Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_worker.c | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | drbd is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation; either version 2, or (at your option) | |
13 | any later version. | |
14 | ||
15 | drbd is distributed in the hope that it will be useful, | |
16 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | GNU General Public License for more details. | |
19 | ||
20 | You should have received a copy of the GNU General Public License | |
21 | along with drbd; see the file COPYING. If not, write to | |
22 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | ||
84b8c06b | 24 | */ |
b411b363 | 25 | |
b411b363 | 26 | #include <linux/module.h> |
b411b363 PR |
27 | #include <linux/drbd.h> |
28 | #include <linux/sched.h> | |
b411b363 PR |
29 | #include <linux/wait.h> |
30 | #include <linux/mm.h> | |
31 | #include <linux/memcontrol.h> | |
32 | #include <linux/mm_inline.h> | |
33 | #include <linux/slab.h> | |
34 | #include <linux/random.h> | |
b411b363 PR |
35 | #include <linux/string.h> |
36 | #include <linux/scatterlist.h> | |
37 | ||
38 | #include "drbd_int.h" | |
a3603a6e | 39 | #include "drbd_protocol.h" |
b411b363 | 40 | #include "drbd_req.h" |
b411b363 | 41 | |
d448a2e1 AG |
42 | static int make_ov_request(struct drbd_device *, int); |
43 | static int make_resync_request(struct drbd_device *, int); | |
b411b363 | 44 | |
c5a91619 | 45 | /* endio handlers: |
ed15b795 | 46 | * drbd_md_endio (defined here) |
fcefa62e AG |
47 | * drbd_request_endio (defined here) |
48 | * drbd_peer_request_endio (defined here) | |
ed15b795 | 49 | * drbd_bm_endio (defined in drbd_bitmap.c) |
c5a91619 | 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 | ||
b411b363 PR |
58 | /* used for synchronous meta data and bitmap IO |
59 | * submitted by drbd_md_sync_page_io() | |
60 | */ | |
4246a0b6 | 61 | void drbd_md_endio(struct bio *bio) |
b411b363 | 62 | { |
b30ab791 | 63 | struct drbd_device *device; |
b411b363 | 64 | |
e37d2438 | 65 | device = bio->bi_private; |
4246a0b6 | 66 | device->md_io.error = bio->bi_error; |
b411b363 | 67 | |
0cfac5dd PR |
68 | /* We grabbed an extra reference in _drbd_md_sync_page_io() to be able |
69 | * to timeout on the lower level device, and eventually detach from it. | |
70 | * If this io completion runs after that timeout expired, this | |
71 | * drbd_md_put_buffer() may allow us to finally try and re-attach. | |
72 | * During normal operation, this only puts that extra reference | |
73 | * down to 1 again. | |
74 | * Make sure we first drop the reference, and only then signal | |
75 | * completion, or we may (in drbd_al_read_log()) cycle so fast into the | |
76 | * next drbd_md_sync_page_io(), that we trigger the | |
b30ab791 | 77 | * ASSERT(atomic_read(&device->md_io_in_use) == 1) there. |
0cfac5dd | 78 | */ |
b30ab791 | 79 | drbd_md_put_buffer(device); |
e37d2438 | 80 | device->md_io.done = 1; |
b30ab791 | 81 | wake_up(&device->misc_wait); |
cdfda633 | 82 | bio_put(bio); |
b30ab791 AG |
83 | if (device->ldev) /* special case: drbd_md_read() during drbd_adm_attach() */ |
84 | put_ldev(device); | |
b411b363 PR |
85 | } |
86 | ||
87 | /* reads on behalf of the partner, | |
88 | * "submitted" by the receiver | |
89 | */ | |
a186e478 | 90 | static void drbd_endio_read_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
91 | { |
92 | unsigned long flags = 0; | |
6780139c AG |
93 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
94 | struct drbd_device *device = peer_device->device; | |
b411b363 | 95 | |
0500813f | 96 | spin_lock_irqsave(&device->resource->req_lock, flags); |
b30ab791 | 97 | device->read_cnt += peer_req->i.size >> 9; |
a8cd15ba | 98 | list_del(&peer_req->w.list); |
b30ab791 AG |
99 | if (list_empty(&device->read_ee)) |
100 | wake_up(&device->ee_wait); | |
db830c46 | 101 | if (test_bit(__EE_WAS_ERROR, &peer_req->flags)) |
b30ab791 | 102 | __drbd_chk_io_error(device, DRBD_READ_ERROR); |
0500813f | 103 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b411b363 | 104 | |
6780139c | 105 | drbd_queue_work(&peer_device->connection->sender_work, &peer_req->w); |
b30ab791 | 106 | put_ldev(device); |
b411b363 PR |
107 | } |
108 | ||
109 | /* writes on behalf of the partner, or resync writes, | |
45bb912b | 110 | * "submitted" by the receiver, final stage. */ |
a0fb3c47 | 111 | void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
112 | { |
113 | unsigned long flags = 0; | |
6780139c AG |
114 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
115 | struct drbd_device *device = peer_device->device; | |
668700b4 | 116 | struct drbd_connection *connection = peer_device->connection; |
181286ad | 117 | struct drbd_interval i; |
b411b363 | 118 | int do_wake; |
579b57ed | 119 | u64 block_id; |
b411b363 | 120 | int do_al_complete_io; |
b411b363 | 121 | |
db830c46 | 122 | /* after we moved peer_req to done_ee, |
b411b363 PR |
123 | * we may no longer access it, |
124 | * it may be freed/reused already! | |
125 | * (as soon as we release the req_lock) */ | |
181286ad | 126 | i = peer_req->i; |
db830c46 AG |
127 | do_al_complete_io = peer_req->flags & EE_CALL_AL_COMPLETE_IO; |
128 | block_id = peer_req->block_id; | |
21ae5d7f | 129 | peer_req->flags &= ~EE_CALL_AL_COMPLETE_IO; |
b411b363 | 130 | |
0500813f | 131 | spin_lock_irqsave(&device->resource->req_lock, flags); |
b30ab791 | 132 | device->writ_cnt += peer_req->i.size >> 9; |
a8cd15ba | 133 | list_move_tail(&peer_req->w.list, &device->done_ee); |
b411b363 | 134 | |
bb3bfe96 | 135 | /* |
5e472264 | 136 | * Do not remove from the write_requests tree here: we did not send the |
bb3bfe96 AG |
137 | * Ack yet and did not wake possibly waiting conflicting requests. |
138 | * Removed from the tree from "drbd_process_done_ee" within the | |
84b8c06b | 139 | * appropriate dw.cb (e_end_block/e_end_resync_block) or from |
bb3bfe96 AG |
140 | * _drbd_clear_done_ee. |
141 | */ | |
b411b363 | 142 | |
b30ab791 | 143 | do_wake = list_empty(block_id == ID_SYNCER ? &device->sync_ee : &device->active_ee); |
b411b363 | 144 | |
a0fb3c47 LE |
145 | /* FIXME do we want to detach for failed REQ_DISCARD? |
146 | * ((peer_req->flags & (EE_WAS_ERROR|EE_IS_TRIM)) == EE_WAS_ERROR) */ | |
147 | if (peer_req->flags & EE_WAS_ERROR) | |
b30ab791 | 148 | __drbd_chk_io_error(device, DRBD_WRITE_ERROR); |
668700b4 PR |
149 | |
150 | if (connection->cstate >= C_WF_REPORT_PARAMS) { | |
151 | kref_get(&device->kref); /* put is in drbd_send_acks_wf() */ | |
152 | if (!queue_work(connection->ack_sender, &peer_device->send_acks_work)) | |
153 | kref_put(&device->kref, drbd_destroy_device); | |
154 | } | |
0500813f | 155 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b411b363 | 156 | |
579b57ed | 157 | if (block_id == ID_SYNCER) |
b30ab791 | 158 | drbd_rs_complete_io(device, i.sector); |
b411b363 PR |
159 | |
160 | if (do_wake) | |
b30ab791 | 161 | wake_up(&device->ee_wait); |
b411b363 PR |
162 | |
163 | if (do_al_complete_io) | |
b30ab791 | 164 | drbd_al_complete_io(device, &i); |
b411b363 | 165 | |
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 | */ | |
4246a0b6 | 172 | void drbd_peer_request_endio(struct bio *bio) |
45bb912b | 173 | { |
db830c46 | 174 | struct drbd_peer_request *peer_req = bio->bi_private; |
a8cd15ba | 175 | struct drbd_device *device = peer_req->peer_device->device; |
45bb912b | 176 | int is_write = bio_data_dir(bio) == WRITE; |
a0fb3c47 | 177 | int is_discard = !!(bio->bi_rw & REQ_DISCARD); |
45bb912b | 178 | |
4246a0b6 | 179 | if (bio->bi_error && __ratelimit(&drbd_ratelimit_state)) |
d0180171 | 180 | drbd_warn(device, "%s: error=%d s=%llus\n", |
a0fb3c47 | 181 | is_write ? (is_discard ? "discard" : "write") |
4246a0b6 | 182 | : "read", bio->bi_error, |
db830c46 | 183 | (unsigned long long)peer_req->i.sector); |
45bb912b | 184 | |
4246a0b6 | 185 | if (bio->bi_error) |
db830c46 | 186 | set_bit(__EE_WAS_ERROR, &peer_req->flags); |
45bb912b LE |
187 | |
188 | bio_put(bio); /* no need for the bio anymore */ | |
db830c46 | 189 | if (atomic_dec_and_test(&peer_req->pending_bios)) { |
45bb912b | 190 | if (is_write) |
db830c46 | 191 | drbd_endio_write_sec_final(peer_req); |
45bb912b | 192 | else |
db830c46 | 193 | drbd_endio_read_sec_final(peer_req); |
45bb912b | 194 | } |
b411b363 PR |
195 | } |
196 | ||
142207f7 LE |
197 | void drbd_panic_after_delayed_completion_of_aborted_request(struct drbd_device *device) |
198 | { | |
199 | panic("drbd%u %s/%u potential random memory corruption caused by delayed completion of aborted local request\n", | |
200 | device->minor, device->resource->name, device->vnr); | |
201 | } | |
202 | ||
b411b363 PR |
203 | /* read, readA or write requests on R_PRIMARY coming from drbd_make_request |
204 | */ | |
4246a0b6 | 205 | void drbd_request_endio(struct bio *bio) |
b411b363 | 206 | { |
a115413d | 207 | unsigned long flags; |
b411b363 | 208 | struct drbd_request *req = bio->bi_private; |
84b8c06b | 209 | struct drbd_device *device = req->device; |
a115413d | 210 | struct bio_and_error m; |
b411b363 | 211 | enum drbd_req_event what; |
1b6dd252 PR |
212 | |
213 | /* If this request was aborted locally before, | |
214 | * but now was completed "successfully", | |
215 | * chances are that this caused arbitrary data corruption. | |
216 | * | |
217 | * "aborting" requests, or force-detaching the disk, is intended for | |
218 | * completely blocked/hung local backing devices which do no longer | |
219 | * complete requests at all, not even do error completions. In this | |
220 | * situation, usually a hard-reset and failover is the only way out. | |
221 | * | |
222 | * By "aborting", basically faking a local error-completion, | |
223 | * we allow for a more graceful swichover by cleanly migrating services. | |
224 | * Still the affected node has to be rebooted "soon". | |
225 | * | |
226 | * By completing these requests, we allow the upper layers to re-use | |
227 | * the associated data pages. | |
228 | * | |
229 | * If later the local backing device "recovers", and now DMAs some data | |
230 | * from disk into the original request pages, in the best case it will | |
231 | * just put random data into unused pages; but typically it will corrupt | |
232 | * meanwhile completely unrelated data, causing all sorts of damage. | |
233 | * | |
234 | * Which means delayed successful completion, | |
235 | * especially for READ requests, | |
236 | * is a reason to panic(). | |
237 | * | |
238 | * We assume that a delayed *error* completion is OK, | |
239 | * though we still will complain noisily about it. | |
240 | */ | |
241 | if (unlikely(req->rq_state & RQ_LOCAL_ABORTED)) { | |
242 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 243 | drbd_emerg(device, "delayed completion of aborted local request; disk-timeout may be too aggressive\n"); |
1b6dd252 | 244 | |
4246a0b6 | 245 | if (!bio->bi_error) |
142207f7 | 246 | drbd_panic_after_delayed_completion_of_aborted_request(device); |
1b6dd252 PR |
247 | } |
248 | ||
b411b363 | 249 | /* to avoid recursion in __req_mod */ |
4246a0b6 | 250 | if (unlikely(bio->bi_error)) { |
2f632aeb | 251 | if (bio->bi_rw & REQ_DISCARD) |
4246a0b6 | 252 | what = (bio->bi_error == -EOPNOTSUPP) |
2f632aeb LE |
253 | ? DISCARD_COMPLETED_NOTSUPP |
254 | : DISCARD_COMPLETED_WITH_ERROR; | |
255 | else | |
256 | what = (bio_data_dir(bio) == WRITE) | |
8554df1c | 257 | ? WRITE_COMPLETED_WITH_ERROR |
5c3c7e64 | 258 | : (bio_rw(bio) == READ) |
8554df1c AG |
259 | ? READ_COMPLETED_WITH_ERROR |
260 | : READ_AHEAD_COMPLETED_WITH_ERROR; | |
b411b363 | 261 | } else |
8554df1c | 262 | what = COMPLETED_OK; |
b411b363 PR |
263 | |
264 | bio_put(req->private_bio); | |
4246a0b6 | 265 | req->private_bio = ERR_PTR(bio->bi_error); |
b411b363 | 266 | |
a115413d | 267 | /* not req_mod(), we need irqsave here! */ |
0500813f | 268 | spin_lock_irqsave(&device->resource->req_lock, flags); |
a115413d | 269 | __req_mod(req, what, &m); |
0500813f | 270 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
b30ab791 | 271 | put_ldev(device); |
a115413d LE |
272 | |
273 | if (m.bio) | |
b30ab791 | 274 | complete_master_bio(device, &m); |
b411b363 PR |
275 | } |
276 | ||
79a3c8d3 | 277 | void drbd_csum_ee(struct crypto_hash *tfm, struct drbd_peer_request *peer_req, void *digest) |
45bb912b LE |
278 | { |
279 | struct hash_desc desc; | |
280 | struct scatterlist sg; | |
db830c46 | 281 | struct page *page = peer_req->pages; |
45bb912b LE |
282 | struct page *tmp; |
283 | unsigned len; | |
284 | ||
285 | desc.tfm = tfm; | |
286 | desc.flags = 0; | |
287 | ||
288 | sg_init_table(&sg, 1); | |
289 | crypto_hash_init(&desc); | |
290 | ||
291 | while ((tmp = page_chain_next(page))) { | |
292 | /* all but the last page will be fully used */ | |
293 | sg_set_page(&sg, page, PAGE_SIZE, 0); | |
294 | crypto_hash_update(&desc, &sg, sg.length); | |
295 | page = tmp; | |
296 | } | |
297 | /* and now the last, possibly only partially used page */ | |
db830c46 | 298 | len = peer_req->i.size & (PAGE_SIZE - 1); |
45bb912b LE |
299 | sg_set_page(&sg, page, len ?: PAGE_SIZE, 0); |
300 | crypto_hash_update(&desc, &sg, sg.length); | |
301 | crypto_hash_final(&desc, digest); | |
302 | } | |
303 | ||
79a3c8d3 | 304 | void drbd_csum_bio(struct crypto_hash *tfm, struct bio *bio, void *digest) |
b411b363 PR |
305 | { |
306 | struct hash_desc desc; | |
307 | struct scatterlist sg; | |
7988613b KO |
308 | struct bio_vec bvec; |
309 | struct bvec_iter iter; | |
b411b363 PR |
310 | |
311 | desc.tfm = tfm; | |
312 | desc.flags = 0; | |
313 | ||
314 | sg_init_table(&sg, 1); | |
315 | crypto_hash_init(&desc); | |
316 | ||
7988613b KO |
317 | bio_for_each_segment(bvec, bio, iter) { |
318 | sg_set_page(&sg, bvec.bv_page, bvec.bv_len, bvec.bv_offset); | |
b411b363 PR |
319 | crypto_hash_update(&desc, &sg, sg.length); |
320 | } | |
321 | crypto_hash_final(&desc, digest); | |
322 | } | |
323 | ||
9676c760 | 324 | /* MAYBE merge common code with w_e_end_ov_req */ |
99920dc5 | 325 | static int w_e_send_csum(struct drbd_work *w, int cancel) |
b411b363 | 326 | { |
a8cd15ba | 327 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
328 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
329 | struct drbd_device *device = peer_device->device; | |
b411b363 PR |
330 | int digest_size; |
331 | void *digest; | |
99920dc5 | 332 | int err = 0; |
b411b363 | 333 | |
53ea4331 LE |
334 | if (unlikely(cancel)) |
335 | goto out; | |
b411b363 | 336 | |
9676c760 | 337 | if (unlikely((peer_req->flags & EE_WAS_ERROR) != 0)) |
53ea4331 | 338 | goto out; |
b411b363 | 339 | |
6780139c | 340 | digest_size = crypto_hash_digestsize(peer_device->connection->csums_tfm); |
53ea4331 LE |
341 | digest = kmalloc(digest_size, GFP_NOIO); |
342 | if (digest) { | |
db830c46 AG |
343 | sector_t sector = peer_req->i.sector; |
344 | unsigned int size = peer_req->i.size; | |
6780139c | 345 | drbd_csum_ee(peer_device->connection->csums_tfm, peer_req, digest); |
9676c760 | 346 | /* Free peer_req and pages before send. |
53ea4331 LE |
347 | * In case we block on congestion, we could otherwise run into |
348 | * some distributed deadlock, if the other side blocks on | |
349 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 350 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 351 | drbd_free_peer_req(device, peer_req); |
db830c46 | 352 | peer_req = NULL; |
b30ab791 | 353 | inc_rs_pending(device); |
6780139c | 354 | err = drbd_send_drequest_csum(peer_device, sector, size, |
db1b0b72 AG |
355 | digest, digest_size, |
356 | P_CSUM_RS_REQUEST); | |
53ea4331 LE |
357 | kfree(digest); |
358 | } else { | |
d0180171 | 359 | drbd_err(device, "kmalloc() of digest failed.\n"); |
99920dc5 | 360 | err = -ENOMEM; |
53ea4331 | 361 | } |
b411b363 | 362 | |
53ea4331 | 363 | out: |
db830c46 | 364 | if (peer_req) |
b30ab791 | 365 | drbd_free_peer_req(device, peer_req); |
b411b363 | 366 | |
99920dc5 | 367 | if (unlikely(err)) |
d0180171 | 368 | drbd_err(device, "drbd_send_drequest(..., csum) failed\n"); |
99920dc5 | 369 | return err; |
b411b363 PR |
370 | } |
371 | ||
372 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
373 | ||
69a22773 | 374 | static int read_for_csum(struct drbd_peer_device *peer_device, sector_t sector, int size) |
b411b363 | 375 | { |
69a22773 | 376 | struct drbd_device *device = peer_device->device; |
db830c46 | 377 | struct drbd_peer_request *peer_req; |
b411b363 | 378 | |
b30ab791 | 379 | if (!get_ldev(device)) |
80a40e43 | 380 | return -EIO; |
b411b363 PR |
381 | |
382 | /* GFP_TRY, because if there is no memory available right now, this may | |
383 | * be rescheduled for later. It is "only" background resync, after all. */ | |
69a22773 | 384 | peer_req = drbd_alloc_peer_req(peer_device, ID_SYNCER /* unused */, sector, |
a0fb3c47 | 385 | size, true /* has real payload */, GFP_TRY); |
db830c46 | 386 | if (!peer_req) |
80a40e43 | 387 | goto defer; |
b411b363 | 388 | |
a8cd15ba | 389 | peer_req->w.cb = w_e_send_csum; |
0500813f | 390 | spin_lock_irq(&device->resource->req_lock); |
b9ed7080 | 391 | list_add_tail(&peer_req->w.list, &device->read_ee); |
0500813f | 392 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 | 393 | |
b30ab791 AG |
394 | atomic_add(size >> 9, &device->rs_sect_ev); |
395 | if (drbd_submit_peer_request(device, peer_req, READ, DRBD_FAULT_RS_RD) == 0) | |
80a40e43 | 396 | return 0; |
b411b363 | 397 | |
10f6d992 LE |
398 | /* If it failed because of ENOMEM, retry should help. If it failed |
399 | * because bio_add_page failed (probably broken lower level driver), | |
400 | * retry may or may not help. | |
401 | * If it does not, you may need to force disconnect. */ | |
0500813f | 402 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 403 | list_del(&peer_req->w.list); |
0500813f | 404 | spin_unlock_irq(&device->resource->req_lock); |
22cc37a9 | 405 | |
b30ab791 | 406 | drbd_free_peer_req(device, peer_req); |
80a40e43 | 407 | defer: |
b30ab791 | 408 | put_ldev(device); |
80a40e43 | 409 | return -EAGAIN; |
b411b363 PR |
410 | } |
411 | ||
99920dc5 | 412 | int w_resync_timer(struct drbd_work *w, int cancel) |
b411b363 | 413 | { |
84b8c06b AG |
414 | struct drbd_device *device = |
415 | container_of(w, struct drbd_device, resync_work); | |
416 | ||
b30ab791 | 417 | switch (device->state.conn) { |
63106d3c | 418 | case C_VERIFY_S: |
d448a2e1 | 419 | make_ov_request(device, cancel); |
63106d3c PR |
420 | break; |
421 | case C_SYNC_TARGET: | |
d448a2e1 | 422 | make_resync_request(device, cancel); |
63106d3c | 423 | break; |
b411b363 PR |
424 | } |
425 | ||
99920dc5 | 426 | return 0; |
794abb75 PR |
427 | } |
428 | ||
429 | void resync_timer_fn(unsigned long data) | |
430 | { | |
b30ab791 | 431 | struct drbd_device *device = (struct drbd_device *) data; |
794abb75 | 432 | |
15e26f6a LE |
433 | drbd_queue_work_if_unqueued( |
434 | &first_peer_device(device)->connection->sender_work, | |
435 | &device->resync_work); | |
b411b363 PR |
436 | } |
437 | ||
778f271d PR |
438 | static void fifo_set(struct fifo_buffer *fb, int value) |
439 | { | |
440 | int i; | |
441 | ||
442 | for (i = 0; i < fb->size; i++) | |
f10f2623 | 443 | fb->values[i] = value; |
778f271d PR |
444 | } |
445 | ||
446 | static int fifo_push(struct fifo_buffer *fb, int value) | |
447 | { | |
448 | int ov; | |
449 | ||
450 | ov = fb->values[fb->head_index]; | |
451 | fb->values[fb->head_index++] = value; | |
452 | ||
453 | if (fb->head_index >= fb->size) | |
454 | fb->head_index = 0; | |
455 | ||
456 | return ov; | |
457 | } | |
458 | ||
459 | static void fifo_add_val(struct fifo_buffer *fb, int value) | |
460 | { | |
461 | int i; | |
462 | ||
463 | for (i = 0; i < fb->size; i++) | |
464 | fb->values[i] += value; | |
465 | } | |
466 | ||
9958c857 PR |
467 | struct fifo_buffer *fifo_alloc(int fifo_size) |
468 | { | |
469 | struct fifo_buffer *fb; | |
470 | ||
8747d30a | 471 | fb = kzalloc(sizeof(struct fifo_buffer) + sizeof(int) * fifo_size, GFP_NOIO); |
9958c857 PR |
472 | if (!fb) |
473 | return NULL; | |
474 | ||
475 | fb->head_index = 0; | |
476 | fb->size = fifo_size; | |
477 | fb->total = 0; | |
478 | ||
479 | return fb; | |
480 | } | |
481 | ||
0e49d7b0 | 482 | static int drbd_rs_controller(struct drbd_device *device, unsigned int sect_in) |
778f271d | 483 | { |
daeda1cc | 484 | struct disk_conf *dc; |
7f34f614 | 485 | unsigned int want; /* The number of sectors we want in-flight */ |
778f271d | 486 | int req_sect; /* Number of sectors to request in this turn */ |
7f34f614 | 487 | int correction; /* Number of sectors more we need in-flight */ |
778f271d PR |
488 | int cps; /* correction per invocation of drbd_rs_controller() */ |
489 | int steps; /* Number of time steps to plan ahead */ | |
490 | int curr_corr; | |
491 | int max_sect; | |
813472ce | 492 | struct fifo_buffer *plan; |
778f271d | 493 | |
b30ab791 AG |
494 | dc = rcu_dereference(device->ldev->disk_conf); |
495 | plan = rcu_dereference(device->rs_plan_s); | |
778f271d | 496 | |
813472ce | 497 | steps = plan->size; /* (dc->c_plan_ahead * 10 * SLEEP_TIME) / HZ; */ |
778f271d | 498 | |
b30ab791 | 499 | if (device->rs_in_flight + sect_in == 0) { /* At start of resync */ |
daeda1cc | 500 | want = ((dc->resync_rate * 2 * SLEEP_TIME) / HZ) * steps; |
778f271d | 501 | } else { /* normal path */ |
daeda1cc PR |
502 | want = dc->c_fill_target ? dc->c_fill_target : |
503 | sect_in * dc->c_delay_target * HZ / (SLEEP_TIME * 10); | |
778f271d PR |
504 | } |
505 | ||
b30ab791 | 506 | correction = want - device->rs_in_flight - plan->total; |
778f271d PR |
507 | |
508 | /* Plan ahead */ | |
509 | cps = correction / steps; | |
813472ce PR |
510 | fifo_add_val(plan, cps); |
511 | plan->total += cps * steps; | |
778f271d PR |
512 | |
513 | /* What we do in this step */ | |
813472ce PR |
514 | curr_corr = fifo_push(plan, 0); |
515 | plan->total -= curr_corr; | |
778f271d PR |
516 | |
517 | req_sect = sect_in + curr_corr; | |
518 | if (req_sect < 0) | |
519 | req_sect = 0; | |
520 | ||
daeda1cc | 521 | max_sect = (dc->c_max_rate * 2 * SLEEP_TIME) / HZ; |
778f271d PR |
522 | if (req_sect > max_sect) |
523 | req_sect = max_sect; | |
524 | ||
525 | /* | |
d0180171 | 526 | drbd_warn(device, "si=%u if=%d wa=%u co=%d st=%d cps=%d pl=%d cc=%d rs=%d\n", |
b30ab791 AG |
527 | sect_in, device->rs_in_flight, want, correction, |
528 | steps, cps, device->rs_planed, curr_corr, req_sect); | |
778f271d PR |
529 | */ |
530 | ||
531 | return req_sect; | |
532 | } | |
533 | ||
b30ab791 | 534 | static int drbd_rs_number_requests(struct drbd_device *device) |
e65f440d | 535 | { |
0e49d7b0 LE |
536 | unsigned int sect_in; /* Number of sectors that came in since the last turn */ |
537 | int number, mxb; | |
538 | ||
539 | sect_in = atomic_xchg(&device->rs_sect_in, 0); | |
540 | device->rs_in_flight -= sect_in; | |
813472ce PR |
541 | |
542 | rcu_read_lock(); | |
0e49d7b0 | 543 | mxb = drbd_get_max_buffers(device) / 2; |
b30ab791 | 544 | if (rcu_dereference(device->rs_plan_s)->size) { |
0e49d7b0 | 545 | number = drbd_rs_controller(device, sect_in) >> (BM_BLOCK_SHIFT - 9); |
b30ab791 | 546 | device->c_sync_rate = number * HZ * (BM_BLOCK_SIZE / 1024) / SLEEP_TIME; |
e65f440d | 547 | } else { |
b30ab791 AG |
548 | device->c_sync_rate = rcu_dereference(device->ldev->disk_conf)->resync_rate; |
549 | number = SLEEP_TIME * device->c_sync_rate / ((BM_BLOCK_SIZE / 1024) * HZ); | |
e65f440d | 550 | } |
813472ce | 551 | rcu_read_unlock(); |
e65f440d | 552 | |
0e49d7b0 LE |
553 | /* Don't have more than "max-buffers"/2 in-flight. |
554 | * Otherwise we may cause the remote site to stall on drbd_alloc_pages(), | |
555 | * potentially causing a distributed deadlock on congestion during | |
556 | * online-verify or (checksum-based) resync, if max-buffers, | |
557 | * socket buffer sizes and resync rate settings are mis-configured. */ | |
7f34f614 LE |
558 | |
559 | /* note that "number" is in units of "BM_BLOCK_SIZE" (which is 4k), | |
560 | * mxb (as used here, and in drbd_alloc_pages on the peer) is | |
561 | * "number of pages" (typically also 4k), | |
562 | * but "rs_in_flight" is in "sectors" (512 Byte). */ | |
563 | if (mxb - device->rs_in_flight/8 < number) | |
564 | number = mxb - device->rs_in_flight/8; | |
0e49d7b0 | 565 | |
e65f440d LE |
566 | return number; |
567 | } | |
568 | ||
44a4d551 | 569 | static int make_resync_request(struct drbd_device *const device, int cancel) |
b411b363 | 570 | { |
44a4d551 LE |
571 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
572 | struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL; | |
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; |
506afb62 | 578 | int align, requeue = 0; |
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); | |
0e49d7b0 | 601 | if (number <= 0) |
0f0601f4 | 602 | goto requeue; |
b411b363 | 603 | |
b411b363 | 604 | for (i = 0; i < number; i++) { |
506afb62 LE |
605 | /* Stop generating RS requests when half of the send buffer is filled, |
606 | * but notify TCP that we'd like to have more space. */ | |
44a4d551 LE |
607 | mutex_lock(&connection->data.mutex); |
608 | if (connection->data.socket) { | |
506afb62 LE |
609 | struct sock *sk = connection->data.socket->sk; |
610 | int queued = sk->sk_wmem_queued; | |
611 | int sndbuf = sk->sk_sndbuf; | |
612 | if (queued > sndbuf / 2) { | |
613 | requeue = 1; | |
614 | if (sk->sk_socket) | |
615 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
616 | } | |
617 | } else | |
618 | requeue = 1; | |
44a4d551 | 619 | mutex_unlock(&connection->data.mutex); |
506afb62 | 620 | if (requeue) |
b411b363 PR |
621 | goto requeue; |
622 | ||
623 | next_sector: | |
624 | size = BM_BLOCK_SIZE; | |
b30ab791 | 625 | bit = drbd_bm_find_next(device, device->bm_resync_fo); |
b411b363 | 626 | |
4b0715f0 | 627 | if (bit == DRBD_END_OF_BITMAP) { |
b30ab791 AG |
628 | device->bm_resync_fo = drbd_bm_bits(device); |
629 | put_ldev(device); | |
99920dc5 | 630 | return 0; |
b411b363 PR |
631 | } |
632 | ||
633 | sector = BM_BIT_TO_SECT(bit); | |
634 | ||
ad3fee79 | 635 | if (drbd_try_rs_begin_io(device, sector)) { |
b30ab791 | 636 | device->bm_resync_fo = bit; |
b411b363 PR |
637 | goto requeue; |
638 | } | |
b30ab791 | 639 | device->bm_resync_fo = bit + 1; |
b411b363 | 640 | |
b30ab791 AG |
641 | if (unlikely(drbd_bm_test_bit(device, bit) == 0)) { |
642 | drbd_rs_complete_io(device, sector); | |
b411b363 PR |
643 | goto next_sector; |
644 | } | |
645 | ||
1816a2b4 | 646 | #if DRBD_MAX_BIO_SIZE > BM_BLOCK_SIZE |
b411b363 PR |
647 | /* try to find some adjacent bits. |
648 | * we stop if we have already the maximum req size. | |
649 | * | |
650 | * Additionally always align bigger requests, in order to | |
651 | * be prepared for all stripe sizes of software RAIDs. | |
b411b363 PR |
652 | */ |
653 | align = 1; | |
d207450c | 654 | rollback_i = i; |
6377b923 | 655 | while (i < number) { |
1816a2b4 | 656 | if (size + BM_BLOCK_SIZE > max_bio_size) |
b411b363 PR |
657 | break; |
658 | ||
659 | /* Be always aligned */ | |
660 | if (sector & ((1<<(align+3))-1)) | |
661 | break; | |
662 | ||
663 | /* do not cross extent boundaries */ | |
664 | if (((bit+1) & BM_BLOCKS_PER_BM_EXT_MASK) == 0) | |
665 | break; | |
666 | /* now, is it actually dirty, after all? | |
667 | * caution, drbd_bm_test_bit is tri-state for some | |
668 | * obscure reason; ( b == 0 ) would get the out-of-band | |
669 | * only accidentally right because of the "oddly sized" | |
670 | * adjustment below */ | |
b30ab791 | 671 | if (drbd_bm_test_bit(device, bit+1) != 1) |
b411b363 PR |
672 | break; |
673 | bit++; | |
674 | size += BM_BLOCK_SIZE; | |
675 | if ((BM_BLOCK_SIZE << align) <= size) | |
676 | align++; | |
677 | i++; | |
678 | } | |
679 | /* if we merged some, | |
680 | * reset the offset to start the next drbd_bm_find_next from */ | |
681 | if (size > BM_BLOCK_SIZE) | |
b30ab791 | 682 | device->bm_resync_fo = bit + 1; |
b411b363 PR |
683 | #endif |
684 | ||
685 | /* adjust very last sectors, in case we are oddly sized */ | |
686 | if (sector + (size>>9) > capacity) | |
687 | size = (capacity-sector)<<9; | |
aaaba345 LE |
688 | |
689 | if (device->use_csums) { | |
44a4d551 | 690 | switch (read_for_csum(peer_device, sector, size)) { |
80a40e43 | 691 | case -EIO: /* Disk failure */ |
b30ab791 | 692 | put_ldev(device); |
99920dc5 | 693 | return -EIO; |
80a40e43 | 694 | case -EAGAIN: /* allocation failed, or ldev busy */ |
b30ab791 AG |
695 | drbd_rs_complete_io(device, sector); |
696 | device->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
d207450c | 697 | i = rollback_i; |
b411b363 | 698 | goto requeue; |
80a40e43 LE |
699 | case 0: |
700 | /* everything ok */ | |
701 | break; | |
702 | default: | |
703 | BUG(); | |
b411b363 PR |
704 | } |
705 | } else { | |
99920dc5 AG |
706 | int err; |
707 | ||
b30ab791 | 708 | inc_rs_pending(device); |
44a4d551 | 709 | err = drbd_send_drequest(peer_device, P_RS_DATA_REQUEST, |
99920dc5 AG |
710 | sector, size, ID_SYNCER); |
711 | if (err) { | |
d0180171 | 712 | drbd_err(device, "drbd_send_drequest() failed, aborting...\n"); |
b30ab791 AG |
713 | dec_rs_pending(device); |
714 | put_ldev(device); | |
99920dc5 | 715 | return err; |
b411b363 PR |
716 | } |
717 | } | |
718 | } | |
719 | ||
b30ab791 | 720 | if (device->bm_resync_fo >= drbd_bm_bits(device)) { |
b411b363 PR |
721 | /* last syncer _request_ was sent, |
722 | * but the P_RS_DATA_REPLY not yet received. sync will end (and | |
723 | * next sync group will resume), as soon as we receive the last | |
724 | * resync data block, and the last bit is cleared. | |
725 | * until then resync "work" is "inactive" ... | |
726 | */ | |
b30ab791 | 727 | put_ldev(device); |
99920dc5 | 728 | return 0; |
b411b363 PR |
729 | } |
730 | ||
731 | requeue: | |
b30ab791 AG |
732 | device->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
733 | mod_timer(&device->resync_timer, jiffies + SLEEP_TIME); | |
734 | put_ldev(device); | |
99920dc5 | 735 | return 0; |
b411b363 PR |
736 | } |
737 | ||
d448a2e1 | 738 | static int make_ov_request(struct drbd_device *device, int cancel) |
b411b363 PR |
739 | { |
740 | int number, i, size; | |
741 | sector_t sector; | |
b30ab791 | 742 | const sector_t capacity = drbd_get_capacity(device->this_bdev); |
58ffa580 | 743 | bool stop_sector_reached = false; |
b411b363 PR |
744 | |
745 | if (unlikely(cancel)) | |
746 | return 1; | |
747 | ||
b30ab791 | 748 | number = drbd_rs_number_requests(device); |
b411b363 | 749 | |
b30ab791 | 750 | sector = device->ov_position; |
b411b363 | 751 | for (i = 0; i < number; i++) { |
58ffa580 | 752 | if (sector >= capacity) |
b411b363 | 753 | return 1; |
58ffa580 LE |
754 | |
755 | /* We check for "finished" only in the reply path: | |
756 | * w_e_end_ov_reply(). | |
757 | * We need to send at least one request out. */ | |
758 | stop_sector_reached = i > 0 | |
b30ab791 AG |
759 | && verify_can_do_stop_sector(device) |
760 | && sector >= device->ov_stop_sector; | |
58ffa580 LE |
761 | if (stop_sector_reached) |
762 | break; | |
b411b363 PR |
763 | |
764 | size = BM_BLOCK_SIZE; | |
765 | ||
ad3fee79 | 766 | if (drbd_try_rs_begin_io(device, sector)) { |
b30ab791 | 767 | device->ov_position = sector; |
b411b363 PR |
768 | goto requeue; |
769 | } | |
770 | ||
771 | if (sector + (size>>9) > capacity) | |
772 | size = (capacity-sector)<<9; | |
773 | ||
b30ab791 | 774 | inc_rs_pending(device); |
69a22773 | 775 | if (drbd_send_ov_request(first_peer_device(device), sector, size)) { |
b30ab791 | 776 | dec_rs_pending(device); |
b411b363 PR |
777 | return 0; |
778 | } | |
779 | sector += BM_SECT_PER_BIT; | |
780 | } | |
b30ab791 | 781 | device->ov_position = sector; |
b411b363 PR |
782 | |
783 | requeue: | |
b30ab791 | 784 | device->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
58ffa580 | 785 | if (i == 0 || !stop_sector_reached) |
b30ab791 | 786 | mod_timer(&device->resync_timer, jiffies + SLEEP_TIME); |
b411b363 PR |
787 | return 1; |
788 | } | |
789 | ||
99920dc5 | 790 | int w_ov_finished(struct drbd_work *w, int cancel) |
b411b363 | 791 | { |
84b8c06b AG |
792 | struct drbd_device_work *dw = |
793 | container_of(w, struct drbd_device_work, w); | |
794 | struct drbd_device *device = dw->device; | |
795 | kfree(dw); | |
b30ab791 AG |
796 | ov_out_of_sync_print(device); |
797 | drbd_resync_finished(device); | |
b411b363 | 798 | |
99920dc5 | 799 | return 0; |
b411b363 PR |
800 | } |
801 | ||
99920dc5 | 802 | static int w_resync_finished(struct drbd_work *w, int cancel) |
b411b363 | 803 | { |
84b8c06b AG |
804 | struct drbd_device_work *dw = |
805 | container_of(w, struct drbd_device_work, w); | |
806 | struct drbd_device *device = dw->device; | |
807 | kfree(dw); | |
b411b363 | 808 | |
b30ab791 | 809 | drbd_resync_finished(device); |
b411b363 | 810 | |
99920dc5 | 811 | return 0; |
b411b363 PR |
812 | } |
813 | ||
b30ab791 | 814 | static void ping_peer(struct drbd_device *device) |
af85e8e8 | 815 | { |
a6b32bc3 | 816 | struct drbd_connection *connection = first_peer_device(device)->connection; |
2a67d8b9 | 817 | |
bde89a9e AG |
818 | clear_bit(GOT_PING_ACK, &connection->flags); |
819 | request_ping(connection); | |
820 | wait_event(connection->ping_wait, | |
821 | test_bit(GOT_PING_ACK, &connection->flags) || device->state.conn < C_CONNECTED); | |
af85e8e8 LE |
822 | } |
823 | ||
b30ab791 | 824 | int drbd_resync_finished(struct drbd_device *device) |
b411b363 PR |
825 | { |
826 | unsigned long db, dt, dbdt; | |
827 | unsigned long n_oos; | |
828 | union drbd_state os, ns; | |
84b8c06b | 829 | struct drbd_device_work *dw; |
b411b363 | 830 | char *khelper_cmd = NULL; |
26525618 | 831 | int verify_done = 0; |
b411b363 PR |
832 | |
833 | /* Remove all elements from the resync LRU. Since future actions | |
834 | * might set bits in the (main) bitmap, then the entries in the | |
835 | * resync LRU would be wrong. */ | |
b30ab791 | 836 | if (drbd_rs_del_all(device)) { |
b411b363 PR |
837 | /* In case this is not possible now, most probably because |
838 | * there are P_RS_DATA_REPLY Packets lingering on the worker's | |
839 | * queue (or even the read operations for those packets | |
840 | * is not finished by now). Retry in 100ms. */ | |
841 | ||
20ee6390 | 842 | schedule_timeout_interruptible(HZ / 10); |
84b8c06b AG |
843 | dw = kmalloc(sizeof(struct drbd_device_work), GFP_ATOMIC); |
844 | if (dw) { | |
845 | dw->w.cb = w_resync_finished; | |
846 | dw->device = device; | |
847 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, | |
848 | &dw->w); | |
b411b363 PR |
849 | return 1; |
850 | } | |
84b8c06b | 851 | drbd_err(device, "Warn failed to drbd_rs_del_all() and to kmalloc(dw).\n"); |
b411b363 PR |
852 | } |
853 | ||
b30ab791 | 854 | dt = (jiffies - device->rs_start - device->rs_paused) / HZ; |
b411b363 PR |
855 | if (dt <= 0) |
856 | dt = 1; | |
84b8c06b | 857 | |
b30ab791 | 858 | db = device->rs_total; |
58ffa580 | 859 | /* adjust for verify start and stop sectors, respective reached position */ |
b30ab791 AG |
860 | if (device->state.conn == C_VERIFY_S || device->state.conn == C_VERIFY_T) |
861 | db -= device->ov_left; | |
58ffa580 | 862 | |
b411b363 | 863 | dbdt = Bit2KB(db/dt); |
b30ab791 | 864 | device->rs_paused /= HZ; |
b411b363 | 865 | |
b30ab791 | 866 | if (!get_ldev(device)) |
b411b363 PR |
867 | goto out; |
868 | ||
b30ab791 | 869 | ping_peer(device); |
af85e8e8 | 870 | |
0500813f | 871 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 872 | os = drbd_read_state(device); |
b411b363 | 873 | |
26525618 LE |
874 | verify_done = (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T); |
875 | ||
b411b363 PR |
876 | /* This protects us against multiple calls (that can happen in the presence |
877 | of application IO), and against connectivity loss just before we arrive here. */ | |
878 | if (os.conn <= C_CONNECTED) | |
879 | goto out_unlock; | |
880 | ||
881 | ns = os; | |
882 | ns.conn = C_CONNECTED; | |
883 | ||
d0180171 | 884 | drbd_info(device, "%s done (total %lu sec; paused %lu sec; %lu K/sec)\n", |
58ffa580 | 885 | verify_done ? "Online verify" : "Resync", |
b30ab791 | 886 | dt + device->rs_paused, device->rs_paused, dbdt); |
b411b363 | 887 | |
b30ab791 | 888 | n_oos = drbd_bm_total_weight(device); |
b411b363 PR |
889 | |
890 | if (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) { | |
891 | if (n_oos) { | |
d0180171 | 892 | drbd_alert(device, "Online verify found %lu %dk block out of sync!\n", |
b411b363 PR |
893 | n_oos, Bit2KB(1)); |
894 | khelper_cmd = "out-of-sync"; | |
895 | } | |
896 | } else { | |
0b0ba1ef | 897 | D_ASSERT(device, (n_oos - device->rs_failed) == 0); |
b411b363 PR |
898 | |
899 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) | |
900 | khelper_cmd = "after-resync-target"; | |
901 | ||
aaaba345 | 902 | if (device->use_csums && device->rs_total) { |
b30ab791 AG |
903 | const unsigned long s = device->rs_same_csum; |
904 | const unsigned long t = device->rs_total; | |
b411b363 PR |
905 | const int ratio = |
906 | (t == 0) ? 0 : | |
907 | (t < 100000) ? ((s*100)/t) : (s/(t/100)); | |
d0180171 | 908 | drbd_info(device, "%u %% had equal checksums, eliminated: %luK; " |
b411b363 PR |
909 | "transferred %luK total %luK\n", |
910 | ratio, | |
b30ab791 AG |
911 | Bit2KB(device->rs_same_csum), |
912 | Bit2KB(device->rs_total - device->rs_same_csum), | |
913 | Bit2KB(device->rs_total)); | |
b411b363 PR |
914 | } |
915 | } | |
916 | ||
b30ab791 | 917 | if (device->rs_failed) { |
d0180171 | 918 | drbd_info(device, " %lu failed blocks\n", device->rs_failed); |
b411b363 PR |
919 | |
920 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
921 | ns.disk = D_INCONSISTENT; | |
922 | ns.pdsk = D_UP_TO_DATE; | |
923 | } else { | |
924 | ns.disk = D_UP_TO_DATE; | |
925 | ns.pdsk = D_INCONSISTENT; | |
926 | } | |
927 | } else { | |
928 | ns.disk = D_UP_TO_DATE; | |
929 | ns.pdsk = D_UP_TO_DATE; | |
930 | ||
931 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
b30ab791 | 932 | if (device->p_uuid) { |
b411b363 PR |
933 | int i; |
934 | for (i = UI_BITMAP ; i <= UI_HISTORY_END ; i++) | |
b30ab791 AG |
935 | _drbd_uuid_set(device, i, device->p_uuid[i]); |
936 | drbd_uuid_set(device, UI_BITMAP, device->ldev->md.uuid[UI_CURRENT]); | |
937 | _drbd_uuid_set(device, UI_CURRENT, device->p_uuid[UI_CURRENT]); | |
b411b363 | 938 | } else { |
d0180171 | 939 | drbd_err(device, "device->p_uuid is NULL! BUG\n"); |
b411b363 PR |
940 | } |
941 | } | |
942 | ||
62b0da3a LE |
943 | if (!(os.conn == C_VERIFY_S || os.conn == C_VERIFY_T)) { |
944 | /* for verify runs, we don't update uuids here, | |
945 | * so there would be nothing to report. */ | |
b30ab791 AG |
946 | drbd_uuid_set_bm(device, 0UL); |
947 | drbd_print_uuids(device, "updated UUIDs"); | |
948 | if (device->p_uuid) { | |
62b0da3a LE |
949 | /* Now the two UUID sets are equal, update what we |
950 | * know of the peer. */ | |
951 | int i; | |
952 | for (i = UI_CURRENT ; i <= UI_HISTORY_END ; i++) | |
b30ab791 | 953 | device->p_uuid[i] = device->ldev->md.uuid[i]; |
62b0da3a | 954 | } |
b411b363 PR |
955 | } |
956 | } | |
957 | ||
b30ab791 | 958 | _drbd_set_state(device, ns, CS_VERBOSE, NULL); |
b411b363 | 959 | out_unlock: |
0500813f | 960 | spin_unlock_irq(&device->resource->req_lock); |
b30ab791 | 961 | put_ldev(device); |
b411b363 | 962 | out: |
b30ab791 AG |
963 | device->rs_total = 0; |
964 | device->rs_failed = 0; | |
965 | device->rs_paused = 0; | |
58ffa580 LE |
966 | |
967 | /* reset start sector, if we reached end of device */ | |
b30ab791 AG |
968 | if (verify_done && device->ov_left == 0) |
969 | device->ov_start_sector = 0; | |
b411b363 | 970 | |
b30ab791 | 971 | drbd_md_sync(device); |
13d42685 | 972 | |
b411b363 | 973 | if (khelper_cmd) |
b30ab791 | 974 | drbd_khelper(device, khelper_cmd); |
b411b363 PR |
975 | |
976 | return 1; | |
977 | } | |
978 | ||
979 | /* helper */ | |
b30ab791 | 980 | static void move_to_net_ee_or_free(struct drbd_device *device, struct drbd_peer_request *peer_req) |
b411b363 | 981 | { |
045417f7 | 982 | if (drbd_peer_req_has_active_page(peer_req)) { |
b411b363 | 983 | /* This might happen if sendpage() has not finished */ |
db830c46 | 984 | int i = (peer_req->i.size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b30ab791 AG |
985 | atomic_add(i, &device->pp_in_use_by_net); |
986 | atomic_sub(i, &device->pp_in_use); | |
0500813f | 987 | spin_lock_irq(&device->resource->req_lock); |
a8cd15ba | 988 | list_add_tail(&peer_req->w.list, &device->net_ee); |
0500813f | 989 | spin_unlock_irq(&device->resource->req_lock); |
435f0740 | 990 | wake_up(&drbd_pp_wait); |
b411b363 | 991 | } else |
b30ab791 | 992 | drbd_free_peer_req(device, peer_req); |
b411b363 PR |
993 | } |
994 | ||
995 | /** | |
996 | * w_e_end_data_req() - Worker callback, to send a P_DATA_REPLY packet in response to a P_DATA_REQUEST | |
b30ab791 | 997 | * @device: DRBD device. |
b411b363 PR |
998 | * @w: work object. |
999 | * @cancel: The connection will be closed anyways | |
1000 | */ | |
99920dc5 | 1001 | int w_e_end_data_req(struct drbd_work *w, int cancel) |
b411b363 | 1002 | { |
a8cd15ba | 1003 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1004 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1005 | struct drbd_device *device = peer_device->device; | |
99920dc5 | 1006 | int err; |
b411b363 PR |
1007 | |
1008 | if (unlikely(cancel)) { | |
b30ab791 AG |
1009 | drbd_free_peer_req(device, peer_req); |
1010 | dec_unacked(device); | |
99920dc5 | 1011 | return 0; |
b411b363 PR |
1012 | } |
1013 | ||
db830c46 | 1014 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
6780139c | 1015 | err = drbd_send_block(peer_device, P_DATA_REPLY, peer_req); |
b411b363 PR |
1016 | } else { |
1017 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1018 | drbd_err(device, "Sending NegDReply. sector=%llus.\n", |
db830c46 | 1019 | (unsigned long long)peer_req->i.sector); |
b411b363 | 1020 | |
6780139c | 1021 | err = drbd_send_ack(peer_device, P_NEG_DREPLY, peer_req); |
b411b363 PR |
1022 | } |
1023 | ||
b30ab791 | 1024 | dec_unacked(device); |
b411b363 | 1025 | |
b30ab791 | 1026 | move_to_net_ee_or_free(device, peer_req); |
b411b363 | 1027 | |
99920dc5 | 1028 | if (unlikely(err)) |
d0180171 | 1029 | drbd_err(device, "drbd_send_block() failed\n"); |
99920dc5 | 1030 | return err; |
b411b363 PR |
1031 | } |
1032 | ||
1033 | /** | |
a209b4ae | 1034 | * 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 |
1035 | * @w: work object. |
1036 | * @cancel: The connection will be closed anyways | |
1037 | */ | |
99920dc5 | 1038 | int w_e_end_rsdata_req(struct drbd_work *w, int cancel) |
b411b363 | 1039 | { |
a8cd15ba | 1040 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1041 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1042 | struct drbd_device *device = peer_device->device; | |
99920dc5 | 1043 | int err; |
b411b363 PR |
1044 | |
1045 | if (unlikely(cancel)) { | |
b30ab791 AG |
1046 | drbd_free_peer_req(device, peer_req); |
1047 | dec_unacked(device); | |
99920dc5 | 1048 | return 0; |
b411b363 PR |
1049 | } |
1050 | ||
b30ab791 AG |
1051 | if (get_ldev_if_state(device, D_FAILED)) { |
1052 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1053 | put_ldev(device); | |
b411b363 PR |
1054 | } |
1055 | ||
b30ab791 | 1056 | if (device->state.conn == C_AHEAD) { |
6780139c | 1057 | err = drbd_send_ack(peer_device, P_RS_CANCEL, peer_req); |
db830c46 | 1058 | } else if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b30ab791 AG |
1059 | if (likely(device->state.pdsk >= D_INCONSISTENT)) { |
1060 | inc_rs_pending(device); | |
6780139c | 1061 | err = drbd_send_block(peer_device, P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1062 | } else { |
1063 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1064 | drbd_err(device, "Not sending RSDataReply, " |
b411b363 | 1065 | "partner DISKLESS!\n"); |
99920dc5 | 1066 | err = 0; |
b411b363 PR |
1067 | } |
1068 | } else { | |
1069 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1070 | drbd_err(device, "Sending NegRSDReply. sector %llus.\n", |
db830c46 | 1071 | (unsigned long long)peer_req->i.sector); |
b411b363 | 1072 | |
6780139c | 1073 | err = drbd_send_ack(peer_device, P_NEG_RS_DREPLY, peer_req); |
b411b363 PR |
1074 | |
1075 | /* update resync data with failure */ | |
b30ab791 | 1076 | drbd_rs_failed_io(device, peer_req->i.sector, peer_req->i.size); |
b411b363 PR |
1077 | } |
1078 | ||
b30ab791 | 1079 | dec_unacked(device); |
b411b363 | 1080 | |
b30ab791 | 1081 | move_to_net_ee_or_free(device, peer_req); |
b411b363 | 1082 | |
99920dc5 | 1083 | if (unlikely(err)) |
d0180171 | 1084 | drbd_err(device, "drbd_send_block() failed\n"); |
99920dc5 | 1085 | return err; |
b411b363 PR |
1086 | } |
1087 | ||
99920dc5 | 1088 | int w_e_end_csum_rs_req(struct drbd_work *w, int cancel) |
b411b363 | 1089 | { |
a8cd15ba | 1090 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1091 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1092 | struct drbd_device *device = peer_device->device; | |
b411b363 PR |
1093 | struct digest_info *di; |
1094 | int digest_size; | |
1095 | void *digest = NULL; | |
99920dc5 | 1096 | int err, eq = 0; |
b411b363 PR |
1097 | |
1098 | if (unlikely(cancel)) { | |
b30ab791 AG |
1099 | drbd_free_peer_req(device, peer_req); |
1100 | dec_unacked(device); | |
99920dc5 | 1101 | return 0; |
b411b363 PR |
1102 | } |
1103 | ||
b30ab791 AG |
1104 | if (get_ldev(device)) { |
1105 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1106 | put_ldev(device); | |
1d53f09e | 1107 | } |
b411b363 | 1108 | |
db830c46 | 1109 | di = peer_req->digest; |
b411b363 | 1110 | |
db830c46 | 1111 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1112 | /* quick hack to try to avoid a race against reconfiguration. |
1113 | * a real fix would be much more involved, | |
1114 | * introducing more locking mechanisms */ | |
6780139c AG |
1115 | if (peer_device->connection->csums_tfm) { |
1116 | digest_size = crypto_hash_digestsize(peer_device->connection->csums_tfm); | |
0b0ba1ef | 1117 | D_ASSERT(device, digest_size == di->digest_size); |
b411b363 PR |
1118 | digest = kmalloc(digest_size, GFP_NOIO); |
1119 | } | |
1120 | if (digest) { | |
6780139c | 1121 | drbd_csum_ee(peer_device->connection->csums_tfm, peer_req, digest); |
b411b363 PR |
1122 | eq = !memcmp(digest, di->digest, digest_size); |
1123 | kfree(digest); | |
1124 | } | |
1125 | ||
1126 | if (eq) { | |
b30ab791 | 1127 | drbd_set_in_sync(device, peer_req->i.sector, peer_req->i.size); |
676396d5 | 1128 | /* rs_same_csums unit is BM_BLOCK_SIZE */ |
b30ab791 | 1129 | device->rs_same_csum += peer_req->i.size >> BM_BLOCK_SHIFT; |
6780139c | 1130 | err = drbd_send_ack(peer_device, P_RS_IS_IN_SYNC, peer_req); |
b411b363 | 1131 | } else { |
b30ab791 | 1132 | inc_rs_pending(device); |
db830c46 AG |
1133 | peer_req->block_id = ID_SYNCER; /* By setting block_id, digest pointer becomes invalid! */ |
1134 | peer_req->flags &= ~EE_HAS_DIGEST; /* This peer request no longer has a digest pointer */ | |
204bba99 | 1135 | kfree(di); |
6780139c | 1136 | err = drbd_send_block(peer_device, P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1137 | } |
1138 | } else { | |
6780139c | 1139 | err = drbd_send_ack(peer_device, P_NEG_RS_DREPLY, peer_req); |
b411b363 | 1140 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 1141 | drbd_err(device, "Sending NegDReply. I guess it gets messy.\n"); |
b411b363 PR |
1142 | } |
1143 | ||
b30ab791 AG |
1144 | dec_unacked(device); |
1145 | move_to_net_ee_or_free(device, peer_req); | |
b411b363 | 1146 | |
99920dc5 | 1147 | if (unlikely(err)) |
d0180171 | 1148 | drbd_err(device, "drbd_send_block/ack() failed\n"); |
99920dc5 | 1149 | return err; |
b411b363 PR |
1150 | } |
1151 | ||
99920dc5 | 1152 | int w_e_end_ov_req(struct drbd_work *w, int cancel) |
b411b363 | 1153 | { |
a8cd15ba | 1154 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1155 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1156 | struct drbd_device *device = peer_device->device; | |
db830c46 AG |
1157 | sector_t sector = peer_req->i.sector; |
1158 | unsigned int size = peer_req->i.size; | |
b411b363 PR |
1159 | int digest_size; |
1160 | void *digest; | |
99920dc5 | 1161 | int err = 0; |
b411b363 PR |
1162 | |
1163 | if (unlikely(cancel)) | |
1164 | goto out; | |
1165 | ||
6780139c | 1166 | digest_size = crypto_hash_digestsize(peer_device->connection->verify_tfm); |
b411b363 | 1167 | digest = kmalloc(digest_size, GFP_NOIO); |
8f21420e | 1168 | if (!digest) { |
99920dc5 | 1169 | err = 1; /* terminate the connection in case the allocation failed */ |
8f21420e | 1170 | goto out; |
b411b363 PR |
1171 | } |
1172 | ||
db830c46 | 1173 | if (likely(!(peer_req->flags & EE_WAS_ERROR))) |
6780139c | 1174 | drbd_csum_ee(peer_device->connection->verify_tfm, peer_req, digest); |
8f21420e PR |
1175 | else |
1176 | memset(digest, 0, digest_size); | |
1177 | ||
53ea4331 LE |
1178 | /* Free e and pages before send. |
1179 | * In case we block on congestion, we could otherwise run into | |
1180 | * some distributed deadlock, if the other side blocks on | |
1181 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1182 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 1183 | drbd_free_peer_req(device, peer_req); |
db830c46 | 1184 | peer_req = NULL; |
b30ab791 | 1185 | inc_rs_pending(device); |
6780139c | 1186 | err = drbd_send_drequest_csum(peer_device, sector, size, digest, digest_size, P_OV_REPLY); |
99920dc5 | 1187 | if (err) |
b30ab791 | 1188 | dec_rs_pending(device); |
8f21420e PR |
1189 | kfree(digest); |
1190 | ||
b411b363 | 1191 | out: |
db830c46 | 1192 | if (peer_req) |
b30ab791 AG |
1193 | drbd_free_peer_req(device, peer_req); |
1194 | dec_unacked(device); | |
99920dc5 | 1195 | return err; |
b411b363 PR |
1196 | } |
1197 | ||
b30ab791 | 1198 | void drbd_ov_out_of_sync_found(struct drbd_device *device, sector_t sector, int size) |
b411b363 | 1199 | { |
b30ab791 AG |
1200 | if (device->ov_last_oos_start + device->ov_last_oos_size == sector) { |
1201 | device->ov_last_oos_size += size>>9; | |
b411b363 | 1202 | } else { |
b30ab791 AG |
1203 | device->ov_last_oos_start = sector; |
1204 | device->ov_last_oos_size = size>>9; | |
b411b363 | 1205 | } |
b30ab791 | 1206 | drbd_set_out_of_sync(device, sector, size); |
b411b363 PR |
1207 | } |
1208 | ||
99920dc5 | 1209 | int w_e_end_ov_reply(struct drbd_work *w, int cancel) |
b411b363 | 1210 | { |
a8cd15ba | 1211 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
6780139c AG |
1212 | struct drbd_peer_device *peer_device = peer_req->peer_device; |
1213 | struct drbd_device *device = peer_device->device; | |
b411b363 | 1214 | struct digest_info *di; |
b411b363 | 1215 | void *digest; |
db830c46 AG |
1216 | sector_t sector = peer_req->i.sector; |
1217 | unsigned int size = peer_req->i.size; | |
53ea4331 | 1218 | int digest_size; |
99920dc5 | 1219 | int err, eq = 0; |
58ffa580 | 1220 | bool stop_sector_reached = false; |
b411b363 PR |
1221 | |
1222 | if (unlikely(cancel)) { | |
b30ab791 AG |
1223 | drbd_free_peer_req(device, peer_req); |
1224 | dec_unacked(device); | |
99920dc5 | 1225 | return 0; |
b411b363 PR |
1226 | } |
1227 | ||
1228 | /* after "cancel", because after drbd_disconnect/drbd_rs_cancel_all | |
1229 | * the resync lru has been cleaned up already */ | |
b30ab791 AG |
1230 | if (get_ldev(device)) { |
1231 | drbd_rs_complete_io(device, peer_req->i.sector); | |
1232 | put_ldev(device); | |
1d53f09e | 1233 | } |
b411b363 | 1234 | |
db830c46 | 1235 | di = peer_req->digest; |
b411b363 | 1236 | |
db830c46 | 1237 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
6780139c | 1238 | digest_size = crypto_hash_digestsize(peer_device->connection->verify_tfm); |
b411b363 PR |
1239 | digest = kmalloc(digest_size, GFP_NOIO); |
1240 | if (digest) { | |
6780139c | 1241 | drbd_csum_ee(peer_device->connection->verify_tfm, peer_req, digest); |
b411b363 | 1242 | |
0b0ba1ef | 1243 | D_ASSERT(device, digest_size == di->digest_size); |
b411b363 PR |
1244 | eq = !memcmp(digest, di->digest, digest_size); |
1245 | kfree(digest); | |
1246 | } | |
b411b363 PR |
1247 | } |
1248 | ||
9676c760 LE |
1249 | /* Free peer_req and pages before send. |
1250 | * In case we block on congestion, we could otherwise run into | |
1251 | * some distributed deadlock, if the other side blocks on | |
1252 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1253 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
b30ab791 | 1254 | drbd_free_peer_req(device, peer_req); |
b411b363 | 1255 | if (!eq) |
b30ab791 | 1256 | drbd_ov_out_of_sync_found(device, sector, size); |
b411b363 | 1257 | else |
b30ab791 | 1258 | ov_out_of_sync_print(device); |
b411b363 | 1259 | |
6780139c | 1260 | err = drbd_send_ack_ex(peer_device, P_OV_RESULT, sector, size, |
fa79abd8 | 1261 | eq ? ID_IN_SYNC : ID_OUT_OF_SYNC); |
b411b363 | 1262 | |
b30ab791 | 1263 | dec_unacked(device); |
b411b363 | 1264 | |
b30ab791 | 1265 | --device->ov_left; |
ea5442af LE |
1266 | |
1267 | /* let's advance progress step marks only for every other megabyte */ | |
b30ab791 AG |
1268 | if ((device->ov_left & 0x200) == 0x200) |
1269 | drbd_advance_rs_marks(device, device->ov_left); | |
ea5442af | 1270 | |
b30ab791 AG |
1271 | stop_sector_reached = verify_can_do_stop_sector(device) && |
1272 | (sector + (size>>9)) >= device->ov_stop_sector; | |
58ffa580 | 1273 | |
b30ab791 AG |
1274 | if (device->ov_left == 0 || stop_sector_reached) { |
1275 | ov_out_of_sync_print(device); | |
1276 | drbd_resync_finished(device); | |
b411b363 PR |
1277 | } |
1278 | ||
99920dc5 | 1279 | return err; |
b411b363 PR |
1280 | } |
1281 | ||
b6dd1a89 LE |
1282 | /* FIXME |
1283 | * We need to track the number of pending barrier acks, | |
1284 | * and to be able to wait for them. | |
1285 | * See also comment in drbd_adm_attach before drbd_suspend_io. | |
1286 | */ | |
bde89a9e | 1287 | static int drbd_send_barrier(struct drbd_connection *connection) |
b411b363 | 1288 | { |
9f5bdc33 | 1289 | struct p_barrier *p; |
b6dd1a89 | 1290 | struct drbd_socket *sock; |
b411b363 | 1291 | |
bde89a9e AG |
1292 | sock = &connection->data; |
1293 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1294 | if (!p) |
1295 | return -EIO; | |
bde89a9e | 1296 | p->barrier = connection->send.current_epoch_nr; |
b6dd1a89 | 1297 | p->pad = 0; |
bde89a9e | 1298 | connection->send.current_epoch_writes = 0; |
84d34f2f | 1299 | connection->send.last_sent_barrier_jif = jiffies; |
b6dd1a89 | 1300 | |
bde89a9e | 1301 | return conn_send_command(connection, sock, P_BARRIER, sizeof(*p), NULL, 0); |
b411b363 PR |
1302 | } |
1303 | ||
99920dc5 | 1304 | int w_send_write_hint(struct drbd_work *w, int cancel) |
b411b363 | 1305 | { |
84b8c06b AG |
1306 | struct drbd_device *device = |
1307 | container_of(w, struct drbd_device, unplug_work); | |
9f5bdc33 AG |
1308 | struct drbd_socket *sock; |
1309 | ||
b411b363 | 1310 | if (cancel) |
99920dc5 | 1311 | return 0; |
a6b32bc3 | 1312 | sock = &first_peer_device(device)->connection->data; |
69a22773 | 1313 | if (!drbd_prepare_command(first_peer_device(device), sock)) |
9f5bdc33 | 1314 | return -EIO; |
69a22773 | 1315 | return drbd_send_command(first_peer_device(device), sock, P_UNPLUG_REMOTE, 0, NULL, 0); |
b411b363 PR |
1316 | } |
1317 | ||
bde89a9e | 1318 | static void re_init_if_first_write(struct drbd_connection *connection, unsigned int epoch) |
4eb9b3cb | 1319 | { |
bde89a9e AG |
1320 | if (!connection->send.seen_any_write_yet) { |
1321 | connection->send.seen_any_write_yet = true; | |
1322 | connection->send.current_epoch_nr = epoch; | |
1323 | connection->send.current_epoch_writes = 0; | |
84d34f2f | 1324 | connection->send.last_sent_barrier_jif = jiffies; |
4eb9b3cb LE |
1325 | } |
1326 | } | |
1327 | ||
bde89a9e | 1328 | static void maybe_send_barrier(struct drbd_connection *connection, unsigned int epoch) |
4eb9b3cb LE |
1329 | { |
1330 | /* re-init if first write on this connection */ | |
bde89a9e | 1331 | if (!connection->send.seen_any_write_yet) |
4eb9b3cb | 1332 | return; |
bde89a9e AG |
1333 | if (connection->send.current_epoch_nr != epoch) { |
1334 | if (connection->send.current_epoch_writes) | |
1335 | drbd_send_barrier(connection); | |
1336 | connection->send.current_epoch_nr = epoch; | |
4eb9b3cb LE |
1337 | } |
1338 | } | |
1339 | ||
8f7bed77 | 1340 | int w_send_out_of_sync(struct drbd_work *w, int cancel) |
73a01a18 PR |
1341 | { |
1342 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1343 | struct drbd_device *device = req->device; |
44a4d551 LE |
1344 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
1345 | struct drbd_connection *const connection = peer_device->connection; | |
99920dc5 | 1346 | int err; |
73a01a18 PR |
1347 | |
1348 | if (unlikely(cancel)) { | |
8554df1c | 1349 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1350 | return 0; |
73a01a18 | 1351 | } |
e5f891b2 | 1352 | req->pre_send_jif = jiffies; |
73a01a18 | 1353 | |
bde89a9e | 1354 | /* this time, no connection->send.current_epoch_writes++; |
b6dd1a89 LE |
1355 | * If it was sent, it was the closing barrier for the last |
1356 | * replicated epoch, before we went into AHEAD mode. | |
1357 | * No more barriers will be sent, until we leave AHEAD mode again. */ | |
bde89a9e | 1358 | maybe_send_barrier(connection, req->epoch); |
b6dd1a89 | 1359 | |
44a4d551 | 1360 | err = drbd_send_out_of_sync(peer_device, req); |
8554df1c | 1361 | req_mod(req, OOS_HANDED_TO_NETWORK); |
73a01a18 | 1362 | |
99920dc5 | 1363 | return err; |
73a01a18 PR |
1364 | } |
1365 | ||
b411b363 PR |
1366 | /** |
1367 | * w_send_dblock() - Worker callback to send a P_DATA packet in order to mirror a write request | |
b411b363 PR |
1368 | * @w: work object. |
1369 | * @cancel: The connection will be closed anyways | |
1370 | */ | |
99920dc5 | 1371 | int w_send_dblock(struct drbd_work *w, int cancel) |
b411b363 PR |
1372 | { |
1373 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1374 | struct drbd_device *device = req->device; |
44a4d551 LE |
1375 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
1376 | struct drbd_connection *connection = peer_device->connection; | |
99920dc5 | 1377 | int err; |
b411b363 PR |
1378 | |
1379 | if (unlikely(cancel)) { | |
8554df1c | 1380 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1381 | return 0; |
b411b363 | 1382 | } |
e5f891b2 | 1383 | req->pre_send_jif = jiffies; |
b411b363 | 1384 | |
bde89a9e AG |
1385 | re_init_if_first_write(connection, req->epoch); |
1386 | maybe_send_barrier(connection, req->epoch); | |
1387 | connection->send.current_epoch_writes++; | |
b6dd1a89 | 1388 | |
44a4d551 | 1389 | err = drbd_send_dblock(peer_device, req); |
99920dc5 | 1390 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1391 | |
99920dc5 | 1392 | return err; |
b411b363 PR |
1393 | } |
1394 | ||
1395 | /** | |
1396 | * w_send_read_req() - Worker callback to send a read request (P_DATA_REQUEST) packet | |
b411b363 PR |
1397 | * @w: work object. |
1398 | * @cancel: The connection will be closed anyways | |
1399 | */ | |
99920dc5 | 1400 | int w_send_read_req(struct drbd_work *w, int cancel) |
b411b363 PR |
1401 | { |
1402 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1403 | struct drbd_device *device = req->device; |
44a4d551 LE |
1404 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
1405 | struct drbd_connection *connection = peer_device->connection; | |
99920dc5 | 1406 | int err; |
b411b363 PR |
1407 | |
1408 | if (unlikely(cancel)) { | |
8554df1c | 1409 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1410 | return 0; |
b411b363 | 1411 | } |
e5f891b2 | 1412 | req->pre_send_jif = jiffies; |
b411b363 | 1413 | |
b6dd1a89 LE |
1414 | /* Even read requests may close a write epoch, |
1415 | * if there was any yet. */ | |
bde89a9e | 1416 | maybe_send_barrier(connection, req->epoch); |
b6dd1a89 | 1417 | |
44a4d551 | 1418 | err = drbd_send_drequest(peer_device, P_DATA_REQUEST, req->i.sector, req->i.size, |
6c1005e7 | 1419 | (unsigned long)req); |
b411b363 | 1420 | |
99920dc5 | 1421 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1422 | |
99920dc5 | 1423 | return err; |
b411b363 PR |
1424 | } |
1425 | ||
99920dc5 | 1426 | int w_restart_disk_io(struct drbd_work *w, int cancel) |
265be2d0 PR |
1427 | { |
1428 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
84b8c06b | 1429 | struct drbd_device *device = req->device; |
265be2d0 | 1430 | |
0778286a | 1431 | if (bio_data_dir(req->master_bio) == WRITE && req->rq_state & RQ_IN_ACT_LOG) |
4dd726f0 | 1432 | drbd_al_begin_io(device, &req->i); |
265be2d0 PR |
1433 | |
1434 | drbd_req_make_private_bio(req, req->master_bio); | |
b30ab791 | 1435 | req->private_bio->bi_bdev = device->ldev->backing_bdev; |
265be2d0 PR |
1436 | generic_make_request(req->private_bio); |
1437 | ||
99920dc5 | 1438 | return 0; |
265be2d0 PR |
1439 | } |
1440 | ||
b30ab791 | 1441 | static int _drbd_may_sync_now(struct drbd_device *device) |
b411b363 | 1442 | { |
b30ab791 | 1443 | struct drbd_device *odev = device; |
95f8efd0 | 1444 | int resync_after; |
b411b363 PR |
1445 | |
1446 | while (1) { | |
a3f8f7dc | 1447 | if (!odev->ldev || odev->state.disk == D_DISKLESS) |
438c8374 | 1448 | return 1; |
daeda1cc | 1449 | rcu_read_lock(); |
95f8efd0 | 1450 | resync_after = rcu_dereference(odev->ldev->disk_conf)->resync_after; |
daeda1cc | 1451 | rcu_read_unlock(); |
95f8efd0 | 1452 | if (resync_after == -1) |
b411b363 | 1453 | return 1; |
b30ab791 | 1454 | odev = minor_to_device(resync_after); |
a3f8f7dc | 1455 | if (!odev) |
841ce241 | 1456 | return 1; |
b411b363 PR |
1457 | if ((odev->state.conn >= C_SYNC_SOURCE && |
1458 | odev->state.conn <= C_PAUSED_SYNC_T) || | |
1459 | odev->state.aftr_isp || odev->state.peer_isp || | |
1460 | odev->state.user_isp) | |
1461 | return 0; | |
1462 | } | |
1463 | } | |
1464 | ||
1465 | /** | |
28bc3b8c | 1466 | * drbd_pause_after() - Pause resync on all devices that may not resync now |
b30ab791 | 1467 | * @device: DRBD device. |
b411b363 PR |
1468 | * |
1469 | * Called from process context only (admin command and after_state_ch). | |
1470 | */ | |
28bc3b8c | 1471 | static bool drbd_pause_after(struct drbd_device *device) |
b411b363 | 1472 | { |
28bc3b8c | 1473 | bool changed = false; |
54761697 | 1474 | struct drbd_device *odev; |
28bc3b8c | 1475 | int i; |
b411b363 | 1476 | |
695d08fa | 1477 | rcu_read_lock(); |
05a10ec7 | 1478 | idr_for_each_entry(&drbd_devices, odev, i) { |
b411b363 PR |
1479 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1480 | continue; | |
28bc3b8c AG |
1481 | if (!_drbd_may_sync_now(odev) && |
1482 | _drbd_set_state(_NS(odev, aftr_isp, 1), | |
1483 | CS_HARD, NULL) != SS_NOTHING_TO_DO) | |
1484 | changed = true; | |
b411b363 | 1485 | } |
695d08fa | 1486 | rcu_read_unlock(); |
b411b363 | 1487 | |
28bc3b8c | 1488 | return changed; |
b411b363 PR |
1489 | } |
1490 | ||
1491 | /** | |
28bc3b8c | 1492 | * drbd_resume_next() - Resume resync on all devices that may resync now |
b30ab791 | 1493 | * @device: DRBD device. |
b411b363 PR |
1494 | * |
1495 | * Called from process context only (admin command and worker). | |
1496 | */ | |
28bc3b8c | 1497 | static bool drbd_resume_next(struct drbd_device *device) |
b411b363 | 1498 | { |
28bc3b8c | 1499 | bool changed = false; |
54761697 | 1500 | struct drbd_device *odev; |
28bc3b8c | 1501 | int i; |
b411b363 | 1502 | |
695d08fa | 1503 | rcu_read_lock(); |
05a10ec7 | 1504 | idr_for_each_entry(&drbd_devices, odev, i) { |
b411b363 PR |
1505 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1506 | continue; | |
1507 | if (odev->state.aftr_isp) { | |
28bc3b8c AG |
1508 | if (_drbd_may_sync_now(odev) && |
1509 | _drbd_set_state(_NS(odev, aftr_isp, 0), | |
1510 | CS_HARD, NULL) != SS_NOTHING_TO_DO) | |
1511 | changed = true; | |
b411b363 PR |
1512 | } |
1513 | } | |
695d08fa | 1514 | rcu_read_unlock(); |
28bc3b8c | 1515 | return changed; |
b411b363 PR |
1516 | } |
1517 | ||
b30ab791 | 1518 | void resume_next_sg(struct drbd_device *device) |
b411b363 | 1519 | { |
28bc3b8c AG |
1520 | lock_all_resources(); |
1521 | drbd_resume_next(device); | |
1522 | unlock_all_resources(); | |
b411b363 PR |
1523 | } |
1524 | ||
b30ab791 | 1525 | void suspend_other_sg(struct drbd_device *device) |
b411b363 | 1526 | { |
28bc3b8c AG |
1527 | lock_all_resources(); |
1528 | drbd_pause_after(device); | |
1529 | unlock_all_resources(); | |
b411b363 PR |
1530 | } |
1531 | ||
28bc3b8c | 1532 | /* caller must lock_all_resources() */ |
b30ab791 | 1533 | enum drbd_ret_code drbd_resync_after_valid(struct drbd_device *device, int o_minor) |
b411b363 | 1534 | { |
54761697 | 1535 | struct drbd_device *odev; |
95f8efd0 | 1536 | int resync_after; |
b411b363 PR |
1537 | |
1538 | if (o_minor == -1) | |
1539 | return NO_ERROR; | |
a3f8f7dc | 1540 | if (o_minor < -1 || o_minor > MINORMASK) |
95f8efd0 | 1541 | return ERR_RESYNC_AFTER; |
b411b363 PR |
1542 | |
1543 | /* check for loops */ | |
b30ab791 | 1544 | odev = minor_to_device(o_minor); |
b411b363 | 1545 | while (1) { |
b30ab791 | 1546 | if (odev == device) |
95f8efd0 | 1547 | return ERR_RESYNC_AFTER_CYCLE; |
b411b363 | 1548 | |
a3f8f7dc LE |
1549 | /* You are free to depend on diskless, non-existing, |
1550 | * or not yet/no longer existing minors. | |
1551 | * We only reject dependency loops. | |
1552 | * We cannot follow the dependency chain beyond a detached or | |
1553 | * missing minor. | |
1554 | */ | |
1555 | if (!odev || !odev->ldev || odev->state.disk == D_DISKLESS) | |
1556 | return NO_ERROR; | |
1557 | ||
daeda1cc | 1558 | rcu_read_lock(); |
95f8efd0 | 1559 | resync_after = rcu_dereference(odev->ldev->disk_conf)->resync_after; |
daeda1cc | 1560 | rcu_read_unlock(); |
b411b363 | 1561 | /* dependency chain ends here, no cycles. */ |
95f8efd0 | 1562 | if (resync_after == -1) |
b411b363 PR |
1563 | return NO_ERROR; |
1564 | ||
1565 | /* follow the dependency chain */ | |
b30ab791 | 1566 | odev = minor_to_device(resync_after); |
b411b363 PR |
1567 | } |
1568 | } | |
1569 | ||
28bc3b8c | 1570 | /* caller must lock_all_resources() */ |
b30ab791 | 1571 | void drbd_resync_after_changed(struct drbd_device *device) |
b411b363 | 1572 | { |
28bc3b8c | 1573 | int changed; |
b411b363 | 1574 | |
dc97b708 | 1575 | do { |
28bc3b8c AG |
1576 | changed = drbd_pause_after(device); |
1577 | changed |= drbd_resume_next(device); | |
1578 | } while (changed); | |
b411b363 PR |
1579 | } |
1580 | ||
b30ab791 | 1581 | void drbd_rs_controller_reset(struct drbd_device *device) |
9bd28d3c | 1582 | { |
ff8bd88b | 1583 | struct gendisk *disk = device->ldev->backing_bdev->bd_contains->bd_disk; |
813472ce PR |
1584 | struct fifo_buffer *plan; |
1585 | ||
b30ab791 AG |
1586 | atomic_set(&device->rs_sect_in, 0); |
1587 | atomic_set(&device->rs_sect_ev, 0); | |
1588 | device->rs_in_flight = 0; | |
ff8bd88b LE |
1589 | device->rs_last_events = |
1590 | (int)part_stat_read(&disk->part0, sectors[0]) + | |
1591 | (int)part_stat_read(&disk->part0, sectors[1]); | |
813472ce PR |
1592 | |
1593 | /* Updating the RCU protected object in place is necessary since | |
1594 | this function gets called from atomic context. | |
1595 | It is valid since all other updates also lead to an completely | |
1596 | empty fifo */ | |
1597 | rcu_read_lock(); | |
b30ab791 | 1598 | plan = rcu_dereference(device->rs_plan_s); |
813472ce PR |
1599 | plan->total = 0; |
1600 | fifo_set(plan, 0); | |
1601 | rcu_read_unlock(); | |
9bd28d3c LE |
1602 | } |
1603 | ||
1f04af33 PR |
1604 | void start_resync_timer_fn(unsigned long data) |
1605 | { | |
b30ab791 | 1606 | struct drbd_device *device = (struct drbd_device *) data; |
ac0acb9e | 1607 | drbd_device_post_work(device, RS_START); |
1f04af33 PR |
1608 | } |
1609 | ||
ac0acb9e | 1610 | static void do_start_resync(struct drbd_device *device) |
1f04af33 | 1611 | { |
b30ab791 | 1612 | if (atomic_read(&device->unacked_cnt) || atomic_read(&device->rs_pending_cnt)) { |
ac0acb9e | 1613 | drbd_warn(device, "postponing start_resync ...\n"); |
b30ab791 AG |
1614 | device->start_resync_timer.expires = jiffies + HZ/10; |
1615 | add_timer(&device->start_resync_timer); | |
ac0acb9e | 1616 | return; |
1f04af33 PR |
1617 | } |
1618 | ||
b30ab791 AG |
1619 | drbd_start_resync(device, C_SYNC_SOURCE); |
1620 | clear_bit(AHEAD_TO_SYNC_SOURCE, &device->flags); | |
1f04af33 PR |
1621 | } |
1622 | ||
aaaba345 LE |
1623 | static bool use_checksum_based_resync(struct drbd_connection *connection, struct drbd_device *device) |
1624 | { | |
1625 | bool csums_after_crash_only; | |
1626 | rcu_read_lock(); | |
1627 | csums_after_crash_only = rcu_dereference(connection->net_conf)->csums_after_crash_only; | |
1628 | rcu_read_unlock(); | |
1629 | return connection->agreed_pro_version >= 89 && /* supported? */ | |
1630 | connection->csums_tfm && /* configured? */ | |
1631 | (csums_after_crash_only == 0 /* use for each resync? */ | |
1632 | || test_bit(CRASHED_PRIMARY, &device->flags)); /* or only after Primary crash? */ | |
1633 | } | |
1634 | ||
b411b363 PR |
1635 | /** |
1636 | * drbd_start_resync() - Start the resync process | |
b30ab791 | 1637 | * @device: DRBD device. |
b411b363 PR |
1638 | * @side: Either C_SYNC_SOURCE or C_SYNC_TARGET |
1639 | * | |
1640 | * This function might bring you directly into one of the | |
1641 | * C_PAUSED_SYNC_* states. | |
1642 | */ | |
b30ab791 | 1643 | void drbd_start_resync(struct drbd_device *device, enum drbd_conns side) |
b411b363 | 1644 | { |
44a4d551 LE |
1645 | struct drbd_peer_device *peer_device = first_peer_device(device); |
1646 | struct drbd_connection *connection = peer_device ? peer_device->connection : NULL; | |
b411b363 PR |
1647 | union drbd_state ns; |
1648 | int r; | |
1649 | ||
b30ab791 | 1650 | if (device->state.conn >= C_SYNC_SOURCE && device->state.conn < C_AHEAD) { |
d0180171 | 1651 | drbd_err(device, "Resync already running!\n"); |
b411b363 PR |
1652 | return; |
1653 | } | |
1654 | ||
b30ab791 | 1655 | if (!test_bit(B_RS_H_DONE, &device->flags)) { |
e64a3294 PR |
1656 | if (side == C_SYNC_TARGET) { |
1657 | /* Since application IO was locked out during C_WF_BITMAP_T and | |
1658 | C_WF_SYNC_UUID we are still unmodified. Before going to C_SYNC_TARGET | |
1659 | we check that we might make the data inconsistent. */ | |
b30ab791 | 1660 | r = drbd_khelper(device, "before-resync-target"); |
e64a3294 PR |
1661 | r = (r >> 8) & 0xff; |
1662 | if (r > 0) { | |
d0180171 | 1663 | drbd_info(device, "before-resync-target handler returned %d, " |
09b9e797 | 1664 | "dropping connection.\n", r); |
44a4d551 | 1665 | conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); |
09b9e797 PR |
1666 | return; |
1667 | } | |
e64a3294 | 1668 | } else /* C_SYNC_SOURCE */ { |
b30ab791 | 1669 | r = drbd_khelper(device, "before-resync-source"); |
e64a3294 PR |
1670 | r = (r >> 8) & 0xff; |
1671 | if (r > 0) { | |
1672 | if (r == 3) { | |
d0180171 | 1673 | drbd_info(device, "before-resync-source handler returned %d, " |
e64a3294 PR |
1674 | "ignoring. Old userland tools?", r); |
1675 | } else { | |
d0180171 | 1676 | drbd_info(device, "before-resync-source handler returned %d, " |
e64a3294 | 1677 | "dropping connection.\n", r); |
44a4d551 | 1678 | conn_request_state(connection, |
a6b32bc3 | 1679 | NS(conn, C_DISCONNECTING), CS_HARD); |
e64a3294 PR |
1680 | return; |
1681 | } | |
1682 | } | |
09b9e797 | 1683 | } |
b411b363 PR |
1684 | } |
1685 | ||
44a4d551 | 1686 | if (current == connection->worker.task) { |
dad20554 | 1687 | /* The worker should not sleep waiting for state_mutex, |
e64a3294 | 1688 | that can take long */ |
b30ab791 AG |
1689 | if (!mutex_trylock(device->state_mutex)) { |
1690 | set_bit(B_RS_H_DONE, &device->flags); | |
1691 | device->start_resync_timer.expires = jiffies + HZ/5; | |
1692 | add_timer(&device->start_resync_timer); | |
e64a3294 PR |
1693 | return; |
1694 | } | |
1695 | } else { | |
b30ab791 | 1696 | mutex_lock(device->state_mutex); |
e64a3294 | 1697 | } |
b411b363 | 1698 | |
28bc3b8c AG |
1699 | lock_all_resources(); |
1700 | clear_bit(B_RS_H_DONE, &device->flags); | |
a700471b | 1701 | /* Did some connection breakage or IO error race with us? */ |
b30ab791 AG |
1702 | if (device->state.conn < C_CONNECTED |
1703 | || !get_ldev_if_state(device, D_NEGOTIATING)) { | |
28bc3b8c AG |
1704 | unlock_all_resources(); |
1705 | goto out; | |
b411b363 PR |
1706 | } |
1707 | ||
b30ab791 | 1708 | ns = drbd_read_state(device); |
b411b363 | 1709 | |
b30ab791 | 1710 | ns.aftr_isp = !_drbd_may_sync_now(device); |
b411b363 PR |
1711 | |
1712 | ns.conn = side; | |
1713 | ||
1714 | if (side == C_SYNC_TARGET) | |
1715 | ns.disk = D_INCONSISTENT; | |
1716 | else /* side == C_SYNC_SOURCE */ | |
1717 | ns.pdsk = D_INCONSISTENT; | |
1718 | ||
28bc3b8c | 1719 | r = _drbd_set_state(device, ns, CS_VERBOSE, NULL); |
b30ab791 | 1720 | ns = drbd_read_state(device); |
b411b363 PR |
1721 | |
1722 | if (ns.conn < C_CONNECTED) | |
1723 | r = SS_UNKNOWN_ERROR; | |
1724 | ||
1725 | if (r == SS_SUCCESS) { | |
b30ab791 | 1726 | unsigned long tw = drbd_bm_total_weight(device); |
1d7734a0 LE |
1727 | unsigned long now = jiffies; |
1728 | int i; | |
1729 | ||
b30ab791 AG |
1730 | device->rs_failed = 0; |
1731 | device->rs_paused = 0; | |
1732 | device->rs_same_csum = 0; | |
b30ab791 AG |
1733 | device->rs_last_sect_ev = 0; |
1734 | device->rs_total = tw; | |
1735 | device->rs_start = now; | |
1d7734a0 | 1736 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
1737 | device->rs_mark_left[i] = tw; |
1738 | device->rs_mark_time[i] = now; | |
1d7734a0 | 1739 | } |
28bc3b8c | 1740 | drbd_pause_after(device); |
5ab7d2c0 LE |
1741 | /* Forget potentially stale cached per resync extent bit-counts. |
1742 | * Open coded drbd_rs_cancel_all(device), we already have IRQs | |
1743 | * disabled, and know the disk state is ok. */ | |
1744 | spin_lock(&device->al_lock); | |
1745 | lc_reset(device->resync); | |
1746 | device->resync_locked = 0; | |
1747 | device->resync_wenr = LC_FREE; | |
1748 | spin_unlock(&device->al_lock); | |
b411b363 | 1749 | } |
28bc3b8c | 1750 | unlock_all_resources(); |
5a22db89 | 1751 | |
b411b363 | 1752 | if (r == SS_SUCCESS) { |
5ab7d2c0 | 1753 | wake_up(&device->al_wait); /* for lc_reset() above */ |
328e0f12 PR |
1754 | /* reset rs_last_bcast when a resync or verify is started, |
1755 | * to deal with potential jiffies wrap. */ | |
b30ab791 | 1756 | device->rs_last_bcast = jiffies - HZ; |
328e0f12 | 1757 | |
d0180171 | 1758 | drbd_info(device, "Began resync as %s (will sync %lu KB [%lu bits set]).\n", |
b411b363 | 1759 | drbd_conn_str(ns.conn), |
b30ab791 AG |
1760 | (unsigned long) device->rs_total << (BM_BLOCK_SHIFT-10), |
1761 | (unsigned long) device->rs_total); | |
aaaba345 | 1762 | if (side == C_SYNC_TARGET) { |
b30ab791 | 1763 | device->bm_resync_fo = 0; |
aaaba345 LE |
1764 | device->use_csums = use_checksum_based_resync(connection, device); |
1765 | } else { | |
1766 | device->use_csums = 0; | |
1767 | } | |
6c922ed5 LE |
1768 | |
1769 | /* Since protocol 96, we must serialize drbd_gen_and_send_sync_uuid | |
1770 | * with w_send_oos, or the sync target will get confused as to | |
1771 | * how much bits to resync. We cannot do that always, because for an | |
1772 | * empty resync and protocol < 95, we need to do it here, as we call | |
1773 | * drbd_resync_finished from here in that case. | |
1774 | * We drbd_gen_and_send_sync_uuid here for protocol < 96, | |
1775 | * and from after_state_ch otherwise. */ | |
44a4d551 LE |
1776 | if (side == C_SYNC_SOURCE && connection->agreed_pro_version < 96) |
1777 | drbd_gen_and_send_sync_uuid(peer_device); | |
b411b363 | 1778 | |
44a4d551 | 1779 | if (connection->agreed_pro_version < 95 && device->rs_total == 0) { |
af85e8e8 LE |
1780 | /* This still has a race (about when exactly the peers |
1781 | * detect connection loss) that can lead to a full sync | |
1782 | * on next handshake. In 8.3.9 we fixed this with explicit | |
1783 | * resync-finished notifications, but the fix | |
1784 | * introduces a protocol change. Sleeping for some | |
1785 | * time longer than the ping interval + timeout on the | |
1786 | * SyncSource, to give the SyncTarget the chance to | |
1787 | * detect connection loss, then waiting for a ping | |
1788 | * response (implicit in drbd_resync_finished) reduces | |
1789 | * the race considerably, but does not solve it. */ | |
44ed167d PR |
1790 | if (side == C_SYNC_SOURCE) { |
1791 | struct net_conf *nc; | |
1792 | int timeo; | |
1793 | ||
1794 | rcu_read_lock(); | |
44a4d551 | 1795 | nc = rcu_dereference(connection->net_conf); |
44ed167d PR |
1796 | timeo = nc->ping_int * HZ + nc->ping_timeo * HZ / 9; |
1797 | rcu_read_unlock(); | |
1798 | schedule_timeout_interruptible(timeo); | |
1799 | } | |
b30ab791 | 1800 | drbd_resync_finished(device); |
b411b363 PR |
1801 | } |
1802 | ||
b30ab791 AG |
1803 | drbd_rs_controller_reset(device); |
1804 | /* ns.conn may already be != device->state.conn, | |
b411b363 PR |
1805 | * we may have been paused in between, or become paused until |
1806 | * the timer triggers. | |
1807 | * No matter, that is handled in resync_timer_fn() */ | |
1808 | if (ns.conn == C_SYNC_TARGET) | |
b30ab791 | 1809 | mod_timer(&device->resync_timer, jiffies); |
b411b363 | 1810 | |
b30ab791 | 1811 | drbd_md_sync(device); |
b411b363 | 1812 | } |
b30ab791 | 1813 | put_ldev(device); |
28bc3b8c | 1814 | out: |
b30ab791 | 1815 | mutex_unlock(device->state_mutex); |
b411b363 PR |
1816 | } |
1817 | ||
e334f550 | 1818 | static void update_on_disk_bitmap(struct drbd_device *device, bool resync_done) |
c7a58db4 LE |
1819 | { |
1820 | struct sib_info sib = { .sib_reason = SIB_SYNC_PROGRESS, }; | |
1821 | device->rs_last_bcast = jiffies; | |
1822 | ||
1823 | if (!get_ldev(device)) | |
1824 | return; | |
1825 | ||
1826 | drbd_bm_write_lazy(device, 0); | |
5ab7d2c0 | 1827 | if (resync_done && is_sync_state(device->state.conn)) |
c7a58db4 | 1828 | drbd_resync_finished(device); |
5ab7d2c0 | 1829 | |
c7a58db4 LE |
1830 | drbd_bcast_event(device, &sib); |
1831 | /* update timestamp, in case it took a while to write out stuff */ | |
1832 | device->rs_last_bcast = jiffies; | |
1833 | put_ldev(device); | |
1834 | } | |
1835 | ||
e334f550 LE |
1836 | static void drbd_ldev_destroy(struct drbd_device *device) |
1837 | { | |
1838 | lc_destroy(device->resync); | |
1839 | device->resync = NULL; | |
1840 | lc_destroy(device->act_log); | |
1841 | device->act_log = NULL; | |
d1b80853 AG |
1842 | |
1843 | __acquire(local); | |
1844 | drbd_free_ldev(device->ldev); | |
1845 | device->ldev = NULL; | |
1846 | __release(local); | |
1847 | ||
e334f550 LE |
1848 | clear_bit(GOING_DISKLESS, &device->flags); |
1849 | wake_up(&device->misc_wait); | |
1850 | } | |
1851 | ||
1852 | static void go_diskless(struct drbd_device *device) | |
1853 | { | |
1854 | D_ASSERT(device, device->state.disk == D_FAILED); | |
1855 | /* we cannot assert local_cnt == 0 here, as get_ldev_if_state will | |
1856 | * inc/dec it frequently. Once we are D_DISKLESS, no one will touch | |
1857 | * the protected members anymore, though, so once put_ldev reaches zero | |
1858 | * again, it will be safe to free them. */ | |
1859 | ||
1860 | /* Try to write changed bitmap pages, read errors may have just | |
1861 | * set some bits outside the area covered by the activity log. | |
1862 | * | |
1863 | * If we have an IO error during the bitmap writeout, | |
1864 | * we will want a full sync next time, just in case. | |
1865 | * (Do we want a specific meta data flag for this?) | |
1866 | * | |
1867 | * If that does not make it to stable storage either, | |
1868 | * we cannot do anything about that anymore. | |
1869 | * | |
1870 | * We still need to check if both bitmap and ldev are present, we may | |
1871 | * end up here after a failed attach, before ldev was even assigned. | |
1872 | */ | |
1873 | if (device->bitmap && device->ldev) { | |
1874 | /* An interrupted resync or similar is allowed to recounts bits | |
1875 | * while we detach. | |
1876 | * Any modifications would not be expected anymore, though. | |
1877 | */ | |
1878 | if (drbd_bitmap_io_from_worker(device, drbd_bm_write, | |
1879 | "detach", BM_LOCKED_TEST_ALLOWED)) { | |
1880 | if (test_bit(WAS_READ_ERROR, &device->flags)) { | |
1881 | drbd_md_set_flag(device, MDF_FULL_SYNC); | |
1882 | drbd_md_sync(device); | |
1883 | } | |
1884 | } | |
1885 | } | |
1886 | ||
1887 | drbd_force_state(device, NS(disk, D_DISKLESS)); | |
1888 | } | |
1889 | ||
ac0acb9e LE |
1890 | static int do_md_sync(struct drbd_device *device) |
1891 | { | |
1892 | drbd_warn(device, "md_sync_timer expired! Worker calls drbd_md_sync().\n"); | |
1893 | drbd_md_sync(device); | |
1894 | return 0; | |
1895 | } | |
1896 | ||
944410e9 LE |
1897 | /* only called from drbd_worker thread, no locking */ |
1898 | void __update_timing_details( | |
1899 | struct drbd_thread_timing_details *tdp, | |
1900 | unsigned int *cb_nr, | |
1901 | void *cb, | |
1902 | const char *fn, const unsigned int line) | |
1903 | { | |
1904 | unsigned int i = *cb_nr % DRBD_THREAD_DETAILS_HIST; | |
1905 | struct drbd_thread_timing_details *td = tdp + i; | |
1906 | ||
1907 | td->start_jif = jiffies; | |
1908 | td->cb_addr = cb; | |
1909 | td->caller_fn = fn; | |
1910 | td->line = line; | |
1911 | td->cb_nr = *cb_nr; | |
1912 | ||
1913 | i = (i+1) % DRBD_THREAD_DETAILS_HIST; | |
1914 | td = tdp + i; | |
1915 | memset(td, 0, sizeof(*td)); | |
1916 | ||
1917 | ++(*cb_nr); | |
1918 | } | |
1919 | ||
e334f550 LE |
1920 | static void do_device_work(struct drbd_device *device, const unsigned long todo) |
1921 | { | |
b47a06d1 | 1922 | if (test_bit(MD_SYNC, &todo)) |
ac0acb9e | 1923 | do_md_sync(device); |
b47a06d1 AG |
1924 | if (test_bit(RS_DONE, &todo) || |
1925 | test_bit(RS_PROGRESS, &todo)) | |
1926 | update_on_disk_bitmap(device, test_bit(RS_DONE, &todo)); | |
1927 | if (test_bit(GO_DISKLESS, &todo)) | |
e334f550 | 1928 | go_diskless(device); |
b47a06d1 | 1929 | if (test_bit(DESTROY_DISK, &todo)) |
e334f550 | 1930 | drbd_ldev_destroy(device); |
b47a06d1 | 1931 | if (test_bit(RS_START, &todo)) |
ac0acb9e | 1932 | do_start_resync(device); |
e334f550 LE |
1933 | } |
1934 | ||
1935 | #define DRBD_DEVICE_WORK_MASK \ | |
1936 | ((1UL << GO_DISKLESS) \ | |
1937 | |(1UL << DESTROY_DISK) \ | |
ac0acb9e LE |
1938 | |(1UL << MD_SYNC) \ |
1939 | |(1UL << RS_START) \ | |
e334f550 LE |
1940 | |(1UL << RS_PROGRESS) \ |
1941 | |(1UL << RS_DONE) \ | |
1942 | ) | |
1943 | ||
1944 | static unsigned long get_work_bits(unsigned long *flags) | |
1945 | { | |
1946 | unsigned long old, new; | |
1947 | do { | |
1948 | old = *flags; | |
1949 | new = old & ~DRBD_DEVICE_WORK_MASK; | |
1950 | } while (cmpxchg(flags, old, new) != old); | |
1951 | return old & DRBD_DEVICE_WORK_MASK; | |
1952 | } | |
1953 | ||
1954 | static void do_unqueued_work(struct drbd_connection *connection) | |
c7a58db4 LE |
1955 | { |
1956 | struct drbd_peer_device *peer_device; | |
1957 | int vnr; | |
1958 | ||
1959 | rcu_read_lock(); | |
1960 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
1961 | struct drbd_device *device = peer_device->device; | |
e334f550 LE |
1962 | unsigned long todo = get_work_bits(&device->flags); |
1963 | if (!todo) | |
c7a58db4 | 1964 | continue; |
5ab7d2c0 | 1965 | |
c7a58db4 LE |
1966 | kref_get(&device->kref); |
1967 | rcu_read_unlock(); | |
e334f550 | 1968 | do_device_work(device, todo); |
c7a58db4 LE |
1969 | kref_put(&device->kref, drbd_destroy_device); |
1970 | rcu_read_lock(); | |
1971 | } | |
1972 | rcu_read_unlock(); | |
1973 | } | |
1974 | ||
a186e478 | 1975 | static bool dequeue_work_batch(struct drbd_work_queue *queue, struct list_head *work_list) |
8c0785a5 LE |
1976 | { |
1977 | spin_lock_irq(&queue->q_lock); | |
15e26f6a | 1978 | list_splice_tail_init(&queue->q, work_list); |
8c0785a5 LE |
1979 | spin_unlock_irq(&queue->q_lock); |
1980 | return !list_empty(work_list); | |
1981 | } | |
1982 | ||
bde89a9e | 1983 | static void wait_for_work(struct drbd_connection *connection, struct list_head *work_list) |
b6dd1a89 LE |
1984 | { |
1985 | DEFINE_WAIT(wait); | |
1986 | struct net_conf *nc; | |
1987 | int uncork, cork; | |
1988 | ||
abde9cc6 | 1989 | dequeue_work_batch(&connection->sender_work, work_list); |
b6dd1a89 LE |
1990 | if (!list_empty(work_list)) |
1991 | return; | |
1992 | ||
1993 | /* Still nothing to do? | |
1994 | * Maybe we still need to close the current epoch, | |
1995 | * even if no new requests are queued yet. | |
1996 | * | |
1997 | * Also, poke TCP, just in case. | |
1998 | * Then wait for new work (or signal). */ | |
1999 | rcu_read_lock(); | |
2000 | nc = rcu_dereference(connection->net_conf); | |
2001 | uncork = nc ? nc->tcp_cork : 0; | |
2002 | rcu_read_unlock(); | |
2003 | if (uncork) { | |
2004 | mutex_lock(&connection->data.mutex); | |
2005 | if (connection->data.socket) | |
2006 | drbd_tcp_uncork(connection->data.socket); | |
2007 | mutex_unlock(&connection->data.mutex); | |
2008 | } | |
2009 | ||
2010 | for (;;) { | |
2011 | int send_barrier; | |
2012 | prepare_to_wait(&connection->sender_work.q_wait, &wait, TASK_INTERRUPTIBLE); | |
0500813f | 2013 | spin_lock_irq(&connection->resource->req_lock); |
b6dd1a89 | 2014 | spin_lock(&connection->sender_work.q_lock); /* FIXME get rid of this one? */ |
bc317a9e | 2015 | if (!list_empty(&connection->sender_work.q)) |
4dd726f0 | 2016 | list_splice_tail_init(&connection->sender_work.q, work_list); |
b6dd1a89 LE |
2017 | spin_unlock(&connection->sender_work.q_lock); /* FIXME get rid of this one? */ |
2018 | if (!list_empty(work_list) || signal_pending(current)) { | |
0500813f | 2019 | spin_unlock_irq(&connection->resource->req_lock); |
b6dd1a89 LE |
2020 | break; |
2021 | } | |
f9c78128 LE |
2022 | |
2023 | /* We found nothing new to do, no to-be-communicated request, | |
2024 | * no other work item. We may still need to close the last | |
2025 | * epoch. Next incoming request epoch will be connection -> | |
2026 | * current transfer log epoch number. If that is different | |
2027 | * from the epoch of the last request we communicated, it is | |
2028 | * safe to send the epoch separating barrier now. | |
2029 | */ | |
2030 | send_barrier = | |
2031 | atomic_read(&connection->current_tle_nr) != | |
2032 | connection->send.current_epoch_nr; | |
0500813f | 2033 | spin_unlock_irq(&connection->resource->req_lock); |
f9c78128 LE |
2034 | |
2035 | if (send_barrier) | |
2036 | maybe_send_barrier(connection, | |
2037 | connection->send.current_epoch_nr + 1); | |
5ab7d2c0 | 2038 | |
e334f550 | 2039 | if (test_bit(DEVICE_WORK_PENDING, &connection->flags)) |
5ab7d2c0 LE |
2040 | break; |
2041 | ||
a80ca1ae LE |
2042 | /* drbd_send() may have called flush_signals() */ |
2043 | if (get_t_state(&connection->worker) != RUNNING) | |
2044 | break; | |
5ab7d2c0 | 2045 | |
b6dd1a89 LE |
2046 | schedule(); |
2047 | /* may be woken up for other things but new work, too, | |
2048 | * e.g. if the current epoch got closed. | |
2049 | * In which case we send the barrier above. */ | |
2050 | } | |
2051 | finish_wait(&connection->sender_work.q_wait, &wait); | |
2052 | ||
2053 | /* someone may have changed the config while we have been waiting above. */ | |
2054 | rcu_read_lock(); | |
2055 | nc = rcu_dereference(connection->net_conf); | |
2056 | cork = nc ? nc->tcp_cork : 0; | |
2057 | rcu_read_unlock(); | |
2058 | mutex_lock(&connection->data.mutex); | |
2059 | if (connection->data.socket) { | |
2060 | if (cork) | |
2061 | drbd_tcp_cork(connection->data.socket); | |
2062 | else if (!uncork) | |
2063 | drbd_tcp_uncork(connection->data.socket); | |
2064 | } | |
2065 | mutex_unlock(&connection->data.mutex); | |
2066 | } | |
2067 | ||
b411b363 PR |
2068 | int drbd_worker(struct drbd_thread *thi) |
2069 | { | |
bde89a9e | 2070 | struct drbd_connection *connection = thi->connection; |
6db7e50a | 2071 | struct drbd_work *w = NULL; |
c06ece6b | 2072 | struct drbd_peer_device *peer_device; |
b411b363 | 2073 | LIST_HEAD(work_list); |
8c0785a5 | 2074 | int vnr; |
b411b363 | 2075 | |
e77a0a5c | 2076 | while (get_t_state(thi) == RUNNING) { |
80822284 | 2077 | drbd_thread_current_set_cpu(thi); |
b411b363 | 2078 | |
944410e9 LE |
2079 | if (list_empty(&work_list)) { |
2080 | update_worker_timing_details(connection, wait_for_work); | |
bde89a9e | 2081 | wait_for_work(connection, &work_list); |
944410e9 | 2082 | } |
b411b363 | 2083 | |
944410e9 LE |
2084 | if (test_and_clear_bit(DEVICE_WORK_PENDING, &connection->flags)) { |
2085 | update_worker_timing_details(connection, do_unqueued_work); | |
e334f550 | 2086 | do_unqueued_work(connection); |
944410e9 | 2087 | } |
5ab7d2c0 | 2088 | |
8c0785a5 | 2089 | if (signal_pending(current)) { |
b411b363 | 2090 | flush_signals(current); |
19393e10 | 2091 | if (get_t_state(thi) == RUNNING) { |
1ec861eb | 2092 | drbd_warn(connection, "Worker got an unexpected signal\n"); |
b411b363 | 2093 | continue; |
19393e10 | 2094 | } |
b411b363 PR |
2095 | break; |
2096 | } | |
2097 | ||
e77a0a5c | 2098 | if (get_t_state(thi) != RUNNING) |
b411b363 | 2099 | break; |
b411b363 | 2100 | |
729e8b87 | 2101 | if (!list_empty(&work_list)) { |
6db7e50a AG |
2102 | w = list_first_entry(&work_list, struct drbd_work, list); |
2103 | list_del_init(&w->list); | |
944410e9 | 2104 | update_worker_timing_details(connection, w->cb); |
6db7e50a | 2105 | if (w->cb(w, connection->cstate < C_WF_REPORT_PARAMS) == 0) |
8c0785a5 | 2106 | continue; |
bde89a9e AG |
2107 | if (connection->cstate >= C_WF_REPORT_PARAMS) |
2108 | conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); | |
b411b363 PR |
2109 | } |
2110 | } | |
b411b363 | 2111 | |
8c0785a5 | 2112 | do { |
944410e9 LE |
2113 | if (test_and_clear_bit(DEVICE_WORK_PENDING, &connection->flags)) { |
2114 | update_worker_timing_details(connection, do_unqueued_work); | |
e334f550 | 2115 | do_unqueued_work(connection); |
944410e9 | 2116 | } |
729e8b87 | 2117 | if (!list_empty(&work_list)) { |
6db7e50a AG |
2118 | w = list_first_entry(&work_list, struct drbd_work, list); |
2119 | list_del_init(&w->list); | |
944410e9 | 2120 | update_worker_timing_details(connection, w->cb); |
6db7e50a | 2121 | w->cb(w, 1); |
729e8b87 LE |
2122 | } else |
2123 | dequeue_work_batch(&connection->sender_work, &work_list); | |
e334f550 | 2124 | } while (!list_empty(&work_list) || test_bit(DEVICE_WORK_PENDING, &connection->flags)); |
b411b363 | 2125 | |
c141ebda | 2126 | rcu_read_lock(); |
c06ece6b AG |
2127 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
2128 | struct drbd_device *device = peer_device->device; | |
0b0ba1ef | 2129 | D_ASSERT(device, device->state.disk == D_DISKLESS && device->state.conn == C_STANDALONE); |
b30ab791 | 2130 | kref_get(&device->kref); |
c141ebda | 2131 | rcu_read_unlock(); |
b30ab791 | 2132 | drbd_device_cleanup(device); |
05a10ec7 | 2133 | kref_put(&device->kref, drbd_destroy_device); |
c141ebda | 2134 | rcu_read_lock(); |
0e29d163 | 2135 | } |
c141ebda | 2136 | rcu_read_unlock(); |
b411b363 PR |
2137 | |
2138 | return 0; | |
2139 | } |