]>
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 AG |
45 | /* endio handlers: |
46 | * drbd_md_io_complete (defined here) | |
fcefa62e AG |
47 | * drbd_request_endio (defined here) |
48 | * drbd_peer_request_endio (defined here) | |
c5a91619 AG |
49 | * bm_async_io_complete (defined in drbd_bitmap.c) |
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 | ||
58 | ||
59 | /* About the global_state_lock | |
60 | Each state transition on an device holds a read lock. In case we have | |
95f8efd0 | 61 | to evaluate the resync after dependencies, we grab a write lock, because |
b411b363 PR |
62 | we need stable states on all devices for that. */ |
63 | rwlock_t global_state_lock; | |
64 | ||
65 | /* used for synchronous meta data and bitmap IO | |
66 | * submitted by drbd_md_sync_page_io() | |
67 | */ | |
68 | void drbd_md_io_complete(struct bio *bio, int error) | |
69 | { | |
70 | struct drbd_md_io *md_io; | |
b30ab791 | 71 | struct drbd_device *device; |
b411b363 PR |
72 | |
73 | md_io = (struct drbd_md_io *)bio->bi_private; | |
b30ab791 | 74 | device = container_of(md_io, struct drbd_device, md_io); |
cdfda633 | 75 | |
b411b363 PR |
76 | md_io->error = error; |
77 | ||
0cfac5dd PR |
78 | /* We grabbed an extra reference in _drbd_md_sync_page_io() to be able |
79 | * to timeout on the lower level device, and eventually detach from it. | |
80 | * If this io completion runs after that timeout expired, this | |
81 | * drbd_md_put_buffer() may allow us to finally try and re-attach. | |
82 | * During normal operation, this only puts that extra reference | |
83 | * down to 1 again. | |
84 | * Make sure we first drop the reference, and only then signal | |
85 | * completion, or we may (in drbd_al_read_log()) cycle so fast into the | |
86 | * next drbd_md_sync_page_io(), that we trigger the | |
b30ab791 | 87 | * ASSERT(atomic_read(&device->md_io_in_use) == 1) there. |
0cfac5dd | 88 | */ |
b30ab791 | 89 | drbd_md_put_buffer(device); |
cdfda633 | 90 | md_io->done = 1; |
b30ab791 | 91 | wake_up(&device->misc_wait); |
cdfda633 | 92 | bio_put(bio); |
b30ab791 AG |
93 | if (device->ldev) /* special case: drbd_md_read() during drbd_adm_attach() */ |
94 | put_ldev(device); | |
b411b363 PR |
95 | } |
96 | ||
97 | /* reads on behalf of the partner, | |
98 | * "submitted" by the receiver | |
99 | */ | |
a186e478 | 100 | static void drbd_endio_read_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
101 | { |
102 | unsigned long flags = 0; | |
6780139c AG |
103 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
104 | struct drbd_device *device = peer_device->device; | |
b411b363 | 105 | |
0500813f | 106 | spin_lock_irqsave(&device->resource->req_lock, flags); |
b30ab791 | 107 | device->read_cnt += peer_req->i.size >> 9; |
a8cd15ba | 108 | list_del(&peer_req->w.list); |
b30ab791 AG |
109 | if (list_empty(&device->read_ee)) |
110 | wake_up(&device->ee_wait); | |
db830c46 | 111 | if (test_bit(__EE_WAS_ERROR, &peer_req->flags)) |
b30ab791 | 112 | __drbd_chk_io_error(device, DRBD_READ_ERROR); |
0500813f | 113 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b411b363 | 114 | |
6780139c | 115 | drbd_queue_work(&peer_device->connection->sender_work, &peer_req->w); |
b30ab791 | 116 | put_ldev(device); |
b411b363 PR |
117 | } |
118 | ||
119 | /* writes on behalf of the partner, or resync writes, | |
45bb912b | 120 | * "submitted" by the receiver, final stage. */ |
a0fb3c47 | 121 | void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
122 | { |
123 | unsigned long flags = 0; | |
6780139c AG |
124 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
125 | struct drbd_device *device = peer_device->device; | |
181286ad | 126 | struct drbd_interval i; |
b411b363 | 127 | int do_wake; |
579b57ed | 128 | u64 block_id; |
b411b363 | 129 | int do_al_complete_io; |
b411b363 | 130 | |
db830c46 | 131 | /* after we moved peer_req to done_ee, |
b411b363 PR |
132 | * we may no longer access it, |
133 | * it may be freed/reused already! | |
134 | * (as soon as we release the req_lock) */ | |
181286ad | 135 | i = peer_req->i; |
db830c46 AG |
136 | do_al_complete_io = peer_req->flags & EE_CALL_AL_COMPLETE_IO; |
137 | block_id = peer_req->block_id; | |
b411b363 | 138 | |
0500813f | 139 | spin_lock_irqsave(&device->resource->req_lock, flags); |
b30ab791 | 140 | device->writ_cnt += peer_req->i.size >> 9; |
a8cd15ba | 141 | list_move_tail(&peer_req->w.list, &device->done_ee); |
b411b363 | 142 | |
bb3bfe96 | 143 | /* |
5e472264 | 144 | * Do not remove from the write_requests tree here: we did not send the |
bb3bfe96 AG |
145 | * Ack yet and did not wake possibly waiting conflicting requests. |
146 | * Removed from the tree from "drbd_process_done_ee" within the | |
84b8c06b | 147 | * appropriate dw.cb (e_end_block/e_end_resync_block) or from |
bb3bfe96 AG |
148 | * _drbd_clear_done_ee. |
149 | */ | |
b411b363 | 150 | |
b30ab791 | 151 | do_wake = list_empty(block_id == ID_SYNCER ? &device->sync_ee : &device->active_ee); |
b411b363 | 152 | |
a0fb3c47 LE |
153 | /* FIXME do we want to detach for failed REQ_DISCARD? |
154 | * ((peer_req->flags & (EE_WAS_ERROR|EE_IS_TRIM)) == EE_WAS_ERROR) */ | |
155 | if (peer_req->flags & EE_WAS_ERROR) | |
b30ab791 | 156 | __drbd_chk_io_error(device, DRBD_WRITE_ERROR); |
0500813f | 157 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b411b363 | 158 | |
579b57ed | 159 | if (block_id == ID_SYNCER) |
b30ab791 | 160 | drbd_rs_complete_io(device, i.sector); |
b411b363 PR |
161 | |
162 | if (do_wake) | |
b30ab791 | 163 | wake_up(&device->ee_wait); |
b411b363 PR |
164 | |
165 | if (do_al_complete_io) | |
b30ab791 | 166 | drbd_al_complete_io(device, &i); |
b411b363 | 167 | |
6780139c | 168 | wake_asender(peer_device->connection); |
b30ab791 | 169 | put_ldev(device); |
45bb912b | 170 | } |
b411b363 | 171 | |
45bb912b LE |
172 | /* writes on behalf of the partner, or resync writes, |
173 | * "submitted" by the receiver. | |
174 | */ | |
fcefa62e | 175 | void drbd_peer_request_endio(struct bio *bio, int error) |
45bb912b | 176 | { |
db830c46 | 177 | struct drbd_peer_request *peer_req = bio->bi_private; |
a8cd15ba | 178 | struct drbd_device *device = peer_req->peer_device->device; |
45bb912b LE |
179 | int uptodate = bio_flagged(bio, BIO_UPTODATE); |
180 | int is_write = bio_data_dir(bio) == WRITE; | |
a0fb3c47 | 181 | int is_discard = !!(bio->bi_rw & REQ_DISCARD); |
45bb912b | 182 | |
07194272 | 183 | if (error && __ratelimit(&drbd_ratelimit_state)) |
d0180171 | 184 | drbd_warn(device, "%s: error=%d s=%llus\n", |
a0fb3c47 LE |
185 | is_write ? (is_discard ? "discard" : "write") |
186 | : "read", error, | |
db830c46 | 187 | (unsigned long long)peer_req->i.sector); |
45bb912b | 188 | if (!error && !uptodate) { |
07194272 | 189 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 190 | drbd_warn(device, "%s: setting error to -EIO s=%llus\n", |
07194272 | 191 | is_write ? "write" : "read", |
db830c46 | 192 | (unsigned long long)peer_req->i.sector); |
45bb912b LE |
193 | /* strange behavior of some lower level drivers... |
194 | * fail the request by clearing the uptodate flag, | |
195 | * but do not return any error?! */ | |
196 | error = -EIO; | |
197 | } | |
198 | ||
199 | if (error) | |
db830c46 | 200 | set_bit(__EE_WAS_ERROR, &peer_req->flags); |
45bb912b LE |
201 | |
202 | bio_put(bio); /* no need for the bio anymore */ | |
db830c46 | 203 | if (atomic_dec_and_test(&peer_req->pending_bios)) { |
45bb912b | 204 | if (is_write) |
db830c46 | 205 | drbd_endio_write_sec_final(peer_req); |
45bb912b | 206 | else |
db830c46 | 207 | drbd_endio_read_sec_final(peer_req); |
45bb912b | 208 | } |
b411b363 PR |
209 | } |
210 | ||
211 | /* read, readA or write requests on R_PRIMARY coming from drbd_make_request | |
212 | */ | |
fcefa62e | 213 | void drbd_request_endio(struct bio *bio, int error) |
b411b363 | 214 | { |
a115413d | 215 | unsigned long flags; |
b411b363 | 216 | struct drbd_request *req = bio->bi_private; |
84b8c06b | 217 | struct drbd_device *device = req->device; |
a115413d | 218 | struct bio_and_error m; |
b411b363 PR |
219 | enum drbd_req_event what; |
220 | int uptodate = bio_flagged(bio, BIO_UPTODATE); | |
221 | ||
b411b363 | 222 | if (!error && !uptodate) { |
d0180171 | 223 | drbd_warn(device, "p %s: setting error to -EIO\n", |
b411b363 PR |
224 | bio_data_dir(bio) == WRITE ? "write" : "read"); |
225 | /* strange behavior of some lower level drivers... | |
226 | * fail the request by clearing the uptodate flag, | |
227 | * but do not return any error?! */ | |
228 | error = -EIO; | |
229 | } | |
230 | ||
1b6dd252 PR |
231 | |
232 | /* If this request was aborted locally before, | |
233 | * but now was completed "successfully", | |
234 | * chances are that this caused arbitrary data corruption. | |
235 | * | |
236 | * "aborting" requests, or force-detaching the disk, is intended for | |
237 | * completely blocked/hung local backing devices which do no longer | |
238 | * complete requests at all, not even do error completions. In this | |
239 | * situation, usually a hard-reset and failover is the only way out. | |
240 | * | |
241 | * By "aborting", basically faking a local error-completion, | |
242 | * we allow for a more graceful swichover by cleanly migrating services. | |
243 | * Still the affected node has to be rebooted "soon". | |
244 | * | |
245 | * By completing these requests, we allow the upper layers to re-use | |
246 | * the associated data pages. | |
247 | * | |
248 | * If later the local backing device "recovers", and now DMAs some data | |
249 | * from disk into the original request pages, in the best case it will | |
250 | * just put random data into unused pages; but typically it will corrupt | |
251 | * meanwhile completely unrelated data, causing all sorts of damage. | |
252 | * | |
253 | * Which means delayed successful completion, | |
254 | * especially for READ requests, | |
255 | * is a reason to panic(). | |
256 | * | |
257 | * We assume that a delayed *error* completion is OK, | |
258 | * though we still will complain noisily about it. | |
259 | */ | |
260 | if (unlikely(req->rq_state & RQ_LOCAL_ABORTED)) { | |
261 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 262 | drbd_emerg(device, "delayed completion of aborted local request; disk-timeout may be too aggressive\n"); |
1b6dd252 PR |
263 | |
264 | if (!error) | |
265 | panic("possible random memory corruption caused by delayed completion of aborted local request\n"); | |
266 | } | |
267 | ||
b411b363 PR |
268 | /* to avoid recursion in __req_mod */ |
269 | if (unlikely(error)) { | |
2f632aeb LE |
270 | if (bio->bi_rw & REQ_DISCARD) |
271 | what = (error == -EOPNOTSUPP) | |
272 | ? DISCARD_COMPLETED_NOTSUPP | |
273 | : DISCARD_COMPLETED_WITH_ERROR; | |
274 | else | |
275 | what = (bio_data_dir(bio) == WRITE) | |
8554df1c | 276 | ? WRITE_COMPLETED_WITH_ERROR |
5c3c7e64 | 277 | : (bio_rw(bio) == READ) |
8554df1c AG |
278 | ? READ_COMPLETED_WITH_ERROR |
279 | : READ_AHEAD_COMPLETED_WITH_ERROR; | |
b411b363 | 280 | } else |
8554df1c | 281 | what = COMPLETED_OK; |
b411b363 PR |
282 | |
283 | bio_put(req->private_bio); | |
284 | req->private_bio = ERR_PTR(error); | |
285 | ||
a115413d | 286 | /* not req_mod(), we need irqsave here! */ |
0500813f | 287 | spin_lock_irqsave(&device->resource->req_lock, flags); |
a115413d | 288 | __req_mod(req, what, &m); |
0500813f | 289 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b30ab791 | 290 | put_ldev(device); |
a115413d LE |
291 | |
292 | if (m.bio) | |
b30ab791 | 293 | complete_master_bio(device, &m); |
b411b363 PR |
294 | } |
295 | ||
79a3c8d3 | 296 | void drbd_csum_ee(struct crypto_hash *tfm, struct drbd_peer_request *peer_req, void *digest) |
45bb912b LE |
297 | { |
298 | struct hash_desc desc; | |
299 | struct scatterlist sg; | |
db830c46 | 300 | struct page *page = peer_req->pages; |
45bb912b LE |
301 | struct page *tmp; |
302 | unsigned len; | |
303 | ||
304 | desc.tfm = tfm; | |
305 | desc.flags = 0; | |
306 | ||
307 | sg_init_table(&sg, 1); | |
308 | crypto_hash_init(&desc); | |
309 | ||
310 | while ((tmp = page_chain_next(page))) { | |
311 | /* all but the last page will be fully used */ | |
312 | sg_set_page(&sg, page, PAGE_SIZE, 0); | |
313 | crypto_hash_update(&desc, &sg, sg.length); | |
314 | page = tmp; | |
315 | } | |
316 | /* and now the last, possibly only partially used page */ | |
db830c46 | 317 | len = peer_req->i.size & (PAGE_SIZE - 1); |
45bb912b LE |
318 | sg_set_page(&sg, page, len ?: PAGE_SIZE, 0); |
319 | crypto_hash_update(&desc, &sg, sg.length); | |
320 | crypto_hash_final(&desc, digest); | |
321 | } | |
322 | ||
79a3c8d3 | 323 | void drbd_csum_bio(struct crypto_hash *tfm, struct bio *bio, void *digest) |
b411b363 PR |
324 | { |
325 | struct hash_desc desc; | |
326 | struct scatterlist sg; | |
7988613b KO |
327 | struct bio_vec bvec; |
328 | struct bvec_iter iter; | |
b411b363 PR |
329 | |
330 | desc.tfm = tfm; | |
331 | desc.flags = 0; | |
332 | ||
333 | sg_init_table(&sg, 1); | |
334 | crypto_hash_init(&desc); | |
335 | ||
7988613b KO |
336 | bio_for_each_segment(bvec, bio, iter) { |
337 | sg_set_page(&sg, bvec.bv_page, bvec.bv_len, bvec.bv_offset); | |
b411b363 PR |
338 | crypto_hash_update(&desc, &sg, sg.length); |
339 | } | |
340 | crypto_hash_final(&desc, digest); | |
341 | } | |
342 | ||
9676c760 | 343 | /* MAYBE merge common code with w_e_end_ov_req */ |
99920dc5 | 344 | static int w_e_send_csum(struct drbd_work *w, int cancel) |
b411b363 | 345 | { |
a8cd15ba | 346 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
347 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
348 | struct drbd_device *device = peer_device->device; | |
b411b363 PR |
349 | int digest_size; |
350 | void *digest; | |
99920dc5 | 351 | int err = 0; |
b411b363 | 352 | |
53ea4331 LE |
353 | if (unlikely(cancel)) |
354 | goto out; | |
b411b363 | 355 | |
9676c760 | 356 | if (unlikely((peer_req->flags & EE_WAS_ERROR) != 0)) |
53ea4331 | 357 | goto out; |
b411b363 | 358 | |
6780139c | 359 | digest_size = crypto_hash_digestsize(peer_device->connection->csums_tfm); |
53ea4331 LE |
360 | digest = kmalloc(digest_size, GFP_NOIO); |
361 | if (digest) { | |
db830c46 AG |
362 | sector_t sector = peer_req->i.sector; |
363 | unsigned int size = peer_req->i.size; | |
6780139c | 364 | drbd_csum_ee(peer_device->connection->csums_tfm, peer_req, digest); |
9676c760 | 365 | /* Free peer_req and pages before send. |
53ea4331 LE |
366 | * In case we block on congestion, we could otherwise run into |
367 | * some distributed deadlock, if the other side blocks on | |
368 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 369 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 370 | drbd_free_peer_req(device, peer_req); |
db830c46 | 371 | peer_req = NULL; |
b30ab791 | 372 | inc_rs_pending(device); |
6780139c | 373 | err = drbd_send_drequest_csum(peer_device, sector, size, |
db1b0b72 AG |
374 | digest, digest_size, |
375 | P_CSUM_RS_REQUEST); | |
53ea4331 LE |
376 | kfree(digest); |
377 | } else { | |
d0180171 | 378 | drbd_err(device, "kmalloc() of digest failed.\n"); |
99920dc5 | 379 | err = -ENOMEM; |
53ea4331 | 380 | } |
b411b363 | 381 | |
53ea4331 | 382 | out: |
db830c46 | 383 | if (peer_req) |
b30ab791 | 384 | drbd_free_peer_req(device, peer_req); |
b411b363 | 385 | |
99920dc5 | 386 | if (unlikely(err)) |
d0180171 | 387 | drbd_err(device, "drbd_send_drequest(..., csum) failed\n"); |
99920dc5 | 388 | return err; |
b411b363 PR |
389 | } |
390 | ||
391 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
392 | ||
69a22773 | 393 | static int read_for_csum(struct drbd_peer_device *peer_device, sector_t sector, int size) |
b411b363 | 394 | { |
69a22773 | 395 | struct drbd_device *device = peer_device->device; |
db830c46 | 396 | struct drbd_peer_request *peer_req; |
b411b363 | 397 | |
b30ab791 | 398 | if (!get_ldev(device)) |
80a40e43 | 399 | return -EIO; |
b411b363 | 400 | |
b30ab791 | 401 | if (drbd_rs_should_slow_down(device, sector)) |
0f0601f4 LE |
402 | goto defer; |
403 | ||
b411b363 PR |
404 | /* GFP_TRY, because if there is no memory available right now, this may |
405 | * be rescheduled for later. It is "only" background resync, after all. */ | |
69a22773 | 406 | peer_req = drbd_alloc_peer_req(peer_device, ID_SYNCER /* unused */, sector, |
a0fb3c47 | 407 | size, true /* has real payload */, GFP_TRY); |
db830c46 | 408 | if (!peer_req) |
80a40e43 | 409 | goto defer; |
b411b363 | 410 | |
a8cd15ba | 411 | peer_req->w.cb = w_e_send_csum; |
0500813f | 412 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 413 | list_add(&peer_req->w.list, &device->read_ee); |
0500813f | 414 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 415 | |
b30ab791 AG |
416 | atomic_add(size >> 9, &device->rs_sect_ev); |
417 | if (drbd_submit_peer_request(device, peer_req, READ, DRBD_FAULT_RS_RD) == 0) | |
80a40e43 | 418 | return 0; |
b411b363 | 419 | |
10f6d992 LE |
420 | /* If it failed because of ENOMEM, retry should help. If it failed |
421 | * because bio_add_page failed (probably broken lower level driver), | |
422 | * retry may or may not help. | |
423 | * If it does not, you may need to force disconnect. */ | |
0500813f | 424 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 425 | list_del(&peer_req->w.list); |
0500813f | 426 | spin_unlock_irq(&device->resource->req_lock); |
22cc37a9 | 427 | |
b30ab791 | 428 | drbd_free_peer_req(device, peer_req); |
80a40e43 | 429 | defer: |
b30ab791 | 430 | put_ldev(device); |
80a40e43 | 431 | return -EAGAIN; |
b411b363 PR |
432 | } |
433 | ||
99920dc5 | 434 | int w_resync_timer(struct drbd_work *w, int cancel) |
b411b363 | 435 | { |
84b8c06b AG |
436 | struct drbd_device *device = |
437 | container_of(w, struct drbd_device, resync_work); | |
438 | ||
b30ab791 | 439 | switch (device->state.conn) { |
63106d3c | 440 | case C_VERIFY_S: |
d448a2e1 | 441 | make_ov_request(device, cancel); |
63106d3c PR |
442 | break; |
443 | case C_SYNC_TARGET: | |
d448a2e1 | 444 | make_resync_request(device, cancel); |
63106d3c | 445 | break; |
b411b363 PR |
446 | } |
447 | ||
99920dc5 | 448 | return 0; |
794abb75 PR |
449 | } |
450 | ||
451 | void resync_timer_fn(unsigned long data) | |
452 | { | |
b30ab791 | 453 | struct drbd_device *device = (struct drbd_device *) data; |
794abb75 | 454 | |
b30ab791 | 455 | if (list_empty(&device->resync_work.list)) |
84b8c06b AG |
456 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, |
457 | &device->resync_work); | |
b411b363 PR |
458 | } |
459 | ||
778f271d PR |
460 | static void fifo_set(struct fifo_buffer *fb, int value) |
461 | { | |
462 | int i; | |
463 | ||
464 | for (i = 0; i < fb->size; i++) | |
f10f2623 | 465 | fb->values[i] = value; |
778f271d PR |
466 | } |
467 | ||
468 | static int fifo_push(struct fifo_buffer *fb, int value) | |
469 | { | |
470 | int ov; | |
471 | ||
472 | ov = fb->values[fb->head_index]; | |
473 | fb->values[fb->head_index++] = value; | |
474 | ||
475 | if (fb->head_index >= fb->size) | |
476 | fb->head_index = 0; | |
477 | ||
478 | return ov; | |
479 | } | |
480 | ||
481 | static void fifo_add_val(struct fifo_buffer *fb, int value) | |
482 | { | |
483 | int i; | |
484 | ||
485 | for (i = 0; i < fb->size; i++) | |
486 | fb->values[i] += value; | |
487 | } | |
488 | ||
9958c857 PR |
489 | struct fifo_buffer *fifo_alloc(int fifo_size) |
490 | { | |
491 | struct fifo_buffer *fb; | |
492 | ||
8747d30a | 493 | fb = kzalloc(sizeof(struct fifo_buffer) + sizeof(int) * fifo_size, GFP_NOIO); |
9958c857 PR |
494 | if (!fb) |
495 | return NULL; | |
496 | ||
497 | fb->head_index = 0; | |
498 | fb->size = fifo_size; | |
499 | fb->total = 0; | |
500 | ||
501 | return fb; | |
502 | } | |
503 | ||
0e49d7b0 | 504 | static int drbd_rs_controller(struct drbd_device *device, unsigned int sect_in) |
778f271d | 505 | { |
daeda1cc | 506 | struct disk_conf *dc; |
778f271d PR |
507 | unsigned int want; /* The number of sectors we want in the proxy */ |
508 | int req_sect; /* Number of sectors to request in this turn */ | |
509 | int correction; /* Number of sectors more we need in the proxy*/ | |
510 | int cps; /* correction per invocation of drbd_rs_controller() */ | |
511 | int steps; /* Number of time steps to plan ahead */ | |
512 | int curr_corr; | |
513 | int max_sect; | |
813472ce | 514 | struct fifo_buffer *plan; |
778f271d | 515 | |
b30ab791 AG |
516 | dc = rcu_dereference(device->ldev->disk_conf); |
517 | plan = rcu_dereference(device->rs_plan_s); | |
778f271d | 518 | |
813472ce | 519 | steps = plan->size; /* (dc->c_plan_ahead * 10 * SLEEP_TIME) / HZ; */ |
778f271d | 520 | |
b30ab791 | 521 | if (device->rs_in_flight + sect_in == 0) { /* At start of resync */ |
daeda1cc | 522 | want = ((dc->resync_rate * 2 * SLEEP_TIME) / HZ) * steps; |
778f271d | 523 | } else { /* normal path */ |
daeda1cc PR |
524 | want = dc->c_fill_target ? dc->c_fill_target : |
525 | sect_in * dc->c_delay_target * HZ / (SLEEP_TIME * 10); | |
778f271d PR |
526 | } |
527 | ||
b30ab791 | 528 | correction = want - device->rs_in_flight - plan->total; |
778f271d PR |
529 | |
530 | /* Plan ahead */ | |
531 | cps = correction / steps; | |
813472ce PR |
532 | fifo_add_val(plan, cps); |
533 | plan->total += cps * steps; | |
778f271d PR |
534 | |
535 | /* What we do in this step */ | |
813472ce PR |
536 | curr_corr = fifo_push(plan, 0); |
537 | plan->total -= curr_corr; | |
778f271d PR |
538 | |
539 | req_sect = sect_in + curr_corr; | |
540 | if (req_sect < 0) | |
541 | req_sect = 0; | |
542 | ||
daeda1cc | 543 | max_sect = (dc->c_max_rate * 2 * SLEEP_TIME) / HZ; |
778f271d PR |
544 | if (req_sect > max_sect) |
545 | req_sect = max_sect; | |
546 | ||
547 | /* | |
d0180171 | 548 | drbd_warn(device, "si=%u if=%d wa=%u co=%d st=%d cps=%d pl=%d cc=%d rs=%d\n", |
b30ab791 AG |
549 | sect_in, device->rs_in_flight, want, correction, |
550 | steps, cps, device->rs_planed, curr_corr, req_sect); | |
778f271d PR |
551 | */ |
552 | ||
553 | return req_sect; | |
554 | } | |
555 | ||
b30ab791 | 556 | static int drbd_rs_number_requests(struct drbd_device *device) |
e65f440d | 557 | { |
0e49d7b0 LE |
558 | unsigned int sect_in; /* Number of sectors that came in since the last turn */ |
559 | int number, mxb; | |
560 | ||
561 | sect_in = atomic_xchg(&device->rs_sect_in, 0); | |
562 | device->rs_in_flight -= sect_in; | |
813472ce PR |
563 | |
564 | rcu_read_lock(); | |
0e49d7b0 | 565 | mxb = drbd_get_max_buffers(device) / 2; |
b30ab791 | 566 | if (rcu_dereference(device->rs_plan_s)->size) { |
0e49d7b0 | 567 | number = drbd_rs_controller(device, sect_in) >> (BM_BLOCK_SHIFT - 9); |
b30ab791 | 568 | device->c_sync_rate = number * HZ * (BM_BLOCK_SIZE / 1024) / SLEEP_TIME; |
e65f440d | 569 | } else { |
b30ab791 AG |
570 | device->c_sync_rate = rcu_dereference(device->ldev->disk_conf)->resync_rate; |
571 | number = SLEEP_TIME * device->c_sync_rate / ((BM_BLOCK_SIZE / 1024) * HZ); | |
e65f440d | 572 | } |
813472ce | 573 | rcu_read_unlock(); |
e65f440d | 574 | |
0e49d7b0 LE |
575 | /* Don't have more than "max-buffers"/2 in-flight. |
576 | * Otherwise we may cause the remote site to stall on drbd_alloc_pages(), | |
577 | * potentially causing a distributed deadlock on congestion during | |
578 | * online-verify or (checksum-based) resync, if max-buffers, | |
579 | * socket buffer sizes and resync rate settings are mis-configured. */ | |
580 | if (mxb - device->rs_in_flight < number) | |
581 | number = mxb - device->rs_in_flight; | |
582 | ||
e65f440d LE |
583 | return number; |
584 | } | |
585 | ||
d448a2e1 | 586 | static int make_resync_request(struct drbd_device *device, int cancel) |
b411b363 PR |
587 | { |
588 | unsigned long bit; | |
589 | sector_t sector; | |
b30ab791 | 590 | const sector_t capacity = drbd_get_capacity(device->this_bdev); |
1816a2b4 | 591 | int max_bio_size; |
e65f440d | 592 | int number, rollback_i, size; |
b411b363 | 593 | int align, queued, sndbuf; |
0f0601f4 | 594 | int i = 0; |
b411b363 PR |
595 | |
596 | if (unlikely(cancel)) | |
99920dc5 | 597 | return 0; |
b411b363 | 598 | |
b30ab791 | 599 | if (device->rs_total == 0) { |
af85e8e8 | 600 | /* empty resync? */ |
b30ab791 | 601 | drbd_resync_finished(device); |
99920dc5 | 602 | return 0; |
af85e8e8 LE |
603 | } |
604 | ||
b30ab791 AG |
605 | if (!get_ldev(device)) { |
606 | /* Since we only need to access device->rsync a | |
607 | get_ldev_if_state(device,D_FAILED) would be sufficient, but | |
b411b363 PR |
608 | to continue resync with a broken disk makes no sense at |
609 | all */ | |
d0180171 | 610 | drbd_err(device, "Disk broke down during resync!\n"); |
99920dc5 | 611 | return 0; |
b411b363 PR |
612 | } |
613 | ||
b30ab791 AG |
614 | max_bio_size = queue_max_hw_sectors(device->rq_queue) << 9; |
615 | number = drbd_rs_number_requests(device); | |
0e49d7b0 | 616 | if (number <= 0) |
0f0601f4 | 617 | goto requeue; |
b411b363 | 618 | |
b411b363 PR |
619 | for (i = 0; i < number; i++) { |
620 | /* Stop generating RS requests, when half of the send buffer is filled */ | |
a6b32bc3 AG |
621 | mutex_lock(&first_peer_device(device)->connection->data.mutex); |
622 | if (first_peer_device(device)->connection->data.socket) { | |
623 | queued = first_peer_device(device)->connection->data.socket->sk->sk_wmem_queued; | |
624 | sndbuf = first_peer_device(device)->connection->data.socket->sk->sk_sndbuf; | |
b411b363 PR |
625 | } else { |
626 | queued = 1; | |
627 | sndbuf = 0; | |
628 | } | |
a6b32bc3 | 629 | mutex_unlock(&first_peer_device(device)->connection->data.mutex); |
b411b363 PR |
630 | if (queued > sndbuf / 2) |
631 | goto requeue; | |
632 | ||
633 | next_sector: | |
634 | size = BM_BLOCK_SIZE; | |
b30ab791 | 635 | bit = drbd_bm_find_next(device, device->bm_resync_fo); |
b411b363 | 636 | |
4b0715f0 | 637 | if (bit == DRBD_END_OF_BITMAP) { |
b30ab791 AG |
638 | device->bm_resync_fo = drbd_bm_bits(device); |
639 | put_ldev(device); | |
99920dc5 | 640 | return 0; |
b411b363 PR |
641 | } |
642 | ||
643 | sector = BM_BIT_TO_SECT(bit); | |
644 | ||
b30ab791 AG |
645 | if (drbd_rs_should_slow_down(device, sector) || |
646 | drbd_try_rs_begin_io(device, sector)) { | |
647 | device->bm_resync_fo = bit; | |
b411b363 PR |
648 | goto requeue; |
649 | } | |
b30ab791 | 650 | device->bm_resync_fo = bit + 1; |
b411b363 | 651 | |
b30ab791 AG |
652 | if (unlikely(drbd_bm_test_bit(device, bit) == 0)) { |
653 | drbd_rs_complete_io(device, sector); | |
b411b363 PR |
654 | goto next_sector; |
655 | } | |
656 | ||
1816a2b4 | 657 | #if DRBD_MAX_BIO_SIZE > BM_BLOCK_SIZE |
b411b363 PR |
658 | /* try to find some adjacent bits. |
659 | * we stop if we have already the maximum req size. | |
660 | * | |
661 | * Additionally always align bigger requests, in order to | |
662 | * be prepared for all stripe sizes of software RAIDs. | |
b411b363 PR |
663 | */ |
664 | align = 1; | |
d207450c | 665 | rollback_i = i; |
6377b923 | 666 | while (i < number) { |
1816a2b4 | 667 | if (size + BM_BLOCK_SIZE > max_bio_size) |
b411b363 PR |
668 | break; |
669 | ||
670 | /* Be always aligned */ | |
671 | if (sector & ((1<<(align+3))-1)) | |
672 | break; | |
673 | ||
674 | /* do not cross extent boundaries */ | |
675 | if (((bit+1) & BM_BLOCKS_PER_BM_EXT_MASK) == 0) | |
676 | break; | |
677 | /* now, is it actually dirty, after all? | |
678 | * caution, drbd_bm_test_bit is tri-state for some | |
679 | * obscure reason; ( b == 0 ) would get the out-of-band | |
680 | * only accidentally right because of the "oddly sized" | |
681 | * adjustment below */ | |
b30ab791 | 682 | if (drbd_bm_test_bit(device, bit+1) != 1) |
b411b363 PR |
683 | break; |
684 | bit++; | |
685 | size += BM_BLOCK_SIZE; | |
686 | if ((BM_BLOCK_SIZE << align) <= size) | |
687 | align++; | |
688 | i++; | |
689 | } | |
690 | /* if we merged some, | |
691 | * reset the offset to start the next drbd_bm_find_next from */ | |
692 | if (size > BM_BLOCK_SIZE) | |
b30ab791 | 693 | device->bm_resync_fo = bit + 1; |
b411b363 PR |
694 | #endif |
695 | ||
696 | /* adjust very last sectors, in case we are oddly sized */ | |
697 | if (sector + (size>>9) > capacity) | |
698 | size = (capacity-sector)<<9; | |
a6b32bc3 AG |
699 | if (first_peer_device(device)->connection->agreed_pro_version >= 89 && |
700 | first_peer_device(device)->connection->csums_tfm) { | |
69a22773 | 701 | switch (read_for_csum(first_peer_device(device), sector, size)) { |
80a40e43 | 702 | case -EIO: /* Disk failure */ |
b30ab791 | 703 | put_ldev(device); |
99920dc5 | 704 | return -EIO; |
80a40e43 | 705 | case -EAGAIN: /* allocation failed, or ldev busy */ |
b30ab791 AG |
706 | drbd_rs_complete_io(device, sector); |
707 | device->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
d207450c | 708 | i = rollback_i; |
b411b363 | 709 | goto requeue; |
80a40e43 LE |
710 | case 0: |
711 | /* everything ok */ | |
712 | break; | |
713 | default: | |
714 | BUG(); | |
b411b363 PR |
715 | } |
716 | } else { | |
99920dc5 AG |
717 | int err; |
718 | ||
b30ab791 | 719 | inc_rs_pending(device); |
69a22773 | 720 | err = drbd_send_drequest(first_peer_device(device), P_RS_DATA_REQUEST, |
99920dc5 AG |
721 | sector, size, ID_SYNCER); |
722 | if (err) { | |
d0180171 | 723 | drbd_err(device, "drbd_send_drequest() failed, aborting...\n"); |
b30ab791 AG |
724 | dec_rs_pending(device); |
725 | put_ldev(device); | |
99920dc5 | 726 | return err; |
b411b363 PR |
727 | } |
728 | } | |
729 | } | |
730 | ||
b30ab791 | 731 | if (device->bm_resync_fo >= drbd_bm_bits(device)) { |
b411b363 PR |
732 | /* last syncer _request_ was sent, |
733 | * but the P_RS_DATA_REPLY not yet received. sync will end (and | |
734 | * next sync group will resume), as soon as we receive the last | |
735 | * resync data block, and the last bit is cleared. | |
736 | * until then resync "work" is "inactive" ... | |
737 | */ | |
b30ab791 | 738 | put_ldev(device); |
99920dc5 | 739 | return 0; |
b411b363 PR |
740 | } |
741 | ||
742 | requeue: | |
b30ab791 AG |
743 | device->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
744 | mod_timer(&device->resync_timer, jiffies + SLEEP_TIME); | |
745 | put_ldev(device); | |
99920dc5 | 746 | return 0; |
b411b363 PR |
747 | } |
748 | ||
d448a2e1 | 749 | static int make_ov_request(struct drbd_device *device, int cancel) |
b411b363 PR |
750 | { |
751 | int number, i, size; | |
752 | sector_t sector; | |
b30ab791 | 753 | const sector_t capacity = drbd_get_capacity(device->this_bdev); |
58ffa580 | 754 | bool stop_sector_reached = false; |
b411b363 PR |
755 | |
756 | if (unlikely(cancel)) | |
757 | return 1; | |
758 | ||
b30ab791 | 759 | number = drbd_rs_number_requests(device); |
b411b363 | 760 | |
b30ab791 | 761 | sector = device->ov_position; |
b411b363 | 762 | for (i = 0; i < number; i++) { |
58ffa580 | 763 | if (sector >= capacity) |
b411b363 | 764 | return 1; |
58ffa580 LE |
765 | |
766 | /* We check for "finished" only in the reply path: | |
767 | * w_e_end_ov_reply(). | |
768 | * We need to send at least one request out. */ | |
769 | stop_sector_reached = i > 0 | |
b30ab791 AG |
770 | && verify_can_do_stop_sector(device) |
771 | && sector >= device->ov_stop_sector; | |
58ffa580 LE |
772 | if (stop_sector_reached) |
773 | break; | |
b411b363 PR |
774 | |
775 | size = BM_BLOCK_SIZE; | |
776 | ||
b30ab791 AG |
777 | if (drbd_rs_should_slow_down(device, sector) || |
778 | drbd_try_rs_begin_io(device, sector)) { | |
779 | device->ov_position = sector; | |
b411b363 PR |
780 | goto requeue; |
781 | } | |
782 | ||
783 | if (sector + (size>>9) > capacity) | |
784 | size = (capacity-sector)<<9; | |
785 | ||
b30ab791 | 786 | inc_rs_pending(device); |
69a22773 | 787 | if (drbd_send_ov_request(first_peer_device(device), sector, size)) { |
b30ab791 | 788 | dec_rs_pending(device); |
b411b363 PR |
789 | return 0; |
790 | } | |
791 | sector += BM_SECT_PER_BIT; | |
792 | } | |
b30ab791 | 793 | device->ov_position = sector; |
b411b363 PR |
794 | |
795 | requeue: | |
b30ab791 | 796 | device->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
58ffa580 | 797 | if (i == 0 || !stop_sector_reached) |
b30ab791 | 798 | mod_timer(&device->resync_timer, jiffies + SLEEP_TIME); |
b411b363 PR |
799 | return 1; |
800 | } | |
801 | ||
99920dc5 | 802 | int w_ov_finished(struct drbd_work *w, int cancel) |
b411b363 | 803 | { |
84b8c06b AG |
804 | struct drbd_device_work *dw = |
805 | container_of(w, struct drbd_device_work, w); | |
806 | struct drbd_device *device = dw->device; | |
807 | kfree(dw); | |
b30ab791 AG |
808 | ov_out_of_sync_print(device); |
809 | drbd_resync_finished(device); | |
b411b363 | 810 | |
99920dc5 | 811 | return 0; |
b411b363 PR |
812 | } |
813 | ||
99920dc5 | 814 | static int w_resync_finished(struct drbd_work *w, int cancel) |
b411b363 | 815 | { |
84b8c06b AG |
816 | struct drbd_device_work *dw = |
817 | container_of(w, struct drbd_device_work, w); | |
818 | struct drbd_device *device = dw->device; | |
819 | kfree(dw); | |
b411b363 | 820 | |
b30ab791 | 821 | drbd_resync_finished(device); |
b411b363 | 822 | |
99920dc5 | 823 | return 0; |
b411b363 PR |
824 | } |
825 | ||
b30ab791 | 826 | static void ping_peer(struct drbd_device *device) |
af85e8e8 | 827 | { |
a6b32bc3 | 828 | struct drbd_connection *connection = first_peer_device(device)->connection; |
2a67d8b9 | 829 | |
bde89a9e AG |
830 | clear_bit(GOT_PING_ACK, &connection->flags); |
831 | request_ping(connection); | |
832 | wait_event(connection->ping_wait, | |
833 | test_bit(GOT_PING_ACK, &connection->flags) || device->state.conn < C_CONNECTED); | |
af85e8e8 LE |
834 | } |
835 | ||
b30ab791 | 836 | int drbd_resync_finished(struct drbd_device *device) |
b411b363 PR |
837 | { |
838 | unsigned long db, dt, dbdt; | |
839 | unsigned long n_oos; | |
840 | union drbd_state os, ns; | |
84b8c06b | 841 | struct drbd_device_work *dw; |
b411b363 | 842 | char *khelper_cmd = NULL; |
26525618 | 843 | int verify_done = 0; |
b411b363 PR |
844 | |
845 | /* Remove all elements from the resync LRU. Since future actions | |
846 | * might set bits in the (main) bitmap, then the entries in the | |
847 | * resync LRU would be wrong. */ | |
b30ab791 | 848 | if (drbd_rs_del_all(device)) { |
b411b363 PR |
849 | /* In case this is not possible now, most probably because |
850 | * there are P_RS_DATA_REPLY Packets lingering on the worker's | |
851 | * queue (or even the read operations for those packets | |
852 | * is not finished by now). Retry in 100ms. */ | |
853 | ||
20ee6390 | 854 | schedule_timeout_interruptible(HZ / 10); |
84b8c06b AG |
855 | dw = kmalloc(sizeof(struct drbd_device_work), GFP_ATOMIC); |
856 | if (dw) { | |
857 | dw->w.cb = w_resync_finished; | |
858 | dw->device = device; | |
859 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, | |
860 | &dw->w); | |
b411b363 PR |
861 | return 1; |
862 | } | |
84b8c06b | 863 | drbd_err(device, "Warn failed to drbd_rs_del_all() and to kmalloc(dw).\n"); |
b411b363 PR |
864 | } |
865 | ||
b30ab791 | 866 | dt = (jiffies - device->rs_start - device->rs_paused) / HZ; |
b411b363 PR |
867 | if (dt <= 0) |
868 | dt = 1; | |
84b8c06b | 869 | |
b30ab791 | 870 | db = device->rs_total; |
58ffa580 | 871 | /* adjust for verify start and stop sectors, respective reached position */ |
b30ab791 AG |
872 | if (device->state.conn == C_VERIFY_S || device->state.conn == C_VERIFY_T) |
873 | db -= device->ov_left; | |
58ffa580 | 874 | |
b411b363 | 875 | dbdt = Bit2KB(db/dt); |
b30ab791 | 876 | device->rs_paused /= HZ; |
b411b363 | 877 | |
b30ab791 | 878 | if (!get_ldev(device)) |
b411b363 PR |
879 | goto out; |
880 | ||
b30ab791 | 881 | ping_peer(device); |
af85e8e8 | 882 | |
0500813f | 883 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 884 | os = drbd_read_state(device); |
b411b363 | 885 | |
26525618 LE |
886 | verify_done = (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T); |
887 | ||
b411b363 PR |
888 | /* This protects us against multiple calls (that can happen in the presence |
889 | of application IO), and against connectivity loss just before we arrive here. */ | |
890 | if (os.conn <= C_CONNECTED) | |
891 | goto out_unlock; | |
892 | ||
893 | ns = os; | |
894 | ns.conn = C_CONNECTED; | |
895 | ||
d0180171 | 896 | drbd_info(device, "%s done (total %lu sec; paused %lu sec; %lu K/sec)\n", |
58ffa580 | 897 | verify_done ? "Online verify" : "Resync", |
b30ab791 | 898 | dt + device->rs_paused, device->rs_paused, dbdt); |
b411b363 | 899 | |
b30ab791 | 900 | n_oos = drbd_bm_total_weight(device); |
b411b363 PR |
901 | |
902 | if (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) { | |
903 | if (n_oos) { | |
d0180171 | 904 | drbd_alert(device, "Online verify found %lu %dk block out of sync!\n", |
b411b363 PR |
905 | n_oos, Bit2KB(1)); |
906 | khelper_cmd = "out-of-sync"; | |
907 | } | |
908 | } else { | |
0b0ba1ef | 909 | D_ASSERT(device, (n_oos - device->rs_failed) == 0); |
b411b363 PR |
910 | |
911 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) | |
912 | khelper_cmd = "after-resync-target"; | |
913 | ||
a6b32bc3 | 914 | if (first_peer_device(device)->connection->csums_tfm && device->rs_total) { |
b30ab791 AG |
915 | const unsigned long s = device->rs_same_csum; |
916 | const unsigned long t = device->rs_total; | |
b411b363 PR |
917 | const int ratio = |
918 | (t == 0) ? 0 : | |
919 | (t < 100000) ? ((s*100)/t) : (s/(t/100)); | |
d0180171 | 920 | drbd_info(device, "%u %% had equal checksums, eliminated: %luK; " |
b411b363 PR |
921 | "transferred %luK total %luK\n", |
922 | ratio, | |
b30ab791 AG |
923 | Bit2KB(device->rs_same_csum), |
924 | Bit2KB(device->rs_total - device->rs_same_csum), | |
925 | Bit2KB(device->rs_total)); | |
b411b363 PR |
926 | } |
927 | } | |
928 | ||
b30ab791 | 929 | if (device->rs_failed) { |
d0180171 | 930 | drbd_info(device, " %lu failed blocks\n", device->rs_failed); |
b411b363 PR |
931 | |
932 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
933 | ns.disk = D_INCONSISTENT; | |
934 | ns.pdsk = D_UP_TO_DATE; | |
935 | } else { | |
936 | ns.disk = D_UP_TO_DATE; | |
937 | ns.pdsk = D_INCONSISTENT; | |
938 | } | |
939 | } else { | |
940 | ns.disk = D_UP_TO_DATE; | |
941 | ns.pdsk = D_UP_TO_DATE; | |
942 | ||
943 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
b30ab791 | 944 | if (device->p_uuid) { |
b411b363 PR |
945 | int i; |
946 | for (i = UI_BITMAP ; i <= UI_HISTORY_END ; i++) | |
b30ab791 AG |
947 | _drbd_uuid_set(device, i, device->p_uuid[i]); |
948 | drbd_uuid_set(device, UI_BITMAP, device->ldev->md.uuid[UI_CURRENT]); | |
949 | _drbd_uuid_set(device, UI_CURRENT, device->p_uuid[UI_CURRENT]); | |
b411b363 | 950 | } else { |
d0180171 | 951 | drbd_err(device, "device->p_uuid is NULL! BUG\n"); |
b411b363 PR |
952 | } |
953 | } | |
954 | ||
62b0da3a LE |
955 | if (!(os.conn == C_VERIFY_S || os.conn == C_VERIFY_T)) { |
956 | /* for verify runs, we don't update uuids here, | |
957 | * so there would be nothing to report. */ | |
b30ab791 AG |
958 | drbd_uuid_set_bm(device, 0UL); |
959 | drbd_print_uuids(device, "updated UUIDs"); | |
960 | if (device->p_uuid) { | |
62b0da3a LE |
961 | /* Now the two UUID sets are equal, update what we |
962 | * know of the peer. */ | |
963 | int i; | |
964 | for (i = UI_CURRENT ; i <= UI_HISTORY_END ; i++) | |
b30ab791 | 965 | device->p_uuid[i] = device->ldev->md.uuid[i]; |
62b0da3a | 966 | } |
b411b363 PR |
967 | } |
968 | } | |
969 | ||
b30ab791 | 970 | _drbd_set_state(device, ns, CS_VERBOSE, NULL); |
b411b363 | 971 | out_unlock: |
0500813f | 972 | spin_unlock_irq(&device->resource->req_lock); |
b30ab791 | 973 | put_ldev(device); |
b411b363 | 974 | out: |
b30ab791 AG |
975 | device->rs_total = 0; |
976 | device->rs_failed = 0; | |
977 | device->rs_paused = 0; | |
58ffa580 LE |
978 | |
979 | /* reset start sector, if we reached end of device */ | |
b30ab791 AG |
980 | if (verify_done && device->ov_left == 0) |
981 | device->ov_start_sector = 0; | |
b411b363 | 982 | |
b30ab791 | 983 | drbd_md_sync(device); |
13d42685 | 984 | |
b411b363 | 985 | if (khelper_cmd) |
b30ab791 | 986 | drbd_khelper(device, khelper_cmd); |
b411b363 PR |
987 | |
988 | return 1; | |
989 | } | |
990 | ||
991 | /* helper */ | |
b30ab791 | 992 | static void move_to_net_ee_or_free(struct drbd_device *device, struct drbd_peer_request *peer_req) |
b411b363 | 993 | { |
045417f7 | 994 | if (drbd_peer_req_has_active_page(peer_req)) { |
b411b363 | 995 | /* This might happen if sendpage() has not finished */ |
db830c46 | 996 | int i = (peer_req->i.size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b30ab791 AG |
997 | atomic_add(i, &device->pp_in_use_by_net); |
998 | atomic_sub(i, &device->pp_in_use); | |
0500813f | 999 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 1000 | list_add_tail(&peer_req->w.list, &device->net_ee); |
0500813f | 1001 | spin_unlock_irq(&device->resource->req_lock); |
435f0740 | 1002 | wake_up(&drbd_pp_wait); |
b411b363 | 1003 | } else |
b30ab791 | 1004 | drbd_free_peer_req(device, peer_req); |
b411b363 PR |
1005 | } |
1006 | ||
1007 | /** | |
1008 | * w_e_end_data_req() - Worker callback, to send a P_DATA_REPLY packet in response to a P_DATA_REQUEST | |
b30ab791 | 1009 | * @device: DRBD device. |
b411b363 PR |
1010 | * @w: work object. |
1011 | * @cancel: The connection will be closed anyways | |
1012 | */ | |
99920dc5 | 1013 | int w_e_end_data_req(struct drbd_work *w, int cancel) |
b411b363 | 1014 | { |
a8cd15ba | 1015 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1016 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1017 | struct drbd_device *device = peer_device->device; | |
99920dc5 | 1018 | int err; |
b411b363 PR |
1019 | |
1020 | if (unlikely(cancel)) { | |
b30ab791 AG |
1021 | drbd_free_peer_req(device, peer_req); |
1022 | dec_unacked(device); | |
99920dc5 | 1023 | return 0; |
b411b363 PR |
1024 | } |
1025 | ||
db830c46 | 1026 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
6780139c | 1027 | err = drbd_send_block(peer_device, P_DATA_REPLY, peer_req); |
b411b363 PR |
1028 | } else { |
1029 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1030 | drbd_err(device, "Sending NegDReply. sector=%llus.\n", |
db830c46 | 1031 | (unsigned long long)peer_req->i.sector); |
b411b363 | 1032 | |
6780139c | 1033 | err = drbd_send_ack(peer_device, P_NEG_DREPLY, peer_req); |
b411b363 PR |
1034 | } |
1035 | ||
b30ab791 | 1036 | dec_unacked(device); |
b411b363 | 1037 | |
b30ab791 | 1038 | move_to_net_ee_or_free(device, peer_req); |
b411b363 | 1039 | |
99920dc5 | 1040 | if (unlikely(err)) |
d0180171 | 1041 | drbd_err(device, "drbd_send_block() failed\n"); |
99920dc5 | 1042 | return err; |
b411b363 PR |
1043 | } |
1044 | ||
1045 | /** | |
a209b4ae | 1046 | * 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 |
1047 | * @w: work object. |
1048 | * @cancel: The connection will be closed anyways | |
1049 | */ | |
99920dc5 | 1050 | int w_e_end_rsdata_req(struct drbd_work *w, int cancel) |
b411b363 | 1051 | { |
a8cd15ba | 1052 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1053 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1054 | struct drbd_device *device = peer_device->device; | |
99920dc5 | 1055 | int err; |
b411b363 PR |
1056 | |
1057 | if (unlikely(cancel)) { | |
b30ab791 AG |
1058 | drbd_free_peer_req(device, peer_req); |
1059 | dec_unacked(device); | |
99920dc5 | 1060 | return 0; |
b411b363 PR |
1061 | } |
1062 | ||
b30ab791 AG |
1063 | if (get_ldev_if_state(device, D_FAILED)) { |
1064 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1065 | put_ldev(device); | |
b411b363 PR |
1066 | } |
1067 | ||
b30ab791 | 1068 | if (device->state.conn == C_AHEAD) { |
6780139c | 1069 | err = drbd_send_ack(peer_device, P_RS_CANCEL, peer_req); |
db830c46 | 1070 | } else if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b30ab791 AG |
1071 | if (likely(device->state.pdsk >= D_INCONSISTENT)) { |
1072 | inc_rs_pending(device); | |
6780139c | 1073 | err = drbd_send_block(peer_device, P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1074 | } else { |
1075 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1076 | drbd_err(device, "Not sending RSDataReply, " |
b411b363 | 1077 | "partner DISKLESS!\n"); |
99920dc5 | 1078 | err = 0; |
b411b363 PR |
1079 | } |
1080 | } else { | |
1081 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1082 | drbd_err(device, "Sending NegRSDReply. sector %llus.\n", |
db830c46 | 1083 | (unsigned long long)peer_req->i.sector); |
b411b363 | 1084 | |
6780139c | 1085 | err = drbd_send_ack(peer_device, P_NEG_RS_DREPLY, peer_req); |
b411b363 PR |
1086 | |
1087 | /* update resync data with failure */ | |
b30ab791 | 1088 | drbd_rs_failed_io(device, peer_req->i.sector, peer_req->i.size); |
b411b363 PR |
1089 | } |
1090 | ||
b30ab791 | 1091 | dec_unacked(device); |
b411b363 | 1092 | |
b30ab791 | 1093 | move_to_net_ee_or_free(device, peer_req); |
b411b363 | 1094 | |
99920dc5 | 1095 | if (unlikely(err)) |
d0180171 | 1096 | drbd_err(device, "drbd_send_block() failed\n"); |
99920dc5 | 1097 | return err; |
b411b363 PR |
1098 | } |
1099 | ||
99920dc5 | 1100 | int w_e_end_csum_rs_req(struct drbd_work *w, int cancel) |
b411b363 | 1101 | { |
a8cd15ba | 1102 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1103 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1104 | struct drbd_device *device = peer_device->device; | |
b411b363 PR |
1105 | struct digest_info *di; |
1106 | int digest_size; | |
1107 | void *digest = NULL; | |
99920dc5 | 1108 | int err, eq = 0; |
b411b363 PR |
1109 | |
1110 | if (unlikely(cancel)) { | |
b30ab791 AG |
1111 | drbd_free_peer_req(device, peer_req); |
1112 | dec_unacked(device); | |
99920dc5 | 1113 | return 0; |
b411b363 PR |
1114 | } |
1115 | ||
b30ab791 AG |
1116 | if (get_ldev(device)) { |
1117 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1118 | put_ldev(device); | |
1d53f09e | 1119 | } |
b411b363 | 1120 | |
db830c46 | 1121 | di = peer_req->digest; |
b411b363 | 1122 | |
db830c46 | 1123 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1124 | /* quick hack to try to avoid a race against reconfiguration. |
1125 | * a real fix would be much more involved, | |
1126 | * introducing more locking mechanisms */ | |
6780139c AG |
1127 | if (peer_device->connection->csums_tfm) { |
1128 | digest_size = crypto_hash_digestsize(peer_device->connection->csums_tfm); | |
0b0ba1ef | 1129 | D_ASSERT(device, digest_size == di->digest_size); |
b411b363 PR |
1130 | digest = kmalloc(digest_size, GFP_NOIO); |
1131 | } | |
1132 | if (digest) { | |
6780139c | 1133 | drbd_csum_ee(peer_device->connection->csums_tfm, peer_req, digest); |
b411b363 PR |
1134 | eq = !memcmp(digest, di->digest, digest_size); |
1135 | kfree(digest); | |
1136 | } | |
1137 | ||
1138 | if (eq) { | |
b30ab791 | 1139 | drbd_set_in_sync(device, peer_req->i.sector, peer_req->i.size); |
676396d5 | 1140 | /* rs_same_csums unit is BM_BLOCK_SIZE */ |
b30ab791 | 1141 | device->rs_same_csum += peer_req->i.size >> BM_BLOCK_SHIFT; |
6780139c | 1142 | err = drbd_send_ack(peer_device, P_RS_IS_IN_SYNC, peer_req); |
b411b363 | 1143 | } else { |
b30ab791 | 1144 | inc_rs_pending(device); |
db830c46 AG |
1145 | peer_req->block_id = ID_SYNCER; /* By setting block_id, digest pointer becomes invalid! */ |
1146 | peer_req->flags &= ~EE_HAS_DIGEST; /* This peer request no longer has a digest pointer */ | |
204bba99 | 1147 | kfree(di); |
6780139c | 1148 | err = drbd_send_block(peer_device, P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1149 | } |
1150 | } else { | |
6780139c | 1151 | err = drbd_send_ack(peer_device, P_NEG_RS_DREPLY, peer_req); |
b411b363 | 1152 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 1153 | drbd_err(device, "Sending NegDReply. I guess it gets messy.\n"); |
b411b363 PR |
1154 | } |
1155 | ||
b30ab791 AG |
1156 | dec_unacked(device); |
1157 | move_to_net_ee_or_free(device, peer_req); | |
b411b363 | 1158 | |
99920dc5 | 1159 | if (unlikely(err)) |
d0180171 | 1160 | drbd_err(device, "drbd_send_block/ack() failed\n"); |
99920dc5 | 1161 | return err; |
b411b363 PR |
1162 | } |
1163 | ||
99920dc5 | 1164 | int w_e_end_ov_req(struct drbd_work *w, int cancel) |
b411b363 | 1165 | { |
a8cd15ba | 1166 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1167 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1168 | struct drbd_device *device = peer_device->device; | |
db830c46 AG |
1169 | sector_t sector = peer_req->i.sector; |
1170 | unsigned int size = peer_req->i.size; | |
b411b363 PR |
1171 | int digest_size; |
1172 | void *digest; | |
99920dc5 | 1173 | int err = 0; |
b411b363 PR |
1174 | |
1175 | if (unlikely(cancel)) | |
1176 | goto out; | |
1177 | ||
6780139c | 1178 | digest_size = crypto_hash_digestsize(peer_device->connection->verify_tfm); |
b411b363 | 1179 | digest = kmalloc(digest_size, GFP_NOIO); |
8f21420e | 1180 | if (!digest) { |
99920dc5 | 1181 | err = 1; /* terminate the connection in case the allocation failed */ |
8f21420e | 1182 | goto out; |
b411b363 PR |
1183 | } |
1184 | ||
db830c46 | 1185 | if (likely(!(peer_req->flags & EE_WAS_ERROR))) |
6780139c | 1186 | drbd_csum_ee(peer_device->connection->verify_tfm, peer_req, digest); |
8f21420e PR |
1187 | else |
1188 | memset(digest, 0, digest_size); | |
1189 | ||
53ea4331 LE |
1190 | /* Free e and pages before send. |
1191 | * In case we block on congestion, we could otherwise run into | |
1192 | * some distributed deadlock, if the other side blocks on | |
1193 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1194 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 1195 | drbd_free_peer_req(device, peer_req); |
db830c46 | 1196 | peer_req = NULL; |
b30ab791 | 1197 | inc_rs_pending(device); |
6780139c | 1198 | err = drbd_send_drequest_csum(peer_device, sector, size, digest, digest_size, P_OV_REPLY); |
99920dc5 | 1199 | if (err) |
b30ab791 | 1200 | dec_rs_pending(device); |
8f21420e PR |
1201 | kfree(digest); |
1202 | ||
b411b363 | 1203 | out: |
db830c46 | 1204 | if (peer_req) |
b30ab791 AG |
1205 | drbd_free_peer_req(device, peer_req); |
1206 | dec_unacked(device); | |
99920dc5 | 1207 | return err; |
b411b363 PR |
1208 | } |
1209 | ||
b30ab791 | 1210 | void drbd_ov_out_of_sync_found(struct drbd_device *device, sector_t sector, int size) |
b411b363 | 1211 | { |
b30ab791 AG |
1212 | if (device->ov_last_oos_start + device->ov_last_oos_size == sector) { |
1213 | device->ov_last_oos_size += size>>9; | |
b411b363 | 1214 | } else { |
b30ab791 AG |
1215 | device->ov_last_oos_start = sector; |
1216 | device->ov_last_oos_size = size>>9; | |
b411b363 | 1217 | } |
b30ab791 | 1218 | drbd_set_out_of_sync(device, sector, size); |
b411b363 PR |
1219 | } |
1220 | ||
99920dc5 | 1221 | int w_e_end_ov_reply(struct drbd_work *w, int cancel) |
b411b363 | 1222 | { |
a8cd15ba | 1223 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1224 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1225 | struct drbd_device *device = peer_device->device; | |
b411b363 | 1226 | struct digest_info *di; |
b411b363 | 1227 | void *digest; |
db830c46 AG |
1228 | sector_t sector = peer_req->i.sector; |
1229 | unsigned int size = peer_req->i.size; | |
53ea4331 | 1230 | int digest_size; |
99920dc5 | 1231 | int err, eq = 0; |
58ffa580 | 1232 | bool stop_sector_reached = false; |
b411b363 PR |
1233 | |
1234 | if (unlikely(cancel)) { | |
b30ab791 AG |
1235 | drbd_free_peer_req(device, peer_req); |
1236 | dec_unacked(device); | |
99920dc5 | 1237 | return 0; |
b411b363 PR |
1238 | } |
1239 | ||
1240 | /* after "cancel", because after drbd_disconnect/drbd_rs_cancel_all | |
1241 | * the resync lru has been cleaned up already */ | |
b30ab791 AG |
1242 | if (get_ldev(device)) { |
1243 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1244 | put_ldev(device); | |
1d53f09e | 1245 | } |
b411b363 | 1246 | |
db830c46 | 1247 | di = peer_req->digest; |
b411b363 | 1248 | |
db830c46 | 1249 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
6780139c | 1250 | digest_size = crypto_hash_digestsize(peer_device->connection->verify_tfm); |
b411b363 PR |
1251 | digest = kmalloc(digest_size, GFP_NOIO); |
1252 | if (digest) { | |
6780139c | 1253 | drbd_csum_ee(peer_device->connection->verify_tfm, peer_req, digest); |
b411b363 | 1254 | |
0b0ba1ef | 1255 | D_ASSERT(device, digest_size == di->digest_size); |
b411b363 PR |
1256 | eq = !memcmp(digest, di->digest, digest_size); |
1257 | kfree(digest); | |
1258 | } | |
b411b363 PR |
1259 | } |
1260 | ||
9676c760 LE |
1261 | /* Free peer_req and pages before send. |
1262 | * In case we block on congestion, we could otherwise run into | |
1263 | * some distributed deadlock, if the other side blocks on | |
1264 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1265 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 1266 | drbd_free_peer_req(device, peer_req); |
b411b363 | 1267 | if (!eq) |
b30ab791 | 1268 | drbd_ov_out_of_sync_found(device, sector, size); |
b411b363 | 1269 | else |
b30ab791 | 1270 | ov_out_of_sync_print(device); |
b411b363 | 1271 | |
6780139c | 1272 | err = drbd_send_ack_ex(peer_device, P_OV_RESULT, sector, size, |
fa79abd8 | 1273 | eq ? ID_IN_SYNC : ID_OUT_OF_SYNC); |
b411b363 | 1274 | |
b30ab791 | 1275 | dec_unacked(device); |
b411b363 | 1276 | |
b30ab791 | 1277 | --device->ov_left; |
ea5442af LE |
1278 | |
1279 | /* let's advance progress step marks only for every other megabyte */ | |
b30ab791 AG |
1280 | if ((device->ov_left & 0x200) == 0x200) |
1281 | drbd_advance_rs_marks(device, device->ov_left); | |
ea5442af | 1282 | |
b30ab791 AG |
1283 | stop_sector_reached = verify_can_do_stop_sector(device) && |
1284 | (sector + (size>>9)) >= device->ov_stop_sector; | |
58ffa580 | 1285 | |
b30ab791 AG |
1286 | if (device->ov_left == 0 || stop_sector_reached) { |
1287 | ov_out_of_sync_print(device); | |
1288 | drbd_resync_finished(device); | |
b411b363 PR |
1289 | } |
1290 | ||
99920dc5 | 1291 | return err; |
b411b363 PR |
1292 | } |
1293 | ||
b6dd1a89 LE |
1294 | /* FIXME |
1295 | * We need to track the number of pending barrier acks, | |
1296 | * and to be able to wait for them. | |
1297 | * See also comment in drbd_adm_attach before drbd_suspend_io. | |
1298 | */ | |
bde89a9e | 1299 | static int drbd_send_barrier(struct drbd_connection *connection) |
b411b363 | 1300 | { |
9f5bdc33 | 1301 | struct p_barrier *p; |
b6dd1a89 | 1302 | struct drbd_socket *sock; |
b411b363 | 1303 | |
bde89a9e AG |
1304 | sock = &connection->data; |
1305 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1306 | if (!p) |
1307 | return -EIO; | |
bde89a9e | 1308 | p->barrier = connection->send.current_epoch_nr; |
b6dd1a89 | 1309 | p->pad = 0; |
bde89a9e | 1310 | connection->send.current_epoch_writes = 0; |
b6dd1a89 | 1311 | |
bde89a9e | 1312 | return conn_send_command(connection, sock, P_BARRIER, sizeof(*p), NULL, 0); |
b411b363 PR |
1313 | } |
1314 | ||
99920dc5 | 1315 | int w_send_write_hint(struct drbd_work *w, int cancel) |
b411b363 | 1316 | { |
84b8c06b AG |
1317 | struct drbd_device *device = |
1318 | container_of(w, struct drbd_device, unplug_work); | |
9f5bdc33 AG |
1319 | struct drbd_socket *sock; |
1320 | ||
b411b363 | 1321 | if (cancel) |
99920dc5 | 1322 | return 0; |
a6b32bc3 | 1323 | sock = &first_peer_device(device)->connection->data; |
69a22773 | 1324 | if (!drbd_prepare_command(first_peer_device(device), sock)) |
9f5bdc33 | 1325 | return -EIO; |
69a22773 | 1326 | return drbd_send_command(first_peer_device(device), sock, P_UNPLUG_REMOTE, 0, NULL, 0); |
b411b363 PR |
1327 | } |
1328 | ||
bde89a9e | 1329 | static void re_init_if_first_write(struct drbd_connection *connection, unsigned int epoch) |
4eb9b3cb | 1330 | { |
bde89a9e AG |
1331 | if (!connection->send.seen_any_write_yet) { |
1332 | connection->send.seen_any_write_yet = true; | |
1333 | connection->send.current_epoch_nr = epoch; | |
1334 | connection->send.current_epoch_writes = 0; | |
4eb9b3cb LE |
1335 | } |
1336 | } | |
1337 | ||
bde89a9e | 1338 | static void maybe_send_barrier(struct drbd_connection *connection, unsigned int epoch) |
4eb9b3cb LE |
1339 | { |
1340 | /* re-init if first write on this connection */ | |
bde89a9e | 1341 | if (!connection->send.seen_any_write_yet) |
4eb9b3cb | 1342 | return; |
bde89a9e AG |
1343 | if (connection->send.current_epoch_nr != epoch) { |
1344 | if (connection->send.current_epoch_writes) | |
1345 | drbd_send_barrier(connection); | |
1346 | connection->send.current_epoch_nr = epoch; | |
4eb9b3cb LE |
1347 | } |
1348 | } | |
1349 | ||
8f7bed77 | 1350 | int w_send_out_of_sync(struct drbd_work *w, int cancel) |
73a01a18 PR |
1351 | { |
1352 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1353 | struct drbd_device *device = req->device; |
a6b32bc3 | 1354 | struct drbd_connection *connection = first_peer_device(device)->connection; |
99920dc5 | 1355 | int err; |
73a01a18 PR |
1356 | |
1357 | if (unlikely(cancel)) { | |
8554df1c | 1358 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1359 | return 0; |
73a01a18 PR |
1360 | } |
1361 | ||
bde89a9e | 1362 | /* this time, no connection->send.current_epoch_writes++; |
b6dd1a89 LE |
1363 | * If it was sent, it was the closing barrier for the last |
1364 | * replicated epoch, before we went into AHEAD mode. | |
1365 | * No more barriers will be sent, until we leave AHEAD mode again. */ | |
bde89a9e | 1366 | maybe_send_barrier(connection, req->epoch); |
b6dd1a89 | 1367 | |
69a22773 | 1368 | err = drbd_send_out_of_sync(first_peer_device(device), req); |
8554df1c | 1369 | req_mod(req, OOS_HANDED_TO_NETWORK); |
73a01a18 | 1370 | |
99920dc5 | 1371 | return err; |
73a01a18 PR |
1372 | } |
1373 | ||
b411b363 PR |
1374 | /** |
1375 | * w_send_dblock() - Worker callback to send a P_DATA packet in order to mirror a write request | |
b411b363 PR |
1376 | * @w: work object. |
1377 | * @cancel: The connection will be closed anyways | |
1378 | */ | |
99920dc5 | 1379 | int w_send_dblock(struct drbd_work *w, int cancel) |
b411b363 PR |
1380 | { |
1381 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1382 | struct drbd_device *device = req->device; |
a6b32bc3 | 1383 | struct drbd_connection *connection = first_peer_device(device)->connection; |
99920dc5 | 1384 | int err; |
b411b363 PR |
1385 | |
1386 | if (unlikely(cancel)) { | |
8554df1c | 1387 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1388 | return 0; |
b411b363 PR |
1389 | } |
1390 | ||
bde89a9e AG |
1391 | re_init_if_first_write(connection, req->epoch); |
1392 | maybe_send_barrier(connection, req->epoch); | |
1393 | connection->send.current_epoch_writes++; | |
b6dd1a89 | 1394 | |
69a22773 | 1395 | err = drbd_send_dblock(first_peer_device(device), req); |
99920dc5 | 1396 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1397 | |
99920dc5 | 1398 | return err; |
b411b363 PR |
1399 | } |
1400 | ||
1401 | /** | |
1402 | * w_send_read_req() - Worker callback to send a read request (P_DATA_REQUEST) packet | |
b411b363 PR |
1403 | * @w: work object. |
1404 | * @cancel: The connection will be closed anyways | |
1405 | */ | |
99920dc5 | 1406 | int w_send_read_req(struct drbd_work *w, int cancel) |
b411b363 PR |
1407 | { |
1408 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1409 | struct drbd_device *device = req->device; |
a6b32bc3 | 1410 | struct drbd_connection *connection = first_peer_device(device)->connection; |
99920dc5 | 1411 | int err; |
b411b363 PR |
1412 | |
1413 | if (unlikely(cancel)) { | |
8554df1c | 1414 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1415 | return 0; |
b411b363 PR |
1416 | } |
1417 | ||
b6dd1a89 LE |
1418 | /* Even read requests may close a write epoch, |
1419 | * if there was any yet. */ | |
bde89a9e | 1420 | maybe_send_barrier(connection, req->epoch); |
b6dd1a89 | 1421 | |
69a22773 | 1422 | err = drbd_send_drequest(first_peer_device(device), P_DATA_REQUEST, req->i.sector, req->i.size, |
6c1005e7 | 1423 | (unsigned long)req); |
b411b363 | 1424 | |
99920dc5 | 1425 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1426 | |
99920dc5 | 1427 | return err; |
b411b363 PR |
1428 | } |
1429 | ||
99920dc5 | 1430 | int w_restart_disk_io(struct drbd_work *w, int cancel) |
265be2d0 PR |
1431 | { |
1432 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1433 | struct drbd_device *device = req->device; |
265be2d0 | 1434 | |
0778286a | 1435 | if (bio_data_dir(req->master_bio) == WRITE && req->rq_state & RQ_IN_ACT_LOG) |
b30ab791 | 1436 | drbd_al_begin_io(device, &req->i, false); |
265be2d0 PR |
1437 | |
1438 | drbd_req_make_private_bio(req, req->master_bio); | |
b30ab791 | 1439 | req->private_bio->bi_bdev = device->ldev->backing_bdev; |
265be2d0 PR |
1440 | generic_make_request(req->private_bio); |
1441 | ||
99920dc5 | 1442 | return 0; |
265be2d0 PR |
1443 | } |
1444 | ||
b30ab791 | 1445 | static int _drbd_may_sync_now(struct drbd_device *device) |
b411b363 | 1446 | { |
b30ab791 | 1447 | struct drbd_device *odev = device; |
95f8efd0 | 1448 | int resync_after; |
b411b363 PR |
1449 | |
1450 | while (1) { | |
a3f8f7dc | 1451 | if (!odev->ldev || odev->state.disk == D_DISKLESS) |
438c8374 | 1452 | return 1; |
daeda1cc | 1453 | rcu_read_lock(); |
95f8efd0 | 1454 | resync_after = rcu_dereference(odev->ldev->disk_conf)->resync_after; |
daeda1cc | 1455 | rcu_read_unlock(); |
95f8efd0 | 1456 | if (resync_after == -1) |
b411b363 | 1457 | return 1; |
b30ab791 | 1458 | odev = minor_to_device(resync_after); |
a3f8f7dc | 1459 | if (!odev) |
841ce241 | 1460 | return 1; |
b411b363 PR |
1461 | if ((odev->state.conn >= C_SYNC_SOURCE && |
1462 | odev->state.conn <= C_PAUSED_SYNC_T) || | |
1463 | odev->state.aftr_isp || odev->state.peer_isp || | |
1464 | odev->state.user_isp) | |
1465 | return 0; | |
1466 | } | |
1467 | } | |
1468 | ||
1469 | /** | |
1470 | * _drbd_pause_after() - Pause resync on all devices that may not resync now | |
b30ab791 | 1471 | * @device: DRBD device. |
b411b363 PR |
1472 | * |
1473 | * Called from process context only (admin command and after_state_ch). | |
1474 | */ | |
b30ab791 | 1475 | static int _drbd_pause_after(struct drbd_device *device) |
b411b363 | 1476 | { |
54761697 | 1477 | struct drbd_device *odev; |
b411b363 PR |
1478 | int i, rv = 0; |
1479 | ||
695d08fa | 1480 | rcu_read_lock(); |
05a10ec7 | 1481 | idr_for_each_entry(&drbd_devices, odev, i) { |
b411b363 PR |
1482 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1483 | continue; | |
1484 | if (!_drbd_may_sync_now(odev)) | |
1485 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 1), CS_HARD, NULL) | |
1486 | != SS_NOTHING_TO_DO); | |
1487 | } | |
695d08fa | 1488 | rcu_read_unlock(); |
b411b363 PR |
1489 | |
1490 | return rv; | |
1491 | } | |
1492 | ||
1493 | /** | |
1494 | * _drbd_resume_next() - Resume resync on all devices that may resync now | |
b30ab791 | 1495 | * @device: DRBD device. |
b411b363 PR |
1496 | * |
1497 | * Called from process context only (admin command and worker). | |
1498 | */ | |
b30ab791 | 1499 | static int _drbd_resume_next(struct drbd_device *device) |
b411b363 | 1500 | { |
54761697 | 1501 | struct drbd_device *odev; |
b411b363 PR |
1502 | int i, rv = 0; |
1503 | ||
695d08fa | 1504 | rcu_read_lock(); |
05a10ec7 | 1505 | idr_for_each_entry(&drbd_devices, odev, i) { |
b411b363 PR |
1506 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1507 | continue; | |
1508 | if (odev->state.aftr_isp) { | |
1509 | if (_drbd_may_sync_now(odev)) | |
1510 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 0), | |
1511 | CS_HARD, NULL) | |
1512 | != SS_NOTHING_TO_DO) ; | |
1513 | } | |
1514 | } | |
695d08fa | 1515 | rcu_read_unlock(); |
b411b363 PR |
1516 | return rv; |
1517 | } | |
1518 | ||
b30ab791 | 1519 | void resume_next_sg(struct drbd_device *device) |
b411b363 PR |
1520 | { |
1521 | write_lock_irq(&global_state_lock); | |
b30ab791 | 1522 | _drbd_resume_next(device); |
b411b363 PR |
1523 | write_unlock_irq(&global_state_lock); |
1524 | } | |
1525 | ||
b30ab791 | 1526 | void suspend_other_sg(struct drbd_device *device) |
b411b363 PR |
1527 | { |
1528 | write_lock_irq(&global_state_lock); | |
b30ab791 | 1529 | _drbd_pause_after(device); |
b411b363 PR |
1530 | write_unlock_irq(&global_state_lock); |
1531 | } | |
1532 | ||
dc97b708 | 1533 | /* caller must hold global_state_lock */ |
b30ab791 | 1534 | enum drbd_ret_code drbd_resync_after_valid(struct drbd_device *device, int o_minor) |
b411b363 | 1535 | { |
54761697 | 1536 | struct drbd_device *odev; |
95f8efd0 | 1537 | int resync_after; |
b411b363 PR |
1538 | |
1539 | if (o_minor == -1) | |
1540 | return NO_ERROR; | |
a3f8f7dc | 1541 | if (o_minor < -1 || o_minor > MINORMASK) |
95f8efd0 | 1542 | return ERR_RESYNC_AFTER; |
b411b363 PR |
1543 | |
1544 | /* check for loops */ | |
b30ab791 | 1545 | odev = minor_to_device(o_minor); |
b411b363 | 1546 | while (1) { |
b30ab791 | 1547 | if (odev == device) |
95f8efd0 | 1548 | return ERR_RESYNC_AFTER_CYCLE; |
b411b363 | 1549 | |
a3f8f7dc LE |
1550 | /* You are free to depend on diskless, non-existing, |
1551 | * or not yet/no longer existing minors. | |
1552 | * We only reject dependency loops. | |
1553 | * We cannot follow the dependency chain beyond a detached or | |
1554 | * missing minor. | |
1555 | */ | |
1556 | if (!odev || !odev->ldev || odev->state.disk == D_DISKLESS) | |
1557 | return NO_ERROR; | |
1558 | ||
daeda1cc | 1559 | rcu_read_lock(); |
95f8efd0 | 1560 | resync_after = rcu_dereference(odev->ldev->disk_conf)->resync_after; |
daeda1cc | 1561 | rcu_read_unlock(); |
b411b363 | 1562 | /* dependency chain ends here, no cycles. */ |
95f8efd0 | 1563 | if (resync_after == -1) |
b411b363 PR |
1564 | return NO_ERROR; |
1565 | ||
1566 | /* follow the dependency chain */ | |
b30ab791 | 1567 | odev = minor_to_device(resync_after); |
b411b363 PR |
1568 | } |
1569 | } | |
1570 | ||
dc97b708 | 1571 | /* caller must hold global_state_lock */ |
b30ab791 | 1572 | void drbd_resync_after_changed(struct drbd_device *device) |
b411b363 PR |
1573 | { |
1574 | int changes; | |
b411b363 | 1575 | |
dc97b708 | 1576 | do { |
b30ab791 AG |
1577 | changes = _drbd_pause_after(device); |
1578 | changes |= _drbd_resume_next(device); | |
dc97b708 | 1579 | } while (changes); |
b411b363 PR |
1580 | } |
1581 | ||
b30ab791 | 1582 | void drbd_rs_controller_reset(struct drbd_device *device) |
9bd28d3c | 1583 | { |
813472ce PR |
1584 | struct fifo_buffer *plan; |
1585 | ||
b30ab791 AG |
1586 | atomic_set(&device->rs_sect_in, 0); |
1587 | atomic_set(&device->rs_sect_ev, 0); | |
1588 | device->rs_in_flight = 0; | |
813472ce PR |
1589 | |
1590 | /* Updating the RCU protected object in place is necessary since | |
1591 | this function gets called from atomic context. | |
1592 | It is valid since all other updates also lead to an completely | |
1593 | empty fifo */ | |
1594 | rcu_read_lock(); | |
b30ab791 | 1595 | plan = rcu_dereference(device->rs_plan_s); |
813472ce PR |
1596 | plan->total = 0; |
1597 | fifo_set(plan, 0); | |
1598 | rcu_read_unlock(); | |
9bd28d3c LE |
1599 | } |
1600 | ||
1f04af33 PR |
1601 | void start_resync_timer_fn(unsigned long data) |
1602 | { | |
b30ab791 | 1603 | struct drbd_device *device = (struct drbd_device *) data; |
1f04af33 | 1604 | |
84b8c06b AG |
1605 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, |
1606 | &device->start_resync_work); | |
1f04af33 PR |
1607 | } |
1608 | ||
99920dc5 | 1609 | int w_start_resync(struct drbd_work *w, int cancel) |
1f04af33 | 1610 | { |
84b8c06b AG |
1611 | struct drbd_device *device = |
1612 | container_of(w, struct drbd_device, start_resync_work); | |
00d56944 | 1613 | |
b30ab791 | 1614 | if (atomic_read(&device->unacked_cnt) || atomic_read(&device->rs_pending_cnt)) { |
d0180171 | 1615 | drbd_warn(device, "w_start_resync later...\n"); |
b30ab791 AG |
1616 | device->start_resync_timer.expires = jiffies + HZ/10; |
1617 | add_timer(&device->start_resync_timer); | |
99920dc5 | 1618 | return 0; |
1f04af33 PR |
1619 | } |
1620 | ||
b30ab791 AG |
1621 | drbd_start_resync(device, C_SYNC_SOURCE); |
1622 | clear_bit(AHEAD_TO_SYNC_SOURCE, &device->flags); | |
99920dc5 | 1623 | return 0; |
1f04af33 PR |
1624 | } |
1625 | ||
b411b363 PR |
1626 | /** |
1627 | * drbd_start_resync() - Start the resync process | |
b30ab791 | 1628 | * @device: DRBD device. |
b411b363 PR |
1629 | * @side: Either C_SYNC_SOURCE or C_SYNC_TARGET |
1630 | * | |
1631 | * This function might bring you directly into one of the | |
1632 | * C_PAUSED_SYNC_* states. | |
1633 | */ | |
b30ab791 | 1634 | void drbd_start_resync(struct drbd_device *device, enum drbd_conns side) |
b411b363 PR |
1635 | { |
1636 | union drbd_state ns; | |
1637 | int r; | |
1638 | ||
b30ab791 | 1639 | if (device->state.conn >= C_SYNC_SOURCE && device->state.conn < C_AHEAD) { |
d0180171 | 1640 | drbd_err(device, "Resync already running!\n"); |
b411b363 PR |
1641 | return; |
1642 | } | |
1643 | ||
b30ab791 | 1644 | if (!test_bit(B_RS_H_DONE, &device->flags)) { |
e64a3294 PR |
1645 | if (side == C_SYNC_TARGET) { |
1646 | /* Since application IO was locked out during C_WF_BITMAP_T and | |
1647 | C_WF_SYNC_UUID we are still unmodified. Before going to C_SYNC_TARGET | |
1648 | we check that we might make the data inconsistent. */ | |
b30ab791 | 1649 | r = drbd_khelper(device, "before-resync-target"); |
e64a3294 PR |
1650 | r = (r >> 8) & 0xff; |
1651 | if (r > 0) { | |
d0180171 | 1652 | drbd_info(device, "before-resync-target handler returned %d, " |
09b9e797 | 1653 | "dropping connection.\n", r); |
a6b32bc3 | 1654 | conn_request_state(first_peer_device(device)->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
09b9e797 PR |
1655 | return; |
1656 | } | |
e64a3294 | 1657 | } else /* C_SYNC_SOURCE */ { |
b30ab791 | 1658 | r = drbd_khelper(device, "before-resync-source"); |
e64a3294 PR |
1659 | r = (r >> 8) & 0xff; |
1660 | if (r > 0) { | |
1661 | if (r == 3) { | |
d0180171 | 1662 | drbd_info(device, "before-resync-source handler returned %d, " |
e64a3294 PR |
1663 | "ignoring. Old userland tools?", r); |
1664 | } else { | |
d0180171 | 1665 | drbd_info(device, "before-resync-source handler returned %d, " |
e64a3294 | 1666 | "dropping connection.\n", r); |
a6b32bc3 AG |
1667 | conn_request_state(first_peer_device(device)->connection, |
1668 | NS(conn, C_DISCONNECTING), CS_HARD); | |
e64a3294 PR |
1669 | return; |
1670 | } | |
1671 | } | |
09b9e797 | 1672 | } |
b411b363 PR |
1673 | } |
1674 | ||
a6b32bc3 | 1675 | if (current == first_peer_device(device)->connection->worker.task) { |
dad20554 | 1676 | /* The worker should not sleep waiting for state_mutex, |
e64a3294 | 1677 | that can take long */ |
b30ab791 AG |
1678 | if (!mutex_trylock(device->state_mutex)) { |
1679 | set_bit(B_RS_H_DONE, &device->flags); | |
1680 | device->start_resync_timer.expires = jiffies + HZ/5; | |
1681 | add_timer(&device->start_resync_timer); | |
e64a3294 PR |
1682 | return; |
1683 | } | |
1684 | } else { | |
b30ab791 | 1685 | mutex_lock(device->state_mutex); |
e64a3294 | 1686 | } |
b30ab791 | 1687 | clear_bit(B_RS_H_DONE, &device->flags); |
b411b363 | 1688 | |
074f4afe LE |
1689 | /* req_lock: serialize with drbd_send_and_submit() and others |
1690 | * global_state_lock: for stable sync-after dependencies */ | |
1691 | spin_lock_irq(&device->resource->req_lock); | |
1692 | write_lock(&global_state_lock); | |
a700471b | 1693 | /* Did some connection breakage or IO error race with us? */ |
b30ab791 AG |
1694 | if (device->state.conn < C_CONNECTED |
1695 | || !get_ldev_if_state(device, D_NEGOTIATING)) { | |
074f4afe LE |
1696 | write_unlock(&global_state_lock); |
1697 | spin_unlock_irq(&device->resource->req_lock); | |
b30ab791 | 1698 | mutex_unlock(device->state_mutex); |
b411b363 PR |
1699 | return; |
1700 | } | |
1701 | ||
b30ab791 | 1702 | ns = drbd_read_state(device); |
b411b363 | 1703 | |
b30ab791 | 1704 | ns.aftr_isp = !_drbd_may_sync_now(device); |
b411b363 PR |
1705 | |
1706 | ns.conn = side; | |
1707 | ||
1708 | if (side == C_SYNC_TARGET) | |
1709 | ns.disk = D_INCONSISTENT; | |
1710 | else /* side == C_SYNC_SOURCE */ | |
1711 | ns.pdsk = D_INCONSISTENT; | |
1712 | ||
b30ab791 AG |
1713 | r = __drbd_set_state(device, ns, CS_VERBOSE, NULL); |
1714 | ns = drbd_read_state(device); | |
b411b363 PR |
1715 | |
1716 | if (ns.conn < C_CONNECTED) | |
1717 | r = SS_UNKNOWN_ERROR; | |
1718 | ||
1719 | if (r == SS_SUCCESS) { | |
b30ab791 | 1720 | unsigned long tw = drbd_bm_total_weight(device); |
1d7734a0 LE |
1721 | unsigned long now = jiffies; |
1722 | int i; | |
1723 | ||
b30ab791 AG |
1724 | device->rs_failed = 0; |
1725 | device->rs_paused = 0; | |
1726 | device->rs_same_csum = 0; | |
1727 | device->rs_last_events = 0; | |
1728 | device->rs_last_sect_ev = 0; | |
1729 | device->rs_total = tw; | |
1730 | device->rs_start = now; | |
1d7734a0 | 1731 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
1732 | device->rs_mark_left[i] = tw; |
1733 | device->rs_mark_time[i] = now; | |
1d7734a0 | 1734 | } |
b30ab791 | 1735 | _drbd_pause_after(device); |
b411b363 | 1736 | } |
074f4afe LE |
1737 | write_unlock(&global_state_lock); |
1738 | spin_unlock_irq(&device->resource->req_lock); | |
5a22db89 | 1739 | |
b411b363 | 1740 | if (r == SS_SUCCESS) { |
328e0f12 PR |
1741 | /* reset rs_last_bcast when a resync or verify is started, |
1742 | * to deal with potential jiffies wrap. */ | |
b30ab791 | 1743 | device->rs_last_bcast = jiffies - HZ; |
328e0f12 | 1744 | |
d0180171 | 1745 | drbd_info(device, "Began resync as %s (will sync %lu KB [%lu bits set]).\n", |
b411b363 | 1746 | drbd_conn_str(ns.conn), |
b30ab791 AG |
1747 | (unsigned long) device->rs_total << (BM_BLOCK_SHIFT-10), |
1748 | (unsigned long) device->rs_total); | |
6c922ed5 | 1749 | if (side == C_SYNC_TARGET) |
b30ab791 | 1750 | device->bm_resync_fo = 0; |
6c922ed5 LE |
1751 | |
1752 | /* Since protocol 96, we must serialize drbd_gen_and_send_sync_uuid | |
1753 | * with w_send_oos, or the sync target will get confused as to | |
1754 | * how much bits to resync. We cannot do that always, because for an | |
1755 | * empty resync and protocol < 95, we need to do it here, as we call | |
1756 | * drbd_resync_finished from here in that case. | |
1757 | * We drbd_gen_and_send_sync_uuid here for protocol < 96, | |
1758 | * and from after_state_ch otherwise. */ | |
a6b32bc3 AG |
1759 | if (side == C_SYNC_SOURCE && |
1760 | first_peer_device(device)->connection->agreed_pro_version < 96) | |
69a22773 | 1761 | drbd_gen_and_send_sync_uuid(first_peer_device(device)); |
b411b363 | 1762 | |
a6b32bc3 AG |
1763 | if (first_peer_device(device)->connection->agreed_pro_version < 95 && |
1764 | device->rs_total == 0) { | |
af85e8e8 LE |
1765 | /* This still has a race (about when exactly the peers |
1766 | * detect connection loss) that can lead to a full sync | |
1767 | * on next handshake. In 8.3.9 we fixed this with explicit | |
1768 | * resync-finished notifications, but the fix | |
1769 | * introduces a protocol change. Sleeping for some | |
1770 | * time longer than the ping interval + timeout on the | |
1771 | * SyncSource, to give the SyncTarget the chance to | |
1772 | * detect connection loss, then waiting for a ping | |
1773 | * response (implicit in drbd_resync_finished) reduces | |
1774 | * the race considerably, but does not solve it. */ | |
44ed167d PR |
1775 | if (side == C_SYNC_SOURCE) { |
1776 | struct net_conf *nc; | |
1777 | int timeo; | |
1778 | ||
1779 | rcu_read_lock(); | |
a6b32bc3 | 1780 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); |
44ed167d PR |
1781 | timeo = nc->ping_int * HZ + nc->ping_timeo * HZ / 9; |
1782 | rcu_read_unlock(); | |
1783 | schedule_timeout_interruptible(timeo); | |
1784 | } | |
b30ab791 | 1785 | drbd_resync_finished(device); |
b411b363 PR |
1786 | } |
1787 | ||
b30ab791 AG |
1788 | drbd_rs_controller_reset(device); |
1789 | /* ns.conn may already be != device->state.conn, | |
b411b363 PR |
1790 | * we may have been paused in between, or become paused until |
1791 | * the timer triggers. | |
1792 | * No matter, that is handled in resync_timer_fn() */ | |
1793 | if (ns.conn == C_SYNC_TARGET) | |
b30ab791 | 1794 | mod_timer(&device->resync_timer, jiffies); |
b411b363 | 1795 | |
b30ab791 | 1796 | drbd_md_sync(device); |
b411b363 | 1797 | } |
b30ab791 AG |
1798 | put_ldev(device); |
1799 | mutex_unlock(device->state_mutex); | |
b411b363 PR |
1800 | } |
1801 | ||
a186e478 | 1802 | static bool dequeue_work_batch(struct drbd_work_queue *queue, struct list_head *work_list) |
8c0785a5 LE |
1803 | { |
1804 | spin_lock_irq(&queue->q_lock); | |
1805 | list_splice_init(&queue->q, work_list); | |
1806 | spin_unlock_irq(&queue->q_lock); | |
1807 | return !list_empty(work_list); | |
1808 | } | |
1809 | ||
a186e478 | 1810 | static bool dequeue_work_item(struct drbd_work_queue *queue, struct list_head *work_list) |
8c0785a5 LE |
1811 | { |
1812 | spin_lock_irq(&queue->q_lock); | |
1813 | if (!list_empty(&queue->q)) | |
1814 | list_move(queue->q.next, work_list); | |
1815 | spin_unlock_irq(&queue->q_lock); | |
1816 | return !list_empty(work_list); | |
1817 | } | |
1818 | ||
bde89a9e | 1819 | static void wait_for_work(struct drbd_connection *connection, struct list_head *work_list) |
b6dd1a89 LE |
1820 | { |
1821 | DEFINE_WAIT(wait); | |
1822 | struct net_conf *nc; | |
1823 | int uncork, cork; | |
1824 | ||
1825 | dequeue_work_item(&connection->sender_work, work_list); | |
1826 | if (!list_empty(work_list)) | |
1827 | return; | |
1828 | ||
1829 | /* Still nothing to do? | |
1830 | * Maybe we still need to close the current epoch, | |
1831 | * even if no new requests are queued yet. | |
1832 | * | |
1833 | * Also, poke TCP, just in case. | |
1834 | * Then wait for new work (or signal). */ | |
1835 | rcu_read_lock(); | |
1836 | nc = rcu_dereference(connection->net_conf); | |
1837 | uncork = nc ? nc->tcp_cork : 0; | |
1838 | rcu_read_unlock(); | |
1839 | if (uncork) { | |
1840 | mutex_lock(&connection->data.mutex); | |
1841 | if (connection->data.socket) | |
1842 | drbd_tcp_uncork(connection->data.socket); | |
1843 | mutex_unlock(&connection->data.mutex); | |
1844 | } | |
1845 | ||
1846 | for (;;) { | |
1847 | int send_barrier; | |
1848 | prepare_to_wait(&connection->sender_work.q_wait, &wait, TASK_INTERRUPTIBLE); | |
0500813f | 1849 | spin_lock_irq(&connection->resource->req_lock); |
b6dd1a89 | 1850 | spin_lock(&connection->sender_work.q_lock); /* FIXME get rid of this one? */ |
bc317a9e LE |
1851 | /* dequeue single item only, |
1852 | * we still use drbd_queue_work_front() in some places */ | |
1853 | if (!list_empty(&connection->sender_work.q)) | |
1854 | list_move(connection->sender_work.q.next, work_list); | |
b6dd1a89 LE |
1855 | spin_unlock(&connection->sender_work.q_lock); /* FIXME get rid of this one? */ |
1856 | if (!list_empty(work_list) || signal_pending(current)) { | |
0500813f | 1857 | spin_unlock_irq(&connection->resource->req_lock); |
b6dd1a89 LE |
1858 | break; |
1859 | } | |
f9c78128 LE |
1860 | |
1861 | /* We found nothing new to do, no to-be-communicated request, | |
1862 | * no other work item. We may still need to close the last | |
1863 | * epoch. Next incoming request epoch will be connection -> | |
1864 | * current transfer log epoch number. If that is different | |
1865 | * from the epoch of the last request we communicated, it is | |
1866 | * safe to send the epoch separating barrier now. | |
1867 | */ | |
1868 | send_barrier = | |
1869 | atomic_read(&connection->current_tle_nr) != | |
1870 | connection->send.current_epoch_nr; | |
0500813f | 1871 | spin_unlock_irq(&connection->resource->req_lock); |
f9c78128 LE |
1872 | |
1873 | if (send_barrier) | |
1874 | maybe_send_barrier(connection, | |
1875 | connection->send.current_epoch_nr + 1); | |
b6dd1a89 LE |
1876 | schedule(); |
1877 | /* may be woken up for other things but new work, too, | |
1878 | * e.g. if the current epoch got closed. | |
1879 | * In which case we send the barrier above. */ | |
1880 | } | |
1881 | finish_wait(&connection->sender_work.q_wait, &wait); | |
1882 | ||
1883 | /* someone may have changed the config while we have been waiting above. */ | |
1884 | rcu_read_lock(); | |
1885 | nc = rcu_dereference(connection->net_conf); | |
1886 | cork = nc ? nc->tcp_cork : 0; | |
1887 | rcu_read_unlock(); | |
1888 | mutex_lock(&connection->data.mutex); | |
1889 | if (connection->data.socket) { | |
1890 | if (cork) | |
1891 | drbd_tcp_cork(connection->data.socket); | |
1892 | else if (!uncork) | |
1893 | drbd_tcp_uncork(connection->data.socket); | |
1894 | } | |
1895 | mutex_unlock(&connection->data.mutex); | |
1896 | } | |
1897 | ||
b411b363 PR |
1898 | int drbd_worker(struct drbd_thread *thi) |
1899 | { | |
bde89a9e | 1900 | struct drbd_connection *connection = thi->connection; |
6db7e50a | 1901 | struct drbd_work *w = NULL; |
c06ece6b | 1902 | struct drbd_peer_device *peer_device; |
b411b363 | 1903 | LIST_HEAD(work_list); |
8c0785a5 | 1904 | int vnr; |
b411b363 | 1905 | |
e77a0a5c | 1906 | while (get_t_state(thi) == RUNNING) { |
80822284 | 1907 | drbd_thread_current_set_cpu(thi); |
b411b363 | 1908 | |
8c0785a5 LE |
1909 | /* as long as we use drbd_queue_work_front(), |
1910 | * we may only dequeue single work items here, not batches. */ | |
1911 | if (list_empty(&work_list)) | |
bde89a9e | 1912 | wait_for_work(connection, &work_list); |
b411b363 | 1913 | |
8c0785a5 | 1914 | if (signal_pending(current)) { |
b411b363 | 1915 | flush_signals(current); |
19393e10 | 1916 | if (get_t_state(thi) == RUNNING) { |
1ec861eb | 1917 | drbd_warn(connection, "Worker got an unexpected signal\n"); |
b411b363 | 1918 | continue; |
19393e10 | 1919 | } |
b411b363 PR |
1920 | break; |
1921 | } | |
1922 | ||
e77a0a5c | 1923 | if (get_t_state(thi) != RUNNING) |
b411b363 | 1924 | break; |
b411b363 | 1925 | |
8c0785a5 | 1926 | while (!list_empty(&work_list)) { |
6db7e50a AG |
1927 | w = list_first_entry(&work_list, struct drbd_work, list); |
1928 | list_del_init(&w->list); | |
1929 | if (w->cb(w, connection->cstate < C_WF_REPORT_PARAMS) == 0) | |
8c0785a5 | 1930 | continue; |
bde89a9e AG |
1931 | if (connection->cstate >= C_WF_REPORT_PARAMS) |
1932 | conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); | |
b411b363 PR |
1933 | } |
1934 | } | |
b411b363 | 1935 | |
8c0785a5 | 1936 | do { |
b411b363 | 1937 | while (!list_empty(&work_list)) { |
6db7e50a AG |
1938 | w = list_first_entry(&work_list, struct drbd_work, list); |
1939 | list_del_init(&w->list); | |
1940 | w->cb(w, 1); | |
b411b363 | 1941 | } |
bde89a9e | 1942 | dequeue_work_batch(&connection->sender_work, &work_list); |
8c0785a5 | 1943 | } while (!list_empty(&work_list)); |
b411b363 | 1944 | |
c141ebda | 1945 | rcu_read_lock(); |
c06ece6b AG |
1946 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
1947 | struct drbd_device *device = peer_device->device; | |
0b0ba1ef | 1948 | D_ASSERT(device, device->state.disk == D_DISKLESS && device->state.conn == C_STANDALONE); |
b30ab791 | 1949 | kref_get(&device->kref); |
c141ebda | 1950 | rcu_read_unlock(); |
b30ab791 | 1951 | drbd_device_cleanup(device); |
05a10ec7 | 1952 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda | 1953 | rcu_read_lock(); |
0e29d163 | 1954 | } |
c141ebda | 1955 | rcu_read_unlock(); |
b411b363 PR |
1956 | |
1957 | return 0; | |
1958 | } |