]>
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 | |
95f8efd0 | 60 | to evaluate the resync after dependencies, we grab a write lock, because |
b411b363 PR |
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; | |
cdfda633 | 70 | struct drbd_conf *mdev; |
b411b363 PR |
71 | |
72 | md_io = (struct drbd_md_io *)bio->bi_private; | |
cdfda633 PR |
73 | mdev = container_of(md_io, struct drbd_conf, md_io); |
74 | ||
b411b363 PR |
75 | md_io->error = error; |
76 | ||
cdfda633 PR |
77 | md_io->done = 1; |
78 | wake_up(&mdev->misc_wait); | |
79 | bio_put(bio); | |
80 | drbd_md_put_buffer(mdev); | |
81 | put_ldev(mdev); | |
b411b363 PR |
82 | } |
83 | ||
84 | /* reads on behalf of the partner, | |
85 | * "submitted" by the receiver | |
86 | */ | |
db830c46 | 87 | void drbd_endio_read_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
88 | { |
89 | unsigned long flags = 0; | |
a21e9298 | 90 | struct drbd_conf *mdev = peer_req->w.mdev; |
b411b363 | 91 | |
87eeee41 | 92 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
db830c46 AG |
93 | mdev->read_cnt += peer_req->i.size >> 9; |
94 | list_del(&peer_req->w.list); | |
b411b363 PR |
95 | if (list_empty(&mdev->read_ee)) |
96 | wake_up(&mdev->ee_wait); | |
db830c46 | 97 | if (test_bit(__EE_WAS_ERROR, &peer_req->flags)) |
81e84650 | 98 | __drbd_chk_io_error(mdev, false); |
87eeee41 | 99 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
b411b363 | 100 | |
db830c46 | 101 | drbd_queue_work(&mdev->tconn->data.work, &peer_req->w); |
b411b363 | 102 | put_ldev(mdev); |
b411b363 PR |
103 | } |
104 | ||
105 | /* writes on behalf of the partner, or resync writes, | |
45bb912b | 106 | * "submitted" by the receiver, final stage. */ |
db830c46 | 107 | static void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
108 | { |
109 | unsigned long flags = 0; | |
a21e9298 | 110 | struct drbd_conf *mdev = peer_req->w.mdev; |
181286ad | 111 | struct drbd_interval i; |
b411b363 | 112 | int do_wake; |
579b57ed | 113 | u64 block_id; |
b411b363 | 114 | int do_al_complete_io; |
b411b363 | 115 | |
db830c46 | 116 | /* after we moved peer_req to done_ee, |
b411b363 PR |
117 | * we may no longer access it, |
118 | * it may be freed/reused already! | |
119 | * (as soon as we release the req_lock) */ | |
181286ad | 120 | i = peer_req->i; |
db830c46 AG |
121 | do_al_complete_io = peer_req->flags & EE_CALL_AL_COMPLETE_IO; |
122 | block_id = peer_req->block_id; | |
b411b363 | 123 | |
87eeee41 | 124 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
db830c46 AG |
125 | mdev->writ_cnt += peer_req->i.size >> 9; |
126 | list_del(&peer_req->w.list); /* has been on active_ee or sync_ee */ | |
127 | list_add_tail(&peer_req->w.list, &mdev->done_ee); | |
b411b363 | 128 | |
bb3bfe96 | 129 | /* |
5e472264 | 130 | * Do not remove from the write_requests tree here: we did not send the |
bb3bfe96 AG |
131 | * Ack yet and did not wake possibly waiting conflicting requests. |
132 | * Removed from the tree from "drbd_process_done_ee" within the | |
133 | * appropriate w.cb (e_end_block/e_end_resync_block) or from | |
134 | * _drbd_clear_done_ee. | |
135 | */ | |
b411b363 | 136 | |
579b57ed | 137 | do_wake = list_empty(block_id == ID_SYNCER ? &mdev->sync_ee : &mdev->active_ee); |
b411b363 | 138 | |
db830c46 | 139 | if (test_bit(__EE_WAS_ERROR, &peer_req->flags)) |
81e84650 | 140 | __drbd_chk_io_error(mdev, false); |
87eeee41 | 141 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
b411b363 | 142 | |
579b57ed | 143 | if (block_id == ID_SYNCER) |
181286ad | 144 | drbd_rs_complete_io(mdev, i.sector); |
b411b363 PR |
145 | |
146 | if (do_wake) | |
147 | wake_up(&mdev->ee_wait); | |
148 | ||
149 | if (do_al_complete_io) | |
181286ad | 150 | drbd_al_complete_io(mdev, &i); |
b411b363 | 151 | |
0625ac19 | 152 | wake_asender(mdev->tconn); |
b411b363 | 153 | put_ldev(mdev); |
45bb912b | 154 | } |
b411b363 | 155 | |
45bb912b LE |
156 | /* writes on behalf of the partner, or resync writes, |
157 | * "submitted" by the receiver. | |
158 | */ | |
fcefa62e | 159 | void drbd_peer_request_endio(struct bio *bio, int error) |
45bb912b | 160 | { |
db830c46 | 161 | struct drbd_peer_request *peer_req = bio->bi_private; |
a21e9298 | 162 | struct drbd_conf *mdev = peer_req->w.mdev; |
45bb912b LE |
163 | int uptodate = bio_flagged(bio, BIO_UPTODATE); |
164 | int is_write = bio_data_dir(bio) == WRITE; | |
165 | ||
07194272 | 166 | if (error && __ratelimit(&drbd_ratelimit_state)) |
45bb912b LE |
167 | dev_warn(DEV, "%s: error=%d s=%llus\n", |
168 | is_write ? "write" : "read", error, | |
db830c46 | 169 | (unsigned long long)peer_req->i.sector); |
45bb912b | 170 | if (!error && !uptodate) { |
07194272 LE |
171 | if (__ratelimit(&drbd_ratelimit_state)) |
172 | dev_warn(DEV, "%s: setting error to -EIO s=%llus\n", | |
173 | is_write ? "write" : "read", | |
db830c46 | 174 | (unsigned long long)peer_req->i.sector); |
45bb912b LE |
175 | /* strange behavior of some lower level drivers... |
176 | * fail the request by clearing the uptodate flag, | |
177 | * but do not return any error?! */ | |
178 | error = -EIO; | |
179 | } | |
180 | ||
181 | if (error) | |
db830c46 | 182 | set_bit(__EE_WAS_ERROR, &peer_req->flags); |
45bb912b LE |
183 | |
184 | bio_put(bio); /* no need for the bio anymore */ | |
db830c46 | 185 | if (atomic_dec_and_test(&peer_req->pending_bios)) { |
45bb912b | 186 | if (is_write) |
db830c46 | 187 | drbd_endio_write_sec_final(peer_req); |
45bb912b | 188 | else |
db830c46 | 189 | drbd_endio_read_sec_final(peer_req); |
45bb912b | 190 | } |
b411b363 PR |
191 | } |
192 | ||
193 | /* read, readA or write requests on R_PRIMARY coming from drbd_make_request | |
194 | */ | |
fcefa62e | 195 | void drbd_request_endio(struct bio *bio, int error) |
b411b363 | 196 | { |
a115413d | 197 | unsigned long flags; |
b411b363 | 198 | struct drbd_request *req = bio->bi_private; |
a21e9298 | 199 | struct drbd_conf *mdev = req->w.mdev; |
a115413d | 200 | struct bio_and_error m; |
b411b363 PR |
201 | enum drbd_req_event what; |
202 | int uptodate = bio_flagged(bio, BIO_UPTODATE); | |
203 | ||
b411b363 PR |
204 | if (!error && !uptodate) { |
205 | dev_warn(DEV, "p %s: setting error to -EIO\n", | |
206 | bio_data_dir(bio) == WRITE ? "write" : "read"); | |
207 | /* strange behavior of some lower level drivers... | |
208 | * fail the request by clearing the uptodate flag, | |
209 | * but do not return any error?! */ | |
210 | error = -EIO; | |
211 | } | |
212 | ||
b411b363 PR |
213 | /* to avoid recursion in __req_mod */ |
214 | if (unlikely(error)) { | |
215 | what = (bio_data_dir(bio) == WRITE) | |
8554df1c | 216 | ? WRITE_COMPLETED_WITH_ERROR |
5c3c7e64 | 217 | : (bio_rw(bio) == READ) |
8554df1c AG |
218 | ? READ_COMPLETED_WITH_ERROR |
219 | : READ_AHEAD_COMPLETED_WITH_ERROR; | |
b411b363 | 220 | } else |
8554df1c | 221 | what = COMPLETED_OK; |
b411b363 PR |
222 | |
223 | bio_put(req->private_bio); | |
224 | req->private_bio = ERR_PTR(error); | |
225 | ||
a115413d | 226 | /* not req_mod(), we need irqsave here! */ |
87eeee41 | 227 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
a115413d | 228 | __req_mod(req, what, &m); |
87eeee41 | 229 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
a115413d LE |
230 | |
231 | if (m.bio) | |
232 | complete_master_bio(mdev, &m); | |
b411b363 PR |
233 | } |
234 | ||
99920dc5 | 235 | int w_read_retry_remote(struct drbd_work *w, int cancel) |
b411b363 PR |
236 | { |
237 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 238 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
239 | |
240 | /* We should not detach for read io-error, | |
241 | * but try to WRITE the P_DATA_REPLY to the failed location, | |
242 | * to give the disk the chance to relocate that block */ | |
243 | ||
87eeee41 | 244 | spin_lock_irq(&mdev->tconn->req_lock); |
d255e5ff | 245 | if (cancel || mdev->state.pdsk != D_UP_TO_DATE) { |
8554df1c | 246 | _req_mod(req, READ_RETRY_REMOTE_CANCELED); |
87eeee41 | 247 | spin_unlock_irq(&mdev->tconn->req_lock); |
99920dc5 | 248 | return 0; |
b411b363 | 249 | } |
87eeee41 | 250 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 251 | |
00d56944 | 252 | return w_send_read_req(w, 0); |
b411b363 PR |
253 | } |
254 | ||
f6ffca9f | 255 | void drbd_csum_ee(struct drbd_conf *mdev, struct crypto_hash *tfm, |
db830c46 | 256 | struct drbd_peer_request *peer_req, void *digest) |
45bb912b LE |
257 | { |
258 | struct hash_desc desc; | |
259 | struct scatterlist sg; | |
db830c46 | 260 | struct page *page = peer_req->pages; |
45bb912b LE |
261 | struct page *tmp; |
262 | unsigned len; | |
263 | ||
264 | desc.tfm = tfm; | |
265 | desc.flags = 0; | |
266 | ||
267 | sg_init_table(&sg, 1); | |
268 | crypto_hash_init(&desc); | |
269 | ||
270 | while ((tmp = page_chain_next(page))) { | |
271 | /* all but the last page will be fully used */ | |
272 | sg_set_page(&sg, page, PAGE_SIZE, 0); | |
273 | crypto_hash_update(&desc, &sg, sg.length); | |
274 | page = tmp; | |
275 | } | |
276 | /* and now the last, possibly only partially used page */ | |
db830c46 | 277 | len = peer_req->i.size & (PAGE_SIZE - 1); |
45bb912b LE |
278 | sg_set_page(&sg, page, len ?: PAGE_SIZE, 0); |
279 | crypto_hash_update(&desc, &sg, sg.length); | |
280 | crypto_hash_final(&desc, digest); | |
281 | } | |
282 | ||
283 | void drbd_csum_bio(struct drbd_conf *mdev, struct crypto_hash *tfm, struct bio *bio, void *digest) | |
b411b363 PR |
284 | { |
285 | struct hash_desc desc; | |
286 | struct scatterlist sg; | |
287 | struct bio_vec *bvec; | |
288 | int i; | |
289 | ||
290 | desc.tfm = tfm; | |
291 | desc.flags = 0; | |
292 | ||
293 | sg_init_table(&sg, 1); | |
294 | crypto_hash_init(&desc); | |
295 | ||
296 | __bio_for_each_segment(bvec, bio, i, 0) { | |
297 | sg_set_page(&sg, bvec->bv_page, bvec->bv_len, bvec->bv_offset); | |
298 | crypto_hash_update(&desc, &sg, sg.length); | |
299 | } | |
300 | crypto_hash_final(&desc, digest); | |
301 | } | |
302 | ||
9676c760 | 303 | /* MAYBE merge common code with w_e_end_ov_req */ |
99920dc5 | 304 | static int w_e_send_csum(struct drbd_work *w, int cancel) |
b411b363 | 305 | { |
00d56944 PR |
306 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
307 | struct drbd_conf *mdev = w->mdev; | |
b411b363 PR |
308 | int digest_size; |
309 | void *digest; | |
99920dc5 | 310 | int err = 0; |
b411b363 | 311 | |
53ea4331 LE |
312 | if (unlikely(cancel)) |
313 | goto out; | |
b411b363 | 314 | |
9676c760 | 315 | if (unlikely((peer_req->flags & EE_WAS_ERROR) != 0)) |
53ea4331 | 316 | goto out; |
b411b363 | 317 | |
f399002e | 318 | digest_size = crypto_hash_digestsize(mdev->tconn->csums_tfm); |
53ea4331 LE |
319 | digest = kmalloc(digest_size, GFP_NOIO); |
320 | if (digest) { | |
db830c46 AG |
321 | sector_t sector = peer_req->i.sector; |
322 | unsigned int size = peer_req->i.size; | |
f399002e | 323 | drbd_csum_ee(mdev, mdev->tconn->csums_tfm, peer_req, digest); |
9676c760 | 324 | /* Free peer_req and pages before send. |
53ea4331 LE |
325 | * In case we block on congestion, we could otherwise run into |
326 | * some distributed deadlock, if the other side blocks on | |
327 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 328 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
3967deb1 | 329 | drbd_free_peer_req(mdev, peer_req); |
db830c46 | 330 | peer_req = NULL; |
53ea4331 | 331 | inc_rs_pending(mdev); |
99920dc5 | 332 | err = drbd_send_drequest_csum(mdev, sector, size, |
db1b0b72 AG |
333 | digest, digest_size, |
334 | P_CSUM_RS_REQUEST); | |
53ea4331 LE |
335 | kfree(digest); |
336 | } else { | |
337 | dev_err(DEV, "kmalloc() of digest failed.\n"); | |
99920dc5 | 338 | err = -ENOMEM; |
53ea4331 | 339 | } |
b411b363 | 340 | |
53ea4331 | 341 | out: |
db830c46 | 342 | if (peer_req) |
3967deb1 | 343 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 344 | |
99920dc5 | 345 | if (unlikely(err)) |
b411b363 | 346 | dev_err(DEV, "drbd_send_drequest(..., csum) failed\n"); |
99920dc5 | 347 | return err; |
b411b363 PR |
348 | } |
349 | ||
350 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
351 | ||
352 | static int read_for_csum(struct drbd_conf *mdev, sector_t sector, int size) | |
353 | { | |
db830c46 | 354 | struct drbd_peer_request *peer_req; |
b411b363 PR |
355 | |
356 | if (!get_ldev(mdev)) | |
80a40e43 | 357 | return -EIO; |
b411b363 | 358 | |
e3555d85 | 359 | if (drbd_rs_should_slow_down(mdev, sector)) |
0f0601f4 LE |
360 | goto defer; |
361 | ||
b411b363 PR |
362 | /* GFP_TRY, because if there is no memory available right now, this may |
363 | * be rescheduled for later. It is "only" background resync, after all. */ | |
0db55363 AG |
364 | peer_req = drbd_alloc_peer_req(mdev, ID_SYNCER /* unused */, sector, |
365 | size, GFP_TRY); | |
db830c46 | 366 | if (!peer_req) |
80a40e43 | 367 | goto defer; |
b411b363 | 368 | |
db830c46 | 369 | peer_req->w.cb = w_e_send_csum; |
87eeee41 | 370 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 371 | list_add(&peer_req->w.list, &mdev->read_ee); |
87eeee41 | 372 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 373 | |
0f0601f4 | 374 | atomic_add(size >> 9, &mdev->rs_sect_ev); |
fbe29dec | 375 | if (drbd_submit_peer_request(mdev, peer_req, READ, DRBD_FAULT_RS_RD) == 0) |
80a40e43 | 376 | return 0; |
b411b363 | 377 | |
10f6d992 LE |
378 | /* If it failed because of ENOMEM, retry should help. If it failed |
379 | * because bio_add_page failed (probably broken lower level driver), | |
380 | * retry may or may not help. | |
381 | * If it does not, you may need to force disconnect. */ | |
87eeee41 | 382 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 383 | list_del(&peer_req->w.list); |
87eeee41 | 384 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 | 385 | |
3967deb1 | 386 | drbd_free_peer_req(mdev, peer_req); |
80a40e43 | 387 | defer: |
45bb912b | 388 | put_ldev(mdev); |
80a40e43 | 389 | return -EAGAIN; |
b411b363 PR |
390 | } |
391 | ||
99920dc5 | 392 | int w_resync_timer(struct drbd_work *w, int cancel) |
b411b363 | 393 | { |
00d56944 | 394 | struct drbd_conf *mdev = w->mdev; |
63106d3c PR |
395 | switch (mdev->state.conn) { |
396 | case C_VERIFY_S: | |
00d56944 | 397 | w_make_ov_request(w, cancel); |
63106d3c PR |
398 | break; |
399 | case C_SYNC_TARGET: | |
00d56944 | 400 | w_make_resync_request(w, cancel); |
63106d3c | 401 | break; |
b411b363 PR |
402 | } |
403 | ||
99920dc5 | 404 | return 0; |
794abb75 PR |
405 | } |
406 | ||
407 | void resync_timer_fn(unsigned long data) | |
408 | { | |
409 | struct drbd_conf *mdev = (struct drbd_conf *) data; | |
410 | ||
411 | if (list_empty(&mdev->resync_work.list)) | |
e42325a5 | 412 | drbd_queue_work(&mdev->tconn->data.work, &mdev->resync_work); |
b411b363 PR |
413 | } |
414 | ||
778f271d PR |
415 | static void fifo_set(struct fifo_buffer *fb, int value) |
416 | { | |
417 | int i; | |
418 | ||
419 | for (i = 0; i < fb->size; i++) | |
f10f2623 | 420 | fb->values[i] = value; |
778f271d PR |
421 | } |
422 | ||
423 | static int fifo_push(struct fifo_buffer *fb, int value) | |
424 | { | |
425 | int ov; | |
426 | ||
427 | ov = fb->values[fb->head_index]; | |
428 | fb->values[fb->head_index++] = value; | |
429 | ||
430 | if (fb->head_index >= fb->size) | |
431 | fb->head_index = 0; | |
432 | ||
433 | return ov; | |
434 | } | |
435 | ||
436 | static void fifo_add_val(struct fifo_buffer *fb, int value) | |
437 | { | |
438 | int i; | |
439 | ||
440 | for (i = 0; i < fb->size; i++) | |
441 | fb->values[i] += value; | |
442 | } | |
443 | ||
9958c857 PR |
444 | struct fifo_buffer *fifo_alloc(int fifo_size) |
445 | { | |
446 | struct fifo_buffer *fb; | |
447 | ||
448 | fb = kzalloc(sizeof(struct fifo_buffer) + sizeof(int) * fifo_size, GFP_KERNEL); | |
449 | if (!fb) | |
450 | return NULL; | |
451 | ||
452 | fb->head_index = 0; | |
453 | fb->size = fifo_size; | |
454 | fb->total = 0; | |
455 | ||
456 | return fb; | |
457 | } | |
458 | ||
9d77a5fe | 459 | static int drbd_rs_controller(struct drbd_conf *mdev) |
778f271d | 460 | { |
daeda1cc | 461 | struct disk_conf *dc; |
778f271d PR |
462 | unsigned int sect_in; /* Number of sectors that came in since the last turn */ |
463 | unsigned int want; /* The number of sectors we want in the proxy */ | |
464 | int req_sect; /* Number of sectors to request in this turn */ | |
465 | int correction; /* Number of sectors more we need in the proxy*/ | |
466 | int cps; /* correction per invocation of drbd_rs_controller() */ | |
467 | int steps; /* Number of time steps to plan ahead */ | |
468 | int curr_corr; | |
469 | int max_sect; | |
813472ce | 470 | struct fifo_buffer *plan; |
778f271d PR |
471 | |
472 | sect_in = atomic_xchg(&mdev->rs_sect_in, 0); /* Number of sectors that came in */ | |
473 | mdev->rs_in_flight -= sect_in; | |
474 | ||
daeda1cc | 475 | dc = rcu_dereference(mdev->ldev->disk_conf); |
813472ce | 476 | plan = rcu_dereference(mdev->rs_plan_s); |
778f271d | 477 | |
813472ce | 478 | steps = plan->size; /* (dc->c_plan_ahead * 10 * SLEEP_TIME) / HZ; */ |
778f271d PR |
479 | |
480 | if (mdev->rs_in_flight + sect_in == 0) { /* At start of resync */ | |
daeda1cc | 481 | want = ((dc->resync_rate * 2 * SLEEP_TIME) / HZ) * steps; |
778f271d | 482 | } else { /* normal path */ |
daeda1cc PR |
483 | want = dc->c_fill_target ? dc->c_fill_target : |
484 | sect_in * dc->c_delay_target * HZ / (SLEEP_TIME * 10); | |
778f271d PR |
485 | } |
486 | ||
813472ce | 487 | correction = want - mdev->rs_in_flight - plan->total; |
778f271d PR |
488 | |
489 | /* Plan ahead */ | |
490 | cps = correction / steps; | |
813472ce PR |
491 | fifo_add_val(plan, cps); |
492 | plan->total += cps * steps; | |
778f271d PR |
493 | |
494 | /* What we do in this step */ | |
813472ce PR |
495 | curr_corr = fifo_push(plan, 0); |
496 | plan->total -= curr_corr; | |
778f271d PR |
497 | |
498 | req_sect = sect_in + curr_corr; | |
499 | if (req_sect < 0) | |
500 | req_sect = 0; | |
501 | ||
daeda1cc | 502 | max_sect = (dc->c_max_rate * 2 * SLEEP_TIME) / HZ; |
778f271d PR |
503 | if (req_sect > max_sect) |
504 | req_sect = max_sect; | |
505 | ||
506 | /* | |
507 | dev_warn(DEV, "si=%u if=%d wa=%u co=%d st=%d cps=%d pl=%d cc=%d rs=%d\n", | |
508 | sect_in, mdev->rs_in_flight, want, correction, | |
509 | steps, cps, mdev->rs_planed, curr_corr, req_sect); | |
510 | */ | |
511 | ||
512 | return req_sect; | |
513 | } | |
514 | ||
9d77a5fe | 515 | static int drbd_rs_number_requests(struct drbd_conf *mdev) |
e65f440d LE |
516 | { |
517 | int number; | |
813472ce PR |
518 | |
519 | rcu_read_lock(); | |
520 | if (rcu_dereference(mdev->rs_plan_s)->size) { | |
e65f440d LE |
521 | number = drbd_rs_controller(mdev) >> (BM_BLOCK_SHIFT - 9); |
522 | mdev->c_sync_rate = number * HZ * (BM_BLOCK_SIZE / 1024) / SLEEP_TIME; | |
523 | } else { | |
daeda1cc | 524 | mdev->c_sync_rate = rcu_dereference(mdev->ldev->disk_conf)->resync_rate; |
e65f440d LE |
525 | number = SLEEP_TIME * mdev->c_sync_rate / ((BM_BLOCK_SIZE / 1024) * HZ); |
526 | } | |
813472ce | 527 | rcu_read_unlock(); |
e65f440d | 528 | |
e65f440d LE |
529 | /* ignore the amount of pending requests, the resync controller should |
530 | * throttle down to incoming reply rate soon enough anyways. */ | |
531 | return number; | |
532 | } | |
533 | ||
99920dc5 | 534 | int w_make_resync_request(struct drbd_work *w, int cancel) |
b411b363 | 535 | { |
00d56944 | 536 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
537 | unsigned long bit; |
538 | sector_t sector; | |
539 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); | |
1816a2b4 | 540 | int max_bio_size; |
e65f440d | 541 | int number, rollback_i, size; |
b411b363 | 542 | int align, queued, sndbuf; |
0f0601f4 | 543 | int i = 0; |
b411b363 PR |
544 | |
545 | if (unlikely(cancel)) | |
99920dc5 | 546 | return 0; |
b411b363 | 547 | |
af85e8e8 LE |
548 | if (mdev->rs_total == 0) { |
549 | /* empty resync? */ | |
550 | drbd_resync_finished(mdev); | |
99920dc5 | 551 | return 0; |
af85e8e8 LE |
552 | } |
553 | ||
b411b363 PR |
554 | if (!get_ldev(mdev)) { |
555 | /* Since we only need to access mdev->rsync a | |
556 | get_ldev_if_state(mdev,D_FAILED) would be sufficient, but | |
557 | to continue resync with a broken disk makes no sense at | |
558 | all */ | |
559 | dev_err(DEV, "Disk broke down during resync!\n"); | |
99920dc5 | 560 | return 0; |
b411b363 PR |
561 | } |
562 | ||
0cfdd247 | 563 | max_bio_size = queue_max_hw_sectors(mdev->rq_queue) << 9; |
e65f440d LE |
564 | number = drbd_rs_number_requests(mdev); |
565 | if (number == 0) | |
0f0601f4 | 566 | goto requeue; |
b411b363 | 567 | |
b411b363 PR |
568 | for (i = 0; i < number; i++) { |
569 | /* Stop generating RS requests, when half of the send buffer is filled */ | |
e42325a5 PR |
570 | mutex_lock(&mdev->tconn->data.mutex); |
571 | if (mdev->tconn->data.socket) { | |
572 | queued = mdev->tconn->data.socket->sk->sk_wmem_queued; | |
573 | sndbuf = mdev->tconn->data.socket->sk->sk_sndbuf; | |
b411b363 PR |
574 | } else { |
575 | queued = 1; | |
576 | sndbuf = 0; | |
577 | } | |
e42325a5 | 578 | mutex_unlock(&mdev->tconn->data.mutex); |
b411b363 PR |
579 | if (queued > sndbuf / 2) |
580 | goto requeue; | |
581 | ||
582 | next_sector: | |
583 | size = BM_BLOCK_SIZE; | |
584 | bit = drbd_bm_find_next(mdev, mdev->bm_resync_fo); | |
585 | ||
4b0715f0 | 586 | if (bit == DRBD_END_OF_BITMAP) { |
b411b363 | 587 | mdev->bm_resync_fo = drbd_bm_bits(mdev); |
b411b363 | 588 | put_ldev(mdev); |
99920dc5 | 589 | return 0; |
b411b363 PR |
590 | } |
591 | ||
592 | sector = BM_BIT_TO_SECT(bit); | |
593 | ||
e3555d85 PR |
594 | if (drbd_rs_should_slow_down(mdev, sector) || |
595 | drbd_try_rs_begin_io(mdev, sector)) { | |
b411b363 PR |
596 | mdev->bm_resync_fo = bit; |
597 | goto requeue; | |
598 | } | |
599 | mdev->bm_resync_fo = bit + 1; | |
600 | ||
601 | if (unlikely(drbd_bm_test_bit(mdev, bit) == 0)) { | |
602 | drbd_rs_complete_io(mdev, sector); | |
603 | goto next_sector; | |
604 | } | |
605 | ||
1816a2b4 | 606 | #if DRBD_MAX_BIO_SIZE > BM_BLOCK_SIZE |
b411b363 PR |
607 | /* try to find some adjacent bits. |
608 | * we stop if we have already the maximum req size. | |
609 | * | |
610 | * Additionally always align bigger requests, in order to | |
611 | * be prepared for all stripe sizes of software RAIDs. | |
b411b363 PR |
612 | */ |
613 | align = 1; | |
d207450c | 614 | rollback_i = i; |
b411b363 | 615 | for (;;) { |
1816a2b4 | 616 | if (size + BM_BLOCK_SIZE > max_bio_size) |
b411b363 PR |
617 | break; |
618 | ||
619 | /* Be always aligned */ | |
620 | if (sector & ((1<<(align+3))-1)) | |
621 | break; | |
622 | ||
623 | /* do not cross extent boundaries */ | |
624 | if (((bit+1) & BM_BLOCKS_PER_BM_EXT_MASK) == 0) | |
625 | break; | |
626 | /* now, is it actually dirty, after all? | |
627 | * caution, drbd_bm_test_bit is tri-state for some | |
628 | * obscure reason; ( b == 0 ) would get the out-of-band | |
629 | * only accidentally right because of the "oddly sized" | |
630 | * adjustment below */ | |
631 | if (drbd_bm_test_bit(mdev, bit+1) != 1) | |
632 | break; | |
633 | bit++; | |
634 | size += BM_BLOCK_SIZE; | |
635 | if ((BM_BLOCK_SIZE << align) <= size) | |
636 | align++; | |
637 | i++; | |
638 | } | |
639 | /* if we merged some, | |
640 | * reset the offset to start the next drbd_bm_find_next from */ | |
641 | if (size > BM_BLOCK_SIZE) | |
642 | mdev->bm_resync_fo = bit + 1; | |
643 | #endif | |
644 | ||
645 | /* adjust very last sectors, in case we are oddly sized */ | |
646 | if (sector + (size>>9) > capacity) | |
647 | size = (capacity-sector)<<9; | |
f399002e | 648 | if (mdev->tconn->agreed_pro_version >= 89 && mdev->tconn->csums_tfm) { |
b411b363 | 649 | switch (read_for_csum(mdev, sector, size)) { |
80a40e43 | 650 | case -EIO: /* Disk failure */ |
b411b363 | 651 | put_ldev(mdev); |
99920dc5 | 652 | return -EIO; |
80a40e43 | 653 | case -EAGAIN: /* allocation failed, or ldev busy */ |
b411b363 PR |
654 | drbd_rs_complete_io(mdev, sector); |
655 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
d207450c | 656 | i = rollback_i; |
b411b363 | 657 | goto requeue; |
80a40e43 LE |
658 | case 0: |
659 | /* everything ok */ | |
660 | break; | |
661 | default: | |
662 | BUG(); | |
b411b363 PR |
663 | } |
664 | } else { | |
99920dc5 AG |
665 | int err; |
666 | ||
b411b363 | 667 | inc_rs_pending(mdev); |
99920dc5 AG |
668 | err = drbd_send_drequest(mdev, P_RS_DATA_REQUEST, |
669 | sector, size, ID_SYNCER); | |
670 | if (err) { | |
b411b363 PR |
671 | dev_err(DEV, "drbd_send_drequest() failed, aborting...\n"); |
672 | dec_rs_pending(mdev); | |
673 | put_ldev(mdev); | |
99920dc5 | 674 | return err; |
b411b363 PR |
675 | } |
676 | } | |
677 | } | |
678 | ||
679 | if (mdev->bm_resync_fo >= drbd_bm_bits(mdev)) { | |
680 | /* last syncer _request_ was sent, | |
681 | * but the P_RS_DATA_REPLY not yet received. sync will end (and | |
682 | * next sync group will resume), as soon as we receive the last | |
683 | * resync data block, and the last bit is cleared. | |
684 | * until then resync "work" is "inactive" ... | |
685 | */ | |
b411b363 | 686 | put_ldev(mdev); |
99920dc5 | 687 | return 0; |
b411b363 PR |
688 | } |
689 | ||
690 | requeue: | |
778f271d | 691 | mdev->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
b411b363 PR |
692 | mod_timer(&mdev->resync_timer, jiffies + SLEEP_TIME); |
693 | put_ldev(mdev); | |
99920dc5 | 694 | return 0; |
b411b363 PR |
695 | } |
696 | ||
00d56944 | 697 | static int w_make_ov_request(struct drbd_work *w, int cancel) |
b411b363 | 698 | { |
00d56944 | 699 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
700 | int number, i, size; |
701 | sector_t sector; | |
702 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); | |
703 | ||
704 | if (unlikely(cancel)) | |
705 | return 1; | |
706 | ||
2649f080 | 707 | number = drbd_rs_number_requests(mdev); |
b411b363 PR |
708 | |
709 | sector = mdev->ov_position; | |
710 | for (i = 0; i < number; i++) { | |
711 | if (sector >= capacity) { | |
b411b363 PR |
712 | return 1; |
713 | } | |
714 | ||
715 | size = BM_BLOCK_SIZE; | |
716 | ||
e3555d85 PR |
717 | if (drbd_rs_should_slow_down(mdev, sector) || |
718 | drbd_try_rs_begin_io(mdev, sector)) { | |
b411b363 PR |
719 | mdev->ov_position = sector; |
720 | goto requeue; | |
721 | } | |
722 | ||
723 | if (sector + (size>>9) > capacity) | |
724 | size = (capacity-sector)<<9; | |
725 | ||
726 | inc_rs_pending(mdev); | |
5b9f499c | 727 | if (drbd_send_ov_request(mdev, sector, size)) { |
b411b363 PR |
728 | dec_rs_pending(mdev); |
729 | return 0; | |
730 | } | |
731 | sector += BM_SECT_PER_BIT; | |
732 | } | |
733 | mdev->ov_position = sector; | |
734 | ||
735 | requeue: | |
2649f080 | 736 | mdev->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
b411b363 PR |
737 | mod_timer(&mdev->resync_timer, jiffies + SLEEP_TIME); |
738 | return 1; | |
739 | } | |
740 | ||
99920dc5 | 741 | int w_ov_finished(struct drbd_work *w, int cancel) |
b411b363 | 742 | { |
00d56944 | 743 | struct drbd_conf *mdev = w->mdev; |
b411b363 | 744 | kfree(w); |
8f7bed77 | 745 | ov_out_of_sync_print(mdev); |
b411b363 PR |
746 | drbd_resync_finished(mdev); |
747 | ||
99920dc5 | 748 | return 0; |
b411b363 PR |
749 | } |
750 | ||
99920dc5 | 751 | static int w_resync_finished(struct drbd_work *w, int cancel) |
b411b363 | 752 | { |
00d56944 | 753 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
754 | kfree(w); |
755 | ||
756 | drbd_resync_finished(mdev); | |
757 | ||
99920dc5 | 758 | return 0; |
b411b363 PR |
759 | } |
760 | ||
af85e8e8 LE |
761 | static void ping_peer(struct drbd_conf *mdev) |
762 | { | |
2a67d8b9 PR |
763 | struct drbd_tconn *tconn = mdev->tconn; |
764 | ||
765 | clear_bit(GOT_PING_ACK, &tconn->flags); | |
766 | request_ping(tconn); | |
767 | wait_event(tconn->ping_wait, | |
768 | test_bit(GOT_PING_ACK, &tconn->flags) || mdev->state.conn < C_CONNECTED); | |
af85e8e8 LE |
769 | } |
770 | ||
b411b363 PR |
771 | int drbd_resync_finished(struct drbd_conf *mdev) |
772 | { | |
773 | unsigned long db, dt, dbdt; | |
774 | unsigned long n_oos; | |
775 | union drbd_state os, ns; | |
776 | struct drbd_work *w; | |
777 | char *khelper_cmd = NULL; | |
26525618 | 778 | int verify_done = 0; |
b411b363 PR |
779 | |
780 | /* Remove all elements from the resync LRU. Since future actions | |
781 | * might set bits in the (main) bitmap, then the entries in the | |
782 | * resync LRU would be wrong. */ | |
783 | if (drbd_rs_del_all(mdev)) { | |
784 | /* In case this is not possible now, most probably because | |
785 | * there are P_RS_DATA_REPLY Packets lingering on the worker's | |
786 | * queue (or even the read operations for those packets | |
787 | * is not finished by now). Retry in 100ms. */ | |
788 | ||
20ee6390 | 789 | schedule_timeout_interruptible(HZ / 10); |
b411b363 PR |
790 | w = kmalloc(sizeof(struct drbd_work), GFP_ATOMIC); |
791 | if (w) { | |
792 | w->cb = w_resync_finished; | |
9b743da9 | 793 | w->mdev = mdev; |
e42325a5 | 794 | drbd_queue_work(&mdev->tconn->data.work, w); |
b411b363 PR |
795 | return 1; |
796 | } | |
797 | dev_err(DEV, "Warn failed to drbd_rs_del_all() and to kmalloc(w).\n"); | |
798 | } | |
799 | ||
800 | dt = (jiffies - mdev->rs_start - mdev->rs_paused) / HZ; | |
801 | if (dt <= 0) | |
802 | dt = 1; | |
803 | db = mdev->rs_total; | |
804 | dbdt = Bit2KB(db/dt); | |
805 | mdev->rs_paused /= HZ; | |
806 | ||
807 | if (!get_ldev(mdev)) | |
808 | goto out; | |
809 | ||
af85e8e8 LE |
810 | ping_peer(mdev); |
811 | ||
87eeee41 | 812 | spin_lock_irq(&mdev->tconn->req_lock); |
78bae59b | 813 | os = drbd_read_state(mdev); |
b411b363 | 814 | |
26525618 LE |
815 | verify_done = (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T); |
816 | ||
b411b363 PR |
817 | /* This protects us against multiple calls (that can happen in the presence |
818 | of application IO), and against connectivity loss just before we arrive here. */ | |
819 | if (os.conn <= C_CONNECTED) | |
820 | goto out_unlock; | |
821 | ||
822 | ns = os; | |
823 | ns.conn = C_CONNECTED; | |
824 | ||
825 | dev_info(DEV, "%s done (total %lu sec; paused %lu sec; %lu K/sec)\n", | |
26525618 | 826 | verify_done ? "Online verify " : "Resync", |
b411b363 PR |
827 | dt + mdev->rs_paused, mdev->rs_paused, dbdt); |
828 | ||
829 | n_oos = drbd_bm_total_weight(mdev); | |
830 | ||
831 | if (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) { | |
832 | if (n_oos) { | |
833 | dev_alert(DEV, "Online verify found %lu %dk block out of sync!\n", | |
834 | n_oos, Bit2KB(1)); | |
835 | khelper_cmd = "out-of-sync"; | |
836 | } | |
837 | } else { | |
838 | D_ASSERT((n_oos - mdev->rs_failed) == 0); | |
839 | ||
840 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) | |
841 | khelper_cmd = "after-resync-target"; | |
842 | ||
f399002e | 843 | if (mdev->tconn->csums_tfm && mdev->rs_total) { |
b411b363 PR |
844 | const unsigned long s = mdev->rs_same_csum; |
845 | const unsigned long t = mdev->rs_total; | |
846 | const int ratio = | |
847 | (t == 0) ? 0 : | |
848 | (t < 100000) ? ((s*100)/t) : (s/(t/100)); | |
24c4830c | 849 | dev_info(DEV, "%u %% had equal checksums, eliminated: %luK; " |
b411b363 PR |
850 | "transferred %luK total %luK\n", |
851 | ratio, | |
852 | Bit2KB(mdev->rs_same_csum), | |
853 | Bit2KB(mdev->rs_total - mdev->rs_same_csum), | |
854 | Bit2KB(mdev->rs_total)); | |
855 | } | |
856 | } | |
857 | ||
858 | if (mdev->rs_failed) { | |
859 | dev_info(DEV, " %lu failed blocks\n", mdev->rs_failed); | |
860 | ||
861 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
862 | ns.disk = D_INCONSISTENT; | |
863 | ns.pdsk = D_UP_TO_DATE; | |
864 | } else { | |
865 | ns.disk = D_UP_TO_DATE; | |
866 | ns.pdsk = D_INCONSISTENT; | |
867 | } | |
868 | } else { | |
869 | ns.disk = D_UP_TO_DATE; | |
870 | ns.pdsk = D_UP_TO_DATE; | |
871 | ||
872 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
873 | if (mdev->p_uuid) { | |
874 | int i; | |
875 | for (i = UI_BITMAP ; i <= UI_HISTORY_END ; i++) | |
876 | _drbd_uuid_set(mdev, i, mdev->p_uuid[i]); | |
877 | drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_CURRENT]); | |
878 | _drbd_uuid_set(mdev, UI_CURRENT, mdev->p_uuid[UI_CURRENT]); | |
879 | } else { | |
880 | dev_err(DEV, "mdev->p_uuid is NULL! BUG\n"); | |
881 | } | |
882 | } | |
883 | ||
62b0da3a LE |
884 | if (!(os.conn == C_VERIFY_S || os.conn == C_VERIFY_T)) { |
885 | /* for verify runs, we don't update uuids here, | |
886 | * so there would be nothing to report. */ | |
887 | drbd_uuid_set_bm(mdev, 0UL); | |
888 | drbd_print_uuids(mdev, "updated UUIDs"); | |
889 | if (mdev->p_uuid) { | |
890 | /* Now the two UUID sets are equal, update what we | |
891 | * know of the peer. */ | |
892 | int i; | |
893 | for (i = UI_CURRENT ; i <= UI_HISTORY_END ; i++) | |
894 | mdev->p_uuid[i] = mdev->ldev->md.uuid[i]; | |
895 | } | |
b411b363 PR |
896 | } |
897 | } | |
898 | ||
899 | _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); | |
900 | out_unlock: | |
87eeee41 | 901 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
902 | put_ldev(mdev); |
903 | out: | |
904 | mdev->rs_total = 0; | |
905 | mdev->rs_failed = 0; | |
906 | mdev->rs_paused = 0; | |
26525618 LE |
907 | if (verify_done) |
908 | mdev->ov_start_sector = 0; | |
b411b363 | 909 | |
13d42685 LE |
910 | drbd_md_sync(mdev); |
911 | ||
b411b363 PR |
912 | if (khelper_cmd) |
913 | drbd_khelper(mdev, khelper_cmd); | |
914 | ||
915 | return 1; | |
916 | } | |
917 | ||
918 | /* helper */ | |
db830c46 | 919 | static void move_to_net_ee_or_free(struct drbd_conf *mdev, struct drbd_peer_request *peer_req) |
b411b363 | 920 | { |
045417f7 | 921 | if (drbd_peer_req_has_active_page(peer_req)) { |
b411b363 | 922 | /* This might happen if sendpage() has not finished */ |
db830c46 | 923 | int i = (peer_req->i.size + PAGE_SIZE -1) >> PAGE_SHIFT; |
435f0740 LE |
924 | atomic_add(i, &mdev->pp_in_use_by_net); |
925 | atomic_sub(i, &mdev->pp_in_use); | |
87eeee41 | 926 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 927 | list_add_tail(&peer_req->w.list, &mdev->net_ee); |
87eeee41 | 928 | spin_unlock_irq(&mdev->tconn->req_lock); |
435f0740 | 929 | wake_up(&drbd_pp_wait); |
b411b363 | 930 | } else |
3967deb1 | 931 | drbd_free_peer_req(mdev, peer_req); |
b411b363 PR |
932 | } |
933 | ||
934 | /** | |
935 | * w_e_end_data_req() - Worker callback, to send a P_DATA_REPLY packet in response to a P_DATA_REQUEST | |
936 | * @mdev: DRBD device. | |
937 | * @w: work object. | |
938 | * @cancel: The connection will be closed anyways | |
939 | */ | |
99920dc5 | 940 | int w_e_end_data_req(struct drbd_work *w, int cancel) |
b411b363 | 941 | { |
db830c46 | 942 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 943 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 944 | int err; |
b411b363 PR |
945 | |
946 | if (unlikely(cancel)) { | |
3967deb1 | 947 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 948 | dec_unacked(mdev); |
99920dc5 | 949 | return 0; |
b411b363 PR |
950 | } |
951 | ||
db830c46 | 952 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
99920dc5 | 953 | err = drbd_send_block(mdev, P_DATA_REPLY, peer_req); |
b411b363 PR |
954 | } else { |
955 | if (__ratelimit(&drbd_ratelimit_state)) | |
956 | dev_err(DEV, "Sending NegDReply. sector=%llus.\n", | |
db830c46 | 957 | (unsigned long long)peer_req->i.sector); |
b411b363 | 958 | |
99920dc5 | 959 | err = drbd_send_ack(mdev, P_NEG_DREPLY, peer_req); |
b411b363 PR |
960 | } |
961 | ||
962 | dec_unacked(mdev); | |
963 | ||
db830c46 | 964 | move_to_net_ee_or_free(mdev, peer_req); |
b411b363 | 965 | |
99920dc5 | 966 | if (unlikely(err)) |
b411b363 | 967 | dev_err(DEV, "drbd_send_block() failed\n"); |
99920dc5 | 968 | return err; |
b411b363 PR |
969 | } |
970 | ||
971 | /** | |
972 | * w_e_end_rsdata_req() - Worker callback to send a P_RS_DATA_REPLY packet in response to a P_RS_DATA_REQUESTRS | |
973 | * @mdev: DRBD device. | |
974 | * @w: work object. | |
975 | * @cancel: The connection will be closed anyways | |
976 | */ | |
99920dc5 | 977 | int w_e_end_rsdata_req(struct drbd_work *w, int cancel) |
b411b363 | 978 | { |
db830c46 | 979 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 980 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 981 | int err; |
b411b363 PR |
982 | |
983 | if (unlikely(cancel)) { | |
3967deb1 | 984 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 985 | dec_unacked(mdev); |
99920dc5 | 986 | return 0; |
b411b363 PR |
987 | } |
988 | ||
989 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
db830c46 | 990 | drbd_rs_complete_io(mdev, peer_req->i.sector); |
b411b363 PR |
991 | put_ldev(mdev); |
992 | } | |
993 | ||
d612d309 | 994 | if (mdev->state.conn == C_AHEAD) { |
99920dc5 | 995 | err = drbd_send_ack(mdev, P_RS_CANCEL, peer_req); |
db830c46 | 996 | } else if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
997 | if (likely(mdev->state.pdsk >= D_INCONSISTENT)) { |
998 | inc_rs_pending(mdev); | |
99920dc5 | 999 | err = drbd_send_block(mdev, P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1000 | } else { |
1001 | if (__ratelimit(&drbd_ratelimit_state)) | |
1002 | dev_err(DEV, "Not sending RSDataReply, " | |
1003 | "partner DISKLESS!\n"); | |
99920dc5 | 1004 | err = 0; |
b411b363 PR |
1005 | } |
1006 | } else { | |
1007 | if (__ratelimit(&drbd_ratelimit_state)) | |
1008 | dev_err(DEV, "Sending NegRSDReply. sector %llus.\n", | |
db830c46 | 1009 | (unsigned long long)peer_req->i.sector); |
b411b363 | 1010 | |
99920dc5 | 1011 | err = drbd_send_ack(mdev, P_NEG_RS_DREPLY, peer_req); |
b411b363 PR |
1012 | |
1013 | /* update resync data with failure */ | |
db830c46 | 1014 | drbd_rs_failed_io(mdev, peer_req->i.sector, peer_req->i.size); |
b411b363 PR |
1015 | } |
1016 | ||
1017 | dec_unacked(mdev); | |
1018 | ||
db830c46 | 1019 | move_to_net_ee_or_free(mdev, peer_req); |
b411b363 | 1020 | |
99920dc5 | 1021 | if (unlikely(err)) |
b411b363 | 1022 | dev_err(DEV, "drbd_send_block() failed\n"); |
99920dc5 | 1023 | return err; |
b411b363 PR |
1024 | } |
1025 | ||
99920dc5 | 1026 | int w_e_end_csum_rs_req(struct drbd_work *w, int cancel) |
b411b363 | 1027 | { |
db830c46 | 1028 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 1029 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
1030 | struct digest_info *di; |
1031 | int digest_size; | |
1032 | void *digest = NULL; | |
99920dc5 | 1033 | int err, eq = 0; |
b411b363 PR |
1034 | |
1035 | if (unlikely(cancel)) { | |
3967deb1 | 1036 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 1037 | dec_unacked(mdev); |
99920dc5 | 1038 | return 0; |
b411b363 PR |
1039 | } |
1040 | ||
1d53f09e | 1041 | if (get_ldev(mdev)) { |
db830c46 | 1042 | drbd_rs_complete_io(mdev, peer_req->i.sector); |
1d53f09e LE |
1043 | put_ldev(mdev); |
1044 | } | |
b411b363 | 1045 | |
db830c46 | 1046 | di = peer_req->digest; |
b411b363 | 1047 | |
db830c46 | 1048 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1049 | /* quick hack to try to avoid a race against reconfiguration. |
1050 | * a real fix would be much more involved, | |
1051 | * introducing more locking mechanisms */ | |
f399002e LE |
1052 | if (mdev->tconn->csums_tfm) { |
1053 | digest_size = crypto_hash_digestsize(mdev->tconn->csums_tfm); | |
b411b363 PR |
1054 | D_ASSERT(digest_size == di->digest_size); |
1055 | digest = kmalloc(digest_size, GFP_NOIO); | |
1056 | } | |
1057 | if (digest) { | |
f399002e | 1058 | drbd_csum_ee(mdev, mdev->tconn->csums_tfm, peer_req, digest); |
b411b363 PR |
1059 | eq = !memcmp(digest, di->digest, digest_size); |
1060 | kfree(digest); | |
1061 | } | |
1062 | ||
1063 | if (eq) { | |
db830c46 | 1064 | drbd_set_in_sync(mdev, peer_req->i.sector, peer_req->i.size); |
676396d5 | 1065 | /* rs_same_csums unit is BM_BLOCK_SIZE */ |
db830c46 | 1066 | mdev->rs_same_csum += peer_req->i.size >> BM_BLOCK_SHIFT; |
99920dc5 | 1067 | err = drbd_send_ack(mdev, P_RS_IS_IN_SYNC, peer_req); |
b411b363 PR |
1068 | } else { |
1069 | inc_rs_pending(mdev); | |
db830c46 AG |
1070 | peer_req->block_id = ID_SYNCER; /* By setting block_id, digest pointer becomes invalid! */ |
1071 | peer_req->flags &= ~EE_HAS_DIGEST; /* This peer request no longer has a digest pointer */ | |
204bba99 | 1072 | kfree(di); |
99920dc5 | 1073 | err = drbd_send_block(mdev, P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1074 | } |
1075 | } else { | |
99920dc5 | 1076 | err = drbd_send_ack(mdev, P_NEG_RS_DREPLY, peer_req); |
b411b363 PR |
1077 | if (__ratelimit(&drbd_ratelimit_state)) |
1078 | dev_err(DEV, "Sending NegDReply. I guess it gets messy.\n"); | |
1079 | } | |
1080 | ||
1081 | dec_unacked(mdev); | |
db830c46 | 1082 | move_to_net_ee_or_free(mdev, peer_req); |
b411b363 | 1083 | |
99920dc5 | 1084 | if (unlikely(err)) |
b411b363 | 1085 | dev_err(DEV, "drbd_send_block/ack() failed\n"); |
99920dc5 | 1086 | return err; |
b411b363 PR |
1087 | } |
1088 | ||
99920dc5 | 1089 | int w_e_end_ov_req(struct drbd_work *w, int cancel) |
b411b363 | 1090 | { |
db830c46 | 1091 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 1092 | struct drbd_conf *mdev = w->mdev; |
db830c46 AG |
1093 | sector_t sector = peer_req->i.sector; |
1094 | unsigned int size = peer_req->i.size; | |
b411b363 PR |
1095 | int digest_size; |
1096 | void *digest; | |
99920dc5 | 1097 | int err = 0; |
b411b363 PR |
1098 | |
1099 | if (unlikely(cancel)) | |
1100 | goto out; | |
1101 | ||
f399002e | 1102 | digest_size = crypto_hash_digestsize(mdev->tconn->verify_tfm); |
b411b363 | 1103 | digest = kmalloc(digest_size, GFP_NOIO); |
8f21420e | 1104 | if (!digest) { |
99920dc5 | 1105 | err = 1; /* terminate the connection in case the allocation failed */ |
8f21420e | 1106 | goto out; |
b411b363 PR |
1107 | } |
1108 | ||
db830c46 | 1109 | if (likely(!(peer_req->flags & EE_WAS_ERROR))) |
f399002e | 1110 | drbd_csum_ee(mdev, mdev->tconn->verify_tfm, peer_req, digest); |
8f21420e PR |
1111 | else |
1112 | memset(digest, 0, digest_size); | |
1113 | ||
53ea4331 LE |
1114 | /* Free e and pages before send. |
1115 | * In case we block on congestion, we could otherwise run into | |
1116 | * some distributed deadlock, if the other side blocks on | |
1117 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1118 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
3967deb1 | 1119 | drbd_free_peer_req(mdev, peer_req); |
db830c46 | 1120 | peer_req = NULL; |
8f21420e | 1121 | inc_rs_pending(mdev); |
99920dc5 AG |
1122 | err = drbd_send_drequest_csum(mdev, sector, size, digest, digest_size, P_OV_REPLY); |
1123 | if (err) | |
8f21420e PR |
1124 | dec_rs_pending(mdev); |
1125 | kfree(digest); | |
1126 | ||
b411b363 | 1127 | out: |
db830c46 | 1128 | if (peer_req) |
3967deb1 | 1129 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 1130 | dec_unacked(mdev); |
99920dc5 | 1131 | return err; |
b411b363 PR |
1132 | } |
1133 | ||
8f7bed77 | 1134 | void drbd_ov_out_of_sync_found(struct drbd_conf *mdev, sector_t sector, int size) |
b411b363 PR |
1135 | { |
1136 | if (mdev->ov_last_oos_start + mdev->ov_last_oos_size == sector) { | |
1137 | mdev->ov_last_oos_size += size>>9; | |
1138 | } else { | |
1139 | mdev->ov_last_oos_start = sector; | |
1140 | mdev->ov_last_oos_size = size>>9; | |
1141 | } | |
1142 | drbd_set_out_of_sync(mdev, sector, size); | |
b411b363 PR |
1143 | } |
1144 | ||
99920dc5 | 1145 | int w_e_end_ov_reply(struct drbd_work *w, int cancel) |
b411b363 | 1146 | { |
db830c46 | 1147 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 1148 | struct drbd_conf *mdev = w->mdev; |
b411b363 | 1149 | struct digest_info *di; |
b411b363 | 1150 | void *digest; |
db830c46 AG |
1151 | sector_t sector = peer_req->i.sector; |
1152 | unsigned int size = peer_req->i.size; | |
53ea4331 | 1153 | int digest_size; |
99920dc5 | 1154 | int err, eq = 0; |
b411b363 PR |
1155 | |
1156 | if (unlikely(cancel)) { | |
3967deb1 | 1157 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 1158 | dec_unacked(mdev); |
99920dc5 | 1159 | return 0; |
b411b363 PR |
1160 | } |
1161 | ||
1162 | /* after "cancel", because after drbd_disconnect/drbd_rs_cancel_all | |
1163 | * the resync lru has been cleaned up already */ | |
1d53f09e | 1164 | if (get_ldev(mdev)) { |
db830c46 | 1165 | drbd_rs_complete_io(mdev, peer_req->i.sector); |
1d53f09e LE |
1166 | put_ldev(mdev); |
1167 | } | |
b411b363 | 1168 | |
db830c46 | 1169 | di = peer_req->digest; |
b411b363 | 1170 | |
db830c46 | 1171 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
f399002e | 1172 | digest_size = crypto_hash_digestsize(mdev->tconn->verify_tfm); |
b411b363 PR |
1173 | digest = kmalloc(digest_size, GFP_NOIO); |
1174 | if (digest) { | |
f399002e | 1175 | drbd_csum_ee(mdev, mdev->tconn->verify_tfm, peer_req, digest); |
b411b363 PR |
1176 | |
1177 | D_ASSERT(digest_size == di->digest_size); | |
1178 | eq = !memcmp(digest, di->digest, digest_size); | |
1179 | kfree(digest); | |
1180 | } | |
b411b363 PR |
1181 | } |
1182 | ||
9676c760 LE |
1183 | /* Free peer_req and pages before send. |
1184 | * In case we block on congestion, we could otherwise run into | |
1185 | * some distributed deadlock, if the other side blocks on | |
1186 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1187 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
3967deb1 | 1188 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 1189 | if (!eq) |
8f7bed77 | 1190 | drbd_ov_out_of_sync_found(mdev, sector, size); |
b411b363 | 1191 | else |
8f7bed77 | 1192 | ov_out_of_sync_print(mdev); |
b411b363 | 1193 | |
99920dc5 | 1194 | err = drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, |
fa79abd8 | 1195 | eq ? ID_IN_SYNC : ID_OUT_OF_SYNC); |
b411b363 | 1196 | |
53ea4331 | 1197 | dec_unacked(mdev); |
b411b363 | 1198 | |
ea5442af LE |
1199 | --mdev->ov_left; |
1200 | ||
1201 | /* let's advance progress step marks only for every other megabyte */ | |
1202 | if ((mdev->ov_left & 0x200) == 0x200) | |
1203 | drbd_advance_rs_marks(mdev, mdev->ov_left); | |
1204 | ||
1205 | if (mdev->ov_left == 0) { | |
8f7bed77 | 1206 | ov_out_of_sync_print(mdev); |
b411b363 PR |
1207 | drbd_resync_finished(mdev); |
1208 | } | |
1209 | ||
99920dc5 | 1210 | return err; |
b411b363 PR |
1211 | } |
1212 | ||
99920dc5 | 1213 | int w_prev_work_done(struct drbd_work *w, int cancel) |
b411b363 PR |
1214 | { |
1215 | struct drbd_wq_barrier *b = container_of(w, struct drbd_wq_barrier, w); | |
00d56944 | 1216 | |
b411b363 | 1217 | complete(&b->done); |
99920dc5 | 1218 | return 0; |
b411b363 PR |
1219 | } |
1220 | ||
99920dc5 | 1221 | int w_send_barrier(struct drbd_work *w, int cancel) |
b411b363 | 1222 | { |
9f5bdc33 | 1223 | struct drbd_socket *sock; |
b411b363 | 1224 | struct drbd_tl_epoch *b = container_of(w, struct drbd_tl_epoch, w); |
00d56944 | 1225 | struct drbd_conf *mdev = w->mdev; |
9f5bdc33 | 1226 | struct p_barrier *p; |
b411b363 PR |
1227 | |
1228 | /* really avoid racing with tl_clear. w.cb may have been referenced | |
1229 | * just before it was reassigned and re-queued, so double check that. | |
1230 | * actually, this race was harmless, since we only try to send the | |
1231 | * barrier packet here, and otherwise do nothing with the object. | |
1232 | * but compare with the head of w_clear_epoch */ | |
87eeee41 | 1233 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
1234 | if (w->cb != w_send_barrier || mdev->state.conn < C_CONNECTED) |
1235 | cancel = 1; | |
87eeee41 | 1236 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 1237 | if (cancel) |
b411b363 | 1238 | return 0; |
99920dc5 | 1239 | |
9f5bdc33 AG |
1240 | sock = &mdev->tconn->data; |
1241 | p = drbd_prepare_command(mdev, sock); | |
1242 | if (!p) | |
1243 | return -EIO; | |
b411b363 PR |
1244 | p->barrier = b->br_number; |
1245 | /* inc_ap_pending was done where this was queued. | |
1246 | * dec_ap_pending will be done in got_BarrierAck | |
1247 | * or (on connection loss) in w_clear_epoch. */ | |
9f5bdc33 | 1248 | return drbd_send_command(mdev, sock, P_BARRIER, sizeof(*p), NULL, 0); |
b411b363 PR |
1249 | } |
1250 | ||
99920dc5 | 1251 | int w_send_write_hint(struct drbd_work *w, int cancel) |
b411b363 | 1252 | { |
00d56944 | 1253 | struct drbd_conf *mdev = w->mdev; |
9f5bdc33 AG |
1254 | struct drbd_socket *sock; |
1255 | ||
b411b363 | 1256 | if (cancel) |
99920dc5 | 1257 | return 0; |
9f5bdc33 AG |
1258 | sock = &mdev->tconn->data; |
1259 | if (!drbd_prepare_command(mdev, sock)) | |
1260 | return -EIO; | |
e658983a | 1261 | return drbd_send_command(mdev, sock, P_UNPLUG_REMOTE, 0, NULL, 0); |
b411b363 PR |
1262 | } |
1263 | ||
8f7bed77 | 1264 | int w_send_out_of_sync(struct drbd_work *w, int cancel) |
73a01a18 PR |
1265 | { |
1266 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 1267 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 1268 | int err; |
73a01a18 PR |
1269 | |
1270 | if (unlikely(cancel)) { | |
8554df1c | 1271 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1272 | return 0; |
73a01a18 PR |
1273 | } |
1274 | ||
8f7bed77 | 1275 | err = drbd_send_out_of_sync(mdev, req); |
8554df1c | 1276 | req_mod(req, OOS_HANDED_TO_NETWORK); |
73a01a18 | 1277 | |
99920dc5 | 1278 | return err; |
73a01a18 PR |
1279 | } |
1280 | ||
b411b363 PR |
1281 | /** |
1282 | * w_send_dblock() - Worker callback to send a P_DATA packet in order to mirror a write request | |
1283 | * @mdev: DRBD device. | |
1284 | * @w: work object. | |
1285 | * @cancel: The connection will be closed anyways | |
1286 | */ | |
99920dc5 | 1287 | int w_send_dblock(struct drbd_work *w, int cancel) |
b411b363 PR |
1288 | { |
1289 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 1290 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 1291 | int err; |
b411b363 PR |
1292 | |
1293 | if (unlikely(cancel)) { | |
8554df1c | 1294 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1295 | return 0; |
b411b363 PR |
1296 | } |
1297 | ||
99920dc5 AG |
1298 | err = drbd_send_dblock(mdev, req); |
1299 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); | |
b411b363 | 1300 | |
99920dc5 | 1301 | return err; |
b411b363 PR |
1302 | } |
1303 | ||
1304 | /** | |
1305 | * w_send_read_req() - Worker callback to send a read request (P_DATA_REQUEST) packet | |
1306 | * @mdev: DRBD device. | |
1307 | * @w: work object. | |
1308 | * @cancel: The connection will be closed anyways | |
1309 | */ | |
99920dc5 | 1310 | int w_send_read_req(struct drbd_work *w, int cancel) |
b411b363 PR |
1311 | { |
1312 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 1313 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 1314 | int err; |
b411b363 PR |
1315 | |
1316 | if (unlikely(cancel)) { | |
8554df1c | 1317 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1318 | return 0; |
b411b363 PR |
1319 | } |
1320 | ||
99920dc5 | 1321 | err = drbd_send_drequest(mdev, P_DATA_REQUEST, req->i.sector, req->i.size, |
6c1005e7 | 1322 | (unsigned long)req); |
b411b363 | 1323 | |
99920dc5 | 1324 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1325 | |
99920dc5 | 1326 | return err; |
b411b363 PR |
1327 | } |
1328 | ||
99920dc5 | 1329 | int w_restart_disk_io(struct drbd_work *w, int cancel) |
265be2d0 PR |
1330 | { |
1331 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 1332 | struct drbd_conf *mdev = w->mdev; |
265be2d0 | 1333 | |
0778286a | 1334 | if (bio_data_dir(req->master_bio) == WRITE && req->rq_state & RQ_IN_ACT_LOG) |
181286ad | 1335 | drbd_al_begin_io(mdev, &req->i); |
265be2d0 PR |
1336 | |
1337 | drbd_req_make_private_bio(req, req->master_bio); | |
1338 | req->private_bio->bi_bdev = mdev->ldev->backing_bdev; | |
1339 | generic_make_request(req->private_bio); | |
1340 | ||
99920dc5 | 1341 | return 0; |
265be2d0 PR |
1342 | } |
1343 | ||
b411b363 PR |
1344 | static int _drbd_may_sync_now(struct drbd_conf *mdev) |
1345 | { | |
1346 | struct drbd_conf *odev = mdev; | |
95f8efd0 | 1347 | int resync_after; |
b411b363 PR |
1348 | |
1349 | while (1) { | |
438c8374 PR |
1350 | if (!odev->ldev) |
1351 | return 1; | |
daeda1cc | 1352 | rcu_read_lock(); |
95f8efd0 | 1353 | resync_after = rcu_dereference(odev->ldev->disk_conf)->resync_after; |
daeda1cc | 1354 | rcu_read_unlock(); |
95f8efd0 | 1355 | if (resync_after == -1) |
b411b363 | 1356 | return 1; |
95f8efd0 | 1357 | odev = minor_to_mdev(resync_after); |
841ce241 AG |
1358 | if (!expect(odev)) |
1359 | return 1; | |
b411b363 PR |
1360 | if ((odev->state.conn >= C_SYNC_SOURCE && |
1361 | odev->state.conn <= C_PAUSED_SYNC_T) || | |
1362 | odev->state.aftr_isp || odev->state.peer_isp || | |
1363 | odev->state.user_isp) | |
1364 | return 0; | |
1365 | } | |
1366 | } | |
1367 | ||
1368 | /** | |
1369 | * _drbd_pause_after() - Pause resync on all devices that may not resync now | |
1370 | * @mdev: DRBD device. | |
1371 | * | |
1372 | * Called from process context only (admin command and after_state_ch). | |
1373 | */ | |
1374 | static int _drbd_pause_after(struct drbd_conf *mdev) | |
1375 | { | |
1376 | struct drbd_conf *odev; | |
1377 | int i, rv = 0; | |
1378 | ||
695d08fa | 1379 | rcu_read_lock(); |
81a5d60e | 1380 | idr_for_each_entry(&minors, odev, i) { |
b411b363 PR |
1381 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1382 | continue; | |
1383 | if (!_drbd_may_sync_now(odev)) | |
1384 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 1), CS_HARD, NULL) | |
1385 | != SS_NOTHING_TO_DO); | |
1386 | } | |
695d08fa | 1387 | rcu_read_unlock(); |
b411b363 PR |
1388 | |
1389 | return rv; | |
1390 | } | |
1391 | ||
1392 | /** | |
1393 | * _drbd_resume_next() - Resume resync on all devices that may resync now | |
1394 | * @mdev: DRBD device. | |
1395 | * | |
1396 | * Called from process context only (admin command and worker). | |
1397 | */ | |
1398 | static int _drbd_resume_next(struct drbd_conf *mdev) | |
1399 | { | |
1400 | struct drbd_conf *odev; | |
1401 | int i, rv = 0; | |
1402 | ||
695d08fa | 1403 | rcu_read_lock(); |
81a5d60e | 1404 | idr_for_each_entry(&minors, odev, i) { |
b411b363 PR |
1405 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1406 | continue; | |
1407 | if (odev->state.aftr_isp) { | |
1408 | if (_drbd_may_sync_now(odev)) | |
1409 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 0), | |
1410 | CS_HARD, NULL) | |
1411 | != SS_NOTHING_TO_DO) ; | |
1412 | } | |
1413 | } | |
695d08fa | 1414 | rcu_read_unlock(); |
b411b363 PR |
1415 | return rv; |
1416 | } | |
1417 | ||
1418 | void resume_next_sg(struct drbd_conf *mdev) | |
1419 | { | |
1420 | write_lock_irq(&global_state_lock); | |
1421 | _drbd_resume_next(mdev); | |
1422 | write_unlock_irq(&global_state_lock); | |
1423 | } | |
1424 | ||
1425 | void suspend_other_sg(struct drbd_conf *mdev) | |
1426 | { | |
1427 | write_lock_irq(&global_state_lock); | |
1428 | _drbd_pause_after(mdev); | |
1429 | write_unlock_irq(&global_state_lock); | |
1430 | } | |
1431 | ||
dc97b708 | 1432 | /* caller must hold global_state_lock */ |
95f8efd0 | 1433 | enum drbd_ret_code drbd_resync_after_valid(struct drbd_conf *mdev, int o_minor) |
b411b363 PR |
1434 | { |
1435 | struct drbd_conf *odev; | |
95f8efd0 | 1436 | int resync_after; |
b411b363 PR |
1437 | |
1438 | if (o_minor == -1) | |
1439 | return NO_ERROR; | |
1440 | if (o_minor < -1 || minor_to_mdev(o_minor) == NULL) | |
95f8efd0 | 1441 | return ERR_RESYNC_AFTER; |
b411b363 PR |
1442 | |
1443 | /* check for loops */ | |
1444 | odev = minor_to_mdev(o_minor); | |
1445 | while (1) { | |
1446 | if (odev == mdev) | |
95f8efd0 | 1447 | return ERR_RESYNC_AFTER_CYCLE; |
b411b363 | 1448 | |
daeda1cc | 1449 | rcu_read_lock(); |
95f8efd0 | 1450 | resync_after = rcu_dereference(odev->ldev->disk_conf)->resync_after; |
daeda1cc | 1451 | rcu_read_unlock(); |
b411b363 | 1452 | /* dependency chain ends here, no cycles. */ |
95f8efd0 | 1453 | if (resync_after == -1) |
b411b363 PR |
1454 | return NO_ERROR; |
1455 | ||
1456 | /* follow the dependency chain */ | |
95f8efd0 | 1457 | odev = minor_to_mdev(resync_after); |
b411b363 PR |
1458 | } |
1459 | } | |
1460 | ||
dc97b708 | 1461 | /* caller must hold global_state_lock */ |
95f8efd0 | 1462 | void drbd_resync_after_changed(struct drbd_conf *mdev) |
b411b363 PR |
1463 | { |
1464 | int changes; | |
b411b363 | 1465 | |
dc97b708 PR |
1466 | do { |
1467 | changes = _drbd_pause_after(mdev); | |
1468 | changes |= _drbd_resume_next(mdev); | |
1469 | } while (changes); | |
b411b363 PR |
1470 | } |
1471 | ||
9bd28d3c LE |
1472 | void drbd_rs_controller_reset(struct drbd_conf *mdev) |
1473 | { | |
813472ce PR |
1474 | struct fifo_buffer *plan; |
1475 | ||
9bd28d3c LE |
1476 | atomic_set(&mdev->rs_sect_in, 0); |
1477 | atomic_set(&mdev->rs_sect_ev, 0); | |
1478 | mdev->rs_in_flight = 0; | |
813472ce PR |
1479 | |
1480 | /* Updating the RCU protected object in place is necessary since | |
1481 | this function gets called from atomic context. | |
1482 | It is valid since all other updates also lead to an completely | |
1483 | empty fifo */ | |
1484 | rcu_read_lock(); | |
1485 | plan = rcu_dereference(mdev->rs_plan_s); | |
1486 | plan->total = 0; | |
1487 | fifo_set(plan, 0); | |
1488 | rcu_read_unlock(); | |
9bd28d3c LE |
1489 | } |
1490 | ||
1f04af33 PR |
1491 | void start_resync_timer_fn(unsigned long data) |
1492 | { | |
1493 | struct drbd_conf *mdev = (struct drbd_conf *) data; | |
1494 | ||
1495 | drbd_queue_work(&mdev->tconn->data.work, &mdev->start_resync_work); | |
1496 | } | |
1497 | ||
99920dc5 | 1498 | int w_start_resync(struct drbd_work *w, int cancel) |
1f04af33 | 1499 | { |
00d56944 PR |
1500 | struct drbd_conf *mdev = w->mdev; |
1501 | ||
1f04af33 PR |
1502 | if (atomic_read(&mdev->unacked_cnt) || atomic_read(&mdev->rs_pending_cnt)) { |
1503 | dev_warn(DEV, "w_start_resync later...\n"); | |
1504 | mdev->start_resync_timer.expires = jiffies + HZ/10; | |
1505 | add_timer(&mdev->start_resync_timer); | |
99920dc5 | 1506 | return 0; |
1f04af33 PR |
1507 | } |
1508 | ||
1509 | drbd_start_resync(mdev, C_SYNC_SOURCE); | |
1510 | clear_bit(AHEAD_TO_SYNC_SOURCE, &mdev->current_epoch->flags); | |
99920dc5 | 1511 | return 0; |
1f04af33 PR |
1512 | } |
1513 | ||
b411b363 PR |
1514 | /** |
1515 | * drbd_start_resync() - Start the resync process | |
1516 | * @mdev: DRBD device. | |
1517 | * @side: Either C_SYNC_SOURCE or C_SYNC_TARGET | |
1518 | * | |
1519 | * This function might bring you directly into one of the | |
1520 | * C_PAUSED_SYNC_* states. | |
1521 | */ | |
1522 | void drbd_start_resync(struct drbd_conf *mdev, enum drbd_conns side) | |
1523 | { | |
1524 | union drbd_state ns; | |
1525 | int r; | |
1526 | ||
c4752ef1 | 1527 | if (mdev->state.conn >= C_SYNC_SOURCE && mdev->state.conn < C_AHEAD) { |
b411b363 PR |
1528 | dev_err(DEV, "Resync already running!\n"); |
1529 | return; | |
1530 | } | |
1531 | ||
59817f4f PR |
1532 | if (mdev->state.conn < C_AHEAD) { |
1533 | /* In case a previous resync run was aborted by an IO error/detach on the peer. */ | |
1534 | drbd_rs_cancel_all(mdev); | |
1535 | /* This should be done when we abort the resync. We definitely do not | |
1536 | want to have this for connections going back and forth between | |
1537 | Ahead/Behind and SyncSource/SyncTarget */ | |
1538 | } | |
b411b363 | 1539 | |
e64a3294 PR |
1540 | if (!test_bit(B_RS_H_DONE, &mdev->flags)) { |
1541 | if (side == C_SYNC_TARGET) { | |
1542 | /* Since application IO was locked out during C_WF_BITMAP_T and | |
1543 | C_WF_SYNC_UUID we are still unmodified. Before going to C_SYNC_TARGET | |
1544 | we check that we might make the data inconsistent. */ | |
1545 | r = drbd_khelper(mdev, "before-resync-target"); | |
1546 | r = (r >> 8) & 0xff; | |
1547 | if (r > 0) { | |
1548 | dev_info(DEV, "before-resync-target handler returned %d, " | |
09b9e797 | 1549 | "dropping connection.\n", r); |
38fa9988 | 1550 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
09b9e797 PR |
1551 | return; |
1552 | } | |
e64a3294 PR |
1553 | } else /* C_SYNC_SOURCE */ { |
1554 | r = drbd_khelper(mdev, "before-resync-source"); | |
1555 | r = (r >> 8) & 0xff; | |
1556 | if (r > 0) { | |
1557 | if (r == 3) { | |
1558 | dev_info(DEV, "before-resync-source handler returned %d, " | |
1559 | "ignoring. Old userland tools?", r); | |
1560 | } else { | |
1561 | dev_info(DEV, "before-resync-source handler returned %d, " | |
1562 | "dropping connection.\n", r); | |
38fa9988 | 1563 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
e64a3294 PR |
1564 | return; |
1565 | } | |
1566 | } | |
09b9e797 | 1567 | } |
b411b363 PR |
1568 | } |
1569 | ||
e64a3294 | 1570 | if (current == mdev->tconn->worker.task) { |
dad20554 | 1571 | /* The worker should not sleep waiting for state_mutex, |
e64a3294 | 1572 | that can take long */ |
8410da8f | 1573 | if (!mutex_trylock(mdev->state_mutex)) { |
e64a3294 PR |
1574 | set_bit(B_RS_H_DONE, &mdev->flags); |
1575 | mdev->start_resync_timer.expires = jiffies + HZ/5; | |
1576 | add_timer(&mdev->start_resync_timer); | |
1577 | return; | |
1578 | } | |
1579 | } else { | |
8410da8f | 1580 | mutex_lock(mdev->state_mutex); |
e64a3294 PR |
1581 | } |
1582 | clear_bit(B_RS_H_DONE, &mdev->flags); | |
b411b363 PR |
1583 | |
1584 | if (!get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
8410da8f | 1585 | mutex_unlock(mdev->state_mutex); |
b411b363 PR |
1586 | return; |
1587 | } | |
1588 | ||
b411b363 | 1589 | write_lock_irq(&global_state_lock); |
78bae59b | 1590 | ns = drbd_read_state(mdev); |
b411b363 PR |
1591 | |
1592 | ns.aftr_isp = !_drbd_may_sync_now(mdev); | |
1593 | ||
1594 | ns.conn = side; | |
1595 | ||
1596 | if (side == C_SYNC_TARGET) | |
1597 | ns.disk = D_INCONSISTENT; | |
1598 | else /* side == C_SYNC_SOURCE */ | |
1599 | ns.pdsk = D_INCONSISTENT; | |
1600 | ||
1601 | r = __drbd_set_state(mdev, ns, CS_VERBOSE, NULL); | |
78bae59b | 1602 | ns = drbd_read_state(mdev); |
b411b363 PR |
1603 | |
1604 | if (ns.conn < C_CONNECTED) | |
1605 | r = SS_UNKNOWN_ERROR; | |
1606 | ||
1607 | if (r == SS_SUCCESS) { | |
1d7734a0 LE |
1608 | unsigned long tw = drbd_bm_total_weight(mdev); |
1609 | unsigned long now = jiffies; | |
1610 | int i; | |
1611 | ||
b411b363 PR |
1612 | mdev->rs_failed = 0; |
1613 | mdev->rs_paused = 0; | |
b411b363 | 1614 | mdev->rs_same_csum = 0; |
0f0601f4 LE |
1615 | mdev->rs_last_events = 0; |
1616 | mdev->rs_last_sect_ev = 0; | |
1d7734a0 LE |
1617 | mdev->rs_total = tw; |
1618 | mdev->rs_start = now; | |
1619 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { | |
1620 | mdev->rs_mark_left[i] = tw; | |
1621 | mdev->rs_mark_time[i] = now; | |
1622 | } | |
b411b363 PR |
1623 | _drbd_pause_after(mdev); |
1624 | } | |
1625 | write_unlock_irq(&global_state_lock); | |
5a22db89 | 1626 | |
b411b363 PR |
1627 | if (r == SS_SUCCESS) { |
1628 | dev_info(DEV, "Began resync as %s (will sync %lu KB [%lu bits set]).\n", | |
1629 | drbd_conn_str(ns.conn), | |
1630 | (unsigned long) mdev->rs_total << (BM_BLOCK_SHIFT-10), | |
1631 | (unsigned long) mdev->rs_total); | |
6c922ed5 LE |
1632 | if (side == C_SYNC_TARGET) |
1633 | mdev->bm_resync_fo = 0; | |
1634 | ||
1635 | /* Since protocol 96, we must serialize drbd_gen_and_send_sync_uuid | |
1636 | * with w_send_oos, or the sync target will get confused as to | |
1637 | * how much bits to resync. We cannot do that always, because for an | |
1638 | * empty resync and protocol < 95, we need to do it here, as we call | |
1639 | * drbd_resync_finished from here in that case. | |
1640 | * We drbd_gen_and_send_sync_uuid here for protocol < 96, | |
1641 | * and from after_state_ch otherwise. */ | |
31890f4a | 1642 | if (side == C_SYNC_SOURCE && mdev->tconn->agreed_pro_version < 96) |
6c922ed5 | 1643 | drbd_gen_and_send_sync_uuid(mdev); |
b411b363 | 1644 | |
31890f4a | 1645 | if (mdev->tconn->agreed_pro_version < 95 && mdev->rs_total == 0) { |
af85e8e8 LE |
1646 | /* This still has a race (about when exactly the peers |
1647 | * detect connection loss) that can lead to a full sync | |
1648 | * on next handshake. In 8.3.9 we fixed this with explicit | |
1649 | * resync-finished notifications, but the fix | |
1650 | * introduces a protocol change. Sleeping for some | |
1651 | * time longer than the ping interval + timeout on the | |
1652 | * SyncSource, to give the SyncTarget the chance to | |
1653 | * detect connection loss, then waiting for a ping | |
1654 | * response (implicit in drbd_resync_finished) reduces | |
1655 | * the race considerably, but does not solve it. */ | |
44ed167d PR |
1656 | if (side == C_SYNC_SOURCE) { |
1657 | struct net_conf *nc; | |
1658 | int timeo; | |
1659 | ||
1660 | rcu_read_lock(); | |
1661 | nc = rcu_dereference(mdev->tconn->net_conf); | |
1662 | timeo = nc->ping_int * HZ + nc->ping_timeo * HZ / 9; | |
1663 | rcu_read_unlock(); | |
1664 | schedule_timeout_interruptible(timeo); | |
1665 | } | |
b411b363 | 1666 | drbd_resync_finished(mdev); |
b411b363 PR |
1667 | } |
1668 | ||
9bd28d3c | 1669 | drbd_rs_controller_reset(mdev); |
b411b363 PR |
1670 | /* ns.conn may already be != mdev->state.conn, |
1671 | * we may have been paused in between, or become paused until | |
1672 | * the timer triggers. | |
1673 | * No matter, that is handled in resync_timer_fn() */ | |
1674 | if (ns.conn == C_SYNC_TARGET) | |
1675 | mod_timer(&mdev->resync_timer, jiffies); | |
1676 | ||
1677 | drbd_md_sync(mdev); | |
1678 | } | |
5a22db89 | 1679 | put_ldev(mdev); |
8410da8f | 1680 | mutex_unlock(mdev->state_mutex); |
b411b363 PR |
1681 | } |
1682 | ||
1683 | int drbd_worker(struct drbd_thread *thi) | |
1684 | { | |
392c8801 | 1685 | struct drbd_tconn *tconn = thi->tconn; |
b411b363 | 1686 | struct drbd_work *w = NULL; |
0e29d163 | 1687 | struct drbd_conf *mdev; |
44ed167d | 1688 | struct net_conf *nc; |
b411b363 | 1689 | LIST_HEAD(work_list); |
f399002e | 1690 | int vnr, intr = 0; |
44ed167d | 1691 | int cork; |
b411b363 | 1692 | |
e77a0a5c | 1693 | while (get_t_state(thi) == RUNNING) { |
80822284 | 1694 | drbd_thread_current_set_cpu(thi); |
b411b363 | 1695 | |
19393e10 PR |
1696 | if (down_trylock(&tconn->data.work.s)) { |
1697 | mutex_lock(&tconn->data.mutex); | |
44ed167d PR |
1698 | |
1699 | rcu_read_lock(); | |
1700 | nc = rcu_dereference(tconn->net_conf); | |
bb77d34e | 1701 | cork = nc ? nc->tcp_cork : 0; |
44ed167d PR |
1702 | rcu_read_unlock(); |
1703 | ||
1704 | if (tconn->data.socket && cork) | |
19393e10 PR |
1705 | drbd_tcp_uncork(tconn->data.socket); |
1706 | mutex_unlock(&tconn->data.mutex); | |
b411b363 | 1707 | |
19393e10 | 1708 | intr = down_interruptible(&tconn->data.work.s); |
b411b363 | 1709 | |
19393e10 | 1710 | mutex_lock(&tconn->data.mutex); |
44ed167d | 1711 | if (tconn->data.socket && cork) |
19393e10 PR |
1712 | drbd_tcp_cork(tconn->data.socket); |
1713 | mutex_unlock(&tconn->data.mutex); | |
b411b363 PR |
1714 | } |
1715 | ||
1716 | if (intr) { | |
b411b363 | 1717 | flush_signals(current); |
19393e10 PR |
1718 | if (get_t_state(thi) == RUNNING) { |
1719 | conn_warn(tconn, "Worker got an unexpected signal\n"); | |
b411b363 | 1720 | continue; |
19393e10 | 1721 | } |
b411b363 PR |
1722 | break; |
1723 | } | |
1724 | ||
e77a0a5c | 1725 | if (get_t_state(thi) != RUNNING) |
b411b363 PR |
1726 | break; |
1727 | /* With this break, we have done a down() but not consumed | |
1728 | the entry from the list. The cleanup code takes care of | |
1729 | this... */ | |
1730 | ||
1731 | w = NULL; | |
19393e10 PR |
1732 | spin_lock_irq(&tconn->data.work.q_lock); |
1733 | if (list_empty(&tconn->data.work.q)) { | |
b411b363 PR |
1734 | /* something terribly wrong in our logic. |
1735 | * we were able to down() the semaphore, | |
1736 | * but the list is empty... doh. | |
1737 | * | |
1738 | * what is the best thing to do now? | |
1739 | * try again from scratch, restarting the receiver, | |
1740 | * asender, whatnot? could break even more ugly, | |
1741 | * e.g. when we are primary, but no good local data. | |
1742 | * | |
1743 | * I'll try to get away just starting over this loop. | |
1744 | */ | |
19393e10 PR |
1745 | conn_warn(tconn, "Work list unexpectedly empty\n"); |
1746 | spin_unlock_irq(&tconn->data.work.q_lock); | |
b411b363 PR |
1747 | continue; |
1748 | } | |
19393e10 | 1749 | w = list_entry(tconn->data.work.q.next, struct drbd_work, list); |
b411b363 | 1750 | list_del_init(&w->list); |
19393e10 | 1751 | spin_unlock_irq(&tconn->data.work.q_lock); |
b411b363 | 1752 | |
99920dc5 | 1753 | if (w->cb(w, tconn->cstate < C_WF_REPORT_PARAMS)) { |
b411b363 | 1754 | /* dev_warn(DEV, "worker: a callback failed! \n"); */ |
bbeb641c PR |
1755 | if (tconn->cstate >= C_WF_REPORT_PARAMS) |
1756 | conn_request_state(tconn, NS(conn, C_NETWORK_FAILURE), CS_HARD); | |
b411b363 PR |
1757 | } |
1758 | } | |
b411b363 | 1759 | |
19393e10 PR |
1760 | spin_lock_irq(&tconn->data.work.q_lock); |
1761 | while (!list_empty(&tconn->data.work.q)) { | |
1762 | list_splice_init(&tconn->data.work.q, &work_list); | |
1763 | spin_unlock_irq(&tconn->data.work.q_lock); | |
b411b363 PR |
1764 | |
1765 | while (!list_empty(&work_list)) { | |
1766 | w = list_entry(work_list.next, struct drbd_work, list); | |
1767 | list_del_init(&w->list); | |
00d56944 | 1768 | w->cb(w, 1); |
b411b363 PR |
1769 | } |
1770 | ||
19393e10 | 1771 | spin_lock_irq(&tconn->data.work.q_lock); |
b411b363 | 1772 | } |
19393e10 | 1773 | sema_init(&tconn->data.work.s, 0); |
b411b363 PR |
1774 | /* DANGEROUS race: if someone did queue his work within the spinlock, |
1775 | * but up() ed outside the spinlock, we could get an up() on the | |
1776 | * semaphore without corresponding list entry. | |
1777 | * So don't do that. | |
1778 | */ | |
19393e10 | 1779 | spin_unlock_irq(&tconn->data.work.q_lock); |
b411b363 | 1780 | |
c141ebda | 1781 | rcu_read_lock(); |
f399002e | 1782 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
0e29d163 | 1783 | D_ASSERT(mdev->state.disk == D_DISKLESS && mdev->state.conn == C_STANDALONE); |
c141ebda PR |
1784 | kref_get(&mdev->kref); |
1785 | rcu_read_unlock(); | |
0e29d163 | 1786 | drbd_mdev_cleanup(mdev); |
c141ebda PR |
1787 | kref_put(&mdev->kref, &drbd_minor_destroy); |
1788 | rcu_read_lock(); | |
0e29d163 | 1789 | } |
c141ebda | 1790 | rcu_read_unlock(); |
b411b363 PR |
1791 | |
1792 | return 0; | |
1793 | } |