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 | ||
24 | */ | |
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" | |
39 | #include "drbd_req.h" | |
b411b363 | 40 | |
00d56944 | 41 | static int w_make_ov_request(struct drbd_work *w, int cancel); |
b411b363 PR |
42 | |
43 | ||
c5a91619 AG |
44 | /* endio handlers: |
45 | * drbd_md_io_complete (defined here) | |
fcefa62e AG |
46 | * drbd_request_endio (defined here) |
47 | * drbd_peer_request_endio (defined here) | |
c5a91619 AG |
48 | * bm_async_io_complete (defined in drbd_bitmap.c) |
49 | * | |
b411b363 PR |
50 | * For all these callbacks, note the following: |
51 | * The callbacks will be called in irq context by the IDE drivers, | |
52 | * and in Softirqs/Tasklets/BH context by the SCSI drivers. | |
53 | * Try to get the locking right :) | |
54 | * | |
55 | */ | |
56 | ||
57 | ||
58 | /* About the global_state_lock | |
59 | Each state transition on an device holds a read lock. In case we have | |
60 | to evaluate the sync after dependencies, we grab a write lock, because | |
61 | we need stable states on all devices for that. */ | |
62 | rwlock_t global_state_lock; | |
63 | ||
64 | /* used for synchronous meta data and bitmap IO | |
65 | * submitted by drbd_md_sync_page_io() | |
66 | */ | |
67 | void drbd_md_io_complete(struct bio *bio, int error) | |
68 | { | |
69 | struct drbd_md_io *md_io; | |
70 | ||
71 | md_io = (struct drbd_md_io *)bio->bi_private; | |
72 | md_io->error = error; | |
73 | ||
b411b363 PR |
74 | complete(&md_io->event); |
75 | } | |
76 | ||
77 | /* reads on behalf of the partner, | |
78 | * "submitted" by the receiver | |
79 | */ | |
db830c46 | 80 | void drbd_endio_read_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
81 | { |
82 | unsigned long flags = 0; | |
a21e9298 | 83 | struct drbd_conf *mdev = peer_req->w.mdev; |
b411b363 | 84 | |
87eeee41 | 85 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
db830c46 AG |
86 | mdev->read_cnt += peer_req->i.size >> 9; |
87 | list_del(&peer_req->w.list); | |
b411b363 PR |
88 | if (list_empty(&mdev->read_ee)) |
89 | wake_up(&mdev->ee_wait); | |
db830c46 | 90 | if (test_bit(__EE_WAS_ERROR, &peer_req->flags)) |
81e84650 | 91 | __drbd_chk_io_error(mdev, false); |
87eeee41 | 92 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
b411b363 | 93 | |
db830c46 | 94 | drbd_queue_work(&mdev->tconn->data.work, &peer_req->w); |
b411b363 | 95 | put_ldev(mdev); |
b411b363 PR |
96 | } |
97 | ||
98 | /* writes on behalf of the partner, or resync writes, | |
45bb912b | 99 | * "submitted" by the receiver, final stage. */ |
db830c46 | 100 | static void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
101 | { |
102 | unsigned long flags = 0; | |
a21e9298 | 103 | struct drbd_conf *mdev = peer_req->w.mdev; |
181286ad | 104 | struct drbd_interval i; |
b411b363 | 105 | int do_wake; |
579b57ed | 106 | u64 block_id; |
b411b363 | 107 | int do_al_complete_io; |
b411b363 | 108 | |
db830c46 | 109 | /* after we moved peer_req to done_ee, |
b411b363 PR |
110 | * we may no longer access it, |
111 | * it may be freed/reused already! | |
112 | * (as soon as we release the req_lock) */ | |
181286ad | 113 | i = peer_req->i; |
db830c46 AG |
114 | do_al_complete_io = peer_req->flags & EE_CALL_AL_COMPLETE_IO; |
115 | block_id = peer_req->block_id; | |
b411b363 | 116 | |
87eeee41 | 117 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
db830c46 AG |
118 | mdev->writ_cnt += peer_req->i.size >> 9; |
119 | list_del(&peer_req->w.list); /* has been on active_ee or sync_ee */ | |
120 | list_add_tail(&peer_req->w.list, &mdev->done_ee); | |
b411b363 | 121 | |
bb3bfe96 | 122 | /* |
5e472264 | 123 | * Do not remove from the write_requests tree here: we did not send the |
bb3bfe96 AG |
124 | * Ack yet and did not wake possibly waiting conflicting requests. |
125 | * Removed from the tree from "drbd_process_done_ee" within the | |
126 | * appropriate w.cb (e_end_block/e_end_resync_block) or from | |
127 | * _drbd_clear_done_ee. | |
128 | */ | |
b411b363 | 129 | |
579b57ed | 130 | do_wake = list_empty(block_id == ID_SYNCER ? &mdev->sync_ee : &mdev->active_ee); |
b411b363 | 131 | |
db830c46 | 132 | if (test_bit(__EE_WAS_ERROR, &peer_req->flags)) |
81e84650 | 133 | __drbd_chk_io_error(mdev, false); |
87eeee41 | 134 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
b411b363 | 135 | |
579b57ed | 136 | if (block_id == ID_SYNCER) |
181286ad | 137 | drbd_rs_complete_io(mdev, i.sector); |
b411b363 PR |
138 | |
139 | if (do_wake) | |
140 | wake_up(&mdev->ee_wait); | |
141 | ||
142 | if (do_al_complete_io) | |
181286ad | 143 | drbd_al_complete_io(mdev, &i); |
b411b363 | 144 | |
0625ac19 | 145 | wake_asender(mdev->tconn); |
b411b363 | 146 | put_ldev(mdev); |
45bb912b | 147 | } |
b411b363 | 148 | |
45bb912b LE |
149 | /* writes on behalf of the partner, or resync writes, |
150 | * "submitted" by the receiver. | |
151 | */ | |
fcefa62e | 152 | void drbd_peer_request_endio(struct bio *bio, int error) |
45bb912b | 153 | { |
db830c46 | 154 | struct drbd_peer_request *peer_req = bio->bi_private; |
a21e9298 | 155 | struct drbd_conf *mdev = peer_req->w.mdev; |
45bb912b LE |
156 | int uptodate = bio_flagged(bio, BIO_UPTODATE); |
157 | int is_write = bio_data_dir(bio) == WRITE; | |
158 | ||
07194272 | 159 | if (error && __ratelimit(&drbd_ratelimit_state)) |
45bb912b LE |
160 | dev_warn(DEV, "%s: error=%d s=%llus\n", |
161 | is_write ? "write" : "read", error, | |
db830c46 | 162 | (unsigned long long)peer_req->i.sector); |
45bb912b | 163 | if (!error && !uptodate) { |
07194272 LE |
164 | if (__ratelimit(&drbd_ratelimit_state)) |
165 | dev_warn(DEV, "%s: setting error to -EIO s=%llus\n", | |
166 | is_write ? "write" : "read", | |
db830c46 | 167 | (unsigned long long)peer_req->i.sector); |
45bb912b LE |
168 | /* strange behavior of some lower level drivers... |
169 | * fail the request by clearing the uptodate flag, | |
170 | * but do not return any error?! */ | |
171 | error = -EIO; | |
172 | } | |
173 | ||
174 | if (error) | |
db830c46 | 175 | set_bit(__EE_WAS_ERROR, &peer_req->flags); |
45bb912b LE |
176 | |
177 | bio_put(bio); /* no need for the bio anymore */ | |
db830c46 | 178 | if (atomic_dec_and_test(&peer_req->pending_bios)) { |
45bb912b | 179 | if (is_write) |
db830c46 | 180 | drbd_endio_write_sec_final(peer_req); |
45bb912b | 181 | else |
db830c46 | 182 | drbd_endio_read_sec_final(peer_req); |
45bb912b | 183 | } |
b411b363 PR |
184 | } |
185 | ||
186 | /* read, readA or write requests on R_PRIMARY coming from drbd_make_request | |
187 | */ | |
fcefa62e | 188 | void drbd_request_endio(struct bio *bio, int error) |
b411b363 | 189 | { |
a115413d | 190 | unsigned long flags; |
b411b363 | 191 | struct drbd_request *req = bio->bi_private; |
a21e9298 | 192 | struct drbd_conf *mdev = req->w.mdev; |
a115413d | 193 | struct bio_and_error m; |
b411b363 PR |
194 | enum drbd_req_event what; |
195 | int uptodate = bio_flagged(bio, BIO_UPTODATE); | |
196 | ||
b411b363 PR |
197 | if (!error && !uptodate) { |
198 | dev_warn(DEV, "p %s: setting error to -EIO\n", | |
199 | bio_data_dir(bio) == WRITE ? "write" : "read"); | |
200 | /* strange behavior of some lower level drivers... | |
201 | * fail the request by clearing the uptodate flag, | |
202 | * but do not return any error?! */ | |
203 | error = -EIO; | |
204 | } | |
205 | ||
b411b363 PR |
206 | /* to avoid recursion in __req_mod */ |
207 | if (unlikely(error)) { | |
208 | what = (bio_data_dir(bio) == WRITE) | |
8554df1c | 209 | ? WRITE_COMPLETED_WITH_ERROR |
5c3c7e64 | 210 | : (bio_rw(bio) == READ) |
8554df1c AG |
211 | ? READ_COMPLETED_WITH_ERROR |
212 | : READ_AHEAD_COMPLETED_WITH_ERROR; | |
b411b363 | 213 | } else |
8554df1c | 214 | what = COMPLETED_OK; |
b411b363 PR |
215 | |
216 | bio_put(req->private_bio); | |
217 | req->private_bio = ERR_PTR(error); | |
218 | ||
a115413d | 219 | /* not req_mod(), we need irqsave here! */ |
87eeee41 | 220 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
a115413d | 221 | __req_mod(req, what, &m); |
87eeee41 | 222 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
a115413d LE |
223 | |
224 | if (m.bio) | |
225 | complete_master_bio(mdev, &m); | |
b411b363 PR |
226 | } |
227 | ||
99920dc5 | 228 | int w_read_retry_remote(struct drbd_work *w, int cancel) |
b411b363 PR |
229 | { |
230 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 231 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
232 | |
233 | /* We should not detach for read io-error, | |
234 | * but try to WRITE the P_DATA_REPLY to the failed location, | |
235 | * to give the disk the chance to relocate that block */ | |
236 | ||
87eeee41 | 237 | spin_lock_irq(&mdev->tconn->req_lock); |
d255e5ff | 238 | if (cancel || mdev->state.pdsk != D_UP_TO_DATE) { |
8554df1c | 239 | _req_mod(req, READ_RETRY_REMOTE_CANCELED); |
87eeee41 | 240 | spin_unlock_irq(&mdev->tconn->req_lock); |
99920dc5 | 241 | return 0; |
b411b363 | 242 | } |
87eeee41 | 243 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 244 | |
00d56944 | 245 | return w_send_read_req(w, 0); |
b411b363 PR |
246 | } |
247 | ||
f6ffca9f | 248 | void drbd_csum_ee(struct drbd_conf *mdev, struct crypto_hash *tfm, |
db830c46 | 249 | struct drbd_peer_request *peer_req, void *digest) |
45bb912b LE |
250 | { |
251 | struct hash_desc desc; | |
252 | struct scatterlist sg; | |
db830c46 | 253 | struct page *page = peer_req->pages; |
45bb912b LE |
254 | struct page *tmp; |
255 | unsigned len; | |
256 | ||
257 | desc.tfm = tfm; | |
258 | desc.flags = 0; | |
259 | ||
260 | sg_init_table(&sg, 1); | |
261 | crypto_hash_init(&desc); | |
262 | ||
263 | while ((tmp = page_chain_next(page))) { | |
264 | /* all but the last page will be fully used */ | |
265 | sg_set_page(&sg, page, PAGE_SIZE, 0); | |
266 | crypto_hash_update(&desc, &sg, sg.length); | |
267 | page = tmp; | |
268 | } | |
269 | /* and now the last, possibly only partially used page */ | |
db830c46 | 270 | len = peer_req->i.size & (PAGE_SIZE - 1); |
45bb912b LE |
271 | sg_set_page(&sg, page, len ?: PAGE_SIZE, 0); |
272 | crypto_hash_update(&desc, &sg, sg.length); | |
273 | crypto_hash_final(&desc, digest); | |
274 | } | |
275 | ||
276 | void drbd_csum_bio(struct drbd_conf *mdev, struct crypto_hash *tfm, struct bio *bio, void *digest) | |
b411b363 PR |
277 | { |
278 | struct hash_desc desc; | |
279 | struct scatterlist sg; | |
280 | struct bio_vec *bvec; | |
281 | int i; | |
282 | ||
283 | desc.tfm = tfm; | |
284 | desc.flags = 0; | |
285 | ||
286 | sg_init_table(&sg, 1); | |
287 | crypto_hash_init(&desc); | |
288 | ||
289 | __bio_for_each_segment(bvec, bio, i, 0) { | |
290 | sg_set_page(&sg, bvec->bv_page, bvec->bv_len, bvec->bv_offset); | |
291 | crypto_hash_update(&desc, &sg, sg.length); | |
292 | } | |
293 | crypto_hash_final(&desc, digest); | |
294 | } | |
295 | ||
9676c760 | 296 | /* MAYBE merge common code with w_e_end_ov_req */ |
99920dc5 | 297 | static int w_e_send_csum(struct drbd_work *w, int cancel) |
b411b363 | 298 | { |
00d56944 PR |
299 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
300 | struct drbd_conf *mdev = w->mdev; | |
b411b363 PR |
301 | int digest_size; |
302 | void *digest; | |
99920dc5 | 303 | int err = 0; |
b411b363 | 304 | |
53ea4331 LE |
305 | if (unlikely(cancel)) |
306 | goto out; | |
b411b363 | 307 | |
9676c760 | 308 | if (unlikely((peer_req->flags & EE_WAS_ERROR) != 0)) |
53ea4331 | 309 | goto out; |
b411b363 | 310 | |
f399002e | 311 | digest_size = crypto_hash_digestsize(mdev->tconn->csums_tfm); |
53ea4331 LE |
312 | digest = kmalloc(digest_size, GFP_NOIO); |
313 | if (digest) { | |
db830c46 AG |
314 | sector_t sector = peer_req->i.sector; |
315 | unsigned int size = peer_req->i.size; | |
f399002e | 316 | drbd_csum_ee(mdev, mdev->tconn->csums_tfm, peer_req, digest); |
9676c760 | 317 | /* Free peer_req and pages before send. |
53ea4331 LE |
318 | * In case we block on congestion, we could otherwise run into |
319 | * some distributed deadlock, if the other side blocks on | |
320 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 321 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
3967deb1 | 322 | drbd_free_peer_req(mdev, peer_req); |
db830c46 | 323 | peer_req = NULL; |
53ea4331 | 324 | inc_rs_pending(mdev); |
99920dc5 | 325 | err = drbd_send_drequest_csum(mdev, sector, size, |
db1b0b72 AG |
326 | digest, digest_size, |
327 | P_CSUM_RS_REQUEST); | |
53ea4331 LE |
328 | kfree(digest); |
329 | } else { | |
330 | dev_err(DEV, "kmalloc() of digest failed.\n"); | |
99920dc5 | 331 | err = -ENOMEM; |
53ea4331 | 332 | } |
b411b363 | 333 | |
53ea4331 | 334 | out: |
db830c46 | 335 | if (peer_req) |
3967deb1 | 336 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 337 | |
99920dc5 | 338 | if (unlikely(err)) |
b411b363 | 339 | dev_err(DEV, "drbd_send_drequest(..., csum) failed\n"); |
99920dc5 | 340 | return err; |
b411b363 PR |
341 | } |
342 | ||
343 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
344 | ||
345 | static int read_for_csum(struct drbd_conf *mdev, sector_t sector, int size) | |
346 | { | |
db830c46 | 347 | struct drbd_peer_request *peer_req; |
b411b363 PR |
348 | |
349 | if (!get_ldev(mdev)) | |
80a40e43 | 350 | return -EIO; |
b411b363 | 351 | |
e3555d85 | 352 | if (drbd_rs_should_slow_down(mdev, sector)) |
0f0601f4 LE |
353 | goto defer; |
354 | ||
b411b363 PR |
355 | /* GFP_TRY, because if there is no memory available right now, this may |
356 | * be rescheduled for later. It is "only" background resync, after all. */ | |
0db55363 AG |
357 | peer_req = drbd_alloc_peer_req(mdev, ID_SYNCER /* unused */, sector, |
358 | size, GFP_TRY); | |
db830c46 | 359 | if (!peer_req) |
80a40e43 | 360 | goto defer; |
b411b363 | 361 | |
db830c46 | 362 | peer_req->w.cb = w_e_send_csum; |
87eeee41 | 363 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 364 | list_add(&peer_req->w.list, &mdev->read_ee); |
87eeee41 | 365 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 366 | |
0f0601f4 | 367 | atomic_add(size >> 9, &mdev->rs_sect_ev); |
fbe29dec | 368 | if (drbd_submit_peer_request(mdev, peer_req, READ, DRBD_FAULT_RS_RD) == 0) |
80a40e43 | 369 | return 0; |
b411b363 | 370 | |
10f6d992 LE |
371 | /* If it failed because of ENOMEM, retry should help. If it failed |
372 | * because bio_add_page failed (probably broken lower level driver), | |
373 | * retry may or may not help. | |
374 | * If it does not, you may need to force disconnect. */ | |
87eeee41 | 375 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 376 | list_del(&peer_req->w.list); |
87eeee41 | 377 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 | 378 | |
3967deb1 | 379 | drbd_free_peer_req(mdev, peer_req); |
80a40e43 | 380 | defer: |
45bb912b | 381 | put_ldev(mdev); |
80a40e43 | 382 | return -EAGAIN; |
b411b363 PR |
383 | } |
384 | ||
99920dc5 | 385 | int w_resync_timer(struct drbd_work *w, int cancel) |
b411b363 | 386 | { |
00d56944 | 387 | struct drbd_conf *mdev = w->mdev; |
63106d3c PR |
388 | switch (mdev->state.conn) { |
389 | case C_VERIFY_S: | |
00d56944 | 390 | w_make_ov_request(w, cancel); |
63106d3c PR |
391 | break; |
392 | case C_SYNC_TARGET: | |
00d56944 | 393 | w_make_resync_request(w, cancel); |
63106d3c | 394 | break; |
b411b363 PR |
395 | } |
396 | ||
99920dc5 | 397 | return 0; |
794abb75 PR |
398 | } |
399 | ||
400 | void resync_timer_fn(unsigned long data) | |
401 | { | |
402 | struct drbd_conf *mdev = (struct drbd_conf *) data; | |
403 | ||
404 | if (list_empty(&mdev->resync_work.list)) | |
e42325a5 | 405 | drbd_queue_work(&mdev->tconn->data.work, &mdev->resync_work); |
b411b363 PR |
406 | } |
407 | ||
778f271d PR |
408 | static void fifo_set(struct fifo_buffer *fb, int value) |
409 | { | |
410 | int i; | |
411 | ||
412 | for (i = 0; i < fb->size; i++) | |
f10f2623 | 413 | fb->values[i] = value; |
778f271d PR |
414 | } |
415 | ||
416 | static int fifo_push(struct fifo_buffer *fb, int value) | |
417 | { | |
418 | int ov; | |
419 | ||
420 | ov = fb->values[fb->head_index]; | |
421 | fb->values[fb->head_index++] = value; | |
422 | ||
423 | if (fb->head_index >= fb->size) | |
424 | fb->head_index = 0; | |
425 | ||
426 | return ov; | |
427 | } | |
428 | ||
429 | static void fifo_add_val(struct fifo_buffer *fb, int value) | |
430 | { | |
431 | int i; | |
432 | ||
433 | for (i = 0; i < fb->size; i++) | |
434 | fb->values[i] += value; | |
435 | } | |
436 | ||
9d77a5fe | 437 | static int drbd_rs_controller(struct drbd_conf *mdev) |
778f271d PR |
438 | { |
439 | unsigned int sect_in; /* Number of sectors that came in since the last turn */ | |
440 | unsigned int want; /* The number of sectors we want in the proxy */ | |
441 | int req_sect; /* Number of sectors to request in this turn */ | |
442 | int correction; /* Number of sectors more we need in the proxy*/ | |
443 | int cps; /* correction per invocation of drbd_rs_controller() */ | |
444 | int steps; /* Number of time steps to plan ahead */ | |
445 | int curr_corr; | |
446 | int max_sect; | |
447 | ||
448 | sect_in = atomic_xchg(&mdev->rs_sect_in, 0); /* Number of sectors that came in */ | |
449 | mdev->rs_in_flight -= sect_in; | |
450 | ||
451 | spin_lock(&mdev->peer_seq_lock); /* get an atomic view on mdev->rs_plan_s */ | |
452 | ||
f399002e | 453 | steps = mdev->rs_plan_s.size; /* (mdev->ldev->dc.c_plan_ahead * 10 * SLEEP_TIME) / HZ; */ |
778f271d PR |
454 | |
455 | if (mdev->rs_in_flight + sect_in == 0) { /* At start of resync */ | |
f399002e | 456 | want = ((mdev->ldev->dc.resync_rate * 2 * SLEEP_TIME) / HZ) * steps; |
778f271d | 457 | } else { /* normal path */ |
f399002e LE |
458 | want = mdev->ldev->dc.c_fill_target ? mdev->ldev->dc.c_fill_target : |
459 | sect_in * mdev->ldev->dc.c_delay_target * HZ / (SLEEP_TIME * 10); | |
778f271d PR |
460 | } |
461 | ||
462 | correction = want - mdev->rs_in_flight - mdev->rs_planed; | |
463 | ||
464 | /* Plan ahead */ | |
465 | cps = correction / steps; | |
466 | fifo_add_val(&mdev->rs_plan_s, cps); | |
467 | mdev->rs_planed += cps * steps; | |
468 | ||
469 | /* What we do in this step */ | |
470 | curr_corr = fifo_push(&mdev->rs_plan_s, 0); | |
471 | spin_unlock(&mdev->peer_seq_lock); | |
472 | mdev->rs_planed -= curr_corr; | |
473 | ||
474 | req_sect = sect_in + curr_corr; | |
475 | if (req_sect < 0) | |
476 | req_sect = 0; | |
477 | ||
f399002e | 478 | max_sect = (mdev->ldev->dc.c_max_rate * 2 * SLEEP_TIME) / HZ; |
778f271d PR |
479 | if (req_sect > max_sect) |
480 | req_sect = max_sect; | |
481 | ||
482 | /* | |
483 | dev_warn(DEV, "si=%u if=%d wa=%u co=%d st=%d cps=%d pl=%d cc=%d rs=%d\n", | |
484 | sect_in, mdev->rs_in_flight, want, correction, | |
485 | steps, cps, mdev->rs_planed, curr_corr, req_sect); | |
486 | */ | |
487 | ||
488 | return req_sect; | |
489 | } | |
490 | ||
9d77a5fe | 491 | static int drbd_rs_number_requests(struct drbd_conf *mdev) |
e65f440d LE |
492 | { |
493 | int number; | |
f399002e | 494 | if (mdev->rs_plan_s.size) { /* mdev->ldev->dc.c_plan_ahead */ |
e65f440d LE |
495 | number = drbd_rs_controller(mdev) >> (BM_BLOCK_SHIFT - 9); |
496 | mdev->c_sync_rate = number * HZ * (BM_BLOCK_SIZE / 1024) / SLEEP_TIME; | |
497 | } else { | |
f399002e | 498 | mdev->c_sync_rate = mdev->ldev->dc.resync_rate; |
e65f440d LE |
499 | number = SLEEP_TIME * mdev->c_sync_rate / ((BM_BLOCK_SIZE / 1024) * HZ); |
500 | } | |
501 | ||
e65f440d LE |
502 | /* ignore the amount of pending requests, the resync controller should |
503 | * throttle down to incoming reply rate soon enough anyways. */ | |
504 | return number; | |
505 | } | |
506 | ||
99920dc5 | 507 | int w_make_resync_request(struct drbd_work *w, int cancel) |
b411b363 | 508 | { |
00d56944 | 509 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
510 | unsigned long bit; |
511 | sector_t sector; | |
512 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); | |
1816a2b4 | 513 | int max_bio_size; |
e65f440d | 514 | int number, rollback_i, size; |
b411b363 | 515 | int align, queued, sndbuf; |
0f0601f4 | 516 | int i = 0; |
b411b363 PR |
517 | |
518 | if (unlikely(cancel)) | |
99920dc5 | 519 | return 0; |
b411b363 | 520 | |
af85e8e8 LE |
521 | if (mdev->rs_total == 0) { |
522 | /* empty resync? */ | |
523 | drbd_resync_finished(mdev); | |
99920dc5 | 524 | return 0; |
af85e8e8 LE |
525 | } |
526 | ||
b411b363 PR |
527 | if (!get_ldev(mdev)) { |
528 | /* Since we only need to access mdev->rsync a | |
529 | get_ldev_if_state(mdev,D_FAILED) would be sufficient, but | |
530 | to continue resync with a broken disk makes no sense at | |
531 | all */ | |
532 | dev_err(DEV, "Disk broke down during resync!\n"); | |
99920dc5 | 533 | return 0; |
b411b363 PR |
534 | } |
535 | ||
0cfdd247 | 536 | max_bio_size = queue_max_hw_sectors(mdev->rq_queue) << 9; |
e65f440d LE |
537 | number = drbd_rs_number_requests(mdev); |
538 | if (number == 0) | |
0f0601f4 | 539 | goto requeue; |
b411b363 | 540 | |
b411b363 PR |
541 | for (i = 0; i < number; i++) { |
542 | /* Stop generating RS requests, when half of the send buffer is filled */ | |
e42325a5 PR |
543 | mutex_lock(&mdev->tconn->data.mutex); |
544 | if (mdev->tconn->data.socket) { | |
545 | queued = mdev->tconn->data.socket->sk->sk_wmem_queued; | |
546 | sndbuf = mdev->tconn->data.socket->sk->sk_sndbuf; | |
b411b363 PR |
547 | } else { |
548 | queued = 1; | |
549 | sndbuf = 0; | |
550 | } | |
e42325a5 | 551 | mutex_unlock(&mdev->tconn->data.mutex); |
b411b363 PR |
552 | if (queued > sndbuf / 2) |
553 | goto requeue; | |
554 | ||
555 | next_sector: | |
556 | size = BM_BLOCK_SIZE; | |
557 | bit = drbd_bm_find_next(mdev, mdev->bm_resync_fo); | |
558 | ||
4b0715f0 | 559 | if (bit == DRBD_END_OF_BITMAP) { |
b411b363 | 560 | mdev->bm_resync_fo = drbd_bm_bits(mdev); |
b411b363 | 561 | put_ldev(mdev); |
99920dc5 | 562 | return 0; |
b411b363 PR |
563 | } |
564 | ||
565 | sector = BM_BIT_TO_SECT(bit); | |
566 | ||
e3555d85 PR |
567 | if (drbd_rs_should_slow_down(mdev, sector) || |
568 | drbd_try_rs_begin_io(mdev, sector)) { | |
b411b363 PR |
569 | mdev->bm_resync_fo = bit; |
570 | goto requeue; | |
571 | } | |
572 | mdev->bm_resync_fo = bit + 1; | |
573 | ||
574 | if (unlikely(drbd_bm_test_bit(mdev, bit) == 0)) { | |
575 | drbd_rs_complete_io(mdev, sector); | |
576 | goto next_sector; | |
577 | } | |
578 | ||
1816a2b4 | 579 | #if DRBD_MAX_BIO_SIZE > BM_BLOCK_SIZE |
b411b363 PR |
580 | /* try to find some adjacent bits. |
581 | * we stop if we have already the maximum req size. | |
582 | * | |
583 | * Additionally always align bigger requests, in order to | |
584 | * be prepared for all stripe sizes of software RAIDs. | |
b411b363 PR |
585 | */ |
586 | align = 1; | |
d207450c | 587 | rollback_i = i; |
b411b363 | 588 | for (;;) { |
1816a2b4 | 589 | if (size + BM_BLOCK_SIZE > max_bio_size) |
b411b363 PR |
590 | break; |
591 | ||
592 | /* Be always aligned */ | |
593 | if (sector & ((1<<(align+3))-1)) | |
594 | break; | |
595 | ||
596 | /* do not cross extent boundaries */ | |
597 | if (((bit+1) & BM_BLOCKS_PER_BM_EXT_MASK) == 0) | |
598 | break; | |
599 | /* now, is it actually dirty, after all? | |
600 | * caution, drbd_bm_test_bit is tri-state for some | |
601 | * obscure reason; ( b == 0 ) would get the out-of-band | |
602 | * only accidentally right because of the "oddly sized" | |
603 | * adjustment below */ | |
604 | if (drbd_bm_test_bit(mdev, bit+1) != 1) | |
605 | break; | |
606 | bit++; | |
607 | size += BM_BLOCK_SIZE; | |
608 | if ((BM_BLOCK_SIZE << align) <= size) | |
609 | align++; | |
610 | i++; | |
611 | } | |
612 | /* if we merged some, | |
613 | * reset the offset to start the next drbd_bm_find_next from */ | |
614 | if (size > BM_BLOCK_SIZE) | |
615 | mdev->bm_resync_fo = bit + 1; | |
616 | #endif | |
617 | ||
618 | /* adjust very last sectors, in case we are oddly sized */ | |
619 | if (sector + (size>>9) > capacity) | |
620 | size = (capacity-sector)<<9; | |
f399002e | 621 | if (mdev->tconn->agreed_pro_version >= 89 && mdev->tconn->csums_tfm) { |
b411b363 | 622 | switch (read_for_csum(mdev, sector, size)) { |
80a40e43 | 623 | case -EIO: /* Disk failure */ |
b411b363 | 624 | put_ldev(mdev); |
99920dc5 | 625 | return -EIO; |
80a40e43 | 626 | case -EAGAIN: /* allocation failed, or ldev busy */ |
b411b363 PR |
627 | drbd_rs_complete_io(mdev, sector); |
628 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
d207450c | 629 | i = rollback_i; |
b411b363 | 630 | goto requeue; |
80a40e43 LE |
631 | case 0: |
632 | /* everything ok */ | |
633 | break; | |
634 | default: | |
635 | BUG(); | |
b411b363 PR |
636 | } |
637 | } else { | |
99920dc5 AG |
638 | int err; |
639 | ||
b411b363 | 640 | inc_rs_pending(mdev); |
99920dc5 AG |
641 | err = drbd_send_drequest(mdev, P_RS_DATA_REQUEST, |
642 | sector, size, ID_SYNCER); | |
643 | if (err) { | |
b411b363 PR |
644 | dev_err(DEV, "drbd_send_drequest() failed, aborting...\n"); |
645 | dec_rs_pending(mdev); | |
646 | put_ldev(mdev); | |
99920dc5 | 647 | return err; |
b411b363 PR |
648 | } |
649 | } | |
650 | } | |
651 | ||
652 | if (mdev->bm_resync_fo >= drbd_bm_bits(mdev)) { | |
653 | /* last syncer _request_ was sent, | |
654 | * but the P_RS_DATA_REPLY not yet received. sync will end (and | |
655 | * next sync group will resume), as soon as we receive the last | |
656 | * resync data block, and the last bit is cleared. | |
657 | * until then resync "work" is "inactive" ... | |
658 | */ | |
b411b363 | 659 | put_ldev(mdev); |
99920dc5 | 660 | return 0; |
b411b363 PR |
661 | } |
662 | ||
663 | requeue: | |
778f271d | 664 | mdev->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
b411b363 PR |
665 | mod_timer(&mdev->resync_timer, jiffies + SLEEP_TIME); |
666 | put_ldev(mdev); | |
99920dc5 | 667 | return 0; |
b411b363 PR |
668 | } |
669 | ||
00d56944 | 670 | static int w_make_ov_request(struct drbd_work *w, int cancel) |
b411b363 | 671 | { |
00d56944 | 672 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
673 | int number, i, size; |
674 | sector_t sector; | |
675 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); | |
676 | ||
677 | if (unlikely(cancel)) | |
678 | return 1; | |
679 | ||
2649f080 | 680 | number = drbd_rs_number_requests(mdev); |
b411b363 PR |
681 | |
682 | sector = mdev->ov_position; | |
683 | for (i = 0; i < number; i++) { | |
684 | if (sector >= capacity) { | |
b411b363 PR |
685 | return 1; |
686 | } | |
687 | ||
688 | size = BM_BLOCK_SIZE; | |
689 | ||
e3555d85 PR |
690 | if (drbd_rs_should_slow_down(mdev, sector) || |
691 | drbd_try_rs_begin_io(mdev, sector)) { | |
b411b363 PR |
692 | mdev->ov_position = sector; |
693 | goto requeue; | |
694 | } | |
695 | ||
696 | if (sector + (size>>9) > capacity) | |
697 | size = (capacity-sector)<<9; | |
698 | ||
699 | inc_rs_pending(mdev); | |
5b9f499c | 700 | if (drbd_send_ov_request(mdev, sector, size)) { |
b411b363 PR |
701 | dec_rs_pending(mdev); |
702 | return 0; | |
703 | } | |
704 | sector += BM_SECT_PER_BIT; | |
705 | } | |
706 | mdev->ov_position = sector; | |
707 | ||
708 | requeue: | |
2649f080 | 709 | mdev->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
b411b363 PR |
710 | mod_timer(&mdev->resync_timer, jiffies + SLEEP_TIME); |
711 | return 1; | |
712 | } | |
713 | ||
99920dc5 | 714 | int w_ov_finished(struct drbd_work *w, int cancel) |
b411b363 | 715 | { |
00d56944 | 716 | struct drbd_conf *mdev = w->mdev; |
b411b363 | 717 | kfree(w); |
8f7bed77 | 718 | ov_out_of_sync_print(mdev); |
b411b363 PR |
719 | drbd_resync_finished(mdev); |
720 | ||
99920dc5 | 721 | return 0; |
b411b363 PR |
722 | } |
723 | ||
99920dc5 | 724 | static int w_resync_finished(struct drbd_work *w, int cancel) |
b411b363 | 725 | { |
00d56944 | 726 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
727 | kfree(w); |
728 | ||
729 | drbd_resync_finished(mdev); | |
730 | ||
99920dc5 | 731 | return 0; |
b411b363 PR |
732 | } |
733 | ||
af85e8e8 LE |
734 | static void ping_peer(struct drbd_conf *mdev) |
735 | { | |
2a67d8b9 PR |
736 | struct drbd_tconn *tconn = mdev->tconn; |
737 | ||
738 | clear_bit(GOT_PING_ACK, &tconn->flags); | |
739 | request_ping(tconn); | |
740 | wait_event(tconn->ping_wait, | |
741 | test_bit(GOT_PING_ACK, &tconn->flags) || mdev->state.conn < C_CONNECTED); | |
af85e8e8 LE |
742 | } |
743 | ||
b411b363 PR |
744 | int drbd_resync_finished(struct drbd_conf *mdev) |
745 | { | |
746 | unsigned long db, dt, dbdt; | |
747 | unsigned long n_oos; | |
748 | union drbd_state os, ns; | |
749 | struct drbd_work *w; | |
750 | char *khelper_cmd = NULL; | |
26525618 | 751 | int verify_done = 0; |
b411b363 PR |
752 | |
753 | /* Remove all elements from the resync LRU. Since future actions | |
754 | * might set bits in the (main) bitmap, then the entries in the | |
755 | * resync LRU would be wrong. */ | |
756 | if (drbd_rs_del_all(mdev)) { | |
757 | /* In case this is not possible now, most probably because | |
758 | * there are P_RS_DATA_REPLY Packets lingering on the worker's | |
759 | * queue (or even the read operations for those packets | |
760 | * is not finished by now). Retry in 100ms. */ | |
761 | ||
20ee6390 | 762 | schedule_timeout_interruptible(HZ / 10); |
b411b363 PR |
763 | w = kmalloc(sizeof(struct drbd_work), GFP_ATOMIC); |
764 | if (w) { | |
765 | w->cb = w_resync_finished; | |
e42325a5 | 766 | drbd_queue_work(&mdev->tconn->data.work, w); |
b411b363 PR |
767 | return 1; |
768 | } | |
769 | dev_err(DEV, "Warn failed to drbd_rs_del_all() and to kmalloc(w).\n"); | |
770 | } | |
771 | ||
772 | dt = (jiffies - mdev->rs_start - mdev->rs_paused) / HZ; | |
773 | if (dt <= 0) | |
774 | dt = 1; | |
775 | db = mdev->rs_total; | |
776 | dbdt = Bit2KB(db/dt); | |
777 | mdev->rs_paused /= HZ; | |
778 | ||
779 | if (!get_ldev(mdev)) | |
780 | goto out; | |
781 | ||
af85e8e8 LE |
782 | ping_peer(mdev); |
783 | ||
87eeee41 | 784 | spin_lock_irq(&mdev->tconn->req_lock); |
78bae59b | 785 | os = drbd_read_state(mdev); |
b411b363 | 786 | |
26525618 LE |
787 | verify_done = (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T); |
788 | ||
b411b363 PR |
789 | /* This protects us against multiple calls (that can happen in the presence |
790 | of application IO), and against connectivity loss just before we arrive here. */ | |
791 | if (os.conn <= C_CONNECTED) | |
792 | goto out_unlock; | |
793 | ||
794 | ns = os; | |
795 | ns.conn = C_CONNECTED; | |
796 | ||
797 | dev_info(DEV, "%s done (total %lu sec; paused %lu sec; %lu K/sec)\n", | |
26525618 | 798 | verify_done ? "Online verify " : "Resync", |
b411b363 PR |
799 | dt + mdev->rs_paused, mdev->rs_paused, dbdt); |
800 | ||
801 | n_oos = drbd_bm_total_weight(mdev); | |
802 | ||
803 | if (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) { | |
804 | if (n_oos) { | |
805 | dev_alert(DEV, "Online verify found %lu %dk block out of sync!\n", | |
806 | n_oos, Bit2KB(1)); | |
807 | khelper_cmd = "out-of-sync"; | |
808 | } | |
809 | } else { | |
810 | D_ASSERT((n_oos - mdev->rs_failed) == 0); | |
811 | ||
812 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) | |
813 | khelper_cmd = "after-resync-target"; | |
814 | ||
f399002e | 815 | if (mdev->tconn->csums_tfm && mdev->rs_total) { |
b411b363 PR |
816 | const unsigned long s = mdev->rs_same_csum; |
817 | const unsigned long t = mdev->rs_total; | |
818 | const int ratio = | |
819 | (t == 0) ? 0 : | |
820 | (t < 100000) ? ((s*100)/t) : (s/(t/100)); | |
24c4830c | 821 | dev_info(DEV, "%u %% had equal checksums, eliminated: %luK; " |
b411b363 PR |
822 | "transferred %luK total %luK\n", |
823 | ratio, | |
824 | Bit2KB(mdev->rs_same_csum), | |
825 | Bit2KB(mdev->rs_total - mdev->rs_same_csum), | |
826 | Bit2KB(mdev->rs_total)); | |
827 | } | |
828 | } | |
829 | ||
830 | if (mdev->rs_failed) { | |
831 | dev_info(DEV, " %lu failed blocks\n", mdev->rs_failed); | |
832 | ||
833 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
834 | ns.disk = D_INCONSISTENT; | |
835 | ns.pdsk = D_UP_TO_DATE; | |
836 | } else { | |
837 | ns.disk = D_UP_TO_DATE; | |
838 | ns.pdsk = D_INCONSISTENT; | |
839 | } | |
840 | } else { | |
841 | ns.disk = D_UP_TO_DATE; | |
842 | ns.pdsk = D_UP_TO_DATE; | |
843 | ||
844 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
845 | if (mdev->p_uuid) { | |
846 | int i; | |
847 | for (i = UI_BITMAP ; i <= UI_HISTORY_END ; i++) | |
848 | _drbd_uuid_set(mdev, i, mdev->p_uuid[i]); | |
849 | drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_CURRENT]); | |
850 | _drbd_uuid_set(mdev, UI_CURRENT, mdev->p_uuid[UI_CURRENT]); | |
851 | } else { | |
852 | dev_err(DEV, "mdev->p_uuid is NULL! BUG\n"); | |
853 | } | |
854 | } | |
855 | ||
62b0da3a LE |
856 | if (!(os.conn == C_VERIFY_S || os.conn == C_VERIFY_T)) { |
857 | /* for verify runs, we don't update uuids here, | |
858 | * so there would be nothing to report. */ | |
859 | drbd_uuid_set_bm(mdev, 0UL); | |
860 | drbd_print_uuids(mdev, "updated UUIDs"); | |
861 | if (mdev->p_uuid) { | |
862 | /* Now the two UUID sets are equal, update what we | |
863 | * know of the peer. */ | |
864 | int i; | |
865 | for (i = UI_CURRENT ; i <= UI_HISTORY_END ; i++) | |
866 | mdev->p_uuid[i] = mdev->ldev->md.uuid[i]; | |
867 | } | |
b411b363 PR |
868 | } |
869 | } | |
870 | ||
871 | _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); | |
872 | out_unlock: | |
87eeee41 | 873 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
874 | put_ldev(mdev); |
875 | out: | |
876 | mdev->rs_total = 0; | |
877 | mdev->rs_failed = 0; | |
878 | mdev->rs_paused = 0; | |
26525618 LE |
879 | if (verify_done) |
880 | mdev->ov_start_sector = 0; | |
b411b363 | 881 | |
13d42685 LE |
882 | drbd_md_sync(mdev); |
883 | ||
b411b363 PR |
884 | if (khelper_cmd) |
885 | drbd_khelper(mdev, khelper_cmd); | |
886 | ||
887 | return 1; | |
888 | } | |
889 | ||
890 | /* helper */ | |
db830c46 | 891 | static void move_to_net_ee_or_free(struct drbd_conf *mdev, struct drbd_peer_request *peer_req) |
b411b363 | 892 | { |
045417f7 | 893 | if (drbd_peer_req_has_active_page(peer_req)) { |
b411b363 | 894 | /* This might happen if sendpage() has not finished */ |
db830c46 | 895 | int i = (peer_req->i.size + PAGE_SIZE -1) >> PAGE_SHIFT; |
435f0740 LE |
896 | atomic_add(i, &mdev->pp_in_use_by_net); |
897 | atomic_sub(i, &mdev->pp_in_use); | |
87eeee41 | 898 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 899 | list_add_tail(&peer_req->w.list, &mdev->net_ee); |
87eeee41 | 900 | spin_unlock_irq(&mdev->tconn->req_lock); |
435f0740 | 901 | wake_up(&drbd_pp_wait); |
b411b363 | 902 | } else |
3967deb1 | 903 | drbd_free_peer_req(mdev, peer_req); |
b411b363 PR |
904 | } |
905 | ||
906 | /** | |
907 | * w_e_end_data_req() - Worker callback, to send a P_DATA_REPLY packet in response to a P_DATA_REQUEST | |
908 | * @mdev: DRBD device. | |
909 | * @w: work object. | |
910 | * @cancel: The connection will be closed anyways | |
911 | */ | |
99920dc5 | 912 | int w_e_end_data_req(struct drbd_work *w, int cancel) |
b411b363 | 913 | { |
db830c46 | 914 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 915 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 916 | int err; |
b411b363 PR |
917 | |
918 | if (unlikely(cancel)) { | |
3967deb1 | 919 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 920 | dec_unacked(mdev); |
99920dc5 | 921 | return 0; |
b411b363 PR |
922 | } |
923 | ||
db830c46 | 924 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
99920dc5 | 925 | err = drbd_send_block(mdev, P_DATA_REPLY, peer_req); |
b411b363 PR |
926 | } else { |
927 | if (__ratelimit(&drbd_ratelimit_state)) | |
928 | dev_err(DEV, "Sending NegDReply. sector=%llus.\n", | |
db830c46 | 929 | (unsigned long long)peer_req->i.sector); |
b411b363 | 930 | |
99920dc5 | 931 | err = drbd_send_ack(mdev, P_NEG_DREPLY, peer_req); |
b411b363 PR |
932 | } |
933 | ||
934 | dec_unacked(mdev); | |
935 | ||
db830c46 | 936 | move_to_net_ee_or_free(mdev, peer_req); |
b411b363 | 937 | |
99920dc5 | 938 | if (unlikely(err)) |
b411b363 | 939 | dev_err(DEV, "drbd_send_block() failed\n"); |
99920dc5 | 940 | return err; |
b411b363 PR |
941 | } |
942 | ||
943 | /** | |
944 | * w_e_end_rsdata_req() - Worker callback to send a P_RS_DATA_REPLY packet in response to a P_RS_DATA_REQUESTRS | |
945 | * @mdev: DRBD device. | |
946 | * @w: work object. | |
947 | * @cancel: The connection will be closed anyways | |
948 | */ | |
99920dc5 | 949 | int w_e_end_rsdata_req(struct drbd_work *w, int cancel) |
b411b363 | 950 | { |
db830c46 | 951 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 952 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 953 | int err; |
b411b363 PR |
954 | |
955 | if (unlikely(cancel)) { | |
3967deb1 | 956 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 957 | dec_unacked(mdev); |
99920dc5 | 958 | return 0; |
b411b363 PR |
959 | } |
960 | ||
961 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
db830c46 | 962 | drbd_rs_complete_io(mdev, peer_req->i.sector); |
b411b363 PR |
963 | put_ldev(mdev); |
964 | } | |
965 | ||
d612d309 | 966 | if (mdev->state.conn == C_AHEAD) { |
99920dc5 | 967 | err = drbd_send_ack(mdev, P_RS_CANCEL, peer_req); |
db830c46 | 968 | } else if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
969 | if (likely(mdev->state.pdsk >= D_INCONSISTENT)) { |
970 | inc_rs_pending(mdev); | |
99920dc5 | 971 | err = drbd_send_block(mdev, P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
972 | } else { |
973 | if (__ratelimit(&drbd_ratelimit_state)) | |
974 | dev_err(DEV, "Not sending RSDataReply, " | |
975 | "partner DISKLESS!\n"); | |
99920dc5 | 976 | err = 0; |
b411b363 PR |
977 | } |
978 | } else { | |
979 | if (__ratelimit(&drbd_ratelimit_state)) | |
980 | dev_err(DEV, "Sending NegRSDReply. sector %llus.\n", | |
db830c46 | 981 | (unsigned long long)peer_req->i.sector); |
b411b363 | 982 | |
99920dc5 | 983 | err = drbd_send_ack(mdev, P_NEG_RS_DREPLY, peer_req); |
b411b363 PR |
984 | |
985 | /* update resync data with failure */ | |
db830c46 | 986 | drbd_rs_failed_io(mdev, peer_req->i.sector, peer_req->i.size); |
b411b363 PR |
987 | } |
988 | ||
989 | dec_unacked(mdev); | |
990 | ||
db830c46 | 991 | move_to_net_ee_or_free(mdev, peer_req); |
b411b363 | 992 | |
99920dc5 | 993 | if (unlikely(err)) |
b411b363 | 994 | dev_err(DEV, "drbd_send_block() failed\n"); |
99920dc5 | 995 | return err; |
b411b363 PR |
996 | } |
997 | ||
99920dc5 | 998 | int w_e_end_csum_rs_req(struct drbd_work *w, int cancel) |
b411b363 | 999 | { |
db830c46 | 1000 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 1001 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
1002 | struct digest_info *di; |
1003 | int digest_size; | |
1004 | void *digest = NULL; | |
99920dc5 | 1005 | int err, eq = 0; |
b411b363 PR |
1006 | |
1007 | if (unlikely(cancel)) { | |
3967deb1 | 1008 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 1009 | dec_unacked(mdev); |
99920dc5 | 1010 | return 0; |
b411b363 PR |
1011 | } |
1012 | ||
1d53f09e | 1013 | if (get_ldev(mdev)) { |
db830c46 | 1014 | drbd_rs_complete_io(mdev, peer_req->i.sector); |
1d53f09e LE |
1015 | put_ldev(mdev); |
1016 | } | |
b411b363 | 1017 | |
db830c46 | 1018 | di = peer_req->digest; |
b411b363 | 1019 | |
db830c46 | 1020 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1021 | /* quick hack to try to avoid a race against reconfiguration. |
1022 | * a real fix would be much more involved, | |
1023 | * introducing more locking mechanisms */ | |
f399002e LE |
1024 | if (mdev->tconn->csums_tfm) { |
1025 | digest_size = crypto_hash_digestsize(mdev->tconn->csums_tfm); | |
b411b363 PR |
1026 | D_ASSERT(digest_size == di->digest_size); |
1027 | digest = kmalloc(digest_size, GFP_NOIO); | |
1028 | } | |
1029 | if (digest) { | |
f399002e | 1030 | drbd_csum_ee(mdev, mdev->tconn->csums_tfm, peer_req, digest); |
b411b363 PR |
1031 | eq = !memcmp(digest, di->digest, digest_size); |
1032 | kfree(digest); | |
1033 | } | |
1034 | ||
1035 | if (eq) { | |
db830c46 | 1036 | drbd_set_in_sync(mdev, peer_req->i.sector, peer_req->i.size); |
676396d5 | 1037 | /* rs_same_csums unit is BM_BLOCK_SIZE */ |
db830c46 | 1038 | mdev->rs_same_csum += peer_req->i.size >> BM_BLOCK_SHIFT; |
99920dc5 | 1039 | err = drbd_send_ack(mdev, P_RS_IS_IN_SYNC, peer_req); |
b411b363 PR |
1040 | } else { |
1041 | inc_rs_pending(mdev); | |
db830c46 AG |
1042 | peer_req->block_id = ID_SYNCER; /* By setting block_id, digest pointer becomes invalid! */ |
1043 | peer_req->flags &= ~EE_HAS_DIGEST; /* This peer request no longer has a digest pointer */ | |
204bba99 | 1044 | kfree(di); |
99920dc5 | 1045 | err = drbd_send_block(mdev, P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1046 | } |
1047 | } else { | |
99920dc5 | 1048 | err = drbd_send_ack(mdev, P_NEG_RS_DREPLY, peer_req); |
b411b363 PR |
1049 | if (__ratelimit(&drbd_ratelimit_state)) |
1050 | dev_err(DEV, "Sending NegDReply. I guess it gets messy.\n"); | |
1051 | } | |
1052 | ||
1053 | dec_unacked(mdev); | |
db830c46 | 1054 | move_to_net_ee_or_free(mdev, peer_req); |
b411b363 | 1055 | |
99920dc5 | 1056 | if (unlikely(err)) |
b411b363 | 1057 | dev_err(DEV, "drbd_send_block/ack() failed\n"); |
99920dc5 | 1058 | return err; |
b411b363 PR |
1059 | } |
1060 | ||
99920dc5 | 1061 | int w_e_end_ov_req(struct drbd_work *w, int cancel) |
b411b363 | 1062 | { |
db830c46 | 1063 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 1064 | struct drbd_conf *mdev = w->mdev; |
db830c46 AG |
1065 | sector_t sector = peer_req->i.sector; |
1066 | unsigned int size = peer_req->i.size; | |
b411b363 PR |
1067 | int digest_size; |
1068 | void *digest; | |
99920dc5 | 1069 | int err = 0; |
b411b363 PR |
1070 | |
1071 | if (unlikely(cancel)) | |
1072 | goto out; | |
1073 | ||
f399002e | 1074 | digest_size = crypto_hash_digestsize(mdev->tconn->verify_tfm); |
b411b363 | 1075 | digest = kmalloc(digest_size, GFP_NOIO); |
8f21420e | 1076 | if (!digest) { |
99920dc5 | 1077 | err = 1; /* terminate the connection in case the allocation failed */ |
8f21420e | 1078 | goto out; |
b411b363 PR |
1079 | } |
1080 | ||
db830c46 | 1081 | if (likely(!(peer_req->flags & EE_WAS_ERROR))) |
f399002e | 1082 | drbd_csum_ee(mdev, mdev->tconn->verify_tfm, peer_req, digest); |
8f21420e PR |
1083 | else |
1084 | memset(digest, 0, digest_size); | |
1085 | ||
53ea4331 LE |
1086 | /* Free e and pages before send. |
1087 | * In case we block on congestion, we could otherwise run into | |
1088 | * some distributed deadlock, if the other side blocks on | |
1089 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1090 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
3967deb1 | 1091 | drbd_free_peer_req(mdev, peer_req); |
db830c46 | 1092 | peer_req = NULL; |
8f21420e | 1093 | inc_rs_pending(mdev); |
99920dc5 AG |
1094 | err = drbd_send_drequest_csum(mdev, sector, size, digest, digest_size, P_OV_REPLY); |
1095 | if (err) | |
8f21420e PR |
1096 | dec_rs_pending(mdev); |
1097 | kfree(digest); | |
1098 | ||
b411b363 | 1099 | out: |
db830c46 | 1100 | if (peer_req) |
3967deb1 | 1101 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 1102 | dec_unacked(mdev); |
99920dc5 | 1103 | return err; |
b411b363 PR |
1104 | } |
1105 | ||
8f7bed77 | 1106 | void drbd_ov_out_of_sync_found(struct drbd_conf *mdev, sector_t sector, int size) |
b411b363 PR |
1107 | { |
1108 | if (mdev->ov_last_oos_start + mdev->ov_last_oos_size == sector) { | |
1109 | mdev->ov_last_oos_size += size>>9; | |
1110 | } else { | |
1111 | mdev->ov_last_oos_start = sector; | |
1112 | mdev->ov_last_oos_size = size>>9; | |
1113 | } | |
1114 | drbd_set_out_of_sync(mdev, sector, size); | |
b411b363 PR |
1115 | } |
1116 | ||
99920dc5 | 1117 | int w_e_end_ov_reply(struct drbd_work *w, int cancel) |
b411b363 | 1118 | { |
db830c46 | 1119 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 1120 | struct drbd_conf *mdev = w->mdev; |
b411b363 | 1121 | struct digest_info *di; |
b411b363 | 1122 | void *digest; |
db830c46 AG |
1123 | sector_t sector = peer_req->i.sector; |
1124 | unsigned int size = peer_req->i.size; | |
53ea4331 | 1125 | int digest_size; |
99920dc5 | 1126 | int err, eq = 0; |
b411b363 PR |
1127 | |
1128 | if (unlikely(cancel)) { | |
3967deb1 | 1129 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 1130 | dec_unacked(mdev); |
99920dc5 | 1131 | return 0; |
b411b363 PR |
1132 | } |
1133 | ||
1134 | /* after "cancel", because after drbd_disconnect/drbd_rs_cancel_all | |
1135 | * the resync lru has been cleaned up already */ | |
1d53f09e | 1136 | if (get_ldev(mdev)) { |
db830c46 | 1137 | drbd_rs_complete_io(mdev, peer_req->i.sector); |
1d53f09e LE |
1138 | put_ldev(mdev); |
1139 | } | |
b411b363 | 1140 | |
db830c46 | 1141 | di = peer_req->digest; |
b411b363 | 1142 | |
db830c46 | 1143 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
f399002e | 1144 | digest_size = crypto_hash_digestsize(mdev->tconn->verify_tfm); |
b411b363 PR |
1145 | digest = kmalloc(digest_size, GFP_NOIO); |
1146 | if (digest) { | |
f399002e | 1147 | drbd_csum_ee(mdev, mdev->tconn->verify_tfm, peer_req, digest); |
b411b363 PR |
1148 | |
1149 | D_ASSERT(digest_size == di->digest_size); | |
1150 | eq = !memcmp(digest, di->digest, digest_size); | |
1151 | kfree(digest); | |
1152 | } | |
b411b363 PR |
1153 | } |
1154 | ||
9676c760 LE |
1155 | /* Free peer_req and pages before send. |
1156 | * In case we block on congestion, we could otherwise run into | |
1157 | * some distributed deadlock, if the other side blocks on | |
1158 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1159 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
3967deb1 | 1160 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 1161 | if (!eq) |
8f7bed77 | 1162 | drbd_ov_out_of_sync_found(mdev, sector, size); |
b411b363 | 1163 | else |
8f7bed77 | 1164 | ov_out_of_sync_print(mdev); |
b411b363 | 1165 | |
99920dc5 | 1166 | err = drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, |
fa79abd8 | 1167 | eq ? ID_IN_SYNC : ID_OUT_OF_SYNC); |
b411b363 | 1168 | |
53ea4331 | 1169 | dec_unacked(mdev); |
b411b363 | 1170 | |
ea5442af LE |
1171 | --mdev->ov_left; |
1172 | ||
1173 | /* let's advance progress step marks only for every other megabyte */ | |
1174 | if ((mdev->ov_left & 0x200) == 0x200) | |
1175 | drbd_advance_rs_marks(mdev, mdev->ov_left); | |
1176 | ||
1177 | if (mdev->ov_left == 0) { | |
8f7bed77 | 1178 | ov_out_of_sync_print(mdev); |
b411b363 PR |
1179 | drbd_resync_finished(mdev); |
1180 | } | |
1181 | ||
99920dc5 | 1182 | return err; |
b411b363 PR |
1183 | } |
1184 | ||
99920dc5 | 1185 | int w_prev_work_done(struct drbd_work *w, int cancel) |
b411b363 PR |
1186 | { |
1187 | struct drbd_wq_barrier *b = container_of(w, struct drbd_wq_barrier, w); | |
00d56944 | 1188 | |
b411b363 | 1189 | complete(&b->done); |
99920dc5 | 1190 | return 0; |
b411b363 PR |
1191 | } |
1192 | ||
99920dc5 | 1193 | int w_send_barrier(struct drbd_work *w, int cancel) |
b411b363 | 1194 | { |
9f5bdc33 | 1195 | struct drbd_socket *sock; |
b411b363 | 1196 | struct drbd_tl_epoch *b = container_of(w, struct drbd_tl_epoch, w); |
00d56944 | 1197 | struct drbd_conf *mdev = w->mdev; |
9f5bdc33 | 1198 | struct p_barrier *p; |
b411b363 PR |
1199 | |
1200 | /* really avoid racing with tl_clear. w.cb may have been referenced | |
1201 | * just before it was reassigned and re-queued, so double check that. | |
1202 | * actually, this race was harmless, since we only try to send the | |
1203 | * barrier packet here, and otherwise do nothing with the object. | |
1204 | * but compare with the head of w_clear_epoch */ | |
87eeee41 | 1205 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
1206 | if (w->cb != w_send_barrier || mdev->state.conn < C_CONNECTED) |
1207 | cancel = 1; | |
87eeee41 | 1208 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 1209 | if (cancel) |
b411b363 | 1210 | return 0; |
99920dc5 | 1211 | |
9f5bdc33 AG |
1212 | sock = &mdev->tconn->data; |
1213 | p = drbd_prepare_command(mdev, sock); | |
1214 | if (!p) | |
1215 | return -EIO; | |
b411b363 PR |
1216 | p->barrier = b->br_number; |
1217 | /* inc_ap_pending was done where this was queued. | |
1218 | * dec_ap_pending will be done in got_BarrierAck | |
1219 | * or (on connection loss) in w_clear_epoch. */ | |
9f5bdc33 | 1220 | return drbd_send_command(mdev, sock, P_BARRIER, sizeof(*p), NULL, 0); |
b411b363 PR |
1221 | } |
1222 | ||
99920dc5 | 1223 | int w_send_write_hint(struct drbd_work *w, int cancel) |
b411b363 | 1224 | { |
00d56944 | 1225 | struct drbd_conf *mdev = w->mdev; |
9f5bdc33 AG |
1226 | struct drbd_socket *sock; |
1227 | ||
b411b363 | 1228 | if (cancel) |
99920dc5 | 1229 | return 0; |
9f5bdc33 AG |
1230 | sock = &mdev->tconn->data; |
1231 | if (!drbd_prepare_command(mdev, sock)) | |
1232 | return -EIO; | |
e658983a | 1233 | return drbd_send_command(mdev, sock, P_UNPLUG_REMOTE, 0, NULL, 0); |
b411b363 PR |
1234 | } |
1235 | ||
8f7bed77 | 1236 | int w_send_out_of_sync(struct drbd_work *w, int cancel) |
73a01a18 PR |
1237 | { |
1238 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 1239 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 1240 | int err; |
73a01a18 PR |
1241 | |
1242 | if (unlikely(cancel)) { | |
8554df1c | 1243 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1244 | return 0; |
73a01a18 PR |
1245 | } |
1246 | ||
8f7bed77 | 1247 | err = drbd_send_out_of_sync(mdev, req); |
8554df1c | 1248 | req_mod(req, OOS_HANDED_TO_NETWORK); |
73a01a18 | 1249 | |
99920dc5 | 1250 | return err; |
73a01a18 PR |
1251 | } |
1252 | ||
b411b363 PR |
1253 | /** |
1254 | * w_send_dblock() - Worker callback to send a P_DATA packet in order to mirror a write request | |
1255 | * @mdev: DRBD device. | |
1256 | * @w: work object. | |
1257 | * @cancel: The connection will be closed anyways | |
1258 | */ | |
99920dc5 | 1259 | int w_send_dblock(struct drbd_work *w, int cancel) |
b411b363 PR |
1260 | { |
1261 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 1262 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 1263 | int err; |
b411b363 PR |
1264 | |
1265 | if (unlikely(cancel)) { | |
8554df1c | 1266 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1267 | return 0; |
b411b363 PR |
1268 | } |
1269 | ||
99920dc5 AG |
1270 | err = drbd_send_dblock(mdev, req); |
1271 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); | |
b411b363 | 1272 | |
99920dc5 | 1273 | return err; |
b411b363 PR |
1274 | } |
1275 | ||
1276 | /** | |
1277 | * w_send_read_req() - Worker callback to send a read request (P_DATA_REQUEST) packet | |
1278 | * @mdev: DRBD device. | |
1279 | * @w: work object. | |
1280 | * @cancel: The connection will be closed anyways | |
1281 | */ | |
99920dc5 | 1282 | int w_send_read_req(struct drbd_work *w, int cancel) |
b411b363 PR |
1283 | { |
1284 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 1285 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 1286 | int err; |
b411b363 PR |
1287 | |
1288 | if (unlikely(cancel)) { | |
8554df1c | 1289 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1290 | return 0; |
b411b363 PR |
1291 | } |
1292 | ||
99920dc5 | 1293 | err = drbd_send_drequest(mdev, P_DATA_REQUEST, req->i.sector, req->i.size, |
6c1005e7 | 1294 | (unsigned long)req); |
b411b363 | 1295 | |
99920dc5 | 1296 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1297 | |
99920dc5 | 1298 | return err; |
b411b363 PR |
1299 | } |
1300 | ||
99920dc5 | 1301 | int w_restart_disk_io(struct drbd_work *w, int cancel) |
265be2d0 PR |
1302 | { |
1303 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 1304 | struct drbd_conf *mdev = w->mdev; |
265be2d0 | 1305 | |
0778286a | 1306 | if (bio_data_dir(req->master_bio) == WRITE && req->rq_state & RQ_IN_ACT_LOG) |
181286ad | 1307 | drbd_al_begin_io(mdev, &req->i); |
265be2d0 PR |
1308 | /* Calling drbd_al_begin_io() out of the worker might deadlocks |
1309 | theoretically. Practically it can not deadlock, since this is | |
1310 | only used when unfreezing IOs. All the extents of the requests | |
1311 | that made it into the TL are already active */ | |
1312 | ||
1313 | drbd_req_make_private_bio(req, req->master_bio); | |
1314 | req->private_bio->bi_bdev = mdev->ldev->backing_bdev; | |
1315 | generic_make_request(req->private_bio); | |
1316 | ||
99920dc5 | 1317 | return 0; |
265be2d0 PR |
1318 | } |
1319 | ||
b411b363 PR |
1320 | static int _drbd_may_sync_now(struct drbd_conf *mdev) |
1321 | { | |
1322 | struct drbd_conf *odev = mdev; | |
1323 | ||
1324 | while (1) { | |
438c8374 PR |
1325 | if (!odev->ldev) |
1326 | return 1; | |
f399002e | 1327 | if (odev->ldev->dc.resync_after == -1) |
b411b363 | 1328 | return 1; |
f399002e | 1329 | odev = minor_to_mdev(odev->ldev->dc.resync_after); |
841ce241 AG |
1330 | if (!expect(odev)) |
1331 | return 1; | |
b411b363 PR |
1332 | if ((odev->state.conn >= C_SYNC_SOURCE && |
1333 | odev->state.conn <= C_PAUSED_SYNC_T) || | |
1334 | odev->state.aftr_isp || odev->state.peer_isp || | |
1335 | odev->state.user_isp) | |
1336 | return 0; | |
1337 | } | |
1338 | } | |
1339 | ||
1340 | /** | |
1341 | * _drbd_pause_after() - Pause resync on all devices that may not resync now | |
1342 | * @mdev: DRBD device. | |
1343 | * | |
1344 | * Called from process context only (admin command and after_state_ch). | |
1345 | */ | |
1346 | static int _drbd_pause_after(struct drbd_conf *mdev) | |
1347 | { | |
1348 | struct drbd_conf *odev; | |
1349 | int i, rv = 0; | |
1350 | ||
695d08fa | 1351 | rcu_read_lock(); |
81a5d60e | 1352 | idr_for_each_entry(&minors, odev, i) { |
b411b363 PR |
1353 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1354 | continue; | |
1355 | if (!_drbd_may_sync_now(odev)) | |
1356 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 1), CS_HARD, NULL) | |
1357 | != SS_NOTHING_TO_DO); | |
1358 | } | |
695d08fa | 1359 | rcu_read_unlock(); |
b411b363 PR |
1360 | |
1361 | return rv; | |
1362 | } | |
1363 | ||
1364 | /** | |
1365 | * _drbd_resume_next() - Resume resync on all devices that may resync now | |
1366 | * @mdev: DRBD device. | |
1367 | * | |
1368 | * Called from process context only (admin command and worker). | |
1369 | */ | |
1370 | static int _drbd_resume_next(struct drbd_conf *mdev) | |
1371 | { | |
1372 | struct drbd_conf *odev; | |
1373 | int i, rv = 0; | |
1374 | ||
695d08fa | 1375 | rcu_read_lock(); |
81a5d60e | 1376 | idr_for_each_entry(&minors, odev, i) { |
b411b363 PR |
1377 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1378 | continue; | |
1379 | if (odev->state.aftr_isp) { | |
1380 | if (_drbd_may_sync_now(odev)) | |
1381 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 0), | |
1382 | CS_HARD, NULL) | |
1383 | != SS_NOTHING_TO_DO) ; | |
1384 | } | |
1385 | } | |
695d08fa | 1386 | rcu_read_unlock(); |
b411b363 PR |
1387 | return rv; |
1388 | } | |
1389 | ||
1390 | void resume_next_sg(struct drbd_conf *mdev) | |
1391 | { | |
1392 | write_lock_irq(&global_state_lock); | |
1393 | _drbd_resume_next(mdev); | |
1394 | write_unlock_irq(&global_state_lock); | |
1395 | } | |
1396 | ||
1397 | void suspend_other_sg(struct drbd_conf *mdev) | |
1398 | { | |
1399 | write_lock_irq(&global_state_lock); | |
1400 | _drbd_pause_after(mdev); | |
1401 | write_unlock_irq(&global_state_lock); | |
1402 | } | |
1403 | ||
dc97b708 PR |
1404 | /* caller must hold global_state_lock */ |
1405 | enum drbd_ret_code drbd_sync_after_valid(struct drbd_conf *mdev, int o_minor) | |
b411b363 PR |
1406 | { |
1407 | struct drbd_conf *odev; | |
1408 | ||
1409 | if (o_minor == -1) | |
1410 | return NO_ERROR; | |
1411 | if (o_minor < -1 || minor_to_mdev(o_minor) == NULL) | |
1412 | return ERR_SYNC_AFTER; | |
1413 | ||
1414 | /* check for loops */ | |
1415 | odev = minor_to_mdev(o_minor); | |
1416 | while (1) { | |
1417 | if (odev == mdev) | |
1418 | return ERR_SYNC_AFTER_CYCLE; | |
1419 | ||
1420 | /* dependency chain ends here, no cycles. */ | |
f399002e | 1421 | if (odev->ldev->dc.resync_after == -1) |
b411b363 PR |
1422 | return NO_ERROR; |
1423 | ||
1424 | /* follow the dependency chain */ | |
f399002e | 1425 | odev = minor_to_mdev(odev->ldev->dc.resync_after); |
b411b363 PR |
1426 | } |
1427 | } | |
1428 | ||
dc97b708 PR |
1429 | /* caller must hold global_state_lock */ |
1430 | void drbd_sync_after_changed(struct drbd_conf *mdev) | |
b411b363 PR |
1431 | { |
1432 | int changes; | |
b411b363 | 1433 | |
dc97b708 PR |
1434 | do { |
1435 | changes = _drbd_pause_after(mdev); | |
1436 | changes |= _drbd_resume_next(mdev); | |
1437 | } while (changes); | |
b411b363 PR |
1438 | } |
1439 | ||
9bd28d3c LE |
1440 | void drbd_rs_controller_reset(struct drbd_conf *mdev) |
1441 | { | |
1442 | atomic_set(&mdev->rs_sect_in, 0); | |
1443 | atomic_set(&mdev->rs_sect_ev, 0); | |
1444 | mdev->rs_in_flight = 0; | |
1445 | mdev->rs_planed = 0; | |
1446 | spin_lock(&mdev->peer_seq_lock); | |
1447 | fifo_set(&mdev->rs_plan_s, 0); | |
1448 | spin_unlock(&mdev->peer_seq_lock); | |
1449 | } | |
1450 | ||
1f04af33 PR |
1451 | void start_resync_timer_fn(unsigned long data) |
1452 | { | |
1453 | struct drbd_conf *mdev = (struct drbd_conf *) data; | |
1454 | ||
1455 | drbd_queue_work(&mdev->tconn->data.work, &mdev->start_resync_work); | |
1456 | } | |
1457 | ||
99920dc5 | 1458 | int w_start_resync(struct drbd_work *w, int cancel) |
1f04af33 | 1459 | { |
00d56944 PR |
1460 | struct drbd_conf *mdev = w->mdev; |
1461 | ||
1f04af33 PR |
1462 | if (atomic_read(&mdev->unacked_cnt) || atomic_read(&mdev->rs_pending_cnt)) { |
1463 | dev_warn(DEV, "w_start_resync later...\n"); | |
1464 | mdev->start_resync_timer.expires = jiffies + HZ/10; | |
1465 | add_timer(&mdev->start_resync_timer); | |
99920dc5 | 1466 | return 0; |
1f04af33 PR |
1467 | } |
1468 | ||
1469 | drbd_start_resync(mdev, C_SYNC_SOURCE); | |
1470 | clear_bit(AHEAD_TO_SYNC_SOURCE, &mdev->current_epoch->flags); | |
99920dc5 | 1471 | return 0; |
1f04af33 PR |
1472 | } |
1473 | ||
b411b363 PR |
1474 | /** |
1475 | * drbd_start_resync() - Start the resync process | |
1476 | * @mdev: DRBD device. | |
1477 | * @side: Either C_SYNC_SOURCE or C_SYNC_TARGET | |
1478 | * | |
1479 | * This function might bring you directly into one of the | |
1480 | * C_PAUSED_SYNC_* states. | |
1481 | */ | |
1482 | void drbd_start_resync(struct drbd_conf *mdev, enum drbd_conns side) | |
1483 | { | |
1484 | union drbd_state ns; | |
1485 | int r; | |
1486 | ||
c4752ef1 | 1487 | if (mdev->state.conn >= C_SYNC_SOURCE && mdev->state.conn < C_AHEAD) { |
b411b363 PR |
1488 | dev_err(DEV, "Resync already running!\n"); |
1489 | return; | |
1490 | } | |
1491 | ||
59817f4f PR |
1492 | if (mdev->state.conn < C_AHEAD) { |
1493 | /* In case a previous resync run was aborted by an IO error/detach on the peer. */ | |
1494 | drbd_rs_cancel_all(mdev); | |
1495 | /* This should be done when we abort the resync. We definitely do not | |
1496 | want to have this for connections going back and forth between | |
1497 | Ahead/Behind and SyncSource/SyncTarget */ | |
1498 | } | |
b411b363 | 1499 | |
e64a3294 PR |
1500 | if (!test_bit(B_RS_H_DONE, &mdev->flags)) { |
1501 | if (side == C_SYNC_TARGET) { | |
1502 | /* Since application IO was locked out during C_WF_BITMAP_T and | |
1503 | C_WF_SYNC_UUID we are still unmodified. Before going to C_SYNC_TARGET | |
1504 | we check that we might make the data inconsistent. */ | |
1505 | r = drbd_khelper(mdev, "before-resync-target"); | |
1506 | r = (r >> 8) & 0xff; | |
1507 | if (r > 0) { | |
1508 | dev_info(DEV, "before-resync-target handler returned %d, " | |
09b9e797 | 1509 | "dropping connection.\n", r); |
38fa9988 | 1510 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
09b9e797 PR |
1511 | return; |
1512 | } | |
e64a3294 PR |
1513 | } else /* C_SYNC_SOURCE */ { |
1514 | r = drbd_khelper(mdev, "before-resync-source"); | |
1515 | r = (r >> 8) & 0xff; | |
1516 | if (r > 0) { | |
1517 | if (r == 3) { | |
1518 | dev_info(DEV, "before-resync-source handler returned %d, " | |
1519 | "ignoring. Old userland tools?", r); | |
1520 | } else { | |
1521 | dev_info(DEV, "before-resync-source handler returned %d, " | |
1522 | "dropping connection.\n", r); | |
38fa9988 | 1523 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
e64a3294 PR |
1524 | return; |
1525 | } | |
1526 | } | |
09b9e797 | 1527 | } |
b411b363 PR |
1528 | } |
1529 | ||
e64a3294 | 1530 | if (current == mdev->tconn->worker.task) { |
dad20554 | 1531 | /* The worker should not sleep waiting for state_mutex, |
e64a3294 | 1532 | that can take long */ |
8410da8f | 1533 | if (!mutex_trylock(mdev->state_mutex)) { |
e64a3294 PR |
1534 | set_bit(B_RS_H_DONE, &mdev->flags); |
1535 | mdev->start_resync_timer.expires = jiffies + HZ/5; | |
1536 | add_timer(&mdev->start_resync_timer); | |
1537 | return; | |
1538 | } | |
1539 | } else { | |
8410da8f | 1540 | mutex_lock(mdev->state_mutex); |
e64a3294 PR |
1541 | } |
1542 | clear_bit(B_RS_H_DONE, &mdev->flags); | |
b411b363 PR |
1543 | |
1544 | if (!get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
8410da8f | 1545 | mutex_unlock(mdev->state_mutex); |
b411b363 PR |
1546 | return; |
1547 | } | |
1548 | ||
b411b363 | 1549 | write_lock_irq(&global_state_lock); |
78bae59b | 1550 | ns = drbd_read_state(mdev); |
b411b363 PR |
1551 | |
1552 | ns.aftr_isp = !_drbd_may_sync_now(mdev); | |
1553 | ||
1554 | ns.conn = side; | |
1555 | ||
1556 | if (side == C_SYNC_TARGET) | |
1557 | ns.disk = D_INCONSISTENT; | |
1558 | else /* side == C_SYNC_SOURCE */ | |
1559 | ns.pdsk = D_INCONSISTENT; | |
1560 | ||
1561 | r = __drbd_set_state(mdev, ns, CS_VERBOSE, NULL); | |
78bae59b | 1562 | ns = drbd_read_state(mdev); |
b411b363 PR |
1563 | |
1564 | if (ns.conn < C_CONNECTED) | |
1565 | r = SS_UNKNOWN_ERROR; | |
1566 | ||
1567 | if (r == SS_SUCCESS) { | |
1d7734a0 LE |
1568 | unsigned long tw = drbd_bm_total_weight(mdev); |
1569 | unsigned long now = jiffies; | |
1570 | int i; | |
1571 | ||
b411b363 PR |
1572 | mdev->rs_failed = 0; |
1573 | mdev->rs_paused = 0; | |
b411b363 | 1574 | mdev->rs_same_csum = 0; |
0f0601f4 LE |
1575 | mdev->rs_last_events = 0; |
1576 | mdev->rs_last_sect_ev = 0; | |
1d7734a0 LE |
1577 | mdev->rs_total = tw; |
1578 | mdev->rs_start = now; | |
1579 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { | |
1580 | mdev->rs_mark_left[i] = tw; | |
1581 | mdev->rs_mark_time[i] = now; | |
1582 | } | |
b411b363 PR |
1583 | _drbd_pause_after(mdev); |
1584 | } | |
1585 | write_unlock_irq(&global_state_lock); | |
5a22db89 | 1586 | |
b411b363 PR |
1587 | if (r == SS_SUCCESS) { |
1588 | dev_info(DEV, "Began resync as %s (will sync %lu KB [%lu bits set]).\n", | |
1589 | drbd_conn_str(ns.conn), | |
1590 | (unsigned long) mdev->rs_total << (BM_BLOCK_SHIFT-10), | |
1591 | (unsigned long) mdev->rs_total); | |
6c922ed5 LE |
1592 | if (side == C_SYNC_TARGET) |
1593 | mdev->bm_resync_fo = 0; | |
1594 | ||
1595 | /* Since protocol 96, we must serialize drbd_gen_and_send_sync_uuid | |
1596 | * with w_send_oos, or the sync target will get confused as to | |
1597 | * how much bits to resync. We cannot do that always, because for an | |
1598 | * empty resync and protocol < 95, we need to do it here, as we call | |
1599 | * drbd_resync_finished from here in that case. | |
1600 | * We drbd_gen_and_send_sync_uuid here for protocol < 96, | |
1601 | * and from after_state_ch otherwise. */ | |
31890f4a | 1602 | if (side == C_SYNC_SOURCE && mdev->tconn->agreed_pro_version < 96) |
6c922ed5 | 1603 | drbd_gen_and_send_sync_uuid(mdev); |
b411b363 | 1604 | |
31890f4a | 1605 | if (mdev->tconn->agreed_pro_version < 95 && mdev->rs_total == 0) { |
af85e8e8 LE |
1606 | /* This still has a race (about when exactly the peers |
1607 | * detect connection loss) that can lead to a full sync | |
1608 | * on next handshake. In 8.3.9 we fixed this with explicit | |
1609 | * resync-finished notifications, but the fix | |
1610 | * introduces a protocol change. Sleeping for some | |
1611 | * time longer than the ping interval + timeout on the | |
1612 | * SyncSource, to give the SyncTarget the chance to | |
1613 | * detect connection loss, then waiting for a ping | |
1614 | * response (implicit in drbd_resync_finished) reduces | |
1615 | * the race considerably, but does not solve it. */ | |
44ed167d PR |
1616 | if (side == C_SYNC_SOURCE) { |
1617 | struct net_conf *nc; | |
1618 | int timeo; | |
1619 | ||
1620 | rcu_read_lock(); | |
1621 | nc = rcu_dereference(mdev->tconn->net_conf); | |
1622 | timeo = nc->ping_int * HZ + nc->ping_timeo * HZ / 9; | |
1623 | rcu_read_unlock(); | |
1624 | schedule_timeout_interruptible(timeo); | |
1625 | } | |
b411b363 | 1626 | drbd_resync_finished(mdev); |
b411b363 PR |
1627 | } |
1628 | ||
9bd28d3c | 1629 | drbd_rs_controller_reset(mdev); |
b411b363 PR |
1630 | /* ns.conn may already be != mdev->state.conn, |
1631 | * we may have been paused in between, or become paused until | |
1632 | * the timer triggers. | |
1633 | * No matter, that is handled in resync_timer_fn() */ | |
1634 | if (ns.conn == C_SYNC_TARGET) | |
1635 | mod_timer(&mdev->resync_timer, jiffies); | |
1636 | ||
1637 | drbd_md_sync(mdev); | |
1638 | } | |
5a22db89 | 1639 | put_ldev(mdev); |
8410da8f | 1640 | mutex_unlock(mdev->state_mutex); |
b411b363 PR |
1641 | } |
1642 | ||
1643 | int drbd_worker(struct drbd_thread *thi) | |
1644 | { | |
392c8801 | 1645 | struct drbd_tconn *tconn = thi->tconn; |
b411b363 | 1646 | struct drbd_work *w = NULL; |
0e29d163 | 1647 | struct drbd_conf *mdev; |
44ed167d | 1648 | struct net_conf *nc; |
b411b363 | 1649 | LIST_HEAD(work_list); |
f399002e | 1650 | int vnr, intr = 0; |
44ed167d | 1651 | int cork; |
b411b363 | 1652 | |
e77a0a5c | 1653 | while (get_t_state(thi) == RUNNING) { |
80822284 | 1654 | drbd_thread_current_set_cpu(thi); |
b411b363 | 1655 | |
19393e10 PR |
1656 | if (down_trylock(&tconn->data.work.s)) { |
1657 | mutex_lock(&tconn->data.mutex); | |
44ed167d PR |
1658 | |
1659 | rcu_read_lock(); | |
1660 | nc = rcu_dereference(tconn->net_conf); | |
1661 | cork = nc ? !nc->no_cork : 0; | |
1662 | rcu_read_unlock(); | |
1663 | ||
1664 | if (tconn->data.socket && cork) | |
19393e10 PR |
1665 | drbd_tcp_uncork(tconn->data.socket); |
1666 | mutex_unlock(&tconn->data.mutex); | |
b411b363 | 1667 | |
19393e10 | 1668 | intr = down_interruptible(&tconn->data.work.s); |
b411b363 | 1669 | |
19393e10 | 1670 | mutex_lock(&tconn->data.mutex); |
44ed167d | 1671 | if (tconn->data.socket && cork) |
19393e10 PR |
1672 | drbd_tcp_cork(tconn->data.socket); |
1673 | mutex_unlock(&tconn->data.mutex); | |
b411b363 PR |
1674 | } |
1675 | ||
1676 | if (intr) { | |
b411b363 | 1677 | flush_signals(current); |
19393e10 PR |
1678 | if (get_t_state(thi) == RUNNING) { |
1679 | conn_warn(tconn, "Worker got an unexpected signal\n"); | |
b411b363 | 1680 | continue; |
19393e10 | 1681 | } |
b411b363 PR |
1682 | break; |
1683 | } | |
1684 | ||
e77a0a5c | 1685 | if (get_t_state(thi) != RUNNING) |
b411b363 PR |
1686 | break; |
1687 | /* With this break, we have done a down() but not consumed | |
1688 | the entry from the list. The cleanup code takes care of | |
1689 | this... */ | |
1690 | ||
1691 | w = NULL; | |
19393e10 PR |
1692 | spin_lock_irq(&tconn->data.work.q_lock); |
1693 | if (list_empty(&tconn->data.work.q)) { | |
b411b363 PR |
1694 | /* something terribly wrong in our logic. |
1695 | * we were able to down() the semaphore, | |
1696 | * but the list is empty... doh. | |
1697 | * | |
1698 | * what is the best thing to do now? | |
1699 | * try again from scratch, restarting the receiver, | |
1700 | * asender, whatnot? could break even more ugly, | |
1701 | * e.g. when we are primary, but no good local data. | |
1702 | * | |
1703 | * I'll try to get away just starting over this loop. | |
1704 | */ | |
19393e10 PR |
1705 | conn_warn(tconn, "Work list unexpectedly empty\n"); |
1706 | spin_unlock_irq(&tconn->data.work.q_lock); | |
b411b363 PR |
1707 | continue; |
1708 | } | |
19393e10 | 1709 | w = list_entry(tconn->data.work.q.next, struct drbd_work, list); |
b411b363 | 1710 | list_del_init(&w->list); |
19393e10 | 1711 | spin_unlock_irq(&tconn->data.work.q_lock); |
b411b363 | 1712 | |
99920dc5 | 1713 | if (w->cb(w, tconn->cstate < C_WF_REPORT_PARAMS)) { |
b411b363 | 1714 | /* dev_warn(DEV, "worker: a callback failed! \n"); */ |
bbeb641c PR |
1715 | if (tconn->cstate >= C_WF_REPORT_PARAMS) |
1716 | conn_request_state(tconn, NS(conn, C_NETWORK_FAILURE), CS_HARD); | |
b411b363 PR |
1717 | } |
1718 | } | |
b411b363 | 1719 | |
19393e10 PR |
1720 | spin_lock_irq(&tconn->data.work.q_lock); |
1721 | while (!list_empty(&tconn->data.work.q)) { | |
1722 | list_splice_init(&tconn->data.work.q, &work_list); | |
1723 | spin_unlock_irq(&tconn->data.work.q_lock); | |
b411b363 PR |
1724 | |
1725 | while (!list_empty(&work_list)) { | |
1726 | w = list_entry(work_list.next, struct drbd_work, list); | |
1727 | list_del_init(&w->list); | |
00d56944 | 1728 | w->cb(w, 1); |
b411b363 PR |
1729 | } |
1730 | ||
19393e10 | 1731 | spin_lock_irq(&tconn->data.work.q_lock); |
b411b363 | 1732 | } |
19393e10 | 1733 | sema_init(&tconn->data.work.s, 0); |
b411b363 PR |
1734 | /* DANGEROUS race: if someone did queue his work within the spinlock, |
1735 | * but up() ed outside the spinlock, we could get an up() on the | |
1736 | * semaphore without corresponding list entry. | |
1737 | * So don't do that. | |
1738 | */ | |
19393e10 | 1739 | spin_unlock_irq(&tconn->data.work.q_lock); |
b411b363 | 1740 | |
d3fcb490 | 1741 | down_read(&drbd_cfg_rwsem); |
f399002e | 1742 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
0e29d163 | 1743 | D_ASSERT(mdev->state.disk == D_DISKLESS && mdev->state.conn == C_STANDALONE); |
0e29d163 PR |
1744 | drbd_mdev_cleanup(mdev); |
1745 | } | |
d3fcb490 | 1746 | up_read(&drbd_cfg_rwsem); |
b411b363 PR |
1747 | |
1748 | return 0; | |
1749 | } |