY_PEER_AND_SIS; |
b411b363 PR |
5150 | break; |
5151 | case P_WRITE_ACK: | |
8554df1c | 5152 | what = WRITE_ACKED_BY_PEER; |
b411b363 PR |
5153 | break; |
5154 | case P_RECV_ACK: | |
8554df1c | 5155 | what = RECV_ACKED_BY_PEER; |
b411b363 | 5156 | break; |
d4dabbe2 LE |
5157 | case P_SUPERSEDED: |
5158 | what = CONFLICT_RESOLVED; | |
b411b363 | 5159 | break; |
7be8da07 | 5160 | case P_RETRY_WRITE: |
7be8da07 | 5161 | what = POSTPONE_WRITE; |
b411b363 PR |
5162 | break; |
5163 | default: | |
2735a594 | 5164 | BUG(); |
b411b363 PR |
5165 | } |
5166 | ||
b30ab791 AG |
5167 | return validate_req_change_req_state(device, p->block_id, sector, |
5168 | &device->write_requests, __func__, | |
2735a594 | 5169 | what, false); |
b411b363 PR |
5170 | } |
5171 | ||
bde89a9e | 5172 | static int got_NegAck(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5173 | { |
9f4fe9ad | 5174 | struct drbd_peer_device *peer_device; |
b30ab791 | 5175 | struct drbd_device *device; |
e658983a | 5176 | struct p_block_ack *p = pi->data; |
b411b363 | 5177 | sector_t sector = be64_to_cpu(p->sector); |
2deb8336 | 5178 | int size = be32_to_cpu(p->blksize); |
85997675 | 5179 | int err; |
b411b363 | 5180 | |
9f4fe9ad AG |
5181 | peer_device = conn_peer_device(connection, pi->vnr); |
5182 | if (!peer_device) | |
2735a594 | 5183 | return -EIO; |
9f4fe9ad | 5184 | device = peer_device->device; |
b411b363 | 5185 | |
69a22773 | 5186 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 | 5187 | |
579b57ed | 5188 | if (p->block_id == ID_SYNCER) { |
b30ab791 AG |
5189 | dec_rs_pending(device); |
5190 | drbd_rs_failed_io(device, sector, size); | |
2735a594 | 5191 | return 0; |
b411b363 | 5192 | } |
2deb8336 | 5193 | |
b30ab791 AG |
5194 | err = validate_req_change_req_state(device, p->block_id, sector, |
5195 | &device->write_requests, __func__, | |
303d1448 | 5196 | NEG_ACKED, true); |
85997675 | 5197 | if (err) { |
c3afd8f5 AG |
5198 | /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs. |
5199 | The master bio might already be completed, therefore the | |
5200 | request is no longer in the collision hash. */ | |
5201 | /* In Protocol B we might already have got a P_RECV_ACK | |
5202 | but then get a P_NEG_ACK afterwards. */ | |
b30ab791 | 5203 | drbd_set_out_of_sync(device, sector, size); |
2deb8336 | 5204 | } |
2735a594 | 5205 | return 0; |
b411b363 PR |
5206 | } |
5207 | ||
bde89a9e | 5208 | static int got_NegDReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5209 | { |
9f4fe9ad | 5210 | struct drbd_peer_device *peer_device; |
b30ab791 | 5211 | struct drbd_device *device; |
e658983a | 5212 | struct p_block_ack *p = pi->data; |
b411b363 PR |
5213 | sector_t sector = be64_to_cpu(p->sector); |
5214 | ||
9f4fe9ad AG |
5215 | peer_device = conn_peer_device(connection, pi->vnr); |
5216 | if (!peer_device) | |
2735a594 | 5217 | return -EIO; |
9f4fe9ad | 5218 | device = peer_device->device; |
1952e916 | 5219 | |
69a22773 | 5220 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
7be8da07 | 5221 | |
d0180171 | 5222 | drbd_err(device, "Got NegDReply; Sector %llus, len %u.\n", |
b411b363 PR |
5223 | (unsigned long long)sector, be32_to_cpu(p->blksize)); |
5224 | ||
b30ab791 AG |
5225 | return validate_req_change_req_state(device, p->block_id, sector, |
5226 | &device->read_requests, __func__, | |
2735a594 | 5227 | NEG_ACKED, false); |
b411b363 PR |
5228 | } |
5229 | ||
bde89a9e | 5230 | static int got_NegRSDReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5231 | { |
9f4fe9ad | 5232 | struct drbd_peer_device *peer_device; |
b30ab791 | 5233 | struct drbd_device *device; |
b411b363 PR |
5234 | sector_t sector; |
5235 | int size; | |
e658983a | 5236 | struct p_block_ack *p = pi->data; |
1952e916 | 5237 | |
9f4fe9ad AG |
5238 | peer_device = conn_peer_device(connection, pi->vnr); |
5239 | if (!peer_device) | |
2735a594 | 5240 | return -EIO; |
9f4fe9ad | 5241 | device = peer_device->device; |
b411b363 PR |
5242 | |
5243 | sector = be64_to_cpu(p->sector); | |
5244 | size = be32_to_cpu(p->blksize); | |
b411b363 | 5245 | |
69a22773 | 5246 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 | 5247 | |
b30ab791 | 5248 | dec_rs_pending(device); |
b411b363 | 5249 | |
b30ab791 AG |
5250 | if (get_ldev_if_state(device, D_FAILED)) { |
5251 | drbd_rs_complete_io(device, sector); | |
e05e1e59 | 5252 | switch (pi->cmd) { |
d612d309 | 5253 | case P_NEG_RS_DREPLY: |
b30ab791 | 5254 | drbd_rs_failed_io(device, sector, size); |
d612d309 PR |
5255 | case P_RS_CANCEL: |
5256 | break; | |
5257 | default: | |
2735a594 | 5258 | BUG(); |
d612d309 | 5259 | } |
b30ab791 | 5260 | put_ldev(device); |
b411b363 PR |
5261 | } |
5262 | ||
2735a594 | 5263 | return 0; |
b411b363 PR |
5264 | } |
5265 | ||
bde89a9e | 5266 | static int got_BarrierAck(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5267 | { |
e658983a | 5268 | struct p_barrier_ack *p = pi->data; |
c06ece6b | 5269 | struct drbd_peer_device *peer_device; |
9ed57dcb | 5270 | int vnr; |
1952e916 | 5271 | |
bde89a9e | 5272 | tl_release(connection, p->barrier, be32_to_cpu(p->set_size)); |
b411b363 | 5273 | |
9ed57dcb | 5274 | rcu_read_lock(); |
c06ece6b AG |
5275 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
5276 | struct drbd_device *device = peer_device->device; | |
5277 | ||
b30ab791 AG |
5278 | if (device->state.conn == C_AHEAD && |
5279 | atomic_read(&device->ap_in_flight) == 0 && | |
5280 | !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &device->flags)) { | |
5281 | device->start_resync_timer.expires = jiffies + HZ; | |
5282 | add_timer(&device->start_resync_timer); | |
9ed57dcb | 5283 | } |
c4752ef1 | 5284 | } |
9ed57dcb | 5285 | rcu_read_unlock(); |
c4752ef1 | 5286 | |
2735a594 | 5287 | return 0; |
b411b363 PR |
5288 | } |
5289 | ||
bde89a9e | 5290 | static int got_OVResult(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5291 | { |
9f4fe9ad | 5292 | struct drbd_peer_device *peer_device; |
b30ab791 | 5293 | struct drbd_device *device; |
e658983a | 5294 | struct p_block_ack *p = pi->data; |
84b8c06b | 5295 | struct drbd_device_work *dw; |
b411b363 PR |
5296 | sector_t sector; |
5297 | int size; | |
5298 | ||
9f4fe9ad AG |
5299 | peer_device = conn_peer_device(connection, pi->vnr); |
5300 | if (!peer_device) | |
2735a594 | 5301 | return -EIO; |
9f4fe9ad | 5302 | device = peer_device->device; |
1952e916 | 5303 | |
b411b363 PR |
5304 | sector = be64_to_cpu(p->sector); |
5305 | size = be32_to_cpu(p->blksize); | |
5306 | ||
69a22773 | 5307 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 PR |
5308 | |
5309 | if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) | |
b30ab791 | 5310 | drbd_ov_out_of_sync_found(device, sector, size); |
b411b363 | 5311 | else |
b30ab791 | 5312 | ov_out_of_sync_print(device); |
b411b363 | 5313 | |
b30ab791 | 5314 | if (!get_ldev(device)) |
2735a594 | 5315 | return 0; |
1d53f09e | 5316 | |
b30ab791 AG |
5317 | drbd_rs_complete_io(device, sector); |
5318 | dec_rs_pending(device); | |
b411b363 | 5319 | |
b30ab791 | 5320 | --device->ov_left; |
ea5442af LE |
5321 | |
5322 | /* let's advance progress step marks only for every other megabyte */ | |
b30ab791 AG |
5323 | if ((device->ov_left & 0x200) == 0x200) |
5324 | drbd_advance_rs_marks(device, device->ov_left); | |
ea5442af | 5325 | |
b30ab791 | 5326 | if (device->ov_left == 0) { |
84b8c06b AG |
5327 | dw = kmalloc(sizeof(*dw), GFP_NOIO); |
5328 | if (dw) { | |
5329 | dw->w.cb = w_ov_finished; | |
5330 | dw->device = device; | |
5331 | drbd_queue_work(&peer_device->connection->sender_work, &dw->w); | |
b411b363 | 5332 | } else { |
84b8c06b | 5333 | drbd_err(device, "kmalloc(dw) failed."); |
b30ab791 AG |
5334 | ov_out_of_sync_print(device); |
5335 | drbd_resync_finished(device); | |
b411b363 PR |
5336 | } |
5337 | } | |
b30ab791 | 5338 | put_ldev(device); |
2735a594 | 5339 | return 0; |
b411b363 PR |
5340 | } |
5341 | ||
bde89a9e | 5342 | static int got_skip(struct drbd_connection *connection, struct packet_info *pi) |
0ced55a3 | 5343 | { |
2735a594 | 5344 | return 0; |
b411b363 PR |
5345 | } |
5346 | ||
bde89a9e | 5347 | static int connection_finish_peer_reqs(struct drbd_connection *connection) |
0ced55a3 | 5348 | { |
c06ece6b | 5349 | struct drbd_peer_device *peer_device; |
c141ebda | 5350 | int vnr, not_empty = 0; |
32862ec7 PR |
5351 | |
5352 | do { | |
bde89a9e | 5353 | clear_bit(SIGNAL_ASENDER, &connection->flags); |
32862ec7 | 5354 | flush_signals(current); |
c141ebda PR |
5355 | |
5356 | rcu_read_lock(); | |
c06ece6b AG |
5357 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
5358 | struct drbd_device *device = peer_device->device; | |
b30ab791 | 5359 | kref_get(&device->kref); |
c141ebda | 5360 | rcu_read_unlock(); |
b30ab791 | 5361 | if (drbd_finish_peer_reqs(device)) { |
05a10ec7 | 5362 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda | 5363 | return 1; |
d3fcb490 | 5364 | } |
05a10ec7 | 5365 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda | 5366 | rcu_read_lock(); |
082a3439 | 5367 | } |
bde89a9e | 5368 | set_bit(SIGNAL_ASENDER, &connection->flags); |
082a3439 | 5369 | |
0500813f | 5370 | spin_lock_irq(&connection->resource->req_lock); |
c06ece6b AG |
5371 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
5372 | struct drbd_device *device = peer_device->device; | |
b30ab791 | 5373 | not_empty = !list_empty(&device->done_ee); |
082a3439 PR |
5374 | if (not_empty) |
5375 | break; | |
5376 | } | |
0500813f | 5377 | spin_unlock_irq(&connection->resource->req_lock); |
c141ebda | 5378 | rcu_read_unlock(); |
32862ec7 PR |
5379 | } while (not_empty); |
5380 | ||
5381 | return 0; | |
0ced55a3 PR |
5382 | } |
5383 | ||
b411b363 PR |
5384 | struct asender_cmd { |
5385 | size_t pkt_size; | |
bde89a9e | 5386 | int (*fn)(struct drbd_connection *connection, struct packet_info *); |
b411b363 PR |
5387 | }; |
5388 | ||
7201b972 | 5389 | static struct asender_cmd asender_tbl[] = { |
e658983a AG |
5390 | [P_PING] = { 0, got_Ping }, |
5391 | [P_PING_ACK] = { 0, got_PingAck }, | |
b411b363 PR |
5392 | [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, |
5393 | [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
5394 | [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
d4dabbe2 | 5395 | [P_SUPERSEDED] = { sizeof(struct p_block_ack), got_BlockAck }, |
b411b363 PR |
5396 | [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, |
5397 | [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, | |
1952e916 | 5398 | [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply }, |
b411b363 PR |
5399 | [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, |
5400 | [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, | |
5401 | [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, | |
5402 | [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, | |
02918be2 | 5403 | [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, |
1952e916 AG |
5404 | [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply }, |
5405 | [P_CONN_ST_CHG_REPLY]={ sizeof(struct p_req_state_reply), got_conn_RqSReply }, | |
5406 | [P_RETRY_WRITE] = { sizeof(struct p_block_ack), got_BlockAck }, | |
7201b972 | 5407 | }; |
b411b363 PR |
5408 | |
5409 | int drbd_asender(struct drbd_thread *thi) | |
5410 | { | |
bde89a9e | 5411 | struct drbd_connection *connection = thi->connection; |
b411b363 | 5412 | struct asender_cmd *cmd = NULL; |
77351055 | 5413 | struct packet_info pi; |
257d0af6 | 5414 | int rv; |
bde89a9e | 5415 | void *buf = connection->meta.rbuf; |
b411b363 | 5416 | int received = 0; |
bde89a9e | 5417 | unsigned int header_size = drbd_header_size(connection); |
52b061a4 | 5418 | int expect = header_size; |
44ed167d PR |
5419 | bool ping_timeout_active = false; |
5420 | struct net_conf *nc; | |
bb77d34e | 5421 | int ping_timeo, tcp_cork, ping_int; |
3990e04d | 5422 | struct sched_param param = { .sched_priority = 2 }; |
b411b363 | 5423 | |
3990e04d PR |
5424 | rv = sched_setscheduler(current, SCHED_RR, ¶m); |
5425 | if (rv < 0) | |
1ec861eb | 5426 | drbd_err(connection, "drbd_asender: ERROR set priority, ret=%d\n", rv); |
b411b363 | 5427 | |
e77a0a5c | 5428 | while (get_t_state(thi) == RUNNING) { |
80822284 | 5429 | drbd_thread_current_set_cpu(thi); |
b411b363 | 5430 | |
44ed167d | 5431 | rcu_read_lock(); |
bde89a9e | 5432 | nc = rcu_dereference(connection->net_conf); |
44ed167d | 5433 | ping_timeo = nc->ping_timeo; |
bb77d34e | 5434 | tcp_cork = nc->tcp_cork; |
44ed167d PR |
5435 | ping_int = nc->ping_int; |
5436 | rcu_read_unlock(); | |
5437 | ||
bde89a9e AG |
5438 | if (test_and_clear_bit(SEND_PING, &connection->flags)) { |
5439 | if (drbd_send_ping(connection)) { | |
1ec861eb | 5440 | drbd_err(connection, "drbd_send_ping has failed\n"); |
b411b363 | 5441 | goto reconnect; |
841ce241 | 5442 | } |
bde89a9e | 5443 | connection->meta.socket->sk->sk_rcvtimeo = ping_timeo * HZ / 10; |
44ed167d | 5444 | ping_timeout_active = true; |
b411b363 PR |
5445 | } |
5446 | ||
32862ec7 PR |
5447 | /* TODO: conditionally cork; it may hurt latency if we cork without |
5448 | much to send */ | |
bb77d34e | 5449 | if (tcp_cork) |
bde89a9e AG |
5450 | drbd_tcp_cork(connection->meta.socket); |
5451 | if (connection_finish_peer_reqs(connection)) { | |
1ec861eb | 5452 | drbd_err(connection, "connection_finish_peer_reqs() failed\n"); |
32862ec7 | 5453 | goto reconnect; |
b411b363 PR |
5454 | } |
5455 | /* but unconditionally uncork unless disabled */ | |
bb77d34e | 5456 | if (tcp_cork) |
bde89a9e | 5457 | drbd_tcp_uncork(connection->meta.socket); |
b411b363 PR |
5458 | |
5459 | /* short circuit, recv_msg would return EINTR anyways. */ | |
5460 | if (signal_pending(current)) | |
5461 | continue; | |
5462 | ||
bde89a9e AG |
5463 | rv = drbd_recv_short(connection->meta.socket, buf, expect-received, 0); |
5464 | clear_bit(SIGNAL_ASENDER, &connection->flags); | |
b411b363 PR |
5465 | |
5466 | flush_signals(current); | |
5467 | ||
5468 | /* Note: | |
5469 | * -EINTR (on meta) we got a signal | |
5470 | * -EAGAIN (on meta) rcvtimeo expired | |
5471 | * -ECONNRESET other side closed the connection | |
5472 | * -ERESTARTSYS (on data) we got a signal | |
5473 | * rv < 0 other than above: unexpected error! | |
5474 | * rv == expected: full header or command | |
5475 | * rv < expected: "woken" by signal during receive | |
5476 | * rv == 0 : "connection shut down by peer" | |
5477 | */ | |
5478 | if (likely(rv > 0)) { | |
5479 | received += rv; | |
5480 | buf += rv; | |
5481 | } else if (rv == 0) { | |
bde89a9e | 5482 | if (test_bit(DISCONNECT_SENT, &connection->flags)) { |
b66623e3 PR |
5483 | long t; |
5484 | rcu_read_lock(); | |
bde89a9e | 5485 | t = rcu_dereference(connection->net_conf)->ping_timeo * HZ/10; |
b66623e3 PR |
5486 | rcu_read_unlock(); |
5487 | ||
bde89a9e AG |
5488 | t = wait_event_timeout(connection->ping_wait, |
5489 | connection->cstate < C_WF_REPORT_PARAMS, | |
b66623e3 | 5490 | t); |
599377ac PR |
5491 | if (t) |
5492 | break; | |
5493 | } | |
1ec861eb | 5494 | drbd_err(connection, "meta connection shut down by peer.\n"); |
b411b363 PR |
5495 | goto reconnect; |
5496 | } else if (rv == -EAGAIN) { | |
cb6518cb LE |
5497 | /* If the data socket received something meanwhile, |
5498 | * that is good enough: peer is still alive. */ | |
bde89a9e AG |
5499 | if (time_after(connection->last_received, |
5500 | jiffies - connection->meta.socket->sk->sk_rcvtimeo)) | |
cb6518cb | 5501 | continue; |
f36af18c | 5502 | if (ping_timeout_active) { |
1ec861eb | 5503 | drbd_err(connection, "PingAck did not arrive in time.\n"); |
b411b363 PR |
5504 | goto reconnect; |
5505 | } | |
bde89a9e | 5506 | set_bit(SEND_PING, &connection->flags); |
b411b363 PR |
5507 | continue; |
5508 | } else if (rv == -EINTR) { | |
5509 | continue; | |
5510 | } else { | |
1ec861eb | 5511 | drbd_err(connection, "sock_recvmsg returned %d\n", rv); |
b411b363 PR |
5512 | goto reconnect; |
5513 | } | |
5514 | ||
5515 | if (received == expect && cmd == NULL) { | |
bde89a9e | 5516 | if (decode_header(connection, connection->meta.rbuf, &pi)) |
b411b363 | 5517 | goto reconnect; |
7201b972 | 5518 | cmd = &asender_tbl[pi.cmd]; |
1952e916 | 5519 | if (pi.cmd >= ARRAY_SIZE(asender_tbl) || !cmd->fn) { |
1ec861eb | 5520 | drbd_err(connection, "Unexpected meta packet %s (0x%04x)\n", |
2fcb8f30 | 5521 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
5522 | goto disconnect; |
5523 | } | |
e658983a | 5524 | expect = header_size + cmd->pkt_size; |
52b061a4 | 5525 | if (pi.size != expect - header_size) { |
1ec861eb | 5526 | drbd_err(connection, "Wrong packet size on meta (c: %d, l: %d)\n", |
77351055 | 5527 | pi.cmd, pi.size); |
b411b363 | 5528 | goto reconnect; |
257d0af6 | 5529 | } |
b411b363 PR |
5530 | } |
5531 | if (received == expect) { | |
2735a594 | 5532 | bool err; |
a4fbda8e | 5533 | |
bde89a9e | 5534 | err = cmd->fn(connection, &pi); |
2735a594 | 5535 | if (err) { |
1ec861eb | 5536 | drbd_err(connection, "%pf failed\n", cmd->fn); |
b411b363 | 5537 | goto reconnect; |
1952e916 | 5538 | } |
b411b363 | 5539 | |
bde89a9e | 5540 | connection->last_received = jiffies; |
f36af18c | 5541 | |
44ed167d PR |
5542 | if (cmd == &asender_tbl[P_PING_ACK]) { |
5543 | /* restore idle timeout */ | |
bde89a9e | 5544 | connection->meta.socket->sk->sk_rcvtimeo = ping_int * HZ; |
44ed167d PR |
5545 | ping_timeout_active = false; |
5546 | } | |
f36af18c | 5547 | |
bde89a9e | 5548 | buf = connection->meta.rbuf; |
b411b363 | 5549 | received = 0; |
52b061a4 | 5550 | expect = header_size; |
b411b363 PR |
5551 | cmd = NULL; |
5552 | } | |
5553 | } | |
5554 | ||
5555 | if (0) { | |
5556 | reconnect: | |
bde89a9e AG |
5557 | conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); |
5558 | conn_md_sync(connection); | |
b411b363 PR |
5559 | } |
5560 | if (0) { | |
5561 | disconnect: | |
bde89a9e | 5562 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 5563 | } |
bde89a9e | 5564 | clear_bit(SIGNAL_ASENDER, &connection->flags); |
b411b363 | 5565 | |
1ec861eb | 5566 | drbd_info(connection, "asender terminated\n"); |
b411b363 PR |
5567 | |
5568 | return 0; | |
5569 | } |