]>
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 | |
84b8c06b | 42 | static int w_make_ov_request(struct drbd_work *, int); |
4d010392 | 43 | static int w_make_resync_request(struct drbd_work *, 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; | |
a8cd15ba | 103 | struct drbd_device *device = peer_req->peer_device->device; |
b411b363 | 104 | |
0500813f | 105 | spin_lock_irqsave(&device->resource->req_lock, flags); |
b30ab791 | 106 | device->read_cnt += peer_req->i.size >> 9; |
a8cd15ba | 107 | list_del(&peer_req->w.list); |
b30ab791 AG |
108 | if (list_empty(&device->read_ee)) |
109 | wake_up(&device->ee_wait); | |
db830c46 | 110 | if (test_bit(__EE_WAS_ERROR, &peer_req->flags)) |
b30ab791 | 111 | __drbd_chk_io_error(device, DRBD_READ_ERROR); |
0500813f | 112 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b411b363 | 113 | |
84b8c06b | 114 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, |
a8cd15ba | 115 | &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. */ |
db830c46 | 121 | static void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
122 | { |
123 | unsigned long flags = 0; | |
a8cd15ba | 124 | struct drbd_device *device = peer_req->peer_device->device; |
181286ad | 125 | struct drbd_interval i; |
b411b363 | 126 | int do_wake; |
579b57ed | 127 | u64 block_id; |
b411b363 | 128 | int do_al_complete_io; |
b411b363 | 129 | |
db830c46 | 130 | /* after we moved peer_req to done_ee, |
b411b363 PR |
131 | * we may no longer access it, |
132 | * it may be freed/reused already! | |
133 | * (as soon as we release the req_lock) */ | |
181286ad | 134 | i = peer_req->i; |
db830c46 AG |
135 | do_al_complete_io = peer_req->flags & EE_CALL_AL_COMPLETE_IO; |
136 | block_id = peer_req->block_id; | |
b411b363 | 137 | |
0500813f | 138 | spin_lock_irqsave(&device->resource->req_lock, flags); |
b30ab791 | 139 | device->writ_cnt += peer_req->i.size >> 9; |
a8cd15ba | 140 | list_move_tail(&peer_req->w.list, &device->done_ee); |
b411b363 | 141 | |
bb3bfe96 | 142 | /* |
5e472264 | 143 | * Do not remove from the write_requests tree here: we did not send the |
bb3bfe96 AG |
144 | * Ack yet and did not wake possibly waiting conflicting requests. |
145 | * Removed from the tree from "drbd_process_done_ee" within the | |
84b8c06b | 146 | * appropriate dw.cb (e_end_block/e_end_resync_block) or from |
bb3bfe96 AG |
147 | * _drbd_clear_done_ee. |
148 | */ | |
b411b363 | 149 | |
b30ab791 | 150 | do_wake = list_empty(block_id == ID_SYNCER ? &device->sync_ee : &device->active_ee); |
b411b363 | 151 | |
db830c46 | 152 | if (test_bit(__EE_WAS_ERROR, &peer_req->flags)) |
b30ab791 | 153 | __drbd_chk_io_error(device, DRBD_WRITE_ERROR); |
0500813f | 154 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b411b363 | 155 | |
579b57ed | 156 | if (block_id == ID_SYNCER) |
b30ab791 | 157 | drbd_rs_complete_io(device, i.sector); |
b411b363 PR |
158 | |
159 | if (do_wake) | |
b30ab791 | 160 | wake_up(&device->ee_wait); |
b411b363 PR |
161 | |
162 | if (do_al_complete_io) | |
b30ab791 | 163 | drbd_al_complete_io(device, &i); |
b411b363 | 164 | |
a6b32bc3 | 165 | wake_asender(first_peer_device(device)->connection); |
b30ab791 | 166 | put_ldev(device); |
45bb912b | 167 | } |
b411b363 | 168 | |
45bb912b LE |
169 | /* writes on behalf of the partner, or resync writes, |
170 | * "submitted" by the receiver. | |
171 | */ | |
fcefa62e | 172 | void drbd_peer_request_endio(struct bio *bio, int error) |
45bb912b | 173 | { |
db830c46 | 174 | struct drbd_peer_request *peer_req = bio->bi_private; |
a8cd15ba | 175 | struct drbd_device *device = peer_req->peer_device->device; |
45bb912b LE |
176 | int uptodate = bio_flagged(bio, BIO_UPTODATE); |
177 | int is_write = bio_data_dir(bio) == WRITE; | |
178 | ||
07194272 | 179 | if (error && __ratelimit(&drbd_ratelimit_state)) |
d0180171 | 180 | drbd_warn(device, "%s: error=%d s=%llus\n", |
45bb912b | 181 | is_write ? "write" : "read", error, |
db830c46 | 182 | (unsigned long long)peer_req->i.sector); |
45bb912b | 183 | if (!error && !uptodate) { |
07194272 | 184 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 185 | drbd_warn(device, "%s: setting error to -EIO s=%llus\n", |
07194272 | 186 | is_write ? "write" : "read", |
db830c46 | 187 | (unsigned long long)peer_req->i.sector); |
45bb912b LE |
188 | /* strange behavior of some lower level drivers... |
189 | * fail the request by clearing the uptodate flag, | |
190 | * but do not return any error?! */ | |
191 | error = -EIO; | |
192 | } | |
193 | ||
194 | if (error) | |
db830c46 | 195 | set_bit(__EE_WAS_ERROR, &peer_req->flags); |
45bb912b LE |
196 | |
197 | bio_put(bio); /* no need for the bio anymore */ | |
db830c46 | 198 | if (atomic_dec_and_test(&peer_req->pending_bios)) { |
45bb912b | 199 | if (is_write) |
db830c46 | 200 | drbd_endio_write_sec_final(peer_req); |
45bb912b | 201 | else |
db830c46 | 202 | drbd_endio_read_sec_final(peer_req); |
45bb912b | 203 | } |
b411b363 PR |
204 | } |
205 | ||
206 | /* read, readA or write requests on R_PRIMARY coming from drbd_make_request | |
207 | */ | |
fcefa62e | 208 | void drbd_request_endio(struct bio *bio, int error) |
b411b363 | 209 | { |
a115413d | 210 | unsigned long flags; |
b411b363 | 211 | struct drbd_request *req = bio->bi_private; |
84b8c06b | 212 | struct drbd_device *device = req->device; |
a115413d | 213 | struct bio_and_error m; |
b411b363 PR |
214 | enum drbd_req_event what; |
215 | int uptodate = bio_flagged(bio, BIO_UPTODATE); | |
216 | ||
b411b363 | 217 | if (!error && !uptodate) { |
d0180171 | 218 | drbd_warn(device, "p %s: setting error to -EIO\n", |
b411b363 PR |
219 | bio_data_dir(bio) == WRITE ? "write" : "read"); |
220 | /* strange behavior of some lower level drivers... | |
221 | * fail the request by clearing the uptodate flag, | |
222 | * but do not return any error?! */ | |
223 | error = -EIO; | |
224 | } | |
225 | ||
1b6dd252 PR |
226 | |
227 | /* If this request was aborted locally before, | |
228 | * but now was completed "successfully", | |
229 | * chances are that this caused arbitrary data corruption. | |
230 | * | |
231 | * "aborting" requests, or force-detaching the disk, is intended for | |
232 | * completely blocked/hung local backing devices which do no longer | |
233 | * complete requests at all, not even do error completions. In this | |
234 | * situation, usually a hard-reset and failover is the only way out. | |
235 | * | |
236 | * By "aborting", basically faking a local error-completion, | |
237 | * we allow for a more graceful swichover by cleanly migrating services. | |
238 | * Still the affected node has to be rebooted "soon". | |
239 | * | |
240 | * By completing these requests, we allow the upper layers to re-use | |
241 | * the associated data pages. | |
242 | * | |
243 | * If later the local backing device "recovers", and now DMAs some data | |
244 | * from disk into the original request pages, in the best case it will | |
245 | * just put random data into unused pages; but typically it will corrupt | |
246 | * meanwhile completely unrelated data, causing all sorts of damage. | |
247 | * | |
248 | * Which means delayed successful completion, | |
249 | * especially for READ requests, | |
250 | * is a reason to panic(). | |
251 | * | |
252 | * We assume that a delayed *error* completion is OK, | |
253 | * though we still will complain noisily about it. | |
254 | */ | |
255 | if (unlikely(req->rq_state & RQ_LOCAL_ABORTED)) { | |
256 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 257 | drbd_emerg(device, "delayed completion of aborted local request; disk-timeout may be too aggressive\n"); |
1b6dd252 PR |
258 | |
259 | if (!error) | |
260 | panic("possible random memory corruption caused by delayed completion of aborted local request\n"); | |
261 | } | |
262 | ||
b411b363 PR |
263 | /* to avoid recursion in __req_mod */ |
264 | if (unlikely(error)) { | |
265 | what = (bio_data_dir(bio) == WRITE) | |
8554df1c | 266 | ? WRITE_COMPLETED_WITH_ERROR |
5c3c7e64 | 267 | : (bio_rw(bio) == READ) |
8554df1c AG |
268 | ? READ_COMPLETED_WITH_ERROR |
269 | : READ_AHEAD_COMPLETED_WITH_ERROR; | |
b411b363 | 270 | } else |
8554df1c | 271 | what = COMPLETED_OK; |
b411b363 PR |
272 | |
273 | bio_put(req->private_bio); | |
274 | req->private_bio = ERR_PTR(error); | |
275 | ||
a115413d | 276 | /* not req_mod(), we need irqsave here! */ |
0500813f | 277 | spin_lock_irqsave(&device->resource->req_lock, flags); |
a115413d | 278 | __req_mod(req, what, &m); |
0500813f | 279 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b30ab791 | 280 | put_ldev(device); |
a115413d LE |
281 | |
282 | if (m.bio) | |
b30ab791 | 283 | complete_master_bio(device, &m); |
b411b363 PR |
284 | } |
285 | ||
79a3c8d3 | 286 | void drbd_csum_ee(struct crypto_hash *tfm, struct drbd_peer_request *peer_req, void *digest) |
45bb912b LE |
287 | { |
288 | struct hash_desc desc; | |
289 | struct scatterlist sg; | |
db830c46 | 290 | struct page *page = peer_req->pages; |
45bb912b LE |
291 | struct page *tmp; |
292 | unsigned len; | |
293 | ||
294 | desc.tfm = tfm; | |
295 | desc.flags = 0; | |
296 | ||
297 | sg_init_table(&sg, 1); | |
298 | crypto_hash_init(&desc); | |
299 | ||
300 | while ((tmp = page_chain_next(page))) { | |
301 | /* all but the last page will be fully used */ | |
302 | sg_set_page(&sg, page, PAGE_SIZE, 0); | |
303 | crypto_hash_update(&desc, &sg, sg.length); | |
304 | page = tmp; | |
305 | } | |
306 | /* and now the last, possibly only partially used page */ | |
db830c46 | 307 | len = peer_req->i.size & (PAGE_SIZE - 1); |
45bb912b LE |
308 | sg_set_page(&sg, page, len ?: PAGE_SIZE, 0); |
309 | crypto_hash_update(&desc, &sg, sg.length); | |
310 | crypto_hash_final(&desc, digest); | |
311 | } | |
312 | ||
79a3c8d3 | 313 | void drbd_csum_bio(struct crypto_hash *tfm, struct bio *bio, void *digest) |
b411b363 PR |
314 | { |
315 | struct hash_desc desc; | |
316 | struct scatterlist sg; | |
7988613b KO |
317 | struct bio_vec bvec; |
318 | struct bvec_iter iter; | |
b411b363 PR |
319 | |
320 | desc.tfm = tfm; | |
321 | desc.flags = 0; | |
322 | ||
323 | sg_init_table(&sg, 1); | |
324 | crypto_hash_init(&desc); | |
325 | ||
7988613b KO |
326 | bio_for_each_segment(bvec, bio, iter) { |
327 | sg_set_page(&sg, bvec.bv_page, bvec.bv_len, bvec.bv_offset); | |
b411b363 PR |
328 | crypto_hash_update(&desc, &sg, sg.length); |
329 | } | |
330 | crypto_hash_final(&desc, digest); | |
331 | } | |
332 | ||
9676c760 | 333 | /* MAYBE merge common code with w_e_end_ov_req */ |
99920dc5 | 334 | static int w_e_send_csum(struct drbd_work *w, int cancel) |
b411b363 | 335 | { |
a8cd15ba AG |
336 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
337 | struct drbd_device *device = peer_req->peer_device->device; | |
b411b363 PR |
338 | int digest_size; |
339 | void *digest; | |
99920dc5 | 340 | int err = 0; |
b411b363 | 341 | |
53ea4331 LE |
342 | if (unlikely(cancel)) |
343 | goto out; | |
b411b363 | 344 | |
9676c760 | 345 | if (unlikely((peer_req->flags & EE_WAS_ERROR) != 0)) |
53ea4331 | 346 | goto out; |
b411b363 | 347 | |
a6b32bc3 | 348 | digest_size = crypto_hash_digestsize(first_peer_device(device)->connection->csums_tfm); |
53ea4331 LE |
349 | digest = kmalloc(digest_size, GFP_NOIO); |
350 | if (digest) { | |
db830c46 AG |
351 | sector_t sector = peer_req->i.sector; |
352 | unsigned int size = peer_req->i.size; | |
79a3c8d3 | 353 | drbd_csum_ee(first_peer_device(device)->connection->csums_tfm, peer_req, digest); |
9676c760 | 354 | /* Free peer_req and pages before send. |
53ea4331 LE |
355 | * In case we block on congestion, we could otherwise run into |
356 | * some distributed deadlock, if the other side blocks on | |
357 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 358 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 359 | drbd_free_peer_req(device, peer_req); |
db830c46 | 360 | peer_req = NULL; |
b30ab791 | 361 | inc_rs_pending(device); |
69a22773 | 362 | err = drbd_send_drequest_csum(first_peer_device(device), sector, size, |
db1b0b72 AG |
363 | digest, digest_size, |
364 | P_CSUM_RS_REQUEST); | |
53ea4331 LE |
365 | kfree(digest); |
366 | } else { | |
d0180171 | 367 | drbd_err(device, "kmalloc() of digest failed.\n"); |
99920dc5 | 368 | err = -ENOMEM; |
53ea4331 | 369 | } |
b411b363 | 370 | |
53ea4331 | 371 | out: |
db830c46 | 372 | if (peer_req) |
b30ab791 | 373 | drbd_free_peer_req(device, peer_req); |
b411b363 | 374 | |
99920dc5 | 375 | if (unlikely(err)) |
d0180171 | 376 | drbd_err(device, "drbd_send_drequest(..., csum) failed\n"); |
99920dc5 | 377 | return err; |
b411b363 PR |
378 | } |
379 | ||
380 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
381 | ||
69a22773 | 382 | static int read_for_csum(struct drbd_peer_device *peer_device, sector_t sector, int size) |
b411b363 | 383 | { |
69a22773 | 384 | struct drbd_device *device = peer_device->device; |
db830c46 | 385 | struct drbd_peer_request *peer_req; |
b411b363 | 386 | |
b30ab791 | 387 | if (!get_ldev(device)) |
80a40e43 | 388 | return -EIO; |
b411b363 | 389 | |
b30ab791 | 390 | if (drbd_rs_should_slow_down(device, sector)) |
0f0601f4 LE |
391 | goto defer; |
392 | ||
b411b363 PR |
393 | /* GFP_TRY, because if there is no memory available right now, this may |
394 | * be rescheduled for later. It is "only" background resync, after all. */ | |
69a22773 | 395 | peer_req = drbd_alloc_peer_req(peer_device, ID_SYNCER /* unused */, sector, |
0db55363 | 396 | size, GFP_TRY); |
db830c46 | 397 | if (!peer_req) |
80a40e43 | 398 | goto defer; |
b411b363 | 399 | |
a8cd15ba | 400 | peer_req->w.cb = w_e_send_csum; |
0500813f | 401 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 402 | list_add(&peer_req->w.list, &device->read_ee); |
0500813f | 403 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 404 | |
b30ab791 AG |
405 | atomic_add(size >> 9, &device->rs_sect_ev); |
406 | if (drbd_submit_peer_request(device, peer_req, READ, DRBD_FAULT_RS_RD) == 0) | |
80a40e43 | 407 | return 0; |
b411b363 | 408 | |
10f6d992 LE |
409 | /* If it failed because of ENOMEM, retry should help. If it failed |
410 | * because bio_add_page failed (probably broken lower level driver), | |
411 | * retry may or may not help. | |
412 | * If it does not, you may need to force disconnect. */ | |
0500813f | 413 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 414 | list_del(&peer_req->w.list); |
0500813f | 415 | spin_unlock_irq(&device->resource->req_lock); |
22cc37a9 | 416 | |
b30ab791 | 417 | drbd_free_peer_req(device, peer_req); |
80a40e43 | 418 | defer: |
b30ab791 | 419 | put_ldev(device); |
80a40e43 | 420 | return -EAGAIN; |
b411b363 PR |
421 | } |
422 | ||
99920dc5 | 423 | int w_resync_timer(struct drbd_work *w, int cancel) |
b411b363 | 424 | { |
84b8c06b AG |
425 | struct drbd_device *device = |
426 | container_of(w, struct drbd_device, resync_work); | |
427 | ||
b30ab791 | 428 | switch (device->state.conn) { |
63106d3c | 429 | case C_VERIFY_S: |
00d56944 | 430 | w_make_ov_request(w, cancel); |
63106d3c PR |
431 | break; |
432 | case C_SYNC_TARGET: | |
00d56944 | 433 | w_make_resync_request(w, cancel); |
63106d3c | 434 | break; |
b411b363 PR |
435 | } |
436 | ||
99920dc5 | 437 | return 0; |
794abb75 PR |
438 | } |
439 | ||
440 | void resync_timer_fn(unsigned long data) | |
441 | { | |
b30ab791 | 442 | struct drbd_device *device = (struct drbd_device *) data; |
794abb75 | 443 | |
b30ab791 | 444 | if (list_empty(&device->resync_work.list)) |
84b8c06b AG |
445 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, |
446 | &device->resync_work); | |
b411b363 PR |
447 | } |
448 | ||
778f271d PR |
449 | static void fifo_set(struct fifo_buffer *fb, int value) |
450 | { | |
451 | int i; | |
452 | ||
453 | for (i = 0; i < fb->size; i++) | |
f10f2623 | 454 | fb->values[i] = value; |
778f271d PR |
455 | } |
456 | ||
457 | static int fifo_push(struct fifo_buffer *fb, int value) | |
458 | { | |
459 | int ov; | |
460 | ||
461 | ov = fb->values[fb->head_index]; | |
462 | fb->values[fb->head_index++] = value; | |
463 | ||
464 | if (fb->head_index >= fb->size) | |
465 | fb->head_index = 0; | |
466 | ||
467 | return ov; | |
468 | } | |
469 | ||
470 | static void fifo_add_val(struct fifo_buffer *fb, int value) | |
471 | { | |
472 | int i; | |
473 | ||
474 | for (i = 0; i < fb->size; i++) | |
475 | fb->values[i] += value; | |
476 | } | |
477 | ||
9958c857 PR |
478 | struct fifo_buffer *fifo_alloc(int fifo_size) |
479 | { | |
480 | struct fifo_buffer *fb; | |
481 | ||
8747d30a | 482 | fb = kzalloc(sizeof(struct fifo_buffer) + sizeof(int) * fifo_size, GFP_NOIO); |
9958c857 PR |
483 | if (!fb) |
484 | return NULL; | |
485 | ||
486 | fb->head_index = 0; | |
487 | fb->size = fifo_size; | |
488 | fb->total = 0; | |
489 | ||
490 | return fb; | |
491 | } | |
492 | ||
b30ab791 | 493 | static int drbd_rs_controller(struct drbd_device *device) |
778f271d | 494 | { |
daeda1cc | 495 | struct disk_conf *dc; |
778f271d PR |
496 | unsigned int sect_in; /* Number of sectors that came in since the last turn */ |
497 | unsigned int want; /* The number of sectors we want in the proxy */ | |
498 | int req_sect; /* Number of sectors to request in this turn */ | |
499 | int correction; /* Number of sectors more we need in the proxy*/ | |
500 | int cps; /* correction per invocation of drbd_rs_controller() */ | |
501 | int steps; /* Number of time steps to plan ahead */ | |
502 | int curr_corr; | |
503 | int max_sect; | |
813472ce | 504 | struct fifo_buffer *plan; |
778f271d | 505 | |
b30ab791 AG |
506 | sect_in = atomic_xchg(&device->rs_sect_in, 0); /* Number of sectors that came in */ |
507 | device->rs_in_flight -= sect_in; | |
778f271d | 508 | |
b30ab791 AG |
509 | dc = rcu_dereference(device->ldev->disk_conf); |
510 | plan = rcu_dereference(device->rs_plan_s); | |
778f271d | 511 | |
813472ce | 512 | steps = plan->size; /* (dc->c_plan_ahead * 10 * SLEEP_TIME) / HZ; */ |
778f271d | 513 | |
b30ab791 | 514 | if (device->rs_in_flight + sect_in == 0) { /* At start of resync */ |
daeda1cc | 515 | want = ((dc->resync_rate * 2 * SLEEP_TIME) / HZ) * steps; |
778f271d | 516 | } else { /* normal path */ |
daeda1cc PR |
517 | want = dc->c_fill_target ? dc->c_fill_target : |
518 | sect_in * dc->c_delay_target * HZ / (SLEEP_TIME * 10); | |
778f271d PR |
519 | } |
520 | ||
b30ab791 | 521 | correction = want - device->rs_in_flight - plan->total; |
778f271d PR |
522 | |
523 | /* Plan ahead */ | |
524 | cps = correction / steps; | |
813472ce PR |
525 | fifo_add_val(plan, cps); |
526 | plan->total += cps * steps; | |
778f271d PR |
527 | |
528 | /* What we do in this step */ | |
813472ce PR |
529 | curr_corr = fifo_push(plan, 0); |
530 | plan->total -= curr_corr; | |
778f271d PR |
531 | |
532 | req_sect = sect_in + curr_corr; | |
533 | if (req_sect < 0) | |
534 | req_sect = 0; | |
535 | ||
daeda1cc | 536 | max_sect = (dc->c_max_rate * 2 * SLEEP_TIME) / HZ; |
778f271d PR |
537 | if (req_sect > max_sect) |
538 | req_sect = max_sect; | |
539 | ||
540 | /* | |
d0180171 | 541 | drbd_warn(device, "si=%u if=%d wa=%u co=%d st=%d cps=%d pl=%d cc=%d rs=%d\n", |
b30ab791 AG |
542 | sect_in, device->rs_in_flight, want, correction, |
543 | steps, cps, device->rs_planed, curr_corr, req_sect); | |
778f271d PR |
544 | */ |
545 | ||
546 | return req_sect; | |
547 | } | |
548 | ||
b30ab791 | 549 | static int drbd_rs_number_requests(struct drbd_device *device) |
e65f440d LE |
550 | { |
551 | int number; | |
813472ce PR |
552 | |
553 | rcu_read_lock(); | |
b30ab791 AG |
554 | if (rcu_dereference(device->rs_plan_s)->size) { |
555 | number = drbd_rs_controller(device) >> (BM_BLOCK_SHIFT - 9); | |
556 | device->c_sync_rate = number * HZ * (BM_BLOCK_SIZE / 1024) / SLEEP_TIME; | |
e65f440d | 557 | } else { |
b30ab791 AG |
558 | device->c_sync_rate = rcu_dereference(device->ldev->disk_conf)->resync_rate; |
559 | number = SLEEP_TIME * device->c_sync_rate / ((BM_BLOCK_SIZE / 1024) * HZ); | |
e65f440d | 560 | } |
813472ce | 561 | rcu_read_unlock(); |
e65f440d | 562 | |
e65f440d LE |
563 | /* ignore the amount of pending requests, the resync controller should |
564 | * throttle down to incoming reply rate soon enough anyways. */ | |
565 | return number; | |
566 | } | |
567 | ||
4d010392 | 568 | static int w_make_resync_request(struct drbd_work *w, int cancel) |
b411b363 | 569 | { |
84b8c06b AG |
570 | struct drbd_device_work *dw = device_work(w); |
571 | struct drbd_device *device = dw->device; | |
b411b363 PR |
572 | unsigned long bit; |
573 | sector_t sector; | |
b30ab791 | 574 | const sector_t capacity = drbd_get_capacity(device->this_bdev); |
1816a2b4 | 575 | int max_bio_size; |
e65f440d | 576 | int number, rollback_i, size; |
b411b363 | 577 | int align, queued, sndbuf; |
0f0601f4 | 578 | int i = 0; |
b411b363 PR |
579 | |
580 | if (unlikely(cancel)) | |
99920dc5 | 581 | return 0; |
b411b363 | 582 | |
b30ab791 | 583 | if (device->rs_total == 0) { |
af85e8e8 | 584 | /* empty resync? */ |
b30ab791 | 585 | drbd_resync_finished(device); |
99920dc5 | 586 | return 0; |
af85e8e8 LE |
587 | } |
588 | ||
b30ab791 AG |
589 | if (!get_ldev(device)) { |
590 | /* Since we only need to access device->rsync a | |
591 | get_ldev_if_state(device,D_FAILED) would be sufficient, but | |
b411b363 PR |
592 | to continue resync with a broken disk makes no sense at |
593 | all */ | |
d0180171 | 594 | drbd_err(device, "Disk broke down during resync!\n"); |
99920dc5 | 595 | return 0; |
b411b363 PR |
596 | } |
597 | ||
b30ab791 AG |
598 | max_bio_size = queue_max_hw_sectors(device->rq_queue) << 9; |
599 | number = drbd_rs_number_requests(device); | |
e65f440d | 600 | if (number == 0) |
0f0601f4 | 601 | goto requeue; |
b411b363 | 602 | |
b411b363 PR |
603 | for (i = 0; i < number; i++) { |
604 | /* Stop generating RS requests, when half of the send buffer is filled */ | |
a6b32bc3 AG |
605 | mutex_lock(&first_peer_device(device)->connection->data.mutex); |
606 | if (first_peer_device(device)->connection->data.socket) { | |
607 | queued = first_peer_device(device)->connection->data.socket->sk->sk_wmem_queued; | |
608 | sndbuf = first_peer_device(device)->connection->data.socket->sk->sk_sndbuf; | |
b411b363 PR |
609 | } else { |
610 | queued = 1; | |
611 | sndbuf = 0; | |
612 | } | |
a6b32bc3 | 613 | mutex_unlock(&first_peer_device(device)->connection->data.mutex); |
b411b363 PR |
614 | if (queued > sndbuf / 2) |
615 | goto requeue; | |
616 | ||
617 | next_sector: | |
618 | size = BM_BLOCK_SIZE; | |
b30ab791 | 619 | bit = drbd_bm_find_next(device, device->bm_resync_fo); |
b411b363 | 620 | |
4b0715f0 | 621 | if (bit == DRBD_END_OF_BITMAP) { |
b30ab791 AG |
622 | device->bm_resync_fo = drbd_bm_bits(device); |
623 | put_ldev(device); | |
99920dc5 | 624 | return 0; |
b411b363 PR |
625 | } |
626 | ||
627 | sector = BM_BIT_TO_SECT(bit); | |
628 | ||
b30ab791 AG |
629 | if (drbd_rs_should_slow_down(device, sector) || |
630 | drbd_try_rs_begin_io(device, sector)) { | |
631 | device->bm_resync_fo = bit; | |
b411b363 PR |
632 | goto requeue; |
633 | } | |
b30ab791 | 634 | device->bm_resync_fo = bit + 1; |
b411b363 | 635 | |
b30ab791 AG |
636 | if (unlikely(drbd_bm_test_bit(device, bit) == 0)) { |
637 | drbd_rs_complete_io(device, sector); | |
b411b363 PR |
638 | goto next_sector; |
639 | } | |
640 | ||
1816a2b4 | 641 | #if DRBD_MAX_BIO_SIZE > BM_BLOCK_SIZE |
b411b363 PR |
642 | /* try to find some adjacent bits. |
643 | * we stop if we have already the maximum req size. | |
644 | * | |
645 | * Additionally always align bigger requests, in order to | |
646 | * be prepared for all stripe sizes of software RAIDs. | |
b411b363 PR |
647 | */ |
648 | align = 1; | |
d207450c | 649 | rollback_i = i; |
b411b363 | 650 | for (;;) { |
1816a2b4 | 651 | if (size + BM_BLOCK_SIZE > max_bio_size) |
b411b363 PR |
652 | break; |
653 | ||
654 | /* Be always aligned */ | |
655 | if (sector & ((1<<(align+3))-1)) | |
656 | break; | |
657 | ||
658 | /* do not cross extent boundaries */ | |
659 | if (((bit+1) & BM_BLOCKS_PER_BM_EXT_MASK) == 0) | |
660 | break; | |
661 | /* now, is it actually dirty, after all? | |
662 | * caution, drbd_bm_test_bit is tri-state for some | |
663 | * obscure reason; ( b == 0 ) would get the out-of-band | |
664 | * only accidentally right because of the "oddly sized" | |
665 | * adjustment below */ | |
b30ab791 | 666 | if (drbd_bm_test_bit(device, bit+1) != 1) |
b411b363 PR |
667 | break; |
668 | bit++; | |
669 | size += BM_BLOCK_SIZE; | |
670 | if ((BM_BLOCK_SIZE << align) <= size) | |
671 | align++; | |
672 | i++; | |
673 | } | |
674 | /* if we merged some, | |
675 | * reset the offset to start the next drbd_bm_find_next from */ | |
676 | if (size > BM_BLOCK_SIZE) | |
b30ab791 | 677 | device->bm_resync_fo = bit + 1; |
b411b363 PR |
678 | #endif |
679 | ||
680 | /* adjust very last sectors, in case we are oddly sized */ | |
681 | if (sector + (size>>9) > capacity) | |
682 | size = (capacity-sector)<<9; | |
a6b32bc3 AG |
683 | if (first_peer_device(device)->connection->agreed_pro_version >= 89 && |
684 | first_peer_device(device)->connection->csums_tfm) { | |
69a22773 | 685 | switch (read_for_csum(first_peer_device(device), sector, size)) { |
80a40e43 | 686 | case -EIO: /* Disk failure */ |
b30ab791 | 687 | put_ldev(device); |
99920dc5 | 688 | return -EIO; |
80a40e43 | 689 | case -EAGAIN: /* allocation failed, or ldev busy */ |
b30ab791 AG |
690 | drbd_rs_complete_io(device, sector); |
691 | device->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
d207450c | 692 | i = rollback_i; |
b411b363 | 693 | goto requeue; |
80a40e43 LE |
694 | case 0: |
695 | /* everything ok */ | |
696 | break; | |
697 | default: | |
698 | BUG(); | |
b411b363 PR |
699 | } |
700 | } else { | |
99920dc5 AG |
701 | int err; |
702 | ||
b30ab791 | 703 | inc_rs_pending(device); |
69a22773 | 704 | err = drbd_send_drequest(first_peer_device(device), P_RS_DATA_REQUEST, |
99920dc5 AG |
705 | sector, size, ID_SYNCER); |
706 | if (err) { | |
d0180171 | 707 | drbd_err(device, "drbd_send_drequest() failed, aborting...\n"); |
b30ab791 AG |
708 | dec_rs_pending(device); |
709 | put_ldev(device); | |
99920dc5 | 710 | return err; |
b411b363 PR |
711 | } |
712 | } | |
713 | } | |
714 | ||
b30ab791 | 715 | if (device->bm_resync_fo >= drbd_bm_bits(device)) { |
b411b363 PR |
716 | /* last syncer _request_ was sent, |
717 | * but the P_RS_DATA_REPLY not yet received. sync will end (and | |
718 | * next sync group will resume), as soon as we receive the last | |
719 | * resync data block, and the last bit is cleared. | |
720 | * until then resync "work" is "inactive" ... | |
721 | */ | |
b30ab791 | 722 | put_ldev(device); |
99920dc5 | 723 | return 0; |
b411b363 PR |
724 | } |
725 | ||
726 | requeue: | |
b30ab791 AG |
727 | device->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
728 | mod_timer(&device->resync_timer, jiffies + SLEEP_TIME); | |
729 | put_ldev(device); | |
99920dc5 | 730 | return 0; |
b411b363 PR |
731 | } |
732 | ||
00d56944 | 733 | static int w_make_ov_request(struct drbd_work *w, int cancel) |
b411b363 | 734 | { |
84b8c06b | 735 | struct drbd_device *device = device_work(w)->device; |
b411b363 PR |
736 | int number, i, size; |
737 | sector_t sector; | |
b30ab791 | 738 | const sector_t capacity = drbd_get_capacity(device->this_bdev); |
58ffa580 | 739 | bool stop_sector_reached = false; |
b411b363 PR |
740 | |
741 | if (unlikely(cancel)) | |
742 | return 1; | |
743 | ||
b30ab791 | 744 | number = drbd_rs_number_requests(device); |
b411b363 | 745 | |
b30ab791 | 746 | sector = device->ov_position; |
b411b363 | 747 | for (i = 0; i < number; i++) { |
58ffa580 | 748 | if (sector >= capacity) |
b411b363 | 749 | return 1; |
58ffa580 LE |
750 | |
751 | /* We check for "finished" only in the reply path: | |
752 | * w_e_end_ov_reply(). | |
753 | * We need to send at least one request out. */ | |
754 | stop_sector_reached = i > 0 | |
b30ab791 AG |
755 | && verify_can_do_stop_sector(device) |
756 | && sector >= device->ov_stop_sector; | |
58ffa580 LE |
757 | if (stop_sector_reached) |
758 | break; | |
b411b363 PR |
759 | |
760 | size = BM_BLOCK_SIZE; | |
761 | ||
b30ab791 AG |
762 | if (drbd_rs_should_slow_down(device, sector) || |
763 | drbd_try_rs_begin_io(device, sector)) { | |
764 | device->ov_position = sector; | |
b411b363 PR |
765 | goto requeue; |
766 | } | |
767 | ||
768 | if (sector + (size>>9) > capacity) | |
769 | size = (capacity-sector)<<9; | |
770 | ||
b30ab791 | 771 | inc_rs_pending(device); |
69a22773 | 772 | if (drbd_send_ov_request(first_peer_device(device), sector, size)) { |
b30ab791 | 773 | dec_rs_pending(device); |
b411b363 PR |
774 | return 0; |
775 | } | |
776 | sector += BM_SECT_PER_BIT; | |
777 | } | |
b30ab791 | 778 | device->ov_position = sector; |
b411b363 PR |
779 | |
780 | requeue: | |
b30ab791 | 781 | device->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
58ffa580 | 782 | if (i == 0 || !stop_sector_reached) |
b30ab791 | 783 | mod_timer(&device->resync_timer, jiffies + SLEEP_TIME); |
b411b363 PR |
784 | return 1; |
785 | } | |
786 | ||
99920dc5 | 787 | int w_ov_finished(struct drbd_work *w, int cancel) |
b411b363 | 788 | { |
84b8c06b AG |
789 | struct drbd_device_work *dw = |
790 | container_of(w, struct drbd_device_work, w); | |
791 | struct drbd_device *device = dw->device; | |
792 | kfree(dw); | |
b30ab791 AG |
793 | ov_out_of_sync_print(device); |
794 | drbd_resync_finished(device); | |
b411b363 | 795 | |
99920dc5 | 796 | return 0; |
b411b363 PR |
797 | } |
798 | ||
99920dc5 | 799 | static int w_resync_finished(struct drbd_work *w, int cancel) |
b411b363 | 800 | { |
84b8c06b AG |
801 | struct drbd_device_work *dw = |
802 | container_of(w, struct drbd_device_work, w); | |
803 | struct drbd_device *device = dw->device; | |
804 | kfree(dw); | |
b411b363 | 805 | |
b30ab791 | 806 | drbd_resync_finished(device); |
b411b363 | 807 | |
99920dc5 | 808 | return 0; |
b411b363 PR |
809 | } |
810 | ||
b30ab791 | 811 | static void ping_peer(struct drbd_device *device) |
af85e8e8 | 812 | { |
a6b32bc3 | 813 | struct drbd_connection *connection = first_peer_device(device)->connection; |
2a67d8b9 | 814 | |
bde89a9e AG |
815 | clear_bit(GOT_PING_ACK, &connection->flags); |
816 | request_ping(connection); | |
817 | wait_event(connection->ping_wait, | |
818 | test_bit(GOT_PING_ACK, &connection->flags) || device->state.conn < C_CONNECTED); | |
af85e8e8 LE |
819 | } |
820 | ||
b30ab791 | 821 | int drbd_resync_finished(struct drbd_device *device) |
b411b363 PR |
822 | { |
823 | unsigned long db, dt, dbdt; | |
824 | unsigned long n_oos; | |
825 | union drbd_state os, ns; | |
84b8c06b | 826 | struct drbd_device_work *dw; |
b411b363 | 827 | char *khelper_cmd = NULL; |
26525618 | 828 | int verify_done = 0; |
b411b363 PR |
829 | |
830 | /* Remove all elements from the resync LRU. Since future actions | |
831 | * might set bits in the (main) bitmap, then the entries in the | |
832 | * resync LRU would be wrong. */ | |
b30ab791 | 833 | if (drbd_rs_del_all(device)) { |
b411b363 PR |
834 | /* In case this is not possible now, most probably because |
835 | * there are P_RS_DATA_REPLY Packets lingering on the worker's | |
836 | * queue (or even the read operations for those packets | |
837 | * is not finished by now). Retry in 100ms. */ | |
838 | ||
20ee6390 | 839 | schedule_timeout_interruptible(HZ / 10); |
84b8c06b AG |
840 | dw = kmalloc(sizeof(struct drbd_device_work), GFP_ATOMIC); |
841 | if (dw) { | |
842 | dw->w.cb = w_resync_finished; | |
843 | dw->device = device; | |
844 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, | |
845 | &dw->w); | |
b411b363 PR |
846 | return 1; |
847 | } | |
84b8c06b | 848 | drbd_err(device, "Warn failed to drbd_rs_del_all() and to kmalloc(dw).\n"); |
b411b363 PR |
849 | } |
850 | ||
b30ab791 | 851 | dt = (jiffies - device->rs_start - device->rs_paused) / HZ; |
b411b363 PR |
852 | if (dt <= 0) |
853 | dt = 1; | |
84b8c06b | 854 | |
b30ab791 | 855 | db = device->rs_total; |
58ffa580 | 856 | /* adjust for verify start and stop sectors, respective reached position */ |
b30ab791 AG |
857 | if (device->state.conn == C_VERIFY_S || device->state.conn == C_VERIFY_T) |
858 | db -= device->ov_left; | |
58ffa580 | 859 | |
b411b363 | 860 | dbdt = Bit2KB(db/dt); |
b30ab791 | 861 | device->rs_paused /= HZ; |
b411b363 | 862 | |
b30ab791 | 863 | if (!get_ldev(device)) |
b411b363 PR |
864 | goto out; |
865 | ||
b30ab791 | 866 | ping_peer(device); |
af85e8e8 | 867 | |
0500813f | 868 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 869 | os = drbd_read_state(device); |
b411b363 | 870 | |
26525618 LE |
871 | verify_done = (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T); |
872 | ||
b411b363 PR |
873 | /* This protects us against multiple calls (that can happen in the presence |
874 | of application IO), and against connectivity loss just before we arrive here. */ | |
875 | if (os.conn <= C_CONNECTED) | |
876 | goto out_unlock; | |
877 | ||
878 | ns = os; | |
879 | ns.conn = C_CONNECTED; | |
880 | ||
d0180171 | 881 | drbd_info(device, "%s done (total %lu sec; paused %lu sec; %lu K/sec)\n", |
58ffa580 | 882 | verify_done ? "Online verify" : "Resync", |
b30ab791 | 883 | dt + device->rs_paused, device->rs_paused, dbdt); |
b411b363 | 884 | |
b30ab791 | 885 | n_oos = drbd_bm_total_weight(device); |
b411b363 PR |
886 | |
887 | if (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) { | |
888 | if (n_oos) { | |
d0180171 | 889 | drbd_alert(device, "Online verify found %lu %dk block out of sync!\n", |
b411b363 PR |
890 | n_oos, Bit2KB(1)); |
891 | khelper_cmd = "out-of-sync"; | |
892 | } | |
893 | } else { | |
0b0ba1ef | 894 | D_ASSERT(device, (n_oos - device->rs_failed) == 0); |
b411b363 PR |
895 | |
896 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) | |
897 | khelper_cmd = "after-resync-target"; | |
898 | ||
a6b32bc3 | 899 | if (first_peer_device(device)->connection->csums_tfm && device->rs_total) { |
b30ab791 AG |
900 | const unsigned long s = device->rs_same_csum; |
901 | const unsigned long t = device->rs_total; | |
b411b363 PR |
902 | const int ratio = |
903 | (t == 0) ? 0 : | |
904 | (t < 100000) ? ((s*100)/t) : (s/(t/100)); | |
d0180171 | 905 | drbd_info(device, "%u %% had equal checksums, eliminated: %luK; " |
b411b363 PR |
906 | "transferred %luK total %luK\n", |
907 | ratio, | |
b30ab791 AG |
908 | Bit2KB(device->rs_same_csum), |
909 | Bit2KB(device->rs_total - device->rs_same_csum), | |
910 | Bit2KB(device->rs_total)); | |
b411b363 PR |
911 | } |
912 | } | |
913 | ||
b30ab791 | 914 | if (device->rs_failed) { |
d0180171 | 915 | drbd_info(device, " %lu failed blocks\n", device->rs_failed); |
b411b363 PR |
916 | |
917 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
918 | ns.disk = D_INCONSISTENT; | |
919 | ns.pdsk = D_UP_TO_DATE; | |
920 | } else { | |
921 | ns.disk = D_UP_TO_DATE; | |
922 | ns.pdsk = D_INCONSISTENT; | |
923 | } | |
924 | } else { | |
925 | ns.disk = D_UP_TO_DATE; | |
926 | ns.pdsk = D_UP_TO_DATE; | |
927 | ||
928 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
b30ab791 | 929 | if (device->p_uuid) { |
b411b363 PR |
930 | int i; |
931 | for (i = UI_BITMAP ; i <= UI_HISTORY_END ; i++) | |
b30ab791 AG |
932 | _drbd_uuid_set(device, i, device->p_uuid[i]); |
933 | drbd_uuid_set(device, UI_BITMAP, device->ldev->md.uuid[UI_CURRENT]); | |
934 | _drbd_uuid_set(device, UI_CURRENT, device->p_uuid[UI_CURRENT]); | |
b411b363 | 935 | } else { |
d0180171 | 936 | drbd_err(device, "device->p_uuid is NULL! BUG\n"); |
b411b363 PR |
937 | } |
938 | } | |
939 | ||
62b0da3a LE |
940 | if (!(os.conn == C_VERIFY_S || os.conn == C_VERIFY_T)) { |
941 | /* for verify runs, we don't update uuids here, | |
942 | * so there would be nothing to report. */ | |
b30ab791 AG |
943 | drbd_uuid_set_bm(device, 0UL); |
944 | drbd_print_uuids(device, "updated UUIDs"); | |
945 | if (device->p_uuid) { | |
62b0da3a LE |
946 | /* Now the two UUID sets are equal, update what we |
947 | * know of the peer. */ | |
948 | int i; | |
949 | for (i = UI_CURRENT ; i <= UI_HISTORY_END ; i++) | |
b30ab791 | 950 | device->p_uuid[i] = device->ldev->md.uuid[i]; |
62b0da3a | 951 | } |
b411b363 PR |
952 | } |
953 | } | |
954 | ||
b30ab791 | 955 | _drbd_set_state(device, ns, CS_VERBOSE, NULL); |
b411b363 | 956 | out_unlock: |
0500813f | 957 | spin_unlock_irq(&device->resource->req_lock); |
b30ab791 | 958 | put_ldev(device); |
b411b363 | 959 | out: |
b30ab791 AG |
960 | device->rs_total = 0; |
961 | device->rs_failed = 0; | |
962 | device->rs_paused = 0; | |
58ffa580 LE |
963 | |
964 | /* reset start sector, if we reached end of device */ | |
b30ab791 AG |
965 | if (verify_done && device->ov_left == 0) |
966 | device->ov_start_sector = 0; | |
b411b363 | 967 | |
b30ab791 | 968 | drbd_md_sync(device); |
13d42685 | 969 | |
b411b363 | 970 | if (khelper_cmd) |
b30ab791 | 971 | drbd_khelper(device, khelper_cmd); |
b411b363 PR |
972 | |
973 | return 1; | |
974 | } | |
975 | ||
976 | /* helper */ | |
b30ab791 | 977 | static void move_to_net_ee_or_free(struct drbd_device *device, struct drbd_peer_request *peer_req) |
b411b363 | 978 | { |
045417f7 | 979 | if (drbd_peer_req_has_active_page(peer_req)) { |
b411b363 | 980 | /* This might happen if sendpage() has not finished */ |
db830c46 | 981 | int i = (peer_req->i.size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b30ab791 AG |
982 | atomic_add(i, &device->pp_in_use_by_net); |
983 | atomic_sub(i, &device->pp_in_use); | |
0500813f | 984 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 985 | list_add_tail(&peer_req->w.list, &device->net_ee); |
0500813f | 986 | spin_unlock_irq(&device->resource->req_lock); |
435f0740 | 987 | wake_up(&drbd_pp_wait); |
b411b363 | 988 | } else |
b30ab791 | 989 | drbd_free_peer_req(device, peer_req); |
b411b363 PR |
990 | } |
991 | ||
992 | /** | |
993 | * w_e_end_data_req() - Worker callback, to send a P_DATA_REPLY packet in response to a P_DATA_REQUEST | |
b30ab791 | 994 | * @device: DRBD device. |
b411b363 PR |
995 | * @w: work object. |
996 | * @cancel: The connection will be closed anyways | |
997 | */ | |
99920dc5 | 998 | int w_e_end_data_req(struct drbd_work *w, int cancel) |
b411b363 | 999 | { |
a8cd15ba AG |
1000 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
1001 | struct drbd_device *device = peer_req->peer_device->device; | |
99920dc5 | 1002 | int err; |
b411b363 PR |
1003 | |
1004 | if (unlikely(cancel)) { | |
b30ab791 AG |
1005 | drbd_free_peer_req(device, peer_req); |
1006 | dec_unacked(device); | |
99920dc5 | 1007 | return 0; |
b411b363 PR |
1008 | } |
1009 | ||
db830c46 | 1010 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
69a22773 | 1011 | err = drbd_send_block(first_peer_device(device), P_DATA_REPLY, peer_req); |
b411b363 PR |
1012 | } else { |
1013 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1014 | drbd_err(device, "Sending NegDReply. sector=%llus.\n", |
db830c46 | 1015 | (unsigned long long)peer_req->i.sector); |
b411b363 | 1016 | |
69a22773 | 1017 | err = drbd_send_ack(first_peer_device(device), P_NEG_DREPLY, peer_req); |
b411b363 PR |
1018 | } |
1019 | ||
b30ab791 | 1020 | dec_unacked(device); |
b411b363 | 1021 | |
b30ab791 | 1022 | move_to_net_ee_or_free(device, peer_req); |
b411b363 | 1023 | |
99920dc5 | 1024 | if (unlikely(err)) |
d0180171 | 1025 | drbd_err(device, "drbd_send_block() failed\n"); |
99920dc5 | 1026 | return err; |
b411b363 PR |
1027 | } |
1028 | ||
1029 | /** | |
a209b4ae | 1030 | * 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 |
1031 | * @w: work object. |
1032 | * @cancel: The connection will be closed anyways | |
1033 | */ | |
99920dc5 | 1034 | int w_e_end_rsdata_req(struct drbd_work *w, int cancel) |
b411b363 | 1035 | { |
a8cd15ba AG |
1036 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
1037 | struct drbd_device *device = peer_req->peer_device->device; | |
99920dc5 | 1038 | int err; |
b411b363 PR |
1039 | |
1040 | if (unlikely(cancel)) { | |
b30ab791 AG |
1041 | drbd_free_peer_req(device, peer_req); |
1042 | dec_unacked(device); | |
99920dc5 | 1043 | return 0; |
b411b363 PR |
1044 | } |
1045 | ||
b30ab791 AG |
1046 | if (get_ldev_if_state(device, D_FAILED)) { |
1047 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1048 | put_ldev(device); | |
b411b363 PR |
1049 | } |
1050 | ||
b30ab791 | 1051 | if (device->state.conn == C_AHEAD) { |
69a22773 | 1052 | err = drbd_send_ack(first_peer_device(device), P_RS_CANCEL, peer_req); |
db830c46 | 1053 | } else if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b30ab791 AG |
1054 | if (likely(device->state.pdsk >= D_INCONSISTENT)) { |
1055 | inc_rs_pending(device); | |
69a22773 | 1056 | err = drbd_send_block(first_peer_device(device), P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1057 | } else { |
1058 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1059 | drbd_err(device, "Not sending RSDataReply, " |
b411b363 | 1060 | "partner DISKLESS!\n"); |
99920dc5 | 1061 | err = 0; |
b411b363 PR |
1062 | } |
1063 | } else { | |
1064 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1065 | drbd_err(device, "Sending NegRSDReply. sector %llus.\n", |
db830c46 | 1066 | (unsigned long long)peer_req->i.sector); |
b411b363 | 1067 | |
69a22773 | 1068 | err = drbd_send_ack(first_peer_device(device), P_NEG_RS_DREPLY, peer_req); |
b411b363 PR |
1069 | |
1070 | /* update resync data with failure */ | |
b30ab791 | 1071 | drbd_rs_failed_io(device, peer_req->i.sector, peer_req->i.size); |
b411b363 PR |
1072 | } |
1073 | ||
b30ab791 | 1074 | dec_unacked(device); |
b411b363 | 1075 | |
b30ab791 | 1076 | move_to_net_ee_or_free(device, peer_req); |
b411b363 | 1077 | |
99920dc5 | 1078 | if (unlikely(err)) |
d0180171 | 1079 | drbd_err(device, "drbd_send_block() failed\n"); |
99920dc5 | 1080 | return err; |
b411b363 PR |
1081 | } |
1082 | ||
99920dc5 | 1083 | int w_e_end_csum_rs_req(struct drbd_work *w, int cancel) |
b411b363 | 1084 | { |
a8cd15ba AG |
1085 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
1086 | struct drbd_device *device = peer_req->peer_device->device; | |
b411b363 PR |
1087 | struct digest_info *di; |
1088 | int digest_size; | |
1089 | void *digest = NULL; | |
99920dc5 | 1090 | int err, eq = 0; |
b411b363 PR |
1091 | |
1092 | if (unlikely(cancel)) { | |
b30ab791 AG |
1093 | drbd_free_peer_req(device, peer_req); |
1094 | dec_unacked(device); | |
99920dc5 | 1095 | return 0; |
b411b363 PR |
1096 | } |
1097 | ||
b30ab791 AG |
1098 | if (get_ldev(device)) { |
1099 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1100 | put_ldev(device); | |
1d53f09e | 1101 | } |
b411b363 | 1102 | |
db830c46 | 1103 | di = peer_req->digest; |
b411b363 | 1104 | |
db830c46 | 1105 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1106 | /* quick hack to try to avoid a race against reconfiguration. |
1107 | * a real fix would be much more involved, | |
1108 | * introducing more locking mechanisms */ | |
a6b32bc3 AG |
1109 | if (first_peer_device(device)->connection->csums_tfm) { |
1110 | digest_size = crypto_hash_digestsize(first_peer_device(device)->connection->csums_tfm); | |
0b0ba1ef | 1111 | D_ASSERT(device, digest_size == di->digest_size); |
b411b363 PR |
1112 | digest = kmalloc(digest_size, GFP_NOIO); |
1113 | } | |
1114 | if (digest) { | |
79a3c8d3 | 1115 | drbd_csum_ee(first_peer_device(device)->connection->csums_tfm, peer_req, digest); |
b411b363 PR |
1116 | eq = !memcmp(digest, di->digest, digest_size); |
1117 | kfree(digest); | |
1118 | } | |
1119 | ||
1120 | if (eq) { | |
b30ab791 | 1121 | drbd_set_in_sync(device, peer_req->i.sector, peer_req->i.size); |
676396d5 | 1122 | /* rs_same_csums unit is BM_BLOCK_SIZE */ |
b30ab791 | 1123 | device->rs_same_csum += peer_req->i.size >> BM_BLOCK_SHIFT; |
69a22773 | 1124 | err = drbd_send_ack(first_peer_device(device), P_RS_IS_IN_SYNC, peer_req); |
b411b363 | 1125 | } else { |
b30ab791 | 1126 | inc_rs_pending(device); |
db830c46 AG |
1127 | peer_req->block_id = ID_SYNCER; /* By setting block_id, digest pointer becomes invalid! */ |
1128 | peer_req->flags &= ~EE_HAS_DIGEST; /* This peer request no longer has a digest pointer */ | |
204bba99 | 1129 | kfree(di); |
69a22773 | 1130 | err = drbd_send_block(first_peer_device(device), P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1131 | } |
1132 | } else { | |
69a22773 | 1133 | err = drbd_send_ack(first_peer_device(device), P_NEG_RS_DREPLY, peer_req); |
b411b363 | 1134 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 1135 | drbd_err(device, "Sending NegDReply. I guess it gets messy.\n"); |
b411b363 PR |
1136 | } |
1137 | ||
b30ab791 AG |
1138 | dec_unacked(device); |
1139 | move_to_net_ee_or_free(device, peer_req); | |
b411b363 | 1140 | |
99920dc5 | 1141 | if (unlikely(err)) |
d0180171 | 1142 | drbd_err(device, "drbd_send_block/ack() failed\n"); |
99920dc5 | 1143 | return err; |
b411b363 PR |
1144 | } |
1145 | ||
99920dc5 | 1146 | int w_e_end_ov_req(struct drbd_work *w, int cancel) |
b411b363 | 1147 | { |
a8cd15ba AG |
1148 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
1149 | struct drbd_device *device = peer_req->peer_device->device; | |
db830c46 AG |
1150 | sector_t sector = peer_req->i.sector; |
1151 | unsigned int size = peer_req->i.size; | |
b411b363 PR |
1152 | int digest_size; |
1153 | void *digest; | |
99920dc5 | 1154 | int err = 0; |
b411b363 PR |
1155 | |
1156 | if (unlikely(cancel)) | |
1157 | goto out; | |
1158 | ||
a6b32bc3 | 1159 | digest_size = crypto_hash_digestsize(first_peer_device(device)->connection->verify_tfm); |
b411b363 | 1160 | digest = kmalloc(digest_size, GFP_NOIO); |
8f21420e | 1161 | if (!digest) { |
99920dc5 | 1162 | err = 1; /* terminate the connection in case the allocation failed */ |
8f21420e | 1163 | goto out; |
b411b363 PR |
1164 | } |
1165 | ||
db830c46 | 1166 | if (likely(!(peer_req->flags & EE_WAS_ERROR))) |
79a3c8d3 | 1167 | drbd_csum_ee(first_peer_device(device)->connection->verify_tfm, peer_req, digest); |
8f21420e PR |
1168 | else |
1169 | memset(digest, 0, digest_size); | |
1170 | ||
53ea4331 LE |
1171 | /* Free e and pages before send. |
1172 | * In case we block on congestion, we could otherwise run into | |
1173 | * some distributed deadlock, if the other side blocks on | |
1174 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1175 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 1176 | drbd_free_peer_req(device, peer_req); |
db830c46 | 1177 | peer_req = NULL; |
b30ab791 | 1178 | inc_rs_pending(device); |
69a22773 | 1179 | err = drbd_send_drequest_csum(first_peer_device(device), sector, size, digest, digest_size, P_OV_REPLY); |
99920dc5 | 1180 | if (err) |
b30ab791 | 1181 | dec_rs_pending(device); |
8f21420e PR |
1182 | kfree(digest); |
1183 | ||
b411b363 | 1184 | out: |
db830c46 | 1185 | if (peer_req) |
b30ab791 AG |
1186 | drbd_free_peer_req(device, peer_req); |
1187 | dec_unacked(device); | |
99920dc5 | 1188 | return err; |
b411b363 PR |
1189 | } |
1190 | ||
b30ab791 | 1191 | void drbd_ov_out_of_sync_found(struct drbd_device *device, sector_t sector, int size) |
b411b363 | 1192 | { |
b30ab791 AG |
1193 | if (device->ov_last_oos_start + device->ov_last_oos_size == sector) { |
1194 | device->ov_last_oos_size += size>>9; | |
b411b363 | 1195 | } else { |
b30ab791 AG |
1196 | device->ov_last_oos_start = sector; |
1197 | device->ov_last_oos_size = size>>9; | |
b411b363 | 1198 | } |
b30ab791 | 1199 | drbd_set_out_of_sync(device, sector, size); |
b411b363 PR |
1200 | } |
1201 | ||
99920dc5 | 1202 | int w_e_end_ov_reply(struct drbd_work *w, int cancel) |
b411b363 | 1203 | { |
a8cd15ba AG |
1204 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
1205 | struct drbd_device *device = peer_req->peer_device->device; | |
b411b363 | 1206 | struct digest_info *di; |
b411b363 | 1207 | void *digest; |
db830c46 AG |
1208 | sector_t sector = peer_req->i.sector; |
1209 | unsigned int size = peer_req->i.size; | |
53ea4331 | 1210 | int digest_size; |
99920dc5 | 1211 | int err, eq = 0; |
58ffa580 | 1212 | bool stop_sector_reached = false; |
b411b363 PR |
1213 | |
1214 | if (unlikely(cancel)) { | |
b30ab791 AG |
1215 | drbd_free_peer_req(device, peer_req); |
1216 | dec_unacked(device); | |
99920dc5 | 1217 | return 0; |
b411b363 PR |
1218 | } |
1219 | ||
1220 | /* after "cancel", because after drbd_disconnect/drbd_rs_cancel_all | |
1221 | * the resync lru has been cleaned up already */ | |
b30ab791 AG |
1222 | if (get_ldev(device)) { |
1223 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1224 | put_ldev(device); | |
1d53f09e | 1225 | } |
b411b363 | 1226 | |
db830c46 | 1227 | di = peer_req->digest; |
b411b363 | 1228 | |
db830c46 | 1229 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
a6b32bc3 | 1230 | digest_size = crypto_hash_digestsize(first_peer_device(device)->connection->verify_tfm); |
b411b363 PR |
1231 | digest = kmalloc(digest_size, GFP_NOIO); |
1232 | if (digest) { | |
79a3c8d3 | 1233 | drbd_csum_ee(first_peer_device(device)->connection->verify_tfm, peer_req, digest); |
b411b363 | 1234 | |
0b0ba1ef | 1235 | D_ASSERT(device, digest_size == di->digest_size); |
b411b363 PR |
1236 | eq = !memcmp(digest, di->digest, digest_size); |
1237 | kfree(digest); | |
1238 | } | |
b411b363 PR |
1239 | } |
1240 | ||
9676c760 LE |
1241 | /* Free peer_req and pages before send. |
1242 | * In case we block on congestion, we could otherwise run into | |
1243 | * some distributed deadlock, if the other side blocks on | |
1244 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1245 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 1246 | drbd_free_peer_req(device, peer_req); |
b411b363 | 1247 | if (!eq) |
b30ab791 | 1248 | drbd_ov_out_of_sync_found(device, sector, size); |
b411b363 | 1249 | else |
b30ab791 | 1250 | ov_out_of_sync_print(device); |
b411b363 | 1251 | |
69a22773 | 1252 | err = drbd_send_ack_ex(first_peer_device(device), P_OV_RESULT, sector, size, |
fa79abd8 | 1253 | eq ? ID_IN_SYNC : ID_OUT_OF_SYNC); |
b411b363 | 1254 | |
b30ab791 | 1255 | dec_unacked(device); |
b411b363 | 1256 | |
b30ab791 | 1257 | --device->ov_left; |
ea5442af LE |
1258 | |
1259 | /* let's advance progress step marks only for every other megabyte */ | |
b30ab791 AG |
1260 | if ((device->ov_left & 0x200) == 0x200) |
1261 | drbd_advance_rs_marks(device, device->ov_left); | |
ea5442af | 1262 | |
b30ab791 AG |
1263 | stop_sector_reached = verify_can_do_stop_sector(device) && |
1264 | (sector + (size>>9)) >= device->ov_stop_sector; | |
58ffa580 | 1265 | |
b30ab791 AG |
1266 | if (device->ov_left == 0 || stop_sector_reached) { |
1267 | ov_out_of_sync_print(device); | |
1268 | drbd_resync_finished(device); | |
b411b363 PR |
1269 | } |
1270 | ||
99920dc5 | 1271 | return err; |
b411b363 PR |
1272 | } |
1273 | ||
b6dd1a89 LE |
1274 | /* FIXME |
1275 | * We need to track the number of pending barrier acks, | |
1276 | * and to be able to wait for them. | |
1277 | * See also comment in drbd_adm_attach before drbd_suspend_io. | |
1278 | */ | |
bde89a9e | 1279 | static int drbd_send_barrier(struct drbd_connection *connection) |
b411b363 | 1280 | { |
9f5bdc33 | 1281 | struct p_barrier *p; |
b6dd1a89 | 1282 | struct drbd_socket *sock; |
b411b363 | 1283 | |
bde89a9e AG |
1284 | sock = &connection->data; |
1285 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1286 | if (!p) |
1287 | return -EIO; | |
bde89a9e | 1288 | p->barrier = connection->send.current_epoch_nr; |
b6dd1a89 | 1289 | p->pad = 0; |
bde89a9e | 1290 | connection->send.current_epoch_writes = 0; |
b6dd1a89 | 1291 | |
bde89a9e | 1292 | return conn_send_command(connection, sock, P_BARRIER, sizeof(*p), NULL, 0); |
b411b363 PR |
1293 | } |
1294 | ||
99920dc5 | 1295 | int w_send_write_hint(struct drbd_work *w, int cancel) |
b411b363 | 1296 | { |
84b8c06b AG |
1297 | struct drbd_device *device = |
1298 | container_of(w, struct drbd_device, unplug_work); | |
9f5bdc33 AG |
1299 | struct drbd_socket *sock; |
1300 | ||
b411b363 | 1301 | if (cancel) |
99920dc5 | 1302 | return 0; |
a6b32bc3 | 1303 | sock = &first_peer_device(device)->connection->data; |
69a22773 | 1304 | if (!drbd_prepare_command(first_peer_device(device), sock)) |
9f5bdc33 | 1305 | return -EIO; |
69a22773 | 1306 | return drbd_send_command(first_peer_device(device), sock, P_UNPLUG_REMOTE, 0, NULL, 0); |
b411b363 PR |
1307 | } |
1308 | ||
bde89a9e | 1309 | static void re_init_if_first_write(struct drbd_connection *connection, unsigned int epoch) |
4eb9b3cb | 1310 | { |
bde89a9e AG |
1311 | if (!connection->send.seen_any_write_yet) { |
1312 | connection->send.seen_any_write_yet = true; | |
1313 | connection->send.current_epoch_nr = epoch; | |
1314 | connection->send.current_epoch_writes = 0; | |
4eb9b3cb LE |
1315 | } |
1316 | } | |
1317 | ||
bde89a9e | 1318 | static void maybe_send_barrier(struct drbd_connection *connection, unsigned int epoch) |
4eb9b3cb LE |
1319 | { |
1320 | /* re-init if first write on this connection */ | |
bde89a9e | 1321 | if (!connection->send.seen_any_write_yet) |
4eb9b3cb | 1322 | return; |
bde89a9e AG |
1323 | if (connection->send.current_epoch_nr != epoch) { |
1324 | if (connection->send.current_epoch_writes) | |
1325 | drbd_send_barrier(connection); | |
1326 | connection->send.current_epoch_nr = epoch; | |
4eb9b3cb LE |
1327 | } |
1328 | } | |
1329 | ||
8f7bed77 | 1330 | int w_send_out_of_sync(struct drbd_work *w, int cancel) |
73a01a18 PR |
1331 | { |
1332 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1333 | struct drbd_device *device = req->device; |
a6b32bc3 | 1334 | struct drbd_connection *connection = first_peer_device(device)->connection; |
99920dc5 | 1335 | int err; |
73a01a18 PR |
1336 | |
1337 | if (unlikely(cancel)) { | |
8554df1c | 1338 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1339 | return 0; |
73a01a18 PR |
1340 | } |
1341 | ||
bde89a9e | 1342 | /* this time, no connection->send.current_epoch_writes++; |
b6dd1a89 LE |
1343 | * If it was sent, it was the closing barrier for the last |
1344 | * replicated epoch, before we went into AHEAD mode. | |
1345 | * No more barriers will be sent, until we leave AHEAD mode again. */ | |
bde89a9e | 1346 | maybe_send_barrier(connection, req->epoch); |
b6dd1a89 | 1347 | |
69a22773 | 1348 | err = drbd_send_out_of_sync(first_peer_device(device), req); |
8554df1c | 1349 | req_mod(req, OOS_HANDED_TO_NETWORK); |
73a01a18 | 1350 | |
99920dc5 | 1351 | return err; |
73a01a18 PR |
1352 | } |
1353 | ||
b411b363 PR |
1354 | /** |
1355 | * w_send_dblock() - Worker callback to send a P_DATA packet in order to mirror a write request | |
b411b363 PR |
1356 | * @w: work object. |
1357 | * @cancel: The connection will be closed anyways | |
1358 | */ | |
99920dc5 | 1359 | int w_send_dblock(struct drbd_work *w, int cancel) |
b411b363 PR |
1360 | { |
1361 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1362 | struct drbd_device *device = req->device; |
a6b32bc3 | 1363 | struct drbd_connection *connection = first_peer_device(device)->connection; |
99920dc5 | 1364 | int err; |
b411b363 PR |
1365 | |
1366 | if (unlikely(cancel)) { | |
8554df1c | 1367 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1368 | return 0; |
b411b363 PR |
1369 | } |
1370 | ||
bde89a9e AG |
1371 | re_init_if_first_write(connection, req->epoch); |
1372 | maybe_send_barrier(connection, req->epoch); | |
1373 | connection->send.current_epoch_writes++; | |
b6dd1a89 | 1374 | |
69a22773 | 1375 | err = drbd_send_dblock(first_peer_device(device), req); |
99920dc5 | 1376 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1377 | |
99920dc5 | 1378 | return err; |
b411b363 PR |
1379 | } |
1380 | ||
1381 | /** | |
1382 | * w_send_read_req() - Worker callback to send a read request (P_DATA_REQUEST) packet | |
b411b363 PR |
1383 | * @w: work object. |
1384 | * @cancel: The connection will be closed anyways | |
1385 | */ | |
99920dc5 | 1386 | int w_send_read_req(struct drbd_work *w, int cancel) |
b411b363 PR |
1387 | { |
1388 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1389 | struct drbd_device *device = req->device; |
a6b32bc3 | 1390 | struct drbd_connection *connection = first_peer_device(device)->connection; |
99920dc5 | 1391 | int err; |
b411b363 PR |
1392 | |
1393 | if (unlikely(cancel)) { | |
8554df1c | 1394 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1395 | return 0; |
b411b363 PR |
1396 | } |
1397 | ||
b6dd1a89 LE |
1398 | /* Even read requests may close a write epoch, |
1399 | * if there was any yet. */ | |
bde89a9e | 1400 | maybe_send_barrier(connection, req->epoch); |
b6dd1a89 | 1401 | |
69a22773 | 1402 | err = drbd_send_drequest(first_peer_device(device), P_DATA_REQUEST, req->i.sector, req->i.size, |
6c1005e7 | 1403 | (unsigned long)req); |
b411b363 | 1404 | |
99920dc5 | 1405 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1406 | |
99920dc5 | 1407 | return err; |
b411b363 PR |
1408 | } |
1409 | ||
99920dc5 | 1410 | int w_restart_disk_io(struct drbd_work *w, int cancel) |
265be2d0 PR |
1411 | { |
1412 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1413 | struct drbd_device *device = req->device; |
265be2d0 | 1414 | |
0778286a | 1415 | if (bio_data_dir(req->master_bio) == WRITE && req->rq_state & RQ_IN_ACT_LOG) |
b30ab791 | 1416 | drbd_al_begin_io(device, &req->i, false); |
265be2d0 PR |
1417 | |
1418 | drbd_req_make_private_bio(req, req->master_bio); | |
b30ab791 | 1419 | req->private_bio->bi_bdev = device->ldev->backing_bdev; |
265be2d0 PR |
1420 | generic_make_request(req->private_bio); |
1421 | ||
99920dc5 | 1422 | return 0; |
265be2d0 PR |
1423 | } |
1424 | ||
b30ab791 | 1425 | static int _drbd_may_sync_now(struct drbd_device *device) |
b411b363 | 1426 | { |
b30ab791 | 1427 | struct drbd_device *odev = device; |
95f8efd0 | 1428 | int resync_after; |
b411b363 PR |
1429 | |
1430 | while (1) { | |
a3f8f7dc | 1431 | if (!odev->ldev || odev->state.disk == D_DISKLESS) |
438c8374 | 1432 | return 1; |
daeda1cc | 1433 | rcu_read_lock(); |
95f8efd0 | 1434 | resync_after = rcu_dereference(odev->ldev->disk_conf)->resync_after; |
daeda1cc | 1435 | rcu_read_unlock(); |
95f8efd0 | 1436 | if (resync_after == -1) |
b411b363 | 1437 | return 1; |
b30ab791 | 1438 | odev = minor_to_device(resync_after); |
a3f8f7dc | 1439 | if (!odev) |
841ce241 | 1440 | return 1; |
b411b363 PR |
1441 | if ((odev->state.conn >= C_SYNC_SOURCE && |
1442 | odev->state.conn <= C_PAUSED_SYNC_T) || | |
1443 | odev->state.aftr_isp || odev->state.peer_isp || | |
1444 | odev->state.user_isp) | |
1445 | return 0; | |
1446 | } | |
1447 | } | |
1448 | ||
1449 | /** | |
1450 | * _drbd_pause_after() - Pause resync on all devices that may not resync now | |
b30ab791 | 1451 | * @device: DRBD device. |
b411b363 PR |
1452 | * |
1453 | * Called from process context only (admin command and after_state_ch). | |
1454 | */ | |
b30ab791 | 1455 | static int _drbd_pause_after(struct drbd_device *device) |
b411b363 | 1456 | { |
54761697 | 1457 | struct drbd_device *odev; |
b411b363 PR |
1458 | int i, rv = 0; |
1459 | ||
695d08fa | 1460 | rcu_read_lock(); |
05a10ec7 | 1461 | idr_for_each_entry(&drbd_devices, odev, i) { |
b411b363 PR |
1462 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1463 | continue; | |
1464 | if (!_drbd_may_sync_now(odev)) | |
1465 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 1), CS_HARD, NULL) | |
1466 | != SS_NOTHING_TO_DO); | |
1467 | } | |
695d08fa | 1468 | rcu_read_unlock(); |
b411b363 PR |
1469 | |
1470 | return rv; | |
1471 | } | |
1472 | ||
1473 | /** | |
1474 | * _drbd_resume_next() - Resume resync on all devices that may resync now | |
b30ab791 | 1475 | * @device: DRBD device. |
b411b363 PR |
1476 | * |
1477 | * Called from process context only (admin command and worker). | |
1478 | */ | |
b30ab791 | 1479 | static int _drbd_resume_next(struct drbd_device *device) |
b411b363 | 1480 | { |
54761697 | 1481 | struct drbd_device *odev; |
b411b363 PR |
1482 | int i, rv = 0; |
1483 | ||
695d08fa | 1484 | rcu_read_lock(); |
05a10ec7 | 1485 | idr_for_each_entry(&drbd_devices, odev, i) { |
b411b363 PR |
1486 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1487 | continue; | |
1488 | if (odev->state.aftr_isp) { | |
1489 | if (_drbd_may_sync_now(odev)) | |
1490 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 0), | |
1491 | CS_HARD, NULL) | |
1492 | != SS_NOTHING_TO_DO) ; | |
1493 | } | |
1494 | } | |
695d08fa | 1495 | rcu_read_unlock(); |
b411b363 PR |
1496 | return rv; |
1497 | } | |
1498 | ||
b30ab791 | 1499 | void resume_next_sg(struct drbd_device *device) |
b411b363 PR |
1500 | { |
1501 | write_lock_irq(&global_state_lock); | |
b30ab791 | 1502 | _drbd_resume_next(device); |
b411b363 PR |
1503 | write_unlock_irq(&global_state_lock); |
1504 | } | |
1505 | ||
b30ab791 | 1506 | void suspend_other_sg(struct drbd_device *device) |
b411b363 PR |
1507 | { |
1508 | write_lock_irq(&global_state_lock); | |
b30ab791 | 1509 | _drbd_pause_after(device); |
b411b363 PR |
1510 | write_unlock_irq(&global_state_lock); |
1511 | } | |
1512 | ||
dc97b708 | 1513 | /* caller must hold global_state_lock */ |
b30ab791 | 1514 | enum drbd_ret_code drbd_resync_after_valid(struct drbd_device *device, int o_minor) |
b411b363 | 1515 | { |
54761697 | 1516 | struct drbd_device *odev; |
95f8efd0 | 1517 | int resync_after; |
b411b363 PR |
1518 | |
1519 | if (o_minor == -1) | |
1520 | return NO_ERROR; | |
a3f8f7dc | 1521 | if (o_minor < -1 || o_minor > MINORMASK) |
95f8efd0 | 1522 | return ERR_RESYNC_AFTER; |
b411b363 PR |
1523 | |
1524 | /* check for loops */ | |
b30ab791 | 1525 | odev = minor_to_device(o_minor); |
b411b363 | 1526 | while (1) { |
b30ab791 | 1527 | if (odev == device) |
95f8efd0 | 1528 | return ERR_RESYNC_AFTER_CYCLE; |
b411b363 | 1529 | |
a3f8f7dc LE |
1530 | /* You are free to depend on diskless, non-existing, |
1531 | * or not yet/no longer existing minors. | |
1532 | * We only reject dependency loops. | |
1533 | * We cannot follow the dependency chain beyond a detached or | |
1534 | * missing minor. | |
1535 | */ | |
1536 | if (!odev || !odev->ldev || odev->state.disk == D_DISKLESS) | |
1537 | return NO_ERROR; | |
1538 | ||
daeda1cc | 1539 | rcu_read_lock(); |
95f8efd0 | 1540 | resync_after = rcu_dereference(odev->ldev->disk_conf)->resync_after; |
daeda1cc | 1541 | rcu_read_unlock(); |
b411b363 | 1542 | /* dependency chain ends here, no cycles. */ |
95f8efd0 | 1543 | if (resync_after == -1) |
b411b363 PR |
1544 | return NO_ERROR; |
1545 | ||
1546 | /* follow the dependency chain */ | |
b30ab791 | 1547 | odev = minor_to_device(resync_after); |
b411b363 PR |
1548 | } |
1549 | } | |
1550 | ||
dc97b708 | 1551 | /* caller must hold global_state_lock */ |
b30ab791 | 1552 | void drbd_resync_after_changed(struct drbd_device *device) |
b411b363 PR |
1553 | { |
1554 | int changes; | |
b411b363 | 1555 | |
dc97b708 | 1556 | do { |
b30ab791 AG |
1557 | changes = _drbd_pause_after(device); |
1558 | changes |= _drbd_resume_next(device); | |
dc97b708 | 1559 | } while (changes); |
b411b363 PR |
1560 | } |
1561 | ||
b30ab791 | 1562 | void drbd_rs_controller_reset(struct drbd_device *device) |
9bd28d3c | 1563 | { |
813472ce PR |
1564 | struct fifo_buffer *plan; |
1565 | ||
b30ab791 AG |
1566 | atomic_set(&device->rs_sect_in, 0); |
1567 | atomic_set(&device->rs_sect_ev, 0); | |
1568 | device->rs_in_flight = 0; | |
813472ce PR |
1569 | |
1570 | /* Updating the RCU protected object in place is necessary since | |
1571 | this function gets called from atomic context. | |
1572 | It is valid since all other updates also lead to an completely | |
1573 | empty fifo */ | |
1574 | rcu_read_lock(); | |
b30ab791 | 1575 | plan = rcu_dereference(device->rs_plan_s); |
813472ce PR |
1576 | plan->total = 0; |
1577 | fifo_set(plan, 0); | |
1578 | rcu_read_unlock(); | |
9bd28d3c LE |
1579 | } |
1580 | ||
1f04af33 PR |
1581 | void start_resync_timer_fn(unsigned long data) |
1582 | { | |
b30ab791 | 1583 | struct drbd_device *device = (struct drbd_device *) data; |
1f04af33 | 1584 | |
84b8c06b AG |
1585 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, |
1586 | &device->start_resync_work); | |
1f04af33 PR |
1587 | } |
1588 | ||
99920dc5 | 1589 | int w_start_resync(struct drbd_work *w, int cancel) |
1f04af33 | 1590 | { |
84b8c06b AG |
1591 | struct drbd_device *device = |
1592 | container_of(w, struct drbd_device, start_resync_work); | |
00d56944 | 1593 | |
b30ab791 | 1594 | if (atomic_read(&device->unacked_cnt) || atomic_read(&device->rs_pending_cnt)) { |
d0180171 | 1595 | drbd_warn(device, "w_start_resync later...\n"); |
b30ab791 AG |
1596 | device->start_resync_timer.expires = jiffies + HZ/10; |
1597 | add_timer(&device->start_resync_timer); | |
99920dc5 | 1598 | return 0; |
1f04af33 PR |
1599 | } |
1600 | ||
b30ab791 AG |
1601 | drbd_start_resync(device, C_SYNC_SOURCE); |
1602 | clear_bit(AHEAD_TO_SYNC_SOURCE, &device->flags); | |
99920dc5 | 1603 | return 0; |
1f04af33 PR |
1604 | } |
1605 | ||
b411b363 PR |
1606 | /** |
1607 | * drbd_start_resync() - Start the resync process | |
b30ab791 | 1608 | * @device: DRBD device. |
b411b363 PR |
1609 | * @side: Either C_SYNC_SOURCE or C_SYNC_TARGET |
1610 | * | |
1611 | * This function might bring you directly into one of the | |
1612 | * C_PAUSED_SYNC_* states. | |
1613 | */ | |
b30ab791 | 1614 | void drbd_start_resync(struct drbd_device *device, enum drbd_conns side) |
b411b363 PR |
1615 | { |
1616 | union drbd_state ns; | |
1617 | int r; | |
1618 | ||
b30ab791 | 1619 | if (device->state.conn >= C_SYNC_SOURCE && device->state.conn < C_AHEAD) { |
d0180171 | 1620 | drbd_err(device, "Resync already running!\n"); |
b411b363 PR |
1621 | return; |
1622 | } | |
1623 | ||
b30ab791 | 1624 | if (!test_bit(B_RS_H_DONE, &device->flags)) { |
e64a3294 PR |
1625 | if (side == C_SYNC_TARGET) { |
1626 | /* Since application IO was locked out during C_WF_BITMAP_T and | |
1627 | C_WF_SYNC_UUID we are still unmodified. Before going to C_SYNC_TARGET | |
1628 | we check that we might make the data inconsistent. */ | |
b30ab791 | 1629 | r = drbd_khelper(device, "before-resync-target"); |
e64a3294 PR |
1630 | r = (r >> 8) & 0xff; |
1631 | if (r > 0) { | |
d0180171 | 1632 | drbd_info(device, "before-resync-target handler returned %d, " |
09b9e797 | 1633 | "dropping connection.\n", r); |
a6b32bc3 | 1634 | conn_request_state(first_peer_device(device)->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
09b9e797 PR |
1635 | return; |
1636 | } | |
e64a3294 | 1637 | } else /* C_SYNC_SOURCE */ { |
b30ab791 | 1638 | r = drbd_khelper(device, "before-resync-source"); |
e64a3294 PR |
1639 | r = (r >> 8) & 0xff; |
1640 | if (r > 0) { | |
1641 | if (r == 3) { | |
d0180171 | 1642 | drbd_info(device, "before-resync-source handler returned %d, " |
e64a3294 PR |
1643 | "ignoring. Old userland tools?", r); |
1644 | } else { | |
d0180171 | 1645 | drbd_info(device, "before-resync-source handler returned %d, " |
e64a3294 | 1646 | "dropping connection.\n", r); |
a6b32bc3 AG |
1647 | conn_request_state(first_peer_device(device)->connection, |
1648 | NS(conn, C_DISCONNECTING), CS_HARD); | |
e64a3294 PR |
1649 | return; |
1650 | } | |
1651 | } | |
09b9e797 | 1652 | } |
b411b363 PR |
1653 | } |
1654 | ||
a6b32bc3 | 1655 | if (current == first_peer_device(device)->connection->worker.task) { |
dad20554 | 1656 | /* The worker should not sleep waiting for state_mutex, |
e64a3294 | 1657 | that can take long */ |
b30ab791 AG |
1658 | if (!mutex_trylock(device->state_mutex)) { |
1659 | set_bit(B_RS_H_DONE, &device->flags); | |
1660 | device->start_resync_timer.expires = jiffies + HZ/5; | |
1661 | add_timer(&device->start_resync_timer); | |
e64a3294 PR |
1662 | return; |
1663 | } | |
1664 | } else { | |
b30ab791 | 1665 | mutex_lock(device->state_mutex); |
e64a3294 | 1666 | } |
b30ab791 | 1667 | clear_bit(B_RS_H_DONE, &device->flags); |
b411b363 | 1668 | |
0cfac5dd | 1669 | write_lock_irq(&global_state_lock); |
a700471b | 1670 | /* Did some connection breakage or IO error race with us? */ |
b30ab791 AG |
1671 | if (device->state.conn < C_CONNECTED |
1672 | || !get_ldev_if_state(device, D_NEGOTIATING)) { | |
0cfac5dd | 1673 | write_unlock_irq(&global_state_lock); |
b30ab791 | 1674 | mutex_unlock(device->state_mutex); |
b411b363 PR |
1675 | return; |
1676 | } | |
1677 | ||
b30ab791 | 1678 | ns = drbd_read_state(device); |
b411b363 | 1679 | |
b30ab791 | 1680 | ns.aftr_isp = !_drbd_may_sync_now(device); |
b411b363 PR |
1681 | |
1682 | ns.conn = side; | |
1683 | ||
1684 | if (side == C_SYNC_TARGET) | |
1685 | ns.disk = D_INCONSISTENT; | |
1686 | else /* side == C_SYNC_SOURCE */ | |
1687 | ns.pdsk = D_INCONSISTENT; | |
1688 | ||
b30ab791 AG |
1689 | r = __drbd_set_state(device, ns, CS_VERBOSE, NULL); |
1690 | ns = drbd_read_state(device); | |
b411b363 PR |
1691 | |
1692 | if (ns.conn < C_CONNECTED) | |
1693 | r = SS_UNKNOWN_ERROR; | |
1694 | ||
1695 | if (r == SS_SUCCESS) { | |
b30ab791 | 1696 | unsigned long tw = drbd_bm_total_weight(device); |
1d7734a0 LE |
1697 | unsigned long now = jiffies; |
1698 | int i; | |
1699 | ||
b30ab791 AG |
1700 | device->rs_failed = 0; |
1701 | device->rs_paused = 0; | |
1702 | device->rs_same_csum = 0; | |
1703 | device->rs_last_events = 0; | |
1704 | device->rs_last_sect_ev = 0; | |
1705 | device->rs_total = tw; | |
1706 | device->rs_start = now; | |
1d7734a0 | 1707 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
1708 | device->rs_mark_left[i] = tw; |
1709 | device->rs_mark_time[i] = now; | |
1d7734a0 | 1710 | } |
b30ab791 | 1711 | _drbd_pause_after(device); |
b411b363 PR |
1712 | } |
1713 | write_unlock_irq(&global_state_lock); | |
5a22db89 | 1714 | |
b411b363 | 1715 | if (r == SS_SUCCESS) { |
328e0f12 PR |
1716 | /* reset rs_last_bcast when a resync or verify is started, |
1717 | * to deal with potential jiffies wrap. */ | |
b30ab791 | 1718 | device->rs_last_bcast = jiffies - HZ; |
328e0f12 | 1719 | |
d0180171 | 1720 | drbd_info(device, "Began resync as %s (will sync %lu KB [%lu bits set]).\n", |
b411b363 | 1721 | drbd_conn_str(ns.conn), |
b30ab791 AG |
1722 | (unsigned long) device->rs_total << (BM_BLOCK_SHIFT-10), |
1723 | (unsigned long) device->rs_total); | |
6c922ed5 | 1724 | if (side == C_SYNC_TARGET) |
b30ab791 | 1725 | device->bm_resync_fo = 0; |
6c922ed5 LE |
1726 | |
1727 | /* Since protocol 96, we must serialize drbd_gen_and_send_sync_uuid | |
1728 | * with w_send_oos, or the sync target will get confused as to | |
1729 | * how much bits to resync. We cannot do that always, because for an | |
1730 | * empty resync and protocol < 95, we need to do it here, as we call | |
1731 | * drbd_resync_finished from here in that case. | |
1732 | * We drbd_gen_and_send_sync_uuid here for protocol < 96, | |
1733 | * and from after_state_ch otherwise. */ | |
a6b32bc3 AG |
1734 | if (side == C_SYNC_SOURCE && |
1735 | first_peer_device(device)->connection->agreed_pro_version < 96) | |
69a22773 | 1736 | drbd_gen_and_send_sync_uuid(first_peer_device(device)); |
b411b363 | 1737 | |
a6b32bc3 AG |
1738 | if (first_peer_device(device)->connection->agreed_pro_version < 95 && |
1739 | device->rs_total == 0) { | |
af85e8e8 LE |
1740 | /* This still has a race (about when exactly the peers |
1741 | * detect connection loss) that can lead to a full sync | |
1742 | * on next handshake. In 8.3.9 we fixed this with explicit | |
1743 | * resync-finished notifications, but the fix | |
1744 | * introduces a protocol change. Sleeping for some | |
1745 | * time longer than the ping interval + timeout on the | |
1746 | * SyncSource, to give the SyncTarget the chance to | |
1747 | * detect connection loss, then waiting for a ping | |
1748 | * response (implicit in drbd_resync_finished) reduces | |
1749 | * the race considerably, but does not solve it. */ | |
44ed167d PR |
1750 | if (side == C_SYNC_SOURCE) { |
1751 | struct net_conf *nc; | |
1752 | int timeo; | |
1753 | ||
1754 | rcu_read_lock(); | |
a6b32bc3 | 1755 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); |
44ed167d PR |
1756 | timeo = nc->ping_int * HZ + nc->ping_timeo * HZ / 9; |
1757 | rcu_read_unlock(); | |
1758 | schedule_timeout_interruptible(timeo); | |
1759 | } | |
b30ab791 | 1760 | drbd_resync_finished(device); |
b411b363 PR |
1761 | } |
1762 | ||
b30ab791 AG |
1763 | drbd_rs_controller_reset(device); |
1764 | /* ns.conn may already be != device->state.conn, | |
b411b363 PR |
1765 | * we may have been paused in between, or become paused until |
1766 | * the timer triggers. | |
1767 | * No matter, that is handled in resync_timer_fn() */ | |
1768 | if (ns.conn == C_SYNC_TARGET) | |
b30ab791 | 1769 | mod_timer(&device->resync_timer, jiffies); |
b411b363 | 1770 | |
b30ab791 | 1771 | drbd_md_sync(device); |
b411b363 | 1772 | } |
b30ab791 AG |
1773 | put_ldev(device); |
1774 | mutex_unlock(device->state_mutex); | |
b411b363 PR |
1775 | } |
1776 | ||
b6dd1a89 LE |
1777 | /* If the resource already closed the current epoch, but we did not |
1778 | * (because we have not yet seen new requests), we should send the | |
1779 | * corresponding barrier now. Must be checked within the same spinlock | |
1780 | * that is used to check for new requests. */ | |
bde89a9e | 1781 | static bool need_to_send_barrier(struct drbd_connection *connection) |
b6dd1a89 LE |
1782 | { |
1783 | if (!connection->send.seen_any_write_yet) | |
1784 | return false; | |
1785 | ||
1786 | /* Skip barriers that do not contain any writes. | |
1787 | * This may happen during AHEAD mode. */ | |
1788 | if (!connection->send.current_epoch_writes) | |
1789 | return false; | |
1790 | ||
1791 | /* ->req_lock is held when requests are queued on | |
1792 | * connection->sender_work, and put into ->transfer_log. | |
1793 | * It is also held when ->current_tle_nr is increased. | |
1794 | * So either there are already new requests queued, | |
1795 | * and corresponding barriers will be send there. | |
1796 | * Or nothing new is queued yet, so the difference will be 1. | |
1797 | */ | |
1798 | if (atomic_read(&connection->current_tle_nr) != | |
1799 | connection->send.current_epoch_nr + 1) | |
1800 | return false; | |
1801 | ||
1802 | return true; | |
1803 | } | |
1804 | ||
a186e478 | 1805 | static bool dequeue_work_batch(struct drbd_work_queue *queue, struct list_head *work_list) |
8c0785a5 LE |
1806 | { |
1807 | spin_lock_irq(&queue->q_lock); | |
1808 | list_splice_init(&queue->q, work_list); | |
1809 | spin_unlock_irq(&queue->q_lock); | |
1810 | return !list_empty(work_list); | |
1811 | } | |
1812 | ||
a186e478 | 1813 | static bool dequeue_work_item(struct drbd_work_queue *queue, struct list_head *work_list) |
8c0785a5 LE |
1814 | { |
1815 | spin_lock_irq(&queue->q_lock); | |
1816 | if (!list_empty(&queue->q)) | |
1817 | list_move(queue->q.next, work_list); | |
1818 | spin_unlock_irq(&queue->q_lock); | |
1819 | return !list_empty(work_list); | |
1820 | } | |
1821 | ||
bde89a9e | 1822 | static void wait_for_work(struct drbd_connection *connection, struct list_head *work_list) |
b6dd1a89 LE |
1823 | { |
1824 | DEFINE_WAIT(wait); | |
1825 | struct net_conf *nc; | |
1826 | int uncork, cork; | |
1827 | ||
1828 | dequeue_work_item(&connection->sender_work, work_list); | |
1829 | if (!list_empty(work_list)) | |
1830 | return; | |
1831 | ||
1832 | /* Still nothing to do? | |
1833 | * Maybe we still need to close the current epoch, | |
1834 | * even if no new requests are queued yet. | |
1835 | * | |
1836 | * Also, poke TCP, just in case. | |
1837 | * Then wait for new work (or signal). */ | |
1838 | rcu_read_lock(); | |
1839 | nc = rcu_dereference(connection->net_conf); | |
1840 | uncork = nc ? nc->tcp_cork : 0; | |
1841 | rcu_read_unlock(); | |
1842 | if (uncork) { | |
1843 | mutex_lock(&connection->data.mutex); | |
1844 | if (connection->data.socket) | |
1845 | drbd_tcp_uncork(connection->data.socket); | |
1846 | mutex_unlock(&connection->data.mutex); | |
1847 | } | |
1848 | ||
1849 | for (;;) { | |
1850 | int send_barrier; | |
1851 | prepare_to_wait(&connection->sender_work.q_wait, &wait, TASK_INTERRUPTIBLE); | |
0500813f | 1852 | spin_lock_irq(&connection->resource->req_lock); |
b6dd1a89 | 1853 | spin_lock(&connection->sender_work.q_lock); /* FIXME get rid of this one? */ |
bc317a9e LE |
1854 | /* dequeue single item only, |
1855 | * we still use drbd_queue_work_front() in some places */ | |
1856 | if (!list_empty(&connection->sender_work.q)) | |
1857 | list_move(connection->sender_work.q.next, work_list); | |
b6dd1a89 LE |
1858 | spin_unlock(&connection->sender_work.q_lock); /* FIXME get rid of this one? */ |
1859 | if (!list_empty(work_list) || signal_pending(current)) { | |
0500813f | 1860 | spin_unlock_irq(&connection->resource->req_lock); |
b6dd1a89 LE |
1861 | break; |
1862 | } | |
1863 | send_barrier = need_to_send_barrier(connection); | |
0500813f | 1864 | spin_unlock_irq(&connection->resource->req_lock); |
b6dd1a89 LE |
1865 | if (send_barrier) { |
1866 | drbd_send_barrier(connection); | |
1867 | connection->send.current_epoch_nr++; | |
1868 | } | |
1869 | schedule(); | |
1870 | /* may be woken up for other things but new work, too, | |
1871 | * e.g. if the current epoch got closed. | |
1872 | * In which case we send the barrier above. */ | |
1873 | } | |
1874 | finish_wait(&connection->sender_work.q_wait, &wait); | |
1875 | ||
1876 | /* someone may have changed the config while we have been waiting above. */ | |
1877 | rcu_read_lock(); | |
1878 | nc = rcu_dereference(connection->net_conf); | |
1879 | cork = nc ? nc->tcp_cork : 0; | |
1880 | rcu_read_unlock(); | |
1881 | mutex_lock(&connection->data.mutex); | |
1882 | if (connection->data.socket) { | |
1883 | if (cork) | |
1884 | drbd_tcp_cork(connection->data.socket); | |
1885 | else if (!uncork) | |
1886 | drbd_tcp_uncork(connection->data.socket); | |
1887 | } | |
1888 | mutex_unlock(&connection->data.mutex); | |
1889 | } | |
1890 | ||
b411b363 PR |
1891 | int drbd_worker(struct drbd_thread *thi) |
1892 | { | |
bde89a9e | 1893 | struct drbd_connection *connection = thi->connection; |
84b8c06b | 1894 | struct drbd_device_work *dw = NULL; |
c06ece6b | 1895 | struct drbd_peer_device *peer_device; |
b411b363 | 1896 | LIST_HEAD(work_list); |
8c0785a5 | 1897 | int vnr; |
b411b363 | 1898 | |
e77a0a5c | 1899 | while (get_t_state(thi) == RUNNING) { |
80822284 | 1900 | drbd_thread_current_set_cpu(thi); |
b411b363 | 1901 | |
8c0785a5 LE |
1902 | /* as long as we use drbd_queue_work_front(), |
1903 | * we may only dequeue single work items here, not batches. */ | |
1904 | if (list_empty(&work_list)) | |
bde89a9e | 1905 | wait_for_work(connection, &work_list); |
b411b363 | 1906 | |
8c0785a5 | 1907 | if (signal_pending(current)) { |
b411b363 | 1908 | flush_signals(current); |
19393e10 | 1909 | if (get_t_state(thi) == RUNNING) { |
1ec861eb | 1910 | drbd_warn(connection, "Worker got an unexpected signal\n"); |
b411b363 | 1911 | continue; |
19393e10 | 1912 | } |
b411b363 PR |
1913 | break; |
1914 | } | |
1915 | ||
e77a0a5c | 1916 | if (get_t_state(thi) != RUNNING) |
b411b363 | 1917 | break; |
b411b363 | 1918 | |
8c0785a5 | 1919 | while (!list_empty(&work_list)) { |
84b8c06b AG |
1920 | dw = list_first_entry(&work_list, struct drbd_device_work, w.list); |
1921 | list_del_init(&dw->w.list); | |
1922 | if (dw->w.cb(&dw->w, connection->cstate < C_WF_REPORT_PARAMS) == 0) | |
8c0785a5 | 1923 | continue; |
bde89a9e AG |
1924 | if (connection->cstate >= C_WF_REPORT_PARAMS) |
1925 | conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); | |
b411b363 PR |
1926 | } |
1927 | } | |
b411b363 | 1928 | |
8c0785a5 | 1929 | do { |
b411b363 | 1930 | while (!list_empty(&work_list)) { |
84b8c06b AG |
1931 | dw = list_first_entry(&work_list, struct drbd_device_work, w.list); |
1932 | list_del_init(&dw->w.list); | |
1933 | dw->w.cb(&dw->w, 1); | |
b411b363 | 1934 | } |
bde89a9e | 1935 | dequeue_work_batch(&connection->sender_work, &work_list); |
8c0785a5 | 1936 | } while (!list_empty(&work_list)); |
b411b363 | 1937 | |
c141ebda | 1938 | rcu_read_lock(); |
c06ece6b AG |
1939 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
1940 | struct drbd_device *device = peer_device->device; | |
0b0ba1ef | 1941 | D_ASSERT(device, device->state.disk == D_DISKLESS && device->state.conn == C_STANDALONE); |
b30ab791 | 1942 | kref_get(&device->kref); |
c141ebda | 1943 | rcu_read_unlock(); |
b30ab791 | 1944 | drbd_device_cleanup(device); |
05a10ec7 | 1945 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda | 1946 | rcu_read_lock(); |
0e29d163 | 1947 | } |
c141ebda | 1948 | rcu_read_unlock(); |
b411b363 PR |
1949 | |
1950 | return 0; | |
1951 | } |