]>
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 | |
c5a2c150 LE |
365 | memset(peer_req, 0, sizeof(*peer_req)); |
366 | INIT_LIST_HEAD(&peer_req->w.list); | |
db830c46 AG |
367 | drbd_clear_interval(&peer_req->i); |
368 | peer_req->i.size = data_size; | |
369 | peer_req->i.sector = sector; | |
c5a2c150 | 370 | peer_req->submit_jif = jiffies; |
a8cd15ba | 371 | peer_req->peer_device = peer_device; |
db830c46 | 372 | peer_req->pages = page; |
9a8e7753 AG |
373 | /* |
374 | * The block_id is opaque to the receiver. It is not endianness | |
375 | * converted, and sent back to the sender unchanged. | |
376 | */ | |
db830c46 | 377 | peer_req->block_id = id; |
b411b363 | 378 | |
db830c46 | 379 | return peer_req; |
b411b363 | 380 | |
45bb912b | 381 | fail: |
db830c46 | 382 | mempool_free(peer_req, drbd_ee_mempool); |
b411b363 PR |
383 | return NULL; |
384 | } | |
385 | ||
b30ab791 | 386 | void __drbd_free_peer_req(struct drbd_device *device, struct drbd_peer_request *peer_req, |
f6ffca9f | 387 | int is_net) |
b411b363 | 388 | { |
21ae5d7f | 389 | might_sleep(); |
db830c46 AG |
390 | if (peer_req->flags & EE_HAS_DIGEST) |
391 | kfree(peer_req->digest); | |
b30ab791 | 392 | drbd_free_pages(device, peer_req->pages, is_net); |
0b0ba1ef AG |
393 | D_ASSERT(device, atomic_read(&peer_req->pending_bios) == 0); |
394 | D_ASSERT(device, drbd_interval_empty(&peer_req->i)); | |
21ae5d7f LE |
395 | if (!expect(!(peer_req->flags & EE_CALL_AL_COMPLETE_IO))) { |
396 | peer_req->flags &= ~EE_CALL_AL_COMPLETE_IO; | |
397 | drbd_al_complete_io(device, &peer_req->i); | |
398 | } | |
db830c46 | 399 | mempool_free(peer_req, drbd_ee_mempool); |
b411b363 PR |
400 | } |
401 | ||
b30ab791 | 402 | int drbd_free_peer_reqs(struct drbd_device *device, struct list_head *list) |
b411b363 PR |
403 | { |
404 | LIST_HEAD(work_list); | |
db830c46 | 405 | struct drbd_peer_request *peer_req, *t; |
b411b363 | 406 | int count = 0; |
b30ab791 | 407 | int is_net = list == &device->net_ee; |
b411b363 | 408 | |
0500813f | 409 | spin_lock_irq(&device->resource->req_lock); |
b411b363 | 410 | list_splice_init(list, &work_list); |
0500813f | 411 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 412 | |
a8cd15ba | 413 | list_for_each_entry_safe(peer_req, t, &work_list, w.list) { |
b30ab791 | 414 | __drbd_free_peer_req(device, peer_req, is_net); |
b411b363 PR |
415 | count++; |
416 | } | |
417 | return count; | |
418 | } | |
419 | ||
b411b363 | 420 | /* |
a990be46 | 421 | * See also comments in _req_mod(,BARRIER_ACKED) and receive_Barrier. |
b411b363 | 422 | */ |
b30ab791 | 423 | static int drbd_finish_peer_reqs(struct drbd_device *device) |
b411b363 PR |
424 | { |
425 | LIST_HEAD(work_list); | |
426 | LIST_HEAD(reclaimed); | |
db830c46 | 427 | struct drbd_peer_request *peer_req, *t; |
e2b3032b | 428 | int err = 0; |
b411b363 | 429 | |
0500813f | 430 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 AG |
431 | reclaim_finished_net_peer_reqs(device, &reclaimed); |
432 | list_splice_init(&device->done_ee, &work_list); | |
0500813f | 433 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 434 | |
a8cd15ba | 435 | list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) |
b30ab791 | 436 | drbd_free_net_peer_req(device, peer_req); |
b411b363 PR |
437 | |
438 | /* possible callbacks here: | |
d4dabbe2 | 439 | * e_end_block, and e_end_resync_block, e_send_superseded. |
b411b363 PR |
440 | * all ignore the last argument. |
441 | */ | |
a8cd15ba | 442 | list_for_each_entry_safe(peer_req, t, &work_list, w.list) { |
e2b3032b AG |
443 | int err2; |
444 | ||
b411b363 | 445 | /* list_del not necessary, next/prev members not touched */ |
a8cd15ba | 446 | err2 = peer_req->w.cb(&peer_req->w, !!err); |
e2b3032b AG |
447 | if (!err) |
448 | err = err2; | |
b30ab791 | 449 | drbd_free_peer_req(device, peer_req); |
b411b363 | 450 | } |
b30ab791 | 451 | wake_up(&device->ee_wait); |
b411b363 | 452 | |
e2b3032b | 453 | return err; |
b411b363 PR |
454 | } |
455 | ||
b30ab791 | 456 | static void _drbd_wait_ee_list_empty(struct drbd_device *device, |
d4da1537 | 457 | struct list_head *head) |
b411b363 PR |
458 | { |
459 | DEFINE_WAIT(wait); | |
460 | ||
461 | /* avoids spin_lock/unlock | |
462 | * and calling prepare_to_wait in the fast path */ | |
463 | while (!list_empty(head)) { | |
b30ab791 | 464 | prepare_to_wait(&device->ee_wait, &wait, TASK_UNINTERRUPTIBLE); |
0500813f | 465 | spin_unlock_irq(&device->resource->req_lock); |
7eaceacc | 466 | io_schedule(); |
b30ab791 | 467 | finish_wait(&device->ee_wait, &wait); |
0500813f | 468 | spin_lock_irq(&device->resource->req_lock); |
b411b363 PR |
469 | } |
470 | } | |
471 | ||
b30ab791 | 472 | static void drbd_wait_ee_list_empty(struct drbd_device *device, |
d4da1537 | 473 | struct list_head *head) |
b411b363 | 474 | { |
0500813f | 475 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 476 | _drbd_wait_ee_list_empty(device, head); |
0500813f | 477 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
478 | } |
479 | ||
dbd9eea0 | 480 | static int drbd_recv_short(struct socket *sock, void *buf, size_t size, int flags) |
b411b363 | 481 | { |
b411b363 PR |
482 | struct kvec iov = { |
483 | .iov_base = buf, | |
484 | .iov_len = size, | |
485 | }; | |
486 | struct msghdr msg = { | |
b411b363 PR |
487 | .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) |
488 | }; | |
f730c848 | 489 | return kernel_recvmsg(sock, &msg, &iov, 1, size, msg.msg_flags); |
b411b363 PR |
490 | } |
491 | ||
bde89a9e | 492 | static int drbd_recv(struct drbd_connection *connection, void *buf, size_t size) |
b411b363 | 493 | { |
b411b363 PR |
494 | int rv; |
495 | ||
bde89a9e | 496 | rv = drbd_recv_short(connection->data.socket, buf, size, 0); |
b411b363 | 497 | |
dbd0820c PR |
498 | if (rv < 0) { |
499 | if (rv == -ECONNRESET) | |
1ec861eb | 500 | drbd_info(connection, "sock was reset by peer\n"); |
dbd0820c | 501 | else if (rv != -ERESTARTSYS) |
1ec861eb | 502 | drbd_err(connection, "sock_recvmsg returned %d\n", rv); |
dbd0820c | 503 | } else if (rv == 0) { |
bde89a9e | 504 | if (test_bit(DISCONNECT_SENT, &connection->flags)) { |
b66623e3 PR |
505 | long t; |
506 | rcu_read_lock(); | |
bde89a9e | 507 | t = rcu_dereference(connection->net_conf)->ping_timeo * HZ/10; |
b66623e3 PR |
508 | rcu_read_unlock(); |
509 | ||
bde89a9e | 510 | t = wait_event_timeout(connection->ping_wait, connection->cstate < C_WF_REPORT_PARAMS, t); |
b66623e3 | 511 | |
599377ac PR |
512 | if (t) |
513 | goto out; | |
514 | } | |
1ec861eb | 515 | drbd_info(connection, "sock was shut down by peer\n"); |
599377ac PR |
516 | } |
517 | ||
b411b363 | 518 | if (rv != size) |
bde89a9e | 519 | conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 | 520 | |
599377ac | 521 | out: |
b411b363 PR |
522 | return rv; |
523 | } | |
524 | ||
bde89a9e | 525 | static int drbd_recv_all(struct drbd_connection *connection, void *buf, size_t size) |
c6967746 AG |
526 | { |
527 | int err; | |
528 | ||
bde89a9e | 529 | err = drbd_recv(connection, buf, size); |
c6967746 AG |
530 | if (err != size) { |
531 | if (err >= 0) | |
532 | err = -EIO; | |
533 | } else | |
534 | err = 0; | |
535 | return err; | |
536 | } | |
537 | ||
bde89a9e | 538 | static int drbd_recv_all_warn(struct drbd_connection *connection, void *buf, size_t size) |
a5c31904 AG |
539 | { |
540 | int err; | |
541 | ||
bde89a9e | 542 | err = drbd_recv_all(connection, buf, size); |
a5c31904 | 543 | if (err && !signal_pending(current)) |
1ec861eb | 544 | drbd_warn(connection, "short read (expected size %d)\n", (int)size); |
a5c31904 AG |
545 | return err; |
546 | } | |
547 | ||
5dbf1673 LE |
548 | /* quoting tcp(7): |
549 | * On individual connections, the socket buffer size must be set prior to the | |
550 | * listen(2) or connect(2) calls in order to have it take effect. | |
551 | * This is our wrapper to do so. | |
552 | */ | |
553 | static void drbd_setbufsize(struct socket *sock, unsigned int snd, | |
554 | unsigned int rcv) | |
555 | { | |
556 | /* open coded SO_SNDBUF, SO_RCVBUF */ | |
557 | if (snd) { | |
558 | sock->sk->sk_sndbuf = snd; | |
559 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
560 | } | |
561 | if (rcv) { | |
562 | sock->sk->sk_rcvbuf = rcv; | |
563 | sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
564 | } | |
565 | } | |
566 | ||
bde89a9e | 567 | static struct socket *drbd_try_connect(struct drbd_connection *connection) |
b411b363 PR |
568 | { |
569 | const char *what; | |
570 | struct socket *sock; | |
571 | struct sockaddr_in6 src_in6; | |
44ed167d PR |
572 | struct sockaddr_in6 peer_in6; |
573 | struct net_conf *nc; | |
574 | int err, peer_addr_len, my_addr_len; | |
69ef82de | 575 | int sndbuf_size, rcvbuf_size, connect_int; |
b411b363 PR |
576 | int disconnect_on_error = 1; |
577 | ||
44ed167d | 578 | rcu_read_lock(); |
bde89a9e | 579 | nc = rcu_dereference(connection->net_conf); |
44ed167d PR |
580 | if (!nc) { |
581 | rcu_read_unlock(); | |
b411b363 | 582 | return NULL; |
44ed167d | 583 | } |
44ed167d PR |
584 | sndbuf_size = nc->sndbuf_size; |
585 | rcvbuf_size = nc->rcvbuf_size; | |
69ef82de | 586 | connect_int = nc->connect_int; |
089c075d | 587 | rcu_read_unlock(); |
44ed167d | 588 | |
bde89a9e AG |
589 | my_addr_len = min_t(int, connection->my_addr_len, sizeof(src_in6)); |
590 | memcpy(&src_in6, &connection->my_addr, my_addr_len); | |
44ed167d | 591 | |
bde89a9e | 592 | if (((struct sockaddr *)&connection->my_addr)->sa_family == AF_INET6) |
44ed167d PR |
593 | src_in6.sin6_port = 0; |
594 | else | |
595 | ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ | |
596 | ||
bde89a9e AG |
597 | peer_addr_len = min_t(int, connection->peer_addr_len, sizeof(src_in6)); |
598 | memcpy(&peer_in6, &connection->peer_addr, peer_addr_len); | |
b411b363 PR |
599 | |
600 | what = "sock_create_kern"; | |
44ed167d PR |
601 | err = sock_create_kern(((struct sockaddr *)&src_in6)->sa_family, |
602 | SOCK_STREAM, IPPROTO_TCP, &sock); | |
b411b363 PR |
603 | if (err < 0) { |
604 | sock = NULL; | |
605 | goto out; | |
606 | } | |
607 | ||
608 | sock->sk->sk_rcvtimeo = | |
69ef82de | 609 | sock->sk->sk_sndtimeo = connect_int * HZ; |
44ed167d | 610 | drbd_setbufsize(sock, sndbuf_size, rcvbuf_size); |
b411b363 PR |
611 | |
612 | /* explicitly bind to the configured IP as source IP | |
613 | * for the outgoing connections. | |
614 | * This is needed for multihomed hosts and to be | |
615 | * able to use lo: interfaces for drbd. | |
616 | * Make sure to use 0 as port number, so linux selects | |
617 | * a free one dynamically. | |
618 | */ | |
b411b363 | 619 | what = "bind before connect"; |
44ed167d | 620 | err = sock->ops->bind(sock, (struct sockaddr *) &src_in6, my_addr_len); |
b411b363 PR |
621 | if (err < 0) |
622 | goto out; | |
623 | ||
624 | /* connect may fail, peer not yet available. | |
625 | * stay C_WF_CONNECTION, don't go Disconnecting! */ | |
626 | disconnect_on_error = 0; | |
627 | what = "connect"; | |
44ed167d | 628 | err = sock->ops->connect(sock, (struct sockaddr *) &peer_in6, peer_addr_len, 0); |
b411b363 PR |
629 | |
630 | out: | |
631 | if (err < 0) { | |
632 | if (sock) { | |
633 | sock_release(sock); | |
634 | sock = NULL; | |
635 | } | |
636 | switch (-err) { | |
637 | /* timeout, busy, signal pending */ | |
638 | case ETIMEDOUT: case EAGAIN: case EINPROGRESS: | |
639 | case EINTR: case ERESTARTSYS: | |
640 | /* peer not (yet) available, network problem */ | |
641 | case ECONNREFUSED: case ENETUNREACH: | |
642 | case EHOSTDOWN: case EHOSTUNREACH: | |
643 | disconnect_on_error = 0; | |
644 | break; | |
645 | default: | |
1ec861eb | 646 | drbd_err(connection, "%s failed, err = %d\n", what, err); |
b411b363 PR |
647 | } |
648 | if (disconnect_on_error) | |
bde89a9e | 649 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 650 | } |
44ed167d | 651 | |
b411b363 PR |
652 | return sock; |
653 | } | |
654 | ||
7a426fd8 | 655 | struct accept_wait_data { |
bde89a9e | 656 | struct drbd_connection *connection; |
7a426fd8 PR |
657 | struct socket *s_listen; |
658 | struct completion door_bell; | |
659 | void (*original_sk_state_change)(struct sock *sk); | |
660 | ||
661 | }; | |
662 | ||
715306f6 | 663 | static void drbd_incoming_connection(struct sock *sk) |
7a426fd8 PR |
664 | { |
665 | struct accept_wait_data *ad = sk->sk_user_data; | |
715306f6 | 666 | void (*state_change)(struct sock *sk); |
7a426fd8 | 667 | |
715306f6 AG |
668 | state_change = ad->original_sk_state_change; |
669 | if (sk->sk_state == TCP_ESTABLISHED) | |
670 | complete(&ad->door_bell); | |
671 | state_change(sk); | |
7a426fd8 PR |
672 | } |
673 | ||
bde89a9e | 674 | static int prepare_listen_socket(struct drbd_connection *connection, struct accept_wait_data *ad) |
b411b363 | 675 | { |
1f3e509b | 676 | int err, sndbuf_size, rcvbuf_size, my_addr_len; |
44ed167d | 677 | struct sockaddr_in6 my_addr; |
1f3e509b | 678 | struct socket *s_listen; |
44ed167d | 679 | struct net_conf *nc; |
b411b363 PR |
680 | const char *what; |
681 | ||
44ed167d | 682 | rcu_read_lock(); |
bde89a9e | 683 | nc = rcu_dereference(connection->net_conf); |
44ed167d PR |
684 | if (!nc) { |
685 | rcu_read_unlock(); | |
7a426fd8 | 686 | return -EIO; |
44ed167d | 687 | } |
44ed167d PR |
688 | sndbuf_size = nc->sndbuf_size; |
689 | rcvbuf_size = nc->rcvbuf_size; | |
44ed167d | 690 | rcu_read_unlock(); |
b411b363 | 691 | |
bde89a9e AG |
692 | my_addr_len = min_t(int, connection->my_addr_len, sizeof(struct sockaddr_in6)); |
693 | memcpy(&my_addr, &connection->my_addr, my_addr_len); | |
b411b363 PR |
694 | |
695 | what = "sock_create_kern"; | |
44ed167d | 696 | err = sock_create_kern(((struct sockaddr *)&my_addr)->sa_family, |
1f3e509b | 697 | SOCK_STREAM, IPPROTO_TCP, &s_listen); |
b411b363 PR |
698 | if (err) { |
699 | s_listen = NULL; | |
700 | goto out; | |
701 | } | |
702 | ||
98683650 | 703 | s_listen->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ |
44ed167d | 704 | drbd_setbufsize(s_listen, sndbuf_size, rcvbuf_size); |
b411b363 PR |
705 | |
706 | what = "bind before listen"; | |
44ed167d | 707 | err = s_listen->ops->bind(s_listen, (struct sockaddr *)&my_addr, my_addr_len); |
b411b363 PR |
708 | if (err < 0) |
709 | goto out; | |
710 | ||
7a426fd8 PR |
711 | ad->s_listen = s_listen; |
712 | write_lock_bh(&s_listen->sk->sk_callback_lock); | |
713 | ad->original_sk_state_change = s_listen->sk->sk_state_change; | |
715306f6 | 714 | s_listen->sk->sk_state_change = drbd_incoming_connection; |
7a426fd8 PR |
715 | s_listen->sk->sk_user_data = ad; |
716 | write_unlock_bh(&s_listen->sk->sk_callback_lock); | |
b411b363 | 717 | |
2820fd39 PR |
718 | what = "listen"; |
719 | err = s_listen->ops->listen(s_listen, 5); | |
720 | if (err < 0) | |
721 | goto out; | |
722 | ||
7a426fd8 | 723 | return 0; |
b411b363 PR |
724 | out: |
725 | if (s_listen) | |
726 | sock_release(s_listen); | |
727 | if (err < 0) { | |
728 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
1ec861eb | 729 | drbd_err(connection, "%s failed, err = %d\n", what, err); |
bde89a9e | 730 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
731 | } |
732 | } | |
b411b363 | 733 | |
7a426fd8 | 734 | return -EIO; |
b411b363 PR |
735 | } |
736 | ||
715306f6 | 737 | static void unregister_state_change(struct sock *sk, struct accept_wait_data *ad) |
b411b363 | 738 | { |
715306f6 AG |
739 | write_lock_bh(&sk->sk_callback_lock); |
740 | sk->sk_state_change = ad->original_sk_state_change; | |
741 | sk->sk_user_data = NULL; | |
742 | write_unlock_bh(&sk->sk_callback_lock); | |
b411b363 PR |
743 | } |
744 | ||
bde89a9e | 745 | static struct socket *drbd_wait_for_connect(struct drbd_connection *connection, struct accept_wait_data *ad) |
b411b363 | 746 | { |
1f3e509b PR |
747 | int timeo, connect_int, err = 0; |
748 | struct socket *s_estab = NULL; | |
1f3e509b PR |
749 | struct net_conf *nc; |
750 | ||
751 | rcu_read_lock(); | |
bde89a9e | 752 | nc = rcu_dereference(connection->net_conf); |
1f3e509b PR |
753 | if (!nc) { |
754 | rcu_read_unlock(); | |
755 | return NULL; | |
756 | } | |
757 | connect_int = nc->connect_int; | |
758 | rcu_read_unlock(); | |
759 | ||
760 | timeo = connect_int * HZ; | |
38b682b2 AM |
761 | /* 28.5% random jitter */ |
762 | timeo += (prandom_u32() & 1) ? timeo / 7 : -timeo / 7; | |
1f3e509b | 763 | |
7a426fd8 PR |
764 | err = wait_for_completion_interruptible_timeout(&ad->door_bell, timeo); |
765 | if (err <= 0) | |
766 | return NULL; | |
b411b363 | 767 | |
7a426fd8 | 768 | err = kernel_accept(ad->s_listen, &s_estab, 0); |
b411b363 PR |
769 | if (err < 0) { |
770 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
1ec861eb | 771 | drbd_err(connection, "accept failed, err = %d\n", err); |
bde89a9e | 772 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
773 | } |
774 | } | |
b411b363 | 775 | |
715306f6 AG |
776 | if (s_estab) |
777 | unregister_state_change(s_estab->sk, ad); | |
b411b363 | 778 | |
b411b363 PR |
779 | return s_estab; |
780 | } | |
b411b363 | 781 | |
bde89a9e | 782 | static int decode_header(struct drbd_connection *, void *, struct packet_info *); |
b411b363 | 783 | |
bde89a9e | 784 | static int send_first_packet(struct drbd_connection *connection, struct drbd_socket *sock, |
9f5bdc33 AG |
785 | enum drbd_packet cmd) |
786 | { | |
bde89a9e | 787 | if (!conn_prepare_command(connection, sock)) |
9f5bdc33 | 788 | return -EIO; |
bde89a9e | 789 | return conn_send_command(connection, sock, cmd, 0, NULL, 0); |
b411b363 PR |
790 | } |
791 | ||
bde89a9e | 792 | static int receive_first_packet(struct drbd_connection *connection, struct socket *sock) |
b411b363 | 793 | { |
bde89a9e | 794 | unsigned int header_size = drbd_header_size(connection); |
9f5bdc33 | 795 | struct packet_info pi; |
4920e37a | 796 | struct net_conf *nc; |
9f5bdc33 | 797 | int err; |
b411b363 | 798 | |
4920e37a PR |
799 | rcu_read_lock(); |
800 | nc = rcu_dereference(connection->net_conf); | |
801 | if (!nc) { | |
802 | rcu_read_unlock(); | |
803 | return -EIO; | |
804 | } | |
805 | sock->sk->sk_rcvtimeo = nc->ping_timeo * 4 * HZ / 10; | |
806 | rcu_read_unlock(); | |
807 | ||
bde89a9e | 808 | err = drbd_recv_short(sock, connection->data.rbuf, header_size, 0); |
9f5bdc33 AG |
809 | if (err != header_size) { |
810 | if (err >= 0) | |
811 | err = -EIO; | |
812 | return err; | |
813 | } | |
bde89a9e | 814 | err = decode_header(connection, connection->data.rbuf, &pi); |
9f5bdc33 AG |
815 | if (err) |
816 | return err; | |
817 | return pi.cmd; | |
b411b363 PR |
818 | } |
819 | ||
820 | /** | |
821 | * drbd_socket_okay() - Free the socket if its connection is not okay | |
b411b363 PR |
822 | * @sock: pointer to the pointer to the socket. |
823 | */ | |
5d0b17f1 | 824 | static bool drbd_socket_okay(struct socket **sock) |
b411b363 PR |
825 | { |
826 | int rr; | |
827 | char tb[4]; | |
828 | ||
829 | if (!*sock) | |
81e84650 | 830 | return false; |
b411b363 | 831 | |
dbd9eea0 | 832 | rr = drbd_recv_short(*sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); |
b411b363 PR |
833 | |
834 | if (rr > 0 || rr == -EAGAIN) { | |
81e84650 | 835 | return true; |
b411b363 PR |
836 | } else { |
837 | sock_release(*sock); | |
838 | *sock = NULL; | |
81e84650 | 839 | return false; |
b411b363 PR |
840 | } |
841 | } | |
5d0b17f1 PR |
842 | |
843 | static bool connection_established(struct drbd_connection *connection, | |
844 | struct socket **sock1, | |
845 | struct socket **sock2) | |
846 | { | |
847 | struct net_conf *nc; | |
848 | int timeout; | |
849 | bool ok; | |
850 | ||
851 | if (!*sock1 || !*sock2) | |
852 | return false; | |
853 | ||
854 | rcu_read_lock(); | |
855 | nc = rcu_dereference(connection->net_conf); | |
856 | timeout = (nc->sock_check_timeo ?: nc->ping_timeo) * HZ / 10; | |
857 | rcu_read_unlock(); | |
858 | schedule_timeout_interruptible(timeout); | |
859 | ||
860 | ok = drbd_socket_okay(sock1); | |
861 | ok = drbd_socket_okay(sock2) && ok; | |
862 | ||
863 | return ok; | |
864 | } | |
865 | ||
2325eb66 PR |
866 | /* Gets called if a connection is established, or if a new minor gets created |
867 | in a connection */ | |
69a22773 | 868 | int drbd_connected(struct drbd_peer_device *peer_device) |
907599e0 | 869 | { |
69a22773 | 870 | struct drbd_device *device = peer_device->device; |
0829f5ed | 871 | int err; |
907599e0 | 872 | |
b30ab791 AG |
873 | atomic_set(&device->packet_seq, 0); |
874 | device->peer_seq = 0; | |
907599e0 | 875 | |
69a22773 AG |
876 | device->state_mutex = peer_device->connection->agreed_pro_version < 100 ? |
877 | &peer_device->connection->cstate_mutex : | |
b30ab791 | 878 | &device->own_state_mutex; |
8410da8f | 879 | |
69a22773 | 880 | err = drbd_send_sync_param(peer_device); |
0829f5ed | 881 | if (!err) |
69a22773 | 882 | err = drbd_send_sizes(peer_device, 0, 0); |
0829f5ed | 883 | if (!err) |
69a22773 | 884 | err = drbd_send_uuids(peer_device); |
0829f5ed | 885 | if (!err) |
69a22773 | 886 | err = drbd_send_current_state(peer_device); |
b30ab791 AG |
887 | clear_bit(USE_DEGR_WFC_T, &device->flags); |
888 | clear_bit(RESIZE_PENDING, &device->flags); | |
889 | atomic_set(&device->ap_in_flight, 0); | |
890 | mod_timer(&device->request_timer, jiffies + HZ); /* just start it here. */ | |
0829f5ed | 891 | return err; |
907599e0 | 892 | } |
b411b363 PR |
893 | |
894 | /* | |
895 | * return values: | |
896 | * 1 yes, we have a valid connection | |
897 | * 0 oops, did not work out, please try again | |
898 | * -1 peer talks different language, | |
899 | * no point in trying again, please go standalone. | |
900 | * -2 We do not have a network config... | |
901 | */ | |
bde89a9e | 902 | static int conn_connect(struct drbd_connection *connection) |
b411b363 | 903 | { |
7da35862 | 904 | struct drbd_socket sock, msock; |
c06ece6b | 905 | struct drbd_peer_device *peer_device; |
44ed167d | 906 | struct net_conf *nc; |
5d0b17f1 PR |
907 | int vnr, timeout, h; |
908 | bool discard_my_data, ok; | |
197296ff | 909 | enum drbd_state_rv rv; |
7a426fd8 | 910 | struct accept_wait_data ad = { |
bde89a9e | 911 | .connection = connection, |
7a426fd8 PR |
912 | .door_bell = COMPLETION_INITIALIZER_ONSTACK(ad.door_bell), |
913 | }; | |
b411b363 | 914 | |
bde89a9e AG |
915 | clear_bit(DISCONNECT_SENT, &connection->flags); |
916 | if (conn_request_state(connection, NS(conn, C_WF_CONNECTION), CS_VERBOSE) < SS_SUCCESS) | |
b411b363 PR |
917 | return -2; |
918 | ||
7da35862 | 919 | mutex_init(&sock.mutex); |
bde89a9e AG |
920 | sock.sbuf = connection->data.sbuf; |
921 | sock.rbuf = connection->data.rbuf; | |
7da35862 PR |
922 | sock.socket = NULL; |
923 | mutex_init(&msock.mutex); | |
bde89a9e AG |
924 | msock.sbuf = connection->meta.sbuf; |
925 | msock.rbuf = connection->meta.rbuf; | |
7da35862 PR |
926 | msock.socket = NULL; |
927 | ||
0916e0e3 | 928 | /* Assume that the peer only understands protocol 80 until we know better. */ |
bde89a9e | 929 | connection->agreed_pro_version = 80; |
b411b363 | 930 | |
bde89a9e | 931 | if (prepare_listen_socket(connection, &ad)) |
7a426fd8 | 932 | return 0; |
b411b363 PR |
933 | |
934 | do { | |
2bf89621 | 935 | struct socket *s; |
b411b363 | 936 | |
bde89a9e | 937 | s = drbd_try_connect(connection); |
b411b363 | 938 | if (s) { |
7da35862 PR |
939 | if (!sock.socket) { |
940 | sock.socket = s; | |
bde89a9e | 941 | send_first_packet(connection, &sock, P_INITIAL_DATA); |
7da35862 | 942 | } else if (!msock.socket) { |
bde89a9e | 943 | clear_bit(RESOLVE_CONFLICTS, &connection->flags); |
7da35862 | 944 | msock.socket = s; |
bde89a9e | 945 | send_first_packet(connection, &msock, P_INITIAL_META); |
b411b363 | 946 | } else { |
1ec861eb | 947 | drbd_err(connection, "Logic error in conn_connect()\n"); |
b411b363 PR |
948 | goto out_release_sockets; |
949 | } | |
950 | } | |
951 | ||
5d0b17f1 PR |
952 | if (connection_established(connection, &sock.socket, &msock.socket)) |
953 | break; | |
b411b363 PR |
954 | |
955 | retry: | |
bde89a9e | 956 | s = drbd_wait_for_connect(connection, &ad); |
b411b363 | 957 | if (s) { |
bde89a9e | 958 | int fp = receive_first_packet(connection, s); |
7da35862 PR |
959 | drbd_socket_okay(&sock.socket); |
960 | drbd_socket_okay(&msock.socket); | |
92f14951 | 961 | switch (fp) { |
e5d6f33a | 962 | case P_INITIAL_DATA: |
7da35862 | 963 | if (sock.socket) { |
1ec861eb | 964 | drbd_warn(connection, "initial packet S crossed\n"); |
7da35862 | 965 | sock_release(sock.socket); |
80c6eed4 PR |
966 | sock.socket = s; |
967 | goto randomize; | |
b411b363 | 968 | } |
7da35862 | 969 | sock.socket = s; |
b411b363 | 970 | break; |
e5d6f33a | 971 | case P_INITIAL_META: |
bde89a9e | 972 | set_bit(RESOLVE_CONFLICTS, &connection->flags); |
7da35862 | 973 | if (msock.socket) { |
1ec861eb | 974 | drbd_warn(connection, "initial packet M crossed\n"); |
7da35862 | 975 | sock_release(msock.socket); |
80c6eed4 PR |
976 | msock.socket = s; |
977 | goto randomize; | |
b411b363 | 978 | } |
7da35862 | 979 | msock.socket = s; |
b411b363 PR |
980 | break; |
981 | default: | |
1ec861eb | 982 | drbd_warn(connection, "Error receiving initial packet\n"); |
b411b363 | 983 | sock_release(s); |
80c6eed4 | 984 | randomize: |
38b682b2 | 985 | if (prandom_u32() & 1) |
b411b363 PR |
986 | goto retry; |
987 | } | |
988 | } | |
989 | ||
bde89a9e | 990 | if (connection->cstate <= C_DISCONNECTING) |
b411b363 PR |
991 | goto out_release_sockets; |
992 | if (signal_pending(current)) { | |
993 | flush_signals(current); | |
994 | smp_rmb(); | |
bde89a9e | 995 | if (get_t_state(&connection->receiver) == EXITING) |
b411b363 PR |
996 | goto out_release_sockets; |
997 | } | |
998 | ||
5d0b17f1 | 999 | ok = connection_established(connection, &sock.socket, &msock.socket); |
b666dbf8 | 1000 | } while (!ok); |
b411b363 | 1001 | |
7a426fd8 PR |
1002 | if (ad.s_listen) |
1003 | sock_release(ad.s_listen); | |
b411b363 | 1004 | |
98683650 PR |
1005 | sock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ |
1006 | msock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ | |
b411b363 | 1007 | |
7da35862 PR |
1008 | sock.socket->sk->sk_allocation = GFP_NOIO; |
1009 | msock.socket->sk->sk_allocation = GFP_NOIO; | |
b411b363 | 1010 | |
7da35862 PR |
1011 | sock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; |
1012 | msock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE; | |
b411b363 | 1013 | |
b411b363 | 1014 | /* NOT YET ... |
bde89a9e | 1015 | * sock.socket->sk->sk_sndtimeo = connection->net_conf->timeout*HZ/10; |
7da35862 | 1016 | * sock.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; |
6038178e | 1017 | * first set it to the P_CONNECTION_FEATURES timeout, |
b411b363 | 1018 | * which we set to 4x the configured ping_timeout. */ |
44ed167d | 1019 | rcu_read_lock(); |
bde89a9e | 1020 | nc = rcu_dereference(connection->net_conf); |
44ed167d | 1021 | |
7da35862 PR |
1022 | sock.socket->sk->sk_sndtimeo = |
1023 | sock.socket->sk->sk_rcvtimeo = nc->ping_timeo*4*HZ/10; | |
b411b363 | 1024 | |
7da35862 | 1025 | msock.socket->sk->sk_rcvtimeo = nc->ping_int*HZ; |
44ed167d | 1026 | timeout = nc->timeout * HZ / 10; |
08b165ba | 1027 | discard_my_data = nc->discard_my_data; |
44ed167d | 1028 | rcu_read_unlock(); |
b411b363 | 1029 | |
7da35862 | 1030 | msock.socket->sk->sk_sndtimeo = timeout; |
b411b363 PR |
1031 | |
1032 | /* we don't want delays. | |
25985edc | 1033 | * we use TCP_CORK where appropriate, though */ |
7da35862 PR |
1034 | drbd_tcp_nodelay(sock.socket); |
1035 | drbd_tcp_nodelay(msock.socket); | |
b411b363 | 1036 | |
bde89a9e AG |
1037 | connection->data.socket = sock.socket; |
1038 | connection->meta.socket = msock.socket; | |
1039 | connection->last_received = jiffies; | |
b411b363 | 1040 | |
bde89a9e | 1041 | h = drbd_do_features(connection); |
b411b363 PR |
1042 | if (h <= 0) |
1043 | return h; | |
1044 | ||
bde89a9e | 1045 | if (connection->cram_hmac_tfm) { |
b30ab791 | 1046 | /* drbd_request_state(device, NS(conn, WFAuth)); */ |
bde89a9e | 1047 | switch (drbd_do_auth(connection)) { |
b10d96cb | 1048 | case -1: |
1ec861eb | 1049 | drbd_err(connection, "Authentication of peer failed\n"); |
b411b363 | 1050 | return -1; |
b10d96cb | 1051 | case 0: |
1ec861eb | 1052 | drbd_err(connection, "Authentication of peer failed, trying again.\n"); |
b10d96cb | 1053 | return 0; |
b411b363 PR |
1054 | } |
1055 | } | |
1056 | ||
bde89a9e AG |
1057 | connection->data.socket->sk->sk_sndtimeo = timeout; |
1058 | connection->data.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; | |
b411b363 | 1059 | |
bde89a9e | 1060 | if (drbd_send_protocol(connection) == -EOPNOTSUPP) |
7e2455c1 | 1061 | return -1; |
b411b363 | 1062 | |
31007745 PR |
1063 | /* Prevent a race between resync-handshake and |
1064 | * being promoted to Primary. | |
1065 | * | |
1066 | * Grab and release the state mutex, so we know that any current | |
1067 | * drbd_set_role() is finished, and any incoming drbd_set_role | |
1068 | * will see the STATE_SENT flag, and wait for it to be cleared. | |
1069 | */ | |
1070 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) | |
1071 | mutex_lock(peer_device->device->state_mutex); | |
1072 | ||
bde89a9e | 1073 | set_bit(STATE_SENT, &connection->flags); |
a1096a6e | 1074 | |
31007745 PR |
1075 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) |
1076 | mutex_unlock(peer_device->device->state_mutex); | |
1077 | ||
c141ebda | 1078 | rcu_read_lock(); |
c06ece6b AG |
1079 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
1080 | struct drbd_device *device = peer_device->device; | |
b30ab791 | 1081 | kref_get(&device->kref); |
26ea8f92 AG |
1082 | rcu_read_unlock(); |
1083 | ||
08b165ba | 1084 | if (discard_my_data) |
b30ab791 | 1085 | set_bit(DISCARD_MY_DATA, &device->flags); |
08b165ba | 1086 | else |
b30ab791 | 1087 | clear_bit(DISCARD_MY_DATA, &device->flags); |
08b165ba | 1088 | |
69a22773 | 1089 | drbd_connected(peer_device); |
05a10ec7 | 1090 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda PR |
1091 | rcu_read_lock(); |
1092 | } | |
1093 | rcu_read_unlock(); | |
1094 | ||
bde89a9e AG |
1095 | rv = conn_request_state(connection, NS(conn, C_WF_REPORT_PARAMS), CS_VERBOSE); |
1096 | if (rv < SS_SUCCESS || connection->cstate != C_WF_REPORT_PARAMS) { | |
1097 | clear_bit(STATE_SENT, &connection->flags); | |
1e86ac48 | 1098 | return 0; |
a1096a6e | 1099 | } |
1e86ac48 | 1100 | |
bde89a9e | 1101 | drbd_thread_start(&connection->asender); |
b411b363 | 1102 | |
0500813f | 1103 | mutex_lock(&connection->resource->conf_update); |
08b165ba PR |
1104 | /* The discard_my_data flag is a single-shot modifier to the next |
1105 | * connection attempt, the handshake of which is now well underway. | |
1106 | * No need for rcu style copying of the whole struct | |
1107 | * just to clear a single value. */ | |
bde89a9e | 1108 | connection->net_conf->discard_my_data = 0; |
0500813f | 1109 | mutex_unlock(&connection->resource->conf_update); |
08b165ba | 1110 | |
d3fcb490 | 1111 | return h; |
b411b363 PR |
1112 | |
1113 | out_release_sockets: | |
7a426fd8 PR |
1114 | if (ad.s_listen) |
1115 | sock_release(ad.s_listen); | |
7da35862 PR |
1116 | if (sock.socket) |
1117 | sock_release(sock.socket); | |
1118 | if (msock.socket) | |
1119 | sock_release(msock.socket); | |
b411b363 PR |
1120 | return -1; |
1121 | } | |
1122 | ||
bde89a9e | 1123 | static int decode_header(struct drbd_connection *connection, void *header, struct packet_info *pi) |
b411b363 | 1124 | { |
bde89a9e | 1125 | unsigned int header_size = drbd_header_size(connection); |
e658983a | 1126 | |
0c8e36d9 AG |
1127 | if (header_size == sizeof(struct p_header100) && |
1128 | *(__be32 *)header == cpu_to_be32(DRBD_MAGIC_100)) { | |
1129 | struct p_header100 *h = header; | |
1130 | if (h->pad != 0) { | |
1ec861eb | 1131 | drbd_err(connection, "Header padding is not zero\n"); |
0c8e36d9 AG |
1132 | return -EINVAL; |
1133 | } | |
1134 | pi->vnr = be16_to_cpu(h->volume); | |
1135 | pi->cmd = be16_to_cpu(h->command); | |
1136 | pi->size = be32_to_cpu(h->length); | |
1137 | } else if (header_size == sizeof(struct p_header95) && | |
1138 | *(__be16 *)header == cpu_to_be16(DRBD_MAGIC_BIG)) { | |
e658983a | 1139 | struct p_header95 *h = header; |
e658983a | 1140 | pi->cmd = be16_to_cpu(h->command); |
b55d84ba AG |
1141 | pi->size = be32_to_cpu(h->length); |
1142 | pi->vnr = 0; | |
e658983a AG |
1143 | } else if (header_size == sizeof(struct p_header80) && |
1144 | *(__be32 *)header == cpu_to_be32(DRBD_MAGIC)) { | |
1145 | struct p_header80 *h = header; | |
1146 | pi->cmd = be16_to_cpu(h->command); | |
1147 | pi->size = be16_to_cpu(h->length); | |
77351055 | 1148 | pi->vnr = 0; |
02918be2 | 1149 | } else { |
1ec861eb | 1150 | drbd_err(connection, "Wrong magic value 0x%08x in protocol version %d\n", |
e658983a | 1151 | be32_to_cpu(*(__be32 *)header), |
bde89a9e | 1152 | connection->agreed_pro_version); |
8172f3e9 | 1153 | return -EINVAL; |
b411b363 | 1154 | } |
e658983a | 1155 | pi->data = header + header_size; |
8172f3e9 | 1156 | return 0; |
257d0af6 | 1157 | } |
b411b363 | 1158 | |
bde89a9e | 1159 | static int drbd_recv_header(struct drbd_connection *connection, struct packet_info *pi) |
257d0af6 | 1160 | { |
bde89a9e | 1161 | void *buffer = connection->data.rbuf; |
69bc7bc3 | 1162 | int err; |
257d0af6 | 1163 | |
bde89a9e | 1164 | err = drbd_recv_all_warn(connection, buffer, drbd_header_size(connection)); |
a5c31904 | 1165 | if (err) |
69bc7bc3 | 1166 | return err; |
257d0af6 | 1167 | |
bde89a9e AG |
1168 | err = decode_header(connection, buffer, pi); |
1169 | connection->last_received = jiffies; | |
b411b363 | 1170 | |
69bc7bc3 | 1171 | return err; |
b411b363 PR |
1172 | } |
1173 | ||
bde89a9e | 1174 | static void drbd_flush(struct drbd_connection *connection) |
b411b363 PR |
1175 | { |
1176 | int rv; | |
c06ece6b | 1177 | struct drbd_peer_device *peer_device; |
4b0007c0 PR |
1178 | int vnr; |
1179 | ||
e9526580 | 1180 | if (connection->resource->write_ordering >= WO_bdev_flush) { |
615e087f | 1181 | rcu_read_lock(); |
c06ece6b AG |
1182 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
1183 | struct drbd_device *device = peer_device->device; | |
1184 | ||
b30ab791 | 1185 | if (!get_ldev(device)) |
615e087f | 1186 | continue; |
b30ab791 | 1187 | kref_get(&device->kref); |
615e087f LE |
1188 | rcu_read_unlock(); |
1189 | ||
f418815f LE |
1190 | /* Right now, we have only this one synchronous code path |
1191 | * for flushes between request epochs. | |
1192 | * We may want to make those asynchronous, | |
1193 | * or at least parallelize the flushes to the volume devices. | |
1194 | */ | |
1195 | device->flush_jif = jiffies; | |
1196 | set_bit(FLUSH_PENDING, &device->flags); | |
b30ab791 | 1197 | rv = blkdev_issue_flush(device->ldev->backing_bdev, |
615e087f | 1198 | GFP_NOIO, NULL); |
f418815f | 1199 | clear_bit(FLUSH_PENDING, &device->flags); |
615e087f | 1200 | if (rv) { |
d0180171 | 1201 | drbd_info(device, "local disk flush failed with status %d\n", rv); |
615e087f LE |
1202 | /* would rather check on EOPNOTSUPP, but that is not reliable. |
1203 | * don't try again for ANY return value != 0 | |
1204 | * if (rv == -EOPNOTSUPP) */ | |
8fe39aac | 1205 | drbd_bump_write_ordering(connection->resource, NULL, WO_drain_io); |
4b0007c0 | 1206 | } |
b30ab791 | 1207 | put_ldev(device); |
05a10ec7 | 1208 | kref_put(&device->kref, drbd_destroy_device); |
b411b363 | 1209 | |
615e087f LE |
1210 | rcu_read_lock(); |
1211 | if (rv) | |
1212 | break; | |
b411b363 | 1213 | } |
615e087f | 1214 | rcu_read_unlock(); |
b411b363 | 1215 | } |
b411b363 PR |
1216 | } |
1217 | ||
1218 | /** | |
1219 | * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. | |
b30ab791 | 1220 | * @device: DRBD device. |
b411b363 PR |
1221 | * @epoch: Epoch object. |
1222 | * @ev: Epoch event. | |
1223 | */ | |
bde89a9e | 1224 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_connection *connection, |
b411b363 PR |
1225 | struct drbd_epoch *epoch, |
1226 | enum epoch_event ev) | |
1227 | { | |
2451fc3b | 1228 | int epoch_size; |
b411b363 | 1229 | struct drbd_epoch *next_epoch; |
b411b363 PR |
1230 | enum finish_epoch rv = FE_STILL_LIVE; |
1231 | ||
bde89a9e | 1232 | spin_lock(&connection->epoch_lock); |
b411b363 PR |
1233 | do { |
1234 | next_epoch = NULL; | |
b411b363 PR |
1235 | |
1236 | epoch_size = atomic_read(&epoch->epoch_size); | |
1237 | ||
1238 | switch (ev & ~EV_CLEANUP) { | |
1239 | case EV_PUT: | |
1240 | atomic_dec(&epoch->active); | |
1241 | break; | |
1242 | case EV_GOT_BARRIER_NR: | |
1243 | set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); | |
b411b363 PR |
1244 | break; |
1245 | case EV_BECAME_LAST: | |
1246 | /* nothing to do*/ | |
1247 | break; | |
1248 | } | |
1249 | ||
b411b363 PR |
1250 | if (epoch_size != 0 && |
1251 | atomic_read(&epoch->active) == 0 && | |
80f9fd55 | 1252 | (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) || ev & EV_CLEANUP)) { |
b411b363 | 1253 | if (!(ev & EV_CLEANUP)) { |
bde89a9e AG |
1254 | spin_unlock(&connection->epoch_lock); |
1255 | drbd_send_b_ack(epoch->connection, epoch->barrier_nr, epoch_size); | |
1256 | spin_lock(&connection->epoch_lock); | |
b411b363 | 1257 | } |
9ed57dcb LE |
1258 | #if 0 |
1259 | /* FIXME: dec unacked on connection, once we have | |
1260 | * something to count pending connection packets in. */ | |
80f9fd55 | 1261 | if (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) |
bde89a9e | 1262 | dec_unacked(epoch->connection); |
9ed57dcb | 1263 | #endif |
b411b363 | 1264 | |
bde89a9e | 1265 | if (connection->current_epoch != epoch) { |
b411b363 PR |
1266 | next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); |
1267 | list_del(&epoch->list); | |
1268 | ev = EV_BECAME_LAST | (ev & EV_CLEANUP); | |
bde89a9e | 1269 | connection->epochs--; |
b411b363 PR |
1270 | kfree(epoch); |
1271 | ||
1272 | if (rv == FE_STILL_LIVE) | |
1273 | rv = FE_DESTROYED; | |
1274 | } else { | |
1275 | epoch->flags = 0; | |
1276 | atomic_set(&epoch->epoch_size, 0); | |
698f9315 | 1277 | /* atomic_set(&epoch->active, 0); is already zero */ |
b411b363 PR |
1278 | if (rv == FE_STILL_LIVE) |
1279 | rv = FE_RECYCLED; | |
1280 | } | |
1281 | } | |
1282 | ||
1283 | if (!next_epoch) | |
1284 | break; | |
1285 | ||
1286 | epoch = next_epoch; | |
1287 | } while (1); | |
1288 | ||
bde89a9e | 1289 | spin_unlock(&connection->epoch_lock); |
b411b363 | 1290 | |
b411b363 PR |
1291 | return rv; |
1292 | } | |
1293 | ||
8fe39aac PR |
1294 | static enum write_ordering_e |
1295 | max_allowed_wo(struct drbd_backing_dev *bdev, enum write_ordering_e wo) | |
1296 | { | |
1297 | struct disk_conf *dc; | |
1298 | ||
1299 | dc = rcu_dereference(bdev->disk_conf); | |
1300 | ||
1301 | if (wo == WO_bdev_flush && !dc->disk_flushes) | |
1302 | wo = WO_drain_io; | |
1303 | if (wo == WO_drain_io && !dc->disk_drain) | |
1304 | wo = WO_none; | |
1305 | ||
1306 | return wo; | |
1307 | } | |
1308 | ||
b411b363 PR |
1309 | /** |
1310 | * drbd_bump_write_ordering() - Fall back to an other write ordering method | |
bde89a9e | 1311 | * @connection: DRBD connection. |
b411b363 PR |
1312 | * @wo: Write ordering method to try. |
1313 | */ | |
8fe39aac PR |
1314 | void drbd_bump_write_ordering(struct drbd_resource *resource, struct drbd_backing_dev *bdev, |
1315 | enum write_ordering_e wo) | |
b411b363 | 1316 | { |
e9526580 | 1317 | struct drbd_device *device; |
b411b363 | 1318 | enum write_ordering_e pwo; |
4b0007c0 | 1319 | int vnr; |
b411b363 PR |
1320 | static char *write_ordering_str[] = { |
1321 | [WO_none] = "none", | |
1322 | [WO_drain_io] = "drain", | |
1323 | [WO_bdev_flush] = "flush", | |
b411b363 PR |
1324 | }; |
1325 | ||
e9526580 | 1326 | pwo = resource->write_ordering; |
70df7092 LE |
1327 | if (wo != WO_bdev_flush) |
1328 | wo = min(pwo, wo); | |
daeda1cc | 1329 | rcu_read_lock(); |
e9526580 | 1330 | idr_for_each_entry(&resource->devices, device, vnr) { |
8fe39aac PR |
1331 | if (get_ldev(device)) { |
1332 | wo = max_allowed_wo(device->ldev, wo); | |
1333 | if (device->ldev == bdev) | |
1334 | bdev = NULL; | |
1335 | put_ldev(device); | |
1336 | } | |
4b0007c0 | 1337 | } |
8fe39aac PR |
1338 | |
1339 | if (bdev) | |
1340 | wo = max_allowed_wo(bdev, wo); | |
1341 | ||
70df7092 LE |
1342 | rcu_read_unlock(); |
1343 | ||
e9526580 PR |
1344 | resource->write_ordering = wo; |
1345 | if (pwo != resource->write_ordering || wo == WO_bdev_flush) | |
1346 | drbd_info(resource, "Method to ensure write ordering: %s\n", write_ordering_str[resource->write_ordering]); | |
b411b363 PR |
1347 | } |
1348 | ||
45bb912b | 1349 | /** |
fbe29dec | 1350 | * drbd_submit_peer_request() |
b30ab791 | 1351 | * @device: DRBD device. |
db830c46 | 1352 | * @peer_req: peer request |
45bb912b | 1353 | * @rw: flag field, see bio->bi_rw |
10f6d992 LE |
1354 | * |
1355 | * May spread the pages to multiple bios, | |
1356 | * depending on bio_add_page restrictions. | |
1357 | * | |
1358 | * Returns 0 if all bios have been submitted, | |
1359 | * -ENOMEM if we could not allocate enough bios, | |
1360 | * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a | |
1361 | * single page to an empty bio (which should never happen and likely indicates | |
1362 | * that the lower level IO stack is in some way broken). This has been observed | |
1363 | * on certain Xen deployments. | |
45bb912b LE |
1364 | */ |
1365 | /* TODO allocate from our own bio_set. */ | |
b30ab791 | 1366 | int drbd_submit_peer_request(struct drbd_device *device, |
fbe29dec AG |
1367 | struct drbd_peer_request *peer_req, |
1368 | const unsigned rw, const int fault_type) | |
45bb912b LE |
1369 | { |
1370 | struct bio *bios = NULL; | |
1371 | struct bio *bio; | |
db830c46 AG |
1372 | struct page *page = peer_req->pages; |
1373 | sector_t sector = peer_req->i.sector; | |
11f8b2b6 | 1374 | unsigned data_size = peer_req->i.size; |
45bb912b | 1375 | unsigned n_bios = 0; |
11f8b2b6 | 1376 | unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT; |
10f6d992 | 1377 | int err = -ENOMEM; |
45bb912b | 1378 | |
a0fb3c47 LE |
1379 | if (peer_req->flags & EE_IS_TRIM_USE_ZEROOUT) { |
1380 | /* wait for all pending IO completions, before we start | |
1381 | * zeroing things out. */ | |
1382 | conn_wait_active_ee_empty(first_peer_device(device)->connection); | |
45d2933c LE |
1383 | /* add it to the active list now, |
1384 | * so we can find it to present it in debugfs */ | |
21ae5d7f LE |
1385 | peer_req->submit_jif = jiffies; |
1386 | peer_req->flags |= EE_SUBMITTED; | |
45d2933c LE |
1387 | spin_lock_irq(&device->resource->req_lock); |
1388 | list_add_tail(&peer_req->w.list, &device->active_ee); | |
1389 | spin_unlock_irq(&device->resource->req_lock); | |
a0fb3c47 | 1390 | if (blkdev_issue_zeroout(device->ldev->backing_bdev, |
d93ba7a5 | 1391 | sector, data_size >> 9, GFP_NOIO, false)) |
a0fb3c47 LE |
1392 | peer_req->flags |= EE_WAS_ERROR; |
1393 | drbd_endio_write_sec_final(peer_req); | |
1394 | return 0; | |
1395 | } | |
1396 | ||
54ed4ed8 LE |
1397 | /* Discards don't have any payload. |
1398 | * But the scsi layer still expects a bio_vec it can use internally, | |
1399 | * see sd_setup_discard_cmnd() and blk_add_request_payload(). */ | |
a0fb3c47 | 1400 | if (peer_req->flags & EE_IS_TRIM) |
54ed4ed8 | 1401 | nr_pages = 1; |
a0fb3c47 | 1402 | |
45bb912b LE |
1403 | /* In most cases, we will only need one bio. But in case the lower |
1404 | * level restrictions happen to be different at this offset on this | |
1405 | * side than those of the sending peer, we may need to submit the | |
9476f39d LE |
1406 | * request in more than one bio. |
1407 | * | |
1408 | * Plain bio_alloc is good enough here, this is no DRBD internally | |
1409 | * generated bio, but a bio allocated on behalf of the peer. | |
1410 | */ | |
45bb912b LE |
1411 | next_bio: |
1412 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
1413 | if (!bio) { | |
a0fb3c47 | 1414 | drbd_err(device, "submit_ee: Allocation of a bio failed (nr_pages=%u)\n", nr_pages); |
45bb912b LE |
1415 | goto fail; |
1416 | } | |
db830c46 | 1417 | /* > peer_req->i.sector, unless this is the first bio */ |
4f024f37 | 1418 | bio->bi_iter.bi_sector = sector; |
b30ab791 | 1419 | bio->bi_bdev = device->ldev->backing_bdev; |
45bb912b | 1420 | bio->bi_rw = rw; |
db830c46 | 1421 | bio->bi_private = peer_req; |
fcefa62e | 1422 | bio->bi_end_io = drbd_peer_request_endio; |
45bb912b LE |
1423 | |
1424 | bio->bi_next = bios; | |
1425 | bios = bio; | |
1426 | ++n_bios; | |
1427 | ||
a0fb3c47 | 1428 | if (rw & REQ_DISCARD) { |
11f8b2b6 | 1429 | bio->bi_iter.bi_size = data_size; |
a0fb3c47 LE |
1430 | goto submit; |
1431 | } | |
1432 | ||
45bb912b | 1433 | page_chain_for_each(page) { |
11f8b2b6 | 1434 | unsigned len = min_t(unsigned, data_size, PAGE_SIZE); |
45bb912b | 1435 | if (!bio_add_page(bio, page, len, 0)) { |
10f6d992 LE |
1436 | /* A single page must always be possible! |
1437 | * But in case it fails anyways, | |
1438 | * we deal with it, and complain (below). */ | |
1439 | if (bio->bi_vcnt == 0) { | |
d0180171 | 1440 | drbd_err(device, |
10f6d992 LE |
1441 | "bio_add_page failed for len=%u, " |
1442 | "bi_vcnt=0 (bi_sector=%llu)\n", | |
4f024f37 | 1443 | len, (uint64_t)bio->bi_iter.bi_sector); |
10f6d992 LE |
1444 | err = -ENOSPC; |
1445 | goto fail; | |
1446 | } | |
45bb912b LE |
1447 | goto next_bio; |
1448 | } | |
11f8b2b6 | 1449 | data_size -= len; |
45bb912b LE |
1450 | sector += len >> 9; |
1451 | --nr_pages; | |
1452 | } | |
11f8b2b6 | 1453 | D_ASSERT(device, data_size == 0); |
a0fb3c47 LE |
1454 | submit: |
1455 | D_ASSERT(device, page == NULL); | |
45bb912b | 1456 | |
db830c46 | 1457 | atomic_set(&peer_req->pending_bios, n_bios); |
21ae5d7f LE |
1458 | /* for debugfs: update timestamp, mark as submitted */ |
1459 | peer_req->submit_jif = jiffies; | |
1460 | peer_req->flags |= EE_SUBMITTED; | |
45bb912b LE |
1461 | do { |
1462 | bio = bios; | |
1463 | bios = bios->bi_next; | |
1464 | bio->bi_next = NULL; | |
1465 | ||
b30ab791 | 1466 | drbd_generic_make_request(device, fault_type, bio); |
45bb912b | 1467 | } while (bios); |
45bb912b LE |
1468 | return 0; |
1469 | ||
1470 | fail: | |
1471 | while (bios) { | |
1472 | bio = bios; | |
1473 | bios = bios->bi_next; | |
1474 | bio_put(bio); | |
1475 | } | |
10f6d992 | 1476 | return err; |
45bb912b LE |
1477 | } |
1478 | ||
b30ab791 | 1479 | static void drbd_remove_epoch_entry_interval(struct drbd_device *device, |
db830c46 | 1480 | struct drbd_peer_request *peer_req) |
53840641 | 1481 | { |
db830c46 | 1482 | struct drbd_interval *i = &peer_req->i; |
53840641 | 1483 | |
b30ab791 | 1484 | drbd_remove_interval(&device->write_requests, i); |
53840641 AG |
1485 | drbd_clear_interval(i); |
1486 | ||
6c852bec | 1487 | /* Wake up any processes waiting for this peer request to complete. */ |
53840641 | 1488 | if (i->waiting) |
b30ab791 | 1489 | wake_up(&device->misc_wait); |
53840641 AG |
1490 | } |
1491 | ||
bde89a9e | 1492 | static void conn_wait_active_ee_empty(struct drbd_connection *connection) |
77fede51 | 1493 | { |
c06ece6b | 1494 | struct drbd_peer_device *peer_device; |
77fede51 PR |
1495 | int vnr; |
1496 | ||
1497 | rcu_read_lock(); | |
c06ece6b AG |
1498 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
1499 | struct drbd_device *device = peer_device->device; | |
1500 | ||
b30ab791 | 1501 | kref_get(&device->kref); |
77fede51 | 1502 | rcu_read_unlock(); |
b30ab791 | 1503 | drbd_wait_ee_list_empty(device, &device->active_ee); |
05a10ec7 | 1504 | kref_put(&device->kref, drbd_destroy_device); |
77fede51 PR |
1505 | rcu_read_lock(); |
1506 | } | |
1507 | rcu_read_unlock(); | |
1508 | } | |
1509 | ||
9f4fe9ad AG |
1510 | static struct drbd_peer_device * |
1511 | conn_peer_device(struct drbd_connection *connection, int volume_number) | |
1512 | { | |
1513 | return idr_find(&connection->peer_devices, volume_number); | |
1514 | } | |
1515 | ||
bde89a9e | 1516 | static int receive_Barrier(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 1517 | { |
2451fc3b | 1518 | int rv; |
e658983a | 1519 | struct p_barrier *p = pi->data; |
b411b363 PR |
1520 | struct drbd_epoch *epoch; |
1521 | ||
9ed57dcb LE |
1522 | /* FIXME these are unacked on connection, |
1523 | * not a specific (peer)device. | |
1524 | */ | |
bde89a9e AG |
1525 | connection->current_epoch->barrier_nr = p->barrier; |
1526 | connection->current_epoch->connection = connection; | |
1527 | rv = drbd_may_finish_epoch(connection, connection->current_epoch, EV_GOT_BARRIER_NR); | |
b411b363 PR |
1528 | |
1529 | /* P_BARRIER_ACK may imply that the corresponding extent is dropped from | |
1530 | * the activity log, which means it would not be resynced in case the | |
1531 | * R_PRIMARY crashes now. | |
1532 | * Therefore we must send the barrier_ack after the barrier request was | |
1533 | * completed. */ | |
e9526580 | 1534 | switch (connection->resource->write_ordering) { |
b411b363 PR |
1535 | case WO_none: |
1536 | if (rv == FE_RECYCLED) | |
82bc0194 | 1537 | return 0; |
2451fc3b PR |
1538 | |
1539 | /* receiver context, in the writeout path of the other node. | |
1540 | * avoid potential distributed deadlock */ | |
1541 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1542 | if (epoch) | |
1543 | break; | |
1544 | else | |
1ec861eb | 1545 | drbd_warn(connection, "Allocation of an epoch failed, slowing down\n"); |
2451fc3b | 1546 | /* Fall through */ |
b411b363 PR |
1547 | |
1548 | case WO_bdev_flush: | |
1549 | case WO_drain_io: | |
bde89a9e AG |
1550 | conn_wait_active_ee_empty(connection); |
1551 | drbd_flush(connection); | |
2451fc3b | 1552 | |
bde89a9e | 1553 | if (atomic_read(&connection->current_epoch->epoch_size)) { |
2451fc3b PR |
1554 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); |
1555 | if (epoch) | |
1556 | break; | |
b411b363 PR |
1557 | } |
1558 | ||
82bc0194 | 1559 | return 0; |
2451fc3b | 1560 | default: |
e9526580 PR |
1561 | drbd_err(connection, "Strangeness in connection->write_ordering %d\n", |
1562 | connection->resource->write_ordering); | |
82bc0194 | 1563 | return -EIO; |
b411b363 PR |
1564 | } |
1565 | ||
1566 | epoch->flags = 0; | |
1567 | atomic_set(&epoch->epoch_size, 0); | |
1568 | atomic_set(&epoch->active, 0); | |
1569 | ||
bde89a9e AG |
1570 | spin_lock(&connection->epoch_lock); |
1571 | if (atomic_read(&connection->current_epoch->epoch_size)) { | |
1572 | list_add(&epoch->list, &connection->current_epoch->list); | |
1573 | connection->current_epoch = epoch; | |
1574 | connection->epochs++; | |
b411b363 PR |
1575 | } else { |
1576 | /* The current_epoch got recycled while we allocated this one... */ | |
1577 | kfree(epoch); | |
1578 | } | |
bde89a9e | 1579 | spin_unlock(&connection->epoch_lock); |
b411b363 | 1580 | |
82bc0194 | 1581 | return 0; |
b411b363 PR |
1582 | } |
1583 | ||
1584 | /* used from receive_RSDataReply (recv_resync_read) | |
1585 | * and from receive_Data */ | |
f6ffca9f | 1586 | static struct drbd_peer_request * |
69a22773 | 1587 | read_in_block(struct drbd_peer_device *peer_device, u64 id, sector_t sector, |
a0fb3c47 | 1588 | struct packet_info *pi) __must_hold(local) |
b411b363 | 1589 | { |
69a22773 | 1590 | struct drbd_device *device = peer_device->device; |
b30ab791 | 1591 | const sector_t capacity = drbd_get_capacity(device->this_bdev); |
db830c46 | 1592 | struct drbd_peer_request *peer_req; |
b411b363 | 1593 | struct page *page; |
11f8b2b6 AG |
1594 | int digest_size, err; |
1595 | unsigned int data_size = pi->size, ds; | |
69a22773 AG |
1596 | void *dig_in = peer_device->connection->int_dig_in; |
1597 | void *dig_vv = peer_device->connection->int_dig_vv; | |
6b4388ac | 1598 | unsigned long *data; |
a0fb3c47 | 1599 | struct p_trim *trim = (pi->cmd == P_TRIM) ? pi->data : NULL; |
b411b363 | 1600 | |
11f8b2b6 | 1601 | digest_size = 0; |
a0fb3c47 | 1602 | if (!trim && peer_device->connection->peer_integrity_tfm) { |
11f8b2b6 | 1603 | digest_size = crypto_hash_digestsize(peer_device->connection->peer_integrity_tfm); |
9f5bdc33 AG |
1604 | /* |
1605 | * FIXME: Receive the incoming digest into the receive buffer | |
1606 | * here, together with its struct p_data? | |
1607 | */ | |
11f8b2b6 | 1608 | err = drbd_recv_all_warn(peer_device->connection, dig_in, digest_size); |
a5c31904 | 1609 | if (err) |
b411b363 | 1610 | return NULL; |
11f8b2b6 | 1611 | data_size -= digest_size; |
b411b363 PR |
1612 | } |
1613 | ||
a0fb3c47 LE |
1614 | if (trim) { |
1615 | D_ASSERT(peer_device, data_size == 0); | |
1616 | data_size = be32_to_cpu(trim->size); | |
1617 | } | |
1618 | ||
841ce241 AG |
1619 | if (!expect(IS_ALIGNED(data_size, 512))) |
1620 | return NULL; | |
a0fb3c47 LE |
1621 | /* prepare for larger trim requests. */ |
1622 | if (!trim && !expect(data_size <= DRBD_MAX_BIO_SIZE)) | |
841ce241 | 1623 | return NULL; |
b411b363 | 1624 | |
6666032a LE |
1625 | /* even though we trust out peer, |
1626 | * we sometimes have to double check. */ | |
1627 | if (sector + (data_size>>9) > capacity) { | |
d0180171 | 1628 | drbd_err(device, "request from peer beyond end of local disk: " |
fdda6544 | 1629 | "capacity: %llus < sector: %llus + size: %u\n", |
6666032a LE |
1630 | (unsigned long long)capacity, |
1631 | (unsigned long long)sector, data_size); | |
1632 | return NULL; | |
1633 | } | |
1634 | ||
b411b363 PR |
1635 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD |
1636 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
1637 | * which in turn might block on the other node at this very place. */ | |
a0fb3c47 | 1638 | peer_req = drbd_alloc_peer_req(peer_device, id, sector, data_size, trim == NULL, GFP_NOIO); |
db830c46 | 1639 | if (!peer_req) |
b411b363 | 1640 | return NULL; |
45bb912b | 1641 | |
21ae5d7f | 1642 | peer_req->flags |= EE_WRITE; |
a0fb3c47 | 1643 | if (trim) |
81a3537a | 1644 | return peer_req; |
a73ff323 | 1645 | |
b411b363 | 1646 | ds = data_size; |
db830c46 | 1647 | page = peer_req->pages; |
45bb912b LE |
1648 | page_chain_for_each(page) { |
1649 | unsigned len = min_t(int, ds, PAGE_SIZE); | |
6b4388ac | 1650 | data = kmap(page); |
69a22773 | 1651 | err = drbd_recv_all_warn(peer_device->connection, data, len); |
b30ab791 | 1652 | if (drbd_insert_fault(device, DRBD_FAULT_RECEIVE)) { |
d0180171 | 1653 | drbd_err(device, "Fault injection: Corrupting data on receive\n"); |
6b4388ac PR |
1654 | data[0] = data[0] ^ (unsigned long)-1; |
1655 | } | |
b411b363 | 1656 | kunmap(page); |
a5c31904 | 1657 | if (err) { |
b30ab791 | 1658 | drbd_free_peer_req(device, peer_req); |
b411b363 PR |
1659 | return NULL; |
1660 | } | |
a5c31904 | 1661 | ds -= len; |
b411b363 PR |
1662 | } |
1663 | ||
11f8b2b6 | 1664 | if (digest_size) { |
69a22773 | 1665 | drbd_csum_ee(peer_device->connection->peer_integrity_tfm, peer_req, dig_vv); |
11f8b2b6 | 1666 | if (memcmp(dig_in, dig_vv, digest_size)) { |
d0180171 | 1667 | drbd_err(device, "Digest integrity check FAILED: %llus +%u\n", |
470be44a | 1668 | (unsigned long long)sector, data_size); |
b30ab791 | 1669 | drbd_free_peer_req(device, peer_req); |
b411b363 PR |
1670 | return NULL; |
1671 | } | |
1672 | } | |
11f8b2b6 | 1673 | device->recv_cnt += data_size >> 9; |
db830c46 | 1674 | return peer_req; |
b411b363 PR |
1675 | } |
1676 | ||
1677 | /* drbd_drain_block() just takes a data block | |
1678 | * out of the socket input buffer, and discards it. | |
1679 | */ | |
69a22773 | 1680 | static int drbd_drain_block(struct drbd_peer_device *peer_device, int data_size) |
b411b363 PR |
1681 | { |
1682 | struct page *page; | |
a5c31904 | 1683 | int err = 0; |
b411b363 PR |
1684 | void *data; |
1685 | ||
c3470cde | 1686 | if (!data_size) |
fc5be839 | 1687 | return 0; |
c3470cde | 1688 | |
69a22773 | 1689 | page = drbd_alloc_pages(peer_device, 1, 1); |
b411b363 PR |
1690 | |
1691 | data = kmap(page); | |
1692 | while (data_size) { | |
fc5be839 AG |
1693 | unsigned int len = min_t(int, data_size, PAGE_SIZE); |
1694 | ||
69a22773 | 1695 | err = drbd_recv_all_warn(peer_device->connection, data, len); |
a5c31904 | 1696 | if (err) |
b411b363 | 1697 | break; |
a5c31904 | 1698 | data_size -= len; |
b411b363 PR |
1699 | } |
1700 | kunmap(page); | |
69a22773 | 1701 | drbd_free_pages(peer_device->device, page, 0); |
fc5be839 | 1702 | return err; |
b411b363 PR |
1703 | } |
1704 | ||
69a22773 | 1705 | static int recv_dless_read(struct drbd_peer_device *peer_device, struct drbd_request *req, |
b411b363 PR |
1706 | sector_t sector, int data_size) |
1707 | { | |
7988613b KO |
1708 | struct bio_vec bvec; |
1709 | struct bvec_iter iter; | |
b411b363 | 1710 | struct bio *bio; |
11f8b2b6 | 1711 | int digest_size, err, expect; |
69a22773 AG |
1712 | void *dig_in = peer_device->connection->int_dig_in; |
1713 | void *dig_vv = peer_device->connection->int_dig_vv; | |
b411b363 | 1714 | |
11f8b2b6 | 1715 | digest_size = 0; |
69a22773 | 1716 | if (peer_device->connection->peer_integrity_tfm) { |
11f8b2b6 AG |
1717 | digest_size = crypto_hash_digestsize(peer_device->connection->peer_integrity_tfm); |
1718 | err = drbd_recv_all_warn(peer_device->connection, dig_in, digest_size); | |
a5c31904 AG |
1719 | if (err) |
1720 | return err; | |
11f8b2b6 | 1721 | data_size -= digest_size; |
b411b363 PR |
1722 | } |
1723 | ||
b411b363 PR |
1724 | /* optimistically update recv_cnt. if receiving fails below, |
1725 | * we disconnect anyways, and counters will be reset. */ | |
69a22773 | 1726 | peer_device->device->recv_cnt += data_size>>9; |
b411b363 PR |
1727 | |
1728 | bio = req->master_bio; | |
69a22773 | 1729 | D_ASSERT(peer_device->device, sector == bio->bi_iter.bi_sector); |
b411b363 | 1730 | |
7988613b KO |
1731 | bio_for_each_segment(bvec, bio, iter) { |
1732 | void *mapped = kmap(bvec.bv_page) + bvec.bv_offset; | |
1733 | expect = min_t(int, data_size, bvec.bv_len); | |
69a22773 | 1734 | err = drbd_recv_all_warn(peer_device->connection, mapped, expect); |
7988613b | 1735 | kunmap(bvec.bv_page); |
a5c31904 AG |
1736 | if (err) |
1737 | return err; | |
1738 | data_size -= expect; | |
b411b363 PR |
1739 | } |
1740 | ||
11f8b2b6 | 1741 | if (digest_size) { |
69a22773 | 1742 | drbd_csum_bio(peer_device->connection->peer_integrity_tfm, bio, dig_vv); |
11f8b2b6 | 1743 | if (memcmp(dig_in, dig_vv, digest_size)) { |
69a22773 | 1744 | drbd_err(peer_device, "Digest integrity check FAILED. Broken NICs?\n"); |
28284cef | 1745 | return -EINVAL; |
b411b363 PR |
1746 | } |
1747 | } | |
1748 | ||
69a22773 | 1749 | D_ASSERT(peer_device->device, data_size == 0); |
28284cef | 1750 | return 0; |
b411b363 PR |
1751 | } |
1752 | ||
a990be46 AG |
1753 | /* |
1754 | * e_end_resync_block() is called in asender context via | |
1755 | * drbd_finish_peer_reqs(). | |
1756 | */ | |
99920dc5 | 1757 | static int e_end_resync_block(struct drbd_work *w, int unused) |
b411b363 | 1758 | { |
8050e6d0 | 1759 | struct drbd_peer_request *peer_req = |
a8cd15ba AG |
1760 | container_of(w, struct drbd_peer_request, w); |
1761 | struct drbd_peer_device *peer_device = peer_req->peer_device; | |
1762 | struct drbd_device *device = peer_device->device; | |
db830c46 | 1763 | sector_t sector = peer_req->i.sector; |
99920dc5 | 1764 | int err; |
b411b363 | 1765 | |
0b0ba1ef | 1766 | D_ASSERT(device, drbd_interval_empty(&peer_req->i)); |
b411b363 | 1767 | |
db830c46 | 1768 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b30ab791 | 1769 | drbd_set_in_sync(device, sector, peer_req->i.size); |
a8cd15ba | 1770 | err = drbd_send_ack(peer_device, P_RS_WRITE_ACK, peer_req); |
b411b363 PR |
1771 | } else { |
1772 | /* Record failure to sync */ | |
b30ab791 | 1773 | drbd_rs_failed_io(device, sector, peer_req->i.size); |
b411b363 | 1774 | |
a8cd15ba | 1775 | err = drbd_send_ack(peer_device, P_NEG_ACK, peer_req); |
b411b363 | 1776 | } |
b30ab791 | 1777 | dec_unacked(device); |
b411b363 | 1778 | |
99920dc5 | 1779 | return err; |
b411b363 PR |
1780 | } |
1781 | ||
69a22773 | 1782 | static int recv_resync_read(struct drbd_peer_device *peer_device, sector_t sector, |
a0fb3c47 | 1783 | struct packet_info *pi) __releases(local) |
b411b363 | 1784 | { |
69a22773 | 1785 | struct drbd_device *device = peer_device->device; |
db830c46 | 1786 | struct drbd_peer_request *peer_req; |
b411b363 | 1787 | |
a0fb3c47 | 1788 | peer_req = read_in_block(peer_device, ID_SYNCER, sector, pi); |
db830c46 | 1789 | if (!peer_req) |
45bb912b | 1790 | goto fail; |
b411b363 | 1791 | |
b30ab791 | 1792 | dec_rs_pending(device); |
b411b363 | 1793 | |
b30ab791 | 1794 | inc_unacked(device); |
b411b363 PR |
1795 | /* corresponding dec_unacked() in e_end_resync_block() |
1796 | * respective _drbd_clear_done_ee */ | |
1797 | ||
a8cd15ba | 1798 | peer_req->w.cb = e_end_resync_block; |
21ae5d7f | 1799 | peer_req->submit_jif = jiffies; |
45bb912b | 1800 | |
0500813f | 1801 | spin_lock_irq(&device->resource->req_lock); |
b9ed7080 | 1802 | list_add_tail(&peer_req->w.list, &device->sync_ee); |
0500813f | 1803 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 1804 | |
a0fb3c47 | 1805 | atomic_add(pi->size >> 9, &device->rs_sect_ev); |
b30ab791 | 1806 | if (drbd_submit_peer_request(device, peer_req, WRITE, DRBD_FAULT_RS_WR) == 0) |
e1c1b0fc | 1807 | return 0; |
b411b363 | 1808 | |
10f6d992 | 1809 | /* don't care for the reason here */ |
d0180171 | 1810 | drbd_err(device, "submit failed, triggering re-connect\n"); |
0500813f | 1811 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 1812 | list_del(&peer_req->w.list); |
0500813f | 1813 | spin_unlock_irq(&device->resource->req_lock); |
22cc37a9 | 1814 | |
b30ab791 | 1815 | drbd_free_peer_req(device, peer_req); |
45bb912b | 1816 | fail: |
b30ab791 | 1817 | put_ldev(device); |
e1c1b0fc | 1818 | return -EIO; |
b411b363 PR |
1819 | } |
1820 | ||
668eebc6 | 1821 | static struct drbd_request * |
b30ab791 | 1822 | find_request(struct drbd_device *device, struct rb_root *root, u64 id, |
bc9c5c41 | 1823 | sector_t sector, bool missing_ok, const char *func) |
51624585 | 1824 | { |
51624585 AG |
1825 | struct drbd_request *req; |
1826 | ||
bc9c5c41 AG |
1827 | /* Request object according to our peer */ |
1828 | req = (struct drbd_request *)(unsigned long)id; | |
5e472264 | 1829 | if (drbd_contains_interval(root, sector, &req->i) && req->i.local) |
668eebc6 | 1830 | return req; |
c3afd8f5 | 1831 | if (!missing_ok) { |
d0180171 | 1832 | drbd_err(device, "%s: failed to find request 0x%lx, sector %llus\n", func, |
c3afd8f5 AG |
1833 | (unsigned long)id, (unsigned long long)sector); |
1834 | } | |
51624585 | 1835 | return NULL; |
b411b363 PR |
1836 | } |
1837 | ||
bde89a9e | 1838 | static int receive_DataReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 1839 | { |
9f4fe9ad | 1840 | struct drbd_peer_device *peer_device; |
b30ab791 | 1841 | struct drbd_device *device; |
b411b363 PR |
1842 | struct drbd_request *req; |
1843 | sector_t sector; | |
82bc0194 | 1844 | int err; |
e658983a | 1845 | struct p_data *p = pi->data; |
4a76b161 | 1846 | |
9f4fe9ad AG |
1847 | peer_device = conn_peer_device(connection, pi->vnr); |
1848 | if (!peer_device) | |
4a76b161 | 1849 | return -EIO; |
9f4fe9ad | 1850 | device = peer_device->device; |
b411b363 PR |
1851 | |
1852 | sector = be64_to_cpu(p->sector); | |
1853 | ||
0500813f | 1854 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 1855 | req = find_request(device, &device->read_requests, p->block_id, sector, false, __func__); |
0500813f | 1856 | spin_unlock_irq(&device->resource->req_lock); |
c3afd8f5 | 1857 | if (unlikely(!req)) |
82bc0194 | 1858 | return -EIO; |
b411b363 | 1859 | |
24c4830c | 1860 | /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid |
b411b363 PR |
1861 | * special casing it there for the various failure cases. |
1862 | * still no race with drbd_fail_pending_reads */ | |
69a22773 | 1863 | err = recv_dless_read(peer_device, req, sector, pi->size); |
82bc0194 | 1864 | if (!err) |
8554df1c | 1865 | req_mod(req, DATA_RECEIVED); |
b411b363 PR |
1866 | /* else: nothing. handled from drbd_disconnect... |
1867 | * I don't think we may complete this just yet | |
1868 | * in case we are "on-disconnect: freeze" */ | |
1869 | ||
82bc0194 | 1870 | return err; |
b411b363 PR |
1871 | } |
1872 | ||
bde89a9e | 1873 | static int receive_RSDataReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 1874 | { |
9f4fe9ad | 1875 | struct drbd_peer_device *peer_device; |
b30ab791 | 1876 | struct drbd_device *device; |
b411b363 | 1877 | sector_t sector; |
82bc0194 | 1878 | int err; |
e658983a | 1879 | struct p_data *p = pi->data; |
4a76b161 | 1880 | |
9f4fe9ad AG |
1881 | peer_device = conn_peer_device(connection, pi->vnr); |
1882 | if (!peer_device) | |
4a76b161 | 1883 | return -EIO; |
9f4fe9ad | 1884 | device = peer_device->device; |
b411b363 PR |
1885 | |
1886 | sector = be64_to_cpu(p->sector); | |
0b0ba1ef | 1887 | D_ASSERT(device, p->block_id == ID_SYNCER); |
b411b363 | 1888 | |
b30ab791 | 1889 | if (get_ldev(device)) { |
b411b363 PR |
1890 | /* data is submitted to disk within recv_resync_read. |
1891 | * corresponding put_ldev done below on error, | |
fcefa62e | 1892 | * or in drbd_peer_request_endio. */ |
a0fb3c47 | 1893 | err = recv_resync_read(peer_device, sector, pi); |
b411b363 PR |
1894 | } else { |
1895 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1896 | drbd_err(device, "Can not write resync data to local disk.\n"); |
b411b363 | 1897 | |
69a22773 | 1898 | err = drbd_drain_block(peer_device, pi->size); |
b411b363 | 1899 | |
69a22773 | 1900 | drbd_send_ack_dp(peer_device, P_NEG_ACK, p, pi->size); |
b411b363 PR |
1901 | } |
1902 | ||
b30ab791 | 1903 | atomic_add(pi->size >> 9, &device->rs_sect_in); |
778f271d | 1904 | |
82bc0194 | 1905 | return err; |
b411b363 PR |
1906 | } |
1907 | ||
b30ab791 | 1908 | static void restart_conflicting_writes(struct drbd_device *device, |
7be8da07 | 1909 | sector_t sector, int size) |
b411b363 | 1910 | { |
7be8da07 AG |
1911 | struct drbd_interval *i; |
1912 | struct drbd_request *req; | |
1913 | ||
b30ab791 | 1914 | drbd_for_each_overlap(i, &device->write_requests, sector, size) { |
7be8da07 AG |
1915 | if (!i->local) |
1916 | continue; | |
1917 | req = container_of(i, struct drbd_request, i); | |
1918 | if (req->rq_state & RQ_LOCAL_PENDING || | |
1919 | !(req->rq_state & RQ_POSTPONED)) | |
1920 | continue; | |
2312f0b3 LE |
1921 | /* as it is RQ_POSTPONED, this will cause it to |
1922 | * be queued on the retry workqueue. */ | |
d4dabbe2 | 1923 | __req_mod(req, CONFLICT_RESOLVED, NULL); |
7be8da07 AG |
1924 | } |
1925 | } | |
b411b363 | 1926 | |
a990be46 AG |
1927 | /* |
1928 | * e_end_block() is called in asender context via drbd_finish_peer_reqs(). | |
b411b363 | 1929 | */ |
99920dc5 | 1930 | static int e_end_block(struct drbd_work *w, int cancel) |
b411b363 | 1931 | { |
8050e6d0 | 1932 | struct drbd_peer_request *peer_req = |
a8cd15ba AG |
1933 | container_of(w, struct drbd_peer_request, w); |
1934 | struct drbd_peer_device *peer_device = peer_req->peer_device; | |
1935 | struct drbd_device *device = peer_device->device; | |
db830c46 | 1936 | sector_t sector = peer_req->i.sector; |
99920dc5 | 1937 | int err = 0, pcmd; |
b411b363 | 1938 | |
303d1448 | 1939 | if (peer_req->flags & EE_SEND_WRITE_ACK) { |
db830c46 | 1940 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b30ab791 AG |
1941 | pcmd = (device->state.conn >= C_SYNC_SOURCE && |
1942 | device->state.conn <= C_PAUSED_SYNC_T && | |
db830c46 | 1943 | peer_req->flags & EE_MAY_SET_IN_SYNC) ? |
b411b363 | 1944 | P_RS_WRITE_ACK : P_WRITE_ACK; |
a8cd15ba | 1945 | err = drbd_send_ack(peer_device, pcmd, peer_req); |
b411b363 | 1946 | if (pcmd == P_RS_WRITE_ACK) |
b30ab791 | 1947 | drbd_set_in_sync(device, sector, peer_req->i.size); |
b411b363 | 1948 | } else { |
a8cd15ba | 1949 | err = drbd_send_ack(peer_device, P_NEG_ACK, peer_req); |
b411b363 PR |
1950 | /* we expect it to be marked out of sync anyways... |
1951 | * maybe assert this? */ | |
1952 | } | |
b30ab791 | 1953 | dec_unacked(device); |
b411b363 | 1954 | } |
08d0dabf | 1955 | |
b411b363 PR |
1956 | /* we delete from the conflict detection hash _after_ we sent out the |
1957 | * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ | |
302bdeae | 1958 | if (peer_req->flags & EE_IN_INTERVAL_TREE) { |
0500813f | 1959 | spin_lock_irq(&device->resource->req_lock); |
0b0ba1ef | 1960 | D_ASSERT(device, !drbd_interval_empty(&peer_req->i)); |
b30ab791 | 1961 | drbd_remove_epoch_entry_interval(device, peer_req); |
7be8da07 | 1962 | if (peer_req->flags & EE_RESTART_REQUESTS) |
b30ab791 | 1963 | restart_conflicting_writes(device, sector, peer_req->i.size); |
0500813f | 1964 | spin_unlock_irq(&device->resource->req_lock); |
bb3bfe96 | 1965 | } else |
0b0ba1ef | 1966 | D_ASSERT(device, drbd_interval_empty(&peer_req->i)); |
b411b363 | 1967 | |
a6b32bc3 | 1968 | drbd_may_finish_epoch(first_peer_device(device)->connection, peer_req->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); |
b411b363 | 1969 | |
99920dc5 | 1970 | return err; |
b411b363 PR |
1971 | } |
1972 | ||
a8cd15ba | 1973 | static int e_send_ack(struct drbd_work *w, enum drbd_packet ack) |
b411b363 | 1974 | { |
8050e6d0 | 1975 | struct drbd_peer_request *peer_req = |
a8cd15ba AG |
1976 | container_of(w, struct drbd_peer_request, w); |
1977 | struct drbd_peer_device *peer_device = peer_req->peer_device; | |
99920dc5 | 1978 | int err; |
b411b363 | 1979 | |
a8cd15ba AG |
1980 | err = drbd_send_ack(peer_device, ack, peer_req); |
1981 | dec_unacked(peer_device->device); | |
b411b363 | 1982 | |
99920dc5 | 1983 | return err; |
b411b363 PR |
1984 | } |
1985 | ||
d4dabbe2 | 1986 | static int e_send_superseded(struct drbd_work *w, int unused) |
7be8da07 | 1987 | { |
a8cd15ba | 1988 | return e_send_ack(w, P_SUPERSEDED); |
7be8da07 AG |
1989 | } |
1990 | ||
99920dc5 | 1991 | static int e_send_retry_write(struct drbd_work *w, int unused) |
7be8da07 | 1992 | { |
a8cd15ba AG |
1993 | struct drbd_peer_request *peer_req = |
1994 | container_of(w, struct drbd_peer_request, w); | |
1995 | struct drbd_connection *connection = peer_req->peer_device->connection; | |
7be8da07 | 1996 | |
a8cd15ba | 1997 | return e_send_ack(w, connection->agreed_pro_version >= 100 ? |
d4dabbe2 | 1998 | P_RETRY_WRITE : P_SUPERSEDED); |
7be8da07 | 1999 | } |
b411b363 | 2000 | |
3e394da1 AG |
2001 | static bool seq_greater(u32 a, u32 b) |
2002 | { | |
2003 | /* | |
2004 | * We assume 32-bit wrap-around here. | |
2005 | * For 24-bit wrap-around, we would have to shift: | |
2006 | * a <<= 8; b <<= 8; | |
2007 | */ | |
2008 | return (s32)a - (s32)b > 0; | |
2009 | } | |
b411b363 | 2010 | |
3e394da1 AG |
2011 | static u32 seq_max(u32 a, u32 b) |
2012 | { | |
2013 | return seq_greater(a, b) ? a : b; | |
b411b363 PR |
2014 | } |
2015 | ||
69a22773 | 2016 | static void update_peer_seq(struct drbd_peer_device *peer_device, unsigned int peer_seq) |
3e394da1 | 2017 | { |
69a22773 | 2018 | struct drbd_device *device = peer_device->device; |
3c13b680 | 2019 | unsigned int newest_peer_seq; |
3e394da1 | 2020 | |
69a22773 | 2021 | if (test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags)) { |
b30ab791 AG |
2022 | spin_lock(&device->peer_seq_lock); |
2023 | newest_peer_seq = seq_max(device->peer_seq, peer_seq); | |
2024 | device->peer_seq = newest_peer_seq; | |
2025 | spin_unlock(&device->peer_seq_lock); | |
2026 | /* wake up only if we actually changed device->peer_seq */ | |
3c13b680 | 2027 | if (peer_seq == newest_peer_seq) |
b30ab791 | 2028 | wake_up(&device->seq_wait); |
7be8da07 | 2029 | } |
b411b363 PR |
2030 | } |
2031 | ||
d93f6302 | 2032 | static inline int overlaps(sector_t s1, int l1, sector_t s2, int l2) |
b6a370ba | 2033 | { |
d93f6302 LE |
2034 | return !((s1 + (l1>>9) <= s2) || (s1 >= s2 + (l2>>9))); |
2035 | } | |
b6a370ba | 2036 | |
d93f6302 | 2037 | /* maybe change sync_ee into interval trees as well? */ |
b30ab791 | 2038 | static bool overlapping_resync_write(struct drbd_device *device, struct drbd_peer_request *peer_req) |
d93f6302 LE |
2039 | { |
2040 | struct drbd_peer_request *rs_req; | |
b6a370ba PR |
2041 | bool rv = 0; |
2042 | ||
0500813f | 2043 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 2044 | list_for_each_entry(rs_req, &device->sync_ee, w.list) { |
d93f6302 LE |
2045 | if (overlaps(peer_req->i.sector, peer_req->i.size, |
2046 | rs_req->i.sector, rs_req->i.size)) { | |
b6a370ba PR |
2047 | rv = 1; |
2048 | break; | |
2049 | } | |
2050 | } | |
0500813f | 2051 | spin_unlock_irq(&device->resource->req_lock); |
b6a370ba PR |
2052 | |
2053 | return rv; | |
2054 | } | |
2055 | ||
b411b363 PR |
2056 | /* Called from receive_Data. |
2057 | * Synchronize packets on sock with packets on msock. | |
2058 | * | |
2059 | * This is here so even when a P_DATA packet traveling via sock overtook an Ack | |
2060 | * packet traveling on msock, they are still processed in the order they have | |
2061 | * been sent. | |
2062 | * | |
2063 | * Note: we don't care for Ack packets overtaking P_DATA packets. | |
2064 | * | |
b30ab791 | 2065 | * In case packet_seq is larger than device->peer_seq number, there are |
b411b363 | 2066 | * outstanding packets on the msock. We wait for them to arrive. |
b30ab791 | 2067 | * In case we are the logically next packet, we update device->peer_seq |
b411b363 PR |
2068 | * ourselves. Correctly handles 32bit wrap around. |
2069 | * | |
2070 | * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, | |
2071 | * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds | |
2072 | * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have | |
2073 | * 1<<9 == 512 seconds aka ages for the 32bit wrap around... | |
2074 | * | |
2075 | * returns 0 if we may process the packet, | |
2076 | * -ERESTARTSYS if we were interrupted (by disconnect signal). */ | |
69a22773 | 2077 | static int wait_for_and_update_peer_seq(struct drbd_peer_device *peer_device, const u32 peer_seq) |
b411b363 | 2078 | { |
69a22773 | 2079 | struct drbd_device *device = peer_device->device; |
b411b363 | 2080 | DEFINE_WAIT(wait); |
b411b363 | 2081 | long timeout; |
b874d231 | 2082 | int ret = 0, tp; |
7be8da07 | 2083 | |
69a22773 | 2084 | if (!test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags)) |
7be8da07 AG |
2085 | return 0; |
2086 | ||
b30ab791 | 2087 | spin_lock(&device->peer_seq_lock); |
b411b363 | 2088 | for (;;) { |
b30ab791 AG |
2089 | if (!seq_greater(peer_seq - 1, device->peer_seq)) { |
2090 | device->peer_seq = seq_max(device->peer_seq, peer_seq); | |
b411b363 | 2091 | break; |
7be8da07 | 2092 | } |
b874d231 | 2093 | |
b411b363 PR |
2094 | if (signal_pending(current)) { |
2095 | ret = -ERESTARTSYS; | |
2096 | break; | |
2097 | } | |
b874d231 PR |
2098 | |
2099 | rcu_read_lock(); | |
a6b32bc3 | 2100 | tp = rcu_dereference(first_peer_device(device)->connection->net_conf)->two_primaries; |
b874d231 PR |
2101 | rcu_read_unlock(); |
2102 | ||
2103 | if (!tp) | |
2104 | break; | |
2105 | ||
2106 | /* Only need to wait if two_primaries is enabled */ | |
b30ab791 AG |
2107 | prepare_to_wait(&device->seq_wait, &wait, TASK_INTERRUPTIBLE); |
2108 | spin_unlock(&device->peer_seq_lock); | |
44ed167d | 2109 | rcu_read_lock(); |
69a22773 | 2110 | timeout = rcu_dereference(peer_device->connection->net_conf)->ping_timeo*HZ/10; |
44ed167d | 2111 | rcu_read_unlock(); |
71b1c1eb | 2112 | timeout = schedule_timeout(timeout); |
b30ab791 | 2113 | spin_lock(&device->peer_seq_lock); |
7be8da07 | 2114 | if (!timeout) { |
b411b363 | 2115 | ret = -ETIMEDOUT; |
d0180171 | 2116 | drbd_err(device, "Timed out waiting for missing ack packets; disconnecting\n"); |
b411b363 PR |
2117 | break; |
2118 | } | |
2119 | } | |
b30ab791 AG |
2120 | spin_unlock(&device->peer_seq_lock); |
2121 | finish_wait(&device->seq_wait, &wait); | |
b411b363 PR |
2122 | return ret; |
2123 | } | |
2124 | ||
688593c5 LE |
2125 | /* see also bio_flags_to_wire() |
2126 | * DRBD_REQ_*, because we need to semantically map the flags to data packet | |
2127 | * flags and back. We may replicate to other kernel versions. */ | |
81f0ffd2 | 2128 | static unsigned long wire_flags_to_bio(u32 dpf) |
76d2e7ec | 2129 | { |
688593c5 LE |
2130 | return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | |
2131 | (dpf & DP_FUA ? REQ_FUA : 0) | | |
2132 | (dpf & DP_FLUSH ? REQ_FLUSH : 0) | | |
2133 | (dpf & DP_DISCARD ? REQ_DISCARD : 0); | |
76d2e7ec PR |
2134 | } |
2135 | ||
b30ab791 | 2136 | static void fail_postponed_requests(struct drbd_device *device, sector_t sector, |
7be8da07 AG |
2137 | unsigned int size) |
2138 | { | |
2139 | struct drbd_interval *i; | |
2140 | ||
2141 | repeat: | |
b30ab791 | 2142 | drbd_for_each_overlap(i, &device->write_requests, sector, size) { |
7be8da07 AG |
2143 | struct drbd_request *req; |
2144 | struct bio_and_error m; | |
2145 | ||
2146 | if (!i->local) | |
2147 | continue; | |
2148 | req = container_of(i, struct drbd_request, i); | |
2149 | if (!(req->rq_state & RQ_POSTPONED)) | |
2150 | continue; | |
2151 | req->rq_state &= ~RQ_POSTPONED; | |
2152 | __req_mod(req, NEG_ACKED, &m); | |
0500813f | 2153 | spin_unlock_irq(&device->resource->req_lock); |
7be8da07 | 2154 | if (m.bio) |
b30ab791 | 2155 | complete_master_bio(device, &m); |
0500813f | 2156 | spin_lock_irq(&device->resource->req_lock); |
7be8da07 AG |
2157 | goto repeat; |
2158 | } | |
2159 | } | |
2160 | ||
b30ab791 | 2161 | static int handle_write_conflicts(struct drbd_device *device, |
7be8da07 AG |
2162 | struct drbd_peer_request *peer_req) |
2163 | { | |
e33b32de | 2164 | struct drbd_connection *connection = peer_req->peer_device->connection; |
bde89a9e | 2165 | bool resolve_conflicts = test_bit(RESOLVE_CONFLICTS, &connection->flags); |
7be8da07 AG |
2166 | sector_t sector = peer_req->i.sector; |
2167 | const unsigned int size = peer_req->i.size; | |
2168 | struct drbd_interval *i; | |
2169 | bool equal; | |
2170 | int err; | |
2171 | ||
2172 | /* | |
2173 | * Inserting the peer request into the write_requests tree will prevent | |
2174 | * new conflicting local requests from being added. | |
2175 | */ | |
b30ab791 | 2176 | drbd_insert_interval(&device->write_requests, &peer_req->i); |
7be8da07 AG |
2177 | |
2178 | repeat: | |
b30ab791 | 2179 | drbd_for_each_overlap(i, &device->write_requests, sector, size) { |
7be8da07 AG |
2180 | if (i == &peer_req->i) |
2181 | continue; | |
08d0dabf LE |
2182 | if (i->completed) |
2183 | continue; | |
7be8da07 AG |
2184 | |
2185 | if (!i->local) { | |
2186 | /* | |
2187 | * Our peer has sent a conflicting remote request; this | |
2188 | * should not happen in a two-node setup. Wait for the | |
2189 | * earlier peer request to complete. | |
2190 | */ | |
b30ab791 | 2191 | err = drbd_wait_misc(device, i); |
7be8da07 AG |
2192 | if (err) |
2193 | goto out; | |
2194 | goto repeat; | |
2195 | } | |
2196 | ||
2197 | equal = i->sector == sector && i->size == size; | |
2198 | if (resolve_conflicts) { | |
2199 | /* | |
2200 | * If the peer request is fully contained within the | |
d4dabbe2 LE |
2201 | * overlapping request, it can be considered overwritten |
2202 | * and thus superseded; otherwise, it will be retried | |
2203 | * once all overlapping requests have completed. | |
7be8da07 | 2204 | */ |
d4dabbe2 | 2205 | bool superseded = i->sector <= sector && i->sector + |
7be8da07 AG |
2206 | (i->size >> 9) >= sector + (size >> 9); |
2207 | ||
2208 | if (!equal) | |
d0180171 | 2209 | drbd_alert(device, "Concurrent writes detected: " |
7be8da07 AG |
2210 | "local=%llus +%u, remote=%llus +%u, " |
2211 | "assuming %s came first\n", | |
2212 | (unsigned long long)i->sector, i->size, | |
2213 | (unsigned long long)sector, size, | |
d4dabbe2 | 2214 | superseded ? "local" : "remote"); |
7be8da07 | 2215 | |
a8cd15ba | 2216 | peer_req->w.cb = superseded ? e_send_superseded : |
7be8da07 | 2217 | e_send_retry_write; |
a8cd15ba | 2218 | list_add_tail(&peer_req->w.list, &device->done_ee); |
e33b32de | 2219 | wake_asender(connection); |
7be8da07 AG |
2220 | |
2221 | err = -ENOENT; | |
2222 | goto out; | |
2223 | } else { | |
2224 | struct drbd_request *req = | |
2225 | container_of(i, struct drbd_request, i); | |
2226 | ||
2227 | if (!equal) | |
d0180171 | 2228 | drbd_alert(device, "Concurrent writes detected: " |
7be8da07 AG |
2229 | "local=%llus +%u, remote=%llus +%u\n", |
2230 | (unsigned long long)i->sector, i->size, | |
2231 | (unsigned long long)sector, size); | |
2232 | ||
2233 | if (req->rq_state & RQ_LOCAL_PENDING || | |
2234 | !(req->rq_state & RQ_POSTPONED)) { | |
2235 | /* | |
2236 | * Wait for the node with the discard flag to | |
d4dabbe2 LE |
2237 | * decide if this request has been superseded |
2238 | * or needs to be retried. | |
2239 | * Requests that have been superseded will | |
7be8da07 AG |
2240 | * disappear from the write_requests tree. |
2241 | * | |
2242 | * In addition, wait for the conflicting | |
2243 | * request to finish locally before submitting | |
2244 | * the conflicting peer request. | |
2245 | */ | |
b30ab791 | 2246 | err = drbd_wait_misc(device, &req->i); |
7be8da07 | 2247 | if (err) { |
e33b32de | 2248 | _conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD); |
b30ab791 | 2249 | fail_postponed_requests(device, sector, size); |
7be8da07 AG |
2250 | goto out; |
2251 | } | |
2252 | goto repeat; | |
2253 | } | |
2254 | /* | |
2255 | * Remember to restart the conflicting requests after | |
2256 | * the new peer request has completed. | |
2257 | */ | |
2258 | peer_req->flags |= EE_RESTART_REQUESTS; | |
2259 | } | |
2260 | } | |
2261 | err = 0; | |
2262 | ||
2263 | out: | |
2264 | if (err) | |
b30ab791 | 2265 | drbd_remove_epoch_entry_interval(device, peer_req); |
7be8da07 AG |
2266 | return err; |
2267 | } | |
2268 | ||
b411b363 | 2269 | /* mirrored write */ |
bde89a9e | 2270 | static int receive_Data(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 2271 | { |
9f4fe9ad | 2272 | struct drbd_peer_device *peer_device; |
b30ab791 | 2273 | struct drbd_device *device; |
21ae5d7f | 2274 | struct net_conf *nc; |
b411b363 | 2275 | sector_t sector; |
db830c46 | 2276 | struct drbd_peer_request *peer_req; |
e658983a | 2277 | struct p_data *p = pi->data; |
7be8da07 | 2278 | u32 peer_seq = be32_to_cpu(p->seq_num); |
b411b363 PR |
2279 | int rw = WRITE; |
2280 | u32 dp_flags; | |
302bdeae | 2281 | int err, tp; |
b411b363 | 2282 | |
9f4fe9ad AG |
2283 | peer_device = conn_peer_device(connection, pi->vnr); |
2284 | if (!peer_device) | |
4a76b161 | 2285 | return -EIO; |
9f4fe9ad | 2286 | device = peer_device->device; |
b411b363 | 2287 | |
b30ab791 | 2288 | if (!get_ldev(device)) { |
82bc0194 AG |
2289 | int err2; |
2290 | ||
69a22773 AG |
2291 | err = wait_for_and_update_peer_seq(peer_device, peer_seq); |
2292 | drbd_send_ack_dp(peer_device, P_NEG_ACK, p, pi->size); | |
bde89a9e | 2293 | atomic_inc(&connection->current_epoch->epoch_size); |
69a22773 | 2294 | err2 = drbd_drain_block(peer_device, pi->size); |
82bc0194 AG |
2295 | if (!err) |
2296 | err = err2; | |
2297 | return err; | |
b411b363 PR |
2298 | } |
2299 | ||
fcefa62e AG |
2300 | /* |
2301 | * Corresponding put_ldev done either below (on various errors), or in | |
2302 | * drbd_peer_request_endio, if we successfully submit the data at the | |
2303 | * end of this function. | |
2304 | */ | |
b411b363 PR |
2305 | |
2306 | sector = be64_to_cpu(p->sector); | |
a0fb3c47 | 2307 | peer_req = read_in_block(peer_device, p->block_id, sector, pi); |
db830c46 | 2308 | if (!peer_req) { |
b30ab791 | 2309 | put_ldev(device); |
82bc0194 | 2310 | return -EIO; |
b411b363 PR |
2311 | } |
2312 | ||
a8cd15ba | 2313 | peer_req->w.cb = e_end_block; |
21ae5d7f LE |
2314 | peer_req->submit_jif = jiffies; |
2315 | peer_req->flags |= EE_APPLICATION; | |
b411b363 | 2316 | |
688593c5 | 2317 | dp_flags = be32_to_cpu(p->dp_flags); |
81f0ffd2 | 2318 | rw |= wire_flags_to_bio(dp_flags); |
a0fb3c47 LE |
2319 | if (pi->cmd == P_TRIM) { |
2320 | struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev); | |
2321 | peer_req->flags |= EE_IS_TRIM; | |
2322 | if (!blk_queue_discard(q)) | |
2323 | peer_req->flags |= EE_IS_TRIM_USE_ZEROOUT; | |
2324 | D_ASSERT(peer_device, peer_req->i.size > 0); | |
2325 | D_ASSERT(peer_device, rw & REQ_DISCARD); | |
2326 | D_ASSERT(peer_device, peer_req->pages == NULL); | |
2327 | } else if (peer_req->pages == NULL) { | |
0b0ba1ef AG |
2328 | D_ASSERT(device, peer_req->i.size == 0); |
2329 | D_ASSERT(device, dp_flags & DP_FLUSH); | |
a73ff323 | 2330 | } |
688593c5 LE |
2331 | |
2332 | if (dp_flags & DP_MAY_SET_IN_SYNC) | |
db830c46 | 2333 | peer_req->flags |= EE_MAY_SET_IN_SYNC; |
688593c5 | 2334 | |
bde89a9e AG |
2335 | spin_lock(&connection->epoch_lock); |
2336 | peer_req->epoch = connection->current_epoch; | |
db830c46 AG |
2337 | atomic_inc(&peer_req->epoch->epoch_size); |
2338 | atomic_inc(&peer_req->epoch->active); | |
bde89a9e | 2339 | spin_unlock(&connection->epoch_lock); |
b411b363 | 2340 | |
302bdeae | 2341 | rcu_read_lock(); |
21ae5d7f LE |
2342 | nc = rcu_dereference(peer_device->connection->net_conf); |
2343 | tp = nc->two_primaries; | |
2344 | if (peer_device->connection->agreed_pro_version < 100) { | |
2345 | switch (nc->wire_protocol) { | |
2346 | case DRBD_PROT_C: | |
2347 | dp_flags |= DP_SEND_WRITE_ACK; | |
2348 | break; | |
2349 | case DRBD_PROT_B: | |
2350 | dp_flags |= DP_SEND_RECEIVE_ACK; | |
2351 | break; | |
2352 | } | |
2353 | } | |
302bdeae | 2354 | rcu_read_unlock(); |
21ae5d7f LE |
2355 | |
2356 | if (dp_flags & DP_SEND_WRITE_ACK) { | |
2357 | peer_req->flags |= EE_SEND_WRITE_ACK; | |
2358 | inc_unacked(device); | |
2359 | /* corresponding dec_unacked() in e_end_block() | |
2360 | * respective _drbd_clear_done_ee */ | |
2361 | } | |
2362 | ||
2363 | if (dp_flags & DP_SEND_RECEIVE_ACK) { | |
2364 | /* I really don't like it that the receiver thread | |
2365 | * sends on the msock, but anyways */ | |
2366 | drbd_send_ack(first_peer_device(device), P_RECV_ACK, peer_req); | |
2367 | } | |
2368 | ||
302bdeae | 2369 | if (tp) { |
21ae5d7f LE |
2370 | /* two primaries implies protocol C */ |
2371 | D_ASSERT(device, dp_flags & DP_SEND_WRITE_ACK); | |
302bdeae | 2372 | peer_req->flags |= EE_IN_INTERVAL_TREE; |
69a22773 | 2373 | err = wait_for_and_update_peer_seq(peer_device, peer_seq); |
7be8da07 | 2374 | if (err) |
b411b363 | 2375 | goto out_interrupted; |
0500813f | 2376 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 2377 | err = handle_write_conflicts(device, peer_req); |
7be8da07 | 2378 | if (err) { |
0500813f | 2379 | spin_unlock_irq(&device->resource->req_lock); |
7be8da07 | 2380 | if (err == -ENOENT) { |
b30ab791 | 2381 | put_ldev(device); |
82bc0194 | 2382 | return 0; |
b411b363 | 2383 | } |
7be8da07 | 2384 | goto out_interrupted; |
b411b363 | 2385 | } |
b874d231 | 2386 | } else { |
69a22773 | 2387 | update_peer_seq(peer_device, peer_seq); |
0500813f | 2388 | spin_lock_irq(&device->resource->req_lock); |
b874d231 | 2389 | } |
a0fb3c47 LE |
2390 | /* if we use the zeroout fallback code, we process synchronously |
2391 | * and we wait for all pending requests, respectively wait for | |
2392 | * active_ee to become empty in drbd_submit_peer_request(); | |
2393 | * better not add ourselves here. */ | |
2394 | if ((peer_req->flags & EE_IS_TRIM_USE_ZEROOUT) == 0) | |
b9ed7080 | 2395 | list_add_tail(&peer_req->w.list, &device->active_ee); |
0500813f | 2396 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 2397 | |
b30ab791 AG |
2398 | if (device->state.conn == C_SYNC_TARGET) |
2399 | wait_event(device->ee_wait, !overlapping_resync_write(device, peer_req)); | |
b411b363 | 2400 | |
b30ab791 | 2401 | if (device->state.pdsk < D_INCONSISTENT) { |
b411b363 | 2402 | /* In case we have the only disk of the cluster, */ |
b30ab791 | 2403 | drbd_set_out_of_sync(device, peer_req->i.sector, peer_req->i.size); |
db830c46 | 2404 | peer_req->flags &= ~EE_MAY_SET_IN_SYNC; |
4dd726f0 | 2405 | drbd_al_begin_io(device, &peer_req->i); |
21ae5d7f | 2406 | peer_req->flags |= EE_CALL_AL_COMPLETE_IO; |
b411b363 PR |
2407 | } |
2408 | ||
b30ab791 | 2409 | err = drbd_submit_peer_request(device, peer_req, rw, DRBD_FAULT_DT_WR); |
82bc0194 AG |
2410 | if (!err) |
2411 | return 0; | |
b411b363 | 2412 | |
10f6d992 | 2413 | /* don't care for the reason here */ |
d0180171 | 2414 | drbd_err(device, "submit failed, triggering re-connect\n"); |
0500813f | 2415 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 2416 | list_del(&peer_req->w.list); |
b30ab791 | 2417 | drbd_remove_epoch_entry_interval(device, peer_req); |
0500813f | 2418 | spin_unlock_irq(&device->resource->req_lock); |
21ae5d7f LE |
2419 | if (peer_req->flags & EE_CALL_AL_COMPLETE_IO) { |
2420 | peer_req->flags &= ~EE_CALL_AL_COMPLETE_IO; | |
b30ab791 | 2421 | drbd_al_complete_io(device, &peer_req->i); |
21ae5d7f | 2422 | } |
22cc37a9 | 2423 | |
b411b363 | 2424 | out_interrupted: |
bde89a9e | 2425 | drbd_may_finish_epoch(connection, peer_req->epoch, EV_PUT + EV_CLEANUP); |
b30ab791 AG |
2426 | put_ldev(device); |
2427 | drbd_free_peer_req(device, peer_req); | |
82bc0194 | 2428 | return err; |
b411b363 PR |
2429 | } |
2430 | ||
0f0601f4 LE |
2431 | /* We may throttle resync, if the lower device seems to be busy, |
2432 | * and current sync rate is above c_min_rate. | |
2433 | * | |
2434 | * To decide whether or not the lower device is busy, we use a scheme similar | |
2435 | * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" | |
2436 | * (more than 64 sectors) of activity we cannot account for with our own resync | |
2437 | * activity, it obviously is "busy". | |
2438 | * | |
2439 | * The current sync rate used here uses only the most recent two step marks, | |
2440 | * to have a short time average so we can react faster. | |
2441 | */ | |
ad3fee79 LE |
2442 | bool drbd_rs_should_slow_down(struct drbd_device *device, sector_t sector, |
2443 | bool throttle_if_app_is_waiting) | |
0f0601f4 | 2444 | { |
e3555d85 | 2445 | struct lc_element *tmp; |
ad3fee79 | 2446 | bool throttle = drbd_rs_c_min_rate_throttle(device); |
daeda1cc | 2447 | |
ad3fee79 LE |
2448 | if (!throttle || throttle_if_app_is_waiting) |
2449 | return throttle; | |
0f0601f4 | 2450 | |
b30ab791 AG |
2451 | spin_lock_irq(&device->al_lock); |
2452 | tmp = lc_find(device->resync, BM_SECT_TO_EXT(sector)); | |
e3555d85 PR |
2453 | if (tmp) { |
2454 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
e8299874 LE |
2455 | if (test_bit(BME_PRIORITY, &bm_ext->flags)) |
2456 | throttle = false; | |
ad3fee79 LE |
2457 | /* Do not slow down if app IO is already waiting for this extent, |
2458 | * and our progress is necessary for application IO to complete. */ | |
e3555d85 | 2459 | } |
b30ab791 | 2460 | spin_unlock_irq(&device->al_lock); |
e3555d85 | 2461 | |
e8299874 LE |
2462 | return throttle; |
2463 | } | |
2464 | ||
2465 | bool drbd_rs_c_min_rate_throttle(struct drbd_device *device) | |
2466 | { | |
2467 | struct gendisk *disk = device->ldev->backing_bdev->bd_contains->bd_disk; | |
2468 | unsigned long db, dt, dbdt; | |
2469 | unsigned int c_min_rate; | |
2470 | int curr_events; | |
2471 | ||
2472 | rcu_read_lock(); | |
2473 | c_min_rate = rcu_dereference(device->ldev->disk_conf)->c_min_rate; | |
2474 | rcu_read_unlock(); | |
2475 | ||
2476 | /* feature disabled? */ | |
2477 | if (c_min_rate == 0) | |
2478 | return false; | |
2479 | ||
0f0601f4 LE |
2480 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
2481 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
b30ab791 | 2482 | atomic_read(&device->rs_sect_ev); |
ad3fee79 LE |
2483 | |
2484 | if (atomic_read(&device->ap_actlog_cnt) | |
ff8bd88b | 2485 | || curr_events - device->rs_last_events > 64) { |
0f0601f4 LE |
2486 | unsigned long rs_left; |
2487 | int i; | |
2488 | ||
b30ab791 | 2489 | device->rs_last_events = curr_events; |
0f0601f4 LE |
2490 | |
2491 | /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, | |
2492 | * approx. */ | |
b30ab791 | 2493 | i = (device->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS; |
2649f080 | 2494 | |
b30ab791 AG |
2495 | if (device->state.conn == C_VERIFY_S || device->state.conn == C_VERIFY_T) |
2496 | rs_left = device->ov_left; | |
2649f080 | 2497 | else |
b30ab791 | 2498 | rs_left = drbd_bm_total_weight(device) - device->rs_failed; |
0f0601f4 | 2499 | |
b30ab791 | 2500 | dt = ((long)jiffies - (long)device->rs_mark_time[i]) / HZ; |
0f0601f4 LE |
2501 | if (!dt) |
2502 | dt++; | |
b30ab791 | 2503 | db = device->rs_mark_left[i] - rs_left; |
0f0601f4 LE |
2504 | dbdt = Bit2KB(db/dt); |
2505 | ||
daeda1cc | 2506 | if (dbdt > c_min_rate) |
e8299874 | 2507 | return true; |
0f0601f4 | 2508 | } |
e8299874 | 2509 | return false; |
0f0601f4 LE |
2510 | } |
2511 | ||
bde89a9e | 2512 | static int receive_DataRequest(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 2513 | { |
9f4fe9ad | 2514 | struct drbd_peer_device *peer_device; |
b30ab791 | 2515 | struct drbd_device *device; |
b411b363 | 2516 | sector_t sector; |
4a76b161 | 2517 | sector_t capacity; |
db830c46 | 2518 | struct drbd_peer_request *peer_req; |
b411b363 | 2519 | struct digest_info *di = NULL; |
b18b37be | 2520 | int size, verb; |
b411b363 | 2521 | unsigned int fault_type; |
e658983a | 2522 | struct p_block_req *p = pi->data; |
4a76b161 | 2523 | |
9f4fe9ad AG |
2524 | peer_device = conn_peer_device(connection, pi->vnr); |
2525 | if (!peer_device) | |
4a76b161 | 2526 | return -EIO; |
9f4fe9ad | 2527 | device = peer_device->device; |
b30ab791 | 2528 | capacity = drbd_get_capacity(device->this_bdev); |
b411b363 PR |
2529 | |
2530 | sector = be64_to_cpu(p->sector); | |
2531 | size = be32_to_cpu(p->blksize); | |
2532 | ||
c670a398 | 2533 | if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { |
d0180171 | 2534 | drbd_err(device, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, |
b411b363 | 2535 | (unsigned long long)sector, size); |
82bc0194 | 2536 | return -EINVAL; |
b411b363 PR |
2537 | } |
2538 | if (sector + (size>>9) > capacity) { | |
d0180171 | 2539 | drbd_err(device, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, |
b411b363 | 2540 | (unsigned long long)sector, size); |
82bc0194 | 2541 | return -EINVAL; |
b411b363 PR |
2542 | } |
2543 | ||
b30ab791 | 2544 | if (!get_ldev_if_state(device, D_UP_TO_DATE)) { |
b18b37be | 2545 | verb = 1; |
e2857216 | 2546 | switch (pi->cmd) { |
b18b37be | 2547 | case P_DATA_REQUEST: |
69a22773 | 2548 | drbd_send_ack_rp(peer_device, P_NEG_DREPLY, p); |
b18b37be PR |
2549 | break; |
2550 | case P_RS_DATA_REQUEST: | |
2551 | case P_CSUM_RS_REQUEST: | |
2552 | case P_OV_REQUEST: | |
69a22773 | 2553 | drbd_send_ack_rp(peer_device, P_NEG_RS_DREPLY , p); |
b18b37be PR |
2554 | break; |
2555 | case P_OV_REPLY: | |
2556 | verb = 0; | |
b30ab791 | 2557 | dec_rs_pending(device); |
69a22773 | 2558 | drbd_send_ack_ex(peer_device, P_OV_RESULT, sector, size, ID_IN_SYNC); |
b18b37be PR |
2559 | break; |
2560 | default: | |
49ba9b1b | 2561 | BUG(); |
b18b37be PR |
2562 | } |
2563 | if (verb && __ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 2564 | drbd_err(device, "Can not satisfy peer's read request, " |
b411b363 | 2565 | "no local data.\n"); |
b18b37be | 2566 | |
a821cc4a | 2567 | /* drain possibly payload */ |
69a22773 | 2568 | return drbd_drain_block(peer_device, pi->size); |
b411b363 PR |
2569 | } |
2570 | ||
2571 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD | |
2572 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
2573 | * which in turn might block on the other node at this very place. */ | |
a0fb3c47 LE |
2574 | peer_req = drbd_alloc_peer_req(peer_device, p->block_id, sector, size, |
2575 | true /* has real payload */, GFP_NOIO); | |
db830c46 | 2576 | if (!peer_req) { |
b30ab791 | 2577 | put_ldev(device); |
82bc0194 | 2578 | return -ENOMEM; |
b411b363 PR |
2579 | } |
2580 | ||
e2857216 | 2581 | switch (pi->cmd) { |
b411b363 | 2582 | case P_DATA_REQUEST: |
a8cd15ba | 2583 | peer_req->w.cb = w_e_end_data_req; |
b411b363 | 2584 | fault_type = DRBD_FAULT_DT_RD; |
80a40e43 | 2585 | /* application IO, don't drbd_rs_begin_io */ |
21ae5d7f | 2586 | peer_req->flags |= EE_APPLICATION; |
80a40e43 LE |
2587 | goto submit; |
2588 | ||
b411b363 | 2589 | case P_RS_DATA_REQUEST: |
a8cd15ba | 2590 | peer_req->w.cb = w_e_end_rsdata_req; |
b411b363 | 2591 | fault_type = DRBD_FAULT_RS_RD; |
5f9915bb | 2592 | /* used in the sector offset progress display */ |
b30ab791 | 2593 | device->bm_resync_fo = BM_SECT_TO_BIT(sector); |
b411b363 PR |
2594 | break; |
2595 | ||
2596 | case P_OV_REPLY: | |
2597 | case P_CSUM_RS_REQUEST: | |
2598 | fault_type = DRBD_FAULT_RS_RD; | |
e2857216 | 2599 | di = kmalloc(sizeof(*di) + pi->size, GFP_NOIO); |
b411b363 PR |
2600 | if (!di) |
2601 | goto out_free_e; | |
2602 | ||
e2857216 | 2603 | di->digest_size = pi->size; |
b411b363 PR |
2604 | di->digest = (((char *)di)+sizeof(struct digest_info)); |
2605 | ||
db830c46 AG |
2606 | peer_req->digest = di; |
2607 | peer_req->flags |= EE_HAS_DIGEST; | |
c36c3ced | 2608 | |
9f4fe9ad | 2609 | if (drbd_recv_all(peer_device->connection, di->digest, pi->size)) |
b411b363 PR |
2610 | goto out_free_e; |
2611 | ||
e2857216 | 2612 | if (pi->cmd == P_CSUM_RS_REQUEST) { |
9f4fe9ad | 2613 | D_ASSERT(device, peer_device->connection->agreed_pro_version >= 89); |
a8cd15ba | 2614 | peer_req->w.cb = w_e_end_csum_rs_req; |
5f9915bb | 2615 | /* used in the sector offset progress display */ |
b30ab791 | 2616 | device->bm_resync_fo = BM_SECT_TO_BIT(sector); |
aaaba345 LE |
2617 | /* remember to report stats in drbd_resync_finished */ |
2618 | device->use_csums = true; | |
e2857216 | 2619 | } else if (pi->cmd == P_OV_REPLY) { |
2649f080 | 2620 | /* track progress, we may need to throttle */ |
b30ab791 | 2621 | atomic_add(size >> 9, &device->rs_sect_in); |
a8cd15ba | 2622 | peer_req->w.cb = w_e_end_ov_reply; |
b30ab791 | 2623 | dec_rs_pending(device); |
0f0601f4 LE |
2624 | /* drbd_rs_begin_io done when we sent this request, |
2625 | * but accounting still needs to be done. */ | |
2626 | goto submit_for_resync; | |
b411b363 PR |
2627 | } |
2628 | break; | |
2629 | ||
2630 | case P_OV_REQUEST: | |
b30ab791 | 2631 | if (device->ov_start_sector == ~(sector_t)0 && |
9f4fe9ad | 2632 | peer_device->connection->agreed_pro_version >= 90) { |
de228bba LE |
2633 | unsigned long now = jiffies; |
2634 | int i; | |
b30ab791 AG |
2635 | device->ov_start_sector = sector; |
2636 | device->ov_position = sector; | |
2637 | device->ov_left = drbd_bm_bits(device) - BM_SECT_TO_BIT(sector); | |
2638 | device->rs_total = device->ov_left; | |
de228bba | 2639 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
2640 | device->rs_mark_left[i] = device->ov_left; |
2641 | device->rs_mark_time[i] = now; | |
de228bba | 2642 | } |
d0180171 | 2643 | drbd_info(device, "Online Verify start sector: %llu\n", |
b411b363 PR |
2644 | (unsigned long long)sector); |
2645 | } | |
a8cd15ba | 2646 | peer_req->w.cb = w_e_end_ov_req; |
b411b363 | 2647 | fault_type = DRBD_FAULT_RS_RD; |
b411b363 PR |
2648 | break; |
2649 | ||
b411b363 | 2650 | default: |
49ba9b1b | 2651 | BUG(); |
b411b363 PR |
2652 | } |
2653 | ||
0f0601f4 LE |
2654 | /* Throttle, drbd_rs_begin_io and submit should become asynchronous |
2655 | * wrt the receiver, but it is not as straightforward as it may seem. | |
2656 | * Various places in the resync start and stop logic assume resync | |
2657 | * requests are processed in order, requeuing this on the worker thread | |
2658 | * introduces a bunch of new code for synchronization between threads. | |
2659 | * | |
2660 | * Unlimited throttling before drbd_rs_begin_io may stall the resync | |
2661 | * "forever", throttling after drbd_rs_begin_io will lock that extent | |
2662 | * for application writes for the same time. For now, just throttle | |
2663 | * here, where the rest of the code expects the receiver to sleep for | |
2664 | * a while, anyways. | |
2665 | */ | |
2666 | ||
2667 | /* Throttle before drbd_rs_begin_io, as that locks out application IO; | |
2668 | * this defers syncer requests for some time, before letting at least | |
2669 | * on request through. The resync controller on the receiving side | |
2670 | * will adapt to the incoming rate accordingly. | |
2671 | * | |
2672 | * We cannot throttle here if remote is Primary/SyncTarget: | |
2673 | * we would also throttle its application reads. | |
2674 | * In that case, throttling is done on the SyncTarget only. | |
2675 | */ | |
c5a2c150 LE |
2676 | |
2677 | /* Even though this may be a resync request, we do add to "read_ee"; | |
2678 | * "sync_ee" is only used for resync WRITEs. | |
2679 | * Add to list early, so debugfs can find this request | |
2680 | * even if we have to sleep below. */ | |
2681 | spin_lock_irq(&device->resource->req_lock); | |
2682 | list_add_tail(&peer_req->w.list, &device->read_ee); | |
2683 | spin_unlock_irq(&device->resource->req_lock); | |
2684 | ||
944410e9 | 2685 | update_receiver_timing_details(connection, drbd_rs_should_slow_down); |
ad3fee79 LE |
2686 | if (device->state.peer != R_PRIMARY |
2687 | && drbd_rs_should_slow_down(device, sector, false)) | |
e3555d85 | 2688 | schedule_timeout_uninterruptible(HZ/10); |
944410e9 | 2689 | update_receiver_timing_details(connection, drbd_rs_begin_io); |
b30ab791 | 2690 | if (drbd_rs_begin_io(device, sector)) |
80a40e43 | 2691 | goto out_free_e; |
b411b363 | 2692 | |
0f0601f4 | 2693 | submit_for_resync: |
b30ab791 | 2694 | atomic_add(size >> 9, &device->rs_sect_ev); |
0f0601f4 | 2695 | |
80a40e43 | 2696 | submit: |
944410e9 | 2697 | update_receiver_timing_details(connection, drbd_submit_peer_request); |
b30ab791 | 2698 | inc_unacked(device); |
b30ab791 | 2699 | if (drbd_submit_peer_request(device, peer_req, READ, fault_type) == 0) |
82bc0194 | 2700 | return 0; |
b411b363 | 2701 | |
10f6d992 | 2702 | /* don't care for the reason here */ |
d0180171 | 2703 | drbd_err(device, "submit failed, triggering re-connect\n"); |
c5a2c150 LE |
2704 | |
2705 | out_free_e: | |
0500813f | 2706 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 2707 | list_del(&peer_req->w.list); |
0500813f | 2708 | spin_unlock_irq(&device->resource->req_lock); |
22cc37a9 LE |
2709 | /* no drbd_rs_complete_io(), we are dropping the connection anyways */ |
2710 | ||
b30ab791 AG |
2711 | put_ldev(device); |
2712 | drbd_free_peer_req(device, peer_req); | |
82bc0194 | 2713 | return -EIO; |
b411b363 PR |
2714 | } |
2715 | ||
69a22773 AG |
2716 | /** |
2717 | * drbd_asb_recover_0p - Recover after split-brain with no remaining primaries | |
2718 | */ | |
2719 | static int drbd_asb_recover_0p(struct drbd_peer_device *peer_device) __must_hold(local) | |
b411b363 | 2720 | { |
69a22773 | 2721 | struct drbd_device *device = peer_device->device; |
b411b363 PR |
2722 | int self, peer, rv = -100; |
2723 | unsigned long ch_self, ch_peer; | |
44ed167d | 2724 | enum drbd_after_sb_p after_sb_0p; |
b411b363 | 2725 | |
b30ab791 AG |
2726 | self = device->ldev->md.uuid[UI_BITMAP] & 1; |
2727 | peer = device->p_uuid[UI_BITMAP] & 1; | |
b411b363 | 2728 | |
b30ab791 AG |
2729 | ch_peer = device->p_uuid[UI_SIZE]; |
2730 | ch_self = device->comm_bm_set; | |
b411b363 | 2731 | |
44ed167d | 2732 | rcu_read_lock(); |
69a22773 | 2733 | after_sb_0p = rcu_dereference(peer_device->connection->net_conf)->after_sb_0p; |
44ed167d PR |
2734 | rcu_read_unlock(); |
2735 | switch (after_sb_0p) { | |
b411b363 PR |
2736 | case ASB_CONSENSUS: |
2737 | case ASB_DISCARD_SECONDARY: | |
2738 | case ASB_CALL_HELPER: | |
44ed167d | 2739 | case ASB_VIOLENTLY: |
d0180171 | 2740 | drbd_err(device, "Configuration error.\n"); |
b411b363 PR |
2741 | break; |
2742 | case ASB_DISCONNECT: | |
2743 | break; | |
2744 | case ASB_DISCARD_YOUNGER_PRI: | |
2745 | if (self == 0 && peer == 1) { | |
2746 | rv = -1; | |
2747 | break; | |
2748 | } | |
2749 | if (self == 1 && peer == 0) { | |
2750 | rv = 1; | |
2751 | break; | |
2752 | } | |
2753 | /* Else fall through to one of the other strategies... */ | |
2754 | case ASB_DISCARD_OLDER_PRI: | |
2755 | if (self == 0 && peer == 1) { | |
2756 | rv = 1; | |
2757 | break; | |
2758 | } | |
2759 | if (self == 1 && peer == 0) { | |
2760 | rv = -1; | |
2761 | break; | |
2762 | } | |
2763 | /* Else fall through to one of the other strategies... */ | |
d0180171 | 2764 | drbd_warn(device, "Discard younger/older primary did not find a decision\n" |
b411b363 PR |
2765 | "Using discard-least-changes instead\n"); |
2766 | case ASB_DISCARD_ZERO_CHG: | |
2767 | if (ch_peer == 0 && ch_self == 0) { | |
69a22773 | 2768 | rv = test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags) |
b411b363 PR |
2769 | ? -1 : 1; |
2770 | break; | |
2771 | } else { | |
2772 | if (ch_peer == 0) { rv = 1; break; } | |
2773 | if (ch_self == 0) { rv = -1; break; } | |
2774 | } | |
44ed167d | 2775 | if (after_sb_0p == ASB_DISCARD_ZERO_CHG) |
b411b363 PR |
2776 | break; |
2777 | case ASB_DISCARD_LEAST_CHG: | |
2778 | if (ch_self < ch_peer) | |
2779 | rv = -1; | |
2780 | else if (ch_self > ch_peer) | |
2781 | rv = 1; | |
2782 | else /* ( ch_self == ch_peer ) */ | |
2783 | /* Well, then use something else. */ | |
69a22773 | 2784 | rv = test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags) |
b411b363 PR |
2785 | ? -1 : 1; |
2786 | break; | |
2787 | case ASB_DISCARD_LOCAL: | |
2788 | rv = -1; | |
2789 | break; | |
2790 | case ASB_DISCARD_REMOTE: | |
2791 | rv = 1; | |
2792 | } | |
2793 | ||
2794 | return rv; | |
2795 | } | |
2796 | ||
69a22773 AG |
2797 | /** |
2798 | * drbd_asb_recover_1p - Recover after split-brain with one remaining primary | |
2799 | */ | |
2800 | static int drbd_asb_recover_1p(struct drbd_peer_device *peer_device) __must_hold(local) | |
b411b363 | 2801 | { |
69a22773 | 2802 | struct drbd_device *device = peer_device->device; |
6184ea21 | 2803 | int hg, rv = -100; |
44ed167d | 2804 | enum drbd_after_sb_p after_sb_1p; |
b411b363 | 2805 | |
44ed167d | 2806 | rcu_read_lock(); |
69a22773 | 2807 | after_sb_1p = rcu_dereference(peer_device->connection->net_conf)->after_sb_1p; |
44ed167d PR |
2808 | rcu_read_unlock(); |
2809 | switch (after_sb_1p) { | |
b411b363 PR |
2810 | case ASB_DISCARD_YOUNGER_PRI: |
2811 | case ASB_DISCARD_OLDER_PRI: | |
2812 | case ASB_DISCARD_LEAST_CHG: | |
2813 | case ASB_DISCARD_LOCAL: | |
2814 | case ASB_DISCARD_REMOTE: | |
44ed167d | 2815 | case ASB_DISCARD_ZERO_CHG: |
d0180171 | 2816 | drbd_err(device, "Configuration error.\n"); |
b411b363 PR |
2817 | break; |
2818 | case ASB_DISCONNECT: | |
2819 | break; | |
2820 | case ASB_CONSENSUS: | |
69a22773 | 2821 | hg = drbd_asb_recover_0p(peer_device); |
b30ab791 | 2822 | if (hg == -1 && device->state.role == R_SECONDARY) |
b411b363 | 2823 | rv = hg; |
b30ab791 | 2824 | if (hg == 1 && device->state.role == R_PRIMARY) |
b411b363 PR |
2825 | rv = hg; |
2826 | break; | |
2827 | case ASB_VIOLENTLY: | |
69a22773 | 2828 | rv = drbd_asb_recover_0p(peer_device); |
b411b363 PR |
2829 | break; |
2830 | case ASB_DISCARD_SECONDARY: | |
b30ab791 | 2831 | return device->state.role == R_PRIMARY ? 1 : -1; |
b411b363 | 2832 | case ASB_CALL_HELPER: |
69a22773 | 2833 | hg = drbd_asb_recover_0p(peer_device); |
b30ab791 | 2834 | if (hg == -1 && device->state.role == R_PRIMARY) { |
bb437946 AG |
2835 | enum drbd_state_rv rv2; |
2836 | ||
b411b363 PR |
2837 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2838 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2839 | * we do not need to wait for the after state change work either. */ | |
b30ab791 | 2840 | rv2 = drbd_change_state(device, CS_VERBOSE, NS(role, R_SECONDARY)); |
bb437946 | 2841 | if (rv2 != SS_SUCCESS) { |
b30ab791 | 2842 | drbd_khelper(device, "pri-lost-after-sb"); |
b411b363 | 2843 | } else { |
d0180171 | 2844 | drbd_warn(device, "Successfully gave up primary role.\n"); |
b411b363 PR |
2845 | rv = hg; |
2846 | } | |
2847 | } else | |
2848 | rv = hg; | |
2849 | } | |
2850 | ||
2851 | return rv; | |
2852 | } | |
2853 | ||
69a22773 AG |
2854 | /** |
2855 | * drbd_asb_recover_2p - Recover after split-brain with two remaining primaries | |
2856 | */ | |
2857 | static int drbd_asb_recover_2p(struct drbd_peer_device *peer_device) __must_hold(local) | |
b411b363 | 2858 | { |
69a22773 | 2859 | struct drbd_device *device = peer_device->device; |
6184ea21 | 2860 | int hg, rv = -100; |
44ed167d | 2861 | enum drbd_after_sb_p after_sb_2p; |
b411b363 | 2862 | |
44ed167d | 2863 | rcu_read_lock(); |
69a22773 | 2864 | after_sb_2p = rcu_dereference(peer_device->connection->net_conf)->after_sb_2p; |
44ed167d PR |
2865 | rcu_read_unlock(); |
2866 | switch (after_sb_2p) { | |
b411b363 PR |
2867 | case ASB_DISCARD_YOUNGER_PRI: |
2868 | case ASB_DISCARD_OLDER_PRI: | |
2869 | case ASB_DISCARD_LEAST_CHG: | |
2870 | case ASB_DISCARD_LOCAL: | |
2871 | case ASB_DISCARD_REMOTE: | |
2872 | case ASB_CONSENSUS: | |
2873 | case ASB_DISCARD_SECONDARY: | |
44ed167d | 2874 | case ASB_DISCARD_ZERO_CHG: |
d0180171 | 2875 | drbd_err(device, "Configuration error.\n"); |
b411b363 PR |
2876 | break; |
2877 | case ASB_VIOLENTLY: | |
69a22773 | 2878 | rv = drbd_asb_recover_0p(peer_device); |
b411b363 PR |
2879 | break; |
2880 | case ASB_DISCONNECT: | |
2881 | break; | |
2882 | case ASB_CALL_HELPER: | |
69a22773 | 2883 | hg = drbd_asb_recover_0p(peer_device); |
b411b363 | 2884 | if (hg == -1) { |
bb437946 AG |
2885 | enum drbd_state_rv rv2; |
2886 | ||
b411b363 PR |
2887 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2888 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2889 | * we do not need to wait for the after state change work either. */ | |
b30ab791 | 2890 | rv2 = drbd_change_state(device, CS_VERBOSE, NS(role, R_SECONDARY)); |
bb437946 | 2891 | if (rv2 != SS_SUCCESS) { |
b30ab791 | 2892 | drbd_khelper(device, "pri-lost-after-sb"); |
b411b363 | 2893 | } else { |
d0180171 | 2894 | drbd_warn(device, "Successfully gave up primary role.\n"); |
b411b363 PR |
2895 | rv = hg; |
2896 | } | |
2897 | } else | |
2898 | rv = hg; | |
2899 | } | |
2900 | ||
2901 | return rv; | |
2902 | } | |
2903 | ||
b30ab791 | 2904 | static void drbd_uuid_dump(struct drbd_device *device, char *text, u64 *uuid, |
b411b363 PR |
2905 | u64 bits, u64 flags) |
2906 | { | |
2907 | if (!uuid) { | |
d0180171 | 2908 | drbd_info(device, "%s uuid info vanished while I was looking!\n", text); |
b411b363 PR |
2909 | return; |
2910 | } | |
d0180171 | 2911 | drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", |
b411b363 PR |
2912 | text, |
2913 | (unsigned long long)uuid[UI_CURRENT], | |
2914 | (unsigned long long)uuid[UI_BITMAP], | |
2915 | (unsigned long long)uuid[UI_HISTORY_START], | |
2916 | (unsigned long long)uuid[UI_HISTORY_END], | |
2917 | (unsigned long long)bits, | |
2918 | (unsigned long long)flags); | |
2919 | } | |
2920 | ||
2921 | /* | |
2922 | 100 after split brain try auto recover | |
2923 | 2 C_SYNC_SOURCE set BitMap | |
2924 | 1 C_SYNC_SOURCE use BitMap | |
2925 | 0 no Sync | |
2926 | -1 C_SYNC_TARGET use BitMap | |
2927 | -2 C_SYNC_TARGET set BitMap | |
2928 | -100 after split brain, disconnect | |
2929 | -1000 unrelated data | |
4a23f264 PR |
2930 | -1091 requires proto 91 |
2931 | -1096 requires proto 96 | |
b411b363 | 2932 | */ |
44a4d551 | 2933 | static int drbd_uuid_compare(struct drbd_device *const device, int *rule_nr) __must_hold(local) |
b411b363 | 2934 | { |
44a4d551 LE |
2935 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
2936 | struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL; | |
b411b363 PR |
2937 | u64 self, peer; |
2938 | int i, j; | |
2939 | ||
b30ab791 AG |
2940 | self = device->ldev->md.uuid[UI_CURRENT] & ~((u64)1); |
2941 | peer = device->p_uuid[UI_CURRENT] & ~((u64)1); | |
b411b363 PR |
2942 | |
2943 | *rule_nr = 10; | |
2944 | if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) | |
2945 | return 0; | |
2946 | ||
2947 | *rule_nr = 20; | |
2948 | if ((self == UUID_JUST_CREATED || self == (u64)0) && | |
2949 | peer != UUID_JUST_CREATED) | |
2950 | return -2; | |
2951 | ||
2952 | *rule_nr = 30; | |
2953 | if (self != UUID_JUST_CREATED && | |
2954 | (peer == UUID_JUST_CREATED || peer == (u64)0)) | |
2955 | return 2; | |
2956 | ||
2957 | if (self == peer) { | |
2958 | int rct, dc; /* roles at crash time */ | |
2959 | ||
b30ab791 | 2960 | if (device->p_uuid[UI_BITMAP] == (u64)0 && device->ldev->md.uuid[UI_BITMAP] != (u64)0) { |
b411b363 | 2961 | |
44a4d551 | 2962 | if (connection->agreed_pro_version < 91) |
4a23f264 | 2963 | return -1091; |
b411b363 | 2964 | |
b30ab791 AG |
2965 | if ((device->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START] & ~((u64)1)) && |
2966 | (device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { | |
d0180171 | 2967 | drbd_info(device, "was SyncSource, missed the resync finished event, corrected myself:\n"); |
b30ab791 AG |
2968 | drbd_uuid_move_history(device); |
2969 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP]; | |
2970 | device->ldev->md.uuid[UI_BITMAP] = 0; | |
b411b363 | 2971 | |
b30ab791 AG |
2972 | drbd_uuid_dump(device, "self", device->ldev->md.uuid, |
2973 | device->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(device) : 0, 0); | |
b411b363 PR |
2974 | *rule_nr = 34; |
2975 | } else { | |
d0180171 | 2976 | drbd_info(device, "was SyncSource (peer failed to write sync_uuid)\n"); |
b411b363 PR |
2977 | *rule_nr = 36; |
2978 | } | |
2979 | ||
2980 | return 1; | |
2981 | } | |
2982 | ||
b30ab791 | 2983 | if (device->ldev->md.uuid[UI_BITMAP] == (u64)0 && device->p_uuid[UI_BITMAP] != (u64)0) { |
b411b363 | 2984 | |
44a4d551 | 2985 | if (connection->agreed_pro_version < 91) |
4a23f264 | 2986 | return -1091; |
b411b363 | 2987 | |
b30ab791 AG |
2988 | if ((device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (device->p_uuid[UI_BITMAP] & ~((u64)1)) && |
2989 | (device->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START] & ~((u64)1))) { | |
d0180171 | 2990 | drbd_info(device, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); |
b411b363 | 2991 | |
b30ab791 AG |
2992 | device->p_uuid[UI_HISTORY_START + 1] = device->p_uuid[UI_HISTORY_START]; |
2993 | device->p_uuid[UI_HISTORY_START] = device->p_uuid[UI_BITMAP]; | |
2994 | device->p_uuid[UI_BITMAP] = 0UL; | |
b411b363 | 2995 | |
b30ab791 | 2996 | drbd_uuid_dump(device, "peer", device->p_uuid, device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); |
b411b363 PR |
2997 | *rule_nr = 35; |
2998 | } else { | |
d0180171 | 2999 | drbd_info(device, "was SyncTarget (failed to write sync_uuid)\n"); |
b411b363 PR |
3000 | *rule_nr = 37; |
3001 | } | |
3002 | ||
3003 | return -1; | |
3004 | } | |
3005 | ||
3006 | /* Common power [off|failure] */ | |
b30ab791 AG |
3007 | rct = (test_bit(CRASHED_PRIMARY, &device->flags) ? 1 : 0) + |
3008 | (device->p_uuid[UI_FLAGS] & 2); | |
b411b363 PR |
3009 | /* lowest bit is set when we were primary, |
3010 | * next bit (weight 2) is set when peer was primary */ | |
3011 | *rule_nr = 40; | |
3012 | ||
3013 | switch (rct) { | |
3014 | case 0: /* !self_pri && !peer_pri */ return 0; | |
3015 | case 1: /* self_pri && !peer_pri */ return 1; | |
3016 | case 2: /* !self_pri && peer_pri */ return -1; | |
3017 | case 3: /* self_pri && peer_pri */ | |
44a4d551 | 3018 | dc = test_bit(RESOLVE_CONFLICTS, &connection->flags); |
b411b363 PR |
3019 | return dc ? -1 : 1; |
3020 | } | |
3021 | } | |
3022 | ||
3023 | *rule_nr = 50; | |
b30ab791 | 3024 | peer = device->p_uuid[UI_BITMAP] & ~((u64)1); |
b411b363 PR |
3025 | if (self == peer) |
3026 | return -1; | |
3027 | ||
3028 | *rule_nr = 51; | |
b30ab791 | 3029 | peer = device->p_uuid[UI_HISTORY_START] & ~((u64)1); |
b411b363 | 3030 | if (self == peer) { |
44a4d551 | 3031 | if (connection->agreed_pro_version < 96 ? |
b30ab791 AG |
3032 | (device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == |
3033 | (device->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) : | |
3034 | peer + UUID_NEW_BM_OFFSET == (device->p_uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
3035 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
3036 | resync as sync source modifications of the peer's UUIDs. */ | |
3037 | ||
44a4d551 | 3038 | if (connection->agreed_pro_version < 91) |
4a23f264 | 3039 | return -1091; |
b411b363 | 3040 | |
b30ab791 AG |
3041 | device->p_uuid[UI_BITMAP] = device->p_uuid[UI_HISTORY_START]; |
3042 | device->p_uuid[UI_HISTORY_START] = device->p_uuid[UI_HISTORY_START + 1]; | |
4a23f264 | 3043 | |
d0180171 | 3044 | drbd_info(device, "Lost last syncUUID packet, corrected:\n"); |
b30ab791 | 3045 | drbd_uuid_dump(device, "peer", device->p_uuid, device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); |
4a23f264 | 3046 | |
b411b363 PR |
3047 | return -1; |
3048 | } | |
3049 | } | |
3050 | ||
3051 | *rule_nr = 60; | |
b30ab791 | 3052 | self = device->ldev->md.uuid[UI_CURRENT] & ~((u64)1); |
b411b363 | 3053 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
b30ab791 | 3054 | peer = device->p_uuid[i] & ~((u64)1); |
b411b363 PR |
3055 | if (self == peer) |
3056 | return -2; | |
3057 | } | |
3058 | ||
3059 | *rule_nr = 70; | |
b30ab791 AG |
3060 | self = device->ldev->md.uuid[UI_BITMAP] & ~((u64)1); |
3061 | peer = device->p_uuid[UI_CURRENT] & ~((u64)1); | |
b411b363 PR |
3062 | if (self == peer) |
3063 | return 1; | |
3064 | ||
3065 | *rule_nr = 71; | |
b30ab791 | 3066 | self = device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); |
b411b363 | 3067 | if (self == peer) { |
44a4d551 | 3068 | if (connection->agreed_pro_version < 96 ? |
b30ab791 AG |
3069 | (device->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == |
3070 | (device->p_uuid[UI_HISTORY_START] & ~((u64)1)) : | |
3071 | self + UUID_NEW_BM_OFFSET == (device->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
3072 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
3073 | resync as sync source modifications of our UUIDs. */ | |
3074 | ||
44a4d551 | 3075 | if (connection->agreed_pro_version < 91) |
4a23f264 | 3076 | return -1091; |
b411b363 | 3077 | |
b30ab791 AG |
3078 | __drbd_uuid_set(device, UI_BITMAP, device->ldev->md.uuid[UI_HISTORY_START]); |
3079 | __drbd_uuid_set(device, UI_HISTORY_START, device->ldev->md.uuid[UI_HISTORY_START + 1]); | |
b411b363 | 3080 | |
d0180171 | 3081 | drbd_info(device, "Last syncUUID did not get through, corrected:\n"); |
b30ab791 AG |
3082 | drbd_uuid_dump(device, "self", device->ldev->md.uuid, |
3083 | device->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(device) : 0, 0); | |
b411b363 PR |
3084 | |
3085 | return 1; | |
3086 | } | |
3087 | } | |
3088 | ||
3089 | ||
3090 | *rule_nr = 80; | |
b30ab791 | 3091 | peer = device->p_uuid[UI_CURRENT] & ~((u64)1); |
b411b363 | 3092 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
b30ab791 | 3093 | self = device->ldev->md.uuid[i] & ~((u64)1); |
b411b363 PR |
3094 | if (self == peer) |
3095 | return 2; | |
3096 | } | |
3097 | ||
3098 | *rule_nr = 90; | |
b30ab791 AG |
3099 | self = device->ldev->md.uuid[UI_BITMAP] & ~((u64)1); |
3100 | peer = device->p_uuid[UI_BITMAP] & ~((u64)1); | |
b411b363 PR |
3101 | if (self == peer && self != ((u64)0)) |
3102 | return 100; | |
3103 | ||
3104 | *rule_nr = 100; | |
3105 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
b30ab791 | 3106 | self = device->ldev->md.uuid[i] & ~((u64)1); |
b411b363 | 3107 | for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { |
b30ab791 | 3108 | peer = device->p_uuid[j] & ~((u64)1); |
b411b363 PR |
3109 | if (self == peer) |
3110 | return -100; | |
3111 | } | |
3112 | } | |
3113 | ||
3114 | return -1000; | |
3115 | } | |
3116 | ||
3117 | /* drbd_sync_handshake() returns the new conn state on success, or | |
3118 | CONN_MASK (-1) on failure. | |
3119 | */ | |
69a22773 AG |
3120 | static enum drbd_conns drbd_sync_handshake(struct drbd_peer_device *peer_device, |
3121 | enum drbd_role peer_role, | |
b411b363 PR |
3122 | enum drbd_disk_state peer_disk) __must_hold(local) |
3123 | { | |
69a22773 | 3124 | struct drbd_device *device = peer_device->device; |
b411b363 PR |
3125 | enum drbd_conns rv = C_MASK; |
3126 | enum drbd_disk_state mydisk; | |
44ed167d | 3127 | struct net_conf *nc; |
6dff2902 | 3128 | int hg, rule_nr, rr_conflict, tentative; |
b411b363 | 3129 | |
b30ab791 | 3130 | mydisk = device->state.disk; |
b411b363 | 3131 | if (mydisk == D_NEGOTIATING) |
b30ab791 | 3132 | mydisk = device->new_state_tmp.disk; |
b411b363 | 3133 | |
d0180171 | 3134 | drbd_info(device, "drbd_sync_handshake:\n"); |
9f2247bb | 3135 | |
b30ab791 AG |
3136 | spin_lock_irq(&device->ldev->md.uuid_lock); |
3137 | drbd_uuid_dump(device, "self", device->ldev->md.uuid, device->comm_bm_set, 0); | |
3138 | drbd_uuid_dump(device, "peer", device->p_uuid, | |
3139 | device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); | |
b411b363 | 3140 | |
b30ab791 AG |
3141 | hg = drbd_uuid_compare(device, &rule_nr); |
3142 | spin_unlock_irq(&device->ldev->md.uuid_lock); | |
b411b363 | 3143 | |
d0180171 | 3144 | drbd_info(device, "uuid_compare()=%d by rule %d\n", hg, rule_nr); |
b411b363 PR |
3145 | |
3146 | if (hg == -1000) { | |
d0180171 | 3147 | drbd_alert(device, "Unrelated data, aborting!\n"); |
b411b363 PR |
3148 | return C_MASK; |
3149 | } | |
4a23f264 | 3150 | if (hg < -1000) { |
d0180171 | 3151 | drbd_alert(device, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000); |
b411b363 PR |
3152 | return C_MASK; |
3153 | } | |
3154 | ||
3155 | if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || | |
3156 | (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { | |
3157 | int f = (hg == -100) || abs(hg) == 2; | |
3158 | hg = mydisk > D_INCONSISTENT ? 1 : -1; | |
3159 | if (f) | |
3160 | hg = hg*2; | |
d0180171 | 3161 | drbd_info(device, "Becoming sync %s due to disk states.\n", |
b411b363 PR |
3162 | hg > 0 ? "source" : "target"); |
3163 | } | |
3164 | ||
3a11a487 | 3165 | if (abs(hg) == 100) |
b30ab791 | 3166 | drbd_khelper(device, "initial-split-brain"); |
3a11a487 | 3167 | |
44ed167d | 3168 | rcu_read_lock(); |
69a22773 | 3169 | nc = rcu_dereference(peer_device->connection->net_conf); |
44ed167d PR |
3170 | |
3171 | if (hg == 100 || (hg == -100 && nc->always_asbp)) { | |
b30ab791 | 3172 | int pcount = (device->state.role == R_PRIMARY) |
b411b363 PR |
3173 | + (peer_role == R_PRIMARY); |
3174 | int forced = (hg == -100); | |
3175 | ||
3176 | switch (pcount) { | |
3177 | case 0: | |
69a22773 | 3178 | hg = drbd_asb_recover_0p(peer_device); |
b411b363 PR |
3179 | break; |
3180 | case 1: | |
69a22773 | 3181 | hg = drbd_asb_recover_1p(peer_device); |
b411b363 PR |
3182 | break; |
3183 | case 2: | |
69a22773 | 3184 | hg = drbd_asb_recover_2p(peer_device); |
b411b363 PR |
3185 | break; |
3186 | } | |
3187 | if (abs(hg) < 100) { | |
d0180171 | 3188 | drbd_warn(device, "Split-Brain detected, %d primaries, " |
b411b363 PR |
3189 | "automatically solved. Sync from %s node\n", |
3190 | pcount, (hg < 0) ? "peer" : "this"); | |
3191 | if (forced) { | |
d0180171 | 3192 | drbd_warn(device, "Doing a full sync, since" |
b411b363 PR |
3193 | " UUIDs where ambiguous.\n"); |
3194 | hg = hg*2; | |
3195 | } | |
3196 | } | |
3197 | } | |
3198 | ||
3199 | if (hg == -100) { | |
b30ab791 | 3200 | if (test_bit(DISCARD_MY_DATA, &device->flags) && !(device->p_uuid[UI_FLAGS]&1)) |
b411b363 | 3201 | hg = -1; |
b30ab791 | 3202 | if (!test_bit(DISCARD_MY_DATA, &device->flags) && (device->p_uuid[UI_FLAGS]&1)) |
b411b363 PR |
3203 | hg = 1; |
3204 | ||
3205 | if (abs(hg) < 100) | |
d0180171 | 3206 | drbd_warn(device, "Split-Brain detected, manually solved. " |
b411b363 PR |
3207 | "Sync from %s node\n", |
3208 | (hg < 0) ? "peer" : "this"); | |
3209 | } | |
44ed167d | 3210 | rr_conflict = nc->rr_conflict; |
6dff2902 | 3211 | tentative = nc->tentative; |
44ed167d | 3212 | rcu_read_unlock(); |
b411b363 PR |
3213 | |
3214 | if (hg == -100) { | |
580b9767 LE |
3215 | /* FIXME this log message is not correct if we end up here |
3216 | * after an attempted attach on a diskless node. | |
3217 | * We just refuse to attach -- well, we drop the "connection" | |
3218 | * to that disk, in a way... */ | |
d0180171 | 3219 | drbd_alert(device, "Split-Brain detected but unresolved, dropping connection!\n"); |
b30ab791 | 3220 | drbd_khelper(device, "split-brain"); |
b411b363 PR |
3221 | return C_MASK; |
3222 | } | |
3223 | ||
3224 | if (hg > 0 && mydisk <= D_INCONSISTENT) { | |
d0180171 | 3225 | drbd_err(device, "I shall become SyncSource, but I am inconsistent!\n"); |
b411b363 PR |
3226 | return C_MASK; |
3227 | } | |
3228 | ||
3229 | if (hg < 0 && /* by intention we do not use mydisk here. */ | |
b30ab791 | 3230 | device->state.role == R_PRIMARY && device->state.disk >= D_CONSISTENT) { |
44ed167d | 3231 | switch (rr_conflict) { |
b411b363 | 3232 | case ASB_CALL_HELPER: |
b30ab791 | 3233 | drbd_khelper(device, "pri-lost"); |
b411b363 PR |
3234 | /* fall through */ |
3235 | case ASB_DISCONNECT: | |
d0180171 | 3236 | drbd_err(device, "I shall become SyncTarget, but I am primary!\n"); |
b411b363 PR |
3237 | return C_MASK; |
3238 | case ASB_VIOLENTLY: | |
d0180171 | 3239 | drbd_warn(device, "Becoming SyncTarget, violating the stable-data" |
b411b363 PR |
3240 | "assumption\n"); |
3241 | } | |
3242 | } | |
3243 | ||
69a22773 | 3244 | if (tentative || test_bit(CONN_DRY_RUN, &peer_device->connection->flags)) { |
cf14c2e9 | 3245 | if (hg == 0) |
d0180171 | 3246 | drbd_info(device, "dry-run connect: No resync, would become Connected immediately.\n"); |
cf14c2e9 | 3247 | else |
d0180171 | 3248 | drbd_info(device, "dry-run connect: Would become %s, doing a %s resync.", |
cf14c2e9 PR |
3249 | drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), |
3250 | abs(hg) >= 2 ? "full" : "bit-map based"); | |
3251 | return C_MASK; | |
3252 | } | |
3253 | ||
b411b363 | 3254 | if (abs(hg) >= 2) { |
d0180171 | 3255 | drbd_info(device, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); |
b30ab791 | 3256 | if (drbd_bitmap_io(device, &drbd_bmio_set_n_write, "set_n_write from sync_handshake", |
20ceb2b2 | 3257 | BM_LOCKED_SET_ALLOWED)) |
b411b363 PR |
3258 | return C_MASK; |
3259 | } | |
3260 | ||
3261 | if (hg > 0) { /* become sync source. */ | |
3262 | rv = C_WF_BITMAP_S; | |
3263 | } else if (hg < 0) { /* become sync target */ | |
3264 | rv = C_WF_BITMAP_T; | |
3265 | } else { | |
3266 | rv = C_CONNECTED; | |
b30ab791 | 3267 | if (drbd_bm_total_weight(device)) { |
d0180171 | 3268 | drbd_info(device, "No resync, but %lu bits in bitmap!\n", |
b30ab791 | 3269 | drbd_bm_total_weight(device)); |
b411b363 PR |
3270 | } |
3271 | } | |
3272 | ||
3273 | return rv; | |
3274 | } | |
3275 | ||
f179d76d | 3276 | static enum drbd_after_sb_p convert_after_sb(enum drbd_after_sb_p peer) |
b411b363 PR |
3277 | { |
3278 | /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ | |
f179d76d PR |
3279 | if (peer == ASB_DISCARD_REMOTE) |
3280 | return ASB_DISCARD_LOCAL; | |
b411b363 PR |
3281 | |
3282 | /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ | |
f179d76d PR |
3283 | if (peer == ASB_DISCARD_LOCAL) |
3284 | return ASB_DISCARD_REMOTE; | |
b411b363 PR |
3285 | |
3286 | /* everything else is valid if they are equal on both sides. */ | |
f179d76d | 3287 | return peer; |
b411b363 PR |
3288 | } |
3289 | ||
bde89a9e | 3290 | static int receive_protocol(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3291 | { |
e658983a | 3292 | struct p_protocol *p = pi->data; |
036b17ea PR |
3293 | enum drbd_after_sb_p p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; |
3294 | int p_proto, p_discard_my_data, p_two_primaries, cf; | |
3295 | struct net_conf *nc, *old_net_conf, *new_net_conf = NULL; | |
3296 | char integrity_alg[SHARED_SECRET_MAX] = ""; | |
accdbcc5 | 3297 | struct crypto_hash *peer_integrity_tfm = NULL; |
7aca6c75 | 3298 | void *int_dig_in = NULL, *int_dig_vv = NULL; |
b411b363 | 3299 | |
b411b363 PR |
3300 | p_proto = be32_to_cpu(p->protocol); |
3301 | p_after_sb_0p = be32_to_cpu(p->after_sb_0p); | |
3302 | p_after_sb_1p = be32_to_cpu(p->after_sb_1p); | |
3303 | p_after_sb_2p = be32_to_cpu(p->after_sb_2p); | |
b411b363 | 3304 | p_two_primaries = be32_to_cpu(p->two_primaries); |
cf14c2e9 | 3305 | cf = be32_to_cpu(p->conn_flags); |
6139f60d | 3306 | p_discard_my_data = cf & CF_DISCARD_MY_DATA; |
cf14c2e9 | 3307 | |
bde89a9e | 3308 | if (connection->agreed_pro_version >= 87) { |
86db0618 | 3309 | int err; |
cf14c2e9 | 3310 | |
88104ca4 | 3311 | if (pi->size > sizeof(integrity_alg)) |
86db0618 | 3312 | return -EIO; |
bde89a9e | 3313 | err = drbd_recv_all(connection, integrity_alg, pi->size); |
86db0618 AG |
3314 | if (err) |
3315 | return err; | |
036b17ea | 3316 | integrity_alg[SHARED_SECRET_MAX - 1] = 0; |
b411b363 PR |
3317 | } |
3318 | ||
7d4c782c | 3319 | if (pi->cmd != P_PROTOCOL_UPDATE) { |
bde89a9e | 3320 | clear_bit(CONN_DRY_RUN, &connection->flags); |
b411b363 | 3321 | |
fbc12f45 | 3322 | if (cf & CF_DRY_RUN) |
bde89a9e | 3323 | set_bit(CONN_DRY_RUN, &connection->flags); |
b411b363 | 3324 | |
fbc12f45 | 3325 | rcu_read_lock(); |
bde89a9e | 3326 | nc = rcu_dereference(connection->net_conf); |
b411b363 | 3327 | |
fbc12f45 | 3328 | if (p_proto != nc->wire_protocol) { |
1ec861eb | 3329 | drbd_err(connection, "incompatible %s settings\n", "protocol"); |
fbc12f45 AG |
3330 | goto disconnect_rcu_unlock; |
3331 | } | |
b411b363 | 3332 | |
fbc12f45 | 3333 | if (convert_after_sb(p_after_sb_0p) != nc->after_sb_0p) { |
1ec861eb | 3334 | drbd_err(connection, "incompatible %s settings\n", "after-sb-0pri"); |
fbc12f45 AG |
3335 | goto disconnect_rcu_unlock; |
3336 | } | |
b411b363 | 3337 | |
fbc12f45 | 3338 | if (convert_after_sb(p_after_sb_1p) != nc->after_sb_1p) { |
1ec861eb | 3339 | drbd_err(connection, "incompatible %s settings\n", "after-sb-1pri"); |
fbc12f45 AG |
3340 | goto disconnect_rcu_unlock; |
3341 | } | |
b411b363 | 3342 | |
fbc12f45 | 3343 | if (convert_after_sb(p_after_sb_2p) != nc->after_sb_2p) { |
1ec861eb | 3344 | drbd_err(connection, "incompatible %s settings\n", "after-sb-2pri"); |
fbc12f45 AG |
3345 | goto disconnect_rcu_unlock; |
3346 | } | |
b411b363 | 3347 | |
fbc12f45 | 3348 | if (p_discard_my_data && nc->discard_my_data) { |
1ec861eb | 3349 | drbd_err(connection, "incompatible %s settings\n", "discard-my-data"); |
fbc12f45 AG |
3350 | goto disconnect_rcu_unlock; |
3351 | } | |
b411b363 | 3352 | |
fbc12f45 | 3353 | if (p_two_primaries != nc->two_primaries) { |
1ec861eb | 3354 | drbd_err(connection, "incompatible %s settings\n", "allow-two-primaries"); |
fbc12f45 AG |
3355 | goto disconnect_rcu_unlock; |
3356 | } | |
b411b363 | 3357 | |
fbc12f45 | 3358 | if (strcmp(integrity_alg, nc->integrity_alg)) { |
1ec861eb | 3359 | drbd_err(connection, "incompatible %s settings\n", "data-integrity-alg"); |
fbc12f45 AG |
3360 | goto disconnect_rcu_unlock; |
3361 | } | |
b411b363 | 3362 | |
fbc12f45 | 3363 | rcu_read_unlock(); |
b411b363 PR |
3364 | } |
3365 | ||
7d4c782c AG |
3366 | if (integrity_alg[0]) { |
3367 | int hash_size; | |
3368 | ||
3369 | /* | |
3370 | * We can only change the peer data integrity algorithm | |
3371 | * here. Changing our own data integrity algorithm | |
3372 | * requires that we send a P_PROTOCOL_UPDATE packet at | |
3373 | * the same time; otherwise, the peer has no way to | |
3374 | * tell between which packets the algorithm should | |
3375 | * change. | |
3376 | */ | |
b411b363 | 3377 | |
7d4c782c AG |
3378 | peer_integrity_tfm = crypto_alloc_hash(integrity_alg, 0, CRYPTO_ALG_ASYNC); |
3379 | if (!peer_integrity_tfm) { | |
1ec861eb | 3380 | drbd_err(connection, "peer data-integrity-alg %s not supported\n", |
7d4c782c AG |
3381 | integrity_alg); |
3382 | goto disconnect; | |
3383 | } | |
b411b363 | 3384 | |
7d4c782c AG |
3385 | hash_size = crypto_hash_digestsize(peer_integrity_tfm); |
3386 | int_dig_in = kmalloc(hash_size, GFP_KERNEL); | |
3387 | int_dig_vv = kmalloc(hash_size, GFP_KERNEL); | |
3388 | if (!(int_dig_in && int_dig_vv)) { | |
1ec861eb | 3389 | drbd_err(connection, "Allocation of buffers for data integrity checking failed\n"); |
b411b363 PR |
3390 | goto disconnect; |
3391 | } | |
b411b363 PR |
3392 | } |
3393 | ||
7d4c782c AG |
3394 | new_net_conf = kmalloc(sizeof(struct net_conf), GFP_KERNEL); |
3395 | if (!new_net_conf) { | |
1ec861eb | 3396 | drbd_err(connection, "Allocation of new net_conf failed\n"); |
7d4c782c AG |
3397 | goto disconnect; |
3398 | } | |
3399 | ||
bde89a9e | 3400 | mutex_lock(&connection->data.mutex); |
0500813f | 3401 | mutex_lock(&connection->resource->conf_update); |
bde89a9e | 3402 | old_net_conf = connection->net_conf; |
7d4c782c AG |
3403 | *new_net_conf = *old_net_conf; |
3404 | ||
3405 | new_net_conf->wire_protocol = p_proto; | |
3406 | new_net_conf->after_sb_0p = convert_after_sb(p_after_sb_0p); | |
3407 | new_net_conf->after_sb_1p = convert_after_sb(p_after_sb_1p); | |
3408 | new_net_conf->after_sb_2p = convert_after_sb(p_after_sb_2p); | |
3409 | new_net_conf->two_primaries = p_two_primaries; | |
3410 | ||
bde89a9e | 3411 | rcu_assign_pointer(connection->net_conf, new_net_conf); |
0500813f | 3412 | mutex_unlock(&connection->resource->conf_update); |
bde89a9e | 3413 | mutex_unlock(&connection->data.mutex); |
7d4c782c | 3414 | |
bde89a9e AG |
3415 | crypto_free_hash(connection->peer_integrity_tfm); |
3416 | kfree(connection->int_dig_in); | |
3417 | kfree(connection->int_dig_vv); | |
3418 | connection->peer_integrity_tfm = peer_integrity_tfm; | |
3419 | connection->int_dig_in = int_dig_in; | |
3420 | connection->int_dig_vv = int_dig_vv; | |
7d4c782c AG |
3421 | |
3422 | if (strcmp(old_net_conf->integrity_alg, integrity_alg)) | |
1ec861eb | 3423 | drbd_info(connection, "peer data-integrity-alg: %s\n", |
7d4c782c AG |
3424 | integrity_alg[0] ? integrity_alg : "(none)"); |
3425 | ||
3426 | synchronize_rcu(); | |
3427 | kfree(old_net_conf); | |
82bc0194 | 3428 | return 0; |
b411b363 | 3429 | |
44ed167d PR |
3430 | disconnect_rcu_unlock: |
3431 | rcu_read_unlock(); | |
b411b363 | 3432 | disconnect: |
b792c35c | 3433 | crypto_free_hash(peer_integrity_tfm); |
036b17ea PR |
3434 | kfree(int_dig_in); |
3435 | kfree(int_dig_vv); | |
bde89a9e | 3436 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3437 | return -EIO; |
b411b363 PR |
3438 | } |
3439 | ||
3440 | /* helper function | |
3441 | * input: alg name, feature name | |
3442 | * return: NULL (alg name was "") | |
3443 | * ERR_PTR(error) if something goes wrong | |
3444 | * or the crypto hash ptr, if it worked out ok. */ | |
8ce953aa | 3445 | static struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_device *device, |
b411b363 PR |
3446 | const char *alg, const char *name) |
3447 | { | |
3448 | struct crypto_hash *tfm; | |
3449 | ||
3450 | if (!alg[0]) | |
3451 | return NULL; | |
3452 | ||
3453 | tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); | |
3454 | if (IS_ERR(tfm)) { | |
d0180171 | 3455 | drbd_err(device, "Can not allocate \"%s\" as %s (reason: %ld)\n", |
b411b363 PR |
3456 | alg, name, PTR_ERR(tfm)); |
3457 | return tfm; | |
3458 | } | |
b411b363 PR |
3459 | return tfm; |
3460 | } | |
3461 | ||
bde89a9e | 3462 | static int ignore_remaining_packet(struct drbd_connection *connection, struct packet_info *pi) |
4a76b161 | 3463 | { |
bde89a9e | 3464 | void *buffer = connection->data.rbuf; |
4a76b161 AG |
3465 | int size = pi->size; |
3466 | ||
3467 | while (size) { | |
3468 | int s = min_t(int, size, DRBD_SOCKET_BUFFER_SIZE); | |
bde89a9e | 3469 | s = drbd_recv(connection, buffer, s); |
4a76b161 AG |
3470 | if (s <= 0) { |
3471 | if (s < 0) | |
3472 | return s; | |
3473 | break; | |
3474 | } | |
3475 | size -= s; | |
3476 | } | |
3477 | if (size) | |
3478 | return -EIO; | |
3479 | return 0; | |
3480 | } | |
3481 | ||
3482 | /* | |
3483 | * config_unknown_volume - device configuration command for unknown volume | |
3484 | * | |
3485 | * When a device is added to an existing connection, the node on which the | |
3486 | * device is added first will send configuration commands to its peer but the | |
3487 | * peer will not know about the device yet. It will warn and ignore these | |
3488 | * commands. Once the device is added on the second node, the second node will | |
3489 | * send the same device configuration commands, but in the other direction. | |
3490 | * | |
3491 | * (We can also end up here if drbd is misconfigured.) | |
3492 | */ | |
bde89a9e | 3493 | static int config_unknown_volume(struct drbd_connection *connection, struct packet_info *pi) |
4a76b161 | 3494 | { |
1ec861eb | 3495 | drbd_warn(connection, "%s packet received for volume %u, which is not configured locally\n", |
2fcb8f30 | 3496 | cmdname(pi->cmd), pi->vnr); |
bde89a9e | 3497 | return ignore_remaining_packet(connection, pi); |
4a76b161 AG |
3498 | } |
3499 | ||
bde89a9e | 3500 | static int receive_SyncParam(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3501 | { |
9f4fe9ad | 3502 | struct drbd_peer_device *peer_device; |
b30ab791 | 3503 | struct drbd_device *device; |
e658983a | 3504 | struct p_rs_param_95 *p; |
b411b363 PR |
3505 | unsigned int header_size, data_size, exp_max_sz; |
3506 | struct crypto_hash *verify_tfm = NULL; | |
3507 | struct crypto_hash *csums_tfm = NULL; | |
2ec91e0e | 3508 | struct net_conf *old_net_conf, *new_net_conf = NULL; |
813472ce | 3509 | struct disk_conf *old_disk_conf = NULL, *new_disk_conf = NULL; |
bde89a9e | 3510 | const int apv = connection->agreed_pro_version; |
813472ce | 3511 | struct fifo_buffer *old_plan = NULL, *new_plan = NULL; |
778f271d | 3512 | int fifo_size = 0; |
82bc0194 | 3513 | int err; |
b411b363 | 3514 | |
9f4fe9ad AG |
3515 | peer_device = conn_peer_device(connection, pi->vnr); |
3516 | if (!peer_device) | |
bde89a9e | 3517 | return config_unknown_volume(connection, pi); |
9f4fe9ad | 3518 | device = peer_device->device; |
b411b363 PR |
3519 | |
3520 | exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) | |
3521 | : apv == 88 ? sizeof(struct p_rs_param) | |
3522 | + SHARED_SECRET_MAX | |
8e26f9cc PR |
3523 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
3524 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 3525 | |
e2857216 | 3526 | if (pi->size > exp_max_sz) { |
d0180171 | 3527 | drbd_err(device, "SyncParam packet too long: received %u, expected <= %u bytes\n", |
e2857216 | 3528 | pi->size, exp_max_sz); |
82bc0194 | 3529 | return -EIO; |
b411b363 PR |
3530 | } |
3531 | ||
3532 | if (apv <= 88) { | |
e658983a | 3533 | header_size = sizeof(struct p_rs_param); |
e2857216 | 3534 | data_size = pi->size - header_size; |
8e26f9cc | 3535 | } else if (apv <= 94) { |
e658983a | 3536 | header_size = sizeof(struct p_rs_param_89); |
e2857216 | 3537 | data_size = pi->size - header_size; |
0b0ba1ef | 3538 | D_ASSERT(device, data_size == 0); |
8e26f9cc | 3539 | } else { |
e658983a | 3540 | header_size = sizeof(struct p_rs_param_95); |
e2857216 | 3541 | data_size = pi->size - header_size; |
0b0ba1ef | 3542 | D_ASSERT(device, data_size == 0); |
b411b363 PR |
3543 | } |
3544 | ||
3545 | /* initialize verify_alg and csums_alg */ | |
e658983a | 3546 | p = pi->data; |
b411b363 PR |
3547 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); |
3548 | ||
9f4fe9ad | 3549 | err = drbd_recv_all(peer_device->connection, p, header_size); |
82bc0194 AG |
3550 | if (err) |
3551 | return err; | |
b411b363 | 3552 | |
0500813f | 3553 | mutex_lock(&connection->resource->conf_update); |
9f4fe9ad | 3554 | old_net_conf = peer_device->connection->net_conf; |
b30ab791 | 3555 | if (get_ldev(device)) { |
813472ce PR |
3556 | new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); |
3557 | if (!new_disk_conf) { | |
b30ab791 | 3558 | put_ldev(device); |
0500813f | 3559 | mutex_unlock(&connection->resource->conf_update); |
d0180171 | 3560 | drbd_err(device, "Allocation of new disk_conf failed\n"); |
813472ce PR |
3561 | return -ENOMEM; |
3562 | } | |
daeda1cc | 3563 | |
b30ab791 | 3564 | old_disk_conf = device->ldev->disk_conf; |
813472ce | 3565 | *new_disk_conf = *old_disk_conf; |
b411b363 | 3566 | |
6394b935 | 3567 | new_disk_conf->resync_rate = be32_to_cpu(p->resync_rate); |
813472ce | 3568 | } |
b411b363 PR |
3569 | |
3570 | if (apv >= 88) { | |
3571 | if (apv == 88) { | |
5de73827 | 3572 | if (data_size > SHARED_SECRET_MAX || data_size == 0) { |
d0180171 | 3573 | drbd_err(device, "verify-alg of wrong size, " |
5de73827 PR |
3574 | "peer wants %u, accepting only up to %u byte\n", |
3575 | data_size, SHARED_SECRET_MAX); | |
813472ce PR |
3576 | err = -EIO; |
3577 | goto reconnect; | |
b411b363 PR |
3578 | } |
3579 | ||
9f4fe9ad | 3580 | err = drbd_recv_all(peer_device->connection, p->verify_alg, data_size); |
813472ce PR |
3581 | if (err) |
3582 | goto reconnect; | |
b411b363 PR |
3583 | /* we expect NUL terminated string */ |
3584 | /* but just in case someone tries to be evil */ | |
0b0ba1ef | 3585 | D_ASSERT(device, p->verify_alg[data_size-1] == 0); |
b411b363 PR |
3586 | p->verify_alg[data_size-1] = 0; |
3587 | ||
3588 | } else /* apv >= 89 */ { | |
3589 | /* we still expect NUL terminated strings */ | |
3590 | /* but just in case someone tries to be evil */ | |
0b0ba1ef AG |
3591 | D_ASSERT(device, p->verify_alg[SHARED_SECRET_MAX-1] == 0); |
3592 | D_ASSERT(device, p->csums_alg[SHARED_SECRET_MAX-1] == 0); | |
b411b363 PR |
3593 | p->verify_alg[SHARED_SECRET_MAX-1] = 0; |
3594 | p->csums_alg[SHARED_SECRET_MAX-1] = 0; | |
3595 | } | |
3596 | ||
2ec91e0e | 3597 | if (strcmp(old_net_conf->verify_alg, p->verify_alg)) { |
b30ab791 | 3598 | if (device->state.conn == C_WF_REPORT_PARAMS) { |
d0180171 | 3599 | drbd_err(device, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", |
2ec91e0e | 3600 | old_net_conf->verify_alg, p->verify_alg); |
b411b363 PR |
3601 | goto disconnect; |
3602 | } | |
b30ab791 | 3603 | verify_tfm = drbd_crypto_alloc_digest_safe(device, |
b411b363 PR |
3604 | p->verify_alg, "verify-alg"); |
3605 | if (IS_ERR(verify_tfm)) { | |
3606 | verify_tfm = NULL; | |
3607 | goto disconnect; | |
3608 | } | |
3609 | } | |
3610 | ||
2ec91e0e | 3611 | if (apv >= 89 && strcmp(old_net_conf->csums_alg, p->csums_alg)) { |
b30ab791 | 3612 | if (device->state.conn == C_WF_REPORT_PARAMS) { |
d0180171 | 3613 | drbd_err(device, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", |
2ec91e0e | 3614 | old_net_conf->csums_alg, p->csums_alg); |
b411b363 PR |
3615 | goto disconnect; |
3616 | } | |
b30ab791 | 3617 | csums_tfm = drbd_crypto_alloc_digest_safe(device, |
b411b363 PR |
3618 | p->csums_alg, "csums-alg"); |
3619 | if (IS_ERR(csums_tfm)) { | |
3620 | csums_tfm = NULL; | |
3621 | goto disconnect; | |
3622 | } | |
3623 | } | |
3624 | ||
813472ce | 3625 | if (apv > 94 && new_disk_conf) { |
daeda1cc PR |
3626 | new_disk_conf->c_plan_ahead = be32_to_cpu(p->c_plan_ahead); |
3627 | new_disk_conf->c_delay_target = be32_to_cpu(p->c_delay_target); | |
3628 | new_disk_conf->c_fill_target = be32_to_cpu(p->c_fill_target); | |
3629 | new_disk_conf->c_max_rate = be32_to_cpu(p->c_max_rate); | |
778f271d | 3630 | |
daeda1cc | 3631 | fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ; |
b30ab791 | 3632 | if (fifo_size != device->rs_plan_s->size) { |
813472ce PR |
3633 | new_plan = fifo_alloc(fifo_size); |
3634 | if (!new_plan) { | |
d0180171 | 3635 | drbd_err(device, "kmalloc of fifo_buffer failed"); |
b30ab791 | 3636 | put_ldev(device); |
778f271d PR |
3637 | goto disconnect; |
3638 | } | |
3639 | } | |
8e26f9cc | 3640 | } |
b411b363 | 3641 | |
91fd4dad | 3642 | if (verify_tfm || csums_tfm) { |
2ec91e0e PR |
3643 | new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL); |
3644 | if (!new_net_conf) { | |
d0180171 | 3645 | drbd_err(device, "Allocation of new net_conf failed\n"); |
91fd4dad PR |
3646 | goto disconnect; |
3647 | } | |
3648 | ||
2ec91e0e | 3649 | *new_net_conf = *old_net_conf; |
91fd4dad PR |
3650 | |
3651 | if (verify_tfm) { | |
2ec91e0e PR |
3652 | strcpy(new_net_conf->verify_alg, p->verify_alg); |
3653 | new_net_conf->verify_alg_len = strlen(p->verify_alg) + 1; | |
9f4fe9ad AG |
3654 | crypto_free_hash(peer_device->connection->verify_tfm); |
3655 | peer_device->connection->verify_tfm = verify_tfm; | |
d0180171 | 3656 | drbd_info(device, "using verify-alg: \"%s\"\n", p->verify_alg); |
91fd4dad PR |
3657 | } |
3658 | if (csums_tfm) { | |
2ec91e0e PR |
3659 | strcpy(new_net_conf->csums_alg, p->csums_alg); |
3660 | new_net_conf->csums_alg_len = strlen(p->csums_alg) + 1; | |
9f4fe9ad AG |
3661 | crypto_free_hash(peer_device->connection->csums_tfm); |
3662 | peer_device->connection->csums_tfm = csums_tfm; | |
d0180171 | 3663 | drbd_info(device, "using csums-alg: \"%s\"\n", p->csums_alg); |
91fd4dad | 3664 | } |
bde89a9e | 3665 | rcu_assign_pointer(connection->net_conf, new_net_conf); |
778f271d | 3666 | } |
b411b363 PR |
3667 | } |
3668 | ||
813472ce | 3669 | if (new_disk_conf) { |
b30ab791 AG |
3670 | rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf); |
3671 | put_ldev(device); | |
813472ce PR |
3672 | } |
3673 | ||
3674 | if (new_plan) { | |
b30ab791 AG |
3675 | old_plan = device->rs_plan_s; |
3676 | rcu_assign_pointer(device->rs_plan_s, new_plan); | |
b411b363 | 3677 | } |
daeda1cc | 3678 | |
0500813f | 3679 | mutex_unlock(&connection->resource->conf_update); |
daeda1cc PR |
3680 | synchronize_rcu(); |
3681 | if (new_net_conf) | |
3682 | kfree(old_net_conf); | |
3683 | kfree(old_disk_conf); | |
813472ce | 3684 | kfree(old_plan); |
daeda1cc | 3685 | |
82bc0194 | 3686 | return 0; |
b411b363 | 3687 | |
813472ce PR |
3688 | reconnect: |
3689 | if (new_disk_conf) { | |
b30ab791 | 3690 | put_ldev(device); |
813472ce PR |
3691 | kfree(new_disk_conf); |
3692 | } | |
0500813f | 3693 | mutex_unlock(&connection->resource->conf_update); |
813472ce PR |
3694 | return -EIO; |
3695 | ||
b411b363 | 3696 | disconnect: |
813472ce PR |
3697 | kfree(new_plan); |
3698 | if (new_disk_conf) { | |
b30ab791 | 3699 | put_ldev(device); |
813472ce PR |
3700 | kfree(new_disk_conf); |
3701 | } | |
0500813f | 3702 | mutex_unlock(&connection->resource->conf_update); |
b411b363 PR |
3703 | /* just for completeness: actually not needed, |
3704 | * as this is not reached if csums_tfm was ok. */ | |
3705 | crypto_free_hash(csums_tfm); | |
3706 | /* but free the verify_tfm again, if csums_tfm did not work out */ | |
3707 | crypto_free_hash(verify_tfm); | |
9f4fe9ad | 3708 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3709 | return -EIO; |
b411b363 PR |
3710 | } |
3711 | ||
b411b363 | 3712 | /* warn if the arguments differ by more than 12.5% */ |
b30ab791 | 3713 | static void warn_if_differ_considerably(struct drbd_device *device, |
b411b363 PR |
3714 | const char *s, sector_t a, sector_t b) |
3715 | { | |
3716 | sector_t d; | |
3717 | if (a == 0 || b == 0) | |
3718 | return; | |
3719 | d = (a > b) ? (a - b) : (b - a); | |
3720 | if (d > (a>>3) || d > (b>>3)) | |
d0180171 | 3721 | drbd_warn(device, "Considerable difference in %s: %llus vs. %llus\n", s, |
b411b363 PR |
3722 | (unsigned long long)a, (unsigned long long)b); |
3723 | } | |
3724 | ||
bde89a9e | 3725 | static int receive_sizes(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3726 | { |
9f4fe9ad | 3727 | struct drbd_peer_device *peer_device; |
b30ab791 | 3728 | struct drbd_device *device; |
e658983a | 3729 | struct p_sizes *p = pi->data; |
e96c9633 | 3730 | enum determine_dev_size dd = DS_UNCHANGED; |
6a8d68b1 | 3731 | sector_t p_size, p_usize, p_csize, my_usize; |
b411b363 | 3732 | int ldsc = 0; /* local disk size changed */ |
e89b591c | 3733 | enum dds_flags ddsf; |
b411b363 | 3734 | |
9f4fe9ad AG |
3735 | peer_device = conn_peer_device(connection, pi->vnr); |
3736 | if (!peer_device) | |
bde89a9e | 3737 | return config_unknown_volume(connection, pi); |
9f4fe9ad | 3738 | device = peer_device->device; |
4a76b161 | 3739 | |
b411b363 PR |
3740 | p_size = be64_to_cpu(p->d_size); |
3741 | p_usize = be64_to_cpu(p->u_size); | |
6a8d68b1 | 3742 | p_csize = be64_to_cpu(p->c_size); |
b411b363 | 3743 | |
b411b363 PR |
3744 | /* just store the peer's disk size for now. |
3745 | * we still need to figure out whether we accept that. */ | |
b30ab791 | 3746 | device->p_size = p_size; |
b411b363 | 3747 | |
b30ab791 | 3748 | if (get_ldev(device)) { |
daeda1cc | 3749 | rcu_read_lock(); |
b30ab791 | 3750 | my_usize = rcu_dereference(device->ldev->disk_conf)->disk_size; |
daeda1cc PR |
3751 | rcu_read_unlock(); |
3752 | ||
b30ab791 AG |
3753 | warn_if_differ_considerably(device, "lower level device sizes", |
3754 | p_size, drbd_get_max_capacity(device->ldev)); | |
3755 | warn_if_differ_considerably(device, "user requested size", | |
daeda1cc | 3756 | p_usize, my_usize); |
b411b363 PR |
3757 | |
3758 | /* if this is the first connect, or an otherwise expected | |
3759 | * param exchange, choose the minimum */ | |
b30ab791 | 3760 | if (device->state.conn == C_WF_REPORT_PARAMS) |
daeda1cc | 3761 | p_usize = min_not_zero(my_usize, p_usize); |
b411b363 PR |
3762 | |
3763 | /* Never shrink a device with usable data during connect. | |
3764 | But allow online shrinking if we are connected. */ | |
b30ab791 AG |
3765 | if (drbd_new_dev_size(device, device->ldev, p_usize, 0) < |
3766 | drbd_get_capacity(device->this_bdev) && | |
3767 | device->state.disk >= D_OUTDATED && | |
3768 | device->state.conn < C_CONNECTED) { | |
d0180171 | 3769 | drbd_err(device, "The peer's disk size is too small!\n"); |
9f4fe9ad | 3770 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b30ab791 | 3771 | put_ldev(device); |
82bc0194 | 3772 | return -EIO; |
b411b363 | 3773 | } |
daeda1cc PR |
3774 | |
3775 | if (my_usize != p_usize) { | |
3776 | struct disk_conf *old_disk_conf, *new_disk_conf = NULL; | |
3777 | ||
3778 | new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); | |
3779 | if (!new_disk_conf) { | |
d0180171 | 3780 | drbd_err(device, "Allocation of new disk_conf failed\n"); |
b30ab791 | 3781 | put_ldev(device); |
daeda1cc PR |
3782 | return -ENOMEM; |
3783 | } | |
3784 | ||
0500813f | 3785 | mutex_lock(&connection->resource->conf_update); |
b30ab791 | 3786 | old_disk_conf = device->ldev->disk_conf; |
daeda1cc PR |
3787 | *new_disk_conf = *old_disk_conf; |
3788 | new_disk_conf->disk_size = p_usize; | |
3789 | ||
b30ab791 | 3790 | rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf); |
0500813f | 3791 | mutex_unlock(&connection->resource->conf_update); |
daeda1cc PR |
3792 | synchronize_rcu(); |
3793 | kfree(old_disk_conf); | |
3794 | ||
d0180171 | 3795 | drbd_info(device, "Peer sets u_size to %lu sectors\n", |
daeda1cc | 3796 | (unsigned long)my_usize); |
b411b363 | 3797 | } |
daeda1cc | 3798 | |
b30ab791 | 3799 | put_ldev(device); |
b411b363 | 3800 | } |
b411b363 | 3801 | |
20c68fde | 3802 | device->peer_max_bio_size = be32_to_cpu(p->max_bio_size); |
20c68fde LE |
3803 | /* Leave drbd_reconsider_max_bio_size() before drbd_determine_dev_size(). |
3804 | In case we cleared the QUEUE_FLAG_DISCARD from our queue in | |
3805 | drbd_reconsider_max_bio_size(), we can be sure that after | |
3806 | drbd_determine_dev_size() no REQ_DISCARDs are in the queue. */ | |
3807 | ||
e89b591c | 3808 | ddsf = be16_to_cpu(p->dds_flags); |
b30ab791 | 3809 | if (get_ldev(device)) { |
8fe39aac | 3810 | drbd_reconsider_max_bio_size(device, device->ldev); |
b30ab791 AG |
3811 | dd = drbd_determine_dev_size(device, ddsf, NULL); |
3812 | put_ldev(device); | |
e96c9633 | 3813 | if (dd == DS_ERROR) |
82bc0194 | 3814 | return -EIO; |
b30ab791 | 3815 | drbd_md_sync(device); |
b411b363 | 3816 | } else { |
6a8d68b1 LE |
3817 | /* |
3818 | * I am diskless, need to accept the peer's *current* size. | |
3819 | * I must NOT accept the peers backing disk size, | |
3820 | * it may have been larger than mine all along... | |
3821 | * | |
3822 | * At this point, the peer knows more about my disk, or at | |
3823 | * least about what we last agreed upon, than myself. | |
3824 | * So if his c_size is less than his d_size, the most likely | |
3825 | * reason is that *my* d_size was smaller last time we checked. | |
3826 | * | |
3827 | * However, if he sends a zero current size, | |
3828 | * take his (user-capped or) backing disk size anyways. | |
3829 | */ | |
8fe39aac | 3830 | drbd_reconsider_max_bio_size(device, NULL); |
6a8d68b1 | 3831 | drbd_set_my_capacity(device, p_csize ?: p_usize ?: p_size); |
b411b363 PR |
3832 | } |
3833 | ||
b30ab791 AG |
3834 | if (get_ldev(device)) { |
3835 | if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev)) { | |
3836 | device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev); | |
b411b363 PR |
3837 | ldsc = 1; |
3838 | } | |
3839 | ||
b30ab791 | 3840 | put_ldev(device); |
b411b363 PR |
3841 | } |
3842 | ||
b30ab791 | 3843 | if (device->state.conn > C_WF_REPORT_PARAMS) { |
b411b363 | 3844 | if (be64_to_cpu(p->c_size) != |
b30ab791 | 3845 | drbd_get_capacity(device->this_bdev) || ldsc) { |
b411b363 PR |
3846 | /* we have different sizes, probably peer |
3847 | * needs to know my new size... */ | |
69a22773 | 3848 | drbd_send_sizes(peer_device, 0, ddsf); |
b411b363 | 3849 | } |
b30ab791 AG |
3850 | if (test_and_clear_bit(RESIZE_PENDING, &device->flags) || |
3851 | (dd == DS_GREW && device->state.conn == C_CONNECTED)) { | |
3852 | if (device->state.pdsk >= D_INCONSISTENT && | |
3853 | device->state.disk >= D_INCONSISTENT) { | |
e89b591c | 3854 | if (ddsf & DDSF_NO_RESYNC) |
d0180171 | 3855 | drbd_info(device, "Resync of new storage suppressed with --assume-clean\n"); |
e89b591c | 3856 | else |
b30ab791 | 3857 | resync_after_online_grow(device); |
e89b591c | 3858 | } else |
b30ab791 | 3859 | set_bit(RESYNC_AFTER_NEG, &device->flags); |
b411b363 PR |
3860 | } |
3861 | } | |
3862 | ||
82bc0194 | 3863 | return 0; |
b411b363 PR |
3864 | } |
3865 | ||
bde89a9e | 3866 | static int receive_uuids(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3867 | { |
9f4fe9ad | 3868 | struct drbd_peer_device *peer_device; |
b30ab791 | 3869 | struct drbd_device *device; |
e658983a | 3870 | struct p_uuids *p = pi->data; |
b411b363 | 3871 | u64 *p_uuid; |
62b0da3a | 3872 | int i, updated_uuids = 0; |
b411b363 | 3873 | |
9f4fe9ad AG |
3874 | peer_device = conn_peer_device(connection, pi->vnr); |
3875 | if (!peer_device) | |
bde89a9e | 3876 | return config_unknown_volume(connection, pi); |
9f4fe9ad | 3877 | device = peer_device->device; |
4a76b161 | 3878 | |
b411b363 | 3879 | p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); |
063eacf8 | 3880 | if (!p_uuid) { |
d0180171 | 3881 | drbd_err(device, "kmalloc of p_uuid failed\n"); |
063eacf8 JW |
3882 | return false; |
3883 | } | |
b411b363 PR |
3884 | |
3885 | for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) | |
3886 | p_uuid[i] = be64_to_cpu(p->uuid[i]); | |
3887 | ||
b30ab791 AG |
3888 | kfree(device->p_uuid); |
3889 | device->p_uuid = p_uuid; | |
b411b363 | 3890 | |
b30ab791 AG |
3891 | if (device->state.conn < C_CONNECTED && |
3892 | device->state.disk < D_INCONSISTENT && | |
3893 | device->state.role == R_PRIMARY && | |
3894 | (device->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { | |
d0180171 | 3895 | drbd_err(device, "Can only connect to data with current UUID=%016llX\n", |
b30ab791 | 3896 | (unsigned long long)device->ed_uuid); |
9f4fe9ad | 3897 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3898 | return -EIO; |
b411b363 PR |
3899 | } |
3900 | ||
b30ab791 | 3901 | if (get_ldev(device)) { |
b411b363 | 3902 | int skip_initial_sync = |
b30ab791 | 3903 | device->state.conn == C_CONNECTED && |
9f4fe9ad | 3904 | peer_device->connection->agreed_pro_version >= 90 && |
b30ab791 | 3905 | device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && |
b411b363 PR |
3906 | (p_uuid[UI_FLAGS] & 8); |
3907 | if (skip_initial_sync) { | |
d0180171 | 3908 | drbd_info(device, "Accepted new current UUID, preparing to skip initial sync\n"); |
b30ab791 | 3909 | drbd_bitmap_io(device, &drbd_bmio_clear_n_write, |
20ceb2b2 LE |
3910 | "clear_n_write from receive_uuids", |
3911 | BM_LOCKED_TEST_ALLOWED); | |
b30ab791 AG |
3912 | _drbd_uuid_set(device, UI_CURRENT, p_uuid[UI_CURRENT]); |
3913 | _drbd_uuid_set(device, UI_BITMAP, 0); | |
3914 | _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), | |
b411b363 | 3915 | CS_VERBOSE, NULL); |
b30ab791 | 3916 | drbd_md_sync(device); |
62b0da3a | 3917 | updated_uuids = 1; |
b411b363 | 3918 | } |
b30ab791 AG |
3919 | put_ldev(device); |
3920 | } else if (device->state.disk < D_INCONSISTENT && | |
3921 | device->state.role == R_PRIMARY) { | |
18a50fa2 PR |
3922 | /* I am a diskless primary, the peer just created a new current UUID |
3923 | for me. */ | |
b30ab791 | 3924 | updated_uuids = drbd_set_ed_uuid(device, p_uuid[UI_CURRENT]); |
b411b363 PR |
3925 | } |
3926 | ||
3927 | /* Before we test for the disk state, we should wait until an eventually | |
3928 | ongoing cluster wide state change is finished. That is important if | |
3929 | we are primary and are detaching from our disk. We need to see the | |
3930 | new disk state... */ | |
b30ab791 AG |
3931 | mutex_lock(device->state_mutex); |
3932 | mutex_unlock(device->state_mutex); | |
3933 | if (device->state.conn >= C_CONNECTED && device->state.disk < D_INCONSISTENT) | |
3934 | updated_uuids |= drbd_set_ed_uuid(device, p_uuid[UI_CURRENT]); | |
62b0da3a LE |
3935 | |
3936 | if (updated_uuids) | |
b30ab791 | 3937 | drbd_print_uuids(device, "receiver updated UUIDs to"); |
b411b363 | 3938 | |
82bc0194 | 3939 | return 0; |
b411b363 PR |
3940 | } |
3941 | ||
3942 | /** | |
3943 | * convert_state() - Converts the peer's view of the cluster state to our point of view | |
3944 | * @ps: The state as seen by the peer. | |
3945 | */ | |
3946 | static union drbd_state convert_state(union drbd_state ps) | |
3947 | { | |
3948 | union drbd_state ms; | |
3949 | ||
3950 | static enum drbd_conns c_tab[] = { | |
369bea63 | 3951 | [C_WF_REPORT_PARAMS] = C_WF_REPORT_PARAMS, |
b411b363 PR |
3952 | [C_CONNECTED] = C_CONNECTED, |
3953 | ||
3954 | [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, | |
3955 | [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, | |
3956 | [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ | |
3957 | [C_VERIFY_S] = C_VERIFY_T, | |
3958 | [C_MASK] = C_MASK, | |
3959 | }; | |
3960 | ||
3961 | ms.i = ps.i; | |
3962 | ||
3963 | ms.conn = c_tab[ps.conn]; | |
3964 | ms.peer = ps.role; | |
3965 | ms.role = ps.peer; | |
3966 | ms.pdsk = ps.disk; | |
3967 | ms.disk = ps.pdsk; | |
3968 | ms.peer_isp = (ps.aftr_isp | ps.user_isp); | |
3969 | ||
3970 | return ms; | |
3971 | } | |
3972 | ||
bde89a9e | 3973 | static int receive_req_state(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 3974 | { |
9f4fe9ad | 3975 | struct drbd_peer_device *peer_device; |
b30ab791 | 3976 | struct drbd_device *device; |
e658983a | 3977 | struct p_req_state *p = pi->data; |
b411b363 | 3978 | union drbd_state mask, val; |
bf885f8a | 3979 | enum drbd_state_rv rv; |
b411b363 | 3980 | |
9f4fe9ad AG |
3981 | peer_device = conn_peer_device(connection, pi->vnr); |
3982 | if (!peer_device) | |
4a76b161 | 3983 | return -EIO; |
9f4fe9ad | 3984 | device = peer_device->device; |
4a76b161 | 3985 | |
b411b363 PR |
3986 | mask.i = be32_to_cpu(p->mask); |
3987 | val.i = be32_to_cpu(p->val); | |
3988 | ||
9f4fe9ad | 3989 | if (test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags) && |
b30ab791 | 3990 | mutex_is_locked(device->state_mutex)) { |
69a22773 | 3991 | drbd_send_sr_reply(peer_device, SS_CONCURRENT_ST_CHG); |
82bc0194 | 3992 | return 0; |
b411b363 PR |
3993 | } |
3994 | ||
3995 | mask = convert_state(mask); | |
3996 | val = convert_state(val); | |
3997 | ||
b30ab791 | 3998 | rv = drbd_change_state(device, CS_VERBOSE, mask, val); |
69a22773 | 3999 | drbd_send_sr_reply(peer_device, rv); |
b411b363 | 4000 | |
b30ab791 | 4001 | drbd_md_sync(device); |
b411b363 | 4002 | |
82bc0194 | 4003 | return 0; |
b411b363 PR |
4004 | } |
4005 | ||
bde89a9e | 4006 | static int receive_req_conn_state(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4007 | { |
e658983a | 4008 | struct p_req_state *p = pi->data; |
b411b363 | 4009 | union drbd_state mask, val; |
bf885f8a | 4010 | enum drbd_state_rv rv; |
b411b363 | 4011 | |
b411b363 PR |
4012 | mask.i = be32_to_cpu(p->mask); |
4013 | val.i = be32_to_cpu(p->val); | |
4014 | ||
bde89a9e AG |
4015 | if (test_bit(RESOLVE_CONFLICTS, &connection->flags) && |
4016 | mutex_is_locked(&connection->cstate_mutex)) { | |
4017 | conn_send_sr_reply(connection, SS_CONCURRENT_ST_CHG); | |
82bc0194 | 4018 | return 0; |
b411b363 PR |
4019 | } |
4020 | ||
4021 | mask = convert_state(mask); | |
4022 | val = convert_state(val); | |
4023 | ||
bde89a9e AG |
4024 | rv = conn_request_state(connection, mask, val, CS_VERBOSE | CS_LOCAL_ONLY | CS_IGN_OUTD_FAIL); |
4025 | conn_send_sr_reply(connection, rv); | |
b411b363 | 4026 | |
82bc0194 | 4027 | return 0; |
b411b363 PR |
4028 | } |
4029 | ||
bde89a9e | 4030 | static int receive_state(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4031 | { |
9f4fe9ad | 4032 | struct drbd_peer_device *peer_device; |
b30ab791 | 4033 | struct drbd_device *device; |
e658983a | 4034 | struct p_state *p = pi->data; |
4ac4aada | 4035 | union drbd_state os, ns, peer_state; |
b411b363 | 4036 | enum drbd_disk_state real_peer_disk; |
65d922c3 | 4037 | enum chg_state_flags cs_flags; |
b411b363 PR |
4038 | int rv; |
4039 | ||
9f4fe9ad AG |
4040 | peer_device = conn_peer_device(connection, pi->vnr); |
4041 | if (!peer_device) | |
bde89a9e | 4042 | return config_unknown_volume(connection, pi); |
9f4fe9ad | 4043 | device = peer_device->device; |
4a76b161 | 4044 | |
b411b363 PR |
4045 | peer_state.i = be32_to_cpu(p->state); |
4046 | ||
4047 | real_peer_disk = peer_state.disk; | |
4048 | if (peer_state.disk == D_NEGOTIATING) { | |
b30ab791 | 4049 | real_peer_disk = device->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; |
d0180171 | 4050 | drbd_info(device, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); |
b411b363 PR |
4051 | } |
4052 | ||
0500813f | 4053 | spin_lock_irq(&device->resource->req_lock); |
b411b363 | 4054 | retry: |
b30ab791 | 4055 | os = ns = drbd_read_state(device); |
0500813f | 4056 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 4057 | |
545752d5 LE |
4058 | /* If some other part of the code (asender thread, timeout) |
4059 | * already decided to close the connection again, | |
4060 | * we must not "re-establish" it here. */ | |
4061 | if (os.conn <= C_TEAR_DOWN) | |
58ffa580 | 4062 | return -ECONNRESET; |
545752d5 | 4063 | |
40424e4a LE |
4064 | /* If this is the "end of sync" confirmation, usually the peer disk |
4065 | * transitions from D_INCONSISTENT to D_UP_TO_DATE. For empty (0 bits | |
4066 | * set) resync started in PausedSyncT, or if the timing of pause-/ | |
4067 | * unpause-sync events has been "just right", the peer disk may | |
4068 | * transition from D_CONSISTENT to D_UP_TO_DATE as well. | |
4069 | */ | |
4070 | if ((os.pdsk == D_INCONSISTENT || os.pdsk == D_CONSISTENT) && | |
4071 | real_peer_disk == D_UP_TO_DATE && | |
e9ef7bb6 LE |
4072 | os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { |
4073 | /* If we are (becoming) SyncSource, but peer is still in sync | |
4074 | * preparation, ignore its uptodate-ness to avoid flapping, it | |
4075 | * will change to inconsistent once the peer reaches active | |
4076 | * syncing states. | |
4077 | * It may have changed syncer-paused flags, however, so we | |
4078 | * cannot ignore this completely. */ | |
4079 | if (peer_state.conn > C_CONNECTED && | |
4080 | peer_state.conn < C_SYNC_SOURCE) | |
4081 | real_peer_disk = D_INCONSISTENT; | |
4082 | ||
4083 | /* if peer_state changes to connected at the same time, | |
4084 | * it explicitly notifies us that it finished resync. | |
4085 | * Maybe we should finish it up, too? */ | |
4086 | else if (os.conn >= C_SYNC_SOURCE && | |
4087 | peer_state.conn == C_CONNECTED) { | |
b30ab791 AG |
4088 | if (drbd_bm_total_weight(device) <= device->rs_failed) |
4089 | drbd_resync_finished(device); | |
82bc0194 | 4090 | return 0; |
e9ef7bb6 LE |
4091 | } |
4092 | } | |
4093 | ||
02b91b55 LE |
4094 | /* explicit verify finished notification, stop sector reached. */ |
4095 | if (os.conn == C_VERIFY_T && os.disk == D_UP_TO_DATE && | |
4096 | peer_state.conn == C_CONNECTED && real_peer_disk == D_UP_TO_DATE) { | |
b30ab791 AG |
4097 | ov_out_of_sync_print(device); |
4098 | drbd_resync_finished(device); | |
58ffa580 | 4099 | return 0; |
02b91b55 LE |
4100 | } |
4101 | ||
e9ef7bb6 LE |
4102 | /* peer says his disk is inconsistent, while we think it is uptodate, |
4103 | * and this happens while the peer still thinks we have a sync going on, | |
4104 | * but we think we are already done with the sync. | |
4105 | * We ignore this to avoid flapping pdsk. | |
4106 | * This should not happen, if the peer is a recent version of drbd. */ | |
4107 | if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && | |
4108 | os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) | |
4109 | real_peer_disk = D_UP_TO_DATE; | |
4110 | ||
4ac4aada LE |
4111 | if (ns.conn == C_WF_REPORT_PARAMS) |
4112 | ns.conn = C_CONNECTED; | |
b411b363 | 4113 | |
67531718 PR |
4114 | if (peer_state.conn == C_AHEAD) |
4115 | ns.conn = C_BEHIND; | |
4116 | ||
b30ab791 AG |
4117 | if (device->p_uuid && peer_state.disk >= D_NEGOTIATING && |
4118 | get_ldev_if_state(device, D_NEGOTIATING)) { | |
b411b363 PR |
4119 | int cr; /* consider resync */ |
4120 | ||
4121 | /* if we established a new connection */ | |
4ac4aada | 4122 | cr = (os.conn < C_CONNECTED); |
b411b363 PR |
4123 | /* if we had an established connection |
4124 | * and one of the nodes newly attaches a disk */ | |
4ac4aada | 4125 | cr |= (os.conn == C_CONNECTED && |
b411b363 | 4126 | (peer_state.disk == D_NEGOTIATING || |
4ac4aada | 4127 | os.disk == D_NEGOTIATING)); |
b411b363 PR |
4128 | /* if we have both been inconsistent, and the peer has been |
4129 | * forced to be UpToDate with --overwrite-data */ | |
b30ab791 | 4130 | cr |= test_bit(CONSIDER_RESYNC, &device->flags); |
b411b363 PR |
4131 | /* if we had been plain connected, and the admin requested to |
4132 | * start a sync by "invalidate" or "invalidate-remote" */ | |
4ac4aada | 4133 | cr |= (os.conn == C_CONNECTED && |
b411b363 PR |
4134 | (peer_state.conn >= C_STARTING_SYNC_S && |
4135 | peer_state.conn <= C_WF_BITMAP_T)); | |
4136 | ||
4137 | if (cr) | |
69a22773 | 4138 | ns.conn = drbd_sync_handshake(peer_device, peer_state.role, real_peer_disk); |
b411b363 | 4139 | |
b30ab791 | 4140 | put_ldev(device); |
4ac4aada LE |
4141 | if (ns.conn == C_MASK) { |
4142 | ns.conn = C_CONNECTED; | |
b30ab791 AG |
4143 | if (device->state.disk == D_NEGOTIATING) { |
4144 | drbd_force_state(device, NS(disk, D_FAILED)); | |
b411b363 | 4145 | } else if (peer_state.disk == D_NEGOTIATING) { |
d0180171 | 4146 | drbd_err(device, "Disk attach process on the peer node was aborted.\n"); |
b411b363 | 4147 | peer_state.disk = D_DISKLESS; |
580b9767 | 4148 | real_peer_disk = D_DISKLESS; |
b411b363 | 4149 | } else { |
9f4fe9ad | 4150 | if (test_and_clear_bit(CONN_DRY_RUN, &peer_device->connection->flags)) |
82bc0194 | 4151 | return -EIO; |
0b0ba1ef | 4152 | D_ASSERT(device, os.conn == C_WF_REPORT_PARAMS); |
9f4fe9ad | 4153 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 4154 | return -EIO; |
b411b363 PR |
4155 | } |
4156 | } | |
4157 | } | |
4158 | ||
0500813f | 4159 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 4160 | if (os.i != drbd_read_state(device).i) |
b411b363 | 4161 | goto retry; |
b30ab791 | 4162 | clear_bit(CONSIDER_RESYNC, &device->flags); |
b411b363 PR |
4163 | ns.peer = peer_state.role; |
4164 | ns.pdsk = real_peer_disk; | |
4165 | ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); | |
4ac4aada | 4166 | if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) |
b30ab791 | 4167 | ns.disk = device->new_state_tmp.disk; |
4ac4aada | 4168 | cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); |
b30ab791 AG |
4169 | if (ns.pdsk == D_CONSISTENT && drbd_suspended(device) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && |
4170 | test_bit(NEW_CUR_UUID, &device->flags)) { | |
8554df1c | 4171 | /* Do not allow tl_restart(RESEND) for a rebooted peer. We can only allow this |
481c6f50 | 4172 | for temporal network outages! */ |
0500813f | 4173 | spin_unlock_irq(&device->resource->req_lock); |
d0180171 | 4174 | drbd_err(device, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); |
9f4fe9ad | 4175 | tl_clear(peer_device->connection); |
b30ab791 AG |
4176 | drbd_uuid_new_current(device); |
4177 | clear_bit(NEW_CUR_UUID, &device->flags); | |
9f4fe9ad | 4178 | conn_request_state(peer_device->connection, NS2(conn, C_PROTOCOL_ERROR, susp, 0), CS_HARD); |
82bc0194 | 4179 | return -EIO; |
481c6f50 | 4180 | } |
b30ab791 AG |
4181 | rv = _drbd_set_state(device, ns, cs_flags, NULL); |
4182 | ns = drbd_read_state(device); | |
0500813f | 4183 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
4184 | |
4185 | if (rv < SS_SUCCESS) { | |
9f4fe9ad | 4186 | conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 4187 | return -EIO; |
b411b363 PR |
4188 | } |
4189 | ||
4ac4aada LE |
4190 | if (os.conn > C_WF_REPORT_PARAMS) { |
4191 | if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && | |
b411b363 PR |
4192 | peer_state.disk != D_NEGOTIATING ) { |
4193 | /* we want resync, peer has not yet decided to sync... */ | |
4194 | /* Nowadays only used when forcing a node into primary role and | |
4195 | setting its disk to UpToDate with that */ | |
69a22773 AG |
4196 | drbd_send_uuids(peer_device); |
4197 | drbd_send_current_state(peer_device); | |
b411b363 PR |
4198 | } |
4199 | } | |
4200 | ||
b30ab791 | 4201 | clear_bit(DISCARD_MY_DATA, &device->flags); |
b411b363 | 4202 | |
b30ab791 | 4203 | drbd_md_sync(device); /* update connected indicator, la_size_sect, ... */ |
b411b363 | 4204 | |
82bc0194 | 4205 | return 0; |
b411b363 PR |
4206 | } |
4207 | ||
bde89a9e | 4208 | static int receive_sync_uuid(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4209 | { |
9f4fe9ad | 4210 | struct drbd_peer_device *peer_device; |
b30ab791 | 4211 | struct drbd_device *device; |
e658983a | 4212 | struct p_rs_uuid *p = pi->data; |
4a76b161 | 4213 | |
9f4fe9ad AG |
4214 | peer_device = conn_peer_device(connection, pi->vnr); |
4215 | if (!peer_device) | |
4a76b161 | 4216 | return -EIO; |
9f4fe9ad | 4217 | device = peer_device->device; |
b411b363 | 4218 | |
b30ab791 AG |
4219 | wait_event(device->misc_wait, |
4220 | device->state.conn == C_WF_SYNC_UUID || | |
4221 | device->state.conn == C_BEHIND || | |
4222 | device->state.conn < C_CONNECTED || | |
4223 | device->state.disk < D_NEGOTIATING); | |
b411b363 | 4224 | |
0b0ba1ef | 4225 | /* D_ASSERT(device, device->state.conn == C_WF_SYNC_UUID ); */ |
b411b363 | 4226 | |
b411b363 PR |
4227 | /* Here the _drbd_uuid_ functions are right, current should |
4228 | _not_ be rotated into the history */ | |
b30ab791 AG |
4229 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
4230 | _drbd_uuid_set(device, UI_CURRENT, be64_to_cpu(p->uuid)); | |
4231 | _drbd_uuid_set(device, UI_BITMAP, 0UL); | |
b411b363 | 4232 | |
b30ab791 AG |
4233 | drbd_print_uuids(device, "updated sync uuid"); |
4234 | drbd_start_resync(device, C_SYNC_TARGET); | |
b411b363 | 4235 | |
b30ab791 | 4236 | put_ldev(device); |
b411b363 | 4237 | } else |
d0180171 | 4238 | drbd_err(device, "Ignoring SyncUUID packet!\n"); |
b411b363 | 4239 | |
82bc0194 | 4240 | return 0; |
b411b363 PR |
4241 | } |
4242 | ||
2c46407d AG |
4243 | /** |
4244 | * receive_bitmap_plain | |
4245 | * | |
4246 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4247 | * code upon failure. | |
4248 | */ | |
4249 | static int | |
69a22773 | 4250 | receive_bitmap_plain(struct drbd_peer_device *peer_device, unsigned int size, |
e658983a | 4251 | unsigned long *p, struct bm_xfer_ctx *c) |
b411b363 | 4252 | { |
50d0b1ad | 4253 | unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - |
69a22773 | 4254 | drbd_header_size(peer_device->connection); |
e658983a | 4255 | unsigned int num_words = min_t(size_t, data_size / sizeof(*p), |
50d0b1ad | 4256 | c->bm_words - c->word_offset); |
e658983a | 4257 | unsigned int want = num_words * sizeof(*p); |
2c46407d | 4258 | int err; |
b411b363 | 4259 | |
50d0b1ad | 4260 | if (want != size) { |
69a22773 | 4261 | drbd_err(peer_device, "%s:want (%u) != size (%u)\n", __func__, want, size); |
2c46407d | 4262 | return -EIO; |
b411b363 PR |
4263 | } |
4264 | if (want == 0) | |
2c46407d | 4265 | return 0; |
69a22773 | 4266 | err = drbd_recv_all(peer_device->connection, p, want); |
82bc0194 | 4267 | if (err) |
2c46407d | 4268 | return err; |
b411b363 | 4269 | |
69a22773 | 4270 | drbd_bm_merge_lel(peer_device->device, c->word_offset, num_words, p); |
b411b363 PR |
4271 | |
4272 | c->word_offset += num_words; | |
4273 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
4274 | if (c->bit_offset > c->bm_bits) | |
4275 | c->bit_offset = c->bm_bits; | |
4276 | ||
2c46407d | 4277 | return 1; |
b411b363 PR |
4278 | } |
4279 | ||
a02d1240 AG |
4280 | static enum drbd_bitmap_code dcbp_get_code(struct p_compressed_bm *p) |
4281 | { | |
4282 | return (enum drbd_bitmap_code)(p->encoding & 0x0f); | |
4283 | } | |
4284 | ||
4285 | static int dcbp_get_start(struct p_compressed_bm *p) | |
4286 | { | |
4287 | return (p->encoding & 0x80) != 0; | |
4288 | } | |
4289 | ||
4290 | static int dcbp_get_pad_bits(struct p_compressed_bm *p) | |
4291 | { | |
4292 | return (p->encoding >> 4) & 0x7; | |
4293 | } | |
4294 | ||
2c46407d AG |
4295 | /** |
4296 | * recv_bm_rle_bits | |
4297 | * | |
4298 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4299 | * code upon failure. | |
4300 | */ | |
4301 | static int | |
69a22773 | 4302 | recv_bm_rle_bits(struct drbd_peer_device *peer_device, |
b411b363 | 4303 | struct p_compressed_bm *p, |
c6d25cfe PR |
4304 | struct bm_xfer_ctx *c, |
4305 | unsigned int len) | |
b411b363 PR |
4306 | { |
4307 | struct bitstream bs; | |
4308 | u64 look_ahead; | |
4309 | u64 rl; | |
4310 | u64 tmp; | |
4311 | unsigned long s = c->bit_offset; | |
4312 | unsigned long e; | |
a02d1240 | 4313 | int toggle = dcbp_get_start(p); |
b411b363 PR |
4314 | int have; |
4315 | int bits; | |
4316 | ||
a02d1240 | 4317 | bitstream_init(&bs, p->code, len, dcbp_get_pad_bits(p)); |
b411b363 PR |
4318 | |
4319 | bits = bitstream_get_bits(&bs, &look_ahead, 64); | |
4320 | if (bits < 0) | |
2c46407d | 4321 | return -EIO; |
b411b363 PR |
4322 | |
4323 | for (have = bits; have > 0; s += rl, toggle = !toggle) { | |
4324 | bits = vli_decode_bits(&rl, look_ahead); | |
4325 | if (bits <= 0) | |
2c46407d | 4326 | return -EIO; |
b411b363 PR |
4327 | |
4328 | if (toggle) { | |
4329 | e = s + rl -1; | |
4330 | if (e >= c->bm_bits) { | |
69a22773 | 4331 | drbd_err(peer_device, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); |
2c46407d | 4332 | return -EIO; |
b411b363 | 4333 | } |
69a22773 | 4334 | _drbd_bm_set_bits(peer_device->device, s, e); |
b411b363 PR |
4335 | } |
4336 | ||
4337 | if (have < bits) { | |
69a22773 | 4338 | drbd_err(peer_device, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", |
b411b363 PR |
4339 | have, bits, look_ahead, |
4340 | (unsigned int)(bs.cur.b - p->code), | |
4341 | (unsigned int)bs.buf_len); | |
2c46407d | 4342 | return -EIO; |
b411b363 | 4343 | } |
d2da5b0c LE |
4344 | /* if we consumed all 64 bits, assign 0; >> 64 is "undefined"; */ |
4345 | if (likely(bits < 64)) | |
4346 | look_ahead >>= bits; | |
4347 | else | |
4348 | look_ahead = 0; | |
b411b363 PR |
4349 | have -= bits; |
4350 | ||
4351 | bits = bitstream_get_bits(&bs, &tmp, 64 - have); | |
4352 | if (bits < 0) | |
2c46407d | 4353 | return -EIO; |
b411b363 PR |
4354 | look_ahead |= tmp << have; |
4355 | have += bits; | |
4356 | } | |
4357 | ||
4358 | c->bit_offset = s; | |
4359 | bm_xfer_ctx_bit_to_word_offset(c); | |
4360 | ||
2c46407d | 4361 | return (s != c->bm_bits); |
b411b363 PR |
4362 | } |
4363 | ||
2c46407d AG |
4364 | /** |
4365 | * decode_bitmap_c | |
4366 | * | |
4367 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
4368 | * code upon failure. | |
4369 | */ | |
4370 | static int | |
69a22773 | 4371 | decode_bitmap_c(struct drbd_peer_device *peer_device, |
b411b363 | 4372 | struct p_compressed_bm *p, |
c6d25cfe PR |
4373 | struct bm_xfer_ctx *c, |
4374 | unsigned int len) | |
b411b363 | 4375 | { |
a02d1240 | 4376 | if (dcbp_get_code(p) == RLE_VLI_Bits) |
69a22773 | 4377 | return recv_bm_rle_bits(peer_device, p, c, len - sizeof(*p)); |
b411b363 PR |
4378 | |
4379 | /* other variants had been implemented for evaluation, | |
4380 | * but have been dropped as this one turned out to be "best" | |
4381 | * during all our tests. */ | |
4382 | ||
69a22773 AG |
4383 | drbd_err(peer_device, "receive_bitmap_c: unknown encoding %u\n", p->encoding); |
4384 | conn_request_state(peer_device->connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); | |
2c46407d | 4385 | return -EIO; |
b411b363 PR |
4386 | } |
4387 | ||
b30ab791 | 4388 | void INFO_bm_xfer_stats(struct drbd_device *device, |
b411b363 PR |
4389 | const char *direction, struct bm_xfer_ctx *c) |
4390 | { | |
4391 | /* what would it take to transfer it "plaintext" */ | |
a6b32bc3 | 4392 | unsigned int header_size = drbd_header_size(first_peer_device(device)->connection); |
50d0b1ad AG |
4393 | unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; |
4394 | unsigned int plain = | |
4395 | header_size * (DIV_ROUND_UP(c->bm_words, data_size) + 1) + | |
4396 | c->bm_words * sizeof(unsigned long); | |
4397 | unsigned int total = c->bytes[0] + c->bytes[1]; | |
4398 | unsigned int r; | |
b411b363 PR |
4399 | |
4400 | /* total can not be zero. but just in case: */ | |
4401 | if (total == 0) | |
4402 | return; | |
4403 | ||
4404 | /* don't report if not compressed */ | |
4405 | if (total >= plain) | |
4406 | return; | |
4407 | ||
4408 | /* total < plain. check for overflow, still */ | |
4409 | r = (total > UINT_MAX/1000) ? (total / (plain/1000)) | |
4410 | : (1000 * total / plain); | |
4411 | ||
4412 | if (r > 1000) | |
4413 | r = 1000; | |
4414 | ||
4415 | r = 1000 - r; | |
d0180171 | 4416 | drbd_info(device, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " |
b411b363 PR |
4417 | "total %u; compression: %u.%u%%\n", |
4418 | direction, | |
4419 | c->bytes[1], c->packets[1], | |
4420 | c->bytes[0], c->packets[0], | |
4421 | total, r/10, r % 10); | |
4422 | } | |
4423 | ||
4424 | /* Since we are processing the bitfield from lower addresses to higher, | |
4425 | it does not matter if the process it in 32 bit chunks or 64 bit | |
4426 | chunks as long as it is little endian. (Understand it as byte stream, | |
4427 | beginning with the lowest byte...) If we would use big endian | |
4428 | we would need to process it from the highest address to the lowest, | |
4429 | in order to be agnostic to the 32 vs 64 bits issue. | |
4430 | ||
4431 | returns 0 on failure, 1 if we successfully received it. */ | |
bde89a9e | 4432 | static int receive_bitmap(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4433 | { |
9f4fe9ad | 4434 | struct drbd_peer_device *peer_device; |
b30ab791 | 4435 | struct drbd_device *device; |
b411b363 | 4436 | struct bm_xfer_ctx c; |
2c46407d | 4437 | int err; |
4a76b161 | 4438 | |
9f4fe9ad AG |
4439 | peer_device = conn_peer_device(connection, pi->vnr); |
4440 | if (!peer_device) | |
4a76b161 | 4441 | return -EIO; |
9f4fe9ad | 4442 | device = peer_device->device; |
b411b363 | 4443 | |
b30ab791 | 4444 | drbd_bm_lock(device, "receive bitmap", BM_LOCKED_SET_ALLOWED); |
20ceb2b2 LE |
4445 | /* you are supposed to send additional out-of-sync information |
4446 | * if you actually set bits during this phase */ | |
b411b363 | 4447 | |
b411b363 | 4448 | c = (struct bm_xfer_ctx) { |
b30ab791 AG |
4449 | .bm_bits = drbd_bm_bits(device), |
4450 | .bm_words = drbd_bm_words(device), | |
b411b363 PR |
4451 | }; |
4452 | ||
2c46407d | 4453 | for(;;) { |
e658983a | 4454 | if (pi->cmd == P_BITMAP) |
69a22773 | 4455 | err = receive_bitmap_plain(peer_device, pi->size, pi->data, &c); |
e658983a | 4456 | else if (pi->cmd == P_COMPRESSED_BITMAP) { |
b411b363 PR |
4457 | /* MAYBE: sanity check that we speak proto >= 90, |
4458 | * and the feature is enabled! */ | |
e658983a | 4459 | struct p_compressed_bm *p = pi->data; |
b411b363 | 4460 | |
bde89a9e | 4461 | if (pi->size > DRBD_SOCKET_BUFFER_SIZE - drbd_header_size(connection)) { |
d0180171 | 4462 | drbd_err(device, "ReportCBitmap packet too large\n"); |
82bc0194 | 4463 | err = -EIO; |
b411b363 PR |
4464 | goto out; |
4465 | } | |
e658983a | 4466 | if (pi->size <= sizeof(*p)) { |
d0180171 | 4467 | drbd_err(device, "ReportCBitmap packet too small (l:%u)\n", pi->size); |
82bc0194 | 4468 | err = -EIO; |
78fcbdae | 4469 | goto out; |
b411b363 | 4470 | } |
9f4fe9ad | 4471 | err = drbd_recv_all(peer_device->connection, p, pi->size); |
e658983a AG |
4472 | if (err) |
4473 | goto out; | |
69a22773 | 4474 | err = decode_bitmap_c(peer_device, p, &c, pi->size); |
b411b363 | 4475 | } else { |
d0180171 | 4476 | drbd_warn(device, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", pi->cmd); |
82bc0194 | 4477 | err = -EIO; |
b411b363 PR |
4478 | goto out; |
4479 | } | |
4480 | ||
e2857216 | 4481 | c.packets[pi->cmd == P_BITMAP]++; |
bde89a9e | 4482 | c.bytes[pi->cmd == P_BITMAP] += drbd_header_size(connection) + pi->size; |
b411b363 | 4483 | |
2c46407d AG |
4484 | if (err <= 0) { |
4485 | if (err < 0) | |
4486 | goto out; | |
b411b363 | 4487 | break; |
2c46407d | 4488 | } |
9f4fe9ad | 4489 | err = drbd_recv_header(peer_device->connection, pi); |
82bc0194 | 4490 | if (err) |
b411b363 | 4491 | goto out; |
2c46407d | 4492 | } |
b411b363 | 4493 | |
b30ab791 | 4494 | INFO_bm_xfer_stats(device, "receive", &c); |
b411b363 | 4495 | |
b30ab791 | 4496 | if (device->state.conn == C_WF_BITMAP_T) { |
de1f8e4a AG |
4497 | enum drbd_state_rv rv; |
4498 | ||
b30ab791 | 4499 | err = drbd_send_bitmap(device); |
82bc0194 | 4500 | if (err) |
b411b363 PR |
4501 | goto out; |
4502 | /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ | |
b30ab791 | 4503 | rv = _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); |
0b0ba1ef | 4504 | D_ASSERT(device, rv == SS_SUCCESS); |
b30ab791 | 4505 | } else if (device->state.conn != C_WF_BITMAP_S) { |
b411b363 PR |
4506 | /* admin may have requested C_DISCONNECTING, |
4507 | * other threads may have noticed network errors */ | |
d0180171 | 4508 | drbd_info(device, "unexpected cstate (%s) in receive_bitmap\n", |
b30ab791 | 4509 | drbd_conn_str(device->state.conn)); |
b411b363 | 4510 | } |
82bc0194 | 4511 | err = 0; |
b411b363 | 4512 | |
b411b363 | 4513 | out: |
b30ab791 AG |
4514 | drbd_bm_unlock(device); |
4515 | if (!err && device->state.conn == C_WF_BITMAP_S) | |
4516 | drbd_start_resync(device, C_SYNC_SOURCE); | |
82bc0194 | 4517 | return err; |
b411b363 PR |
4518 | } |
4519 | ||
bde89a9e | 4520 | static int receive_skip(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 4521 | { |
1ec861eb | 4522 | drbd_warn(connection, "skipping unknown optional packet type %d, l: %d!\n", |
e2857216 | 4523 | pi->cmd, pi->size); |
b411b363 | 4524 | |
bde89a9e | 4525 | return ignore_remaining_packet(connection, pi); |
b411b363 PR |
4526 | } |
4527 | ||
bde89a9e | 4528 | static int receive_UnplugRemote(struct drbd_connection *connection, struct packet_info *pi) |
0ced55a3 | 4529 | { |
e7f52dfb LE |
4530 | /* Make sure we've acked all the TCP data associated |
4531 | * with the data requests being unplugged */ | |
bde89a9e | 4532 | drbd_tcp_quickack(connection->data.socket); |
0ced55a3 | 4533 | |
82bc0194 | 4534 | return 0; |
0ced55a3 PR |
4535 | } |
4536 | ||
bde89a9e | 4537 | static int receive_out_of_sync(struct drbd_connection *connection, struct packet_info *pi) |
73a01a18 | 4538 | { |
9f4fe9ad | 4539 | struct drbd_peer_device *peer_device; |
b30ab791 | 4540 | struct drbd_device *device; |
e658983a | 4541 | struct p_block_desc *p = pi->data; |
4a76b161 | 4542 | |
9f4fe9ad AG |
4543 | peer_device = conn_peer_device(connection, pi->vnr); |
4544 | if (!peer_device) | |
4a76b161 | 4545 | return -EIO; |
9f4fe9ad | 4546 | device = peer_device->device; |
73a01a18 | 4547 | |
b30ab791 | 4548 | switch (device->state.conn) { |
f735e363 LE |
4549 | case C_WF_SYNC_UUID: |
4550 | case C_WF_BITMAP_T: | |
4551 | case C_BEHIND: | |
4552 | break; | |
4553 | default: | |
d0180171 | 4554 | drbd_err(device, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n", |
b30ab791 | 4555 | drbd_conn_str(device->state.conn)); |
f735e363 LE |
4556 | } |
4557 | ||
b30ab791 | 4558 | drbd_set_out_of_sync(device, be64_to_cpu(p->sector), be32_to_cpu(p->blksize)); |
73a01a18 | 4559 | |
82bc0194 | 4560 | return 0; |
73a01a18 PR |
4561 | } |
4562 | ||
02918be2 PR |
4563 | struct data_cmd { |
4564 | int expect_payload; | |
4565 | size_t pkt_size; | |
bde89a9e | 4566 | int (*fn)(struct drbd_connection *, struct packet_info *); |
02918be2 PR |
4567 | }; |
4568 | ||
4569 | static struct data_cmd drbd_cmd_handler[] = { | |
4570 | [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, | |
4571 | [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, | |
4572 | [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , | |
4573 | [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , | |
e658983a AG |
4574 | [P_BITMAP] = { 1, 0, receive_bitmap } , |
4575 | [P_COMPRESSED_BITMAP] = { 1, 0, receive_bitmap } , | |
4576 | [P_UNPLUG_REMOTE] = { 0, 0, receive_UnplugRemote }, | |
02918be2 PR |
4577 | [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, |
4578 | [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
e658983a AG |
4579 | [P_SYNC_PARAM] = { 1, 0, receive_SyncParam }, |
4580 | [P_SYNC_PARAM89] = { 1, 0, receive_SyncParam }, | |
02918be2 PR |
4581 | [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, |
4582 | [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, | |
4583 | [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, | |
4584 | [P_STATE] = { 0, sizeof(struct p_state), receive_state }, | |
4585 | [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, | |
4586 | [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, | |
4587 | [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
4588 | [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
4589 | [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
4590 | [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, | |
73a01a18 | 4591 | [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync }, |
4a76b161 | 4592 | [P_CONN_ST_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_conn_state }, |
036b17ea | 4593 | [P_PROTOCOL_UPDATE] = { 1, sizeof(struct p_protocol), receive_protocol }, |
a0fb3c47 | 4594 | [P_TRIM] = { 0, sizeof(struct p_trim), receive_Data }, |
b411b363 PR |
4595 | }; |
4596 | ||
bde89a9e | 4597 | static void drbdd(struct drbd_connection *connection) |
b411b363 | 4598 | { |
77351055 | 4599 | struct packet_info pi; |
02918be2 | 4600 | size_t shs; /* sub header size */ |
82bc0194 | 4601 | int err; |
b411b363 | 4602 | |
bde89a9e | 4603 | while (get_t_state(&connection->receiver) == RUNNING) { |
deebe195 | 4604 | struct data_cmd *cmd; |
b411b363 | 4605 | |
bde89a9e | 4606 | drbd_thread_current_set_cpu(&connection->receiver); |
944410e9 | 4607 | update_receiver_timing_details(connection, drbd_recv_header); |
bde89a9e | 4608 | if (drbd_recv_header(connection, &pi)) |
02918be2 | 4609 | goto err_out; |
b411b363 | 4610 | |
deebe195 | 4611 | cmd = &drbd_cmd_handler[pi.cmd]; |
4a76b161 | 4612 | if (unlikely(pi.cmd >= ARRAY_SIZE(drbd_cmd_handler) || !cmd->fn)) { |
1ec861eb | 4613 | drbd_err(connection, "Unexpected data packet %s (0x%04x)", |
2fcb8f30 | 4614 | cmdname(pi.cmd), pi.cmd); |
02918be2 | 4615 | goto err_out; |
0b33a916 | 4616 | } |
b411b363 | 4617 | |
e658983a AG |
4618 | shs = cmd->pkt_size; |
4619 | if (pi.size > shs && !cmd->expect_payload) { | |
1ec861eb | 4620 | drbd_err(connection, "No payload expected %s l:%d\n", |
2fcb8f30 | 4621 | cmdname(pi.cmd), pi.size); |
02918be2 | 4622 | goto err_out; |
b411b363 | 4623 | } |
b411b363 | 4624 | |
c13f7e1a | 4625 | if (shs) { |
944410e9 | 4626 | update_receiver_timing_details(connection, drbd_recv_all_warn); |
bde89a9e | 4627 | err = drbd_recv_all_warn(connection, pi.data, shs); |
a5c31904 | 4628 | if (err) |
c13f7e1a | 4629 | goto err_out; |
e2857216 | 4630 | pi.size -= shs; |
c13f7e1a LE |
4631 | } |
4632 | ||
944410e9 | 4633 | update_receiver_timing_details(connection, cmd->fn); |
bde89a9e | 4634 | err = cmd->fn(connection, &pi); |
4a76b161 | 4635 | if (err) { |
1ec861eb | 4636 | drbd_err(connection, "error receiving %s, e: %d l: %d!\n", |
9f5bdc33 | 4637 | cmdname(pi.cmd), err, pi.size); |
02918be2 | 4638 | goto err_out; |
b411b363 PR |
4639 | } |
4640 | } | |
82bc0194 | 4641 | return; |
b411b363 | 4642 | |
82bc0194 | 4643 | err_out: |
bde89a9e | 4644 | conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); |
b411b363 PR |
4645 | } |
4646 | ||
bde89a9e | 4647 | static void conn_disconnect(struct drbd_connection *connection) |
b411b363 | 4648 | { |
c06ece6b | 4649 | struct drbd_peer_device *peer_device; |
bbeb641c | 4650 | enum drbd_conns oc; |
376694a0 | 4651 | int vnr; |
b411b363 | 4652 | |
bde89a9e | 4653 | if (connection->cstate == C_STANDALONE) |
b411b363 | 4654 | return; |
b411b363 | 4655 | |
545752d5 LE |
4656 | /* We are about to start the cleanup after connection loss. |
4657 | * Make sure drbd_make_request knows about that. | |
4658 | * Usually we should be in some network failure state already, | |
4659 | * but just in case we are not, we fix it up here. | |
4660 | */ | |
bde89a9e | 4661 | conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); |
545752d5 | 4662 | |
b411b363 | 4663 | /* asender does not clean up anything. it must not interfere, either */ |
bde89a9e AG |
4664 | drbd_thread_stop(&connection->asender); |
4665 | drbd_free_sock(connection); | |
360cc740 | 4666 | |
c141ebda | 4667 | rcu_read_lock(); |
c06ece6b AG |
4668 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
4669 | struct drbd_device *device = peer_device->device; | |
b30ab791 | 4670 | kref_get(&device->kref); |
c141ebda | 4671 | rcu_read_unlock(); |
69a22773 | 4672 | drbd_disconnected(peer_device); |
c06ece6b | 4673 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda PR |
4674 | rcu_read_lock(); |
4675 | } | |
4676 | rcu_read_unlock(); | |
4677 | ||
bde89a9e | 4678 | if (!list_empty(&connection->current_epoch->list)) |
1ec861eb | 4679 | drbd_err(connection, "ASSERTION FAILED: connection->current_epoch->list not empty\n"); |
12038a3a | 4680 | /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ |
bde89a9e AG |
4681 | atomic_set(&connection->current_epoch->epoch_size, 0); |
4682 | connection->send.seen_any_write_yet = false; | |
12038a3a | 4683 | |
1ec861eb | 4684 | drbd_info(connection, "Connection closed\n"); |
360cc740 | 4685 | |
bde89a9e AG |
4686 | if (conn_highest_role(connection) == R_PRIMARY && conn_highest_pdsk(connection) >= D_UNKNOWN) |
4687 | conn_try_outdate_peer_async(connection); | |
cb703454 | 4688 | |
0500813f | 4689 | spin_lock_irq(&connection->resource->req_lock); |
bde89a9e | 4690 | oc = connection->cstate; |
bbeb641c | 4691 | if (oc >= C_UNCONNECTED) |
bde89a9e | 4692 | _conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE); |
bbeb641c | 4693 | |
0500813f | 4694 | spin_unlock_irq(&connection->resource->req_lock); |
360cc740 | 4695 | |
f3dfa40a | 4696 | if (oc == C_DISCONNECTING) |
bde89a9e | 4697 | conn_request_state(connection, NS(conn, C_STANDALONE), CS_VERBOSE | CS_HARD); |
360cc740 PR |
4698 | } |
4699 | ||
69a22773 | 4700 | static int drbd_disconnected(struct drbd_peer_device *peer_device) |
360cc740 | 4701 | { |
69a22773 | 4702 | struct drbd_device *device = peer_device->device; |
360cc740 | 4703 | unsigned int i; |
b411b363 | 4704 | |
85719573 | 4705 | /* wait for current activity to cease. */ |
0500813f | 4706 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 AG |
4707 | _drbd_wait_ee_list_empty(device, &device->active_ee); |
4708 | _drbd_wait_ee_list_empty(device, &device->sync_ee); | |
4709 | _drbd_wait_ee_list_empty(device, &device->read_ee); | |
0500813f | 4710 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
4711 | |
4712 | /* We do not have data structures that would allow us to | |
4713 | * get the rs_pending_cnt down to 0 again. | |
4714 | * * On C_SYNC_TARGET we do not have any data structures describing | |
4715 | * the pending RSDataRequest's we have sent. | |
4716 | * * On C_SYNC_SOURCE there is no data structure that tracks | |
4717 | * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. | |
4718 | * And no, it is not the sum of the reference counts in the | |
4719 | * resync_LRU. The resync_LRU tracks the whole operation including | |
4720 | * the disk-IO, while the rs_pending_cnt only tracks the blocks | |
4721 | * on the fly. */ | |
b30ab791 AG |
4722 | drbd_rs_cancel_all(device); |
4723 | device->rs_total = 0; | |
4724 | device->rs_failed = 0; | |
4725 | atomic_set(&device->rs_pending_cnt, 0); | |
4726 | wake_up(&device->misc_wait); | |
b411b363 | 4727 | |
b30ab791 AG |
4728 | del_timer_sync(&device->resync_timer); |
4729 | resync_timer_fn((unsigned long)device); | |
b411b363 | 4730 | |
b411b363 PR |
4731 | /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, |
4732 | * w_make_resync_request etc. which may still be on the worker queue | |
4733 | * to be "canceled" */ | |
b5043c5e | 4734 | drbd_flush_workqueue(&peer_device->connection->sender_work); |
b411b363 | 4735 | |
b30ab791 | 4736 | drbd_finish_peer_reqs(device); |
b411b363 | 4737 | |
d10b4ea3 PR |
4738 | /* This second workqueue flush is necessary, since drbd_finish_peer_reqs() |
4739 | might have issued a work again. The one before drbd_finish_peer_reqs() is | |
4740 | necessary to reclain net_ee in drbd_finish_peer_reqs(). */ | |
b5043c5e | 4741 | drbd_flush_workqueue(&peer_device->connection->sender_work); |
d10b4ea3 | 4742 | |
08332d73 LE |
4743 | /* need to do it again, drbd_finish_peer_reqs() may have populated it |
4744 | * again via drbd_try_clear_on_disk_bm(). */ | |
b30ab791 | 4745 | drbd_rs_cancel_all(device); |
b411b363 | 4746 | |
b30ab791 AG |
4747 | kfree(device->p_uuid); |
4748 | device->p_uuid = NULL; | |
b411b363 | 4749 | |
b30ab791 | 4750 | if (!drbd_suspended(device)) |
69a22773 | 4751 | tl_clear(peer_device->connection); |
b411b363 | 4752 | |
b30ab791 | 4753 | drbd_md_sync(device); |
b411b363 | 4754 | |
20ceb2b2 LE |
4755 | /* serialize with bitmap writeout triggered by the state change, |
4756 | * if any. */ | |
b30ab791 | 4757 | wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags)); |
20ceb2b2 | 4758 | |
b411b363 PR |
4759 | /* tcp_close and release of sendpage pages can be deferred. I don't |
4760 | * want to use SO_LINGER, because apparently it can be deferred for | |
4761 | * more than 20 seconds (longest time I checked). | |
4762 | * | |
4763 | * Actually we don't care for exactly when the network stack does its | |
4764 | * put_page(), but release our reference on these pages right here. | |
4765 | */ | |
b30ab791 | 4766 | i = drbd_free_peer_reqs(device, &device->net_ee); |
b411b363 | 4767 | if (i) |
d0180171 | 4768 | drbd_info(device, "net_ee not empty, killed %u entries\n", i); |
b30ab791 | 4769 | i = atomic_read(&device->pp_in_use_by_net); |
435f0740 | 4770 | if (i) |
d0180171 | 4771 | drbd_info(device, "pp_in_use_by_net = %d, expected 0\n", i); |
b30ab791 | 4772 | i = atomic_read(&device->pp_in_use); |
b411b363 | 4773 | if (i) |
d0180171 | 4774 | drbd_info(device, "pp_in_use = %d, expected 0\n", i); |
b411b363 | 4775 | |
0b0ba1ef AG |
4776 | D_ASSERT(device, list_empty(&device->read_ee)); |
4777 | D_ASSERT(device, list_empty(&device->active_ee)); | |
4778 | D_ASSERT(device, list_empty(&device->sync_ee)); | |
4779 | D_ASSERT(device, list_empty(&device->done_ee)); | |
b411b363 | 4780 | |
360cc740 | 4781 | return 0; |
b411b363 PR |
4782 | } |
4783 | ||
4784 | /* | |
4785 | * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version | |
4786 | * we can agree on is stored in agreed_pro_version. | |
4787 | * | |
4788 | * feature flags and the reserved array should be enough room for future | |
4789 | * enhancements of the handshake protocol, and possible plugins... | |
4790 | * | |
4791 | * for now, they are expected to be zero, but ignored. | |
4792 | */ | |
bde89a9e | 4793 | static int drbd_send_features(struct drbd_connection *connection) |
b411b363 | 4794 | { |
9f5bdc33 AG |
4795 | struct drbd_socket *sock; |
4796 | struct p_connection_features *p; | |
b411b363 | 4797 | |
bde89a9e AG |
4798 | sock = &connection->data; |
4799 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 | 4800 | if (!p) |
e8d17b01 | 4801 | return -EIO; |
b411b363 PR |
4802 | memset(p, 0, sizeof(*p)); |
4803 | p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); | |
4804 | p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); | |
20c68fde | 4805 | p->feature_flags = cpu_to_be32(PRO_FEATURES); |
bde89a9e | 4806 | return conn_send_command(connection, sock, P_CONNECTION_FEATURES, sizeof(*p), NULL, 0); |
b411b363 PR |
4807 | } |
4808 | ||
4809 | /* | |
4810 | * return values: | |
4811 | * 1 yes, we have a valid connection | |
4812 | * 0 oops, did not work out, please try again | |
4813 | * -1 peer talks different language, | |
4814 | * no point in trying again, please go standalone. | |
4815 | */ | |
bde89a9e | 4816 | static int drbd_do_features(struct drbd_connection *connection) |
b411b363 | 4817 | { |
bde89a9e | 4818 | /* ASSERT current == connection->receiver ... */ |
e658983a AG |
4819 | struct p_connection_features *p; |
4820 | const int expect = sizeof(struct p_connection_features); | |
77351055 | 4821 | struct packet_info pi; |
a5c31904 | 4822 | int err; |
b411b363 | 4823 | |
bde89a9e | 4824 | err = drbd_send_features(connection); |
e8d17b01 | 4825 | if (err) |
b411b363 PR |
4826 | return 0; |
4827 | ||
bde89a9e | 4828 | err = drbd_recv_header(connection, &pi); |
69bc7bc3 | 4829 | if (err) |
b411b363 PR |
4830 | return 0; |
4831 | ||
6038178e | 4832 | if (pi.cmd != P_CONNECTION_FEATURES) { |
1ec861eb | 4833 | drbd_err(connection, "expected ConnectionFeatures packet, received: %s (0x%04x)\n", |
2fcb8f30 | 4834 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4835 | return -1; |
4836 | } | |
4837 | ||
77351055 | 4838 | if (pi.size != expect) { |
1ec861eb | 4839 | drbd_err(connection, "expected ConnectionFeatures length: %u, received: %u\n", |
77351055 | 4840 | expect, pi.size); |
b411b363 PR |
4841 | return -1; |
4842 | } | |
4843 | ||
e658983a | 4844 | p = pi.data; |
bde89a9e | 4845 | err = drbd_recv_all_warn(connection, p, expect); |
a5c31904 | 4846 | if (err) |
b411b363 | 4847 | return 0; |
b411b363 | 4848 | |
b411b363 PR |
4849 | p->protocol_min = be32_to_cpu(p->protocol_min); |
4850 | p->protocol_max = be32_to_cpu(p->protocol_max); | |
4851 | if (p->protocol_max == 0) | |
4852 | p->protocol_max = p->protocol_min; | |
4853 | ||
4854 | if (PRO_VERSION_MAX < p->protocol_min || | |
4855 | PRO_VERSION_MIN > p->protocol_max) | |
4856 | goto incompat; | |
4857 | ||
bde89a9e | 4858 | connection->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); |
20c68fde | 4859 | connection->agreed_features = PRO_FEATURES & be32_to_cpu(p->feature_flags); |
b411b363 | 4860 | |
1ec861eb | 4861 | drbd_info(connection, "Handshake successful: " |
bde89a9e | 4862 | "Agreed network protocol version %d\n", connection->agreed_pro_version); |
b411b363 | 4863 | |
20c68fde LE |
4864 | drbd_info(connection, "Agreed to%ssupport TRIM on protocol level\n", |
4865 | connection->agreed_features & FF_TRIM ? " " : " not "); | |
4866 | ||
b411b363 PR |
4867 | return 1; |
4868 | ||
4869 | incompat: | |
1ec861eb | 4870 | drbd_err(connection, "incompatible DRBD dialects: " |
b411b363 PR |
4871 | "I support %d-%d, peer supports %d-%d\n", |
4872 | PRO_VERSION_MIN, PRO_VERSION_MAX, | |
4873 | p->protocol_min, p->protocol_max); | |
4874 | return -1; | |
4875 | } | |
4876 | ||
4877 | #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) | |
bde89a9e | 4878 | static int drbd_do_auth(struct drbd_connection *connection) |
b411b363 | 4879 | { |
1ec861eb AG |
4880 | drbd_err(connection, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); |
4881 | drbd_err(connection, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); | |
b10d96cb | 4882 | return -1; |
b411b363 PR |
4883 | } |
4884 | #else | |
4885 | #define CHALLENGE_LEN 64 | |
b10d96cb JT |
4886 | |
4887 | /* Return value: | |
4888 | 1 - auth succeeded, | |
4889 | 0 - failed, try again (network error), | |
4890 | -1 - auth failed, don't try again. | |
4891 | */ | |
4892 | ||
bde89a9e | 4893 | static int drbd_do_auth(struct drbd_connection *connection) |
b411b363 | 4894 | { |
9f5bdc33 | 4895 | struct drbd_socket *sock; |
b411b363 PR |
4896 | char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ |
4897 | struct scatterlist sg; | |
4898 | char *response = NULL; | |
4899 | char *right_response = NULL; | |
4900 | char *peers_ch = NULL; | |
44ed167d PR |
4901 | unsigned int key_len; |
4902 | char secret[SHARED_SECRET_MAX]; /* 64 byte */ | |
b411b363 PR |
4903 | unsigned int resp_size; |
4904 | struct hash_desc desc; | |
77351055 | 4905 | struct packet_info pi; |
44ed167d | 4906 | struct net_conf *nc; |
69bc7bc3 | 4907 | int err, rv; |
b411b363 | 4908 | |
9f5bdc33 | 4909 | /* FIXME: Put the challenge/response into the preallocated socket buffer. */ |
b411b363 | 4910 | |
44ed167d | 4911 | rcu_read_lock(); |
bde89a9e | 4912 | nc = rcu_dereference(connection->net_conf); |
44ed167d PR |
4913 | key_len = strlen(nc->shared_secret); |
4914 | memcpy(secret, nc->shared_secret, key_len); | |
4915 | rcu_read_unlock(); | |
4916 | ||
bde89a9e | 4917 | desc.tfm = connection->cram_hmac_tfm; |
b411b363 PR |
4918 | desc.flags = 0; |
4919 | ||
bde89a9e | 4920 | rv = crypto_hash_setkey(connection->cram_hmac_tfm, (u8 *)secret, key_len); |
b411b363 | 4921 | if (rv) { |
1ec861eb | 4922 | drbd_err(connection, "crypto_hash_setkey() failed with %d\n", rv); |
b10d96cb | 4923 | rv = -1; |
b411b363 PR |
4924 | goto fail; |
4925 | } | |
4926 | ||
4927 | get_random_bytes(my_challenge, CHALLENGE_LEN); | |
4928 | ||
bde89a9e AG |
4929 | sock = &connection->data; |
4930 | if (!conn_prepare_command(connection, sock)) { | |
9f5bdc33 AG |
4931 | rv = 0; |
4932 | goto fail; | |
4933 | } | |
bde89a9e | 4934 | rv = !conn_send_command(connection, sock, P_AUTH_CHALLENGE, 0, |
9f5bdc33 | 4935 | my_challenge, CHALLENGE_LEN); |
b411b363 PR |
4936 | if (!rv) |
4937 | goto fail; | |
4938 | ||
bde89a9e | 4939 | err = drbd_recv_header(connection, &pi); |
69bc7bc3 AG |
4940 | if (err) { |
4941 | rv = 0; | |
b411b363 | 4942 | goto fail; |
69bc7bc3 | 4943 | } |
b411b363 | 4944 | |
77351055 | 4945 | if (pi.cmd != P_AUTH_CHALLENGE) { |
1ec861eb | 4946 | drbd_err(connection, "expected AuthChallenge packet, received: %s (0x%04x)\n", |
2fcb8f30 | 4947 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4948 | rv = 0; |
4949 | goto fail; | |
4950 | } | |
4951 | ||
77351055 | 4952 | if (pi.size > CHALLENGE_LEN * 2) { |
1ec861eb | 4953 | drbd_err(connection, "expected AuthChallenge payload too big.\n"); |
b10d96cb | 4954 | rv = -1; |
b411b363 PR |
4955 | goto fail; |
4956 | } | |
4957 | ||
67cca286 PR |
4958 | if (pi.size < CHALLENGE_LEN) { |
4959 | drbd_err(connection, "AuthChallenge payload too small.\n"); | |
4960 | rv = -1; | |
4961 | goto fail; | |
4962 | } | |
4963 | ||
77351055 | 4964 | peers_ch = kmalloc(pi.size, GFP_NOIO); |
b411b363 | 4965 | if (peers_ch == NULL) { |
1ec861eb | 4966 | drbd_err(connection, "kmalloc of peers_ch failed\n"); |
b10d96cb | 4967 | rv = -1; |
b411b363 PR |
4968 | goto fail; |
4969 | } | |
4970 | ||
bde89a9e | 4971 | err = drbd_recv_all_warn(connection, peers_ch, pi.size); |
a5c31904 | 4972 | if (err) { |
b411b363 PR |
4973 | rv = 0; |
4974 | goto fail; | |
4975 | } | |
4976 | ||
67cca286 PR |
4977 | if (!memcmp(my_challenge, peers_ch, CHALLENGE_LEN)) { |
4978 | drbd_err(connection, "Peer presented the same challenge!\n"); | |
4979 | rv = -1; | |
4980 | goto fail; | |
4981 | } | |
4982 | ||
bde89a9e | 4983 | resp_size = crypto_hash_digestsize(connection->cram_hmac_tfm); |
b411b363 PR |
4984 | response = kmalloc(resp_size, GFP_NOIO); |
4985 | if (response == NULL) { | |
1ec861eb | 4986 | drbd_err(connection, "kmalloc of response failed\n"); |
b10d96cb | 4987 | rv = -1; |
b411b363 PR |
4988 | goto fail; |
4989 | } | |
4990 | ||
4991 | sg_init_table(&sg, 1); | |
77351055 | 4992 | sg_set_buf(&sg, peers_ch, pi.size); |
b411b363 PR |
4993 | |
4994 | rv = crypto_hash_digest(&desc, &sg, sg.length, response); | |
4995 | if (rv) { | |
1ec861eb | 4996 | drbd_err(connection, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 4997 | rv = -1; |
b411b363 PR |
4998 | goto fail; |
4999 | } | |
5000 | ||
bde89a9e | 5001 | if (!conn_prepare_command(connection, sock)) { |
9f5bdc33 | 5002 | rv = 0; |
b411b363 | 5003 | goto fail; |
9f5bdc33 | 5004 | } |
bde89a9e | 5005 | rv = !conn_send_command(connection, sock, P_AUTH_RESPONSE, 0, |
9f5bdc33 | 5006 | response, resp_size); |
b411b363 PR |
5007 | if (!rv) |
5008 | goto fail; | |
5009 | ||
bde89a9e | 5010 | err = drbd_recv_header(connection, &pi); |
69bc7bc3 | 5011 | if (err) { |
b411b363 PR |
5012 | rv = 0; |
5013 | goto fail; | |
5014 | } | |
5015 | ||
77351055 | 5016 | if (pi.cmd != P_AUTH_RESPONSE) { |
1ec861eb | 5017 | drbd_err(connection, "expected AuthResponse packet, received: %s (0x%04x)\n", |
2fcb8f30 | 5018 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
5019 | rv = 0; |
5020 | goto fail; | |
5021 | } | |
5022 | ||
77351055 | 5023 | if (pi.size != resp_size) { |
1ec861eb | 5024 | drbd_err(connection, "expected AuthResponse payload of wrong size\n"); |
b411b363 PR |
5025 | rv = 0; |
5026 | goto fail; | |
5027 | } | |
b411b363 | 5028 | |
bde89a9e | 5029 | err = drbd_recv_all_warn(connection, response , resp_size); |
a5c31904 | 5030 | if (err) { |
b411b363 PR |
5031 | rv = 0; |
5032 | goto fail; | |
5033 | } | |
5034 | ||
5035 | right_response = kmalloc(resp_size, GFP_NOIO); | |
2d1ee87d | 5036 | if (right_response == NULL) { |
1ec861eb | 5037 | drbd_err(connection, "kmalloc of right_response failed\n"); |
b10d96cb | 5038 | rv = -1; |
b411b363 PR |
5039 | goto fail; |
5040 | } | |
5041 | ||
5042 | sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); | |
5043 | ||
5044 | rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); | |
5045 | if (rv) { | |
1ec861eb | 5046 | drbd_err(connection, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 5047 | rv = -1; |
b411b363 PR |
5048 | goto fail; |
5049 | } | |
5050 | ||
5051 | rv = !memcmp(response, right_response, resp_size); | |
5052 | ||
5053 | if (rv) | |
1ec861eb | 5054 | drbd_info(connection, "Peer authenticated using %d bytes HMAC\n", |
44ed167d | 5055 | resp_size); |
b10d96cb JT |
5056 | else |
5057 | rv = -1; | |
b411b363 PR |
5058 | |
5059 | fail: | |
5060 | kfree(peers_ch); | |
5061 | kfree(response); | |
5062 | kfree(right_response); | |
5063 | ||
5064 | return rv; | |
5065 | } | |
5066 | #endif | |
5067 | ||
8fe60551 | 5068 | int drbd_receiver(struct drbd_thread *thi) |
b411b363 | 5069 | { |
bde89a9e | 5070 | struct drbd_connection *connection = thi->connection; |
b411b363 PR |
5071 | int h; |
5072 | ||
1ec861eb | 5073 | drbd_info(connection, "receiver (re)started\n"); |
b411b363 PR |
5074 | |
5075 | do { | |
bde89a9e | 5076 | h = conn_connect(connection); |
b411b363 | 5077 | if (h == 0) { |
bde89a9e | 5078 | conn_disconnect(connection); |
20ee6390 | 5079 | schedule_timeout_interruptible(HZ); |
b411b363 PR |
5080 | } |
5081 | if (h == -1) { | |
1ec861eb | 5082 | drbd_warn(connection, "Discarding network configuration.\n"); |
bde89a9e | 5083 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
5084 | } |
5085 | } while (h == 0); | |
5086 | ||
91fd4dad | 5087 | if (h > 0) |
bde89a9e | 5088 | drbdd(connection); |
b411b363 | 5089 | |
bde89a9e | 5090 | conn_disconnect(connection); |
b411b363 | 5091 | |
1ec861eb | 5092 | drbd_info(connection, "receiver terminated\n"); |
b411b363 PR |
5093 | return 0; |
5094 | } | |
5095 | ||
5096 | /* ********* acknowledge sender ******** */ | |
5097 | ||
bde89a9e | 5098 | static int got_conn_RqSReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5099 | { |
e658983a | 5100 | struct p_req_state_reply *p = pi->data; |
e4f78ede PR |
5101 | int retcode = be32_to_cpu(p->retcode); |
5102 | ||
5103 | if (retcode >= SS_SUCCESS) { | |
bde89a9e | 5104 | set_bit(CONN_WD_ST_CHG_OKAY, &connection->flags); |
e4f78ede | 5105 | } else { |
bde89a9e | 5106 | set_bit(CONN_WD_ST_CHG_FAIL, &connection->flags); |
1ec861eb | 5107 | drbd_err(connection, "Requested state change failed by peer: %s (%d)\n", |
e4f78ede PR |
5108 | drbd_set_st_err_str(retcode), retcode); |
5109 | } | |
bde89a9e | 5110 | wake_up(&connection->ping_wait); |
e4f78ede | 5111 | |
2735a594 | 5112 | return 0; |
e4f78ede | 5113 | } |
b411b363 | 5114 | |
bde89a9e | 5115 | static int got_RqSReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5116 | { |
9f4fe9ad | 5117 | struct drbd_peer_device *peer_device; |
b30ab791 | 5118 | struct drbd_device *device; |
e658983a | 5119 | struct p_req_state_reply *p = pi->data; |
b411b363 PR |
5120 | int retcode = be32_to_cpu(p->retcode); |
5121 | ||
9f4fe9ad AG |
5122 | peer_device = conn_peer_device(connection, pi->vnr); |
5123 | if (!peer_device) | |
2735a594 | 5124 | return -EIO; |
9f4fe9ad | 5125 | device = peer_device->device; |
1952e916 | 5126 | |
bde89a9e | 5127 | if (test_bit(CONN_WD_ST_CHG_REQ, &connection->flags)) { |
0b0ba1ef | 5128 | D_ASSERT(device, connection->agreed_pro_version < 100); |
bde89a9e | 5129 | return got_conn_RqSReply(connection, pi); |
4d0fc3fd PR |
5130 | } |
5131 | ||
b411b363 | 5132 | if (retcode >= SS_SUCCESS) { |
b30ab791 | 5133 | set_bit(CL_ST_CHG_SUCCESS, &device->flags); |
b411b363 | 5134 | } else { |
b30ab791 | 5135 | set_bit(CL_ST_CHG_FAIL, &device->flags); |
d0180171 | 5136 | drbd_err(device, "Requested state change failed by peer: %s (%d)\n", |
e4f78ede | 5137 | drbd_set_st_err_str(retcode), retcode); |
b411b363 | 5138 | } |
b30ab791 | 5139 | wake_up(&device->state_wait); |
b411b363 | 5140 | |
2735a594 | 5141 | return 0; |
b411b363 PR |
5142 | } |
5143 | ||
bde89a9e | 5144 | static int got_Ping(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5145 | { |
bde89a9e | 5146 | return drbd_send_ping_ack(connection); |
b411b363 PR |
5147 | |
5148 | } | |
5149 | ||
bde89a9e | 5150 | static int got_PingAck(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 PR |
5151 | { |
5152 | /* restore idle timeout */ | |
bde89a9e AG |
5153 | connection->meta.socket->sk->sk_rcvtimeo = connection->net_conf->ping_int*HZ; |
5154 | if (!test_and_set_bit(GOT_PING_ACK, &connection->flags)) | |
5155 | wake_up(&connection->ping_wait); | |
b411b363 | 5156 | |
2735a594 | 5157 | return 0; |
b411b363 PR |
5158 | } |
5159 | ||
bde89a9e | 5160 | static int got_IsInSync(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5161 | { |
9f4fe9ad | 5162 | struct drbd_peer_device *peer_device; |
b30ab791 | 5163 | struct drbd_device *device; |
e658983a | 5164 | struct p_block_ack *p = pi->data; |
b411b363 PR |
5165 | sector_t sector = be64_to_cpu(p->sector); |
5166 | int blksize = be32_to_cpu(p->blksize); | |
5167 | ||
9f4fe9ad AG |
5168 | peer_device = conn_peer_device(connection, pi->vnr); |
5169 | if (!peer_device) | |
2735a594 | 5170 | return -EIO; |
9f4fe9ad | 5171 | device = peer_device->device; |
1952e916 | 5172 | |
9f4fe9ad | 5173 | D_ASSERT(device, peer_device->connection->agreed_pro_version >= 89); |
b411b363 | 5174 | |
69a22773 | 5175 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 | 5176 | |
b30ab791 AG |
5177 | if (get_ldev(device)) { |
5178 | drbd_rs_complete_io(device, sector); | |
5179 | drbd_set_in_sync(device, sector, blksize); | |
1d53f09e | 5180 | /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ |
b30ab791 AG |
5181 | device->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); |
5182 | put_ldev(device); | |
1d53f09e | 5183 | } |
b30ab791 AG |
5184 | dec_rs_pending(device); |
5185 | atomic_add(blksize >> 9, &device->rs_sect_in); | |
b411b363 | 5186 | |
2735a594 | 5187 | return 0; |
b411b363 PR |
5188 | } |
5189 | ||
bc9c5c41 | 5190 | static int |
b30ab791 | 5191 | validate_req_change_req_state(struct drbd_device *device, u64 id, sector_t sector, |
bc9c5c41 AG |
5192 | struct rb_root *root, const char *func, |
5193 | enum drbd_req_event what, bool missing_ok) | |
b411b363 PR |
5194 | { |
5195 | struct drbd_request *req; | |
5196 | struct bio_and_error m; | |
5197 | ||
0500813f | 5198 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 5199 | req = find_request(device, root, id, sector, missing_ok, func); |
b411b363 | 5200 | if (unlikely(!req)) { |
0500813f | 5201 | spin_unlock_irq(&device->resource->req_lock); |
85997675 | 5202 | return -EIO; |
b411b363 PR |
5203 | } |
5204 | __req_mod(req, what, &m); | |
0500813f | 5205 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
5206 | |
5207 | if (m.bio) | |
b30ab791 | 5208 | complete_master_bio(device, &m); |
85997675 | 5209 | return 0; |
b411b363 PR |
5210 | } |
5211 | ||
bde89a9e | 5212 | static int got_BlockAck(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5213 | { |
9f4fe9ad | 5214 | struct drbd_peer_device *peer_device; |
b30ab791 | 5215 | struct drbd_device *device; |
e658983a | 5216 | struct p_block_ack *p = pi->data; |
b411b363 PR |
5217 | sector_t sector = be64_to_cpu(p->sector); |
5218 | int blksize = be32_to_cpu(p->blksize); | |
5219 | enum drbd_req_event what; | |
5220 | ||
9f4fe9ad AG |
5221 | peer_device = conn_peer_device(connection, pi->vnr); |
5222 | if (!peer_device) | |
2735a594 | 5223 | return -EIO; |
9f4fe9ad | 5224 | device = peer_device->device; |
1952e916 | 5225 | |
69a22773 | 5226 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 | 5227 | |
579b57ed | 5228 | if (p->block_id == ID_SYNCER) { |
b30ab791 AG |
5229 | drbd_set_in_sync(device, sector, blksize); |
5230 | dec_rs_pending(device); | |
2735a594 | 5231 | return 0; |
b411b363 | 5232 | } |
e05e1e59 | 5233 | switch (pi->cmd) { |
b411b363 | 5234 | case P_RS_WRITE_ACK: |
8554df1c | 5235 | what = WRITE_ACKED_BY_PEER_AND_SIS; |
b411b363 PR |
5236 | break; |
5237 | case P_WRITE_ACK: | |
8554df1c | 5238 | what = WRITE_ACKED_BY_PEER; |
b411b363 PR |
5239 | break; |
5240 | case P_RECV_ACK: | |
8554df1c | 5241 | what = RECV_ACKED_BY_PEER; |
b411b363 | 5242 | break; |
d4dabbe2 LE |
5243 | case P_SUPERSEDED: |
5244 | what = CONFLICT_RESOLVED; | |
b411b363 | 5245 | break; |
7be8da07 | 5246 | case P_RETRY_WRITE: |
7be8da07 | 5247 | what = POSTPONE_WRITE; |
b411b363 PR |
5248 | break; |
5249 | default: | |
2735a594 | 5250 | BUG(); |
b411b363 PR |
5251 | } |
5252 | ||
b30ab791 AG |
5253 | return validate_req_change_req_state(device, p->block_id, sector, |
5254 | &device->write_requests, __func__, | |
2735a594 | 5255 | what, false); |
b411b363 PR |
5256 | } |
5257 | ||
bde89a9e | 5258 | static int got_NegAck(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5259 | { |
9f4fe9ad | 5260 | struct drbd_peer_device *peer_device; |
b30ab791 | 5261 | struct drbd_device *device; |
e658983a | 5262 | struct p_block_ack *p = pi->data; |
b411b363 | 5263 | sector_t sector = be64_to_cpu(p->sector); |
2deb8336 | 5264 | int size = be32_to_cpu(p->blksize); |
85997675 | 5265 | int err; |
b411b363 | 5266 | |
9f4fe9ad AG |
5267 | peer_device = conn_peer_device(connection, pi->vnr); |
5268 | if (!peer_device) | |
2735a594 | 5269 | return -EIO; |
9f4fe9ad | 5270 | device = peer_device->device; |
b411b363 | 5271 | |
69a22773 | 5272 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 | 5273 | |
579b57ed | 5274 | if (p->block_id == ID_SYNCER) { |
b30ab791 AG |
5275 | dec_rs_pending(device); |
5276 | drbd_rs_failed_io(device, sector, size); | |
2735a594 | 5277 | return 0; |
b411b363 | 5278 | } |
2deb8336 | 5279 | |
b30ab791 AG |
5280 | err = validate_req_change_req_state(device, p->block_id, sector, |
5281 | &device->write_requests, __func__, | |
303d1448 | 5282 | NEG_ACKED, true); |
85997675 | 5283 | if (err) { |
c3afd8f5 AG |
5284 | /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs. |
5285 | The master bio might already be completed, therefore the | |
5286 | request is no longer in the collision hash. */ | |
5287 | /* In Protocol B we might already have got a P_RECV_ACK | |
5288 | but then get a P_NEG_ACK afterwards. */ | |
b30ab791 | 5289 | drbd_set_out_of_sync(device, sector, size); |
2deb8336 | 5290 | } |
2735a594 | 5291 | return 0; |
b411b363 PR |
5292 | } |
5293 | ||
bde89a9e | 5294 | static int got_NegDReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5295 | { |
9f4fe9ad | 5296 | struct drbd_peer_device *peer_device; |
b30ab791 | 5297 | struct drbd_device *device; |
e658983a | 5298 | struct p_block_ack *p = pi->data; |
b411b363 PR |
5299 | sector_t sector = be64_to_cpu(p->sector); |
5300 | ||
9f4fe9ad AG |
5301 | peer_device = conn_peer_device(connection, pi->vnr); |
5302 | if (!peer_device) | |
2735a594 | 5303 | return -EIO; |
9f4fe9ad | 5304 | device = peer_device->device; |
1952e916 | 5305 | |
69a22773 | 5306 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
7be8da07 | 5307 | |
d0180171 | 5308 | drbd_err(device, "Got NegDReply; Sector %llus, len %u.\n", |
b411b363 PR |
5309 | (unsigned long long)sector, be32_to_cpu(p->blksize)); |
5310 | ||
b30ab791 AG |
5311 | return validate_req_change_req_state(device, p->block_id, sector, |
5312 | &device->read_requests, __func__, | |
2735a594 | 5313 | NEG_ACKED, false); |
b411b363 PR |
5314 | } |
5315 | ||
bde89a9e | 5316 | static int got_NegRSDReply(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5317 | { |
9f4fe9ad | 5318 | struct drbd_peer_device *peer_device; |
b30ab791 | 5319 | struct drbd_device *device; |
b411b363 PR |
5320 | sector_t sector; |
5321 | int size; | |
e658983a | 5322 | struct p_block_ack *p = pi->data; |
1952e916 | 5323 | |
9f4fe9ad AG |
5324 | peer_device = conn_peer_device(connection, pi->vnr); |
5325 | if (!peer_device) | |
2735a594 | 5326 | return -EIO; |
9f4fe9ad | 5327 | device = peer_device->device; |
b411b363 PR |
5328 | |
5329 | sector = be64_to_cpu(p->sector); | |
5330 | size = be32_to_cpu(p->blksize); | |
b411b363 | 5331 | |
69a22773 | 5332 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 | 5333 | |
b30ab791 | 5334 | dec_rs_pending(device); |
b411b363 | 5335 | |
b30ab791 AG |
5336 | if (get_ldev_if_state(device, D_FAILED)) { |
5337 | drbd_rs_complete_io(device, sector); | |
e05e1e59 | 5338 | switch (pi->cmd) { |
d612d309 | 5339 | case P_NEG_RS_DREPLY: |
b30ab791 | 5340 | drbd_rs_failed_io(device, sector, size); |
d612d309 PR |
5341 | case P_RS_CANCEL: |
5342 | break; | |
5343 | default: | |
2735a594 | 5344 | BUG(); |
d612d309 | 5345 | } |
b30ab791 | 5346 | put_ldev(device); |
b411b363 PR |
5347 | } |
5348 | ||
2735a594 | 5349 | return 0; |
b411b363 PR |
5350 | } |
5351 | ||
bde89a9e | 5352 | static int got_BarrierAck(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5353 | { |
e658983a | 5354 | struct p_barrier_ack *p = pi->data; |
c06ece6b | 5355 | struct drbd_peer_device *peer_device; |
9ed57dcb | 5356 | int vnr; |
1952e916 | 5357 | |
bde89a9e | 5358 | tl_release(connection, p->barrier, be32_to_cpu(p->set_size)); |
b411b363 | 5359 | |
9ed57dcb | 5360 | rcu_read_lock(); |
c06ece6b AG |
5361 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
5362 | struct drbd_device *device = peer_device->device; | |
5363 | ||
b30ab791 AG |
5364 | if (device->state.conn == C_AHEAD && |
5365 | atomic_read(&device->ap_in_flight) == 0 && | |
5366 | !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &device->flags)) { | |
5367 | device->start_resync_timer.expires = jiffies + HZ; | |
5368 | add_timer(&device->start_resync_timer); | |
9ed57dcb | 5369 | } |
c4752ef1 | 5370 | } |
9ed57dcb | 5371 | rcu_read_unlock(); |
c4752ef1 | 5372 | |
2735a594 | 5373 | return 0; |
b411b363 PR |
5374 | } |
5375 | ||
bde89a9e | 5376 | static int got_OVResult(struct drbd_connection *connection, struct packet_info *pi) |
b411b363 | 5377 | { |
9f4fe9ad | 5378 | struct drbd_peer_device *peer_device; |
b30ab791 | 5379 | struct drbd_device *device; |
e658983a | 5380 | struct p_block_ack *p = pi->data; |
84b8c06b | 5381 | struct drbd_device_work *dw; |
b411b363 PR |
5382 | sector_t sector; |
5383 | int size; | |
5384 | ||
9f4fe9ad AG |
5385 | peer_device = conn_peer_device(connection, pi->vnr); |
5386 | if (!peer_device) | |
2735a594 | 5387 | return -EIO; |
9f4fe9ad | 5388 | device = peer_device->device; |
1952e916 | 5389 | |
b411b363 PR |
5390 | sector = be64_to_cpu(p->sector); |
5391 | size = be32_to_cpu(p->blksize); | |
5392 | ||
69a22773 | 5393 | update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); |
b411b363 PR |
5394 | |
5395 | if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) | |
b30ab791 | 5396 | drbd_ov_out_of_sync_found(device, sector, size); |
b411b363 | 5397 | else |
b30ab791 | 5398 | ov_out_of_sync_print(device); |
b411b363 | 5399 | |
b30ab791 | 5400 | if (!get_ldev(device)) |
2735a594 | 5401 | return 0; |
1d53f09e | 5402 | |
b30ab791 AG |
5403 | drbd_rs_complete_io(device, sector); |
5404 | dec_rs_pending(device); | |
b411b363 | 5405 | |
b30ab791 | 5406 | --device->ov_left; |
ea5442af LE |
5407 | |
5408 | /* let's advance progress step marks only for every other megabyte */ | |
b30ab791 AG |
5409 | if ((device->ov_left & 0x200) == 0x200) |
5410 | drbd_advance_rs_marks(device, device->ov_left); | |
ea5442af | 5411 | |
b30ab791 | 5412 | if (device->ov_left == 0) { |
84b8c06b AG |
5413 | dw = kmalloc(sizeof(*dw), GFP_NOIO); |
5414 | if (dw) { | |
5415 | dw->w.cb = w_ov_finished; | |
5416 | dw->device = device; | |
5417 | drbd_queue_work(&peer_device->connection->sender_work, &dw->w); | |
b411b363 | 5418 | } else { |
84b8c06b | 5419 | drbd_err(device, "kmalloc(dw) failed."); |
b30ab791 AG |
5420 | ov_out_of_sync_print(device); |
5421 | drbd_resync_finished(device); | |
b411b363 PR |
5422 | } |
5423 | } | |
b30ab791 | 5424 | put_ldev(device); |
2735a594 | 5425 | return 0; |
b411b363 PR |
5426 | } |
5427 | ||
bde89a9e | 5428 | static int got_skip(struct drbd_connection *connection, struct packet_info *pi) |
0ced55a3 | 5429 | { |
2735a594 | 5430 | return 0; |
b411b363 PR |
5431 | } |
5432 | ||
bde89a9e | 5433 | static int connection_finish_peer_reqs(struct drbd_connection *connection) |
0ced55a3 | 5434 | { |
c06ece6b | 5435 | struct drbd_peer_device *peer_device; |
c141ebda | 5436 | int vnr, not_empty = 0; |
32862ec7 PR |
5437 | |
5438 | do { | |
bde89a9e | 5439 | clear_bit(SIGNAL_ASENDER, &connection->flags); |
32862ec7 | 5440 | flush_signals(current); |
c141ebda PR |
5441 | |
5442 | rcu_read_lock(); | |
c06ece6b AG |
5443 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
5444 | struct drbd_device *device = peer_device->device; | |
b30ab791 | 5445 | kref_get(&device->kref); |
c141ebda | 5446 | rcu_read_unlock(); |
b30ab791 | 5447 | if (drbd_finish_peer_reqs(device)) { |
05a10ec7 | 5448 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda | 5449 | return 1; |
d3fcb490 | 5450 | } |
05a10ec7 | 5451 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda | 5452 | rcu_read_lock(); |
082a3439 | 5453 | } |
bde89a9e | 5454 | set_bit(SIGNAL_ASENDER, &connection->flags); |
082a3439 | 5455 | |
0500813f | 5456 | spin_lock_irq(&connection->resource->req_lock); |
c06ece6b AG |
5457 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
5458 | struct drbd_device *device = peer_device->device; | |
b30ab791 | 5459 | not_empty = !list_empty(&device->done_ee); |
082a3439 PR |
5460 | if (not_empty) |
5461 | break; | |
5462 | } | |
0500813f | 5463 | spin_unlock_irq(&connection->resource->req_lock); |
c141ebda | 5464 | rcu_read_unlock(); |
32862ec7 PR |
5465 | } while (not_empty); |
5466 | ||
5467 | return 0; | |
0ced55a3 PR |
5468 | } |
5469 | ||
b411b363 PR |
5470 | struct asender_cmd { |
5471 | size_t pkt_size; | |
bde89a9e | 5472 | int (*fn)(struct drbd_connection *connection, struct packet_info *); |
b411b363 PR |
5473 | }; |
5474 | ||
7201b972 | 5475 | static struct asender_cmd asender_tbl[] = { |
e658983a AG |
5476 | [P_PING] = { 0, got_Ping }, |
5477 | [P_PING_ACK] = { 0, got_PingAck }, | |
b411b363 PR |
5478 | [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, |
5479 | [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
5480 | [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
d4dabbe2 | 5481 | [P_SUPERSEDED] = { sizeof(struct p_block_ack), got_BlockAck }, |
b411b363 PR |
5482 | [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, |
5483 | [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, | |
1952e916 | 5484 | [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply }, |
b411b363 PR |
5485 | [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, |
5486 | [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, | |
5487 | [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, | |
5488 | [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, | |
02918be2 | 5489 | [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, |
1952e916 AG |
5490 | [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply }, |
5491 | [P_CONN_ST_CHG_REPLY]={ sizeof(struct p_req_state_reply), got_conn_RqSReply }, | |
5492 | [P_RETRY_WRITE] = { sizeof(struct p_block_ack), got_BlockAck }, | |
7201b972 | 5493 | }; |
b411b363 PR |
5494 | |
5495 | int drbd_asender(struct drbd_thread *thi) | |
5496 | { | |
bde89a9e | 5497 | struct drbd_connection *connection = thi->connection; |
b411b363 | 5498 | struct asender_cmd *cmd = NULL; |
77351055 | 5499 | struct packet_info pi; |
257d0af6 | 5500 | int rv; |
bde89a9e | 5501 | void *buf = connection->meta.rbuf; |
b411b363 | 5502 | int received = 0; |
bde89a9e | 5503 | unsigned int header_size = drbd_header_size(connection); |
52b061a4 | 5504 | int expect = header_size; |
44ed167d PR |
5505 | bool ping_timeout_active = false; |
5506 | struct net_conf *nc; | |
bb77d34e | 5507 | int ping_timeo, tcp_cork, ping_int; |
3990e04d | 5508 | struct sched_param param = { .sched_priority = 2 }; |
b411b363 | 5509 | |
3990e04d PR |
5510 | rv = sched_setscheduler(current, SCHED_RR, ¶m); |
5511 | if (rv < 0) | |
1ec861eb | 5512 | drbd_err(connection, "drbd_asender: ERROR set priority, ret=%d\n", rv); |
b411b363 | 5513 | |
e77a0a5c | 5514 | while (get_t_state(thi) == RUNNING) { |
80822284 | 5515 | drbd_thread_current_set_cpu(thi); |
b411b363 | 5516 | |
44ed167d | 5517 | rcu_read_lock(); |
bde89a9e | 5518 | nc = rcu_dereference(connection->net_conf); |
44ed167d | 5519 | ping_timeo = nc->ping_timeo; |
bb77d34e | 5520 | tcp_cork = nc->tcp_cork; |
44ed167d PR |
5521 | ping_int = nc->ping_int; |
5522 | rcu_read_unlock(); | |
5523 | ||
bde89a9e AG |
5524 | if (test_and_clear_bit(SEND_PING, &connection->flags)) { |
5525 | if (drbd_send_ping(connection)) { | |
1ec861eb | 5526 | drbd_err(connection, "drbd_send_ping has failed\n"); |
b411b363 | 5527 | goto reconnect; |
841ce241 | 5528 | } |
bde89a9e | 5529 | connection->meta.socket->sk->sk_rcvtimeo = ping_timeo * HZ / 10; |
44ed167d | 5530 | ping_timeout_active = true; |
b411b363 PR |
5531 | } |
5532 | ||
32862ec7 PR |
5533 | /* TODO: conditionally cork; it may hurt latency if we cork without |
5534 | much to send */ | |
bb77d34e | 5535 | if (tcp_cork) |
bde89a9e AG |
5536 | drbd_tcp_cork(connection->meta.socket); |
5537 | if (connection_finish_peer_reqs(connection)) { | |
1ec861eb | 5538 | drbd_err(connection, "connection_finish_peer_reqs() failed\n"); |
32862ec7 | 5539 | goto reconnect; |
b411b363 PR |
5540 | } |
5541 | /* but unconditionally uncork unless disabled */ | |
bb77d34e | 5542 | if (tcp_cork) |
bde89a9e | 5543 | drbd_tcp_uncork(connection->meta.socket); |
b411b363 PR |
5544 | |
5545 | /* short circuit, recv_msg would return EINTR anyways. */ | |
5546 | if (signal_pending(current)) | |
5547 | continue; | |
5548 | ||
bde89a9e AG |
5549 | rv = drbd_recv_short(connection->meta.socket, buf, expect-received, 0); |
5550 | clear_bit(SIGNAL_ASENDER, &connection->flags); | |
b411b363 PR |
5551 | |
5552 | flush_signals(current); | |
5553 | ||
5554 | /* Note: | |
5555 | * -EINTR (on meta) we got a signal | |
5556 | * -EAGAIN (on meta) rcvtimeo expired | |
5557 | * -ECONNRESET other side closed the connection | |
5558 | * -ERESTARTSYS (on data) we got a signal | |
5559 | * rv < 0 other than above: unexpected error! | |
5560 | * rv == expected: full header or command | |
5561 | * rv < expected: "woken" by signal during receive | |
5562 | * rv == 0 : "connection shut down by peer" | |
5563 | */ | |
abde9cc6 | 5564 | received_more: |
b411b363 PR |
5565 | if (likely(rv > 0)) { |
5566 | received += rv; | |
5567 | buf += rv; | |
5568 | } else if (rv == 0) { | |
bde89a9e | 5569 | if (test_bit(DISCONNECT_SENT, &connection->flags)) { |
b66623e3 PR |
5570 | long t; |
5571 | rcu_read_lock(); | |
bde89a9e | 5572 | t = rcu_dereference(connection->net_conf)->ping_timeo * HZ/10; |
b66623e3 PR |
5573 | rcu_read_unlock(); |
5574 | ||
bde89a9e AG |
5575 | t = wait_event_timeout(connection->ping_wait, |
5576 | connection->cstate < C_WF_REPORT_PARAMS, | |
b66623e3 | 5577 | t); |
599377ac PR |
5578 | if (t) |
5579 | break; | |
5580 | } | |
1ec861eb | 5581 | drbd_err(connection, "meta connection shut down by peer.\n"); |
b411b363 PR |
5582 | goto reconnect; |
5583 | } else if (rv == -EAGAIN) { | |
cb6518cb LE |
5584 | /* If the data socket received something meanwhile, |
5585 | * that is good enough: peer is still alive. */ | |
bde89a9e AG |
5586 | if (time_after(connection->last_received, |
5587 | jiffies - connection->meta.socket->sk->sk_rcvtimeo)) | |
cb6518cb | 5588 | continue; |
f36af18c | 5589 | if (ping_timeout_active) { |
1ec861eb | 5590 | drbd_err(connection, "PingAck did not arrive in time.\n"); |
b411b363 PR |
5591 | goto reconnect; |
5592 | } | |
bde89a9e | 5593 | set_bit(SEND_PING, &connection->flags); |
b411b363 PR |
5594 | continue; |
5595 | } else if (rv == -EINTR) { | |
5596 | continue; | |
5597 | } else { | |
1ec861eb | 5598 | drbd_err(connection, "sock_recvmsg returned %d\n", rv); |
b411b363 PR |
5599 | goto reconnect; |
5600 | } | |
5601 | ||
5602 | if (received == expect && cmd == NULL) { | |
bde89a9e | 5603 | if (decode_header(connection, connection->meta.rbuf, &pi)) |
b411b363 | 5604 | goto reconnect; |
7201b972 | 5605 | cmd = &asender_tbl[pi.cmd]; |
1952e916 | 5606 | if (pi.cmd >= ARRAY_SIZE(asender_tbl) || !cmd->fn) { |
1ec861eb | 5607 | drbd_err(connection, "Unexpected meta packet %s (0x%04x)\n", |
2fcb8f30 | 5608 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
5609 | goto disconnect; |
5610 | } | |
e658983a | 5611 | expect = header_size + cmd->pkt_size; |
52b061a4 | 5612 | if (pi.size != expect - header_size) { |
1ec861eb | 5613 | drbd_err(connection, "Wrong packet size on meta (c: %d, l: %d)\n", |
77351055 | 5614 | pi.cmd, pi.size); |
b411b363 | 5615 | goto reconnect; |
257d0af6 | 5616 | } |
b411b363 PR |
5617 | } |
5618 | if (received == expect) { | |
2735a594 | 5619 | bool err; |
a4fbda8e | 5620 | |
bde89a9e | 5621 | err = cmd->fn(connection, &pi); |
2735a594 | 5622 | if (err) { |
1ec861eb | 5623 | drbd_err(connection, "%pf failed\n", cmd->fn); |
b411b363 | 5624 | goto reconnect; |
1952e916 | 5625 | } |
b411b363 | 5626 | |
bde89a9e | 5627 | connection->last_received = jiffies; |
f36af18c | 5628 | |
44ed167d PR |
5629 | if (cmd == &asender_tbl[P_PING_ACK]) { |
5630 | /* restore idle timeout */ | |
bde89a9e | 5631 | connection->meta.socket->sk->sk_rcvtimeo = ping_int * HZ; |
44ed167d PR |
5632 | ping_timeout_active = false; |
5633 | } | |
f36af18c | 5634 | |
bde89a9e | 5635 | buf = connection->meta.rbuf; |
b411b363 | 5636 | received = 0; |
52b061a4 | 5637 | expect = header_size; |
b411b363 PR |
5638 | cmd = NULL; |
5639 | } | |
abde9cc6 LE |
5640 | if (test_bit(SEND_PING, &connection->flags)) |
5641 | continue; | |
5642 | rv = drbd_recv_short(connection->meta.socket, buf, expect-received, MSG_DONTWAIT); | |
5643 | if (rv > 0) | |
5644 | goto received_more; | |
b411b363 PR |
5645 | } |
5646 | ||
5647 | if (0) { | |
5648 | reconnect: | |
bde89a9e AG |
5649 | conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); |
5650 | conn_md_sync(connection); | |
b411b363 PR |
5651 | } |
5652 | if (0) { | |
5653 | disconnect: | |
bde89a9e | 5654 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 5655 | } |
bde89a9e | 5656 | clear_bit(SIGNAL_ASENDER, &connection->flags); |
b411b363 | 5657 | |
1ec861eb | 5658 | drbd_info(connection, "asender terminated\n"); |
b411b363 PR |
5659 | |
5660 | return 0; | |
5661 | } |