Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_worker.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 | ||
84b8c06b | 24 | */ |
b411b363 | 25 | |
b411b363 | 26 | #include <linux/module.h> |
b411b363 PR |
27 | #include <linux/drbd.h> |
28 | #include <linux/sched.h> | |
b411b363 PR |
29 | #include <linux/wait.h> |
30 | #include <linux/mm.h> | |
31 | #include <linux/memcontrol.h> | |
32 | #include <linux/mm_inline.h> | |
33 | #include <linux/slab.h> | |
34 | #include <linux/random.h> | |
b411b363 PR |
35 | #include <linux/string.h> |
36 | #include <linux/scatterlist.h> | |
37 | ||
38 | #include "drbd_int.h" | |
a3603a6e | 39 | #include "drbd_protocol.h" |
b411b363 | 40 | #include "drbd_req.h" |
b411b363 | 41 | |
d448a2e1 AG |
42 | static int make_ov_request(struct drbd_device *, int); |
43 | static int make_resync_request(struct drbd_device *, int); | |
b411b363 | 44 | |
c5a91619 | 45 | /* endio handlers: |
ed15b795 | 46 | * drbd_md_endio (defined here) |
fcefa62e AG |
47 | * drbd_request_endio (defined here) |
48 | * drbd_peer_request_endio (defined here) | |
ed15b795 | 49 | * drbd_bm_endio (defined in drbd_bitmap.c) |
c5a91619 | 50 | * |
b411b363 PR |
51 | * For all these callbacks, note the following: |
52 | * The callbacks will be called in irq context by the IDE drivers, | |
53 | * and in Softirqs/Tasklets/BH context by the SCSI drivers. | |
54 | * Try to get the locking right :) | |
55 | * | |
56 | */ | |
57 | ||
b411b363 PR |
58 | /* used for synchronous meta data and bitmap IO |
59 | * submitted by drbd_md_sync_page_io() | |
60 | */ | |
4246a0b6 | 61 | void drbd_md_endio(struct bio *bio) |
b411b363 | 62 | { |
b30ab791 | 63 | struct drbd_device *device; |
b411b363 | 64 | |
e37d2438 | 65 | device = bio->bi_private; |
4246a0b6 | 66 | device->md_io.error = bio->bi_error; |
b411b363 | 67 | |
0cfac5dd PR |
68 | /* We grabbed an extra reference in _drbd_md_sync_page_io() to be able |
69 | * to timeout on the lower level device, and eventually detach from it. | |
70 | * If this io completion runs after that timeout expired, this | |
71 | * drbd_md_put_buffer() may allow us to finally try and re-attach. | |
72 | * During normal operation, this only puts that extra reference | |
73 | * down to 1 again. | |
74 | * Make sure we first drop the reference, and only then signal | |
75 | * completion, or we may (in drbd_al_read_log()) cycle so fast into the | |
76 | * next drbd_md_sync_page_io(), that we trigger the | |
b30ab791 | 77 | * ASSERT(atomic_read(&device->md_io_in_use) == 1) there. |
0cfac5dd | 78 | */ |
b30ab791 | 79 | drbd_md_put_buffer(device); |
e37d2438 | 80 | device->md_io.done = 1; |
b30ab791 | 81 | wake_up(&device->misc_wait); |
cdfda633 | 82 | bio_put(bio); |
b30ab791 AG |
83 | if (device->ldev) /* special case: drbd_md_read() during drbd_adm_attach() */ |
84 | put_ldev(device); | |
b411b363 PR |
85 | } |
86 | ||
87 | /* reads on behalf of the partner, | |
88 | * "submitted" by the receiver | |
89 | */ | |
a186e478 | 90 | static void drbd_endio_read_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
91 | { |
92 | unsigned long flags = 0; | |
6780139c AG |
93 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
94 | struct drbd_device *device = peer_device->device; | |
b411b363 | 95 | |
0500813f | 96 | spin_lock_irqsave(&device->resource->req_lock, flags); |
b30ab791 | 97 | device->read_cnt += peer_req->i.size >> 9; |
a8cd15ba | 98 | list_del(&peer_req->w.list); |
b30ab791 AG |
99 | if (list_empty(&device->read_ee)) |
100 | wake_up(&device->ee_wait); | |
db830c46 | 101 | if (test_bit(__EE_WAS_ERROR, &peer_req->flags)) |
b30ab791 | 102 | __drbd_chk_io_error(device, DRBD_READ_ERROR); |
0500813f | 103 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b411b363 | 104 | |
6780139c | 105 | drbd_queue_work(&peer_device->connection->sender_work, &peer_req->w); |
b30ab791 | 106 | put_ldev(device); |
b411b363 PR |
107 | } |
108 | ||
109 | /* writes on behalf of the partner, or resync writes, | |
45bb912b | 110 | * "submitted" by the receiver, final stage. */ |
a0fb3c47 | 111 | void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
112 | { |
113 | unsigned long flags = 0; | |
6780139c AG |
114 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
115 | struct drbd_device *device = peer_device->device; | |
668700b4 | 116 | struct drbd_connection *connection = peer_device->connection; |
181286ad | 117 | struct drbd_interval i; |
b411b363 | 118 | int do_wake; |
579b57ed | 119 | u64 block_id; |
b411b363 | 120 | int do_al_complete_io; |
b411b363 | 121 | |
db830c46 | 122 | /* after we moved peer_req to done_ee, |
b411b363 PR |
123 | * we may no longer access it, |
124 | * it may be freed/reused already! | |
125 | * (as soon as we release the req_lock) */ | |
181286ad | 126 | i = peer_req->i; |
db830c46 AG |
127 | do_al_complete_io = peer_req->flags & EE_CALL_AL_COMPLETE_IO; |
128 | block_id = peer_req->block_id; | |
21ae5d7f | 129 | peer_req->flags &= ~EE_CALL_AL_COMPLETE_IO; |
b411b363 | 130 | |
0500813f | 131 | spin_lock_irqsave(&device->resource->req_lock, flags); |
b30ab791 | 132 | device->writ_cnt += peer_req->i.size >> 9; |
a8cd15ba | 133 | list_move_tail(&peer_req->w.list, &device->done_ee); |
b411b363 | 134 | |
bb3bfe96 | 135 | /* |
5e472264 | 136 | * Do not remove from the write_requests tree here: we did not send the |
bb3bfe96 AG |
137 | * Ack yet and did not wake possibly waiting conflicting requests. |
138 | * Removed from the tree from "drbd_process_done_ee" within the | |
84b8c06b | 139 | * appropriate dw.cb (e_end_block/e_end_resync_block) or from |
bb3bfe96 AG |
140 | * _drbd_clear_done_ee. |
141 | */ | |
b411b363 | 142 | |
b30ab791 | 143 | do_wake = list_empty(block_id == ID_SYNCER ? &device->sync_ee : &device->active_ee); |
b411b363 | 144 | |
a0fb3c47 LE |
145 | /* FIXME do we want to detach for failed REQ_DISCARD? |
146 | * ((peer_req->flags & (EE_WAS_ERROR|EE_IS_TRIM)) == EE_WAS_ERROR) */ | |
147 | if (peer_req->flags & EE_WAS_ERROR) | |
b30ab791 | 148 | __drbd_chk_io_error(device, DRBD_WRITE_ERROR); |
668700b4 PR |
149 | |
150 | if (connection->cstate >= C_WF_REPORT_PARAMS) { | |
151 | kref_get(&device->kref); /* put is in drbd_send_acks_wf() */ | |
152 | if (!queue_work(connection->ack_sender, &peer_device->send_acks_work)) | |
153 | kref_put(&device->kref, drbd_destroy_device); | |
154 | } | |
0500813f | 155 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b411b363 | 156 | |
579b57ed | 157 | if (block_id == ID_SYNCER) |
b30ab791 | 158 | drbd_rs_complete_io(device, i.sector); |
b411b363 PR |
159 | |
160 | if (do_wake) | |
b30ab791 | 161 | wake_up(&device->ee_wait); |
b411b363 PR |
162 | |
163 | if (do_al_complete_io) | |
b30ab791 | 164 | drbd_al_complete_io(device, &i); |
b411b363 | 165 | |
b30ab791 | 166 | put_ldev(device); |
45bb912b | 167 | } |
b411b363 | 168 | |
45bb912b LE |
169 | /* writes on behalf of the partner, or resync writes, |
170 | * "submitted" by the receiver. | |
171 | */ | |
4246a0b6 | 172 | void drbd_peer_request_endio(struct bio *bio) |
45bb912b | 173 | { |
db830c46 | 174 | struct drbd_peer_request *peer_req = bio->bi_private; |
a8cd15ba | 175 | struct drbd_device *device = peer_req->peer_device->device; |
7e5fec31 FF |
176 | bool is_write = bio_data_dir(bio) == WRITE; |
177 | bool is_discard = !!(bio_op(bio) == REQ_OP_DISCARD); | |
45bb912b | 178 | |
4246a0b6 | 179 | if (bio->bi_error && __ratelimit(&drbd_ratelimit_state)) |
d0180171 | 180 | drbd_warn(device, "%s: error=%d s=%llus\n", |
a0fb3c47 | 181 | is_write ? (is_discard ? "discard" : "write") |
4246a0b6 | 182 | : "read", bio->bi_error, |
db830c46 | 183 | (unsigned long long)peer_req->i.sector); |
45bb912b | 184 | |
4246a0b6 | 185 | if (bio->bi_error) |
db830c46 | 186 | set_bit(__EE_WAS_ERROR, &peer_req->flags); |
45bb912b LE |
187 | |
188 | bio_put(bio); /* no need for the bio anymore */ | |
db830c46 | 189 | if (atomic_dec_and_test(&peer_req->pending_bios)) { |
45bb912b | 190 | if (is_write) |
db830c46 | 191 | drbd_endio_write_sec_final(peer_req); |
45bb912b | 192 | else |
db830c46 | 193 | drbd_endio_read_sec_final(peer_req); |
45bb912b | 194 | } |
b411b363 PR |
195 | } |
196 | ||
142207f7 LE |
197 | void drbd_panic_after_delayed_completion_of_aborted_request(struct drbd_device *device) |
198 | { | |
199 | panic("drbd%u %s/%u potential random memory corruption caused by delayed completion of aborted local request\n", | |
200 | device->minor, device->resource->name, device->vnr); | |
201 | } | |
202 | ||
b411b363 PR |
203 | /* read, readA or write requests on R_PRIMARY coming from drbd_make_request |
204 | */ | |
4246a0b6 | 205 | void drbd_request_endio(struct bio *bio) |
b411b363 | 206 | { |
a115413d | 207 | unsigned long flags; |
b411b363 | 208 | struct drbd_request *req = bio->bi_private; |
84b8c06b | 209 | struct drbd_device *device = req->device; |
a115413d | 210 | struct bio_and_error m; |
b411b363 | 211 | enum drbd_req_event what; |
1b6dd252 PR |
212 | |
213 | /* If this request was aborted locally before, | |
214 | * but now was completed "successfully", | |
215 | * chances are that this caused arbitrary data corruption. | |
216 | * | |
217 | * "aborting" requests, or force-detaching the disk, is intended for | |
218 | * completely blocked/hung local backing devices which do no longer | |
219 | * complete requests at all, not even do error completions. In this | |
220 | * situation, usually a hard-reset and failover is the only way out. | |
221 | * | |
222 | * By "aborting", basically faking a local error-completion, | |
223 | * we allow for a more graceful swichover by cleanly migrating services. | |
224 | * Still the affected node has to be rebooted "soon". | |
225 | * | |
226 | * By completing these requests, we allow the upper layers to re-use | |
227 | * the associated data pages. | |
228 | * | |
229 | * If later the local backing device "recovers", and now DMAs some data | |
230 | * from disk into the original request pages, in the best case it will | |
231 | * just put random data into unused pages; but typically it will corrupt | |
232 | * meanwhile completely unrelated data, causing all sorts of damage. | |
233 | * | |
234 | * Which means delayed successful completion, | |
235 | * especially for READ requests, | |
236 | * is a reason to panic(). | |
237 | * | |
238 | * We assume that a delayed *error* completion is OK, | |
239 | * though we still will complain noisily about it. | |
240 | */ | |
241 | if (unlikely(req->rq_state & RQ_LOCAL_ABORTED)) { | |
242 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 243 | drbd_emerg(device, "delayed completion of aborted local request; disk-timeout may be too aggressive\n"); |
1b6dd252 | 244 | |
4246a0b6 | 245 | if (!bio->bi_error) |
142207f7 | 246 | drbd_panic_after_delayed_completion_of_aborted_request(device); |
1b6dd252 PR |
247 | } |
248 | ||
b411b363 | 249 | /* to avoid recursion in __req_mod */ |
4246a0b6 | 250 | if (unlikely(bio->bi_error)) { |
70246286 CH |
251 | switch (bio_op(bio)) { |
252 | case REQ_OP_DISCARD: | |
253 | if (bio->bi_error == -EOPNOTSUPP) | |
254 | what = DISCARD_COMPLETED_NOTSUPP; | |
255 | else | |
256 | what = DISCARD_COMPLETED_WITH_ERROR; | |
257 | break; | |
258 | case REQ_OP_READ: | |
1eff9d32 | 259 | if (bio->bi_opf & REQ_RAHEAD) |
70246286 CH |
260 | what = READ_AHEAD_COMPLETED_WITH_ERROR; |
261 | else | |
262 | what = READ_COMPLETED_WITH_ERROR; | |
263 | break; | |
264 | default: | |
265 | what = WRITE_COMPLETED_WITH_ERROR; | |
266 | break; | |
267 | } | |
268 | } else { | |
8554df1c | 269 | what = COMPLETED_OK; |
70246286 | 270 | } |
b411b363 PR |
271 | |
272 | bio_put(req->private_bio); | |
4246a0b6 | 273 | req->private_bio = ERR_PTR(bio->bi_error); |
b411b363 | 274 | |
a115413d | 275 | /* not req_mod(), we need irqsave here! */ |
0500813f | 276 | spin_lock_irqsave(&device->resource->req_lock, flags); |
a115413d | 277 | __req_mod(req, what, &m); |
0500813f | 278 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b30ab791 | 279 | put_ldev(device); |
a115413d LE |
280 | |
281 | if (m.bio) | |
b30ab791 | 282 | complete_master_bio(device, &m); |
b411b363 PR |
283 | } |
284 | ||
9534d671 | 285 | void drbd_csum_ee(struct crypto_ahash *tfm, struct drbd_peer_request *peer_req, void *digest) |
45bb912b | 286 | { |
9534d671 | 287 | AHASH_REQUEST_ON_STACK(req, tfm); |
45bb912b | 288 | struct scatterlist sg; |
db830c46 | 289 | struct page *page = peer_req->pages; |
45bb912b LE |
290 | struct page *tmp; |
291 | unsigned len; | |
292 | ||
9534d671 HX |
293 | ahash_request_set_tfm(req, tfm); |
294 | ahash_request_set_callback(req, 0, NULL, NULL); | |
45bb912b LE |
295 | |
296 | sg_init_table(&sg, 1); | |
9534d671 | 297 | crypto_ahash_init(req); |
45bb912b LE |
298 | |
299 | while ((tmp = page_chain_next(page))) { | |
300 | /* all but the last page will be fully used */ | |
301 | sg_set_page(&sg, page, PAGE_SIZE, 0); | |
9534d671 HX |
302 | ahash_request_set_crypt(req, &sg, NULL, sg.length); |
303 | crypto_ahash_update(req); | |
45bb912b LE |
304 | page = tmp; |
305 | } | |
306 | /* and now the last, possibly only partially used page */ | |
db830c46 | 307 | len = peer_req->i.size & (PAGE_SIZE - 1); |
45bb912b | 308 | sg_set_page(&sg, page, len ?: PAGE_SIZE, 0); |
9534d671 HX |
309 | ahash_request_set_crypt(req, &sg, digest, sg.length); |
310 | crypto_ahash_finup(req); | |
311 | ahash_request_zero(req); | |
45bb912b LE |
312 | } |
313 | ||
9534d671 | 314 | void drbd_csum_bio(struct crypto_ahash *tfm, struct bio *bio, void *digest) |
b411b363 | 315 | { |
9534d671 | 316 | AHASH_REQUEST_ON_STACK(req, tfm); |
b411b363 | 317 | struct scatterlist sg; |
7988613b KO |
318 | struct bio_vec bvec; |
319 | struct bvec_iter iter; | |
b411b363 | 320 | |
9534d671 HX |
321 | ahash_request_set_tfm(req, tfm); |
322 | ahash_request_set_callback(req, 0, NULL, NULL); | |
b411b363 PR |
323 | |
324 | sg_init_table(&sg, 1); | |
9534d671 | 325 | crypto_ahash_init(req); |
b411b363 | 326 | |
7988613b KO |
327 | bio_for_each_segment(bvec, bio, iter) { |
328 | sg_set_page(&sg, bvec.bv_page, bvec.bv_len, bvec.bv_offset); | |
9534d671 HX |
329 | ahash_request_set_crypt(req, &sg, NULL, sg.length); |
330 | crypto_ahash_update(req); | |
9104d31a LE |
331 | /* REQ_OP_WRITE_SAME has only one segment, |
332 | * checksum the payload only once. */ | |
333 | if (bio_op(bio) == REQ_OP_WRITE_SAME) | |
334 | break; | |
b411b363 | 335 | } |
9534d671 HX |
336 | ahash_request_set_crypt(req, NULL, digest, 0); |
337 | crypto_ahash_final(req); | |
338 | ahash_request_zero(req); | |
b411b363 PR |
339 | } |
340 | ||
9676c760 | 341 | /* MAYBE merge common code with w_e_end_ov_req */ |
99920dc5 | 342 | static int w_e_send_csum(struct drbd_work *w, int cancel) |
b411b363 | 343 | { |
a8cd15ba | 344 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
345 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
346 | struct drbd_device *device = peer_device->device; | |
b411b363 PR |
347 | int digest_size; |
348 | void *digest; | |
99920dc5 | 349 | int err = 0; |
b411b363 | 350 | |
53ea4331 LE |
351 | if (unlikely(cancel)) |
352 | goto out; | |
b411b363 | 353 | |
9676c760 | 354 | if (unlikely((peer_req->flags & EE_WAS_ERROR) != 0)) |
53ea4331 | 355 | goto out; |
b411b363 | 356 | |
9534d671 | 357 | digest_size = crypto_ahash_digestsize(peer_device->connection->csums_tfm); |
53ea4331 LE |
358 | digest = kmalloc(digest_size, GFP_NOIO); |
359 | if (digest) { | |
db830c46 AG |
360 | sector_t sector = peer_req->i.sector; |
361 | unsigned int size = peer_req->i.size; | |
6780139c | 362 | drbd_csum_ee(peer_device->connection->csums_tfm, peer_req, digest); |
9676c760 | 363 | /* Free peer_req and pages before send. |
53ea4331 LE |
364 | * In case we block on congestion, we could otherwise run into |
365 | * some distributed deadlock, if the other side blocks on | |
366 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 367 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 368 | drbd_free_peer_req(device, peer_req); |
db830c46 | 369 | peer_req = NULL; |
b30ab791 | 370 | inc_rs_pending(device); |
6780139c | 371 | err = drbd_send_drequest_csum(peer_device, sector, size, |
db1b0b72 AG |
372 | digest, digest_size, |
373 | P_CSUM_RS_REQUEST); | |
53ea4331 LE |
374 | kfree(digest); |
375 | } else { | |
d0180171 | 376 | drbd_err(device, "kmalloc() of digest failed.\n"); |
99920dc5 | 377 | err = -ENOMEM; |
53ea4331 | 378 | } |
b411b363 | 379 | |
53ea4331 | 380 | out: |
db830c46 | 381 | if (peer_req) |
b30ab791 | 382 | drbd_free_peer_req(device, peer_req); |
b411b363 | 383 | |
99920dc5 | 384 | if (unlikely(err)) |
d0180171 | 385 | drbd_err(device, "drbd_send_drequest(..., csum) failed\n"); |
99920dc5 | 386 | return err; |
b411b363 PR |
387 | } |
388 | ||
389 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
390 | ||
69a22773 | 391 | static int read_for_csum(struct drbd_peer_device *peer_device, sector_t sector, int size) |
b411b363 | 392 | { |
69a22773 | 393 | struct drbd_device *device = peer_device->device; |
db830c46 | 394 | struct drbd_peer_request *peer_req; |
b411b363 | 395 | |
b30ab791 | 396 | if (!get_ldev(device)) |
80a40e43 | 397 | return -EIO; |
b411b363 PR |
398 | |
399 | /* GFP_TRY, because if there is no memory available right now, this may | |
400 | * be rescheduled for later. It is "only" background resync, after all. */ | |
69a22773 | 401 | peer_req = drbd_alloc_peer_req(peer_device, ID_SYNCER /* unused */, sector, |
9104d31a | 402 | size, size, GFP_TRY); |
db830c46 | 403 | if (!peer_req) |
80a40e43 | 404 | goto defer; |
b411b363 | 405 | |
a8cd15ba | 406 | peer_req->w.cb = w_e_send_csum; |
0500813f | 407 | spin_lock_irq(&device->resource->req_lock); |
b9ed7080 | 408 | list_add_tail(&peer_req->w.list, &device->read_ee); |
0500813f | 409 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 410 | |
b30ab791 | 411 | atomic_add(size >> 9, &device->rs_sect_ev); |
bb3cc85e MC |
412 | if (drbd_submit_peer_request(device, peer_req, REQ_OP_READ, 0, |
413 | DRBD_FAULT_RS_RD) == 0) | |
80a40e43 | 414 | return 0; |
b411b363 | 415 | |
10f6d992 LE |
416 | /* If it failed because of ENOMEM, retry should help. If it failed |
417 | * because bio_add_page failed (probably broken lower level driver), | |
418 | * retry may or may not help. | |
419 | * If it does not, you may need to force disconnect. */ | |
0500813f | 420 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 421 | list_del(&peer_req->w.list); |
0500813f | 422 | spin_unlock_irq(&device->resource->req_lock); |
22cc37a9 | 423 | |
b30ab791 | 424 | drbd_free_peer_req(device, peer_req); |
80a40e43 | 425 | defer: |
b30ab791 | 426 | put_ldev(device); |
80a40e43 | 427 | return -EAGAIN; |
b411b363 PR |
428 | } |
429 | ||
99920dc5 | 430 | int w_resync_timer(struct drbd_work *w, int cancel) |
b411b363 | 431 | { |
84b8c06b AG |
432 | struct drbd_device *device = |
433 | container_of(w, struct drbd_device, resync_work); | |
434 | ||
b30ab791 | 435 | switch (device->state.conn) { |
63106d3c | 436 | case C_VERIFY_S: |
d448a2e1 | 437 | make_ov_request(device, cancel); |
63106d3c PR |
438 | break; |
439 | case C_SYNC_TARGET: | |
d448a2e1 | 440 | make_resync_request(device, cancel); |
63106d3c | 441 | break; |
b411b363 PR |
442 | } |
443 | ||
99920dc5 | 444 | return 0; |
794abb75 PR |
445 | } |
446 | ||
447 | void resync_timer_fn(unsigned long data) | |
448 | { | |
b30ab791 | 449 | struct drbd_device *device = (struct drbd_device *) data; |
794abb75 | 450 | |
15e26f6a LE |
451 | drbd_queue_work_if_unqueued( |
452 | &first_peer_device(device)->connection->sender_work, | |
453 | &device->resync_work); | |
b411b363 PR |
454 | } |
455 | ||
778f271d PR |
456 | static void fifo_set(struct fifo_buffer *fb, int value) |
457 | { | |
458 | int i; | |
459 | ||
460 | for (i = 0; i < fb->size; i++) | |
f10f2623 | 461 | fb->values[i] = value; |
778f271d PR |
462 | } |
463 | ||
464 | static int fifo_push(struct fifo_buffer *fb, int value) | |
465 | { | |
466 | int ov; | |
467 | ||
468 | ov = fb->values[fb->head_index]; | |
469 | fb->values[fb->head_index++] = value; | |
470 | ||
471 | if (fb->head_index >= fb->size) | |
472 | fb->head_index = 0; | |
473 | ||
474 | return ov; | |
475 | } | |
476 | ||
477 | static void fifo_add_val(struct fifo_buffer *fb, int value) | |
478 | { | |
479 | int i; | |
480 | ||
481 | for (i = 0; i < fb->size; i++) | |
482 | fb->values[i] += value; | |
483 | } | |
484 | ||
9958c857 PR |
485 | struct fifo_buffer *fifo_alloc(int fifo_size) |
486 | { | |
487 | struct fifo_buffer *fb; | |
488 | ||
8747d30a | 489 | fb = kzalloc(sizeof(struct fifo_buffer) + sizeof(int) * fifo_size, GFP_NOIO); |
9958c857 PR |
490 | if (!fb) |
491 | return NULL; | |
492 | ||
493 | fb->head_index = 0; | |
494 | fb->size = fifo_size; | |
495 | fb->total = 0; | |
496 | ||
497 | return fb; | |
498 | } | |
499 | ||
0e49d7b0 | 500 | static int drbd_rs_controller(struct drbd_device *device, unsigned int sect_in) |
778f271d | 501 | { |
daeda1cc | 502 | struct disk_conf *dc; |
7f34f614 | 503 | unsigned int want; /* The number of sectors we want in-flight */ |
778f271d | 504 | int req_sect; /* Number of sectors to request in this turn */ |
7f34f614 | 505 | int correction; /* Number of sectors more we need in-flight */ |
778f271d PR |
506 | int cps; /* correction per invocation of drbd_rs_controller() */ |
507 | int steps; /* Number of time steps to plan ahead */ | |
508 | int curr_corr; | |
509 | int max_sect; | |
813472ce | 510 | struct fifo_buffer *plan; |
778f271d | 511 | |
b30ab791 AG |
512 | dc = rcu_dereference(device->ldev->disk_conf); |
513 | plan = rcu_dereference(device->rs_plan_s); | |
778f271d | 514 | |
813472ce | 515 | steps = plan->size; /* (dc->c_plan_ahead * 10 * SLEEP_TIME) / HZ; */ |
778f271d | 516 | |
b30ab791 | 517 | if (device->rs_in_flight + sect_in == 0) { /* At start of resync */ |
daeda1cc | 518 | want = ((dc->resync_rate * 2 * SLEEP_TIME) / HZ) * steps; |
778f271d | 519 | } else { /* normal path */ |
daeda1cc PR |
520 | want = dc->c_fill_target ? dc->c_fill_target : |
521 | sect_in * dc->c_delay_target * HZ / (SLEEP_TIME * 10); | |
778f271d PR |
522 | } |
523 | ||
b30ab791 | 524 | correction = want - device->rs_in_flight - plan->total; |
778f271d PR |
525 | |
526 | /* Plan ahead */ | |
527 | cps = correction / steps; | |
813472ce PR |
528 | fifo_add_val(plan, cps); |
529 | plan->total += cps * steps; | |
778f271d PR |
530 | |
531 | /* What we do in this step */ | |
813472ce PR |
532 | curr_corr = fifo_push(plan, 0); |
533 | plan->total -= curr_corr; | |
778f271d PR |
534 | |
535 | req_sect = sect_in + curr_corr; | |
536 | if (req_sect < 0) | |
537 | req_sect = 0; | |
538 | ||
daeda1cc | 539 | max_sect = (dc->c_max_rate * 2 * SLEEP_TIME) / HZ; |
778f271d PR |
540 | if (req_sect > max_sect) |
541 | req_sect = max_sect; | |
542 | ||
543 | /* | |
d0180171 | 544 | drbd_warn(device, "si=%u if=%d wa=%u co=%d st=%d cps=%d pl=%d cc=%d rs=%d\n", |
b30ab791 AG |
545 | sect_in, device->rs_in_flight, want, correction, |
546 | steps, cps, device->rs_planed, curr_corr, req_sect); | |
778f271d PR |
547 | */ |
548 | ||
549 | return req_sect; | |
550 | } | |
551 | ||
b30ab791 | 552 | static int drbd_rs_number_requests(struct drbd_device *device) |
e65f440d | 553 | { |
0e49d7b0 LE |
554 | unsigned int sect_in; /* Number of sectors that came in since the last turn */ |
555 | int number, mxb; | |
556 | ||
557 | sect_in = atomic_xchg(&device->rs_sect_in, 0); | |
558 | device->rs_in_flight -= sect_in; | |
813472ce PR |
559 | |
560 | rcu_read_lock(); | |
0e49d7b0 | 561 | mxb = drbd_get_max_buffers(device) / 2; |
b30ab791 | 562 | if (rcu_dereference(device->rs_plan_s)->size) { |
0e49d7b0 | 563 | number = drbd_rs_controller(device, sect_in) >> (BM_BLOCK_SHIFT - 9); |
b30ab791 | 564 | device->c_sync_rate = number * HZ * (BM_BLOCK_SIZE / 1024) / SLEEP_TIME; |
e65f440d | 565 | } else { |
b30ab791 AG |
566 | device->c_sync_rate = rcu_dereference(device->ldev->disk_conf)->resync_rate; |
567 | number = SLEEP_TIME * device->c_sync_rate / ((BM_BLOCK_SIZE / 1024) * HZ); | |
e65f440d | 568 | } |
813472ce | 569 | rcu_read_unlock(); |
e65f440d | 570 | |
0e49d7b0 LE |
571 | /* Don't have more than "max-buffers"/2 in-flight. |
572 | * Otherwise we may cause the remote site to stall on drbd_alloc_pages(), | |
573 | * potentially causing a distributed deadlock on congestion during | |
574 | * online-verify or (checksum-based) resync, if max-buffers, | |
575 | * socket buffer sizes and resync rate settings are mis-configured. */ | |
7f34f614 LE |
576 | |
577 | /* note that "number" is in units of "BM_BLOCK_SIZE" (which is 4k), | |
578 | * mxb (as used here, and in drbd_alloc_pages on the peer) is | |
579 | * "number of pages" (typically also 4k), | |
580 | * but "rs_in_flight" is in "sectors" (512 Byte). */ | |
581 | if (mxb - device->rs_in_flight/8 < number) | |
582 | number = mxb - device->rs_in_flight/8; | |
0e49d7b0 | 583 | |
e65f440d LE |
584 | return number; |
585 | } | |
586 | ||
44a4d551 | 587 | static int make_resync_request(struct drbd_device *const device, int cancel) |
b411b363 | 588 | { |
44a4d551 LE |
589 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
590 | struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL; | |
b411b363 PR |
591 | unsigned long bit; |
592 | sector_t sector; | |
b30ab791 | 593 | const sector_t capacity = drbd_get_capacity(device->this_bdev); |
1816a2b4 | 594 | int max_bio_size; |
e65f440d | 595 | int number, rollback_i, size; |
506afb62 | 596 | int align, requeue = 0; |
0f0601f4 | 597 | int i = 0; |
92d94ae6 | 598 | int discard_granularity = 0; |
b411b363 PR |
599 | |
600 | if (unlikely(cancel)) | |
99920dc5 | 601 | return 0; |
b411b363 | 602 | |
b30ab791 | 603 | if (device->rs_total == 0) { |
af85e8e8 | 604 | /* empty resync? */ |
b30ab791 | 605 | drbd_resync_finished(device); |
99920dc5 | 606 | return 0; |
af85e8e8 LE |
607 | } |
608 | ||
b30ab791 AG |
609 | if (!get_ldev(device)) { |
610 | /* Since we only need to access device->rsync a | |
611 | get_ldev_if_state(device,D_FAILED) would be sufficient, but | |
b411b363 PR |
612 | to continue resync with a broken disk makes no sense at |
613 | all */ | |
d0180171 | 614 | drbd_err(device, "Disk broke down during resync!\n"); |
99920dc5 | 615 | return 0; |
b411b363 PR |
616 | } |
617 | ||
9104d31a | 618 | if (connection->agreed_features & DRBD_FF_THIN_RESYNC) { |
92d94ae6 PR |
619 | rcu_read_lock(); |
620 | discard_granularity = rcu_dereference(device->ldev->disk_conf)->rs_discard_granularity; | |
621 | rcu_read_unlock(); | |
622 | } | |
623 | ||
b30ab791 AG |
624 | max_bio_size = queue_max_hw_sectors(device->rq_queue) << 9; |
625 | number = drbd_rs_number_requests(device); | |
0e49d7b0 | 626 | if (number <= 0) |
0f0601f4 | 627 | goto requeue; |
b411b363 | 628 | |
b411b363 | 629 | for (i = 0; i < number; i++) { |
506afb62 LE |
630 | /* Stop generating RS requests when half of the send buffer is filled, |
631 | * but notify TCP that we'd like to have more space. */ | |
44a4d551 LE |
632 | mutex_lock(&connection->data.mutex); |
633 | if (connection->data.socket) { | |
506afb62 LE |
634 | struct sock *sk = connection->data.socket->sk; |
635 | int queued = sk->sk_wmem_queued; | |
636 | int sndbuf = sk->sk_sndbuf; | |
637 | if (queued > sndbuf / 2) { | |
638 | requeue = 1; | |
639 | if (sk->sk_socket) | |
640 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
641 | } | |
642 | } else | |
643 | requeue = 1; | |
44a4d551 | 644 | mutex_unlock(&connection->data.mutex); |
506afb62 | 645 | if (requeue) |
b411b363 PR |
646 | goto requeue; |
647 | ||
648 | next_sector: | |
649 | size = BM_BLOCK_SIZE; | |
b30ab791 | 650 | bit = drbd_bm_find_next(device, device->bm_resync_fo); |
b411b363 | 651 | |
4b0715f0 | 652 | if (bit == DRBD_END_OF_BITMAP) { |
b30ab791 AG |
653 | device->bm_resync_fo = drbd_bm_bits(device); |
654 | put_ldev(device); | |
99920dc5 | 655 | return 0; |
b411b363 PR |
656 | } |
657 | ||
658 | sector = BM_BIT_TO_SECT(bit); | |
659 | ||
ad3fee79 | 660 | if (drbd_try_rs_begin_io(device, sector)) { |
b30ab791 | 661 | device->bm_resync_fo = bit; |
b411b363 PR |
662 | goto requeue; |
663 | } | |
b30ab791 | 664 | device->bm_resync_fo = bit + 1; |
b411b363 | 665 | |
b30ab791 AG |
666 | if (unlikely(drbd_bm_test_bit(device, bit) == 0)) { |
667 | drbd_rs_complete_io(device, sector); | |
b411b363 PR |
668 | goto next_sector; |
669 | } | |
670 | ||
1816a2b4 | 671 | #if DRBD_MAX_BIO_SIZE > BM_BLOCK_SIZE |
b411b363 PR |
672 | /* try to find some adjacent bits. |
673 | * we stop if we have already the maximum req size. | |
674 | * | |
675 | * Additionally always align bigger requests, in order to | |
676 | * be prepared for all stripe sizes of software RAIDs. | |
b411b363 PR |
677 | */ |
678 | align = 1; | |
d207450c | 679 | rollback_i = i; |
6377b923 | 680 | while (i < number) { |
1816a2b4 | 681 | if (size + BM_BLOCK_SIZE > max_bio_size) |
b411b363 PR |
682 | break; |
683 | ||
684 | /* Be always aligned */ | |
685 | if (sector & ((1<<(align+3))-1)) | |
686 | break; | |
687 | ||
92d94ae6 PR |
688 | if (discard_granularity && size == discard_granularity) |
689 | break; | |
690 | ||
b411b363 PR |
691 | /* do not cross extent boundaries */ |
692 | if (((bit+1) & BM_BLOCKS_PER_BM_EXT_MASK) == 0) | |
693 | break; | |
694 | /* now, is it actually dirty, after all? | |
695 | * caution, drbd_bm_test_bit is tri-state for some | |
696 | * obscure reason; ( b == 0 ) would get the out-of-band | |
697 | * only accidentally right because of the "oddly sized" | |
698 | * adjustment below */ | |
b30ab791 | 699 | if (drbd_bm_test_bit(device, bit+1) != 1) |
b411b363 PR |
700 | break; |
701 | bit++; | |
702 | size += BM_BLOCK_SIZE; | |
703 | if ((BM_BLOCK_SIZE << align) <= size) | |
704 | align++; | |
705 | i++; | |
706 | } | |
707 | /* if we merged some, | |
708 | * reset the offset to start the next drbd_bm_find_next from */ | |
709 | if (size > BM_BLOCK_SIZE) | |
b30ab791 | 710 | device->bm_resync_fo = bit + 1; |
b411b363 PR |
711 | #endif |
712 | ||
713 | /* adjust very last sectors, in case we are oddly sized */ | |
714 | if (sector + (size>>9) > capacity) | |
715 | size = (capacity-sector)<<9; | |
aaaba345 LE |
716 | |
717 | if (device->use_csums) { | |
44a4d551 | 718 | switch (read_for_csum(peer_device, sector, size)) { |
80a40e43 | 719 | case -EIO: /* Disk failure */ |
b30ab791 | 720 | put_ldev(device); |
99920dc5 | 721 | return -EIO; |
80a40e43 | 722 | case -EAGAIN: /* allocation failed, or ldev busy */ |
b30ab791 AG |
723 | drbd_rs_complete_io(device, sector); |
724 | device->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
d207450c | 725 | i = rollback_i; |
b411b363 | 726 | goto requeue; |
80a40e43 LE |
727 | case 0: |
728 | /* everything ok */ | |
729 | break; | |
730 | default: | |
731 | BUG(); | |
b411b363 PR |
732 | } |
733 | } else { | |
99920dc5 AG |
734 | int err; |
735 | ||
b30ab791 | 736 | inc_rs_pending(device); |
92d94ae6 PR |
737 | err = drbd_send_drequest(peer_device, |
738 | size == discard_granularity ? P_RS_THIN_REQ : P_RS_DATA_REQUEST, | |
99920dc5 AG |
739 | sector, size, ID_SYNCER); |
740 | if (err) { | |
d0180171 | 741 | drbd_err(device, "drbd_send_drequest() failed, aborting...\n"); |
b30ab791 AG |
742 | dec_rs_pending(device); |
743 | put_ldev(device); | |
99920dc5 | 744 | return err; |
b411b363 PR |
745 | } |
746 | } | |
747 | } | |
748 | ||
b30ab791 | 749 | if (device->bm_resync_fo >= drbd_bm_bits(device)) { |
b411b363 PR |
750 | /* last syncer _request_ was sent, |
751 | * but the P_RS_DATA_REPLY not yet received. sync will end (and | |
752 | * next sync group will resume), as soon as we receive the last | |
753 | * resync data block, and the last bit is cleared. | |
754 | * until then resync "work" is "inactive" ... | |
755 | */ | |
b30ab791 | 756 | put_ldev(device); |
99920dc5 | 757 | return 0; |
b411b363 PR |
758 | } |
759 | ||
760 | requeue: | |
b30ab791 AG |
761 | device->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
762 | mod_timer(&device->resync_timer, jiffies + SLEEP_TIME); | |
763 | put_ldev(device); | |
99920dc5 | 764 | return 0; |
b411b363 PR |
765 | } |
766 | ||
d448a2e1 | 767 | static int make_ov_request(struct drbd_device *device, int cancel) |
b411b363 PR |
768 | { |
769 | int number, i, size; | |
770 | sector_t sector; | |
b30ab791 | 771 | const sector_t capacity = drbd_get_capacity(device->this_bdev); |
58ffa580 | 772 | bool stop_sector_reached = false; |
b411b363 PR |
773 | |
774 | if (unlikely(cancel)) | |
775 | return 1; | |
776 | ||
b30ab791 | 777 | number = drbd_rs_number_requests(device); |
b411b363 | 778 | |
b30ab791 | 779 | sector = device->ov_position; |
b411b363 | 780 | for (i = 0; i < number; i++) { |
58ffa580 | 781 | if (sector >= capacity) |
b411b363 | 782 | return 1; |
58ffa580 LE |
783 | |
784 | /* We check for "finished" only in the reply path: | |
785 | * w_e_end_ov_reply(). | |
786 | * We need to send at least one request out. */ | |
787 | stop_sector_reached = i > 0 | |
b30ab791 AG |
788 | && verify_can_do_stop_sector(device) |
789 | && sector >= device->ov_stop_sector; | |
58ffa580 LE |
790 | if (stop_sector_reached) |
791 | break; | |
b411b363 PR |
792 | |
793 | size = BM_BLOCK_SIZE; | |
794 | ||
ad3fee79 | 795 | if (drbd_try_rs_begin_io(device, sector)) { |
b30ab791 | 796 | device->ov_position = sector; |
b411b363 PR |
797 | goto requeue; |
798 | } | |
799 | ||
800 | if (sector + (size>>9) > capacity) | |
801 | size = (capacity-sector)<<9; | |
802 | ||
b30ab791 | 803 | inc_rs_pending(device); |
69a22773 | 804 | if (drbd_send_ov_request(first_peer_device(device), sector, size)) { |
b30ab791 | 805 | dec_rs_pending(device); |
b411b363 PR |
806 | return 0; |
807 | } | |
808 | sector += BM_SECT_PER_BIT; | |
809 | } | |
b30ab791 | 810 | device->ov_position = sector; |
b411b363 PR |
811 | |
812 | requeue: | |
b30ab791 | 813 | device->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
58ffa580 | 814 | if (i == 0 || !stop_sector_reached) |
b30ab791 | 815 | mod_timer(&device->resync_timer, jiffies + SLEEP_TIME); |
b411b363 PR |
816 | return 1; |
817 | } | |
818 | ||
99920dc5 | 819 | int w_ov_finished(struct drbd_work *w, int cancel) |
b411b363 | 820 | { |
84b8c06b AG |
821 | struct drbd_device_work *dw = |
822 | container_of(w, struct drbd_device_work, w); | |
823 | struct drbd_device *device = dw->device; | |
824 | kfree(dw); | |
b30ab791 AG |
825 | ov_out_of_sync_print(device); |
826 | drbd_resync_finished(device); | |
b411b363 | 827 | |
99920dc5 | 828 | return 0; |
b411b363 PR |
829 | } |
830 | ||
99920dc5 | 831 | static int w_resync_finished(struct drbd_work *w, int cancel) |
b411b363 | 832 | { |
84b8c06b AG |
833 | struct drbd_device_work *dw = |
834 | container_of(w, struct drbd_device_work, w); | |
835 | struct drbd_device *device = dw->device; | |
836 | kfree(dw); | |
b411b363 | 837 | |
b30ab791 | 838 | drbd_resync_finished(device); |
b411b363 | 839 | |
99920dc5 | 840 | return 0; |
b411b363 PR |
841 | } |
842 | ||
b30ab791 | 843 | static void ping_peer(struct drbd_device *device) |
af85e8e8 | 844 | { |
a6b32bc3 | 845 | struct drbd_connection *connection = first_peer_device(device)->connection; |
2a67d8b9 | 846 | |
bde89a9e AG |
847 | clear_bit(GOT_PING_ACK, &connection->flags); |
848 | request_ping(connection); | |
849 | wait_event(connection->ping_wait, | |
850 | test_bit(GOT_PING_ACK, &connection->flags) || device->state.conn < C_CONNECTED); | |
af85e8e8 LE |
851 | } |
852 | ||
b30ab791 | 853 | int drbd_resync_finished(struct drbd_device *device) |
b411b363 | 854 | { |
26a96110 | 855 | struct drbd_connection *connection = first_peer_device(device)->connection; |
b411b363 PR |
856 | unsigned long db, dt, dbdt; |
857 | unsigned long n_oos; | |
858 | union drbd_state os, ns; | |
84b8c06b | 859 | struct drbd_device_work *dw; |
b411b363 | 860 | char *khelper_cmd = NULL; |
26525618 | 861 | int verify_done = 0; |
b411b363 PR |
862 | |
863 | /* Remove all elements from the resync LRU. Since future actions | |
864 | * might set bits in the (main) bitmap, then the entries in the | |
865 | * resync LRU would be wrong. */ | |
b30ab791 | 866 | if (drbd_rs_del_all(device)) { |
b411b363 PR |
867 | /* In case this is not possible now, most probably because |
868 | * there are P_RS_DATA_REPLY Packets lingering on the worker's | |
869 | * queue (or even the read operations for those packets | |
870 | * is not finished by now). Retry in 100ms. */ | |
871 | ||
20ee6390 | 872 | schedule_timeout_interruptible(HZ / 10); |
84b8c06b AG |
873 | dw = kmalloc(sizeof(struct drbd_device_work), GFP_ATOMIC); |
874 | if (dw) { | |
875 | dw->w.cb = w_resync_finished; | |
876 | dw->device = device; | |
26a96110 | 877 | drbd_queue_work(&connection->sender_work, &dw->w); |
b411b363 PR |
878 | return 1; |
879 | } | |
84b8c06b | 880 | drbd_err(device, "Warn failed to drbd_rs_del_all() and to kmalloc(dw).\n"); |
b411b363 PR |
881 | } |
882 | ||
b30ab791 | 883 | dt = (jiffies - device->rs_start - device->rs_paused) / HZ; |
b411b363 PR |
884 | if (dt <= 0) |
885 | dt = 1; | |
84b8c06b | 886 | |
b30ab791 | 887 | db = device->rs_total; |
58ffa580 | 888 | /* adjust for verify start and stop sectors, respective reached position */ |
b30ab791 AG |
889 | if (device->state.conn == C_VERIFY_S || device->state.conn == C_VERIFY_T) |
890 | db -= device->ov_left; | |
58ffa580 | 891 | |
b411b363 | 892 | dbdt = Bit2KB(db/dt); |
b30ab791 | 893 | device->rs_paused /= HZ; |
b411b363 | 894 | |
b30ab791 | 895 | if (!get_ldev(device)) |
b411b363 PR |
896 | goto out; |
897 | ||
b30ab791 | 898 | ping_peer(device); |
af85e8e8 | 899 | |
0500813f | 900 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 901 | os = drbd_read_state(device); |
b411b363 | 902 | |
26525618 LE |
903 | verify_done = (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T); |
904 | ||
b411b363 PR |
905 | /* This protects us against multiple calls (that can happen in the presence |
906 | of application IO), and against connectivity loss just before we arrive here. */ | |
907 | if (os.conn <= C_CONNECTED) | |
908 | goto out_unlock; | |
909 | ||
910 | ns = os; | |
911 | ns.conn = C_CONNECTED; | |
912 | ||
d0180171 | 913 | drbd_info(device, "%s done (total %lu sec; paused %lu sec; %lu K/sec)\n", |
58ffa580 | 914 | verify_done ? "Online verify" : "Resync", |
b30ab791 | 915 | dt + device->rs_paused, device->rs_paused, dbdt); |
b411b363 | 916 | |
b30ab791 | 917 | n_oos = drbd_bm_total_weight(device); |
b411b363 PR |
918 | |
919 | if (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) { | |
920 | if (n_oos) { | |
d0180171 | 921 | drbd_alert(device, "Online verify found %lu %dk block out of sync!\n", |
b411b363 PR |
922 | n_oos, Bit2KB(1)); |
923 | khelper_cmd = "out-of-sync"; | |
924 | } | |
925 | } else { | |
0b0ba1ef | 926 | D_ASSERT(device, (n_oos - device->rs_failed) == 0); |
b411b363 PR |
927 | |
928 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) | |
929 | khelper_cmd = "after-resync-target"; | |
930 | ||
aaaba345 | 931 | if (device->use_csums && device->rs_total) { |
b30ab791 AG |
932 | const unsigned long s = device->rs_same_csum; |
933 | const unsigned long t = device->rs_total; | |
b411b363 PR |
934 | const int ratio = |
935 | (t == 0) ? 0 : | |
936 | (t < 100000) ? ((s*100)/t) : (s/(t/100)); | |
d0180171 | 937 | drbd_info(device, "%u %% had equal checksums, eliminated: %luK; " |
b411b363 PR |
938 | "transferred %luK total %luK\n", |
939 | ratio, | |
b30ab791 AG |
940 | Bit2KB(device->rs_same_csum), |
941 | Bit2KB(device->rs_total - device->rs_same_csum), | |
942 | Bit2KB(device->rs_total)); | |
b411b363 PR |
943 | } |
944 | } | |
945 | ||
b30ab791 | 946 | if (device->rs_failed) { |
d0180171 | 947 | drbd_info(device, " %lu failed blocks\n", device->rs_failed); |
b411b363 PR |
948 | |
949 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
950 | ns.disk = D_INCONSISTENT; | |
951 | ns.pdsk = D_UP_TO_DATE; | |
952 | } else { | |
953 | ns.disk = D_UP_TO_DATE; | |
954 | ns.pdsk = D_INCONSISTENT; | |
955 | } | |
956 | } else { | |
957 | ns.disk = D_UP_TO_DATE; | |
958 | ns.pdsk = D_UP_TO_DATE; | |
959 | ||
960 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
b30ab791 | 961 | if (device->p_uuid) { |
b411b363 PR |
962 | int i; |
963 | for (i = UI_BITMAP ; i <= UI_HISTORY_END ; i++) | |
b30ab791 AG |
964 | _drbd_uuid_set(device, i, device->p_uuid[i]); |
965 | drbd_uuid_set(device, UI_BITMAP, device->ldev->md.uuid[UI_CURRENT]); | |
966 | _drbd_uuid_set(device, UI_CURRENT, device->p_uuid[UI_CURRENT]); | |
b411b363 | 967 | } else { |
d0180171 | 968 | drbd_err(device, "device->p_uuid is NULL! BUG\n"); |
b411b363 PR |
969 | } |
970 | } | |
971 | ||
62b0da3a LE |
972 | if (!(os.conn == C_VERIFY_S || os.conn == C_VERIFY_T)) { |
973 | /* for verify runs, we don't update uuids here, | |
974 | * so there would be nothing to report. */ | |
b30ab791 AG |
975 | drbd_uuid_set_bm(device, 0UL); |
976 | drbd_print_uuids(device, "updated UUIDs"); | |
977 | if (device->p_uuid) { | |
62b0da3a LE |
978 | /* Now the two UUID sets are equal, update what we |
979 | * know of the peer. */ | |
980 | int i; | |
981 | for (i = UI_CURRENT ; i <= UI_HISTORY_END ; i++) | |
b30ab791 | 982 | device->p_uuid[i] = device->ldev->md.uuid[i]; |
62b0da3a | 983 | } |
b411b363 PR |
984 | } |
985 | } | |
986 | ||
b30ab791 | 987 | _drbd_set_state(device, ns, CS_VERBOSE, NULL); |
b411b363 | 988 | out_unlock: |
0500813f | 989 | spin_unlock_irq(&device->resource->req_lock); |
26a96110 LE |
990 | |
991 | /* If we have been sync source, and have an effective fencing-policy, | |
992 | * once *all* volumes are back in sync, call "unfence". */ | |
993 | if (os.conn == C_SYNC_SOURCE) { | |
994 | enum drbd_disk_state disk_state = D_MASK; | |
995 | enum drbd_disk_state pdsk_state = D_MASK; | |
996 | enum drbd_fencing_p fp = FP_DONT_CARE; | |
997 | ||
998 | rcu_read_lock(); | |
999 | fp = rcu_dereference(device->ldev->disk_conf)->fencing; | |
1000 | if (fp != FP_DONT_CARE) { | |
1001 | struct drbd_peer_device *peer_device; | |
1002 | int vnr; | |
1003 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
1004 | struct drbd_device *device = peer_device->device; | |
1005 | disk_state = min_t(enum drbd_disk_state, disk_state, device->state.disk); | |
1006 | pdsk_state = min_t(enum drbd_disk_state, pdsk_state, device->state.pdsk); | |
1007 | } | |
1008 | } | |
1009 | rcu_read_unlock(); | |
1010 | if (disk_state == D_UP_TO_DATE && pdsk_state == D_UP_TO_DATE) | |
1011 | conn_khelper(connection, "unfence-peer"); | |
1012 | } | |
1013 | ||
b30ab791 | 1014 | put_ldev(device); |
b411b363 | 1015 | out: |
b30ab791 AG |
1016 | device->rs_total = 0; |
1017 | device->rs_failed = 0; | |
1018 | device->rs_paused = 0; | |
58ffa580 LE |
1019 | |
1020 | /* reset start sector, if we reached end of device */ | |
b30ab791 AG |
1021 | if (verify_done && device->ov_left == 0) |
1022 | device->ov_start_sector = 0; | |
b411b363 | 1023 | |
b30ab791 | 1024 | drbd_md_sync(device); |
13d42685 | 1025 | |
b411b363 | 1026 | if (khelper_cmd) |
b30ab791 | 1027 | drbd_khelper(device, khelper_cmd); |
b411b363 PR |
1028 | |
1029 | return 1; | |
1030 | } | |
1031 | ||
1032 | /* helper */ | |
b30ab791 | 1033 | static void move_to_net_ee_or_free(struct drbd_device *device, struct drbd_peer_request *peer_req) |
b411b363 | 1034 | { |
045417f7 | 1035 | if (drbd_peer_req_has_active_page(peer_req)) { |
b411b363 | 1036 | /* This might happen if sendpage() has not finished */ |
db830c46 | 1037 | int i = (peer_req->i.size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b30ab791 AG |
1038 | atomic_add(i, &device->pp_in_use_by_net); |
1039 | atomic_sub(i, &device->pp_in_use); | |
0500813f | 1040 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 1041 | list_add_tail(&peer_req->w.list, &device->net_ee); |
0500813f | 1042 | spin_unlock_irq(&device->resource->req_lock); |
435f0740 | 1043 | wake_up(&drbd_pp_wait); |
b411b363 | 1044 | } else |
b30ab791 | 1045 | drbd_free_peer_req(device, peer_req); |
b411b363 PR |
1046 | } |
1047 | ||
1048 | /** | |
1049 | * w_e_end_data_req() - Worker callback, to send a P_DATA_REPLY packet in response to a P_DATA_REQUEST | |
b411b363 PR |
1050 | * @w: work object. |
1051 | * @cancel: The connection will be closed anyways | |
1052 | */ | |
99920dc5 | 1053 | int w_e_end_data_req(struct drbd_work *w, int cancel) |
b411b363 | 1054 | { |
a8cd15ba | 1055 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1056 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1057 | struct drbd_device *device = peer_device->device; | |
99920dc5 | 1058 | int err; |
b411b363 PR |
1059 | |
1060 | if (unlikely(cancel)) { | |
b30ab791 AG |
1061 | drbd_free_peer_req(device, peer_req); |
1062 | dec_unacked(device); | |
99920dc5 | 1063 | return 0; |
b411b363 PR |
1064 | } |
1065 | ||
db830c46 | 1066 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
6780139c | 1067 | err = drbd_send_block(peer_device, P_DATA_REPLY, peer_req); |
b411b363 PR |
1068 | } else { |
1069 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1070 | drbd_err(device, "Sending NegDReply. sector=%llus.\n", |
db830c46 | 1071 | (unsigned long long)peer_req->i.sector); |
b411b363 | 1072 | |
6780139c | 1073 | err = drbd_send_ack(peer_device, P_NEG_DREPLY, peer_req); |
b411b363 PR |
1074 | } |
1075 | ||
b30ab791 | 1076 | dec_unacked(device); |
b411b363 | 1077 | |
b30ab791 | 1078 | move_to_net_ee_or_free(device, peer_req); |
b411b363 | 1079 | |
99920dc5 | 1080 | if (unlikely(err)) |
d0180171 | 1081 | drbd_err(device, "drbd_send_block() failed\n"); |
99920dc5 | 1082 | return err; |
b411b363 PR |
1083 | } |
1084 | ||
700ca8c0 PR |
1085 | static bool all_zero(struct drbd_peer_request *peer_req) |
1086 | { | |
1087 | struct page *page = peer_req->pages; | |
1088 | unsigned int len = peer_req->i.size; | |
1089 | ||
1090 | page_chain_for_each(page) { | |
1091 | unsigned int l = min_t(unsigned int, len, PAGE_SIZE); | |
1092 | unsigned int i, words = l / sizeof(long); | |
1093 | unsigned long *d; | |
1094 | ||
1095 | d = kmap_atomic(page); | |
1096 | for (i = 0; i < words; i++) { | |
1097 | if (d[i]) { | |
1098 | kunmap_atomic(d); | |
1099 | return false; | |
1100 | } | |
1101 | } | |
1102 | kunmap_atomic(d); | |
1103 | len -= l; | |
1104 | } | |
1105 | ||
1106 | return true; | |
1107 | } | |
1108 | ||
b411b363 | 1109 | /** |
a209b4ae | 1110 | * w_e_end_rsdata_req() - Worker callback to send a P_RS_DATA_REPLY packet in response to a P_RS_DATA_REQUEST |
b411b363 PR |
1111 | * @w: work object. |
1112 | * @cancel: The connection will be closed anyways | |
1113 | */ | |
99920dc5 | 1114 | int w_e_end_rsdata_req(struct drbd_work *w, int cancel) |
b411b363 | 1115 | { |
a8cd15ba | 1116 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1117 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1118 | struct drbd_device *device = peer_device->device; | |
99920dc5 | 1119 | int err; |
b411b363 PR |
1120 | |
1121 | if (unlikely(cancel)) { | |
b30ab791 AG |
1122 | drbd_free_peer_req(device, peer_req); |
1123 | dec_unacked(device); | |
99920dc5 | 1124 | return 0; |
b411b363 PR |
1125 | } |
1126 | ||
b30ab791 AG |
1127 | if (get_ldev_if_state(device, D_FAILED)) { |
1128 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1129 | put_ldev(device); | |
b411b363 PR |
1130 | } |
1131 | ||
b30ab791 | 1132 | if (device->state.conn == C_AHEAD) { |
6780139c | 1133 | err = drbd_send_ack(peer_device, P_RS_CANCEL, peer_req); |
db830c46 | 1134 | } else if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b30ab791 AG |
1135 | if (likely(device->state.pdsk >= D_INCONSISTENT)) { |
1136 | inc_rs_pending(device); | |
700ca8c0 PR |
1137 | if (peer_req->flags & EE_RS_THIN_REQ && all_zero(peer_req)) |
1138 | err = drbd_send_rs_deallocated(peer_device, peer_req); | |
1139 | else | |
1140 | err = drbd_send_block(peer_device, P_RS_DATA_REPLY, peer_req); | |
b411b363 PR |
1141 | } else { |
1142 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1143 | drbd_err(device, "Not sending RSDataReply, " |
b411b363 | 1144 | "partner DISKLESS!\n"); |
99920dc5 | 1145 | err = 0; |
b411b363 PR |
1146 | } |
1147 | } else { | |
1148 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1149 | drbd_err(device, "Sending NegRSDReply. sector %llus.\n", |
db830c46 | 1150 | (unsigned long long)peer_req->i.sector); |
b411b363 | 1151 | |
6780139c | 1152 | err = drbd_send_ack(peer_device, P_NEG_RS_DREPLY, peer_req); |
b411b363 PR |
1153 | |
1154 | /* update resync data with failure */ | |
b30ab791 | 1155 | drbd_rs_failed_io(device, peer_req->i.sector, peer_req->i.size); |
b411b363 PR |
1156 | } |
1157 | ||
b30ab791 | 1158 | dec_unacked(device); |
b411b363 | 1159 | |
b30ab791 | 1160 | move_to_net_ee_or_free(device, peer_req); |
b411b363 | 1161 | |
99920dc5 | 1162 | if (unlikely(err)) |
d0180171 | 1163 | drbd_err(device, "drbd_send_block() failed\n"); |
99920dc5 | 1164 | return err; |
b411b363 PR |
1165 | } |
1166 | ||
99920dc5 | 1167 | int w_e_end_csum_rs_req(struct drbd_work *w, int cancel) |
b411b363 | 1168 | { |
a8cd15ba | 1169 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1170 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1171 | struct drbd_device *device = peer_device->device; | |
b411b363 PR |
1172 | struct digest_info *di; |
1173 | int digest_size; | |
1174 | void *digest = NULL; | |
99920dc5 | 1175 | int err, eq = 0; |
b411b363 PR |
1176 | |
1177 | if (unlikely(cancel)) { | |
b30ab791 AG |
1178 | drbd_free_peer_req(device, peer_req); |
1179 | dec_unacked(device); | |
99920dc5 | 1180 | return 0; |
b411b363 PR |
1181 | } |
1182 | ||
b30ab791 AG |
1183 | if (get_ldev(device)) { |
1184 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1185 | put_ldev(device); | |
1d53f09e | 1186 | } |
b411b363 | 1187 | |
db830c46 | 1188 | di = peer_req->digest; |
b411b363 | 1189 | |
db830c46 | 1190 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1191 | /* quick hack to try to avoid a race against reconfiguration. |
1192 | * a real fix would be much more involved, | |
1193 | * introducing more locking mechanisms */ | |
6780139c | 1194 | if (peer_device->connection->csums_tfm) { |
9534d671 | 1195 | digest_size = crypto_ahash_digestsize(peer_device->connection->csums_tfm); |
0b0ba1ef | 1196 | D_ASSERT(device, digest_size == di->digest_size); |
b411b363 PR |
1197 | digest = kmalloc(digest_size, GFP_NOIO); |
1198 | } | |
1199 | if (digest) { | |
6780139c | 1200 | drbd_csum_ee(peer_device->connection->csums_tfm, peer_req, digest); |
b411b363 PR |
1201 | eq = !memcmp(digest, di->digest, digest_size); |
1202 | kfree(digest); | |
1203 | } | |
1204 | ||
1205 | if (eq) { | |
b30ab791 | 1206 | drbd_set_in_sync(device, peer_req->i.sector, peer_req->i.size); |
676396d5 | 1207 | /* rs_same_csums unit is BM_BLOCK_SIZE */ |
b30ab791 | 1208 | device->rs_same_csum += peer_req->i.size >> BM_BLOCK_SHIFT; |
6780139c | 1209 | err = drbd_send_ack(peer_device, P_RS_IS_IN_SYNC, peer_req); |
b411b363 | 1210 | } else { |
b30ab791 | 1211 | inc_rs_pending(device); |
db830c46 AG |
1212 | peer_req->block_id = ID_SYNCER; /* By setting block_id, digest pointer becomes invalid! */ |
1213 | peer_req->flags &= ~EE_HAS_DIGEST; /* This peer request no longer has a digest pointer */ | |
204bba99 | 1214 | kfree(di); |
6780139c | 1215 | err = drbd_send_block(peer_device, P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1216 | } |
1217 | } else { | |
6780139c | 1218 | err = drbd_send_ack(peer_device, P_NEG_RS_DREPLY, peer_req); |
b411b363 | 1219 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 1220 | drbd_err(device, "Sending NegDReply. I guess it gets messy.\n"); |
b411b363 PR |
1221 | } |
1222 | ||
b30ab791 AG |
1223 | dec_unacked(device); |
1224 | move_to_net_ee_or_free(device, peer_req); | |
b411b363 | 1225 | |
99920dc5 | 1226 | if (unlikely(err)) |
d0180171 | 1227 | drbd_err(device, "drbd_send_block/ack() failed\n"); |
99920dc5 | 1228 | return err; |
b411b363 PR |
1229 | } |
1230 | ||
99920dc5 | 1231 | int w_e_end_ov_req(struct drbd_work *w, int cancel) |
b411b363 | 1232 | { |
a8cd15ba | 1233 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1234 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1235 | struct drbd_device *device = peer_device->device; | |
db830c46 AG |
1236 | sector_t sector = peer_req->i.sector; |
1237 | unsigned int size = peer_req->i.size; | |
b411b363 PR |
1238 | int digest_size; |
1239 | void *digest; | |
99920dc5 | 1240 | int err = 0; |
b411b363 PR |
1241 | |
1242 | if (unlikely(cancel)) | |
1243 | goto out; | |
1244 | ||
9534d671 | 1245 | digest_size = crypto_ahash_digestsize(peer_device->connection->verify_tfm); |
b411b363 | 1246 | digest = kmalloc(digest_size, GFP_NOIO); |
8f21420e | 1247 | if (!digest) { |
99920dc5 | 1248 | err = 1; /* terminate the connection in case the allocation failed */ |
8f21420e | 1249 | goto out; |
b411b363 PR |
1250 | } |
1251 | ||
db830c46 | 1252 | if (likely(!(peer_req->flags & EE_WAS_ERROR))) |
6780139c | 1253 | drbd_csum_ee(peer_device->connection->verify_tfm, peer_req, digest); |
8f21420e PR |
1254 | else |
1255 | memset(digest, 0, digest_size); | |
1256 | ||
53ea4331 LE |
1257 | /* Free e and pages before send. |
1258 | * In case we block on congestion, we could otherwise run into | |
1259 | * some distributed deadlock, if the other side blocks on | |
1260 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1261 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 1262 | drbd_free_peer_req(device, peer_req); |
db830c46 | 1263 | peer_req = NULL; |
b30ab791 | 1264 | inc_rs_pending(device); |
6780139c | 1265 | err = drbd_send_drequest_csum(peer_device, sector, size, digest, digest_size, P_OV_REPLY); |
99920dc5 | 1266 | if (err) |
b30ab791 | 1267 | dec_rs_pending(device); |
8f21420e PR |
1268 | kfree(digest); |
1269 | ||
b411b363 | 1270 | out: |
db830c46 | 1271 | if (peer_req) |
b30ab791 AG |
1272 | drbd_free_peer_req(device, peer_req); |
1273 | dec_unacked(device); | |
99920dc5 | 1274 | return err; |
b411b363 PR |
1275 | } |
1276 | ||
b30ab791 | 1277 | void drbd_ov_out_of_sync_found(struct drbd_device *device, sector_t sector, int size) |
b411b363 | 1278 | { |
b30ab791 AG |
1279 | if (device->ov_last_oos_start + device->ov_last_oos_size == sector) { |
1280 | device->ov_last_oos_size += size>>9; | |
b411b363 | 1281 | } else { |
b30ab791 AG |
1282 | device->ov_last_oos_start = sector; |
1283 | device->ov_last_oos_size = size>>9; | |
b411b363 | 1284 | } |
b30ab791 | 1285 | drbd_set_out_of_sync(device, sector, size); |
b411b363 PR |
1286 | } |
1287 | ||
99920dc5 | 1288 | int w_e_end_ov_reply(struct drbd_work *w, int cancel) |
b411b363 | 1289 | { |
a8cd15ba | 1290 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1291 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1292 | struct drbd_device *device = peer_device->device; | |
b411b363 | 1293 | struct digest_info *di; |
b411b363 | 1294 | void *digest; |
db830c46 AG |
1295 | sector_t sector = peer_req->i.sector; |
1296 | unsigned int size = peer_req->i.size; | |
53ea4331 | 1297 | int digest_size; |
99920dc5 | 1298 | int err, eq = 0; |
58ffa580 | 1299 | bool stop_sector_reached = false; |
b411b363 PR |
1300 | |
1301 | if (unlikely(cancel)) { | |
b30ab791 AG |
1302 | drbd_free_peer_req(device, peer_req); |
1303 | dec_unacked(device); | |
99920dc5 | 1304 | return 0; |
b411b363 PR |
1305 | } |
1306 | ||
1307 | /* after "cancel", because after drbd_disconnect/drbd_rs_cancel_all | |
1308 | * the resync lru has been cleaned up already */ | |
b30ab791 AG |
1309 | if (get_ldev(device)) { |
1310 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1311 | put_ldev(device); | |
1d53f09e | 1312 | } |
b411b363 | 1313 | |
db830c46 | 1314 | di = peer_req->digest; |
b411b363 | 1315 | |
db830c46 | 1316 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
9534d671 | 1317 | digest_size = crypto_ahash_digestsize(peer_device->connection->verify_tfm); |
b411b363 PR |
1318 | digest = kmalloc(digest_size, GFP_NOIO); |
1319 | if (digest) { | |
6780139c | 1320 | drbd_csum_ee(peer_device->connection->verify_tfm, peer_req, digest); |
b411b363 | 1321 | |
0b0ba1ef | 1322 | D_ASSERT(device, digest_size == di->digest_size); |
b411b363 PR |
1323 | eq = !memcmp(digest, di->digest, digest_size); |
1324 | kfree(digest); | |
1325 | } | |
b411b363 PR |
1326 | } |
1327 | ||
9676c760 LE |
1328 | /* Free peer_req and pages before send. |
1329 | * In case we block on congestion, we could otherwise run into | |
1330 | * some distributed deadlock, if the other side blocks on | |
1331 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1332 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 1333 | drbd_free_peer_req(device, peer_req); |
b411b363 | 1334 | if (!eq) |
b30ab791 | 1335 | drbd_ov_out_of_sync_found(device, sector, size); |
b411b363 | 1336 | else |
b30ab791 | 1337 | ov_out_of_sync_print(device); |
b411b363 | 1338 | |
6780139c | 1339 | err = drbd_send_ack_ex(peer_device, P_OV_RESULT, sector, size, |
fa79abd8 | 1340 | eq ? ID_IN_SYNC : ID_OUT_OF_SYNC); |
b411b363 | 1341 | |
b30ab791 | 1342 | dec_unacked(device); |
b411b363 | 1343 | |
b30ab791 | 1344 | --device->ov_left; |
ea5442af LE |
1345 | |
1346 | /* let's advance progress step marks only for every other megabyte */ | |
b30ab791 AG |
1347 | if ((device->ov_left & 0x200) == 0x200) |
1348 | drbd_advance_rs_marks(device, device->ov_left); | |
ea5442af | 1349 | |
b30ab791 AG |
1350 | stop_sector_reached = verify_can_do_stop_sector(device) && |
1351 | (sector + (size>>9)) >= device->ov_stop_sector; | |
58ffa580 | 1352 | |
b30ab791 AG |
1353 | if (device->ov_left == 0 || stop_sector_reached) { |
1354 | ov_out_of_sync_print(device); | |
1355 | drbd_resync_finished(device); | |
b411b363 PR |
1356 | } |
1357 | ||
99920dc5 | 1358 | return err; |
b411b363 PR |
1359 | } |
1360 | ||
b6dd1a89 LE |
1361 | /* FIXME |
1362 | * We need to track the number of pending barrier acks, | |
1363 | * and to be able to wait for them. | |
1364 | * See also comment in drbd_adm_attach before drbd_suspend_io. | |
1365 | */ | |
bde89a9e | 1366 | static int drbd_send_barrier(struct drbd_connection *connection) |
b411b363 | 1367 | { |
9f5bdc33 | 1368 | struct p_barrier *p; |
b6dd1a89 | 1369 | struct drbd_socket *sock; |
b411b363 | 1370 | |
bde89a9e AG |
1371 | sock = &connection->data; |
1372 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1373 | if (!p) |
1374 | return -EIO; | |
bde89a9e | 1375 | p->barrier = connection->send.current_epoch_nr; |
b6dd1a89 | 1376 | p->pad = 0; |
bde89a9e | 1377 | connection->send.current_epoch_writes = 0; |
84d34f2f | 1378 | connection->send.last_sent_barrier_jif = jiffies; |
b6dd1a89 | 1379 | |
bde89a9e | 1380 | return conn_send_command(connection, sock, P_BARRIER, sizeof(*p), NULL, 0); |
b411b363 PR |
1381 | } |
1382 | ||
99920dc5 | 1383 | int w_send_write_hint(struct drbd_work *w, int cancel) |
b411b363 | 1384 | { |
84b8c06b AG |
1385 | struct drbd_device *device = |
1386 | container_of(w, struct drbd_device, unplug_work); | |
9f5bdc33 AG |
1387 | struct drbd_socket *sock; |
1388 | ||
b411b363 | 1389 | if (cancel) |
99920dc5 | 1390 | return 0; |
a6b32bc3 | 1391 | sock = &first_peer_device(device)->connection->data; |
69a22773 | 1392 | if (!drbd_prepare_command(first_peer_device(device), sock)) |
9f5bdc33 | 1393 | return -EIO; |
69a22773 | 1394 | return drbd_send_command(first_peer_device(device), sock, P_UNPLUG_REMOTE, 0, NULL, 0); |
b411b363 PR |
1395 | } |
1396 | ||
bde89a9e | 1397 | static void re_init_if_first_write(struct drbd_connection *connection, unsigned int epoch) |
4eb9b3cb | 1398 | { |
bde89a9e AG |
1399 | if (!connection->send.seen_any_write_yet) { |
1400 | connection->send.seen_any_write_yet = true; | |
1401 | connection->send.current_epoch_nr = epoch; | |
1402 | connection->send.current_epoch_writes = 0; | |
84d34f2f | 1403 | connection->send.last_sent_barrier_jif = jiffies; |
4eb9b3cb LE |
1404 | } |
1405 | } | |
1406 | ||
bde89a9e | 1407 | static void maybe_send_barrier(struct drbd_connection *connection, unsigned int epoch) |
4eb9b3cb LE |
1408 | { |
1409 | /* re-init if first write on this connection */ | |
bde89a9e | 1410 | if (!connection->send.seen_any_write_yet) |
4eb9b3cb | 1411 | return; |
bde89a9e AG |
1412 | if (connection->send.current_epoch_nr != epoch) { |
1413 | if (connection->send.current_epoch_writes) | |
1414 | drbd_send_barrier(connection); | |
1415 | connection->send.current_epoch_nr = epoch; | |
4eb9b3cb LE |
1416 | } |
1417 | } | |
1418 | ||
8f7bed77 | 1419 | int w_send_out_of_sync(struct drbd_work *w, int cancel) |
73a01a18 PR |
1420 | { |
1421 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1422 | struct drbd_device *device = req->device; |
44a4d551 LE |
1423 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
1424 | struct drbd_connection *const connection = peer_device->connection; | |
99920dc5 | 1425 | int err; |
73a01a18 PR |
1426 | |
1427 | if (unlikely(cancel)) { | |
8554df1c | 1428 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1429 | return 0; |
73a01a18 | 1430 | } |
e5f891b2 | 1431 | req->pre_send_jif = jiffies; |
73a01a18 | 1432 | |
bde89a9e | 1433 | /* this time, no connection->send.current_epoch_writes++; |
b6dd1a89 LE |
1434 | * If it was sent, it was the closing barrier for the last |
1435 | * replicated epoch, before we went into AHEAD mode. | |
1436 | * No more barriers will be sent, until we leave AHEAD mode again. */ | |
bde89a9e | 1437 | maybe_send_barrier(connection, req->epoch); |
b6dd1a89 | 1438 | |
44a4d551 | 1439 | err = drbd_send_out_of_sync(peer_device, req); |
8554df1c | 1440 | req_mod(req, OOS_HANDED_TO_NETWORK); |
73a01a18 | 1441 | |
99920dc5 | 1442 | return err; |
73a01a18 PR |
1443 | } |
1444 | ||
b411b363 PR |
1445 | /** |
1446 | * w_send_dblock() - Worker callback to send a P_DATA packet in order to mirror a write request | |
b411b363 PR |
1447 | * @w: work object. |
1448 | * @cancel: The connection will be closed anyways | |
1449 | */ | |
99920dc5 | 1450 | int w_send_dblock(struct drbd_work *w, int cancel) |
b411b363 PR |
1451 | { |
1452 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1453 | struct drbd_device *device = req->device; |
44a4d551 LE |
1454 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
1455 | struct drbd_connection *connection = peer_device->connection; | |
99920dc5 | 1456 | int err; |
b411b363 PR |
1457 | |
1458 | if (unlikely(cancel)) { | |
8554df1c | 1459 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1460 | return 0; |
b411b363 | 1461 | } |
e5f891b2 | 1462 | req->pre_send_jif = jiffies; |
b411b363 | 1463 | |
bde89a9e AG |
1464 | re_init_if_first_write(connection, req->epoch); |
1465 | maybe_send_barrier(connection, req->epoch); | |
1466 | connection->send.current_epoch_writes++; | |
b6dd1a89 | 1467 | |
44a4d551 | 1468 | err = drbd_send_dblock(peer_device, req); |
99920dc5 | 1469 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1470 | |
99920dc5 | 1471 | return err; |
b411b363 PR |
1472 | } |
1473 | ||
1474 | /** | |
1475 | * w_send_read_req() - Worker callback to send a read request (P_DATA_REQUEST) packet | |
b411b363 PR |
1476 | * @w: work object. |
1477 | * @cancel: The connection will be closed anyways | |
1478 | */ | |
99920dc5 | 1479 | int w_send_read_req(struct drbd_work *w, int cancel) |
b411b363 PR |
1480 | { |
1481 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1482 | struct drbd_device *device = req->device; |
44a4d551 LE |
1483 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
1484 | struct drbd_connection *connection = peer_device->connection; | |
99920dc5 | 1485 | int err; |
b411b363 PR |
1486 | |
1487 | if (unlikely(cancel)) { | |
8554df1c | 1488 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1489 | return 0; |
b411b363 | 1490 | } |
e5f891b2 | 1491 | req->pre_send_jif = jiffies; |
b411b363 | 1492 | |
b6dd1a89 LE |
1493 | /* Even read requests may close a write epoch, |
1494 | * if there was any yet. */ | |
bde89a9e | 1495 | maybe_send_barrier(connection, req->epoch); |
b6dd1a89 | 1496 | |
44a4d551 | 1497 | err = drbd_send_drequest(peer_device, P_DATA_REQUEST, req->i.sector, req->i.size, |
6c1005e7 | 1498 | (unsigned long)req); |
b411b363 | 1499 | |
99920dc5 | 1500 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1501 | |
99920dc5 | 1502 | return err; |
b411b363 PR |
1503 | } |
1504 | ||
99920dc5 | 1505 | int w_restart_disk_io(struct drbd_work *w, int cancel) |
265be2d0 PR |
1506 | { |
1507 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1508 | struct drbd_device *device = req->device; |
265be2d0 | 1509 | |
0778286a | 1510 | if (bio_data_dir(req->master_bio) == WRITE && req->rq_state & RQ_IN_ACT_LOG) |
4dd726f0 | 1511 | drbd_al_begin_io(device, &req->i); |
265be2d0 PR |
1512 | |
1513 | drbd_req_make_private_bio(req, req->master_bio); | |
b30ab791 | 1514 | req->private_bio->bi_bdev = device->ldev->backing_bdev; |
265be2d0 PR |
1515 | generic_make_request(req->private_bio); |
1516 | ||
99920dc5 | 1517 | return 0; |
265be2d0 PR |
1518 | } |
1519 | ||
b30ab791 | 1520 | static int _drbd_may_sync_now(struct drbd_device *device) |
b411b363 | 1521 | { |
b30ab791 | 1522 | struct drbd_device *odev = device; |
95f8efd0 | 1523 | int resync_after; |
b411b363 PR |
1524 | |
1525 | while (1) { | |
a3f8f7dc | 1526 | if (!odev->ldev || odev->state.disk == D_DISKLESS) |
438c8374 | 1527 | return 1; |
daeda1cc | 1528 | rcu_read_lock(); |
95f8efd0 | 1529 | resync_after = rcu_dereference(odev->ldev->disk_conf)->resync_after; |
daeda1cc | 1530 | rcu_read_unlock(); |
95f8efd0 | 1531 | if (resync_after == -1) |
b411b363 | 1532 | return 1; |
b30ab791 | 1533 | odev = minor_to_device(resync_after); |
a3f8f7dc | 1534 | if (!odev) |
841ce241 | 1535 | return 1; |
b411b363 PR |
1536 | if ((odev->state.conn >= C_SYNC_SOURCE && |
1537 | odev->state.conn <= C_PAUSED_SYNC_T) || | |
1538 | odev->state.aftr_isp || odev->state.peer_isp || | |
1539 | odev->state.user_isp) | |
1540 | return 0; | |
1541 | } | |
1542 | } | |
1543 | ||
1544 | /** | |
28bc3b8c | 1545 | * drbd_pause_after() - Pause resync on all devices that may not resync now |
b30ab791 | 1546 | * @device: DRBD device. |
b411b363 PR |
1547 | * |
1548 | * Called from process context only (admin command and after_state_ch). | |
1549 | */ | |
28bc3b8c | 1550 | static bool drbd_pause_after(struct drbd_device *device) |
b411b363 | 1551 | { |
28bc3b8c | 1552 | bool changed = false; |
54761697 | 1553 | struct drbd_device *odev; |
28bc3b8c | 1554 | int i; |
b411b363 | 1555 | |
695d08fa | 1556 | rcu_read_lock(); |
05a10ec7 | 1557 | idr_for_each_entry(&drbd_devices, odev, i) { |
b411b363 PR |
1558 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1559 | continue; | |
28bc3b8c AG |
1560 | if (!_drbd_may_sync_now(odev) && |
1561 | _drbd_set_state(_NS(odev, aftr_isp, 1), | |
1562 | CS_HARD, NULL) != SS_NOTHING_TO_DO) | |
1563 | changed = true; | |
b411b363 | 1564 | } |
695d08fa | 1565 | rcu_read_unlock(); |
b411b363 | 1566 | |
28bc3b8c | 1567 | return changed; |
b411b363 PR |
1568 | } |
1569 | ||
1570 | /** | |
28bc3b8c | 1571 | * drbd_resume_next() - Resume resync on all devices that may resync now |
b30ab791 | 1572 | * @device: DRBD device. |
b411b363 PR |
1573 | * |
1574 | * Called from process context only (admin command and worker). | |
1575 | */ | |
28bc3b8c | 1576 | static bool drbd_resume_next(struct drbd_device *device) |
b411b363 | 1577 | { |
28bc3b8c | 1578 | bool changed = false; |
54761697 | 1579 | struct drbd_device *odev; |
28bc3b8c | 1580 | int i; |
b411b363 | 1581 | |
695d08fa | 1582 | rcu_read_lock(); |
05a10ec7 | 1583 | idr_for_each_entry(&drbd_devices, odev, i) { |
b411b363 PR |
1584 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1585 | continue; | |
1586 | if (odev->state.aftr_isp) { | |
28bc3b8c AG |
1587 | if (_drbd_may_sync_now(odev) && |
1588 | _drbd_set_state(_NS(odev, aftr_isp, 0), | |
1589 | CS_HARD, NULL) != SS_NOTHING_TO_DO) | |
1590 | changed = true; | |
b411b363 PR |
1591 | } |
1592 | } | |
695d08fa | 1593 | rcu_read_unlock(); |
28bc3b8c | 1594 | return changed; |
b411b363 PR |
1595 | } |
1596 | ||
b30ab791 | 1597 | void resume_next_sg(struct drbd_device *device) |
b411b363 | 1598 | { |
28bc3b8c AG |
1599 | lock_all_resources(); |
1600 | drbd_resume_next(device); | |
1601 | unlock_all_resources(); | |
b411b363 PR |
1602 | } |
1603 | ||
b30ab791 | 1604 | void suspend_other_sg(struct drbd_device *device) |
b411b363 | 1605 | { |
28bc3b8c AG |
1606 | lock_all_resources(); |
1607 | drbd_pause_after(device); | |
1608 | unlock_all_resources(); | |
b411b363 PR |
1609 | } |
1610 | ||
28bc3b8c | 1611 | /* caller must lock_all_resources() */ |
b30ab791 | 1612 | enum drbd_ret_code drbd_resync_after_valid(struct drbd_device *device, int o_minor) |
b411b363 | 1613 | { |
54761697 | 1614 | struct drbd_device *odev; |
95f8efd0 | 1615 | int resync_after; |
b411b363 PR |
1616 | |
1617 | if (o_minor == -1) | |
1618 | return NO_ERROR; | |
a3f8f7dc | 1619 | if (o_minor < -1 || o_minor > MINORMASK) |
95f8efd0 | 1620 | return ERR_RESYNC_AFTER; |
b411b363 PR |
1621 | |
1622 | /* check for loops */ | |
b30ab791 | 1623 | odev = minor_to_device(o_minor); |
b411b363 | 1624 | while (1) { |
b30ab791 | 1625 | if (odev == device) |
95f8efd0 | 1626 | return ERR_RESYNC_AFTER_CYCLE; |
b411b363 | 1627 | |
a3f8f7dc LE |
1628 | /* You are free to depend on diskless, non-existing, |
1629 | * or not yet/no longer existing minors. | |
1630 | * We only reject dependency loops. | |
1631 | * We cannot follow the dependency chain beyond a detached or | |
1632 | * missing minor. | |
1633 | */ | |
1634 | if (!odev || !odev->ldev || odev->state.disk == D_DISKLESS) | |
1635 | return NO_ERROR; | |
1636 | ||
daeda1cc | 1637 | rcu_read_lock(); |
95f8efd0 | 1638 | resync_after = rcu_dereference(odev->ldev->disk_conf)->resync_after; |
daeda1cc | 1639 | rcu_read_unlock(); |
b411b363 | 1640 | /* dependency chain ends here, no cycles. */ |
95f8efd0 | 1641 | if (resync_after == -1) |
b411b363 PR |
1642 | return NO_ERROR; |
1643 | ||
1644 | /* follow the dependency chain */ | |
b30ab791 | 1645 | odev = minor_to_device(resync_after); |
b411b363 PR |
1646 | } |
1647 | } | |
1648 | ||
28bc3b8c | 1649 | /* caller must lock_all_resources() */ |
b30ab791 | 1650 | void drbd_resync_after_changed(struct drbd_device *device) |
b411b363 | 1651 | { |
28bc3b8c | 1652 | int changed; |
b411b363 | 1653 | |
dc97b708 | 1654 | do { |
28bc3b8c AG |
1655 | changed = drbd_pause_after(device); |
1656 | changed |= drbd_resume_next(device); | |
1657 | } while (changed); | |
b411b363 PR |
1658 | } |
1659 | ||
b30ab791 | 1660 | void drbd_rs_controller_reset(struct drbd_device *device) |
9bd28d3c | 1661 | { |
ff8bd88b | 1662 | struct gendisk *disk = device->ldev->backing_bdev->bd_contains->bd_disk; |
813472ce PR |
1663 | struct fifo_buffer *plan; |
1664 | ||
b30ab791 AG |
1665 | atomic_set(&device->rs_sect_in, 0); |
1666 | atomic_set(&device->rs_sect_ev, 0); | |
1667 | device->rs_in_flight = 0; | |
ff8bd88b LE |
1668 | device->rs_last_events = |
1669 | (int)part_stat_read(&disk->part0, sectors[0]) + | |
1670 | (int)part_stat_read(&disk->part0, sectors[1]); | |
813472ce PR |
1671 | |
1672 | /* Updating the RCU protected object in place is necessary since | |
1673 | this function gets called from atomic context. | |
1674 | It is valid since all other updates also lead to an completely | |
1675 | empty fifo */ | |
1676 | rcu_read_lock(); | |
b30ab791 | 1677 | plan = rcu_dereference(device->rs_plan_s); |
813472ce PR |
1678 | plan->total = 0; |
1679 | fifo_set(plan, 0); | |
1680 | rcu_read_unlock(); | |
9bd28d3c LE |
1681 | } |
1682 | ||
1f04af33 PR |
1683 | void start_resync_timer_fn(unsigned long data) |
1684 | { | |
b30ab791 | 1685 | struct drbd_device *device = (struct drbd_device *) data; |
ac0acb9e | 1686 | drbd_device_post_work(device, RS_START); |
1f04af33 PR |
1687 | } |
1688 | ||
ac0acb9e | 1689 | static void do_start_resync(struct drbd_device *device) |
1f04af33 | 1690 | { |
b30ab791 | 1691 | if (atomic_read(&device->unacked_cnt) || atomic_read(&device->rs_pending_cnt)) { |
ac0acb9e | 1692 | drbd_warn(device, "postponing start_resync ...\n"); |
b30ab791 AG |
1693 | device->start_resync_timer.expires = jiffies + HZ/10; |
1694 | add_timer(&device->start_resync_timer); | |
ac0acb9e | 1695 | return; |
1f04af33 PR |
1696 | } |
1697 | ||
b30ab791 AG |
1698 | drbd_start_resync(device, C_SYNC_SOURCE); |
1699 | clear_bit(AHEAD_TO_SYNC_SOURCE, &device->flags); | |
1f04af33 PR |
1700 | } |
1701 | ||
aaaba345 LE |
1702 | static bool use_checksum_based_resync(struct drbd_connection *connection, struct drbd_device *device) |
1703 | { | |
1704 | bool csums_after_crash_only; | |
1705 | rcu_read_lock(); | |
1706 | csums_after_crash_only = rcu_dereference(connection->net_conf)->csums_after_crash_only; | |
1707 | rcu_read_unlock(); | |
1708 | return connection->agreed_pro_version >= 89 && /* supported? */ | |
1709 | connection->csums_tfm && /* configured? */ | |
7e5fec31 | 1710 | (csums_after_crash_only == false /* use for each resync? */ |
aaaba345 LE |
1711 | || test_bit(CRASHED_PRIMARY, &device->flags)); /* or only after Primary crash? */ |
1712 | } | |
1713 | ||
b411b363 PR |
1714 | /** |
1715 | * drbd_start_resync() - Start the resync process | |
b30ab791 | 1716 | * @device: DRBD device. |
b411b363 PR |
1717 | * @side: Either C_SYNC_SOURCE or C_SYNC_TARGET |
1718 | * | |
1719 | * This function might bring you directly into one of the | |
1720 | * C_PAUSED_SYNC_* states. | |
1721 | */ | |
b30ab791 | 1722 | void drbd_start_resync(struct drbd_device *device, enum drbd_conns side) |
b411b363 | 1723 | { |
44a4d551 LE |
1724 | struct drbd_peer_device *peer_device = first_peer_device(device); |
1725 | struct drbd_connection *connection = peer_device ? peer_device->connection : NULL; | |
b411b363 PR |
1726 | union drbd_state ns; |
1727 | int r; | |
1728 | ||
b30ab791 | 1729 | if (device->state.conn >= C_SYNC_SOURCE && device->state.conn < C_AHEAD) { |
d0180171 | 1730 | drbd_err(device, "Resync already running!\n"); |
b411b363 PR |
1731 | return; |
1732 | } | |
1733 | ||
b30ab791 | 1734 | if (!test_bit(B_RS_H_DONE, &device->flags)) { |
e64a3294 PR |
1735 | if (side == C_SYNC_TARGET) { |
1736 | /* Since application IO was locked out during C_WF_BITMAP_T and | |
1737 | C_WF_SYNC_UUID we are still unmodified. Before going to C_SYNC_TARGET | |
1738 | we check that we might make the data inconsistent. */ | |
b30ab791 | 1739 | r = drbd_khelper(device, "before-resync-target"); |
e64a3294 PR |
1740 | r = (r >> 8) & 0xff; |
1741 | if (r > 0) { | |
d0180171 | 1742 | drbd_info(device, "before-resync-target handler returned %d, " |
09b9e797 | 1743 | "dropping connection.\n", r); |
44a4d551 | 1744 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
09b9e797 PR |
1745 | return; |
1746 | } | |
e64a3294 | 1747 | } else /* C_SYNC_SOURCE */ { |
b30ab791 | 1748 | r = drbd_khelper(device, "before-resync-source"); |
e64a3294 PR |
1749 | r = (r >> 8) & 0xff; |
1750 | if (r > 0) { | |
1751 | if (r == 3) { | |
d0180171 | 1752 | drbd_info(device, "before-resync-source handler returned %d, " |
e64a3294 PR |
1753 | "ignoring. Old userland tools?", r); |
1754 | } else { | |
d0180171 | 1755 | drbd_info(device, "before-resync-source handler returned %d, " |
e64a3294 | 1756 | "dropping connection.\n", r); |
44a4d551 | 1757 | conn_request_state(connection, |
a6b32bc3 | 1758 | NS(conn, C_DISCONNECTING), CS_HARD); |
e64a3294 PR |
1759 | return; |
1760 | } | |
1761 | } | |
09b9e797 | 1762 | } |
b411b363 PR |
1763 | } |
1764 | ||
44a4d551 | 1765 | if (current == connection->worker.task) { |
dad20554 | 1766 | /* The worker should not sleep waiting for state_mutex, |
e64a3294 | 1767 | that can take long */ |
b30ab791 AG |
1768 | if (!mutex_trylock(device->state_mutex)) { |
1769 | set_bit(B_RS_H_DONE, &device->flags); | |
1770 | device->start_resync_timer.expires = jiffies + HZ/5; | |
1771 | add_timer(&device->start_resync_timer); | |
e64a3294 PR |
1772 | return; |
1773 | } | |
1774 | } else { | |
b30ab791 | 1775 | mutex_lock(device->state_mutex); |
e64a3294 | 1776 | } |
b411b363 | 1777 | |
28bc3b8c AG |
1778 | lock_all_resources(); |
1779 | clear_bit(B_RS_H_DONE, &device->flags); | |
a700471b | 1780 | /* Did some connection breakage or IO error race with us? */ |
b30ab791 AG |
1781 | if (device->state.conn < C_CONNECTED |
1782 | || !get_ldev_if_state(device, D_NEGOTIATING)) { | |
28bc3b8c AG |
1783 | unlock_all_resources(); |
1784 | goto out; | |
b411b363 PR |
1785 | } |
1786 | ||
b30ab791 | 1787 | ns = drbd_read_state(device); |
b411b363 | 1788 | |
b30ab791 | 1789 | ns.aftr_isp = !_drbd_may_sync_now(device); |
b411b363 PR |
1790 | |
1791 | ns.conn = side; | |
1792 | ||
1793 | if (side == C_SYNC_TARGET) | |
1794 | ns.disk = D_INCONSISTENT; | |
1795 | else /* side == C_SYNC_SOURCE */ | |
1796 | ns.pdsk = D_INCONSISTENT; | |
1797 | ||
28bc3b8c | 1798 | r = _drbd_set_state(device, ns, CS_VERBOSE, NULL); |
b30ab791 | 1799 | ns = drbd_read_state(device); |
b411b363 PR |
1800 | |
1801 | if (ns.conn < C_CONNECTED) | |
1802 | r = SS_UNKNOWN_ERROR; | |
1803 | ||
1804 | if (r == SS_SUCCESS) { | |
b30ab791 | 1805 | unsigned long tw = drbd_bm_total_weight(device); |
1d7734a0 LE |
1806 | unsigned long now = jiffies; |
1807 | int i; | |
1808 | ||
b30ab791 AG |
1809 | device->rs_failed = 0; |
1810 | device->rs_paused = 0; | |
1811 | device->rs_same_csum = 0; | |
b30ab791 AG |
1812 | device->rs_last_sect_ev = 0; |
1813 | device->rs_total = tw; | |
1814 | device->rs_start = now; | |
1d7734a0 | 1815 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
1816 | device->rs_mark_left[i] = tw; |
1817 | device->rs_mark_time[i] = now; | |
1d7734a0 | 1818 | } |
28bc3b8c | 1819 | drbd_pause_after(device); |
5ab7d2c0 LE |
1820 | /* Forget potentially stale cached per resync extent bit-counts. |
1821 | * Open coded drbd_rs_cancel_all(device), we already have IRQs | |
1822 | * disabled, and know the disk state is ok. */ | |
1823 | spin_lock(&device->al_lock); | |
1824 | lc_reset(device->resync); | |
1825 | device->resync_locked = 0; | |
1826 | device->resync_wenr = LC_FREE; | |
1827 | spin_unlock(&device->al_lock); | |
b411b363 | 1828 | } |
28bc3b8c | 1829 | unlock_all_resources(); |
5a22db89 | 1830 | |
b411b363 | 1831 | if (r == SS_SUCCESS) { |
5ab7d2c0 | 1832 | wake_up(&device->al_wait); /* for lc_reset() above */ |
328e0f12 PR |
1833 | /* reset rs_last_bcast when a resync or verify is started, |
1834 | * to deal with potential jiffies wrap. */ | |
b30ab791 | 1835 | device->rs_last_bcast = jiffies - HZ; |
328e0f12 | 1836 | |
d0180171 | 1837 | drbd_info(device, "Began resync as %s (will sync %lu KB [%lu bits set]).\n", |
b411b363 | 1838 | drbd_conn_str(ns.conn), |
b30ab791 AG |
1839 | (unsigned long) device->rs_total << (BM_BLOCK_SHIFT-10), |
1840 | (unsigned long) device->rs_total); | |
aaaba345 | 1841 | if (side == C_SYNC_TARGET) { |
b30ab791 | 1842 | device->bm_resync_fo = 0; |
aaaba345 LE |
1843 | device->use_csums = use_checksum_based_resync(connection, device); |
1844 | } else { | |
7e5fec31 | 1845 | device->use_csums = false; |
aaaba345 | 1846 | } |
6c922ed5 LE |
1847 | |
1848 | /* Since protocol 96, we must serialize drbd_gen_and_send_sync_uuid | |
1849 | * with w_send_oos, or the sync target will get confused as to | |
1850 | * how much bits to resync. We cannot do that always, because for an | |
1851 | * empty resync and protocol < 95, we need to do it here, as we call | |
1852 | * drbd_resync_finished from here in that case. | |
1853 | * We drbd_gen_and_send_sync_uuid here for protocol < 96, | |
1854 | * and from after_state_ch otherwise. */ | |
44a4d551 LE |
1855 | if (side == C_SYNC_SOURCE && connection->agreed_pro_version < 96) |
1856 | drbd_gen_and_send_sync_uuid(peer_device); | |
b411b363 | 1857 | |
44a4d551 | 1858 | if (connection->agreed_pro_version < 95 && device->rs_total == 0) { |
af85e8e8 LE |
1859 | /* This still has a race (about when exactly the peers |
1860 | * detect connection loss) that can lead to a full sync | |
1861 | * on next handshake. In 8.3.9 we fixed this with explicit | |
1862 | * resync-finished notifications, but the fix | |
1863 | * introduces a protocol change. Sleeping for some | |
1864 | * time longer than the ping interval + timeout on the | |
1865 | * SyncSource, to give the SyncTarget the chance to | |
1866 | * detect connection loss, then waiting for a ping | |
1867 | * response (implicit in drbd_resync_finished) reduces | |
1868 | * the race considerably, but does not solve it. */ | |
44ed167d PR |
1869 | if (side == C_SYNC_SOURCE) { |
1870 | struct net_conf *nc; | |
1871 | int timeo; | |
1872 | ||
1873 | rcu_read_lock(); | |
44a4d551 | 1874 | nc = rcu_dereference(connection->net_conf); |
44ed167d PR |
1875 | timeo = nc->ping_int * HZ + nc->ping_timeo * HZ / 9; |
1876 | rcu_read_unlock(); | |
1877 | schedule_timeout_interruptible(timeo); | |
1878 | } | |
b30ab791 | 1879 | drbd_resync_finished(device); |
b411b363 PR |
1880 | } |
1881 | ||
b30ab791 AG |
1882 | drbd_rs_controller_reset(device); |
1883 | /* ns.conn may already be != device->state.conn, | |
b411b363 PR |
1884 | * we may have been paused in between, or become paused until |
1885 | * the timer triggers. | |
1886 | * No matter, that is handled in resync_timer_fn() */ | |
1887 | if (ns.conn == C_SYNC_TARGET) | |
b30ab791 | 1888 | mod_timer(&device->resync_timer, jiffies); |
b411b363 | 1889 | |
b30ab791 | 1890 | drbd_md_sync(device); |
b411b363 | 1891 | } |
b30ab791 | 1892 | put_ldev(device); |
28bc3b8c | 1893 | out: |
b30ab791 | 1894 | mutex_unlock(device->state_mutex); |
b411b363 PR |
1895 | } |
1896 | ||
e334f550 | 1897 | static void update_on_disk_bitmap(struct drbd_device *device, bool resync_done) |
c7a58db4 LE |
1898 | { |
1899 | struct sib_info sib = { .sib_reason = SIB_SYNC_PROGRESS, }; | |
1900 | device->rs_last_bcast = jiffies; | |
1901 | ||
1902 | if (!get_ldev(device)) | |
1903 | return; | |
1904 | ||
1905 | drbd_bm_write_lazy(device, 0); | |
5ab7d2c0 | 1906 | if (resync_done && is_sync_state(device->state.conn)) |
c7a58db4 | 1907 | drbd_resync_finished(device); |
5ab7d2c0 | 1908 | |
c7a58db4 LE |
1909 | drbd_bcast_event(device, &sib); |
1910 | /* update timestamp, in case it took a while to write out stuff */ | |
1911 | device->rs_last_bcast = jiffies; | |
1912 | put_ldev(device); | |
1913 | } | |
1914 | ||
e334f550 LE |
1915 | static void drbd_ldev_destroy(struct drbd_device *device) |
1916 | { | |
1917 | lc_destroy(device->resync); | |
1918 | device->resync = NULL; | |
1919 | lc_destroy(device->act_log); | |
1920 | device->act_log = NULL; | |
d1b80853 AG |
1921 | |
1922 | __acquire(local); | |
63a7c8ad | 1923 | drbd_backing_dev_free(device, device->ldev); |
d1b80853 AG |
1924 | device->ldev = NULL; |
1925 | __release(local); | |
1926 | ||
e334f550 LE |
1927 | clear_bit(GOING_DISKLESS, &device->flags); |
1928 | wake_up(&device->misc_wait); | |
1929 | } | |
1930 | ||
1931 | static void go_diskless(struct drbd_device *device) | |
1932 | { | |
1933 | D_ASSERT(device, device->state.disk == D_FAILED); | |
1934 | /* we cannot assert local_cnt == 0 here, as get_ldev_if_state will | |
1935 | * inc/dec it frequently. Once we are D_DISKLESS, no one will touch | |
1936 | * the protected members anymore, though, so once put_ldev reaches zero | |
1937 | * again, it will be safe to free them. */ | |
1938 | ||
1939 | /* Try to write changed bitmap pages, read errors may have just | |
1940 | * set some bits outside the area covered by the activity log. | |
1941 | * | |
1942 | * If we have an IO error during the bitmap writeout, | |
1943 | * we will want a full sync next time, just in case. | |
1944 | * (Do we want a specific meta data flag for this?) | |
1945 | * | |
1946 | * If that does not make it to stable storage either, | |
1947 | * we cannot do anything about that anymore. | |
1948 | * | |
1949 | * We still need to check if both bitmap and ldev are present, we may | |
1950 | * end up here after a failed attach, before ldev was even assigned. | |
1951 | */ | |
1952 | if (device->bitmap && device->ldev) { | |
1953 | /* An interrupted resync or similar is allowed to recounts bits | |
1954 | * while we detach. | |
1955 | * Any modifications would not be expected anymore, though. | |
1956 | */ | |
1957 | if (drbd_bitmap_io_from_worker(device, drbd_bm_write, | |
1958 | "detach", BM_LOCKED_TEST_ALLOWED)) { | |
1959 | if (test_bit(WAS_READ_ERROR, &device->flags)) { | |
1960 | drbd_md_set_flag(device, MDF_FULL_SYNC); | |
1961 | drbd_md_sync(device); | |
1962 | } | |
1963 | } | |
1964 | } | |
1965 | ||
1966 | drbd_force_state(device, NS(disk, D_DISKLESS)); | |
1967 | } | |
1968 | ||
ac0acb9e LE |
1969 | static int do_md_sync(struct drbd_device *device) |
1970 | { | |
1971 | drbd_warn(device, "md_sync_timer expired! Worker calls drbd_md_sync().\n"); | |
1972 | drbd_md_sync(device); | |
1973 | return 0; | |
1974 | } | |
1975 | ||
944410e9 LE |
1976 | /* only called from drbd_worker thread, no locking */ |
1977 | void __update_timing_details( | |
1978 | struct drbd_thread_timing_details *tdp, | |
1979 | unsigned int *cb_nr, | |
1980 | void *cb, | |
1981 | const char *fn, const unsigned int line) | |
1982 | { | |
1983 | unsigned int i = *cb_nr % DRBD_THREAD_DETAILS_HIST; | |
1984 | struct drbd_thread_timing_details *td = tdp + i; | |
1985 | ||
1986 | td->start_jif = jiffies; | |
1987 | td->cb_addr = cb; | |
1988 | td->caller_fn = fn; | |
1989 | td->line = line; | |
1990 | td->cb_nr = *cb_nr; | |
1991 | ||
1992 | i = (i+1) % DRBD_THREAD_DETAILS_HIST; | |
1993 | td = tdp + i; | |
1994 | memset(td, 0, sizeof(*td)); | |
1995 | ||
1996 | ++(*cb_nr); | |
1997 | } | |
1998 | ||
e334f550 LE |
1999 | static void do_device_work(struct drbd_device *device, const unsigned long todo) |
2000 | { | |
b47a06d1 | 2001 | if (test_bit(MD_SYNC, &todo)) |
ac0acb9e | 2002 | do_md_sync(device); |
b47a06d1 AG |
2003 | if (test_bit(RS_DONE, &todo) || |
2004 | test_bit(RS_PROGRESS, &todo)) | |
2005 | update_on_disk_bitmap(device, test_bit(RS_DONE, &todo)); | |
2006 | if (test_bit(GO_DISKLESS, &todo)) | |
e334f550 | 2007 | go_diskless(device); |
b47a06d1 | 2008 | if (test_bit(DESTROY_DISK, &todo)) |
e334f550 | 2009 | drbd_ldev_destroy(device); |
b47a06d1 | 2010 | if (test_bit(RS_START, &todo)) |
ac0acb9e | 2011 | do_start_resync(device); |
e334f550 LE |
2012 | } |
2013 | ||
2014 | #define DRBD_DEVICE_WORK_MASK \ | |
2015 | ((1UL << GO_DISKLESS) \ | |
2016 | |(1UL << DESTROY_DISK) \ | |
ac0acb9e LE |
2017 | |(1UL << MD_SYNC) \ |
2018 | |(1UL << RS_START) \ | |
e334f550 LE |
2019 | |(1UL << RS_PROGRESS) \ |
2020 | |(1UL << RS_DONE) \ | |
2021 | ) | |
2022 | ||
2023 | static unsigned long get_work_bits(unsigned long *flags) | |
2024 | { | |
2025 | unsigned long old, new; | |
2026 | do { | |
2027 | old = *flags; | |
2028 | new = old & ~DRBD_DEVICE_WORK_MASK; | |
2029 | } while (cmpxchg(flags, old, new) != old); | |
2030 | return old & DRBD_DEVICE_WORK_MASK; | |
2031 | } | |
2032 | ||
2033 | static void do_unqueued_work(struct drbd_connection *connection) | |
c7a58db4 LE |
2034 | { |
2035 | struct drbd_peer_device *peer_device; | |
2036 | int vnr; | |
2037 | ||
2038 | rcu_read_lock(); | |
2039 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
2040 | struct drbd_device *device = peer_device->device; | |
e334f550 LE |
2041 | unsigned long todo = get_work_bits(&device->flags); |
2042 | if (!todo) | |
c7a58db4 | 2043 | continue; |
5ab7d2c0 | 2044 | |
c7a58db4 LE |
2045 | kref_get(&device->kref); |
2046 | rcu_read_unlock(); | |
e334f550 | 2047 | do_device_work(device, todo); |
c7a58db4 LE |
2048 | kref_put(&device->kref, drbd_destroy_device); |
2049 | rcu_read_lock(); | |
2050 | } | |
2051 | rcu_read_unlock(); | |
2052 | } | |
2053 | ||
a186e478 | 2054 | static bool dequeue_work_batch(struct drbd_work_queue *queue, struct list_head *work_list) |
8c0785a5 LE |
2055 | { |
2056 | spin_lock_irq(&queue->q_lock); | |
15e26f6a | 2057 | list_splice_tail_init(&queue->q, work_list); |
8c0785a5 LE |
2058 | spin_unlock_irq(&queue->q_lock); |
2059 | return !list_empty(work_list); | |
2060 | } | |
2061 | ||
bde89a9e | 2062 | static void wait_for_work(struct drbd_connection *connection, struct list_head *work_list) |
b6dd1a89 LE |
2063 | { |
2064 | DEFINE_WAIT(wait); | |
2065 | struct net_conf *nc; | |
2066 | int uncork, cork; | |
2067 | ||
abde9cc6 | 2068 | dequeue_work_batch(&connection->sender_work, work_list); |
b6dd1a89 LE |
2069 | if (!list_empty(work_list)) |
2070 | return; | |
2071 | ||
2072 | /* Still nothing to do? | |
2073 | * Maybe we still need to close the current epoch, | |
2074 | * even if no new requests are queued yet. | |
2075 | * | |
2076 | * Also, poke TCP, just in case. | |
2077 | * Then wait for new work (or signal). */ | |
2078 | rcu_read_lock(); | |
2079 | nc = rcu_dereference(connection->net_conf); | |
2080 | uncork = nc ? nc->tcp_cork : 0; | |
2081 | rcu_read_unlock(); | |
2082 | if (uncork) { | |
2083 | mutex_lock(&connection->data.mutex); | |
2084 | if (connection->data.socket) | |
2085 | drbd_tcp_uncork(connection->data.socket); | |
2086 | mutex_unlock(&connection->data.mutex); | |
2087 | } | |
2088 | ||
2089 | for (;;) { | |
2090 | int send_barrier; | |
2091 | prepare_to_wait(&connection->sender_work.q_wait, &wait, TASK_INTERRUPTIBLE); | |
0500813f | 2092 | spin_lock_irq(&connection->resource->req_lock); |
b6dd1a89 | 2093 | spin_lock(&connection->sender_work.q_lock); /* FIXME get rid of this one? */ |
bc317a9e | 2094 | if (!list_empty(&connection->sender_work.q)) |
4dd726f0 | 2095 | list_splice_tail_init(&connection->sender_work.q, work_list); |
b6dd1a89 LE |
2096 | spin_unlock(&connection->sender_work.q_lock); /* FIXME get rid of this one? */ |
2097 | if (!list_empty(work_list) || signal_pending(current)) { | |
0500813f | 2098 | spin_unlock_irq(&connection->resource->req_lock); |
b6dd1a89 LE |
2099 | break; |
2100 | } | |
f9c78128 LE |
2101 | |
2102 | /* We found nothing new to do, no to-be-communicated request, | |
2103 | * no other work item. We may still need to close the last | |
2104 | * epoch. Next incoming request epoch will be connection -> | |
2105 | * current transfer log epoch number. If that is different | |
2106 | * from the epoch of the last request we communicated, it is | |
2107 | * safe to send the epoch separating barrier now. | |
2108 | */ | |
2109 | send_barrier = | |
2110 | atomic_read(&connection->current_tle_nr) != | |
2111 | connection->send.current_epoch_nr; | |
0500813f | 2112 | spin_unlock_irq(&connection->resource->req_lock); |
f9c78128 LE |
2113 | |
2114 | if (send_barrier) | |
2115 | maybe_send_barrier(connection, | |
2116 | connection->send.current_epoch_nr + 1); | |
5ab7d2c0 | 2117 | |
e334f550 | 2118 | if (test_bit(DEVICE_WORK_PENDING, &connection->flags)) |
5ab7d2c0 LE |
2119 | break; |
2120 | ||
a80ca1ae LE |
2121 | /* drbd_send() may have called flush_signals() */ |
2122 | if (get_t_state(&connection->worker) != RUNNING) | |
2123 | break; | |
5ab7d2c0 | 2124 | |
b6dd1a89 LE |
2125 | schedule(); |
2126 | /* may be woken up for other things but new work, too, | |
2127 | * e.g. if the current epoch got closed. | |
2128 | * In which case we send the barrier above. */ | |
2129 | } | |
2130 | finish_wait(&connection->sender_work.q_wait, &wait); | |
2131 | ||
2132 | /* someone may have changed the config while we have been waiting above. */ | |
2133 | rcu_read_lock(); | |
2134 | nc = rcu_dereference(connection->net_conf); | |
2135 | cork = nc ? nc->tcp_cork : 0; | |
2136 | rcu_read_unlock(); | |
2137 | mutex_lock(&connection->data.mutex); | |
2138 | if (connection->data.socket) { | |
2139 | if (cork) | |
2140 | drbd_tcp_cork(connection->data.socket); | |
2141 | else if (!uncork) | |
2142 | drbd_tcp_uncork(connection->data.socket); | |
2143 | } | |
2144 | mutex_unlock(&connection->data.mutex); | |
2145 | } | |
2146 | ||
b411b363 PR |
2147 | int drbd_worker(struct drbd_thread *thi) |
2148 | { | |
bde89a9e | 2149 | struct drbd_connection *connection = thi->connection; |
6db7e50a | 2150 | struct drbd_work *w = NULL; |
c06ece6b | 2151 | struct drbd_peer_device *peer_device; |
b411b363 | 2152 | LIST_HEAD(work_list); |
8c0785a5 | 2153 | int vnr; |
b411b363 | 2154 | |
e77a0a5c | 2155 | while (get_t_state(thi) == RUNNING) { |
80822284 | 2156 | drbd_thread_current_set_cpu(thi); |
b411b363 | 2157 | |
944410e9 LE |
2158 | if (list_empty(&work_list)) { |
2159 | update_worker_timing_details(connection, wait_for_work); | |
bde89a9e | 2160 | wait_for_work(connection, &work_list); |
944410e9 | 2161 | } |
b411b363 | 2162 | |
944410e9 LE |
2163 | if (test_and_clear_bit(DEVICE_WORK_PENDING, &connection->flags)) { |
2164 | update_worker_timing_details(connection, do_unqueued_work); | |
e334f550 | 2165 | do_unqueued_work(connection); |
944410e9 | 2166 | } |
5ab7d2c0 | 2167 | |
8c0785a5 | 2168 | if (signal_pending(current)) { |
b411b363 | 2169 | flush_signals(current); |
19393e10 | 2170 | if (get_t_state(thi) == RUNNING) { |
1ec861eb | 2171 | drbd_warn(connection, "Worker got an unexpected signal\n"); |
b411b363 | 2172 | continue; |
19393e10 | 2173 | } |
b411b363 PR |
2174 | break; |
2175 | } | |
2176 | ||
e77a0a5c | 2177 | if (get_t_state(thi) != RUNNING) |
b411b363 | 2178 | break; |
b411b363 | 2179 | |
729e8b87 | 2180 | if (!list_empty(&work_list)) { |
6db7e50a AG |
2181 | w = list_first_entry(&work_list, struct drbd_work, list); |
2182 | list_del_init(&w->list); | |
944410e9 | 2183 | update_worker_timing_details(connection, w->cb); |
6db7e50a | 2184 | if (w->cb(w, connection->cstate < C_WF_REPORT_PARAMS) == 0) |
8c0785a5 | 2185 | continue; |
bde89a9e AG |
2186 | if (connection->cstate >= C_WF_REPORT_PARAMS) |
2187 | conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); | |
b411b363 PR |
2188 | } |
2189 | } | |
b411b363 | 2190 | |
8c0785a5 | 2191 | do { |
944410e9 LE |
2192 | if (test_and_clear_bit(DEVICE_WORK_PENDING, &connection->flags)) { |
2193 | update_worker_timing_details(connection, do_unqueued_work); | |
e334f550 | 2194 | do_unqueued_work(connection); |
944410e9 | 2195 | } |
729e8b87 | 2196 | if (!list_empty(&work_list)) { |
6db7e50a AG |
2197 | w = list_first_entry(&work_list, struct drbd_work, list); |
2198 | list_del_init(&w->list); | |
944410e9 | 2199 | update_worker_timing_details(connection, w->cb); |
6db7e50a | 2200 | w->cb(w, 1); |
729e8b87 LE |
2201 | } else |
2202 | dequeue_work_batch(&connection->sender_work, &work_list); | |
e334f550 | 2203 | } while (!list_empty(&work_list) || test_bit(DEVICE_WORK_PENDING, &connection->flags)); |
b411b363 | 2204 | |
c141ebda | 2205 | rcu_read_lock(); |
c06ece6b AG |
2206 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
2207 | struct drbd_device *device = peer_device->device; | |
0b0ba1ef | 2208 | D_ASSERT(device, device->state.disk == D_DISKLESS && device->state.conn == C_STANDALONE); |
b30ab791 | 2209 | kref_get(&device->kref); |
c141ebda | 2210 | rcu_read_unlock(); |
b30ab791 | 2211 | drbd_device_cleanup(device); |
05a10ec7 | 2212 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda | 2213 | rcu_read_lock(); |
0e29d163 | 2214 | } |
c141ebda | 2215 | rcu_read_unlock(); |
b411b363 PR |
2216 | |
2217 | return 0; | |
2218 | } |