]>
Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_worker.c | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | drbd is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation; either version 2, or (at your option) | |
13 | any later version. | |
14 | ||
15 | drbd is distributed in the hope that it will be useful, | |
16 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | GNU General Public License for more details. | |
19 | ||
20 | You should have received a copy of the GNU General Public License | |
21 | along with drbd; see the file COPYING. If not, write to | |
22 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | ||
84b8c06b | 24 | */ |
b411b363 | 25 | |
b411b363 | 26 | #include <linux/module.h> |
b411b363 PR |
27 | #include <linux/drbd.h> |
28 | #include <linux/sched.h> | |
b411b363 PR |
29 | #include <linux/wait.h> |
30 | #include <linux/mm.h> | |
31 | #include <linux/memcontrol.h> | |
32 | #include <linux/mm_inline.h> | |
33 | #include <linux/slab.h> | |
34 | #include <linux/random.h> | |
b411b363 PR |
35 | #include <linux/string.h> |
36 | #include <linux/scatterlist.h> | |
37 | ||
38 | #include "drbd_int.h" | |
a3603a6e | 39 | #include "drbd_protocol.h" |
b411b363 | 40 | #include "drbd_req.h" |
b411b363 | 41 | |
84b8c06b | 42 | static int w_make_ov_request(struct drbd_work *, int); |
b411b363 PR |
43 | |
44 | ||
c5a91619 AG |
45 | /* endio handlers: |
46 | * drbd_md_io_complete (defined here) | |
fcefa62e AG |
47 | * drbd_request_endio (defined here) |
48 | * drbd_peer_request_endio (defined here) | |
c5a91619 AG |
49 | * bm_async_io_complete (defined in drbd_bitmap.c) |
50 | * | |
b411b363 PR |
51 | * For all these callbacks, note the following: |
52 | * The callbacks will be called in irq context by the IDE drivers, | |
53 | * and in Softirqs/Tasklets/BH context by the SCSI drivers. | |
54 | * Try to get the locking right :) | |
55 | * | |
56 | */ | |
57 | ||
58 | ||
59 | /* About the global_state_lock | |
60 | Each state transition on an device holds a read lock. In case we have | |
95f8efd0 | 61 | to evaluate the resync after dependencies, we grab a write lock, because |
b411b363 PR |
62 | we need stable states on all devices for that. */ |
63 | rwlock_t global_state_lock; | |
64 | ||
65 | /* used for synchronous meta data and bitmap IO | |
66 | * submitted by drbd_md_sync_page_io() | |
67 | */ | |
68 | void drbd_md_io_complete(struct bio *bio, int error) | |
69 | { | |
70 | struct drbd_md_io *md_io; | |
b30ab791 | 71 | struct drbd_device *device; |
b411b363 PR |
72 | |
73 | md_io = (struct drbd_md_io *)bio->bi_private; | |
b30ab791 | 74 | device = container_of(md_io, struct drbd_device, md_io); |
cdfda633 | 75 | |
b411b363 PR |
76 | md_io->error = error; |
77 | ||
0cfac5dd PR |
78 | /* We grabbed an extra reference in _drbd_md_sync_page_io() to be able |
79 | * to timeout on the lower level device, and eventually detach from it. | |
80 | * If this io completion runs after that timeout expired, this | |
81 | * drbd_md_put_buffer() may allow us to finally try and re-attach. | |
82 | * During normal operation, this only puts that extra reference | |
83 | * down to 1 again. | |
84 | * Make sure we first drop the reference, and only then signal | |
85 | * completion, or we may (in drbd_al_read_log()) cycle so fast into the | |
86 | * next drbd_md_sync_page_io(), that we trigger the | |
b30ab791 | 87 | * ASSERT(atomic_read(&device->md_io_in_use) == 1) there. |
0cfac5dd | 88 | */ |
b30ab791 | 89 | drbd_md_put_buffer(device); |
cdfda633 | 90 | md_io->done = 1; |
b30ab791 | 91 | wake_up(&device->misc_wait); |
cdfda633 | 92 | bio_put(bio); |
b30ab791 AG |
93 | if (device->ldev) /* special case: drbd_md_read() during drbd_adm_attach() */ |
94 | put_ldev(device); | |
b411b363 PR |
95 | } |
96 | ||
97 | /* reads on behalf of the partner, | |
98 | * "submitted" by the receiver | |
99 | */ | |
a186e478 | 100 | static void drbd_endio_read_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
101 | { |
102 | unsigned long flags = 0; | |
84b8c06b | 103 | struct drbd_device *device = peer_req->dw.device; |
b411b363 | 104 | |
0500813f | 105 | spin_lock_irqsave(&device->resource->req_lock, flags); |
b30ab791 | 106 | device->read_cnt += peer_req->i.size >> 9; |
84b8c06b | 107 | list_del(&peer_req->dw.w.list); |
b30ab791 AG |
108 | if (list_empty(&device->read_ee)) |
109 | wake_up(&device->ee_wait); | |
db830c46 | 110 | if (test_bit(__EE_WAS_ERROR, &peer_req->flags)) |
b30ab791 | 111 | __drbd_chk_io_error(device, DRBD_READ_ERROR); |
0500813f | 112 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b411b363 | 113 | |
84b8c06b AG |
114 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, |
115 | &peer_req->dw.w); | |
b30ab791 | 116 | put_ldev(device); |
b411b363 PR |
117 | } |
118 | ||
119 | /* writes on behalf of the partner, or resync writes, | |
45bb912b | 120 | * "submitted" by the receiver, final stage. */ |
db830c46 | 121 | static void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
122 | { |
123 | unsigned long flags = 0; | |
84b8c06b | 124 | struct drbd_device *device = peer_req->dw.device; |
181286ad | 125 | struct drbd_interval i; |
b411b363 | 126 | int do_wake; |
579b57ed | 127 | u64 block_id; |
b411b363 | 128 | int do_al_complete_io; |
b411b363 | 129 | |
db830c46 | 130 | /* after we moved peer_req to done_ee, |
b411b363 PR |
131 | * we may no longer access it, |
132 | * it may be freed/reused already! | |
133 | * (as soon as we release the req_lock) */ | |
181286ad | 134 | i = peer_req->i; |
db830c46 AG |
135 | do_al_complete_io = peer_req->flags & EE_CALL_AL_COMPLETE_IO; |
136 | block_id = peer_req->block_id; | |
b411b363 | 137 | |
0500813f | 138 | spin_lock_irqsave(&device->resource->req_lock, flags); |
b30ab791 | 139 | device->writ_cnt += peer_req->i.size >> 9; |
84b8c06b | 140 | list_move_tail(&peer_req->dw.w.list, &device->done_ee); |
b411b363 | 141 | |
bb3bfe96 | 142 | /* |
5e472264 | 143 | * Do not remove from the write_requests tree here: we did not send the |
bb3bfe96 AG |
144 | * Ack yet and did not wake possibly waiting conflicting requests. |
145 | * Removed from the tree from "drbd_process_done_ee" within the | |
84b8c06b | 146 | * appropriate dw.cb (e_end_block/e_end_resync_block) or from |
bb3bfe96 AG |
147 | * _drbd_clear_done_ee. |
148 | */ | |
b411b363 | 149 | |
b30ab791 | 150 | do_wake = list_empty(block_id == ID_SYNCER ? &device->sync_ee : &device->active_ee); |
b411b363 | 151 | |
db830c46 | 152 | if (test_bit(__EE_WAS_ERROR, &peer_req->flags)) |
b30ab791 | 153 | __drbd_chk_io_error(device, DRBD_WRITE_ERROR); |
0500813f | 154 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b411b363 | 155 | |
579b57ed | 156 | if (block_id == ID_SYNCER) |
b30ab791 | 157 | drbd_rs_complete_io(device, i.sector); |
b411b363 PR |
158 | |
159 | if (do_wake) | |
b30ab791 | 160 | wake_up(&device->ee_wait); |
b411b363 PR |
161 | |
162 | if (do_al_complete_io) | |
b30ab791 | 163 | drbd_al_complete_io(device, &i); |
b411b363 | 164 | |
a6b32bc3 | 165 | wake_asender(first_peer_device(device)->connection); |
b30ab791 | 166 | put_ldev(device); |
45bb912b | 167 | } |
b411b363 | 168 | |
45bb912b LE |
169 | /* writes on behalf of the partner, or resync writes, |
170 | * "submitted" by the receiver. | |
171 | */ | |
fcefa62e | 172 | void drbd_peer_request_endio(struct bio *bio, int error) |
45bb912b | 173 | { |
db830c46 | 174 | struct drbd_peer_request *peer_req = bio->bi_private; |
84b8c06b | 175 | struct drbd_device *device = peer_req->dw.device; |
45bb912b LE |
176 | int uptodate = bio_flagged(bio, BIO_UPTODATE); |
177 | int is_write = bio_data_dir(bio) == WRITE; | |
178 | ||
07194272 | 179 | if (error && __ratelimit(&drbd_ratelimit_state)) |
d0180171 | 180 | drbd_warn(device, "%s: error=%d s=%llus\n", |
45bb912b | 181 | is_write ? "write" : "read", error, |
db830c46 | 182 | (unsigned long long)peer_req->i.sector); |
45bb912b | 183 | if (!error && !uptodate) { |
07194272 | 184 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 185 | drbd_warn(device, "%s: setting error to -EIO s=%llus\n", |
07194272 | 186 | is_write ? "write" : "read", |
db830c46 | 187 | (unsigned long long)peer_req->i.sector); |
45bb912b LE |
188 | /* strange behavior of some lower level drivers... |
189 | * fail the request by clearing the uptodate flag, | |
190 | * but do not return any error?! */ | |
191 | error = -EIO; | |
192 | } | |
193 | ||
194 | if (error) | |
db830c46 | 195 | set_bit(__EE_WAS_ERROR, &peer_req->flags); |
45bb912b LE |
196 | |
197 | bio_put(bio); /* no need for the bio anymore */ | |
db830c46 | 198 | if (atomic_dec_and_test(&peer_req->pending_bios)) { |
45bb912b | 199 | if (is_write) |
db830c46 | 200 | drbd_endio_write_sec_final(peer_req); |
45bb912b | 201 | else |
db830c46 | 202 | drbd_endio_read_sec_final(peer_req); |
45bb912b | 203 | } |
b411b363 PR |
204 | } |
205 | ||
206 | /* read, readA or write requests on R_PRIMARY coming from drbd_make_request | |
207 | */ | |
fcefa62e | 208 | void drbd_request_endio(struct bio *bio, int error) |
b411b363 | 209 | { |
a115413d | 210 | unsigned long flags; |
b411b363 | 211 | struct drbd_request *req = bio->bi_private; |
84b8c06b | 212 | struct drbd_device *device = req->device; |
a115413d | 213 | struct bio_and_error m; |
b411b363 PR |
214 | enum drbd_req_event what; |
215 | int uptodate = bio_flagged(bio, BIO_UPTODATE); | |
216 | ||
b411b363 | 217 | if (!error && !uptodate) { |
d0180171 | 218 | drbd_warn(device, "p %s: setting error to -EIO\n", |
b411b363 PR |
219 | bio_data_dir(bio) == WRITE ? "write" : "read"); |
220 | /* strange behavior of some lower level drivers... | |
221 | * fail the request by clearing the uptodate flag, | |
222 | * but do not return any error?! */ | |
223 | error = -EIO; | |
224 | } | |
225 | ||
1b6dd252 PR |
226 | |
227 | /* If this request was aborted locally before, | |
228 | * but now was completed "successfully", | |
229 | * chances are that this caused arbitrary data corruption. | |
230 | * | |
231 | * "aborting" requests, or force-detaching the disk, is intended for | |
232 | * completely blocked/hung local backing devices which do no longer | |
233 | * complete requests at all, not even do error completions. In this | |
234 | * situation, usually a hard-reset and failover is the only way out. | |
235 | * | |
236 | * By "aborting", basically faking a local error-completion, | |
237 | * we allow for a more graceful swichover by cleanly migrating services. | |
238 | * Still the affected node has to be rebooted "soon". | |
239 | * | |
240 | * By completing these requests, we allow the upper layers to re-use | |
241 | * the associated data pages. | |
242 | * | |
243 | * If later the local backing device "recovers", and now DMAs some data | |
244 | * from disk into the original request pages, in the best case it will | |
245 | * just put random data into unused pages; but typically it will corrupt | |
246 | * meanwhile completely unrelated data, causing all sorts of damage. | |
247 | * | |
248 | * Which means delayed successful completion, | |
249 | * especially for READ requests, | |
250 | * is a reason to panic(). | |
251 | * | |
252 | * We assume that a delayed *error* completion is OK, | |
253 | * though we still will complain noisily about it. | |
254 | */ | |
255 | if (unlikely(req->rq_state & RQ_LOCAL_ABORTED)) { | |
256 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 257 | drbd_emerg(device, "delayed completion of aborted local request; disk-timeout may be too aggressive\n"); |
1b6dd252 PR |
258 | |
259 | if (!error) | |
260 | panic("possible random memory corruption caused by delayed completion of aborted local request\n"); | |
261 | } | |
262 | ||
b411b363 PR |
263 | /* to avoid recursion in __req_mod */ |
264 | if (unlikely(error)) { | |
265 | what = (bio_data_dir(bio) == WRITE) | |
8554df1c | 266 | ? WRITE_COMPLETED_WITH_ERROR |
5c3c7e64 | 267 | : (bio_rw(bio) == READ) |
8554df1c AG |
268 | ? READ_COMPLETED_WITH_ERROR |
269 | : READ_AHEAD_COMPLETED_WITH_ERROR; | |
b411b363 | 270 | } else |
8554df1c | 271 | what = COMPLETED_OK; |
b411b363 PR |
272 | |
273 | bio_put(req->private_bio); | |
274 | req->private_bio = ERR_PTR(error); | |
275 | ||
a115413d | 276 | /* not req_mod(), we need irqsave here! */ |
0500813f | 277 | spin_lock_irqsave(&device->resource->req_lock, flags); |
a115413d | 278 | __req_mod(req, what, &m); |
0500813f | 279 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b30ab791 | 280 | put_ldev(device); |
a115413d LE |
281 | |
282 | if (m.bio) | |
b30ab791 | 283 | complete_master_bio(device, &m); |
b411b363 PR |
284 | } |
285 | ||
79a3c8d3 | 286 | void drbd_csum_ee(struct crypto_hash *tfm, struct drbd_peer_request *peer_req, void *digest) |
45bb912b LE |
287 | { |
288 | struct hash_desc desc; | |
289 | struct scatterlist sg; | |
db830c46 | 290 | struct page *page = peer_req->pages; |
45bb912b LE |
291 | struct page *tmp; |
292 | unsigned len; | |
293 | ||
294 | desc.tfm = tfm; | |
295 | desc.flags = 0; | |
296 | ||
297 | sg_init_table(&sg, 1); | |
298 | crypto_hash_init(&desc); | |
299 | ||
300 | while ((tmp = page_chain_next(page))) { | |
301 | /* all but the last page will be fully used */ | |
302 | sg_set_page(&sg, page, PAGE_SIZE, 0); | |
303 | crypto_hash_update(&desc, &sg, sg.length); | |
304 | page = tmp; | |
305 | } | |
306 | /* and now the last, possibly only partially used page */ | |
db830c46 | 307 | len = peer_req->i.size & (PAGE_SIZE - 1); |
45bb912b LE |
308 | sg_set_page(&sg, page, len ?: PAGE_SIZE, 0); |
309 | crypto_hash_update(&desc, &sg, sg.length); | |
310 | crypto_hash_final(&desc, digest); | |
311 | } | |
312 | ||
79a3c8d3 | 313 | void drbd_csum_bio(struct crypto_hash *tfm, struct bio *bio, void *digest) |
b411b363 PR |
314 | { |
315 | struct hash_desc desc; | |
316 | struct scatterlist sg; | |
7988613b KO |
317 | struct bio_vec bvec; |
318 | struct bvec_iter iter; | |
b411b363 PR |
319 | |
320 | desc.tfm = tfm; | |
321 | desc.flags = 0; | |
322 | ||
323 | sg_init_table(&sg, 1); | |
324 | crypto_hash_init(&desc); | |
325 | ||
7988613b KO |
326 | bio_for_each_segment(bvec, bio, iter) { |
327 | sg_set_page(&sg, bvec.bv_page, bvec.bv_len, bvec.bv_offset); | |
b411b363 PR |
328 | crypto_hash_update(&desc, &sg, sg.length); |
329 | } | |
330 | crypto_hash_final(&desc, digest); | |
331 | } | |
332 | ||
9676c760 | 333 | /* MAYBE merge common code with w_e_end_ov_req */ |
99920dc5 | 334 | static int w_e_send_csum(struct drbd_work *w, int cancel) |
b411b363 | 335 | { |
84b8c06b AG |
336 | struct drbd_device_work *dw = device_work(w); |
337 | struct drbd_peer_request *peer_req = container_of(dw, struct drbd_peer_request, dw); | |
338 | struct drbd_device *device = dw->device; | |
b411b363 PR |
339 | int digest_size; |
340 | void *digest; | |
99920dc5 | 341 | int err = 0; |
b411b363 | 342 | |
53ea4331 LE |
343 | if (unlikely(cancel)) |
344 | goto out; | |
b411b363 | 345 | |
9676c760 | 346 | if (unlikely((peer_req->flags & EE_WAS_ERROR) != 0)) |
53ea4331 | 347 | goto out; |
b411b363 | 348 | |
a6b32bc3 | 349 | digest_size = crypto_hash_digestsize(first_peer_device(device)->connection->csums_tfm); |
53ea4331 LE |
350 | digest = kmalloc(digest_size, GFP_NOIO); |
351 | if (digest) { | |
db830c46 AG |
352 | sector_t sector = peer_req->i.sector; |
353 | unsigned int size = peer_req->i.size; | |
79a3c8d3 | 354 | drbd_csum_ee(first_peer_device(device)->connection->csums_tfm, peer_req, digest); |
9676c760 | 355 | /* Free peer_req and pages before send. |
53ea4331 LE |
356 | * In case we block on congestion, we could otherwise run into |
357 | * some distributed deadlock, if the other side blocks on | |
358 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 359 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 360 | drbd_free_peer_req(device, peer_req); |
db830c46 | 361 | peer_req = NULL; |
b30ab791 | 362 | inc_rs_pending(device); |
69a22773 | 363 | err = drbd_send_drequest_csum(first_peer_device(device), sector, size, |
db1b0b72 AG |
364 | digest, digest_size, |
365 | P_CSUM_RS_REQUEST); | |
53ea4331 LE |
366 | kfree(digest); |
367 | } else { | |
d0180171 | 368 | drbd_err(device, "kmalloc() of digest failed.\n"); |
99920dc5 | 369 | err = -ENOMEM; |
53ea4331 | 370 | } |
b411b363 | 371 | |
53ea4331 | 372 | out: |
db830c46 | 373 | if (peer_req) |
b30ab791 | 374 | drbd_free_peer_req(device, peer_req); |
b411b363 | 375 | |
99920dc5 | 376 | if (unlikely(err)) |
d0180171 | 377 | drbd_err(device, "drbd_send_drequest(..., csum) failed\n"); |
99920dc5 | 378 | return err; |
b411b363 PR |
379 | } |
380 | ||
381 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
382 | ||
69a22773 | 383 | static int read_for_csum(struct drbd_peer_device *peer_device, sector_t sector, int size) |
b411b363 | 384 | { |
69a22773 | 385 | struct drbd_device *device = peer_device->device; |
db830c46 | 386 | struct drbd_peer_request *peer_req; |
b411b363 | 387 | |
b30ab791 | 388 | if (!get_ldev(device)) |
80a40e43 | 389 | return -EIO; |
b411b363 | 390 | |
b30ab791 | 391 | if (drbd_rs_should_slow_down(device, sector)) |
0f0601f4 LE |
392 | goto defer; |
393 | ||
b411b363 PR |
394 | /* GFP_TRY, because if there is no memory available right now, this may |
395 | * be rescheduled for later. It is "only" background resync, after all. */ | |
69a22773 | 396 | peer_req = drbd_alloc_peer_req(peer_device, ID_SYNCER /* unused */, sector, |
0db55363 | 397 | size, GFP_TRY); |
db830c46 | 398 | if (!peer_req) |
80a40e43 | 399 | goto defer; |
b411b363 | 400 | |
84b8c06b | 401 | peer_req->dw.w.cb = w_e_send_csum; |
0500813f | 402 | spin_lock_irq(&device->resource->req_lock); |
84b8c06b | 403 | list_add(&peer_req->dw.w.list, &device->read_ee); |
0500813f | 404 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 405 | |
b30ab791 AG |
406 | atomic_add(size >> 9, &device->rs_sect_ev); |
407 | if (drbd_submit_peer_request(device, peer_req, READ, DRBD_FAULT_RS_RD) == 0) | |
80a40e43 | 408 | return 0; |
b411b363 | 409 | |
10f6d992 LE |
410 | /* If it failed because of ENOMEM, retry should help. If it failed |
411 | * because bio_add_page failed (probably broken lower level driver), | |
412 | * retry may or may not help. | |
413 | * If it does not, you may need to force disconnect. */ | |
0500813f | 414 | spin_lock_irq(&device->resource->req_lock); |
84b8c06b | 415 | list_del(&peer_req->dw.w.list); |
0500813f | 416 | spin_unlock_irq(&device->resource->req_lock); |
22cc37a9 | 417 | |
b30ab791 | 418 | drbd_free_peer_req(device, peer_req); |
80a40e43 | 419 | defer: |
b30ab791 | 420 | put_ldev(device); |
80a40e43 | 421 | return -EAGAIN; |
b411b363 PR |
422 | } |
423 | ||
99920dc5 | 424 | int w_resync_timer(struct drbd_work *w, int cancel) |
b411b363 | 425 | { |
84b8c06b AG |
426 | struct drbd_device *device = |
427 | container_of(w, struct drbd_device, resync_work); | |
428 | ||
b30ab791 | 429 | switch (device->state.conn) { |
63106d3c | 430 | case C_VERIFY_S: |
00d56944 | 431 | w_make_ov_request(w, cancel); |
63106d3c PR |
432 | break; |
433 | case C_SYNC_TARGET: | |
00d56944 | 434 | w_make_resync_request(w, cancel); |
63106d3c | 435 | break; |
b411b363 PR |
436 | } |
437 | ||
99920dc5 | 438 | return 0; |
794abb75 PR |
439 | } |
440 | ||
441 | void resync_timer_fn(unsigned long data) | |
442 | { | |
b30ab791 | 443 | struct drbd_device *device = (struct drbd_device *) data; |
794abb75 | 444 | |
b30ab791 | 445 | if (list_empty(&device->resync_work.list)) |
84b8c06b AG |
446 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, |
447 | &device->resync_work); | |
b411b363 PR |
448 | } |
449 | ||
778f271d PR |
450 | static void fifo_set(struct fifo_buffer *fb, int value) |
451 | { | |
452 | int i; | |
453 | ||
454 | for (i = 0; i < fb->size; i++) | |
f10f2623 | 455 | fb->values[i] = value; |
778f271d PR |
456 | } |
457 | ||
458 | static int fifo_push(struct fifo_buffer *fb, int value) | |
459 | { | |
460 | int ov; | |
461 | ||
462 | ov = fb->values[fb->head_index]; | |
463 | fb->values[fb->head_index++] = value; | |
464 | ||
465 | if (fb->head_index >= fb->size) | |
466 | fb->head_index = 0; | |
467 | ||
468 | return ov; | |
469 | } | |
470 | ||
471 | static void fifo_add_val(struct fifo_buffer *fb, int value) | |
472 | { | |
473 | int i; | |
474 | ||
475 | for (i = 0; i < fb->size; i++) | |
476 | fb->values[i] += value; | |
477 | } | |
478 | ||
9958c857 PR |
479 | struct fifo_buffer *fifo_alloc(int fifo_size) |
480 | { | |
481 | struct fifo_buffer *fb; | |
482 | ||
8747d30a | 483 | fb = kzalloc(sizeof(struct fifo_buffer) + sizeof(int) * fifo_size, GFP_NOIO); |
9958c857 PR |
484 | if (!fb) |
485 | return NULL; | |
486 | ||
487 | fb->head_index = 0; | |
488 | fb->size = fifo_size; | |
489 | fb->total = 0; | |
490 | ||
491 | return fb; | |
492 | } | |
493 | ||
b30ab791 | 494 | static int drbd_rs_controller(struct drbd_device *device) |
778f271d | 495 | { |
daeda1cc | 496 | struct disk_conf *dc; |
778f271d PR |
497 | unsigned int sect_in; /* Number of sectors that came in since the last turn */ |
498 | unsigned int want; /* The number of sectors we want in the proxy */ | |
499 | int req_sect; /* Number of sectors to request in this turn */ | |
500 | int correction; /* Number of sectors more we need in the proxy*/ | |
501 | int cps; /* correction per invocation of drbd_rs_controller() */ | |
502 | int steps; /* Number of time steps to plan ahead */ | |
503 | int curr_corr; | |
504 | int max_sect; | |
813472ce | 505 | struct fifo_buffer *plan; |
778f271d | 506 | |
b30ab791 AG |
507 | sect_in = atomic_xchg(&device->rs_sect_in, 0); /* Number of sectors that came in */ |
508 | device->rs_in_flight -= sect_in; | |
778f271d | 509 | |
b30ab791 AG |
510 | dc = rcu_dereference(device->ldev->disk_conf); |
511 | plan = rcu_dereference(device->rs_plan_s); | |
778f271d | 512 | |
813472ce | 513 | steps = plan->size; /* (dc->c_plan_ahead * 10 * SLEEP_TIME) / HZ; */ |
778f271d | 514 | |
b30ab791 | 515 | if (device->rs_in_flight + sect_in == 0) { /* At start of resync */ |
daeda1cc | 516 | want = ((dc->resync_rate * 2 * SLEEP_TIME) / HZ) * steps; |
778f271d | 517 | } else { /* normal path */ |
daeda1cc PR |
518 | want = dc->c_fill_target ? dc->c_fill_target : |
519 | sect_in * dc->c_delay_target * HZ / (SLEEP_TIME * 10); | |
778f271d PR |
520 | } |
521 | ||
b30ab791 | 522 | correction = want - device->rs_in_flight - plan->total; |
778f271d PR |
523 | |
524 | /* Plan ahead */ | |
525 | cps = correction / steps; | |
813472ce PR |
526 | fifo_add_val(plan, cps); |
527 | plan->total += cps * steps; | |
778f271d PR |
528 | |
529 | /* What we do in this step */ | |
813472ce PR |
530 | curr_corr = fifo_push(plan, 0); |
531 | plan->total -= curr_corr; | |
778f271d PR |
532 | |
533 | req_sect = sect_in + curr_corr; | |
534 | if (req_sect < 0) | |
535 | req_sect = 0; | |
536 | ||
daeda1cc | 537 | max_sect = (dc->c_max_rate * 2 * SLEEP_TIME) / HZ; |
778f271d PR |
538 | if (req_sect > max_sect) |
539 | req_sect = max_sect; | |
540 | ||
541 | /* | |
d0180171 | 542 | drbd_warn(device, "si=%u if=%d wa=%u co=%d st=%d cps=%d pl=%d cc=%d rs=%d\n", |
b30ab791 AG |
543 | sect_in, device->rs_in_flight, want, correction, |
544 | steps, cps, device->rs_planed, curr_corr, req_sect); | |
778f271d PR |
545 | */ |
546 | ||
547 | return req_sect; | |
548 | } | |
549 | ||
b30ab791 | 550 | static int drbd_rs_number_requests(struct drbd_device *device) |
e65f440d LE |
551 | { |
552 | int number; | |
813472ce PR |
553 | |
554 | rcu_read_lock(); | |
b30ab791 AG |
555 | if (rcu_dereference(device->rs_plan_s)->size) { |
556 | number = drbd_rs_controller(device) >> (BM_BLOCK_SHIFT - 9); | |
557 | device->c_sync_rate = number * HZ * (BM_BLOCK_SIZE / 1024) / SLEEP_TIME; | |
e65f440d | 558 | } else { |
b30ab791 AG |
559 | device->c_sync_rate = rcu_dereference(device->ldev->disk_conf)->resync_rate; |
560 | number = SLEEP_TIME * device->c_sync_rate / ((BM_BLOCK_SIZE / 1024) * HZ); | |
e65f440d | 561 | } |
813472ce | 562 | rcu_read_unlock(); |
e65f440d | 563 | |
e65f440d LE |
564 | /* ignore the amount of pending requests, the resync controller should |
565 | * throttle down to incoming reply rate soon enough anyways. */ | |
566 | return number; | |
567 | } | |
568 | ||
99920dc5 | 569 | int w_make_resync_request(struct drbd_work *w, int cancel) |
b411b363 | 570 | { |
84b8c06b AG |
571 | struct drbd_device_work *dw = device_work(w); |
572 | struct drbd_device *device = dw->device; | |
b411b363 PR |
573 | unsigned long bit; |
574 | sector_t sector; | |
b30ab791 | 575 | const sector_t capacity = drbd_get_capacity(device->this_bdev); |
1816a2b4 | 576 | int max_bio_size; |
e65f440d | 577 | int number, rollback_i, size; |
b411b363 | 578 | int align, queued, sndbuf; |
0f0601f4 | 579 | int i = 0; |
b411b363 PR |
580 | |
581 | if (unlikely(cancel)) | |
99920dc5 | 582 | return 0; |
b411b363 | 583 | |
b30ab791 | 584 | if (device->rs_total == 0) { |
af85e8e8 | 585 | /* empty resync? */ |
b30ab791 | 586 | drbd_resync_finished(device); |
99920dc5 | 587 | return 0; |
af85e8e8 LE |
588 | } |
589 | ||
b30ab791 AG |
590 | if (!get_ldev(device)) { |
591 | /* Since we only need to access device->rsync a | |
592 | get_ldev_if_state(device,D_FAILED) would be sufficient, but | |
b411b363 PR |
593 | to continue resync with a broken disk makes no sense at |
594 | all */ | |
d0180171 | 595 | drbd_err(device, "Disk broke down during resync!\n"); |
99920dc5 | 596 | return 0; |
b411b363 PR |
597 | } |
598 | ||
b30ab791 AG |
599 | max_bio_size = queue_max_hw_sectors(device->rq_queue) << 9; |
600 | number = drbd_rs_number_requests(device); | |
e65f440d | 601 | if (number == 0) |
0f0601f4 | 602 | goto requeue; |
b411b363 | 603 | |
b411b363 PR |
604 | for (i = 0; i < number; i++) { |
605 | /* Stop generating RS requests, when half of the send buffer is filled */ | |
a6b32bc3 AG |
606 | mutex_lock(&first_peer_device(device)->connection->data.mutex); |
607 | if (first_peer_device(device)->connection->data.socket) { | |
608 | queued = first_peer_device(device)->connection->data.socket->sk->sk_wmem_queued; | |
609 | sndbuf = first_peer_device(device)->connection->data.socket->sk->sk_sndbuf; | |
b411b363 PR |
610 | } else { |
611 | queued = 1; | |
612 | sndbuf = 0; | |
613 | } | |
a6b32bc3 | 614 | mutex_unlock(&first_peer_device(device)->connection->data.mutex); |
b411b363 PR |
615 | if (queued > sndbuf / 2) |
616 | goto requeue; | |
617 | ||
618 | next_sector: | |
619 | size = BM_BLOCK_SIZE; | |
b30ab791 | 620 | bit = drbd_bm_find_next(device, device->bm_resync_fo); |
b411b363 | 621 | |
4b0715f0 | 622 | if (bit == DRBD_END_OF_BITMAP) { |
b30ab791 AG |
623 | device->bm_resync_fo = drbd_bm_bits(device); |
624 | put_ldev(device); | |
99920dc5 | 625 | return 0; |
b411b363 PR |
626 | } |
627 | ||
628 | sector = BM_BIT_TO_SECT(bit); | |
629 | ||
b30ab791 AG |
630 | if (drbd_rs_should_slow_down(device, sector) || |
631 | drbd_try_rs_begin_io(device, sector)) { | |
632 | device->bm_resync_fo = bit; | |
b411b363 PR |
633 | goto requeue; |
634 | } | |
b30ab791 | 635 | device->bm_resync_fo = bit + 1; |
b411b363 | 636 | |
b30ab791 AG |
637 | if (unlikely(drbd_bm_test_bit(device, bit) == 0)) { |
638 | drbd_rs_complete_io(device, sector); | |
b411b363 PR |
639 | goto next_sector; |
640 | } | |
641 | ||
1816a2b4 | 642 | #if DRBD_MAX_BIO_SIZE > BM_BLOCK_SIZE |
b411b363 PR |
643 | /* try to find some adjacent bits. |
644 | * we stop if we have already the maximum req size. | |
645 | * | |
646 | * Additionally always align bigger requests, in order to | |
647 | * be prepared for all stripe sizes of software RAIDs. | |
b411b363 PR |
648 | */ |
649 | align = 1; | |
d207450c | 650 | rollback_i = i; |
b411b363 | 651 | for (;;) { |
1816a2b4 | 652 | if (size + BM_BLOCK_SIZE > max_bio_size) |
b411b363 PR |
653 | break; |
654 | ||
655 | /* Be always aligned */ | |
656 | if (sector & ((1<<(align+3))-1)) | |
657 | break; | |
658 | ||
659 | /* do not cross extent boundaries */ | |
660 | if (((bit+1) & BM_BLOCKS_PER_BM_EXT_MASK) == 0) | |
661 | break; | |
662 | /* now, is it actually dirty, after all? | |
663 | * caution, drbd_bm_test_bit is tri-state for some | |
664 | * obscure reason; ( b == 0 ) would get the out-of-band | |
665 | * only accidentally right because of the "oddly sized" | |
666 | * adjustment below */ | |
b30ab791 | 667 | if (drbd_bm_test_bit(device, bit+1) != 1) |
b411b363 PR |
668 | break; |
669 | bit++; | |
670 | size += BM_BLOCK_SIZE; | |
671 | if ((BM_BLOCK_SIZE << align) <= size) | |
672 | align++; | |
673 | i++; | |
674 | } | |
675 | /* if we merged some, | |
676 | * reset the offset to start the next drbd_bm_find_next from */ | |
677 | if (size > BM_BLOCK_SIZE) | |
b30ab791 | 678 | device->bm_resync_fo = bit + 1; |
b411b363 PR |
679 | #endif |
680 | ||
681 | /* adjust very last sectors, in case we are oddly sized */ | |
682 | if (sector + (size>>9) > capacity) | |
683 | size = (capacity-sector)<<9; | |
a6b32bc3 AG |
684 | if (first_peer_device(device)->connection->agreed_pro_version >= 89 && |
685 | first_peer_device(device)->connection->csums_tfm) { | |
69a22773 | 686 | switch (read_for_csum(first_peer_device(device), sector, size)) { |
80a40e43 | 687 | case -EIO: /* Disk failure */ |
b30ab791 | 688 | put_ldev(device); |
99920dc5 | 689 | return -EIO; |
80a40e43 | 690 | case -EAGAIN: /* allocation failed, or ldev busy */ |
b30ab791 AG |
691 | drbd_rs_complete_io(device, sector); |
692 | device->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
d207450c | 693 | i = rollback_i; |
b411b363 | 694 | goto requeue; |
80a40e43 LE |
695 | case 0: |
696 | /* everything ok */ | |
697 | break; | |
698 | default: | |
699 | BUG(); | |
b411b363 PR |
700 | } |
701 | } else { | |
99920dc5 AG |
702 | int err; |
703 | ||
b30ab791 | 704 | inc_rs_pending(device); |
69a22773 | 705 | err = drbd_send_drequest(first_peer_device(device), P_RS_DATA_REQUEST, |
99920dc5 AG |
706 | sector, size, ID_SYNCER); |
707 | if (err) { | |
d0180171 | 708 | drbd_err(device, "drbd_send_drequest() failed, aborting...\n"); |
b30ab791 AG |
709 | dec_rs_pending(device); |
710 | put_ldev(device); | |
99920dc5 | 711 | return err; |
b411b363 PR |
712 | } |
713 | } | |
714 | } | |
715 | ||
b30ab791 | 716 | if (device->bm_resync_fo >= drbd_bm_bits(device)) { |
b411b363 PR |
717 | /* last syncer _request_ was sent, |
718 | * but the P_RS_DATA_REPLY not yet received. sync will end (and | |
719 | * next sync group will resume), as soon as we receive the last | |
720 | * resync data block, and the last bit is cleared. | |
721 | * until then resync "work" is "inactive" ... | |
722 | */ | |
b30ab791 | 723 | put_ldev(device); |
99920dc5 | 724 | return 0; |
b411b363 PR |
725 | } |
726 | ||
727 | requeue: | |
b30ab791 AG |
728 | device->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
729 | mod_timer(&device->resync_timer, jiffies + SLEEP_TIME); | |
730 | put_ldev(device); | |
99920dc5 | 731 | return 0; |
b411b363 PR |
732 | } |
733 | ||
00d56944 | 734 | static int w_make_ov_request(struct drbd_work *w, int cancel) |
b411b363 | 735 | { |
84b8c06b | 736 | struct drbd_device *device = device_work(w)->device; |
b411b363 PR |
737 | int number, i, size; |
738 | sector_t sector; | |
b30ab791 | 739 | const sector_t capacity = drbd_get_capacity(device->this_bdev); |
58ffa580 | 740 | bool stop_sector_reached = false; |
b411b363 PR |
741 | |
742 | if (unlikely(cancel)) | |
743 | return 1; | |
744 | ||
b30ab791 | 745 | number = drbd_rs_number_requests(device); |
b411b363 | 746 | |
b30ab791 | 747 | sector = device->ov_position; |
b411b363 | 748 | for (i = 0; i < number; i++) { |
58ffa580 | 749 | if (sector >= capacity) |
b411b363 | 750 | return 1; |
58ffa580 LE |
751 | |
752 | /* We check for "finished" only in the reply path: | |
753 | * w_e_end_ov_reply(). | |
754 | * We need to send at least one request out. */ | |
755 | stop_sector_reached = i > 0 | |
b30ab791 AG |
756 | && verify_can_do_stop_sector(device) |
757 | && sector >= device->ov_stop_sector; | |
58ffa580 LE |
758 | if (stop_sector_reached) |
759 | break; | |
b411b363 PR |
760 | |
761 | size = BM_BLOCK_SIZE; | |
762 | ||
b30ab791 AG |
763 | if (drbd_rs_should_slow_down(device, sector) || |
764 | drbd_try_rs_begin_io(device, sector)) { | |
765 | device->ov_position = sector; | |
b411b363 PR |
766 | goto requeue; |
767 | } | |
768 | ||
769 | if (sector + (size>>9) > capacity) | |
770 | size = (capacity-sector)<<9; | |
771 | ||
b30ab791 | 772 | inc_rs_pending(device); |
69a22773 | 773 | if (drbd_send_ov_request(first_peer_device(device), sector, size)) { |
b30ab791 | 774 | dec_rs_pending(device); |
b411b363 PR |
775 | return 0; |
776 | } | |
777 | sector += BM_SECT_PER_BIT; | |
778 | } | |
b30ab791 | 779 | device->ov_position = sector; |
b411b363 PR |
780 | |
781 | requeue: | |
b30ab791 | 782 | device->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
58ffa580 | 783 | if (i == 0 || !stop_sector_reached) |
b30ab791 | 784 | mod_timer(&device->resync_timer, jiffies + SLEEP_TIME); |
b411b363 PR |
785 | return 1; |
786 | } | |
787 | ||
99920dc5 | 788 | int w_ov_finished(struct drbd_work *w, int cancel) |
b411b363 | 789 | { |
84b8c06b AG |
790 | struct drbd_device_work *dw = |
791 | container_of(w, struct drbd_device_work, w); | |
792 | struct drbd_device *device = dw->device; | |
793 | kfree(dw); | |
b30ab791 AG |
794 | ov_out_of_sync_print(device); |
795 | drbd_resync_finished(device); | |
b411b363 | 796 | |
99920dc5 | 797 | return 0; |
b411b363 PR |
798 | } |
799 | ||
99920dc5 | 800 | static int w_resync_finished(struct drbd_work *w, int cancel) |
b411b363 | 801 | { |
84b8c06b AG |
802 | struct drbd_device_work *dw = |
803 | container_of(w, struct drbd_device_work, w); | |
804 | struct drbd_device *device = dw->device; | |
805 | kfree(dw); | |
b411b363 | 806 | |
b30ab791 | 807 | drbd_resync_finished(device); |
b411b363 | 808 | |
99920dc5 | 809 | return 0; |
b411b363 PR |
810 | } |
811 | ||
b30ab791 | 812 | static void ping_peer(struct drbd_device *device) |
af85e8e8 | 813 | { |
a6b32bc3 | 814 | struct drbd_connection *connection = first_peer_device(device)->connection; |
2a67d8b9 | 815 | |
bde89a9e AG |
816 | clear_bit(GOT_PING_ACK, &connection->flags); |
817 | request_ping(connection); | |
818 | wait_event(connection->ping_wait, | |
819 | test_bit(GOT_PING_ACK, &connection->flags) || device->state.conn < C_CONNECTED); | |
af85e8e8 LE |
820 | } |
821 | ||
b30ab791 | 822 | int drbd_resync_finished(struct drbd_device *device) |
b411b363 PR |
823 | { |
824 | unsigned long db, dt, dbdt; | |
825 | unsigned long n_oos; | |
826 | union drbd_state os, ns; | |
84b8c06b | 827 | struct drbd_device_work *dw; |
b411b363 | 828 | char *khelper_cmd = NULL; |
26525618 | 829 | int verify_done = 0; |
b411b363 PR |
830 | |
831 | /* Remove all elements from the resync LRU. Since future actions | |
832 | * might set bits in the (main) bitmap, then the entries in the | |
833 | * resync LRU would be wrong. */ | |
b30ab791 | 834 | if (drbd_rs_del_all(device)) { |
b411b363 PR |
835 | /* In case this is not possible now, most probably because |
836 | * there are P_RS_DATA_REPLY Packets lingering on the worker's | |
837 | * queue (or even the read operations for those packets | |
838 | * is not finished by now). Retry in 100ms. */ | |
839 | ||
20ee6390 | 840 | schedule_timeout_interruptible(HZ / 10); |
84b8c06b AG |
841 | dw = kmalloc(sizeof(struct drbd_device_work), GFP_ATOMIC); |
842 | if (dw) { | |
843 | dw->w.cb = w_resync_finished; | |
844 | dw->device = device; | |
845 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, | |
846 | &dw->w); | |
b411b363 PR |
847 | return 1; |
848 | } | |
84b8c06b | 849 | drbd_err(device, "Warn failed to drbd_rs_del_all() and to kmalloc(dw).\n"); |
b411b363 PR |
850 | } |
851 | ||
b30ab791 | 852 | dt = (jiffies - device->rs_start - device->rs_paused) / HZ; |
b411b363 PR |
853 | if (dt <= 0) |
854 | dt = 1; | |
84b8c06b | 855 | |
b30ab791 | 856 | db = device->rs_total; |
58ffa580 | 857 | /* adjust for verify start and stop sectors, respective reached position */ |
b30ab791 AG |
858 | if (device->state.conn == C_VERIFY_S || device->state.conn == C_VERIFY_T) |
859 | db -= device->ov_left; | |
58ffa580 | 860 | |
b411b363 | 861 | dbdt = Bit2KB(db/dt); |
b30ab791 | 862 | device->rs_paused /= HZ; |
b411b363 | 863 | |
b30ab791 | 864 | if (!get_ldev(device)) |
b411b363 PR |
865 | goto out; |
866 | ||
b30ab791 | 867 | ping_peer(device); |
af85e8e8 | 868 | |
0500813f | 869 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 870 | os = drbd_read_state(device); |
b411b363 | 871 | |
26525618 LE |
872 | verify_done = (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T); |
873 | ||
b411b363 PR |
874 | /* This protects us against multiple calls (that can happen in the presence |
875 | of application IO), and against connectivity loss just before we arrive here. */ | |
876 | if (os.conn <= C_CONNECTED) | |
877 | goto out_unlock; | |
878 | ||
879 | ns = os; | |
880 | ns.conn = C_CONNECTED; | |
881 | ||
d0180171 | 882 | drbd_info(device, "%s done (total %lu sec; paused %lu sec; %lu K/sec)\n", |
58ffa580 | 883 | verify_done ? "Online verify" : "Resync", |
b30ab791 | 884 | dt + device->rs_paused, device->rs_paused, dbdt); |
b411b363 | 885 | |
b30ab791 | 886 | n_oos = drbd_bm_total_weight(device); |
b411b363 PR |
887 | |
888 | if (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) { | |
889 | if (n_oos) { | |
d0180171 | 890 | drbd_alert(device, "Online verify found %lu %dk block out of sync!\n", |
b411b363 PR |
891 | n_oos, Bit2KB(1)); |
892 | khelper_cmd = "out-of-sync"; | |
893 | } | |
894 | } else { | |
0b0ba1ef | 895 | D_ASSERT(device, (n_oos - device->rs_failed) == 0); |
b411b363 PR |
896 | |
897 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) | |
898 | khelper_cmd = "after-resync-target"; | |
899 | ||
a6b32bc3 | 900 | if (first_peer_device(device)->connection->csums_tfm && device->rs_total) { |
b30ab791 AG |
901 | const unsigned long s = device->rs_same_csum; |
902 | const unsigned long t = device->rs_total; | |
b411b363 PR |
903 | const int ratio = |
904 | (t == 0) ? 0 : | |
905 | (t < 100000) ? ((s*100)/t) : (s/(t/100)); | |
d0180171 | 906 | drbd_info(device, "%u %% had equal checksums, eliminated: %luK; " |
b411b363 PR |
907 | "transferred %luK total %luK\n", |
908 | ratio, | |
b30ab791 AG |
909 | Bit2KB(device->rs_same_csum), |
910 | Bit2KB(device->rs_total - device->rs_same_csum), | |
911 | Bit2KB(device->rs_total)); | |
b411b363 PR |
912 | } |
913 | } | |
914 | ||
b30ab791 | 915 | if (device->rs_failed) { |
d0180171 | 916 | drbd_info(device, " %lu failed blocks\n", device->rs_failed); |
b411b363 PR |
917 | |
918 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
919 | ns.disk = D_INCONSISTENT; | |
920 | ns.pdsk = D_UP_TO_DATE; | |
921 | } else { | |
922 | ns.disk = D_UP_TO_DATE; | |
923 | ns.pdsk = D_INCONSISTENT; | |
924 | } | |
925 | } else { | |
926 | ns.disk = D_UP_TO_DATE; | |
927 | ns.pdsk = D_UP_TO_DATE; | |
928 | ||
929 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
b30ab791 | 930 | if (device->p_uuid) { |
b411b363 PR |
931 | int i; |
932 | for (i = UI_BITMAP ; i <= UI_HISTORY_END ; i++) | |
b30ab791 AG |
933 | _drbd_uuid_set(device, i, device->p_uuid[i]); |
934 | drbd_uuid_set(device, UI_BITMAP, device->ldev->md.uuid[UI_CURRENT]); | |
935 | _drbd_uuid_set(device, UI_CURRENT, device->p_uuid[UI_CURRENT]); | |
b411b363 | 936 | } else { |
d0180171 | 937 | drbd_err(device, "device->p_uuid is NULL! BUG\n"); |
b411b363 PR |
938 | } |
939 | } | |
940 | ||
62b0da3a LE |
941 | if (!(os.conn == C_VERIFY_S || os.conn == C_VERIFY_T)) { |
942 | /* for verify runs, we don't update uuids here, | |
943 | * so there would be nothing to report. */ | |
b30ab791 AG |
944 | drbd_uuid_set_bm(device, 0UL); |
945 | drbd_print_uuids(device, "updated UUIDs"); | |
946 | if (device->p_uuid) { | |
62b0da3a LE |
947 | /* Now the two UUID sets are equal, update what we |
948 | * know of the peer. */ | |
949 | int i; | |
950 | for (i = UI_CURRENT ; i <= UI_HISTORY_END ; i++) | |
b30ab791 | 951 | device->p_uuid[i] = device->ldev->md.uuid[i]; |
62b0da3a | 952 | } |
b411b363 PR |
953 | } |
954 | } | |
955 | ||
b30ab791 | 956 | _drbd_set_state(device, ns, CS_VERBOSE, NULL); |
b411b363 | 957 | out_unlock: |
0500813f | 958 | spin_unlock_irq(&device->resource->req_lock); |
b30ab791 | 959 | put_ldev(device); |
b411b363 | 960 | out: |
b30ab791 AG |
961 | device->rs_total = 0; |
962 | device->rs_failed = 0; | |
963 | device->rs_paused = 0; | |
58ffa580 LE |
964 | |
965 | /* reset start sector, if we reached end of device */ | |
b30ab791 AG |
966 | if (verify_done && device->ov_left == 0) |
967 | device->ov_start_sector = 0; | |
b411b363 | 968 | |
b30ab791 | 969 | drbd_md_sync(device); |
13d42685 | 970 | |
b411b363 | 971 | if (khelper_cmd) |
b30ab791 | 972 | drbd_khelper(device, khelper_cmd); |
b411b363 PR |
973 | |
974 | return 1; | |
975 | } | |
976 | ||
977 | /* helper */ | |
b30ab791 | 978 | static void move_to_net_ee_or_free(struct drbd_device *device, struct drbd_peer_request *peer_req) |
b411b363 | 979 | { |
045417f7 | 980 | if (drbd_peer_req_has_active_page(peer_req)) { |
b411b363 | 981 | /* This might happen if sendpage() has not finished */ |
db830c46 | 982 | int i = (peer_req->i.size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b30ab791 AG |
983 | atomic_add(i, &device->pp_in_use_by_net); |
984 | atomic_sub(i, &device->pp_in_use); | |
0500813f | 985 | spin_lock_irq(&device->resource->req_lock); |
84b8c06b | 986 | list_add_tail(&peer_req->dw.w.list, &device->net_ee); |
0500813f | 987 | spin_unlock_irq(&device->resource->req_lock); |
435f0740 | 988 | wake_up(&drbd_pp_wait); |
b411b363 | 989 | } else |
b30ab791 | 990 | drbd_free_peer_req(device, peer_req); |
b411b363 PR |
991 | } |
992 | ||
993 | /** | |
994 | * w_e_end_data_req() - Worker callback, to send a P_DATA_REPLY packet in response to a P_DATA_REQUEST | |
b30ab791 | 995 | * @device: DRBD device. |
b411b363 PR |
996 | * @w: work object. |
997 | * @cancel: The connection will be closed anyways | |
998 | */ | |
99920dc5 | 999 | int w_e_end_data_req(struct drbd_work *w, int cancel) |
b411b363 | 1000 | { |
84b8c06b AG |
1001 | struct drbd_device_work *dw = device_work(w); |
1002 | struct drbd_peer_request *peer_req = container_of(dw, struct drbd_peer_request, dw); | |
1003 | struct drbd_device *device = dw->device; | |
99920dc5 | 1004 | int err; |
b411b363 PR |
1005 | |
1006 | if (unlikely(cancel)) { | |
b30ab791 AG |
1007 | drbd_free_peer_req(device, peer_req); |
1008 | dec_unacked(device); | |
99920dc5 | 1009 | return 0; |
b411b363 PR |
1010 | } |
1011 | ||
db830c46 | 1012 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
69a22773 | 1013 | err = drbd_send_block(first_peer_device(device), P_DATA_REPLY, peer_req); |
b411b363 PR |
1014 | } else { |
1015 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1016 | drbd_err(device, "Sending NegDReply. sector=%llus.\n", |
db830c46 | 1017 | (unsigned long long)peer_req->i.sector); |
b411b363 | 1018 | |
69a22773 | 1019 | err = drbd_send_ack(first_peer_device(device), P_NEG_DREPLY, peer_req); |
b411b363 PR |
1020 | } |
1021 | ||
b30ab791 | 1022 | dec_unacked(device); |
b411b363 | 1023 | |
b30ab791 | 1024 | move_to_net_ee_or_free(device, peer_req); |
b411b363 | 1025 | |
99920dc5 | 1026 | if (unlikely(err)) |
d0180171 | 1027 | drbd_err(device, "drbd_send_block() failed\n"); |
99920dc5 | 1028 | return err; |
b411b363 PR |
1029 | } |
1030 | ||
1031 | /** | |
a209b4ae | 1032 | * w_e_end_rsdata_req() - Worker callback to send a P_RS_DATA_REPLY packet in response to a P_RS_DATA_REQUEST |
b411b363 PR |
1033 | * @w: work object. |
1034 | * @cancel: The connection will be closed anyways | |
1035 | */ | |
99920dc5 | 1036 | int w_e_end_rsdata_req(struct drbd_work *w, int cancel) |
b411b363 | 1037 | { |
84b8c06b AG |
1038 | struct drbd_device_work *dw = device_work(w); |
1039 | struct drbd_peer_request *peer_req = container_of(dw, struct drbd_peer_request, dw); | |
1040 | struct drbd_device *device = dw->device; | |
99920dc5 | 1041 | int err; |
b411b363 PR |
1042 | |
1043 | if (unlikely(cancel)) { | |
b30ab791 AG |
1044 | drbd_free_peer_req(device, peer_req); |
1045 | dec_unacked(device); | |
99920dc5 | 1046 | return 0; |
b411b363 PR |
1047 | } |
1048 | ||
b30ab791 AG |
1049 | if (get_ldev_if_state(device, D_FAILED)) { |
1050 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1051 | put_ldev(device); | |
b411b363 PR |
1052 | } |
1053 | ||
b30ab791 | 1054 | if (device->state.conn == C_AHEAD) { |
69a22773 | 1055 | err = drbd_send_ack(first_peer_device(device), P_RS_CANCEL, peer_req); |
db830c46 | 1056 | } else if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b30ab791 AG |
1057 | if (likely(device->state.pdsk >= D_INCONSISTENT)) { |
1058 | inc_rs_pending(device); | |
69a22773 | 1059 | err = drbd_send_block(first_peer_device(device), P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1060 | } else { |
1061 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1062 | drbd_err(device, "Not sending RSDataReply, " |
b411b363 | 1063 | "partner DISKLESS!\n"); |
99920dc5 | 1064 | err = 0; |
b411b363 PR |
1065 | } |
1066 | } else { | |
1067 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1068 | drbd_err(device, "Sending NegRSDReply. sector %llus.\n", |
db830c46 | 1069 | (unsigned long long)peer_req->i.sector); |
b411b363 | 1070 | |
69a22773 | 1071 | err = drbd_send_ack(first_peer_device(device), P_NEG_RS_DREPLY, peer_req); |
b411b363 PR |
1072 | |
1073 | /* update resync data with failure */ | |
b30ab791 | 1074 | drbd_rs_failed_io(device, peer_req->i.sector, peer_req->i.size); |
b411b363 PR |
1075 | } |
1076 | ||
b30ab791 | 1077 | dec_unacked(device); |
b411b363 | 1078 | |
b30ab791 | 1079 | move_to_net_ee_or_free(device, peer_req); |
b411b363 | 1080 | |
99920dc5 | 1081 | if (unlikely(err)) |
d0180171 | 1082 | drbd_err(device, "drbd_send_block() failed\n"); |
99920dc5 | 1083 | return err; |
b411b363 PR |
1084 | } |
1085 | ||
99920dc5 | 1086 | int w_e_end_csum_rs_req(struct drbd_work *w, int cancel) |
b411b363 | 1087 | { |
84b8c06b AG |
1088 | struct drbd_device_work *dw = device_work(w); |
1089 | struct drbd_peer_request *peer_req = container_of(dw, struct drbd_peer_request, dw); | |
1090 | struct drbd_device *device = dw->device; | |
b411b363 PR |
1091 | struct digest_info *di; |
1092 | int digest_size; | |
1093 | void *digest = NULL; | |
99920dc5 | 1094 | int err, eq = 0; |
b411b363 PR |
1095 | |
1096 | if (unlikely(cancel)) { | |
b30ab791 AG |
1097 | drbd_free_peer_req(device, peer_req); |
1098 | dec_unacked(device); | |
99920dc5 | 1099 | return 0; |
b411b363 PR |
1100 | } |
1101 | ||
b30ab791 AG |
1102 | if (get_ldev(device)) { |
1103 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1104 | put_ldev(device); | |
1d53f09e | 1105 | } |
b411b363 | 1106 | |
db830c46 | 1107 | di = peer_req->digest; |
b411b363 | 1108 | |
db830c46 | 1109 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1110 | /* quick hack to try to avoid a race against reconfiguration. |
1111 | * a real fix would be much more involved, | |
1112 | * introducing more locking mechanisms */ | |
a6b32bc3 AG |
1113 | if (first_peer_device(device)->connection->csums_tfm) { |
1114 | digest_size = crypto_hash_digestsize(first_peer_device(device)->connection->csums_tfm); | |
0b0ba1ef | 1115 | D_ASSERT(device, digest_size == di->digest_size); |
b411b363 PR |
1116 | digest = kmalloc(digest_size, GFP_NOIO); |
1117 | } | |
1118 | if (digest) { | |
79a3c8d3 | 1119 | drbd_csum_ee(first_peer_device(device)->connection->csums_tfm, peer_req, digest); |
b411b363 PR |
1120 | eq = !memcmp(digest, di->digest, digest_size); |
1121 | kfree(digest); | |
1122 | } | |
1123 | ||
1124 | if (eq) { | |
b30ab791 | 1125 | drbd_set_in_sync(device, peer_req->i.sector, peer_req->i.size); |
676396d5 | 1126 | /* rs_same_csums unit is BM_BLOCK_SIZE */ |
b30ab791 | 1127 | device->rs_same_csum += peer_req->i.size >> BM_BLOCK_SHIFT; |
69a22773 | 1128 | err = drbd_send_ack(first_peer_device(device), P_RS_IS_IN_SYNC, peer_req); |
b411b363 | 1129 | } else { |
b30ab791 | 1130 | inc_rs_pending(device); |
db830c46 AG |
1131 | peer_req->block_id = ID_SYNCER; /* By setting block_id, digest pointer becomes invalid! */ |
1132 | peer_req->flags &= ~EE_HAS_DIGEST; /* This peer request no longer has a digest pointer */ | |
204bba99 | 1133 | kfree(di); |
69a22773 | 1134 | err = drbd_send_block(first_peer_device(device), P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1135 | } |
1136 | } else { | |
69a22773 | 1137 | err = drbd_send_ack(first_peer_device(device), P_NEG_RS_DREPLY, peer_req); |
b411b363 | 1138 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 1139 | drbd_err(device, "Sending NegDReply. I guess it gets messy.\n"); |
b411b363 PR |
1140 | } |
1141 | ||
b30ab791 AG |
1142 | dec_unacked(device); |
1143 | move_to_net_ee_or_free(device, peer_req); | |
b411b363 | 1144 | |
99920dc5 | 1145 | if (unlikely(err)) |
d0180171 | 1146 | drbd_err(device, "drbd_send_block/ack() failed\n"); |
99920dc5 | 1147 | return err; |
b411b363 PR |
1148 | } |
1149 | ||
99920dc5 | 1150 | int w_e_end_ov_req(struct drbd_work *w, int cancel) |
b411b363 | 1151 | { |
84b8c06b AG |
1152 | struct drbd_device_work *dw = device_work(w); |
1153 | struct drbd_peer_request *peer_req = container_of(dw, struct drbd_peer_request, dw); | |
1154 | struct drbd_device *device = dw->device; | |
db830c46 AG |
1155 | sector_t sector = peer_req->i.sector; |
1156 | unsigned int size = peer_req->i.size; | |
b411b363 PR |
1157 | int digest_size; |
1158 | void *digest; | |
99920dc5 | 1159 | int err = 0; |
b411b363 PR |
1160 | |
1161 | if (unlikely(cancel)) | |
1162 | goto out; | |
1163 | ||
a6b32bc3 | 1164 | digest_size = crypto_hash_digestsize(first_peer_device(device)->connection->verify_tfm); |
b411b363 | 1165 | digest = kmalloc(digest_size, GFP_NOIO); |
8f21420e | 1166 | if (!digest) { |
99920dc5 | 1167 | err = 1; /* terminate the connection in case the allocation failed */ |
8f21420e | 1168 | goto out; |
b411b363 PR |
1169 | } |
1170 | ||
db830c46 | 1171 | if (likely(!(peer_req->flags & EE_WAS_ERROR))) |
79a3c8d3 | 1172 | drbd_csum_ee(first_peer_device(device)->connection->verify_tfm, peer_req, digest); |
8f21420e PR |
1173 | else |
1174 | memset(digest, 0, digest_size); | |
1175 | ||
53ea4331 LE |
1176 | /* Free e and pages before send. |
1177 | * In case we block on congestion, we could otherwise run into | |
1178 | * some distributed deadlock, if the other side blocks on | |
1179 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1180 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 1181 | drbd_free_peer_req(device, peer_req); |
db830c46 | 1182 | peer_req = NULL; |
b30ab791 | 1183 | inc_rs_pending(device); |
69a22773 | 1184 | err = drbd_send_drequest_csum(first_peer_device(device), sector, size, digest, digest_size, P_OV_REPLY); |
99920dc5 | 1185 | if (err) |
b30ab791 | 1186 | dec_rs_pending(device); |
8f21420e PR |
1187 | kfree(digest); |
1188 | ||
b411b363 | 1189 | out: |
db830c46 | 1190 | if (peer_req) |
b30ab791 AG |
1191 | drbd_free_peer_req(device, peer_req); |
1192 | dec_unacked(device); | |
99920dc5 | 1193 | return err; |
b411b363 PR |
1194 | } |
1195 | ||
b30ab791 | 1196 | void drbd_ov_out_of_sync_found(struct drbd_device *device, sector_t sector, int size) |
b411b363 | 1197 | { |
b30ab791 AG |
1198 | if (device->ov_last_oos_start + device->ov_last_oos_size == sector) { |
1199 | device->ov_last_oos_size += size>>9; | |
b411b363 | 1200 | } else { |
b30ab791 AG |
1201 | device->ov_last_oos_start = sector; |
1202 | device->ov_last_oos_size = size>>9; | |
b411b363 | 1203 | } |
b30ab791 | 1204 | drbd_set_out_of_sync(device, sector, size); |
b411b363 PR |
1205 | } |
1206 | ||
99920dc5 | 1207 | int w_e_end_ov_reply(struct drbd_work *w, int cancel) |
b411b363 | 1208 | { |
84b8c06b AG |
1209 | struct drbd_device_work *dw = device_work(w); |
1210 | struct drbd_peer_request *peer_req = container_of(dw, struct drbd_peer_request, dw); | |
1211 | struct drbd_device *device = dw->device; | |
b411b363 | 1212 | struct digest_info *di; |
b411b363 | 1213 | void *digest; |
db830c46 AG |
1214 | sector_t sector = peer_req->i.sector; |
1215 | unsigned int size = peer_req->i.size; | |
53ea4331 | 1216 | int digest_size; |
99920dc5 | 1217 | int err, eq = 0; |
58ffa580 | 1218 | bool stop_sector_reached = false; |
b411b363 PR |
1219 | |
1220 | if (unlikely(cancel)) { | |
b30ab791 AG |
1221 | drbd_free_peer_req(device, peer_req); |
1222 | dec_unacked(device); | |
99920dc5 | 1223 | return 0; |
b411b363 PR |
1224 | } |
1225 | ||
1226 | /* after "cancel", because after drbd_disconnect/drbd_rs_cancel_all | |
1227 | * the resync lru has been cleaned up already */ | |
b30ab791 AG |
1228 | if (get_ldev(device)) { |
1229 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1230 | put_ldev(device); | |
1d53f09e | 1231 | } |
b411b363 | 1232 | |
db830c46 | 1233 | di = peer_req->digest; |
b411b363 | 1234 | |
db830c46 | 1235 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
a6b32bc3 | 1236 | digest_size = crypto_hash_digestsize(first_peer_device(device)->connection->verify_tfm); |
b411b363 PR |
1237 | digest = kmalloc(digest_size, GFP_NOIO); |
1238 | if (digest) { | |
79a3c8d3 | 1239 | drbd_csum_ee(first_peer_device(device)->connection->verify_tfm, peer_req, digest); |
b411b363 | 1240 | |
0b0ba1ef | 1241 | D_ASSERT(device, digest_size == di->digest_size); |
b411b363 PR |
1242 | eq = !memcmp(digest, di->digest, digest_size); |
1243 | kfree(digest); | |
1244 | } | |
b411b363 PR |
1245 | } |
1246 | ||
9676c760 LE |
1247 | /* Free peer_req and pages before send. |
1248 | * In case we block on congestion, we could otherwise run into | |
1249 | * some distributed deadlock, if the other side blocks on | |
1250 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1251 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 1252 | drbd_free_peer_req(device, peer_req); |
b411b363 | 1253 | if (!eq) |
b30ab791 | 1254 | drbd_ov_out_of_sync_found(device, sector, size); |
b411b363 | 1255 | else |
b30ab791 | 1256 | ov_out_of_sync_print(device); |
b411b363 | 1257 | |
69a22773 | 1258 | err = drbd_send_ack_ex(first_peer_device(device), P_OV_RESULT, sector, size, |
fa79abd8 | 1259 | eq ? ID_IN_SYNC : ID_OUT_OF_SYNC); |
b411b363 | 1260 | |
b30ab791 | 1261 | dec_unacked(device); |
b411b363 | 1262 | |
b30ab791 | 1263 | --device->ov_left; |
ea5442af LE |
1264 | |
1265 | /* let's advance progress step marks only for every other megabyte */ | |
b30ab791 AG |
1266 | if ((device->ov_left & 0x200) == 0x200) |
1267 | drbd_advance_rs_marks(device, device->ov_left); | |
ea5442af | 1268 | |
b30ab791 AG |
1269 | stop_sector_reached = verify_can_do_stop_sector(device) && |
1270 | (sector + (size>>9)) >= device->ov_stop_sector; | |
58ffa580 | 1271 | |
b30ab791 AG |
1272 | if (device->ov_left == 0 || stop_sector_reached) { |
1273 | ov_out_of_sync_print(device); | |
1274 | drbd_resync_finished(device); | |
b411b363 PR |
1275 | } |
1276 | ||
99920dc5 | 1277 | return err; |
b411b363 PR |
1278 | } |
1279 | ||
b6dd1a89 LE |
1280 | /* FIXME |
1281 | * We need to track the number of pending barrier acks, | |
1282 | * and to be able to wait for them. | |
1283 | * See also comment in drbd_adm_attach before drbd_suspend_io. | |
1284 | */ | |
bde89a9e | 1285 | static int drbd_send_barrier(struct drbd_connection *connection) |
b411b363 | 1286 | { |
9f5bdc33 | 1287 | struct p_barrier *p; |
b6dd1a89 | 1288 | struct drbd_socket *sock; |
b411b363 | 1289 | |
bde89a9e AG |
1290 | sock = &connection->data; |
1291 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1292 | if (!p) |
1293 | return -EIO; | |
bde89a9e | 1294 | p->barrier = connection->send.current_epoch_nr; |
b6dd1a89 | 1295 | p->pad = 0; |
bde89a9e | 1296 | connection->send.current_epoch_writes = 0; |
b6dd1a89 | 1297 | |
bde89a9e | 1298 | return conn_send_command(connection, sock, P_BARRIER, sizeof(*p), NULL, 0); |
b411b363 PR |
1299 | } |
1300 | ||
99920dc5 | 1301 | int w_send_write_hint(struct drbd_work *w, int cancel) |
b411b363 | 1302 | { |
84b8c06b AG |
1303 | struct drbd_device *device = |
1304 | container_of(w, struct drbd_device, unplug_work); | |
9f5bdc33 AG |
1305 | struct drbd_socket *sock; |
1306 | ||
b411b363 | 1307 | if (cancel) |
99920dc5 | 1308 | return 0; |
a6b32bc3 | 1309 | sock = &first_peer_device(device)->connection->data; |
69a22773 | 1310 | if (!drbd_prepare_command(first_peer_device(device), sock)) |
9f5bdc33 | 1311 | return -EIO; |
69a22773 | 1312 | return drbd_send_command(first_peer_device(device), sock, P_UNPLUG_REMOTE, 0, NULL, 0); |
b411b363 PR |
1313 | } |
1314 | ||
bde89a9e | 1315 | static void re_init_if_first_write(struct drbd_connection *connection, unsigned int epoch) |
4eb9b3cb | 1316 | { |
bde89a9e AG |
1317 | if (!connection->send.seen_any_write_yet) { |
1318 | connection->send.seen_any_write_yet = true; | |
1319 | connection->send.current_epoch_nr = epoch; | |
1320 | connection->send.current_epoch_writes = 0; | |
4eb9b3cb LE |
1321 | } |
1322 | } | |
1323 | ||
bde89a9e | 1324 | static void maybe_send_barrier(struct drbd_connection *connection, unsigned int epoch) |
4eb9b3cb LE |
1325 | { |
1326 | /* re-init if first write on this connection */ | |
bde89a9e | 1327 | if (!connection->send.seen_any_write_yet) |
4eb9b3cb | 1328 | return; |
bde89a9e AG |
1329 | if (connection->send.current_epoch_nr != epoch) { |
1330 | if (connection->send.current_epoch_writes) | |
1331 | drbd_send_barrier(connection); | |
1332 | connection->send.current_epoch_nr = epoch; | |
4eb9b3cb LE |
1333 | } |
1334 | } | |
1335 | ||
8f7bed77 | 1336 | int w_send_out_of_sync(struct drbd_work *w, int cancel) |
73a01a18 PR |
1337 | { |
1338 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1339 | struct drbd_device *device = req->device; |
a6b32bc3 | 1340 | struct drbd_connection *connection = first_peer_device(device)->connection; |
99920dc5 | 1341 | int err; |
73a01a18 PR |
1342 | |
1343 | if (unlikely(cancel)) { | |
8554df1c | 1344 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1345 | return 0; |
73a01a18 PR |
1346 | } |
1347 | ||
bde89a9e | 1348 | /* this time, no connection->send.current_epoch_writes++; |
b6dd1a89 LE |
1349 | * If it was sent, it was the closing barrier for the last |
1350 | * replicated epoch, before we went into AHEAD mode. | |
1351 | * No more barriers will be sent, until we leave AHEAD mode again. */ | |
bde89a9e | 1352 | maybe_send_barrier(connection, req->epoch); |
b6dd1a89 | 1353 | |
69a22773 | 1354 | err = drbd_send_out_of_sync(first_peer_device(device), req); |
8554df1c | 1355 | req_mod(req, OOS_HANDED_TO_NETWORK); |
73a01a18 | 1356 | |
99920dc5 | 1357 | return err; |
73a01a18 PR |
1358 | } |
1359 | ||
b411b363 PR |
1360 | /** |
1361 | * w_send_dblock() - Worker callback to send a P_DATA packet in order to mirror a write request | |
b411b363 PR |
1362 | * @w: work object. |
1363 | * @cancel: The connection will be closed anyways | |
1364 | */ | |
99920dc5 | 1365 | int w_send_dblock(struct drbd_work *w, int cancel) |
b411b363 PR |
1366 | { |
1367 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1368 | struct drbd_device *device = req->device; |
a6b32bc3 | 1369 | struct drbd_connection *connection = first_peer_device(device)->connection; |
99920dc5 | 1370 | int err; |
b411b363 PR |
1371 | |
1372 | if (unlikely(cancel)) { | |
8554df1c | 1373 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1374 | return 0; |
b411b363 PR |
1375 | } |
1376 | ||
bde89a9e AG |
1377 | re_init_if_first_write(connection, req->epoch); |
1378 | maybe_send_barrier(connection, req->epoch); | |
1379 | connection->send.current_epoch_writes++; | |
b6dd1a89 | 1380 | |
69a22773 | 1381 | err = drbd_send_dblock(first_peer_device(device), req); |
99920dc5 | 1382 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1383 | |
99920dc5 | 1384 | return err; |
b411b363 PR |
1385 | } |
1386 | ||
1387 | /** | |
1388 | * w_send_read_req() - Worker callback to send a read request (P_DATA_REQUEST) packet | |
b411b363 PR |
1389 | * @w: work object. |
1390 | * @cancel: The connection will be closed anyways | |
1391 | */ | |
99920dc5 | 1392 | int w_send_read_req(struct drbd_work *w, int cancel) |
b411b363 PR |
1393 | { |
1394 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1395 | struct drbd_device *device = req->device; |
a6b32bc3 | 1396 | struct drbd_connection *connection = first_peer_device(device)->connection; |
99920dc5 | 1397 | int err; |
b411b363 PR |
1398 | |
1399 | if (unlikely(cancel)) { | |
8554df1c | 1400 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1401 | return 0; |
b411b363 PR |
1402 | } |
1403 | ||
b6dd1a89 LE |
1404 | /* Even read requests may close a write epoch, |
1405 | * if there was any yet. */ | |
bde89a9e | 1406 | maybe_send_barrier(connection, req->epoch); |
b6dd1a89 | 1407 | |
69a22773 | 1408 | err = drbd_send_drequest(first_peer_device(device), P_DATA_REQUEST, req->i.sector, req->i.size, |
6c1005e7 | 1409 | (unsigned long)req); |
b411b363 | 1410 | |
99920dc5 | 1411 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1412 | |
99920dc5 | 1413 | return err; |
b411b363 PR |
1414 | } |
1415 | ||
99920dc5 | 1416 | int w_restart_disk_io(struct drbd_work *w, int cancel) |
265be2d0 PR |
1417 | { |
1418 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1419 | struct drbd_device *device = req->device; |
265be2d0 | 1420 | |
0778286a | 1421 | if (bio_data_dir(req->master_bio) == WRITE && req->rq_state & RQ_IN_ACT_LOG) |
b30ab791 | 1422 | drbd_al_begin_io(device, &req->i, false); |
265be2d0 PR |
1423 | |
1424 | drbd_req_make_private_bio(req, req->master_bio); | |
b30ab791 | 1425 | req->private_bio->bi_bdev = device->ldev->backing_bdev; |
265be2d0 PR |
1426 | generic_make_request(req->private_bio); |
1427 | ||
99920dc5 | 1428 | return 0; |
265be2d0 PR |
1429 | } |
1430 | ||
b30ab791 | 1431 | static int _drbd_may_sync_now(struct drbd_device *device) |
b411b363 | 1432 | { |
b30ab791 | 1433 | struct drbd_device *odev = device; |
95f8efd0 | 1434 | int resync_after; |
b411b363 PR |
1435 | |
1436 | while (1) { | |
a3f8f7dc | 1437 | if (!odev->ldev || odev->state.disk == D_DISKLESS) |
438c8374 | 1438 | return 1; |
daeda1cc | 1439 | rcu_read_lock(); |
95f8efd0 | 1440 | resync_after = rcu_dereference(odev->ldev->disk_conf)->resync_after; |
daeda1cc | 1441 | rcu_read_unlock(); |
95f8efd0 | 1442 | if (resync_after == -1) |
b411b363 | 1443 | return 1; |
b30ab791 | 1444 | odev = minor_to_device(resync_after); |
a3f8f7dc | 1445 | if (!odev) |
841ce241 | 1446 | return 1; |
b411b363 PR |
1447 | if ((odev->state.conn >= C_SYNC_SOURCE && |
1448 | odev->state.conn <= C_PAUSED_SYNC_T) || | |
1449 | odev->state.aftr_isp || odev->state.peer_isp || | |
1450 | odev->state.user_isp) | |
1451 | return 0; | |
1452 | } | |
1453 | } | |
1454 | ||
1455 | /** | |
1456 | * _drbd_pause_after() - Pause resync on all devices that may not resync now | |
b30ab791 | 1457 | * @device: DRBD device. |
b411b363 PR |
1458 | * |
1459 | * Called from process context only (admin command and after_state_ch). | |
1460 | */ | |
b30ab791 | 1461 | static int _drbd_pause_after(struct drbd_device *device) |
b411b363 | 1462 | { |
54761697 | 1463 | struct drbd_device *odev; |
b411b363 PR |
1464 | int i, rv = 0; |
1465 | ||
695d08fa | 1466 | rcu_read_lock(); |
05a10ec7 | 1467 | idr_for_each_entry(&drbd_devices, odev, i) { |
b411b363 PR |
1468 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1469 | continue; | |
1470 | if (!_drbd_may_sync_now(odev)) | |
1471 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 1), CS_HARD, NULL) | |
1472 | != SS_NOTHING_TO_DO); | |
1473 | } | |
695d08fa | 1474 | rcu_read_unlock(); |
b411b363 PR |
1475 | |
1476 | return rv; | |
1477 | } | |
1478 | ||
1479 | /** | |
1480 | * _drbd_resume_next() - Resume resync on all devices that may resync now | |
b30ab791 | 1481 | * @device: DRBD device. |
b411b363 PR |
1482 | * |
1483 | * Called from process context only (admin command and worker). | |
1484 | */ | |
b30ab791 | 1485 | static int _drbd_resume_next(struct drbd_device *device) |
b411b363 | 1486 | { |
54761697 | 1487 | struct drbd_device *odev; |
b411b363 PR |
1488 | int i, rv = 0; |
1489 | ||
695d08fa | 1490 | rcu_read_lock(); |
05a10ec7 | 1491 | idr_for_each_entry(&drbd_devices, odev, i) { |
b411b363 PR |
1492 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1493 | continue; | |
1494 | if (odev->state.aftr_isp) { | |
1495 | if (_drbd_may_sync_now(odev)) | |
1496 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 0), | |
1497 | CS_HARD, NULL) | |
1498 | != SS_NOTHING_TO_DO) ; | |
1499 | } | |
1500 | } | |
695d08fa | 1501 | rcu_read_unlock(); |
b411b363 PR |
1502 | return rv; |
1503 | } | |
1504 | ||
b30ab791 | 1505 | void resume_next_sg(struct drbd_device *device) |
b411b363 PR |
1506 | { |
1507 | write_lock_irq(&global_state_lock); | |
b30ab791 | 1508 | _drbd_resume_next(device); |
b411b363 PR |
1509 | write_unlock_irq(&global_state_lock); |
1510 | } | |
1511 | ||
b30ab791 | 1512 | void suspend_other_sg(struct drbd_device *device) |
b411b363 PR |
1513 | { |
1514 | write_lock_irq(&global_state_lock); | |
b30ab791 | 1515 | _drbd_pause_after(device); |
b411b363 PR |
1516 | write_unlock_irq(&global_state_lock); |
1517 | } | |
1518 | ||
dc97b708 | 1519 | /* caller must hold global_state_lock */ |
b30ab791 | 1520 | enum drbd_ret_code drbd_resync_after_valid(struct drbd_device *device, int o_minor) |
b411b363 | 1521 | { |
54761697 | 1522 | struct drbd_device *odev; |
95f8efd0 | 1523 | int resync_after; |
b411b363 PR |
1524 | |
1525 | if (o_minor == -1) | |
1526 | return NO_ERROR; | |
a3f8f7dc | 1527 | if (o_minor < -1 || o_minor > MINORMASK) |
95f8efd0 | 1528 | return ERR_RESYNC_AFTER; |
b411b363 PR |
1529 | |
1530 | /* check for loops */ | |
b30ab791 | 1531 | odev = minor_to_device(o_minor); |
b411b363 | 1532 | while (1) { |
b30ab791 | 1533 | if (odev == device) |
95f8efd0 | 1534 | return ERR_RESYNC_AFTER_CYCLE; |
b411b363 | 1535 | |
a3f8f7dc LE |
1536 | /* You are free to depend on diskless, non-existing, |
1537 | * or not yet/no longer existing minors. | |
1538 | * We only reject dependency loops. | |
1539 | * We cannot follow the dependency chain beyond a detached or | |
1540 | * missing minor. | |
1541 | */ | |
1542 | if (!odev || !odev->ldev || odev->state.disk == D_DISKLESS) | |
1543 | return NO_ERROR; | |
1544 | ||
daeda1cc | 1545 | rcu_read_lock(); |
95f8efd0 | 1546 | resync_after = rcu_dereference(odev->ldev->disk_conf)->resync_after; |
daeda1cc | 1547 | rcu_read_unlock(); |
b411b363 | 1548 | /* dependency chain ends here, no cycles. */ |
95f8efd0 | 1549 | if (resync_after == -1) |
b411b363 PR |
1550 | return NO_ERROR; |
1551 | ||
1552 | /* follow the dependency chain */ | |
b30ab791 | 1553 | odev = minor_to_device(resync_after); |
b411b363 PR |
1554 | } |
1555 | } | |
1556 | ||
dc97b708 | 1557 | /* caller must hold global_state_lock */ |
b30ab791 | 1558 | void drbd_resync_after_changed(struct drbd_device *device) |
b411b363 PR |
1559 | { |
1560 | int changes; | |
b411b363 | 1561 | |
dc97b708 | 1562 | do { |
b30ab791 AG |
1563 | changes = _drbd_pause_after(device); |
1564 | changes |= _drbd_resume_next(device); | |
dc97b708 | 1565 | } while (changes); |
b411b363 PR |
1566 | } |
1567 | ||
b30ab791 | 1568 | void drbd_rs_controller_reset(struct drbd_device *device) |
9bd28d3c | 1569 | { |
813472ce PR |
1570 | struct fifo_buffer *plan; |
1571 | ||
b30ab791 AG |
1572 | atomic_set(&device->rs_sect_in, 0); |
1573 | atomic_set(&device->rs_sect_ev, 0); | |
1574 | device->rs_in_flight = 0; | |
813472ce PR |
1575 | |
1576 | /* Updating the RCU protected object in place is necessary since | |
1577 | this function gets called from atomic context. | |
1578 | It is valid since all other updates also lead to an completely | |
1579 | empty fifo */ | |
1580 | rcu_read_lock(); | |
b30ab791 | 1581 | plan = rcu_dereference(device->rs_plan_s); |
813472ce PR |
1582 | plan->total = 0; |
1583 | fifo_set(plan, 0); | |
1584 | rcu_read_unlock(); | |
9bd28d3c LE |
1585 | } |
1586 | ||
1f04af33 PR |
1587 | void start_resync_timer_fn(unsigned long data) |
1588 | { | |
b30ab791 | 1589 | struct drbd_device *device = (struct drbd_device *) data; |
1f04af33 | 1590 | |
84b8c06b AG |
1591 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, |
1592 | &device->start_resync_work); | |
1f04af33 PR |
1593 | } |
1594 | ||
99920dc5 | 1595 | int w_start_resync(struct drbd_work *w, int cancel) |
1f04af33 | 1596 | { |
84b8c06b AG |
1597 | struct drbd_device *device = |
1598 | container_of(w, struct drbd_device, start_resync_work); | |
00d56944 | 1599 | |
b30ab791 | 1600 | if (atomic_read(&device->unacked_cnt) || atomic_read(&device->rs_pending_cnt)) { |
d0180171 | 1601 | drbd_warn(device, "w_start_resync later...\n"); |
b30ab791 AG |
1602 | device->start_resync_timer.expires = jiffies + HZ/10; |
1603 | add_timer(&device->start_resync_timer); | |
99920dc5 | 1604 | return 0; |
1f04af33 PR |
1605 | } |
1606 | ||
b30ab791 AG |
1607 | drbd_start_resync(device, C_SYNC_SOURCE); |
1608 | clear_bit(AHEAD_TO_SYNC_SOURCE, &device->flags); | |
99920dc5 | 1609 | return 0; |
1f04af33 PR |
1610 | } |
1611 | ||
b411b363 PR |
1612 | /** |
1613 | * drbd_start_resync() - Start the resync process | |
b30ab791 | 1614 | * @device: DRBD device. |
b411b363 PR |
1615 | * @side: Either C_SYNC_SOURCE or C_SYNC_TARGET |
1616 | * | |
1617 | * This function might bring you directly into one of the | |
1618 | * C_PAUSED_SYNC_* states. | |
1619 | */ | |
b30ab791 | 1620 | void drbd_start_resync(struct drbd_device *device, enum drbd_conns side) |
b411b363 PR |
1621 | { |
1622 | union drbd_state ns; | |
1623 | int r; | |
1624 | ||
b30ab791 | 1625 | if (device->state.conn >= C_SYNC_SOURCE && device->state.conn < C_AHEAD) { |
d0180171 | 1626 | drbd_err(device, "Resync already running!\n"); |
b411b363 PR |
1627 | return; |
1628 | } | |
1629 | ||
b30ab791 | 1630 | if (!test_bit(B_RS_H_DONE, &device->flags)) { |
e64a3294 PR |
1631 | if (side == C_SYNC_TARGET) { |
1632 | /* Since application IO was locked out during C_WF_BITMAP_T and | |
1633 | C_WF_SYNC_UUID we are still unmodified. Before going to C_SYNC_TARGET | |
1634 | we check that we might make the data inconsistent. */ | |
b30ab791 | 1635 | r = drbd_khelper(device, "before-resync-target"); |
e64a3294 PR |
1636 | r = (r >> 8) & 0xff; |
1637 | if (r > 0) { | |
d0180171 | 1638 | drbd_info(device, "before-resync-target handler returned %d, " |
09b9e797 | 1639 | "dropping connection.\n", r); |
a6b32bc3 | 1640 | conn_request_state(first_peer_device(device)->connection, NS(conn, C_DISCONNECTING), CS_HARD); |
09b9e797 PR |
1641 | return; |
1642 | } | |
e64a3294 | 1643 | } else /* C_SYNC_SOURCE */ { |
b30ab791 | 1644 | r = drbd_khelper(device, "before-resync-source"); |
e64a3294 PR |
1645 | r = (r >> 8) & 0xff; |
1646 | if (r > 0) { | |
1647 | if (r == 3) { | |
d0180171 | 1648 | drbd_info(device, "before-resync-source handler returned %d, " |
e64a3294 PR |
1649 | "ignoring. Old userland tools?", r); |
1650 | } else { | |
d0180171 | 1651 | drbd_info(device, "before-resync-source handler returned %d, " |
e64a3294 | 1652 | "dropping connection.\n", r); |
a6b32bc3 AG |
1653 | conn_request_state(first_peer_device(device)->connection, |
1654 | NS(conn, C_DISCONNECTING), CS_HARD); | |
e64a3294 PR |
1655 | return; |
1656 | } | |
1657 | } | |
09b9e797 | 1658 | } |
b411b363 PR |
1659 | } |
1660 | ||
a6b32bc3 | 1661 | if (current == first_peer_device(device)->connection->worker.task) { |
dad20554 | 1662 | /* The worker should not sleep waiting for state_mutex, |
e64a3294 | 1663 | that can take long */ |
b30ab791 AG |
1664 | if (!mutex_trylock(device->state_mutex)) { |
1665 | set_bit(B_RS_H_DONE, &device->flags); | |
1666 | device->start_resync_timer.expires = jiffies + HZ/5; | |
1667 | add_timer(&device->start_resync_timer); | |
e64a3294 PR |
1668 | return; |
1669 | } | |
1670 | } else { | |
b30ab791 | 1671 | mutex_lock(device->state_mutex); |
e64a3294 | 1672 | } |
b30ab791 | 1673 | clear_bit(B_RS_H_DONE, &device->flags); |
b411b363 | 1674 | |
0cfac5dd | 1675 | write_lock_irq(&global_state_lock); |
a700471b | 1676 | /* Did some connection breakage or IO error race with us? */ |
b30ab791 AG |
1677 | if (device->state.conn < C_CONNECTED |
1678 | || !get_ldev_if_state(device, D_NEGOTIATING)) { | |
0cfac5dd | 1679 | write_unlock_irq(&global_state_lock); |
b30ab791 | 1680 | mutex_unlock(device->state_mutex); |
b411b363 PR |
1681 | return; |
1682 | } | |
1683 | ||
b30ab791 | 1684 | ns = drbd_read_state(device); |
b411b363 | 1685 | |
b30ab791 | 1686 | ns.aftr_isp = !_drbd_may_sync_now(device); |
b411b363 PR |
1687 | |
1688 | ns.conn = side; | |
1689 | ||
1690 | if (side == C_SYNC_TARGET) | |
1691 | ns.disk = D_INCONSISTENT; | |
1692 | else /* side == C_SYNC_SOURCE */ | |
1693 | ns.pdsk = D_INCONSISTENT; | |
1694 | ||
b30ab791 AG |
1695 | r = __drbd_set_state(device, ns, CS_VERBOSE, NULL); |
1696 | ns = drbd_read_state(device); | |
b411b363 PR |
1697 | |
1698 | if (ns.conn < C_CONNECTED) | |
1699 | r = SS_UNKNOWN_ERROR; | |
1700 | ||
1701 | if (r == SS_SUCCESS) { | |
b30ab791 | 1702 | unsigned long tw = drbd_bm_total_weight(device); |
1d7734a0 LE |
1703 | unsigned long now = jiffies; |
1704 | int i; | |
1705 | ||
b30ab791 AG |
1706 | device->rs_failed = 0; |
1707 | device->rs_paused = 0; | |
1708 | device->rs_same_csum = 0; | |
1709 | device->rs_last_events = 0; | |
1710 | device->rs_last_sect_ev = 0; | |
1711 | device->rs_total = tw; | |
1712 | device->rs_start = now; | |
1d7734a0 | 1713 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
1714 | device->rs_mark_left[i] = tw; |
1715 | device->rs_mark_time[i] = now; | |
1d7734a0 | 1716 | } |
b30ab791 | 1717 | _drbd_pause_after(device); |
b411b363 PR |
1718 | } |
1719 | write_unlock_irq(&global_state_lock); | |
5a22db89 | 1720 | |
b411b363 | 1721 | if (r == SS_SUCCESS) { |
328e0f12 PR |
1722 | /* reset rs_last_bcast when a resync or verify is started, |
1723 | * to deal with potential jiffies wrap. */ | |
b30ab791 | 1724 | device->rs_last_bcast = jiffies - HZ; |
328e0f12 | 1725 | |
d0180171 | 1726 | drbd_info(device, "Began resync as %s (will sync %lu KB [%lu bits set]).\n", |
b411b363 | 1727 | drbd_conn_str(ns.conn), |
b30ab791 AG |
1728 | (unsigned long) device->rs_total << (BM_BLOCK_SHIFT-10), |
1729 | (unsigned long) device->rs_total); | |
6c922ed5 | 1730 | if (side == C_SYNC_TARGET) |
b30ab791 | 1731 | device->bm_resync_fo = 0; |
6c922ed5 LE |
1732 | |
1733 | /* Since protocol 96, we must serialize drbd_gen_and_send_sync_uuid | |
1734 | * with w_send_oos, or the sync target will get confused as to | |
1735 | * how much bits to resync. We cannot do that always, because for an | |
1736 | * empty resync and protocol < 95, we need to do it here, as we call | |
1737 | * drbd_resync_finished from here in that case. | |
1738 | * We drbd_gen_and_send_sync_uuid here for protocol < 96, | |
1739 | * and from after_state_ch otherwise. */ | |
a6b32bc3 AG |
1740 | if (side == C_SYNC_SOURCE && |
1741 | first_peer_device(device)->connection->agreed_pro_version < 96) | |
69a22773 | 1742 | drbd_gen_and_send_sync_uuid(first_peer_device(device)); |
b411b363 | 1743 | |
a6b32bc3 AG |
1744 | if (first_peer_device(device)->connection->agreed_pro_version < 95 && |
1745 | device->rs_total == 0) { | |
af85e8e8 LE |
1746 | /* This still has a race (about when exactly the peers |
1747 | * detect connection loss) that can lead to a full sync | |
1748 | * on next handshake. In 8.3.9 we fixed this with explicit | |
1749 | * resync-finished notifications, but the fix | |
1750 | * introduces a protocol change. Sleeping for some | |
1751 | * time longer than the ping interval + timeout on the | |
1752 | * SyncSource, to give the SyncTarget the chance to | |
1753 | * detect connection loss, then waiting for a ping | |
1754 | * response (implicit in drbd_resync_finished) reduces | |
1755 | * the race considerably, but does not solve it. */ | |
44ed167d PR |
1756 | if (side == C_SYNC_SOURCE) { |
1757 | struct net_conf *nc; | |
1758 | int timeo; | |
1759 | ||
1760 | rcu_read_lock(); | |
a6b32bc3 | 1761 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); |
44ed167d PR |
1762 | timeo = nc->ping_int * HZ + nc->ping_timeo * HZ / 9; |
1763 | rcu_read_unlock(); | |
1764 | schedule_timeout_interruptible(timeo); | |
1765 | } | |
b30ab791 | 1766 | drbd_resync_finished(device); |
b411b363 PR |
1767 | } |
1768 | ||
b30ab791 AG |
1769 | drbd_rs_controller_reset(device); |
1770 | /* ns.conn may already be != device->state.conn, | |
b411b363 PR |
1771 | * we may have been paused in between, or become paused until |
1772 | * the timer triggers. | |
1773 | * No matter, that is handled in resync_timer_fn() */ | |
1774 | if (ns.conn == C_SYNC_TARGET) | |
b30ab791 | 1775 | mod_timer(&device->resync_timer, jiffies); |
b411b363 | 1776 | |
b30ab791 | 1777 | drbd_md_sync(device); |
b411b363 | 1778 | } |
b30ab791 AG |
1779 | put_ldev(device); |
1780 | mutex_unlock(device->state_mutex); | |
b411b363 PR |
1781 | } |
1782 | ||
b6dd1a89 LE |
1783 | /* If the resource already closed the current epoch, but we did not |
1784 | * (because we have not yet seen new requests), we should send the | |
1785 | * corresponding barrier now. Must be checked within the same spinlock | |
1786 | * that is used to check for new requests. */ | |
bde89a9e | 1787 | static bool need_to_send_barrier(struct drbd_connection *connection) |
b6dd1a89 LE |
1788 | { |
1789 | if (!connection->send.seen_any_write_yet) | |
1790 | return false; | |
1791 | ||
1792 | /* Skip barriers that do not contain any writes. | |
1793 | * This may happen during AHEAD mode. */ | |
1794 | if (!connection->send.current_epoch_writes) | |
1795 | return false; | |
1796 | ||
1797 | /* ->req_lock is held when requests are queued on | |
1798 | * connection->sender_work, and put into ->transfer_log. | |
1799 | * It is also held when ->current_tle_nr is increased. | |
1800 | * So either there are already new requests queued, | |
1801 | * and corresponding barriers will be send there. | |
1802 | * Or nothing new is queued yet, so the difference will be 1. | |
1803 | */ | |
1804 | if (atomic_read(&connection->current_tle_nr) != | |
1805 | connection->send.current_epoch_nr + 1) | |
1806 | return false; | |
1807 | ||
1808 | return true; | |
1809 | } | |
1810 | ||
a186e478 | 1811 | static bool dequeue_work_batch(struct drbd_work_queue *queue, struct list_head *work_list) |
8c0785a5 LE |
1812 | { |
1813 | spin_lock_irq(&queue->q_lock); | |
1814 | list_splice_init(&queue->q, work_list); | |
1815 | spin_unlock_irq(&queue->q_lock); | |
1816 | return !list_empty(work_list); | |
1817 | } | |
1818 | ||
a186e478 | 1819 | static bool dequeue_work_item(struct drbd_work_queue *queue, struct list_head *work_list) |
8c0785a5 LE |
1820 | { |
1821 | spin_lock_irq(&queue->q_lock); | |
1822 | if (!list_empty(&queue->q)) | |
1823 | list_move(queue->q.next, work_list); | |
1824 | spin_unlock_irq(&queue->q_lock); | |
1825 | return !list_empty(work_list); | |
1826 | } | |
1827 | ||
bde89a9e | 1828 | static void wait_for_work(struct drbd_connection *connection, struct list_head *work_list) |
b6dd1a89 LE |
1829 | { |
1830 | DEFINE_WAIT(wait); | |
1831 | struct net_conf *nc; | |
1832 | int uncork, cork; | |
1833 | ||
1834 | dequeue_work_item(&connection->sender_work, work_list); | |
1835 | if (!list_empty(work_list)) | |
1836 | return; | |
1837 | ||
1838 | /* Still nothing to do? | |
1839 | * Maybe we still need to close the current epoch, | |
1840 | * even if no new requests are queued yet. | |
1841 | * | |
1842 | * Also, poke TCP, just in case. | |
1843 | * Then wait for new work (or signal). */ | |
1844 | rcu_read_lock(); | |
1845 | nc = rcu_dereference(connection->net_conf); | |
1846 | uncork = nc ? nc->tcp_cork : 0; | |
1847 | rcu_read_unlock(); | |
1848 | if (uncork) { | |
1849 | mutex_lock(&connection->data.mutex); | |
1850 | if (connection->data.socket) | |
1851 | drbd_tcp_uncork(connection->data.socket); | |
1852 | mutex_unlock(&connection->data.mutex); | |
1853 | } | |
1854 | ||
1855 | for (;;) { | |
1856 | int send_barrier; | |
1857 | prepare_to_wait(&connection->sender_work.q_wait, &wait, TASK_INTERRUPTIBLE); | |
0500813f | 1858 | spin_lock_irq(&connection->resource->req_lock); |
b6dd1a89 | 1859 | spin_lock(&connection->sender_work.q_lock); /* FIXME get rid of this one? */ |
bc317a9e LE |
1860 | /* dequeue single item only, |
1861 | * we still use drbd_queue_work_front() in some places */ | |
1862 | if (!list_empty(&connection->sender_work.q)) | |
1863 | list_move(connection->sender_work.q.next, work_list); | |
b6dd1a89 LE |
1864 | spin_unlock(&connection->sender_work.q_lock); /* FIXME get rid of this one? */ |
1865 | if (!list_empty(work_list) || signal_pending(current)) { | |
0500813f | 1866 | spin_unlock_irq(&connection->resource->req_lock); |
b6dd1a89 LE |
1867 | break; |
1868 | } | |
1869 | send_barrier = need_to_send_barrier(connection); | |
0500813f | 1870 | spin_unlock_irq(&connection->resource->req_lock); |
b6dd1a89 LE |
1871 | if (send_barrier) { |
1872 | drbd_send_barrier(connection); | |
1873 | connection->send.current_epoch_nr++; | |
1874 | } | |
1875 | schedule(); | |
1876 | /* may be woken up for other things but new work, too, | |
1877 | * e.g. if the current epoch got closed. | |
1878 | * In which case we send the barrier above. */ | |
1879 | } | |
1880 | finish_wait(&connection->sender_work.q_wait, &wait); | |
1881 | ||
1882 | /* someone may have changed the config while we have been waiting above. */ | |
1883 | rcu_read_lock(); | |
1884 | nc = rcu_dereference(connection->net_conf); | |
1885 | cork = nc ? nc->tcp_cork : 0; | |
1886 | rcu_read_unlock(); | |
1887 | mutex_lock(&connection->data.mutex); | |
1888 | if (connection->data.socket) { | |
1889 | if (cork) | |
1890 | drbd_tcp_cork(connection->data.socket); | |
1891 | else if (!uncork) | |
1892 | drbd_tcp_uncork(connection->data.socket); | |
1893 | } | |
1894 | mutex_unlock(&connection->data.mutex); | |
1895 | } | |
1896 | ||
b411b363 PR |
1897 | int drbd_worker(struct drbd_thread *thi) |
1898 | { | |
bde89a9e | 1899 | struct drbd_connection *connection = thi->connection; |
84b8c06b | 1900 | struct drbd_device_work *dw = NULL; |
c06ece6b | 1901 | struct drbd_peer_device *peer_device; |
b411b363 | 1902 | LIST_HEAD(work_list); |
8c0785a5 | 1903 | int vnr; |
b411b363 | 1904 | |
e77a0a5c | 1905 | while (get_t_state(thi) == RUNNING) { |
80822284 | 1906 | drbd_thread_current_set_cpu(thi); |
b411b363 | 1907 | |
8c0785a5 LE |
1908 | /* as long as we use drbd_queue_work_front(), |
1909 | * we may only dequeue single work items here, not batches. */ | |
1910 | if (list_empty(&work_list)) | |
bde89a9e | 1911 | wait_for_work(connection, &work_list); |
b411b363 | 1912 | |
8c0785a5 | 1913 | if (signal_pending(current)) { |
b411b363 | 1914 | flush_signals(current); |
19393e10 | 1915 | if (get_t_state(thi) == RUNNING) { |
1ec861eb | 1916 | drbd_warn(connection, "Worker got an unexpected signal\n"); |
b411b363 | 1917 | continue; |
19393e10 | 1918 | } |
b411b363 PR |
1919 | break; |
1920 | } | |
1921 | ||
e77a0a5c | 1922 | if (get_t_state(thi) != RUNNING) |
b411b363 | 1923 | break; |
b411b363 | 1924 | |
8c0785a5 | 1925 | while (!list_empty(&work_list)) { |
84b8c06b AG |
1926 | dw = list_first_entry(&work_list, struct drbd_device_work, w.list); |
1927 | list_del_init(&dw->w.list); | |
1928 | if (dw->w.cb(&dw->w, connection->cstate < C_WF_REPORT_PARAMS) == 0) | |
8c0785a5 | 1929 | continue; |
bde89a9e AG |
1930 | if (connection->cstate >= C_WF_REPORT_PARAMS) |
1931 | conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); | |
b411b363 PR |
1932 | } |
1933 | } | |
b411b363 | 1934 | |
8c0785a5 | 1935 | do { |
b411b363 | 1936 | while (!list_empty(&work_list)) { |
84b8c06b AG |
1937 | dw = list_first_entry(&work_list, struct drbd_device_work, w.list); |
1938 | list_del_init(&dw->w.list); | |
1939 | dw->w.cb(&dw->w, 1); | |
b411b363 | 1940 | } |
bde89a9e | 1941 | dequeue_work_batch(&connection->sender_work, &work_list); |
8c0785a5 | 1942 | } while (!list_empty(&work_list)); |
b411b363 | 1943 | |
c141ebda | 1944 | rcu_read_lock(); |
c06ece6b AG |
1945 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
1946 | struct drbd_device *device = peer_device->device; | |
0b0ba1ef | 1947 | D_ASSERT(device, device->state.disk == D_DISKLESS && device->state.conn == C_STANDALONE); |
b30ab791 | 1948 | kref_get(&device->kref); |
c141ebda | 1949 | rcu_read_unlock(); |
b30ab791 | 1950 | drbd_device_cleanup(device); |
05a10ec7 | 1951 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda | 1952 | rcu_read_lock(); |
0e29d163 | 1953 | } |
c141ebda | 1954 | rcu_read_unlock(); |
b411b363 PR |
1955 | |
1956 | return 0; | |
1957 | } |