]>
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 | ||
44a4d551 | 586 | static int make_resync_request(struct drbd_device *const device, int cancel) |
b411b363 | 587 | { |
44a4d551 LE |
588 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
589 | struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL; | |
b411b363 PR |
590 | unsigned long bit; |
591 | sector_t sector; | |
b30ab791 | 592 | const sector_t capacity = drbd_get_capacity(device->this_bdev); |
1816a2b4 | 593 | int max_bio_size; |
e65f440d | 594 | int number, rollback_i, size; |
b411b363 | 595 | int align, queued, sndbuf; |
0f0601f4 | 596 | int i = 0; |
b411b363 PR |
597 | |
598 | if (unlikely(cancel)) | |
99920dc5 | 599 | return 0; |
b411b363 | 600 | |
b30ab791 | 601 | if (device->rs_total == 0) { |
af85e8e8 | 602 | /* empty resync? */ |
b30ab791 | 603 | drbd_resync_finished(device); |
99920dc5 | 604 | return 0; |
af85e8e8 LE |
605 | } |
606 | ||
b30ab791 AG |
607 | if (!get_ldev(device)) { |
608 | /* Since we only need to access device->rsync a | |
609 | get_ldev_if_state(device,D_FAILED) would be sufficient, but | |
b411b363 PR |
610 | to continue resync with a broken disk makes no sense at |
611 | all */ | |
d0180171 | 612 | drbd_err(device, "Disk broke down during resync!\n"); |
99920dc5 | 613 | return 0; |
b411b363 PR |
614 | } |
615 | ||
b30ab791 AG |
616 | max_bio_size = queue_max_hw_sectors(device->rq_queue) << 9; |
617 | number = drbd_rs_number_requests(device); | |
0e49d7b0 | 618 | if (number <= 0) |
0f0601f4 | 619 | goto requeue; |
b411b363 | 620 | |
b411b363 PR |
621 | for (i = 0; i < number; i++) { |
622 | /* Stop generating RS requests, when half of the send buffer is filled */ | |
44a4d551 LE |
623 | mutex_lock(&connection->data.mutex); |
624 | if (connection->data.socket) { | |
625 | queued = connection->data.socket->sk->sk_wmem_queued; | |
626 | sndbuf = connection->data.socket->sk->sk_sndbuf; | |
b411b363 PR |
627 | } else { |
628 | queued = 1; | |
629 | sndbuf = 0; | |
630 | } | |
44a4d551 | 631 | mutex_unlock(&connection->data.mutex); |
b411b363 PR |
632 | if (queued > sndbuf / 2) |
633 | goto requeue; | |
634 | ||
635 | next_sector: | |
636 | size = BM_BLOCK_SIZE; | |
b30ab791 | 637 | bit = drbd_bm_find_next(device, device->bm_resync_fo); |
b411b363 | 638 | |
4b0715f0 | 639 | if (bit == DRBD_END_OF_BITMAP) { |
b30ab791 AG |
640 | device->bm_resync_fo = drbd_bm_bits(device); |
641 | put_ldev(device); | |
99920dc5 | 642 | return 0; |
b411b363 PR |
643 | } |
644 | ||
645 | sector = BM_BIT_TO_SECT(bit); | |
646 | ||
b30ab791 AG |
647 | if (drbd_rs_should_slow_down(device, sector) || |
648 | drbd_try_rs_begin_io(device, sector)) { | |
649 | device->bm_resync_fo = bit; | |
b411b363 PR |
650 | goto requeue; |
651 | } | |
b30ab791 | 652 | device->bm_resync_fo = bit + 1; |
b411b363 | 653 | |
b30ab791 AG |
654 | if (unlikely(drbd_bm_test_bit(device, bit) == 0)) { |
655 | drbd_rs_complete_io(device, sector); | |
b411b363 PR |
656 | goto next_sector; |
657 | } | |
658 | ||
1816a2b4 | 659 | #if DRBD_MAX_BIO_SIZE > BM_BLOCK_SIZE |
b411b363 PR |
660 | /* try to find some adjacent bits. |
661 | * we stop if we have already the maximum req size. | |
662 | * | |
663 | * Additionally always align bigger requests, in order to | |
664 | * be prepared for all stripe sizes of software RAIDs. | |
b411b363 PR |
665 | */ |
666 | align = 1; | |
d207450c | 667 | rollback_i = i; |
6377b923 | 668 | while (i < number) { |
1816a2b4 | 669 | if (size + BM_BLOCK_SIZE > max_bio_size) |
b411b363 PR |
670 | break; |
671 | ||
672 | /* Be always aligned */ | |
673 | if (sector & ((1<<(align+3))-1)) | |
674 | break; | |
675 | ||
676 | /* do not cross extent boundaries */ | |
677 | if (((bit+1) & BM_BLOCKS_PER_BM_EXT_MASK) == 0) | |
678 | break; | |
679 | /* now, is it actually dirty, after all? | |
680 | * caution, drbd_bm_test_bit is tri-state for some | |
681 | * obscure reason; ( b == 0 ) would get the out-of-band | |
682 | * only accidentally right because of the "oddly sized" | |
683 | * adjustment below */ | |
b30ab791 | 684 | if (drbd_bm_test_bit(device, bit+1) != 1) |
b411b363 PR |
685 | break; |
686 | bit++; | |
687 | size += BM_BLOCK_SIZE; | |
688 | if ((BM_BLOCK_SIZE << align) <= size) | |
689 | align++; | |
690 | i++; | |
691 | } | |
692 | /* if we merged some, | |
693 | * reset the offset to start the next drbd_bm_find_next from */ | |
694 | if (size > BM_BLOCK_SIZE) | |
b30ab791 | 695 | device->bm_resync_fo = bit + 1; |
b411b363 PR |
696 | #endif |
697 | ||
698 | /* adjust very last sectors, in case we are oddly sized */ | |
699 | if (sector + (size>>9) > capacity) | |
700 | size = (capacity-sector)<<9; | |
44a4d551 LE |
701 | if (connection->agreed_pro_version >= 89 && |
702 | connection->csums_tfm) { | |
703 | switch (read_for_csum(peer_device, sector, size)) { | |
80a40e43 | 704 | case -EIO: /* Disk failure */ |
b30ab791 | 705 | put_ldev(device); |
99920dc5 | 706 | return -EIO; |
80a40e43 | 707 | case -EAGAIN: /* allocation failed, or ldev busy */ |
b30ab791 AG |
708 | drbd_rs_complete_io(device, sector); |
709 | device->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
d207450c | 710 | i = rollback_i; |
b411b363 | 711 | goto requeue; |
80a40e43 LE |
712 | case 0: |
713 | /* everything ok */ | |
714 | break; | |
715 | default: | |
716 | BUG(); | |
b411b363 PR |
717 | } |
718 | } else { | |
99920dc5 AG |
719 | int err; |
720 | ||
b30ab791 | 721 | inc_rs_pending(device); |
44a4d551 | 722 | err = drbd_send_drequest(peer_device, P_RS_DATA_REQUEST, |
99920dc5 AG |
723 | sector, size, ID_SYNCER); |
724 | if (err) { | |
d0180171 | 725 | drbd_err(device, "drbd_send_drequest() failed, aborting...\n"); |
b30ab791 AG |
726 | dec_rs_pending(device); |
727 | put_ldev(device); | |
99920dc5 | 728 | return err; |
b411b363 PR |
729 | } |
730 | } | |
731 | } | |
732 | ||
b30ab791 | 733 | if (device->bm_resync_fo >= drbd_bm_bits(device)) { |
b411b363 PR |
734 | /* last syncer _request_ was sent, |
735 | * but the P_RS_DATA_REPLY not yet received. sync will end (and | |
736 | * next sync group will resume), as soon as we receive the last | |
737 | * resync data block, and the last bit is cleared. | |
738 | * until then resync "work" is "inactive" ... | |
739 | */ | |
b30ab791 | 740 | put_ldev(device); |
99920dc5 | 741 | return 0; |
b411b363 PR |
742 | } |
743 | ||
744 | requeue: | |
b30ab791 AG |
745 | device->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
746 | mod_timer(&device->resync_timer, jiffies + SLEEP_TIME); | |
747 | put_ldev(device); | |
99920dc5 | 748 | return 0; |
b411b363 PR |
749 | } |
750 | ||
d448a2e1 | 751 | static int make_ov_request(struct drbd_device *device, int cancel) |
b411b363 PR |
752 | { |
753 | int number, i, size; | |
754 | sector_t sector; | |
b30ab791 | 755 | const sector_t capacity = drbd_get_capacity(device->this_bdev); |
58ffa580 | 756 | bool stop_sector_reached = false; |
b411b363 PR |
757 | |
758 | if (unlikely(cancel)) | |
759 | return 1; | |
760 | ||
b30ab791 | 761 | number = drbd_rs_number_requests(device); |
b411b363 | 762 | |
b30ab791 | 763 | sector = device->ov_position; |
b411b363 | 764 | for (i = 0; i < number; i++) { |
58ffa580 | 765 | if (sector >= capacity) |
b411b363 | 766 | return 1; |
58ffa580 LE |
767 | |
768 | /* We check for "finished" only in the reply path: | |
769 | * w_e_end_ov_reply(). | |
770 | * We need to send at least one request out. */ | |
771 | stop_sector_reached = i > 0 | |
b30ab791 AG |
772 | && verify_can_do_stop_sector(device) |
773 | && sector >= device->ov_stop_sector; | |
58ffa580 LE |
774 | if (stop_sector_reached) |
775 | break; | |
b411b363 PR |
776 | |
777 | size = BM_BLOCK_SIZE; | |
778 | ||
b30ab791 AG |
779 | if (drbd_rs_should_slow_down(device, sector) || |
780 | drbd_try_rs_begin_io(device, sector)) { | |
781 | device->ov_position = sector; | |
b411b363 PR |
782 | goto requeue; |
783 | } | |
784 | ||
785 | if (sector + (size>>9) > capacity) | |
786 | size = (capacity-sector)<<9; | |
787 | ||
b30ab791 | 788 | inc_rs_pending(device); |
69a22773 | 789 | if (drbd_send_ov_request(first_peer_device(device), sector, size)) { |
b30ab791 | 790 | dec_rs_pending(device); |
b411b363 PR |
791 | return 0; |
792 | } | |
793 | sector += BM_SECT_PER_BIT; | |
794 | } | |
b30ab791 | 795 | device->ov_position = sector; |
b411b363 PR |
796 | |
797 | requeue: | |
b30ab791 | 798 | device->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
58ffa580 | 799 | if (i == 0 || !stop_sector_reached) |
b30ab791 | 800 | mod_timer(&device->resync_timer, jiffies + SLEEP_TIME); |
b411b363 PR |
801 | return 1; |
802 | } | |
803 | ||
99920dc5 | 804 | int w_ov_finished(struct drbd_work *w, int cancel) |
b411b363 | 805 | { |
84b8c06b AG |
806 | struct drbd_device_work *dw = |
807 | container_of(w, struct drbd_device_work, w); | |
808 | struct drbd_device *device = dw->device; | |
809 | kfree(dw); | |
b30ab791 AG |
810 | ov_out_of_sync_print(device); |
811 | drbd_resync_finished(device); | |
b411b363 | 812 | |
99920dc5 | 813 | return 0; |
b411b363 PR |
814 | } |
815 | ||
99920dc5 | 816 | static int w_resync_finished(struct drbd_work *w, int cancel) |
b411b363 | 817 | { |
84b8c06b AG |
818 | struct drbd_device_work *dw = |
819 | container_of(w, struct drbd_device_work, w); | |
820 | struct drbd_device *device = dw->device; | |
821 | kfree(dw); | |
b411b363 | 822 | |
b30ab791 | 823 | drbd_resync_finished(device); |
b411b363 | 824 | |
99920dc5 | 825 | return 0; |
b411b363 PR |
826 | } |
827 | ||
b30ab791 | 828 | static void ping_peer(struct drbd_device *device) |
af85e8e8 | 829 | { |
a6b32bc3 | 830 | struct drbd_connection *connection = first_peer_device(device)->connection; |
2a67d8b9 | 831 | |
bde89a9e AG |
832 | clear_bit(GOT_PING_ACK, &connection->flags); |
833 | request_ping(connection); | |
834 | wait_event(connection->ping_wait, | |
835 | test_bit(GOT_PING_ACK, &connection->flags) || device->state.conn < C_CONNECTED); | |
af85e8e8 LE |
836 | } |
837 | ||
b30ab791 | 838 | int drbd_resync_finished(struct drbd_device *device) |
b411b363 PR |
839 | { |
840 | unsigned long db, dt, dbdt; | |
841 | unsigned long n_oos; | |
842 | union drbd_state os, ns; | |
84b8c06b | 843 | struct drbd_device_work *dw; |
b411b363 | 844 | char *khelper_cmd = NULL; |
26525618 | 845 | int verify_done = 0; |
b411b363 PR |
846 | |
847 | /* Remove all elements from the resync LRU. Since future actions | |
848 | * might set bits in the (main) bitmap, then the entries in the | |
849 | * resync LRU would be wrong. */ | |
b30ab791 | 850 | if (drbd_rs_del_all(device)) { |
b411b363 PR |
851 | /* In case this is not possible now, most probably because |
852 | * there are P_RS_DATA_REPLY Packets lingering on the worker's | |
853 | * queue (or even the read operations for those packets | |
854 | * is not finished by now). Retry in 100ms. */ | |
855 | ||
20ee6390 | 856 | schedule_timeout_interruptible(HZ / 10); |
84b8c06b AG |
857 | dw = kmalloc(sizeof(struct drbd_device_work), GFP_ATOMIC); |
858 | if (dw) { | |
859 | dw->w.cb = w_resync_finished; | |
860 | dw->device = device; | |
861 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, | |
862 | &dw->w); | |
b411b363 PR |
863 | return 1; |
864 | } | |
84b8c06b | 865 | drbd_err(device, "Warn failed to drbd_rs_del_all() and to kmalloc(dw).\n"); |
b411b363 PR |
866 | } |
867 | ||
b30ab791 | 868 | dt = (jiffies - device->rs_start - device->rs_paused) / HZ; |
b411b363 PR |
869 | if (dt <= 0) |
870 | dt = 1; | |
84b8c06b | 871 | |
b30ab791 | 872 | db = device->rs_total; |
58ffa580 | 873 | /* adjust for verify start and stop sectors, respective reached position */ |
b30ab791 AG |
874 | if (device->state.conn == C_VERIFY_S || device->state.conn == C_VERIFY_T) |
875 | db -= device->ov_left; | |
58ffa580 | 876 | |
b411b363 | 877 | dbdt = Bit2KB(db/dt); |
b30ab791 | 878 | device->rs_paused /= HZ; |
b411b363 | 879 | |
b30ab791 | 880 | if (!get_ldev(device)) |
b411b363 PR |
881 | goto out; |
882 | ||
b30ab791 | 883 | ping_peer(device); |
af85e8e8 | 884 | |
0500813f | 885 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 886 | os = drbd_read_state(device); |
b411b363 | 887 | |
26525618 LE |
888 | verify_done = (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T); |
889 | ||
b411b363 PR |
890 | /* This protects us against multiple calls (that can happen in the presence |
891 | of application IO), and against connectivity loss just before we arrive here. */ | |
892 | if (os.conn <= C_CONNECTED) | |
893 | goto out_unlock; | |
894 | ||
895 | ns = os; | |
896 | ns.conn = C_CONNECTED; | |
897 | ||
d0180171 | 898 | drbd_info(device, "%s done (total %lu sec; paused %lu sec; %lu K/sec)\n", |
58ffa580 | 899 | verify_done ? "Online verify" : "Resync", |
b30ab791 | 900 | dt + device->rs_paused, device->rs_paused, dbdt); |
b411b363 | 901 | |
b30ab791 | 902 | n_oos = drbd_bm_total_weight(device); |
b411b363 PR |
903 | |
904 | if (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) { | |
905 | if (n_oos) { | |
d0180171 | 906 | drbd_alert(device, "Online verify found %lu %dk block out of sync!\n", |
b411b363 PR |
907 | n_oos, Bit2KB(1)); |
908 | khelper_cmd = "out-of-sync"; | |
909 | } | |
910 | } else { | |
0b0ba1ef | 911 | D_ASSERT(device, (n_oos - device->rs_failed) == 0); |
b411b363 PR |
912 | |
913 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) | |
914 | khelper_cmd = "after-resync-target"; | |
915 | ||
a6b32bc3 | 916 | if (first_peer_device(device)->connection->csums_tfm && device->rs_total) { |
b30ab791 AG |
917 | const unsigned long s = device->rs_same_csum; |
918 | const unsigned long t = device->rs_total; | |
b411b363 PR |
919 | const int ratio = |
920 | (t == 0) ? 0 : | |
921 | (t < 100000) ? ((s*100)/t) : (s/(t/100)); | |
d0180171 | 922 | drbd_info(device, "%u %% had equal checksums, eliminated: %luK; " |
b411b363 PR |
923 | "transferred %luK total %luK\n", |
924 | ratio, | |
b30ab791 AG |
925 | Bit2KB(device->rs_same_csum), |
926 | Bit2KB(device->rs_total - device->rs_same_csum), | |
927 | Bit2KB(device->rs_total)); | |
b411b363 PR |
928 | } |
929 | } | |
930 | ||
b30ab791 | 931 | if (device->rs_failed) { |
d0180171 | 932 | drbd_info(device, " %lu failed blocks\n", device->rs_failed); |
b411b363 PR |
933 | |
934 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
935 | ns.disk = D_INCONSISTENT; | |
936 | ns.pdsk = D_UP_TO_DATE; | |
937 | } else { | |
938 | ns.disk = D_UP_TO_DATE; | |
939 | ns.pdsk = D_INCONSISTENT; | |
940 | } | |
941 | } else { | |
942 | ns.disk = D_UP_TO_DATE; | |
943 | ns.pdsk = D_UP_TO_DATE; | |
944 | ||
945 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
b30ab791 | 946 | if (device->p_uuid) { |
b411b363 PR |
947 | int i; |
948 | for (i = UI_BITMAP ; i <= UI_HISTORY_END ; i++) | |
b30ab791 AG |
949 | _drbd_uuid_set(device, i, device->p_uuid[i]); |
950 | drbd_uuid_set(device, UI_BITMAP, device->ldev->md.uuid[UI_CURRENT]); | |
951 | _drbd_uuid_set(device, UI_CURRENT, device->p_uuid[UI_CURRENT]); | |
b411b363 | 952 | } else { |
d0180171 | 953 | drbd_err(device, "device->p_uuid is NULL! BUG\n"); |
b411b363 PR |
954 | } |
955 | } | |
956 | ||
62b0da3a LE |
957 | if (!(os.conn == C_VERIFY_S || os.conn == C_VERIFY_T)) { |
958 | /* for verify runs, we don't update uuids here, | |
959 | * so there would be nothing to report. */ | |
b30ab791 AG |
960 | drbd_uuid_set_bm(device, 0UL); |
961 | drbd_print_uuids(device, "updated UUIDs"); | |
962 | if (device->p_uuid) { | |
62b0da3a LE |
963 | /* Now the two UUID sets are equal, update what we |
964 | * know of the peer. */ | |
965 | int i; | |
966 | for (i = UI_CURRENT ; i <= UI_HISTORY_END ; i++) | |
b30ab791 | 967 | device->p_uuid[i] = device->ldev->md.uuid[i]; |
62b0da3a | 968 | } |
b411b363 PR |
969 | } |
970 | } | |
971 | ||
b30ab791 | 972 | _drbd_set_state(device, ns, CS_VERBOSE, NULL); |
b411b363 | 973 | out_unlock: |
0500813f | 974 | spin_unlock_irq(&device->resource->req_lock); |
b30ab791 | 975 | put_ldev(device); |
b411b363 | 976 | out: |
b30ab791 AG |
977 | device->rs_total = 0; |
978 | device->rs_failed = 0; | |
979 | device->rs_paused = 0; | |
58ffa580 LE |
980 | |
981 | /* reset start sector, if we reached end of device */ | |
b30ab791 AG |
982 | if (verify_done && device->ov_left == 0) |
983 | device->ov_start_sector = 0; | |
b411b363 | 984 | |
b30ab791 | 985 | drbd_md_sync(device); |
13d42685 | 986 | |
b411b363 | 987 | if (khelper_cmd) |
b30ab791 | 988 | drbd_khelper(device, khelper_cmd); |
b411b363 PR |
989 | |
990 | return 1; | |
991 | } | |
992 | ||
993 | /* helper */ | |
b30ab791 | 994 | static void move_to_net_ee_or_free(struct drbd_device *device, struct drbd_peer_request *peer_req) |
b411b363 | 995 | { |
045417f7 | 996 | if (drbd_peer_req_has_active_page(peer_req)) { |
b411b363 | 997 | /* This might happen if sendpage() has not finished */ |
db830c46 | 998 | int i = (peer_req->i.size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b30ab791 AG |
999 | atomic_add(i, &device->pp_in_use_by_net); |
1000 | atomic_sub(i, &device->pp_in_use); | |
0500813f | 1001 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 1002 | list_add_tail(&peer_req->w.list, &device->net_ee); |
0500813f | 1003 | spin_unlock_irq(&device->resource->req_lock); |
435f0740 | 1004 | wake_up(&drbd_pp_wait); |
b411b363 | 1005 | } else |
b30ab791 | 1006 | drbd_free_peer_req(device, peer_req); |
b411b363 PR |
1007 | } |
1008 | ||
1009 | /** | |
1010 | * w_e_end_data_req() - Worker callback, to send a P_DATA_REPLY packet in response to a P_DATA_REQUEST | |
b30ab791 | 1011 | * @device: DRBD device. |
b411b363 PR |
1012 | * @w: work object. |
1013 | * @cancel: The connection will be closed anyways | |
1014 | */ | |
99920dc5 | 1015 | int w_e_end_data_req(struct drbd_work *w, int cancel) |
b411b363 | 1016 | { |
a8cd15ba | 1017 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1018 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1019 | struct drbd_device *device = peer_device->device; | |
99920dc5 | 1020 | int err; |
b411b363 PR |
1021 | |
1022 | if (unlikely(cancel)) { | |
b30ab791 AG |
1023 | drbd_free_peer_req(device, peer_req); |
1024 | dec_unacked(device); | |
99920dc5 | 1025 | return 0; |
b411b363 PR |
1026 | } |
1027 | ||
db830c46 | 1028 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
6780139c | 1029 | err = drbd_send_block(peer_device, P_DATA_REPLY, peer_req); |
b411b363 PR |
1030 | } else { |
1031 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1032 | drbd_err(device, "Sending NegDReply. sector=%llus.\n", |
db830c46 | 1033 | (unsigned long long)peer_req->i.sector); |
b411b363 | 1034 | |
6780139c | 1035 | err = drbd_send_ack(peer_device, P_NEG_DREPLY, peer_req); |
b411b363 PR |
1036 | } |
1037 | ||
b30ab791 | 1038 | dec_unacked(device); |
b411b363 | 1039 | |
b30ab791 | 1040 | move_to_net_ee_or_free(device, peer_req); |
b411b363 | 1041 | |
99920dc5 | 1042 | if (unlikely(err)) |
d0180171 | 1043 | drbd_err(device, "drbd_send_block() failed\n"); |
99920dc5 | 1044 | return err; |
b411b363 PR |
1045 | } |
1046 | ||
1047 | /** | |
a209b4ae | 1048 | * 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 |
1049 | * @w: work object. |
1050 | * @cancel: The connection will be closed anyways | |
1051 | */ | |
99920dc5 | 1052 | int w_e_end_rsdata_req(struct drbd_work *w, int cancel) |
b411b363 | 1053 | { |
a8cd15ba | 1054 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1055 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1056 | struct drbd_device *device = peer_device->device; | |
99920dc5 | 1057 | int err; |
b411b363 PR |
1058 | |
1059 | if (unlikely(cancel)) { | |
b30ab791 AG |
1060 | drbd_free_peer_req(device, peer_req); |
1061 | dec_unacked(device); | |
99920dc5 | 1062 | return 0; |
b411b363 PR |
1063 | } |
1064 | ||
b30ab791 AG |
1065 | if (get_ldev_if_state(device, D_FAILED)) { |
1066 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1067 | put_ldev(device); | |
b411b363 PR |
1068 | } |
1069 | ||
b30ab791 | 1070 | if (device->state.conn == C_AHEAD) { |
6780139c | 1071 | err = drbd_send_ack(peer_device, P_RS_CANCEL, peer_req); |
db830c46 | 1072 | } else if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b30ab791 AG |
1073 | if (likely(device->state.pdsk >= D_INCONSISTENT)) { |
1074 | inc_rs_pending(device); | |
6780139c | 1075 | err = drbd_send_block(peer_device, P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1076 | } else { |
1077 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1078 | drbd_err(device, "Not sending RSDataReply, " |
b411b363 | 1079 | "partner DISKLESS!\n"); |
99920dc5 | 1080 | err = 0; |
b411b363 PR |
1081 | } |
1082 | } else { | |
1083 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1084 | drbd_err(device, "Sending NegRSDReply. sector %llus.\n", |
db830c46 | 1085 | (unsigned long long)peer_req->i.sector); |
b411b363 | 1086 | |
6780139c | 1087 | err = drbd_send_ack(peer_device, P_NEG_RS_DREPLY, peer_req); |
b411b363 PR |
1088 | |
1089 | /* update resync data with failure */ | |
b30ab791 | 1090 | drbd_rs_failed_io(device, peer_req->i.sector, peer_req->i.size); |
b411b363 PR |
1091 | } |
1092 | ||
b30ab791 | 1093 | dec_unacked(device); |
b411b363 | 1094 | |
b30ab791 | 1095 | move_to_net_ee_or_free(device, peer_req); |
b411b363 | 1096 | |
99920dc5 | 1097 | if (unlikely(err)) |
d0180171 | 1098 | drbd_err(device, "drbd_send_block() failed\n"); |
99920dc5 | 1099 | return err; |
b411b363 PR |
1100 | } |
1101 | ||
99920dc5 | 1102 | int w_e_end_csum_rs_req(struct drbd_work *w, int cancel) |
b411b363 | 1103 | { |
a8cd15ba | 1104 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1105 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1106 | struct drbd_device *device = peer_device->device; | |
b411b363 PR |
1107 | struct digest_info *di; |
1108 | int digest_size; | |
1109 | void *digest = NULL; | |
99920dc5 | 1110 | int err, eq = 0; |
b411b363 PR |
1111 | |
1112 | if (unlikely(cancel)) { | |
b30ab791 AG |
1113 | drbd_free_peer_req(device, peer_req); |
1114 | dec_unacked(device); | |
99920dc5 | 1115 | return 0; |
b411b363 PR |
1116 | } |
1117 | ||
b30ab791 AG |
1118 | if (get_ldev(device)) { |
1119 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1120 | put_ldev(device); | |
1d53f09e | 1121 | } |
b411b363 | 1122 | |
db830c46 | 1123 | di = peer_req->digest; |
b411b363 | 1124 | |
db830c46 | 1125 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1126 | /* quick hack to try to avoid a race against reconfiguration. |
1127 | * a real fix would be much more involved, | |
1128 | * introducing more locking mechanisms */ | |
6780139c AG |
1129 | if (peer_device->connection->csums_tfm) { |
1130 | digest_size = crypto_hash_digestsize(peer_device->connection->csums_tfm); | |
0b0ba1ef | 1131 | D_ASSERT(device, digest_size == di->digest_size); |
b411b363 PR |
1132 | digest = kmalloc(digest_size, GFP_NOIO); |
1133 | } | |
1134 | if (digest) { | |
6780139c | 1135 | drbd_csum_ee(peer_device->connection->csums_tfm, peer_req, digest); |
b411b363 PR |
1136 | eq = !memcmp(digest, di->digest, digest_size); |
1137 | kfree(digest); | |
1138 | } | |
1139 | ||
1140 | if (eq) { | |
b30ab791 | 1141 | drbd_set_in_sync(device, peer_req->i.sector, peer_req->i.size); |
676396d5 | 1142 | /* rs_same_csums unit is BM_BLOCK_SIZE */ |
b30ab791 | 1143 | device->rs_same_csum += peer_req->i.size >> BM_BLOCK_SHIFT; |
6780139c | 1144 | err = drbd_send_ack(peer_device, P_RS_IS_IN_SYNC, peer_req); |
b411b363 | 1145 | } else { |
b30ab791 | 1146 | inc_rs_pending(device); |
db830c46 AG |
1147 | peer_req->block_id = ID_SYNCER; /* By setting block_id, digest pointer becomes invalid! */ |
1148 | peer_req->flags &= ~EE_HAS_DIGEST; /* This peer request no longer has a digest pointer */ | |
204bba99 | 1149 | kfree(di); |
6780139c | 1150 | err = drbd_send_block(peer_device, P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1151 | } |
1152 | } else { | |
6780139c | 1153 | err = drbd_send_ack(peer_device, P_NEG_RS_DREPLY, peer_req); |
b411b363 | 1154 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 1155 | drbd_err(device, "Sending NegDReply. I guess it gets messy.\n"); |
b411b363 PR |
1156 | } |
1157 | ||
b30ab791 AG |
1158 | dec_unacked(device); |
1159 | move_to_net_ee_or_free(device, peer_req); | |
b411b363 | 1160 | |
99920dc5 | 1161 | if (unlikely(err)) |
d0180171 | 1162 | drbd_err(device, "drbd_send_block/ack() failed\n"); |
99920dc5 | 1163 | return err; |
b411b363 PR |
1164 | } |
1165 | ||
99920dc5 | 1166 | int w_e_end_ov_req(struct drbd_work *w, int cancel) |
b411b363 | 1167 | { |
a8cd15ba | 1168 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1169 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1170 | struct drbd_device *device = peer_device->device; | |
db830c46 AG |
1171 | sector_t sector = peer_req->i.sector; |
1172 | unsigned int size = peer_req->i.size; | |
b411b363 PR |
1173 | int digest_size; |
1174 | void *digest; | |
99920dc5 | 1175 | int err = 0; |
b411b363 PR |
1176 | |
1177 | if (unlikely(cancel)) | |
1178 | goto out; | |
1179 | ||
6780139c | 1180 | digest_size = crypto_hash_digestsize(peer_device->connection->verify_tfm); |
b411b363 | 1181 | digest = kmalloc(digest_size, GFP_NOIO); |
8f21420e | 1182 | if (!digest) { |
99920dc5 | 1183 | err = 1; /* terminate the connection in case the allocation failed */ |
8f21420e | 1184 | goto out; |
b411b363 PR |
1185 | } |
1186 | ||
db830c46 | 1187 | if (likely(!(peer_req->flags & EE_WAS_ERROR))) |
6780139c | 1188 | drbd_csum_ee(peer_device->connection->verify_tfm, peer_req, digest); |
8f21420e PR |
1189 | else |
1190 | memset(digest, 0, digest_size); | |
1191 | ||
53ea4331 LE |
1192 | /* Free e and pages before send. |
1193 | * In case we block on congestion, we could otherwise run into | |
1194 | * some distributed deadlock, if the other side blocks on | |
1195 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1196 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 1197 | drbd_free_peer_req(device, peer_req); |
db830c46 | 1198 | peer_req = NULL; |
b30ab791 | 1199 | inc_rs_pending(device); |
6780139c | 1200 | err = drbd_send_drequest_csum(peer_device, sector, size, digest, digest_size, P_OV_REPLY); |
99920dc5 | 1201 | if (err) |
b30ab791 | 1202 | dec_rs_pending(device); |
8f21420e PR |
1203 | kfree(digest); |
1204 | ||
b411b363 | 1205 | out: |
db830c46 | 1206 | if (peer_req) |
b30ab791 AG |
1207 | drbd_free_peer_req(device, peer_req); |
1208 | dec_unacked(device); | |
99920dc5 | 1209 | return err; |
b411b363 PR |
1210 | } |
1211 | ||
b30ab791 | 1212 | void drbd_ov_out_of_sync_found(struct drbd_device *device, sector_t sector, int size) |
b411b363 | 1213 | { |
b30ab791 AG |
1214 | if (device->ov_last_oos_start + device->ov_last_oos_size == sector) { |
1215 | device->ov_last_oos_size += size>>9; | |
b411b363 | 1216 | } else { |
b30ab791 AG |
1217 | device->ov_last_oos_start = sector; |
1218 | device->ov_last_oos_size = size>>9; | |
b411b363 | 1219 | } |
b30ab791 | 1220 | drbd_set_out_of_sync(device, sector, size); |
b411b363 PR |
1221 | } |
1222 | ||
99920dc5 | 1223 | int w_e_end_ov_reply(struct drbd_work *w, int cancel) |
b411b363 | 1224 | { |
a8cd15ba | 1225 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1226 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1227 | struct drbd_device *device = peer_device->device; | |
b411b363 | 1228 | struct digest_info *di; |
b411b363 | 1229 | void *digest; |
db830c46 AG |
1230 | sector_t sector = peer_req->i.sector; |
1231 | unsigned int size = peer_req->i.size; | |
53ea4331 | 1232 | int digest_size; |
99920dc5 | 1233 | int err, eq = 0; |
58ffa580 | 1234 | bool stop_sector_reached = false; |
b411b363 PR |
1235 | |
1236 | if (unlikely(cancel)) { | |
b30ab791 AG |
1237 | drbd_free_peer_req(device, peer_req); |
1238 | dec_unacked(device); | |
99920dc5 | 1239 | return 0; |
b411b363 PR |
1240 | } |
1241 | ||
1242 | /* after "cancel", because after drbd_disconnect/drbd_rs_cancel_all | |
1243 | * the resync lru has been cleaned up already */ | |
b30ab791 AG |
1244 | if (get_ldev(device)) { |
1245 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1246 | put_ldev(device); | |
1d53f09e | 1247 | } |
b411b363 | 1248 | |
db830c46 | 1249 | di = peer_req->digest; |
b411b363 | 1250 | |
db830c46 | 1251 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
6780139c | 1252 | digest_size = crypto_hash_digestsize(peer_device->connection->verify_tfm); |
b411b363 PR |
1253 | digest = kmalloc(digest_size, GFP_NOIO); |
1254 | if (digest) { | |
6780139c | 1255 | drbd_csum_ee(peer_device->connection->verify_tfm, peer_req, digest); |
b411b363 | 1256 | |
0b0ba1ef | 1257 | D_ASSERT(device, digest_size == di->digest_size); |
b411b363 PR |
1258 | eq = !memcmp(digest, di->digest, digest_size); |
1259 | kfree(digest); | |
1260 | } | |
b411b363 PR |
1261 | } |
1262 | ||
9676c760 LE |
1263 | /* Free peer_req and pages before send. |
1264 | * In case we block on congestion, we could otherwise run into | |
1265 | * some distributed deadlock, if the other side blocks on | |
1266 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1267 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 1268 | drbd_free_peer_req(device, peer_req); |
b411b363 | 1269 | if (!eq) |
b30ab791 | 1270 | drbd_ov_out_of_sync_found(device, sector, size); |
b411b363 | 1271 | else |
b30ab791 | 1272 | ov_out_of_sync_print(device); |
b411b363 | 1273 | |
6780139c | 1274 | err = drbd_send_ack_ex(peer_device, P_OV_RESULT, sector, size, |
fa79abd8 | 1275 | eq ? ID_IN_SYNC : ID_OUT_OF_SYNC); |
b411b363 | 1276 | |
b30ab791 | 1277 | dec_unacked(device); |
b411b363 | 1278 | |
b30ab791 | 1279 | --device->ov_left; |
ea5442af LE |
1280 | |
1281 | /* let's advance progress step marks only for every other megabyte */ | |
b30ab791 AG |
1282 | if ((device->ov_left & 0x200) == 0x200) |
1283 | drbd_advance_rs_marks(device, device->ov_left); | |
ea5442af | 1284 | |
b30ab791 AG |
1285 | stop_sector_reached = verify_can_do_stop_sector(device) && |
1286 | (sector + (size>>9)) >= device->ov_stop_sector; | |
58ffa580 | 1287 | |
b30ab791 AG |
1288 | if (device->ov_left == 0 || stop_sector_reached) { |
1289 | ov_out_of_sync_print(device); | |
1290 | drbd_resync_finished(device); | |
b411b363 PR |
1291 | } |
1292 | ||
99920dc5 | 1293 | return err; |
b411b363 PR |
1294 | } |
1295 | ||
b6dd1a89 LE |
1296 | /* FIXME |
1297 | * We need to track the number of pending barrier acks, | |
1298 | * and to be able to wait for them. | |
1299 | * See also comment in drbd_adm_attach before drbd_suspend_io. | |
1300 | */ | |
bde89a9e | 1301 | static int drbd_send_barrier(struct drbd_connection *connection) |
b411b363 | 1302 | { |
9f5bdc33 | 1303 | struct p_barrier *p; |
b6dd1a89 | 1304 | struct drbd_socket *sock; |
b411b363 | 1305 | |
bde89a9e AG |
1306 | sock = &connection->data; |
1307 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1308 | if (!p) |
1309 | return -EIO; | |
bde89a9e | 1310 | p->barrier = connection->send.current_epoch_nr; |
b6dd1a89 | 1311 | p->pad = 0; |
bde89a9e | 1312 | connection->send.current_epoch_writes = 0; |
b6dd1a89 | 1313 | |
bde89a9e | 1314 | return conn_send_command(connection, sock, P_BARRIER, sizeof(*p), NULL, 0); |
b411b363 PR |
1315 | } |
1316 | ||
99920dc5 | 1317 | int w_send_write_hint(struct drbd_work *w, int cancel) |
b411b363 | 1318 | { |
84b8c06b AG |
1319 | struct drbd_device *device = |
1320 | container_of(w, struct drbd_device, unplug_work); | |
9f5bdc33 AG |
1321 | struct drbd_socket *sock; |
1322 | ||
b411b363 | 1323 | if (cancel) |
99920dc5 | 1324 | return 0; |
a6b32bc3 | 1325 | sock = &first_peer_device(device)->connection->data; |
69a22773 | 1326 | if (!drbd_prepare_command(first_peer_device(device), sock)) |
9f5bdc33 | 1327 | return -EIO; |
69a22773 | 1328 | return drbd_send_command(first_peer_device(device), sock, P_UNPLUG_REMOTE, 0, NULL, 0); |
b411b363 PR |
1329 | } |
1330 | ||
bde89a9e | 1331 | static void re_init_if_first_write(struct drbd_connection *connection, unsigned int epoch) |
4eb9b3cb | 1332 | { |
bde89a9e AG |
1333 | if (!connection->send.seen_any_write_yet) { |
1334 | connection->send.seen_any_write_yet = true; | |
1335 | connection->send.current_epoch_nr = epoch; | |
1336 | connection->send.current_epoch_writes = 0; | |
4eb9b3cb LE |
1337 | } |
1338 | } | |
1339 | ||
bde89a9e | 1340 | static void maybe_send_barrier(struct drbd_connection *connection, unsigned int epoch) |
4eb9b3cb LE |
1341 | { |
1342 | /* re-init if first write on this connection */ | |
bde89a9e | 1343 | if (!connection->send.seen_any_write_yet) |
4eb9b3cb | 1344 | return; |
bde89a9e AG |
1345 | if (connection->send.current_epoch_nr != epoch) { |
1346 | if (connection->send.current_epoch_writes) | |
1347 | drbd_send_barrier(connection); | |
1348 | connection->send.current_epoch_nr = epoch; | |
4eb9b3cb LE |
1349 | } |
1350 | } | |
1351 | ||
8f7bed77 | 1352 | int w_send_out_of_sync(struct drbd_work *w, int cancel) |
73a01a18 PR |
1353 | { |
1354 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1355 | struct drbd_device *device = req->device; |
44a4d551 LE |
1356 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
1357 | struct drbd_connection *const connection = peer_device->connection; | |
99920dc5 | 1358 | int err; |
73a01a18 PR |
1359 | |
1360 | if (unlikely(cancel)) { | |
8554df1c | 1361 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1362 | return 0; |
73a01a18 PR |
1363 | } |
1364 | ||
bde89a9e | 1365 | /* this time, no connection->send.current_epoch_writes++; |
b6dd1a89 LE |
1366 | * If it was sent, it was the closing barrier for the last |
1367 | * replicated epoch, before we went into AHEAD mode. | |
1368 | * No more barriers will be sent, until we leave AHEAD mode again. */ | |
bde89a9e | 1369 | maybe_send_barrier(connection, req->epoch); |
b6dd1a89 | 1370 | |
44a4d551 | 1371 | err = drbd_send_out_of_sync(peer_device, req); |
8554df1c | 1372 | req_mod(req, OOS_HANDED_TO_NETWORK); |
73a01a18 | 1373 | |
99920dc5 | 1374 | return err; |
73a01a18 PR |
1375 | } |
1376 | ||
b411b363 PR |
1377 | /** |
1378 | * w_send_dblock() - Worker callback to send a P_DATA packet in order to mirror a write request | |
b411b363 PR |
1379 | * @w: work object. |
1380 | * @cancel: The connection will be closed anyways | |
1381 | */ | |
99920dc5 | 1382 | int w_send_dblock(struct drbd_work *w, int cancel) |
b411b363 PR |
1383 | { |
1384 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1385 | struct drbd_device *device = req->device; |
44a4d551 LE |
1386 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
1387 | struct drbd_connection *connection = peer_device->connection; | |
99920dc5 | 1388 | int err; |
b411b363 PR |
1389 | |
1390 | if (unlikely(cancel)) { | |
8554df1c | 1391 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1392 | return 0; |
b411b363 PR |
1393 | } |
1394 | ||
bde89a9e AG |
1395 | re_init_if_first_write(connection, req->epoch); |
1396 | maybe_send_barrier(connection, req->epoch); | |
1397 | connection->send.current_epoch_writes++; | |
b6dd1a89 | 1398 | |
44a4d551 | 1399 | err = drbd_send_dblock(peer_device, req); |
99920dc5 | 1400 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1401 | |
99920dc5 | 1402 | return err; |
b411b363 PR |
1403 | } |
1404 | ||
1405 | /** | |
1406 | * w_send_read_req() - Worker callback to send a read request (P_DATA_REQUEST) packet | |
b411b363 PR |
1407 | * @w: work object. |
1408 | * @cancel: The connection will be closed anyways | |
1409 | */ | |
99920dc5 | 1410 | int w_send_read_req(struct drbd_work *w, int cancel) |
b411b363 PR |
1411 | { |
1412 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1413 | struct drbd_device *device = req->device; |
44a4d551 LE |
1414 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
1415 | struct drbd_connection *connection = peer_device->connection; | |
99920dc5 | 1416 | int err; |
b411b363 PR |
1417 | |
1418 | if (unlikely(cancel)) { | |
8554df1c | 1419 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1420 | return 0; |
b411b363 PR |
1421 | } |
1422 | ||
b6dd1a89 LE |
1423 | /* Even read requests may close a write epoch, |
1424 | * if there was any yet. */ | |
bde89a9e | 1425 | maybe_send_barrier(connection, req->epoch); |
b6dd1a89 | 1426 | |
44a4d551 | 1427 | err = drbd_send_drequest(peer_device, P_DATA_REQUEST, req->i.sector, req->i.size, |
6c1005e7 | 1428 | (unsigned long)req); |
b411b363 | 1429 | |
99920dc5 | 1430 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1431 | |
99920dc5 | 1432 | return err; |
b411b363 PR |
1433 | } |
1434 | ||
99920dc5 | 1435 | int w_restart_disk_io(struct drbd_work *w, int cancel) |
265be2d0 PR |
1436 | { |
1437 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1438 | struct drbd_device *device = req->device; |
265be2d0 | 1439 | |
0778286a | 1440 | if (bio_data_dir(req->master_bio) == WRITE && req->rq_state & RQ_IN_ACT_LOG) |
4dd726f0 | 1441 | drbd_al_begin_io(device, &req->i); |
265be2d0 PR |
1442 | |
1443 | drbd_req_make_private_bio(req, req->master_bio); | |
b30ab791 | 1444 | req->private_bio->bi_bdev = device->ldev->backing_bdev; |
265be2d0 PR |
1445 | generic_make_request(req->private_bio); |
1446 | ||
99920dc5 | 1447 | return 0; |
265be2d0 PR |
1448 | } |
1449 | ||
b30ab791 | 1450 | static int _drbd_may_sync_now(struct drbd_device *device) |
b411b363 | 1451 | { |
b30ab791 | 1452 | struct drbd_device *odev = device; |
95f8efd0 | 1453 | int resync_after; |
b411b363 PR |
1454 | |
1455 | while (1) { | |
a3f8f7dc | 1456 | if (!odev->ldev || odev->state.disk == D_DISKLESS) |
438c8374 | 1457 | return 1; |
daeda1cc | 1458 | rcu_read_lock(); |
95f8efd0 | 1459 | resync_after = rcu_dereference(odev->ldev->disk_conf)->resync_after; |
daeda1cc | 1460 | rcu_read_unlock(); |
95f8efd0 | 1461 | if (resync_after == -1) |
b411b363 | 1462 | return 1; |
b30ab791 | 1463 | odev = minor_to_device(resync_after); |
a3f8f7dc | 1464 | if (!odev) |
841ce241 | 1465 | return 1; |
b411b363 PR |
1466 | if ((odev->state.conn >= C_SYNC_SOURCE && |
1467 | odev->state.conn <= C_PAUSED_SYNC_T) || | |
1468 | odev->state.aftr_isp || odev->state.peer_isp || | |
1469 | odev->state.user_isp) | |
1470 | return 0; | |
1471 | } | |
1472 | } | |
1473 | ||
1474 | /** | |
1475 | * _drbd_pause_after() - Pause resync on all devices that may not resync now | |
b30ab791 | 1476 | * @device: DRBD device. |
b411b363 PR |
1477 | * |
1478 | * Called from process context only (admin command and after_state_ch). | |
1479 | */ | |
b30ab791 | 1480 | static int _drbd_pause_after(struct drbd_device *device) |
b411b363 | 1481 | { |
54761697 | 1482 | struct drbd_device *odev; |
b411b363 PR |
1483 | int i, rv = 0; |
1484 | ||
695d08fa | 1485 | rcu_read_lock(); |
05a10ec7 | 1486 | idr_for_each_entry(&drbd_devices, odev, i) { |
b411b363 PR |
1487 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1488 | continue; | |
1489 | if (!_drbd_may_sync_now(odev)) | |
1490 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 1), CS_HARD, NULL) | |
1491 | != SS_NOTHING_TO_DO); | |
1492 | } | |
695d08fa | 1493 | rcu_read_unlock(); |
b411b363 PR |
1494 | |
1495 | return rv; | |
1496 | } | |
1497 | ||
1498 | /** | |
1499 | * _drbd_resume_next() - Resume resync on all devices that may resync now | |
b30ab791 | 1500 | * @device: DRBD device. |
b411b363 PR |
1501 | * |
1502 | * Called from process context only (admin command and worker). | |
1503 | */ | |
b30ab791 | 1504 | static int _drbd_resume_next(struct drbd_device *device) |
b411b363 | 1505 | { |
54761697 | 1506 | struct drbd_device *odev; |
b411b363 PR |
1507 | int i, rv = 0; |
1508 | ||
695d08fa | 1509 | rcu_read_lock(); |
05a10ec7 | 1510 | idr_for_each_entry(&drbd_devices, odev, i) { |
b411b363 PR |
1511 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1512 | continue; | |
1513 | if (odev->state.aftr_isp) { | |
1514 | if (_drbd_may_sync_now(odev)) | |
1515 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 0), | |
1516 | CS_HARD, NULL) | |
1517 | != SS_NOTHING_TO_DO) ; | |
1518 | } | |
1519 | } | |
695d08fa | 1520 | rcu_read_unlock(); |
b411b363 PR |
1521 | return rv; |
1522 | } | |
1523 | ||
b30ab791 | 1524 | void resume_next_sg(struct drbd_device *device) |
b411b363 PR |
1525 | { |
1526 | write_lock_irq(&global_state_lock); | |
b30ab791 | 1527 | _drbd_resume_next(device); |
b411b363 PR |
1528 | write_unlock_irq(&global_state_lock); |
1529 | } | |
1530 | ||
b30ab791 | 1531 | void suspend_other_sg(struct drbd_device *device) |
b411b363 PR |
1532 | { |
1533 | write_lock_irq(&global_state_lock); | |
b30ab791 | 1534 | _drbd_pause_after(device); |
b411b363 PR |
1535 | write_unlock_irq(&global_state_lock); |
1536 | } | |
1537 | ||
dc97b708 | 1538 | /* caller must hold global_state_lock */ |
b30ab791 | 1539 | enum drbd_ret_code drbd_resync_after_valid(struct drbd_device *device, int o_minor) |
b411b363 | 1540 | { |
54761697 | 1541 | struct drbd_device *odev; |
95f8efd0 | 1542 | int resync_after; |
b411b363 PR |
1543 | |
1544 | if (o_minor == -1) | |
1545 | return NO_ERROR; | |
a3f8f7dc | 1546 | if (o_minor < -1 || o_minor > MINORMASK) |
95f8efd0 | 1547 | return ERR_RESYNC_AFTER; |
b411b363 PR |
1548 | |
1549 | /* check for loops */ | |
b30ab791 | 1550 | odev = minor_to_device(o_minor); |
b411b363 | 1551 | while (1) { |
b30ab791 | 1552 | if (odev == device) |
95f8efd0 | 1553 | return ERR_RESYNC_AFTER_CYCLE; |
b411b363 | 1554 | |
a3f8f7dc LE |
1555 | /* You are free to depend on diskless, non-existing, |
1556 | * or not yet/no longer existing minors. | |
1557 | * We only reject dependency loops. | |
1558 | * We cannot follow the dependency chain beyond a detached or | |
1559 | * missing minor. | |
1560 | */ | |
1561 | if (!odev || !odev->ldev || odev->state.disk == D_DISKLESS) | |
1562 | return NO_ERROR; | |
1563 | ||
daeda1cc | 1564 | rcu_read_lock(); |
95f8efd0 | 1565 | resync_after = rcu_dereference(odev->ldev->disk_conf)->resync_after; |
daeda1cc | 1566 | rcu_read_unlock(); |
b411b363 | 1567 | /* dependency chain ends here, no cycles. */ |
95f8efd0 | 1568 | if (resync_after == -1) |
b411b363 PR |
1569 | return NO_ERROR; |
1570 | ||
1571 | /* follow the dependency chain */ | |
b30ab791 | 1572 | odev = minor_to_device(resync_after); |
b411b363 PR |
1573 | } |
1574 | } | |
1575 | ||
dc97b708 | 1576 | /* caller must hold global_state_lock */ |
b30ab791 | 1577 | void drbd_resync_after_changed(struct drbd_device *device) |
b411b363 PR |
1578 | { |
1579 | int changes; | |
b411b363 | 1580 | |
dc97b708 | 1581 | do { |
b30ab791 AG |
1582 | changes = _drbd_pause_after(device); |
1583 | changes |= _drbd_resume_next(device); | |
dc97b708 | 1584 | } while (changes); |
b411b363 PR |
1585 | } |
1586 | ||
b30ab791 | 1587 | void drbd_rs_controller_reset(struct drbd_device *device) |
9bd28d3c | 1588 | { |
813472ce PR |
1589 | struct fifo_buffer *plan; |
1590 | ||
b30ab791 AG |
1591 | atomic_set(&device->rs_sect_in, 0); |
1592 | atomic_set(&device->rs_sect_ev, 0); | |
1593 | device->rs_in_flight = 0; | |
813472ce PR |
1594 | |
1595 | /* Updating the RCU protected object in place is necessary since | |
1596 | this function gets called from atomic context. | |
1597 | It is valid since all other updates also lead to an completely | |
1598 | empty fifo */ | |
1599 | rcu_read_lock(); | |
b30ab791 | 1600 | plan = rcu_dereference(device->rs_plan_s); |
813472ce PR |
1601 | plan->total = 0; |
1602 | fifo_set(plan, 0); | |
1603 | rcu_read_unlock(); | |
9bd28d3c LE |
1604 | } |
1605 | ||
1f04af33 PR |
1606 | void start_resync_timer_fn(unsigned long data) |
1607 | { | |
b30ab791 | 1608 | struct drbd_device *device = (struct drbd_device *) data; |
ac0acb9e | 1609 | drbd_device_post_work(device, RS_START); |
1f04af33 PR |
1610 | } |
1611 | ||
ac0acb9e | 1612 | static void do_start_resync(struct drbd_device *device) |
1f04af33 | 1613 | { |
b30ab791 | 1614 | if (atomic_read(&device->unacked_cnt) || atomic_read(&device->rs_pending_cnt)) { |
ac0acb9e | 1615 | drbd_warn(device, "postponing start_resync ...\n"); |
b30ab791 AG |
1616 | device->start_resync_timer.expires = jiffies + HZ/10; |
1617 | add_timer(&device->start_resync_timer); | |
ac0acb9e | 1618 | return; |
1f04af33 PR |
1619 | } |
1620 | ||
b30ab791 AG |
1621 | drbd_start_resync(device, C_SYNC_SOURCE); |
1622 | clear_bit(AHEAD_TO_SYNC_SOURCE, &device->flags); | |
1f04af33 PR |
1623 | } |
1624 | ||
b411b363 PR |
1625 | /** |
1626 | * drbd_start_resync() - Start the resync process | |
b30ab791 | 1627 | * @device: DRBD device. |
b411b363 PR |
1628 | * @side: Either C_SYNC_SOURCE or C_SYNC_TARGET |
1629 | * | |
1630 | * This function might bring you directly into one of the | |
1631 | * C_PAUSED_SYNC_* states. | |
1632 | */ | |
b30ab791 | 1633 | void drbd_start_resync(struct drbd_device *device, enum drbd_conns side) |
b411b363 | 1634 | { |
44a4d551 LE |
1635 | struct drbd_peer_device *peer_device = first_peer_device(device); |
1636 | struct drbd_connection *connection = peer_device ? peer_device->connection : NULL; | |
b411b363 PR |
1637 | union drbd_state ns; |
1638 | int r; | |
1639 | ||
b30ab791 | 1640 | if (device->state.conn >= C_SYNC_SOURCE && device->state.conn < C_AHEAD) { |
d0180171 | 1641 | drbd_err(device, "Resync already running!\n"); |
b411b363 PR |
1642 | return; |
1643 | } | |
1644 | ||
b30ab791 | 1645 | if (!test_bit(B_RS_H_DONE, &device->flags)) { |
e64a3294 PR |
1646 | if (side == C_SYNC_TARGET) { |
1647 | /* Since application IO was locked out during C_WF_BITMAP_T and | |
1648 | C_WF_SYNC_UUID we are still unmodified. Before going to C_SYNC_TARGET | |
1649 | we check that we might make the data inconsistent. */ | |
b30ab791 | 1650 | r = drbd_khelper(device, "before-resync-target"); |
e64a3294 PR |
1651 | r = (r >> 8) & 0xff; |
1652 | if (r > 0) { | |
d0180171 | 1653 | drbd_info(device, "before-resync-target handler returned %d, " |
09b9e797 | 1654 | "dropping connection.\n", r); |
44a4d551 | 1655 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
09b9e797 PR |
1656 | return; |
1657 | } | |
e64a3294 | 1658 | } else /* C_SYNC_SOURCE */ { |
b30ab791 | 1659 | r = drbd_khelper(device, "before-resync-source"); |
e64a3294 PR |
1660 | r = (r >> 8) & 0xff; |
1661 | if (r > 0) { | |
1662 | if (r == 3) { | |
d0180171 | 1663 | drbd_info(device, "before-resync-source handler returned %d, " |
e64a3294 PR |
1664 | "ignoring. Old userland tools?", r); |
1665 | } else { | |
d0180171 | 1666 | drbd_info(device, "before-resync-source handler returned %d, " |
e64a3294 | 1667 | "dropping connection.\n", r); |
44a4d551 | 1668 | conn_request_state(connection, |
a6b32bc3 | 1669 | NS(conn, C_DISCONNECTING), CS_HARD); |
e64a3294 PR |
1670 | return; |
1671 | } | |
1672 | } | |
09b9e797 | 1673 | } |
b411b363 PR |
1674 | } |
1675 | ||
44a4d551 | 1676 | if (current == connection->worker.task) { |
dad20554 | 1677 | /* The worker should not sleep waiting for state_mutex, |
e64a3294 | 1678 | that can take long */ |
b30ab791 AG |
1679 | if (!mutex_trylock(device->state_mutex)) { |
1680 | set_bit(B_RS_H_DONE, &device->flags); | |
1681 | device->start_resync_timer.expires = jiffies + HZ/5; | |
1682 | add_timer(&device->start_resync_timer); | |
e64a3294 PR |
1683 | return; |
1684 | } | |
1685 | } else { | |
b30ab791 | 1686 | mutex_lock(device->state_mutex); |
e64a3294 | 1687 | } |
b30ab791 | 1688 | clear_bit(B_RS_H_DONE, &device->flags); |
b411b363 | 1689 | |
074f4afe LE |
1690 | /* req_lock: serialize with drbd_send_and_submit() and others |
1691 | * global_state_lock: for stable sync-after dependencies */ | |
1692 | spin_lock_irq(&device->resource->req_lock); | |
1693 | write_lock(&global_state_lock); | |
a700471b | 1694 | /* Did some connection breakage or IO error race with us? */ |
b30ab791 AG |
1695 | if (device->state.conn < C_CONNECTED |
1696 | || !get_ldev_if_state(device, D_NEGOTIATING)) { | |
074f4afe LE |
1697 | write_unlock(&global_state_lock); |
1698 | spin_unlock_irq(&device->resource->req_lock); | |
b30ab791 | 1699 | mutex_unlock(device->state_mutex); |
b411b363 PR |
1700 | return; |
1701 | } | |
1702 | ||
b30ab791 | 1703 | ns = drbd_read_state(device); |
b411b363 | 1704 | |
b30ab791 | 1705 | ns.aftr_isp = !_drbd_may_sync_now(device); |
b411b363 PR |
1706 | |
1707 | ns.conn = side; | |
1708 | ||
1709 | if (side == C_SYNC_TARGET) | |
1710 | ns.disk = D_INCONSISTENT; | |
1711 | else /* side == C_SYNC_SOURCE */ | |
1712 | ns.pdsk = D_INCONSISTENT; | |
1713 | ||
b30ab791 AG |
1714 | r = __drbd_set_state(device, ns, CS_VERBOSE, NULL); |
1715 | ns = drbd_read_state(device); | |
b411b363 PR |
1716 | |
1717 | if (ns.conn < C_CONNECTED) | |
1718 | r = SS_UNKNOWN_ERROR; | |
1719 | ||
1720 | if (r == SS_SUCCESS) { | |
b30ab791 | 1721 | unsigned long tw = drbd_bm_total_weight(device); |
1d7734a0 LE |
1722 | unsigned long now = jiffies; |
1723 | int i; | |
1724 | ||
b30ab791 AG |
1725 | device->rs_failed = 0; |
1726 | device->rs_paused = 0; | |
1727 | device->rs_same_csum = 0; | |
1728 | device->rs_last_events = 0; | |
1729 | device->rs_last_sect_ev = 0; | |
1730 | device->rs_total = tw; | |
1731 | device->rs_start = now; | |
1d7734a0 | 1732 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
1733 | device->rs_mark_left[i] = tw; |
1734 | device->rs_mark_time[i] = now; | |
1d7734a0 | 1735 | } |
b30ab791 | 1736 | _drbd_pause_after(device); |
5ab7d2c0 LE |
1737 | /* Forget potentially stale cached per resync extent bit-counts. |
1738 | * Open coded drbd_rs_cancel_all(device), we already have IRQs | |
1739 | * disabled, and know the disk state is ok. */ | |
1740 | spin_lock(&device->al_lock); | |
1741 | lc_reset(device->resync); | |
1742 | device->resync_locked = 0; | |
1743 | device->resync_wenr = LC_FREE; | |
1744 | spin_unlock(&device->al_lock); | |
b411b363 | 1745 | } |
074f4afe LE |
1746 | write_unlock(&global_state_lock); |
1747 | spin_unlock_irq(&device->resource->req_lock); | |
5a22db89 | 1748 | |
b411b363 | 1749 | if (r == SS_SUCCESS) { |
5ab7d2c0 | 1750 | wake_up(&device->al_wait); /* for lc_reset() above */ |
328e0f12 PR |
1751 | /* reset rs_last_bcast when a resync or verify is started, |
1752 | * to deal with potential jiffies wrap. */ | |
b30ab791 | 1753 | device->rs_last_bcast = jiffies - HZ; |
328e0f12 | 1754 | |
d0180171 | 1755 | drbd_info(device, "Began resync as %s (will sync %lu KB [%lu bits set]).\n", |
b411b363 | 1756 | drbd_conn_str(ns.conn), |
b30ab791 AG |
1757 | (unsigned long) device->rs_total << (BM_BLOCK_SHIFT-10), |
1758 | (unsigned long) device->rs_total); | |
6c922ed5 | 1759 | if (side == C_SYNC_TARGET) |
b30ab791 | 1760 | device->bm_resync_fo = 0; |
6c922ed5 LE |
1761 | |
1762 | /* Since protocol 96, we must serialize drbd_gen_and_send_sync_uuid | |
1763 | * with w_send_oos, or the sync target will get confused as to | |
1764 | * how much bits to resync. We cannot do that always, because for an | |
1765 | * empty resync and protocol < 95, we need to do it here, as we call | |
1766 | * drbd_resync_finished from here in that case. | |
1767 | * We drbd_gen_and_send_sync_uuid here for protocol < 96, | |
1768 | * and from after_state_ch otherwise. */ | |
44a4d551 LE |
1769 | if (side == C_SYNC_SOURCE && connection->agreed_pro_version < 96) |
1770 | drbd_gen_and_send_sync_uuid(peer_device); | |
b411b363 | 1771 | |
44a4d551 | 1772 | if (connection->agreed_pro_version < 95 && device->rs_total == 0) { |
af85e8e8 LE |
1773 | /* This still has a race (about when exactly the peers |
1774 | * detect connection loss) that can lead to a full sync | |
1775 | * on next handshake. In 8.3.9 we fixed this with explicit | |
1776 | * resync-finished notifications, but the fix | |
1777 | * introduces a protocol change. Sleeping for some | |
1778 | * time longer than the ping interval + timeout on the | |
1779 | * SyncSource, to give the SyncTarget the chance to | |
1780 | * detect connection loss, then waiting for a ping | |
1781 | * response (implicit in drbd_resync_finished) reduces | |
1782 | * the race considerably, but does not solve it. */ | |
44ed167d PR |
1783 | if (side == C_SYNC_SOURCE) { |
1784 | struct net_conf *nc; | |
1785 | int timeo; | |
1786 | ||
1787 | rcu_read_lock(); | |
44a4d551 | 1788 | nc = rcu_dereference(connection->net_conf); |
44ed167d PR |
1789 | timeo = nc->ping_int * HZ + nc->ping_timeo * HZ / 9; |
1790 | rcu_read_unlock(); | |
1791 | schedule_timeout_interruptible(timeo); | |
1792 | } | |
b30ab791 | 1793 | drbd_resync_finished(device); |
b411b363 PR |
1794 | } |
1795 | ||
b30ab791 AG |
1796 | drbd_rs_controller_reset(device); |
1797 | /* ns.conn may already be != device->state.conn, | |
b411b363 PR |
1798 | * we may have been paused in between, or become paused until |
1799 | * the timer triggers. | |
1800 | * No matter, that is handled in resync_timer_fn() */ | |
1801 | if (ns.conn == C_SYNC_TARGET) | |
b30ab791 | 1802 | mod_timer(&device->resync_timer, jiffies); |
b411b363 | 1803 | |
b30ab791 | 1804 | drbd_md_sync(device); |
b411b363 | 1805 | } |
b30ab791 AG |
1806 | put_ldev(device); |
1807 | mutex_unlock(device->state_mutex); | |
b411b363 PR |
1808 | } |
1809 | ||
e334f550 | 1810 | static void update_on_disk_bitmap(struct drbd_device *device, bool resync_done) |
c7a58db4 LE |
1811 | { |
1812 | struct sib_info sib = { .sib_reason = SIB_SYNC_PROGRESS, }; | |
1813 | device->rs_last_bcast = jiffies; | |
1814 | ||
1815 | if (!get_ldev(device)) | |
1816 | return; | |
1817 | ||
1818 | drbd_bm_write_lazy(device, 0); | |
5ab7d2c0 | 1819 | if (resync_done && is_sync_state(device->state.conn)) |
c7a58db4 | 1820 | drbd_resync_finished(device); |
5ab7d2c0 | 1821 | |
c7a58db4 LE |
1822 | drbd_bcast_event(device, &sib); |
1823 | /* update timestamp, in case it took a while to write out stuff */ | |
1824 | device->rs_last_bcast = jiffies; | |
1825 | put_ldev(device); | |
1826 | } | |
1827 | ||
e334f550 LE |
1828 | static void drbd_ldev_destroy(struct drbd_device *device) |
1829 | { | |
1830 | lc_destroy(device->resync); | |
1831 | device->resync = NULL; | |
1832 | lc_destroy(device->act_log); | |
1833 | device->act_log = NULL; | |
1834 | __no_warn(local, | |
1835 | drbd_free_ldev(device->ldev); | |
1836 | device->ldev = NULL;); | |
1837 | clear_bit(GOING_DISKLESS, &device->flags); | |
1838 | wake_up(&device->misc_wait); | |
1839 | } | |
1840 | ||
1841 | static void go_diskless(struct drbd_device *device) | |
1842 | { | |
1843 | D_ASSERT(device, device->state.disk == D_FAILED); | |
1844 | /* we cannot assert local_cnt == 0 here, as get_ldev_if_state will | |
1845 | * inc/dec it frequently. Once we are D_DISKLESS, no one will touch | |
1846 | * the protected members anymore, though, so once put_ldev reaches zero | |
1847 | * again, it will be safe to free them. */ | |
1848 | ||
1849 | /* Try to write changed bitmap pages, read errors may have just | |
1850 | * set some bits outside the area covered by the activity log. | |
1851 | * | |
1852 | * If we have an IO error during the bitmap writeout, | |
1853 | * we will want a full sync next time, just in case. | |
1854 | * (Do we want a specific meta data flag for this?) | |
1855 | * | |
1856 | * If that does not make it to stable storage either, | |
1857 | * we cannot do anything about that anymore. | |
1858 | * | |
1859 | * We still need to check if both bitmap and ldev are present, we may | |
1860 | * end up here after a failed attach, before ldev was even assigned. | |
1861 | */ | |
1862 | if (device->bitmap && device->ldev) { | |
1863 | /* An interrupted resync or similar is allowed to recounts bits | |
1864 | * while we detach. | |
1865 | * Any modifications would not be expected anymore, though. | |
1866 | */ | |
1867 | if (drbd_bitmap_io_from_worker(device, drbd_bm_write, | |
1868 | "detach", BM_LOCKED_TEST_ALLOWED)) { | |
1869 | if (test_bit(WAS_READ_ERROR, &device->flags)) { | |
1870 | drbd_md_set_flag(device, MDF_FULL_SYNC); | |
1871 | drbd_md_sync(device); | |
1872 | } | |
1873 | } | |
1874 | } | |
1875 | ||
1876 | drbd_force_state(device, NS(disk, D_DISKLESS)); | |
1877 | } | |
1878 | ||
ac0acb9e LE |
1879 | static int do_md_sync(struct drbd_device *device) |
1880 | { | |
1881 | drbd_warn(device, "md_sync_timer expired! Worker calls drbd_md_sync().\n"); | |
1882 | drbd_md_sync(device); | |
1883 | return 0; | |
1884 | } | |
1885 | ||
e334f550 LE |
1886 | #define WORK_PENDING(work_bit, todo) (todo & (1UL << work_bit)) |
1887 | static void do_device_work(struct drbd_device *device, const unsigned long todo) | |
1888 | { | |
ac0acb9e LE |
1889 | if (WORK_PENDING(MD_SYNC, todo)) |
1890 | do_md_sync(device); | |
e334f550 LE |
1891 | if (WORK_PENDING(RS_DONE, todo) || |
1892 | WORK_PENDING(RS_PROGRESS, todo)) | |
1893 | update_on_disk_bitmap(device, WORK_PENDING(RS_DONE, todo)); | |
1894 | if (WORK_PENDING(GO_DISKLESS, todo)) | |
1895 | go_diskless(device); | |
1896 | if (WORK_PENDING(DESTROY_DISK, todo)) | |
1897 | drbd_ldev_destroy(device); | |
ac0acb9e LE |
1898 | if (WORK_PENDING(RS_START, todo)) |
1899 | do_start_resync(device); | |
e334f550 LE |
1900 | } |
1901 | ||
1902 | #define DRBD_DEVICE_WORK_MASK \ | |
1903 | ((1UL << GO_DISKLESS) \ | |
1904 | |(1UL << DESTROY_DISK) \ | |
ac0acb9e LE |
1905 | |(1UL << MD_SYNC) \ |
1906 | |(1UL << RS_START) \ | |
e334f550 LE |
1907 | |(1UL << RS_PROGRESS) \ |
1908 | |(1UL << RS_DONE) \ | |
1909 | ) | |
1910 | ||
1911 | static unsigned long get_work_bits(unsigned long *flags) | |
1912 | { | |
1913 | unsigned long old, new; | |
1914 | do { | |
1915 | old = *flags; | |
1916 | new = old & ~DRBD_DEVICE_WORK_MASK; | |
1917 | } while (cmpxchg(flags, old, new) != old); | |
1918 | return old & DRBD_DEVICE_WORK_MASK; | |
1919 | } | |
1920 | ||
1921 | static void do_unqueued_work(struct drbd_connection *connection) | |
c7a58db4 LE |
1922 | { |
1923 | struct drbd_peer_device *peer_device; | |
1924 | int vnr; | |
1925 | ||
1926 | rcu_read_lock(); | |
1927 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
1928 | struct drbd_device *device = peer_device->device; | |
e334f550 LE |
1929 | unsigned long todo = get_work_bits(&device->flags); |
1930 | if (!todo) | |
c7a58db4 | 1931 | continue; |
5ab7d2c0 | 1932 | |
c7a58db4 LE |
1933 | kref_get(&device->kref); |
1934 | rcu_read_unlock(); | |
e334f550 | 1935 | do_device_work(device, todo); |
c7a58db4 LE |
1936 | kref_put(&device->kref, drbd_destroy_device); |
1937 | rcu_read_lock(); | |
1938 | } | |
1939 | rcu_read_unlock(); | |
1940 | } | |
1941 | ||
a186e478 | 1942 | static bool dequeue_work_batch(struct drbd_work_queue *queue, struct list_head *work_list) |
8c0785a5 LE |
1943 | { |
1944 | spin_lock_irq(&queue->q_lock); | |
1945 | list_splice_init(&queue->q, work_list); | |
1946 | spin_unlock_irq(&queue->q_lock); | |
1947 | return !list_empty(work_list); | |
1948 | } | |
1949 | ||
a186e478 | 1950 | static bool dequeue_work_item(struct drbd_work_queue *queue, struct list_head *work_list) |
8c0785a5 LE |
1951 | { |
1952 | spin_lock_irq(&queue->q_lock); | |
1953 | if (!list_empty(&queue->q)) | |
1954 | list_move(queue->q.next, work_list); | |
1955 | spin_unlock_irq(&queue->q_lock); | |
1956 | return !list_empty(work_list); | |
1957 | } | |
1958 | ||
bde89a9e | 1959 | static void wait_for_work(struct drbd_connection *connection, struct list_head *work_list) |
b6dd1a89 LE |
1960 | { |
1961 | DEFINE_WAIT(wait); | |
1962 | struct net_conf *nc; | |
1963 | int uncork, cork; | |
1964 | ||
1965 | dequeue_work_item(&connection->sender_work, work_list); | |
1966 | if (!list_empty(work_list)) | |
1967 | return; | |
1968 | ||
1969 | /* Still nothing to do? | |
1970 | * Maybe we still need to close the current epoch, | |
1971 | * even if no new requests are queued yet. | |
1972 | * | |
1973 | * Also, poke TCP, just in case. | |
1974 | * Then wait for new work (or signal). */ | |
1975 | rcu_read_lock(); | |
1976 | nc = rcu_dereference(connection->net_conf); | |
1977 | uncork = nc ? nc->tcp_cork : 0; | |
1978 | rcu_read_unlock(); | |
1979 | if (uncork) { | |
1980 | mutex_lock(&connection->data.mutex); | |
1981 | if (connection->data.socket) | |
1982 | drbd_tcp_uncork(connection->data.socket); | |
1983 | mutex_unlock(&connection->data.mutex); | |
1984 | } | |
1985 | ||
1986 | for (;;) { | |
1987 | int send_barrier; | |
1988 | prepare_to_wait(&connection->sender_work.q_wait, &wait, TASK_INTERRUPTIBLE); | |
0500813f | 1989 | spin_lock_irq(&connection->resource->req_lock); |
b6dd1a89 | 1990 | spin_lock(&connection->sender_work.q_lock); /* FIXME get rid of this one? */ |
bc317a9e LE |
1991 | /* dequeue single item only, |
1992 | * we still use drbd_queue_work_front() in some places */ | |
1993 | if (!list_empty(&connection->sender_work.q)) | |
4dd726f0 | 1994 | list_splice_tail_init(&connection->sender_work.q, work_list); |
b6dd1a89 LE |
1995 | spin_unlock(&connection->sender_work.q_lock); /* FIXME get rid of this one? */ |
1996 | if (!list_empty(work_list) || signal_pending(current)) { | |
0500813f | 1997 | spin_unlock_irq(&connection->resource->req_lock); |
b6dd1a89 LE |
1998 | break; |
1999 | } | |
f9c78128 LE |
2000 | |
2001 | /* We found nothing new to do, no to-be-communicated request, | |
2002 | * no other work item. We may still need to close the last | |
2003 | * epoch. Next incoming request epoch will be connection -> | |
2004 | * current transfer log epoch number. If that is different | |
2005 | * from the epoch of the last request we communicated, it is | |
2006 | * safe to send the epoch separating barrier now. | |
2007 | */ | |
2008 | send_barrier = | |
2009 | atomic_read(&connection->current_tle_nr) != | |
2010 | connection->send.current_epoch_nr; | |
0500813f | 2011 | spin_unlock_irq(&connection->resource->req_lock); |
f9c78128 LE |
2012 | |
2013 | if (send_barrier) | |
2014 | maybe_send_barrier(connection, | |
2015 | connection->send.current_epoch_nr + 1); | |
5ab7d2c0 | 2016 | |
e334f550 | 2017 | if (test_bit(DEVICE_WORK_PENDING, &connection->flags)) |
5ab7d2c0 LE |
2018 | break; |
2019 | ||
a80ca1ae LE |
2020 | /* drbd_send() may have called flush_signals() */ |
2021 | if (get_t_state(&connection->worker) != RUNNING) | |
2022 | break; | |
5ab7d2c0 | 2023 | |
b6dd1a89 LE |
2024 | schedule(); |
2025 | /* may be woken up for other things but new work, too, | |
2026 | * e.g. if the current epoch got closed. | |
2027 | * In which case we send the barrier above. */ | |
2028 | } | |
2029 | finish_wait(&connection->sender_work.q_wait, &wait); | |
2030 | ||
2031 | /* someone may have changed the config while we have been waiting above. */ | |
2032 | rcu_read_lock(); | |
2033 | nc = rcu_dereference(connection->net_conf); | |
2034 | cork = nc ? nc->tcp_cork : 0; | |
2035 | rcu_read_unlock(); | |
2036 | mutex_lock(&connection->data.mutex); | |
2037 | if (connection->data.socket) { | |
2038 | if (cork) | |
2039 | drbd_tcp_cork(connection->data.socket); | |
2040 | else if (!uncork) | |
2041 | drbd_tcp_uncork(connection->data.socket); | |
2042 | } | |
2043 | mutex_unlock(&connection->data.mutex); | |
2044 | } | |
2045 | ||
b411b363 PR |
2046 | int drbd_worker(struct drbd_thread *thi) |
2047 | { | |
bde89a9e | 2048 | struct drbd_connection *connection = thi->connection; |
6db7e50a | 2049 | struct drbd_work *w = NULL; |
c06ece6b | 2050 | struct drbd_peer_device *peer_device; |
b411b363 | 2051 | LIST_HEAD(work_list); |
8c0785a5 | 2052 | int vnr; |
b411b363 | 2053 | |
e77a0a5c | 2054 | while (get_t_state(thi) == RUNNING) { |
80822284 | 2055 | drbd_thread_current_set_cpu(thi); |
b411b363 | 2056 | |
8c0785a5 | 2057 | if (list_empty(&work_list)) |
bde89a9e | 2058 | wait_for_work(connection, &work_list); |
b411b363 | 2059 | |
e334f550 LE |
2060 | if (test_and_clear_bit(DEVICE_WORK_PENDING, &connection->flags)) |
2061 | do_unqueued_work(connection); | |
5ab7d2c0 | 2062 | |
8c0785a5 | 2063 | if (signal_pending(current)) { |
b411b363 | 2064 | flush_signals(current); |
19393e10 | 2065 | if (get_t_state(thi) == RUNNING) { |
1ec861eb | 2066 | drbd_warn(connection, "Worker got an unexpected signal\n"); |
b411b363 | 2067 | continue; |
19393e10 | 2068 | } |
b411b363 PR |
2069 | break; |
2070 | } | |
2071 | ||
e77a0a5c | 2072 | if (get_t_state(thi) != RUNNING) |
b411b363 | 2073 | break; |
b411b363 | 2074 | |
8c0785a5 | 2075 | while (!list_empty(&work_list)) { |
6db7e50a AG |
2076 | w = list_first_entry(&work_list, struct drbd_work, list); |
2077 | list_del_init(&w->list); | |
2078 | if (w->cb(w, connection->cstate < C_WF_REPORT_PARAMS) == 0) | |
8c0785a5 | 2079 | continue; |
bde89a9e AG |
2080 | if (connection->cstate >= C_WF_REPORT_PARAMS) |
2081 | conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); | |
b411b363 PR |
2082 | } |
2083 | } | |
b411b363 | 2084 | |
8c0785a5 | 2085 | do { |
e334f550 LE |
2086 | if (test_and_clear_bit(DEVICE_WORK_PENDING, &connection->flags)) |
2087 | do_unqueued_work(connection); | |
b411b363 | 2088 | while (!list_empty(&work_list)) { |
6db7e50a AG |
2089 | w = list_first_entry(&work_list, struct drbd_work, list); |
2090 | list_del_init(&w->list); | |
2091 | w->cb(w, 1); | |
b411b363 | 2092 | } |
bde89a9e | 2093 | dequeue_work_batch(&connection->sender_work, &work_list); |
e334f550 | 2094 | } while (!list_empty(&work_list) || test_bit(DEVICE_WORK_PENDING, &connection->flags)); |
b411b363 | 2095 | |
c141ebda | 2096 | rcu_read_lock(); |
c06ece6b AG |
2097 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
2098 | struct drbd_device *device = peer_device->device; | |
0b0ba1ef | 2099 | D_ASSERT(device, device->state.disk == D_DISKLESS && device->state.conn == C_STANDALONE); |
b30ab791 | 2100 | kref_get(&device->kref); |
c141ebda | 2101 | rcu_read_unlock(); |
b30ab791 | 2102 | drbd_device_cleanup(device); |
05a10ec7 | 2103 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda | 2104 | rcu_read_lock(); |
0e29d163 | 2105 | } |
c141ebda | 2106 | rcu_read_unlock(); |
b411b363 PR |
2107 | |
2108 | return 0; | |
2109 | } |