Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_req.c | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | drbd is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation; either version 2, or (at your option) | |
13 | any later version. | |
14 | ||
15 | drbd is distributed in the hope that it will be useful, | |
16 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | GNU General Public License for more details. | |
19 | ||
20 | You should have received a copy of the GNU General Public License | |
21 | along with drbd; see the file COPYING. If not, write to | |
22 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | ||
24 | */ | |
25 | ||
b411b363 PR |
26 | #include <linux/module.h> |
27 | ||
28 | #include <linux/slab.h> | |
29 | #include <linux/drbd.h> | |
30 | #include "drbd_int.h" | |
b411b363 PR |
31 | #include "drbd_req.h" |
32 | ||
33 | ||
34 | /* Update disk stats at start of I/O request */ | |
35 | static void _drbd_start_io_acct(struct drbd_conf *mdev, struct drbd_request *req, struct bio *bio) | |
36 | { | |
37 | const int rw = bio_data_dir(bio); | |
38 | int cpu; | |
39 | cpu = part_stat_lock(); | |
40 | part_stat_inc(cpu, &mdev->vdisk->part0, ios[rw]); | |
41 | part_stat_add(cpu, &mdev->vdisk->part0, sectors[rw], bio_sectors(bio)); | |
753c8913 | 42 | part_inc_in_flight(&mdev->vdisk->part0, rw); |
b411b363 | 43 | part_stat_unlock(); |
b411b363 PR |
44 | } |
45 | ||
46 | /* Update disk stats when completing request upwards */ | |
47 | static void _drbd_end_io_acct(struct drbd_conf *mdev, struct drbd_request *req) | |
48 | { | |
49 | int rw = bio_data_dir(req->master_bio); | |
50 | unsigned long duration = jiffies - req->start_time; | |
51 | int cpu; | |
52 | cpu = part_stat_lock(); | |
53 | part_stat_add(cpu, &mdev->vdisk->part0, ticks[rw], duration); | |
54 | part_round_stats(cpu, &mdev->vdisk->part0); | |
753c8913 | 55 | part_dec_in_flight(&mdev->vdisk->part0, rw); |
b411b363 | 56 | part_stat_unlock(); |
b411b363 PR |
57 | } |
58 | ||
9e204cdd AG |
59 | static struct drbd_request *drbd_req_new(struct drbd_conf *mdev, |
60 | struct bio *bio_src) | |
61 | { | |
62 | struct drbd_request *req; | |
63 | ||
64 | req = mempool_alloc(drbd_request_mempool, GFP_NOIO); | |
65 | if (!req) | |
66 | return NULL; | |
67 | ||
68 | drbd_req_make_private_bio(req, bio_src); | |
69 | req->rq_state = bio_data_dir(bio_src) == WRITE ? RQ_WRITE : 0; | |
a21e9298 | 70 | req->w.mdev = mdev; |
9e204cdd AG |
71 | req->master_bio = bio_src; |
72 | req->epoch = 0; | |
53840641 | 73 | |
9e204cdd AG |
74 | drbd_clear_interval(&req->i); |
75 | req->i.sector = bio_src->bi_sector; | |
76 | req->i.size = bio_src->bi_size; | |
5e472264 | 77 | req->i.local = true; |
53840641 AG |
78 | req->i.waiting = false; |
79 | ||
9e204cdd AG |
80 | INIT_LIST_HEAD(&req->tl_requests); |
81 | INIT_LIST_HEAD(&req->w.list); | |
82 | ||
83 | return req; | |
84 | } | |
85 | ||
86 | static void drbd_req_free(struct drbd_request *req) | |
87 | { | |
88 | mempool_free(req, drbd_request_mempool); | |
89 | } | |
90 | ||
91 | /* rw is bio_data_dir(), only READ or WRITE */ | |
b411b363 PR |
92 | static void _req_is_done(struct drbd_conf *mdev, struct drbd_request *req, const int rw) |
93 | { | |
94 | const unsigned long s = req->rq_state; | |
288f422e PR |
95 | |
96 | /* remove it from the transfer log. | |
97 | * well, only if it had been there in the first | |
98 | * place... if it had not (local only or conflicting | |
99 | * and never sent), it should still be "empty" as | |
100 | * initialized in drbd_req_new(), so we can list_del() it | |
101 | * here unconditionally */ | |
102 | list_del(&req->tl_requests); | |
103 | ||
b411b363 PR |
104 | /* if it was a write, we may have to set the corresponding |
105 | * bit(s) out-of-sync first. If it had a local part, we need to | |
106 | * release the reference to the activity log. */ | |
107 | if (rw == WRITE) { | |
b411b363 PR |
108 | /* Set out-of-sync unless both OK flags are set |
109 | * (local only or remote failed). | |
110 | * Other places where we set out-of-sync: | |
111 | * READ with local io-error */ | |
112 | if (!(s & RQ_NET_OK) || !(s & RQ_LOCAL_OK)) | |
ace652ac | 113 | drbd_set_out_of_sync(mdev, req->i.sector, req->i.size); |
b411b363 PR |
114 | |
115 | if ((s & RQ_NET_OK) && (s & RQ_LOCAL_OK) && (s & RQ_NET_SIS)) | |
ace652ac | 116 | drbd_set_in_sync(mdev, req->i.sector, req->i.size); |
b411b363 PR |
117 | |
118 | /* one might be tempted to move the drbd_al_complete_io | |
fcefa62e | 119 | * to the local io completion callback drbd_request_endio. |
b411b363 PR |
120 | * but, if this was a mirror write, we may only |
121 | * drbd_al_complete_io after this is RQ_NET_DONE, | |
122 | * otherwise the extent could be dropped from the al | |
123 | * before it has actually been written on the peer. | |
124 | * if we crash before our peer knows about the request, | |
125 | * but after the extent has been dropped from the al, | |
126 | * we would forget to resync the corresponding extent. | |
127 | */ | |
128 | if (s & RQ_LOCAL_MASK) { | |
129 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
0778286a | 130 | if (s & RQ_IN_ACT_LOG) |
181286ad | 131 | drbd_al_complete_io(mdev, &req->i); |
b411b363 PR |
132 | put_ldev(mdev); |
133 | } else if (__ratelimit(&drbd_ratelimit_state)) { | |
181286ad LE |
134 | dev_warn(DEV, "Should have called drbd_al_complete_io(, %llu, %u), " |
135 | "but my Disk seems to have failed :(\n", | |
136 | (unsigned long long) req->i.sector, req->i.size); | |
b411b363 PR |
137 | } |
138 | } | |
139 | } | |
140 | ||
32fa7e91 | 141 | drbd_req_free(req); |
b411b363 PR |
142 | } |
143 | ||
144 | static void queue_barrier(struct drbd_conf *mdev) | |
145 | { | |
146 | struct drbd_tl_epoch *b; | |
147 | ||
148 | /* We are within the req_lock. Once we queued the barrier for sending, | |
149 | * we set the CREATE_BARRIER bit. It is cleared as soon as a new | |
150 | * barrier/epoch object is added. This is the only place this bit is | |
151 | * set. It indicates that the barrier for this epoch is already queued, | |
152 | * and no new epoch has been created yet. */ | |
153 | if (test_bit(CREATE_BARRIER, &mdev->flags)) | |
154 | return; | |
155 | ||
87eeee41 | 156 | b = mdev->tconn->newest_tle; |
b411b363 | 157 | b->w.cb = w_send_barrier; |
a21e9298 | 158 | b->w.mdev = mdev; |
b411b363 PR |
159 | /* inc_ap_pending done here, so we won't |
160 | * get imbalanced on connection loss. | |
161 | * dec_ap_pending will be done in got_BarrierAck | |
162 | * or (on connection loss) in tl_clear. */ | |
163 | inc_ap_pending(mdev); | |
e42325a5 | 164 | drbd_queue_work(&mdev->tconn->data.work, &b->w); |
b411b363 PR |
165 | set_bit(CREATE_BARRIER, &mdev->flags); |
166 | } | |
167 | ||
168 | static void _about_to_complete_local_write(struct drbd_conf *mdev, | |
169 | struct drbd_request *req) | |
170 | { | |
171 | const unsigned long s = req->rq_state; | |
b411b363 | 172 | |
8a3c1044 LE |
173 | /* Before we can signal completion to the upper layers, |
174 | * we may need to close the current epoch. | |
175 | * We can skip this, if this request has not even been sent, because we | |
176 | * did not have a fully established connection yet/anymore, during | |
177 | * bitmap exchange, or while we are C_AHEAD due to congestion policy. | |
178 | */ | |
179 | if (mdev->state.conn >= C_CONNECTED && | |
180 | (s & RQ_NET_SENT) != 0 && | |
87eeee41 | 181 | req->epoch == mdev->tconn->newest_tle->br_number) |
b411b363 | 182 | queue_barrier(mdev); |
b411b363 PR |
183 | } |
184 | ||
185 | void complete_master_bio(struct drbd_conf *mdev, | |
186 | struct bio_and_error *m) | |
187 | { | |
b411b363 PR |
188 | bio_endio(m->bio, m->error); |
189 | dec_ap_bio(mdev); | |
190 | } | |
191 | ||
53840641 AG |
192 | |
193 | static void drbd_remove_request_interval(struct rb_root *root, | |
194 | struct drbd_request *req) | |
195 | { | |
a21e9298 | 196 | struct drbd_conf *mdev = req->w.mdev; |
53840641 AG |
197 | struct drbd_interval *i = &req->i; |
198 | ||
199 | drbd_remove_interval(root, i); | |
200 | ||
201 | /* Wake up any processes waiting for this request to complete. */ | |
202 | if (i->waiting) | |
203 | wake_up(&mdev->misc_wait); | |
204 | } | |
205 | ||
b411b363 PR |
206 | /* Helper for __req_mod(). |
207 | * Set m->bio to the master bio, if it is fit to be completed, | |
208 | * or leave it alone (it is initialized to NULL in __req_mod), | |
209 | * if it has already been completed, or cannot be completed yet. | |
210 | * If m->bio is set, the error status to be returned is placed in m->error. | |
211 | */ | |
212 | void _req_may_be_done(struct drbd_request *req, struct bio_and_error *m) | |
213 | { | |
214 | const unsigned long s = req->rq_state; | |
a21e9298 | 215 | struct drbd_conf *mdev = req->w.mdev; |
b411b363 PR |
216 | /* only WRITES may end up here without a master bio (on barrier ack) */ |
217 | int rw = req->master_bio ? bio_data_dir(req->master_bio) : WRITE; | |
218 | ||
b411b363 PR |
219 | /* we must not complete the master bio, while it is |
220 | * still being processed by _drbd_send_zc_bio (drbd_send_dblock) | |
221 | * not yet acknowledged by the peer | |
222 | * not yet completed by the local io subsystem | |
223 | * these flags may get cleared in any order by | |
224 | * the worker, | |
225 | * the receiver, | |
226 | * the bio_endio completion callbacks. | |
227 | */ | |
7be8da07 AG |
228 | if (s & RQ_LOCAL_PENDING) |
229 | return; | |
230 | if (req->i.waiting) { | |
231 | /* Retry all conflicting peer requests. */ | |
232 | wake_up(&mdev->misc_wait); | |
233 | } | |
b411b363 PR |
234 | if (s & RQ_NET_QUEUED) |
235 | return; | |
236 | if (s & RQ_NET_PENDING) | |
237 | return; | |
b411b363 PR |
238 | |
239 | if (req->master_bio) { | |
8554df1c | 240 | /* this is DATA_RECEIVED (remote read) |
b411b363 PR |
241 | * or protocol C P_WRITE_ACK |
242 | * or protocol B P_RECV_ACK | |
8554df1c | 243 | * or protocol A "HANDED_OVER_TO_NETWORK" (SendAck) |
b411b363 PR |
244 | * or canceled or failed, |
245 | * or killed from the transfer log due to connection loss. | |
246 | */ | |
247 | ||
248 | /* | |
249 | * figure out whether to report success or failure. | |
250 | * | |
251 | * report success when at least one of the operations succeeded. | |
252 | * or, to put the other way, | |
253 | * only report failure, when both operations failed. | |
254 | * | |
255 | * what to do about the failures is handled elsewhere. | |
256 | * what we need to do here is just: complete the master_bio. | |
257 | * | |
258 | * local completion error, if any, has been stored as ERR_PTR | |
fcefa62e | 259 | * in private_bio within drbd_request_endio. |
b411b363 PR |
260 | */ |
261 | int ok = (s & RQ_LOCAL_OK) || (s & RQ_NET_OK); | |
262 | int error = PTR_ERR(req->private_bio); | |
263 | ||
264 | /* remove the request from the conflict detection | |
265 | * respective block_id verification hash */ | |
dac1389c AG |
266 | if (!drbd_interval_empty(&req->i)) { |
267 | struct rb_root *root; | |
268 | ||
dac1389c AG |
269 | if (rw == WRITE) |
270 | root = &mdev->write_requests; | |
271 | else | |
272 | root = &mdev->read_requests; | |
53840641 | 273 | drbd_remove_request_interval(root, req); |
7be8da07 | 274 | } else if (!(s & RQ_POSTPONED)) |
8825f7c3 | 275 | D_ASSERT((s & (RQ_NET_MASK & ~RQ_NET_DONE)) == 0); |
b411b363 PR |
276 | |
277 | /* for writes we need to do some extra housekeeping */ | |
278 | if (rw == WRITE) | |
279 | _about_to_complete_local_write(mdev, req); | |
280 | ||
281 | /* Update disk stats */ | |
282 | _drbd_end_io_acct(mdev, req); | |
283 | ||
7be8da07 AG |
284 | if (!(s & RQ_POSTPONED)) { |
285 | m->error = ok ? 0 : (error ?: -EIO); | |
286 | m->bio = req->master_bio; | |
287 | } | |
b411b363 PR |
288 | req->master_bio = NULL; |
289 | } | |
290 | ||
291 | if ((s & RQ_NET_MASK) == 0 || (s & RQ_NET_DONE)) { | |
292 | /* this is disconnected (local only) operation, | |
293 | * or protocol C P_WRITE_ACK, | |
294 | * or protocol A or B P_BARRIER_ACK, | |
295 | * or killed from the transfer log due to connection loss. */ | |
296 | _req_is_done(mdev, req, rw); | |
297 | } | |
298 | /* else: network part and not DONE yet. that is | |
299 | * protocol A or B, barrier ack still pending... */ | |
300 | } | |
301 | ||
cfa03415 PR |
302 | static void _req_may_be_done_not_susp(struct drbd_request *req, struct bio_and_error *m) |
303 | { | |
a21e9298 | 304 | struct drbd_conf *mdev = req->w.mdev; |
cfa03415 | 305 | |
2aebfabb | 306 | if (!drbd_suspended(mdev)) |
cfa03415 PR |
307 | _req_may_be_done(req, m); |
308 | } | |
309 | ||
b411b363 PR |
310 | /* obviously this could be coded as many single functions |
311 | * instead of one huge switch, | |
312 | * or by putting the code directly in the respective locations | |
313 | * (as it has been before). | |
314 | * | |
315 | * but having it this way | |
316 | * enforces that it is all in this one place, where it is easier to audit, | |
317 | * it makes it obvious that whatever "event" "happens" to a request should | |
318 | * happen "atomically" within the req_lock, | |
319 | * and it enforces that we have to think in a very structured manner | |
320 | * about the "events" that may happen to a request during its life time ... | |
321 | */ | |
2a80699f | 322 | int __req_mod(struct drbd_request *req, enum drbd_req_event what, |
b411b363 PR |
323 | struct bio_and_error *m) |
324 | { | |
a21e9298 | 325 | struct drbd_conf *mdev = req->w.mdev; |
2a80699f | 326 | int rv = 0; |
7be8da07 AG |
327 | |
328 | if (m) | |
329 | m->bio = NULL; | |
b411b363 | 330 | |
b411b363 PR |
331 | switch (what) { |
332 | default: | |
333 | dev_err(DEV, "LOGIC BUG in %s:%u\n", __FILE__ , __LINE__); | |
334 | break; | |
335 | ||
336 | /* does not happen... | |
337 | * initialization done in drbd_req_new | |
8554df1c | 338 | case CREATED: |
b411b363 PR |
339 | break; |
340 | */ | |
341 | ||
8554df1c | 342 | case TO_BE_SENT: /* via network */ |
7be8da07 | 343 | /* reached via __drbd_make_request |
b411b363 PR |
344 | * and from w_read_retry_remote */ |
345 | D_ASSERT(!(req->rq_state & RQ_NET_MASK)); | |
346 | req->rq_state |= RQ_NET_PENDING; | |
347 | inc_ap_pending(mdev); | |
348 | break; | |
349 | ||
8554df1c | 350 | case TO_BE_SUBMITTED: /* locally */ |
7be8da07 | 351 | /* reached via __drbd_make_request */ |
b411b363 PR |
352 | D_ASSERT(!(req->rq_state & RQ_LOCAL_MASK)); |
353 | req->rq_state |= RQ_LOCAL_PENDING; | |
354 | break; | |
355 | ||
8554df1c | 356 | case COMPLETED_OK: |
b411b363 | 357 | if (bio_data_dir(req->master_bio) == WRITE) |
ace652ac | 358 | mdev->writ_cnt += req->i.size >> 9; |
b411b363 | 359 | else |
ace652ac | 360 | mdev->read_cnt += req->i.size >> 9; |
b411b363 PR |
361 | |
362 | req->rq_state |= (RQ_LOCAL_COMPLETED|RQ_LOCAL_OK); | |
363 | req->rq_state &= ~RQ_LOCAL_PENDING; | |
364 | ||
cfa03415 | 365 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
366 | put_ldev(mdev); |
367 | break; | |
368 | ||
8554df1c | 369 | case WRITE_COMPLETED_WITH_ERROR: |
b411b363 PR |
370 | req->rq_state |= RQ_LOCAL_COMPLETED; |
371 | req->rq_state &= ~RQ_LOCAL_PENDING; | |
372 | ||
81e84650 | 373 | __drbd_chk_io_error(mdev, false); |
cfa03415 | 374 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
375 | put_ldev(mdev); |
376 | break; | |
377 | ||
8554df1c | 378 | case READ_AHEAD_COMPLETED_WITH_ERROR: |
b411b363 PR |
379 | /* it is legal to fail READA */ |
380 | req->rq_state |= RQ_LOCAL_COMPLETED; | |
381 | req->rq_state &= ~RQ_LOCAL_PENDING; | |
cfa03415 | 382 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
383 | put_ldev(mdev); |
384 | break; | |
385 | ||
8554df1c | 386 | case READ_COMPLETED_WITH_ERROR: |
ace652ac | 387 | drbd_set_out_of_sync(mdev, req->i.sector, req->i.size); |
b411b363 PR |
388 | |
389 | req->rq_state |= RQ_LOCAL_COMPLETED; | |
390 | req->rq_state &= ~RQ_LOCAL_PENDING; | |
391 | ||
b411b363 | 392 | D_ASSERT(!(req->rq_state & RQ_NET_MASK)); |
b411b363 | 393 | |
81e84650 | 394 | __drbd_chk_io_error(mdev, false); |
b411b363 | 395 | put_ldev(mdev); |
b411b363 | 396 | |
d255e5ff LE |
397 | /* no point in retrying if there is no good remote data, |
398 | * or we have no connection. */ | |
399 | if (mdev->state.pdsk != D_UP_TO_DATE) { | |
cfa03415 | 400 | _req_may_be_done_not_susp(req, m); |
d255e5ff LE |
401 | break; |
402 | } | |
403 | ||
8554df1c | 404 | /* _req_mod(req,TO_BE_SENT); oops, recursion... */ |
d255e5ff LE |
405 | req->rq_state |= RQ_NET_PENDING; |
406 | inc_ap_pending(mdev); | |
8554df1c | 407 | /* fall through: _req_mod(req,QUEUE_FOR_NET_READ); */ |
b411b363 | 408 | |
8554df1c | 409 | case QUEUE_FOR_NET_READ: |
b411b363 PR |
410 | /* READ or READA, and |
411 | * no local disk, | |
412 | * or target area marked as invalid, | |
413 | * or just got an io-error. */ | |
7be8da07 | 414 | /* from __drbd_make_request |
b411b363 PR |
415 | * or from bio_endio during read io-error recovery */ |
416 | ||
417 | /* so we can verify the handle in the answer packet | |
418 | * corresponding hlist_del is in _req_may_be_done() */ | |
dac1389c | 419 | drbd_insert_interval(&mdev->read_requests, &req->i); |
b411b363 | 420 | |
83c38830 | 421 | set_bit(UNPLUG_REMOTE, &mdev->flags); |
b411b363 PR |
422 | |
423 | D_ASSERT(req->rq_state & RQ_NET_PENDING); | |
424 | req->rq_state |= RQ_NET_QUEUED; | |
425 | req->w.cb = (req->rq_state & RQ_LOCAL_MASK) | |
426 | ? w_read_retry_remote | |
427 | : w_send_read_req; | |
e42325a5 | 428 | drbd_queue_work(&mdev->tconn->data.work, &req->w); |
b411b363 PR |
429 | break; |
430 | ||
8554df1c | 431 | case QUEUE_FOR_NET_WRITE: |
b411b363 | 432 | /* assert something? */ |
7be8da07 | 433 | /* from __drbd_make_request only */ |
b411b363 | 434 | |
b411b363 | 435 | /* corresponding hlist_del is in _req_may_be_done() */ |
de696716 | 436 | drbd_insert_interval(&mdev->write_requests, &req->i); |
b411b363 PR |
437 | |
438 | /* NOTE | |
439 | * In case the req ended up on the transfer log before being | |
440 | * queued on the worker, it could lead to this request being | |
441 | * missed during cleanup after connection loss. | |
442 | * So we have to do both operations here, | |
443 | * within the same lock that protects the transfer log. | |
444 | * | |
445 | * _req_add_to_epoch(req); this has to be after the | |
446 | * _maybe_start_new_epoch(req); which happened in | |
7be8da07 | 447 | * __drbd_make_request, because we now may set the bit |
b411b363 PR |
448 | * again ourselves to close the current epoch. |
449 | * | |
450 | * Add req to the (now) current epoch (barrier). */ | |
451 | ||
83c38830 LE |
452 | /* otherwise we may lose an unplug, which may cause some remote |
453 | * io-scheduler timeout to expire, increasing maximum latency, | |
454 | * hurting performance. */ | |
455 | set_bit(UNPLUG_REMOTE, &mdev->flags); | |
456 | ||
7be8da07 | 457 | /* see __drbd_make_request, |
b411b363 PR |
458 | * just after it grabs the req_lock */ |
459 | D_ASSERT(test_bit(CREATE_BARRIER, &mdev->flags) == 0); | |
460 | ||
87eeee41 | 461 | req->epoch = mdev->tconn->newest_tle->br_number; |
b411b363 PR |
462 | |
463 | /* increment size of current epoch */ | |
87eeee41 | 464 | mdev->tconn->newest_tle->n_writes++; |
b411b363 PR |
465 | |
466 | /* queue work item to send data */ | |
467 | D_ASSERT(req->rq_state & RQ_NET_PENDING); | |
468 | req->rq_state |= RQ_NET_QUEUED; | |
469 | req->w.cb = w_send_dblock; | |
e42325a5 | 470 | drbd_queue_work(&mdev->tconn->data.work, &req->w); |
b411b363 PR |
471 | |
472 | /* close the epoch, in case it outgrew the limit */ | |
87eeee41 | 473 | if (mdev->tconn->newest_tle->n_writes >= mdev->tconn->net_conf->max_epoch_size) |
b411b363 PR |
474 | queue_barrier(mdev); |
475 | ||
476 | break; | |
477 | ||
8554df1c | 478 | case QUEUE_FOR_SEND_OOS: |
73a01a18 | 479 | req->rq_state |= RQ_NET_QUEUED; |
8f7bed77 | 480 | req->w.cb = w_send_out_of_sync; |
e42325a5 | 481 | drbd_queue_work(&mdev->tconn->data.work, &req->w); |
73a01a18 PR |
482 | break; |
483 | ||
8554df1c | 484 | case OOS_HANDED_TO_NETWORK: |
73a01a18 | 485 | /* actually the same */ |
8554df1c | 486 | case SEND_CANCELED: |
b411b363 | 487 | /* treat it the same */ |
8554df1c | 488 | case SEND_FAILED: |
b411b363 PR |
489 | /* real cleanup will be done from tl_clear. just update flags |
490 | * so it is no longer marked as on the worker queue */ | |
491 | req->rq_state &= ~RQ_NET_QUEUED; | |
492 | /* if we did it right, tl_clear should be scheduled only after | |
493 | * this, so this should not be necessary! */ | |
cfa03415 | 494 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
495 | break; |
496 | ||
8554df1c | 497 | case HANDED_OVER_TO_NETWORK: |
b411b363 | 498 | /* assert something? */ |
759fbdfb | 499 | if (bio_data_dir(req->master_bio) == WRITE) |
ace652ac | 500 | atomic_add(req->i.size >> 9, &mdev->ap_in_flight); |
759fbdfb | 501 | |
b411b363 | 502 | if (bio_data_dir(req->master_bio) == WRITE && |
89e58e75 | 503 | mdev->tconn->net_conf->wire_protocol == DRBD_PROT_A) { |
b411b363 PR |
504 | /* this is what is dangerous about protocol A: |
505 | * pretend it was successfully written on the peer. */ | |
506 | if (req->rq_state & RQ_NET_PENDING) { | |
507 | dec_ap_pending(mdev); | |
508 | req->rq_state &= ~RQ_NET_PENDING; | |
509 | req->rq_state |= RQ_NET_OK; | |
510 | } /* else: neg-ack was faster... */ | |
511 | /* it is still not yet RQ_NET_DONE until the | |
512 | * corresponding epoch barrier got acked as well, | |
513 | * so we know what to dirty on connection loss */ | |
514 | } | |
515 | req->rq_state &= ~RQ_NET_QUEUED; | |
516 | req->rq_state |= RQ_NET_SENT; | |
517 | /* because _drbd_send_zc_bio could sleep, and may want to | |
8554df1c AG |
518 | * dereference the bio even after the "WRITE_ACKED_BY_PEER" and |
519 | * "COMPLETED_OK" events came in, once we return from | |
b411b363 PR |
520 | * _drbd_send_zc_bio (drbd_send_dblock), we have to check |
521 | * whether it is done already, and end it. */ | |
cfa03415 | 522 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
523 | break; |
524 | ||
8554df1c | 525 | case READ_RETRY_REMOTE_CANCELED: |
d255e5ff LE |
526 | req->rq_state &= ~RQ_NET_QUEUED; |
527 | /* fall through, in case we raced with drbd_disconnect */ | |
8554df1c | 528 | case CONNECTION_LOST_WHILE_PENDING: |
b411b363 PR |
529 | /* transfer log cleanup after connection loss */ |
530 | /* assert something? */ | |
531 | if (req->rq_state & RQ_NET_PENDING) | |
532 | dec_ap_pending(mdev); | |
533 | req->rq_state &= ~(RQ_NET_OK|RQ_NET_PENDING); | |
534 | req->rq_state |= RQ_NET_DONE; | |
759fbdfb | 535 | if (req->rq_state & RQ_NET_SENT && req->rq_state & RQ_WRITE) |
ace652ac | 536 | atomic_sub(req->i.size >> 9, &mdev->ap_in_flight); |
759fbdfb | 537 | |
b411b363 PR |
538 | /* if it is still queued, we may not complete it here. |
539 | * it will be canceled soon. */ | |
540 | if (!(req->rq_state & RQ_NET_QUEUED)) | |
cfa03415 | 541 | _req_may_be_done(req, m); /* Allowed while state.susp */ |
b411b363 PR |
542 | break; |
543 | ||
8554df1c | 544 | case WRITE_ACKED_BY_PEER_AND_SIS: |
b411b363 | 545 | req->rq_state |= RQ_NET_SIS; |
7be8da07 | 546 | case DISCARD_WRITE: |
b411b363 PR |
547 | /* for discarded conflicting writes of multiple primaries, |
548 | * there is no need to keep anything in the tl, potential | |
549 | * node crashes are covered by the activity log. */ | |
b411b363 PR |
550 | req->rq_state |= RQ_NET_DONE; |
551 | /* fall through */ | |
8554df1c | 552 | case WRITE_ACKED_BY_PEER: |
b411b363 PR |
553 | /* protocol C; successfully written on peer. |
554 | * Nothing to do here. | |
555 | * We want to keep the tl in place for all protocols, to cater | |
556 | * for volatile write-back caches on lower level devices. | |
557 | * | |
558 | * A barrier request is expected to have forced all prior | |
559 | * requests onto stable storage, so completion of a barrier | |
560 | * request could set NET_DONE right here, and not wait for the | |
561 | * P_BARRIER_ACK, but that is an unnecessary optimization. */ | |
562 | ||
563 | /* this makes it effectively the same as for: */ | |
8554df1c | 564 | case RECV_ACKED_BY_PEER: |
b411b363 | 565 | /* protocol B; pretends to be successfully written on peer. |
8554df1c | 566 | * see also notes above in HANDED_OVER_TO_NETWORK about |
b411b363 PR |
567 | * protocol != C */ |
568 | req->rq_state |= RQ_NET_OK; | |
569 | D_ASSERT(req->rq_state & RQ_NET_PENDING); | |
570 | dec_ap_pending(mdev); | |
ace652ac | 571 | atomic_sub(req->i.size >> 9, &mdev->ap_in_flight); |
b411b363 | 572 | req->rq_state &= ~RQ_NET_PENDING; |
cfa03415 | 573 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
574 | break; |
575 | ||
7be8da07 AG |
576 | case POSTPONE_WRITE: |
577 | /* | |
578 | * If this node has already detected the write conflict, the | |
579 | * worker will be waiting on misc_wait. Wake it up once this | |
580 | * request has completed locally. | |
581 | */ | |
582 | D_ASSERT(req->rq_state & RQ_NET_PENDING); | |
583 | req->rq_state |= RQ_POSTPONED; | |
584 | _req_may_be_done_not_susp(req, m); | |
585 | break; | |
586 | ||
8554df1c | 587 | case NEG_ACKED: |
b411b363 | 588 | /* assert something? */ |
759fbdfb | 589 | if (req->rq_state & RQ_NET_PENDING) { |
b411b363 | 590 | dec_ap_pending(mdev); |
ace652ac | 591 | atomic_sub(req->i.size >> 9, &mdev->ap_in_flight); |
759fbdfb | 592 | } |
b411b363 PR |
593 | req->rq_state &= ~(RQ_NET_OK|RQ_NET_PENDING); |
594 | ||
595 | req->rq_state |= RQ_NET_DONE; | |
cfa03415 | 596 | _req_may_be_done_not_susp(req, m); |
8554df1c | 597 | /* else: done by HANDED_OVER_TO_NETWORK */ |
b411b363 PR |
598 | break; |
599 | ||
8554df1c | 600 | case FAIL_FROZEN_DISK_IO: |
265be2d0 PR |
601 | if (!(req->rq_state & RQ_LOCAL_COMPLETED)) |
602 | break; | |
603 | ||
cfa03415 | 604 | _req_may_be_done(req, m); /* Allowed while state.susp */ |
265be2d0 PR |
605 | break; |
606 | ||
8554df1c | 607 | case RESTART_FROZEN_DISK_IO: |
265be2d0 PR |
608 | if (!(req->rq_state & RQ_LOCAL_COMPLETED)) |
609 | break; | |
610 | ||
611 | req->rq_state &= ~RQ_LOCAL_COMPLETED; | |
612 | ||
613 | rv = MR_READ; | |
614 | if (bio_data_dir(req->master_bio) == WRITE) | |
615 | rv = MR_WRITE; | |
616 | ||
617 | get_ldev(mdev); | |
618 | req->w.cb = w_restart_disk_io; | |
e42325a5 | 619 | drbd_queue_work(&mdev->tconn->data.work, &req->w); |
265be2d0 PR |
620 | break; |
621 | ||
8554df1c | 622 | case RESEND: |
11b58e73 | 623 | /* If RQ_NET_OK is already set, we got a P_WRITE_ACK or P_RECV_ACK |
47ff2d0a | 624 | before the connection loss (B&C only); only P_BARRIER_ACK was missing. |
11b58e73 | 625 | Trowing them out of the TL here by pretending we got a BARRIER_ACK |
481c6f50 | 626 | We ensure that the peer was not rebooted */ |
11b58e73 PR |
627 | if (!(req->rq_state & RQ_NET_OK)) { |
628 | if (req->w.cb) { | |
e42325a5 | 629 | drbd_queue_work(&mdev->tconn->data.work, &req->w); |
11b58e73 PR |
630 | rv = req->rq_state & RQ_WRITE ? MR_WRITE : MR_READ; |
631 | } | |
632 | break; | |
633 | } | |
8554df1c | 634 | /* else, fall through to BARRIER_ACKED */ |
11b58e73 | 635 | |
8554df1c | 636 | case BARRIER_ACKED: |
288f422e PR |
637 | if (!(req->rq_state & RQ_WRITE)) |
638 | break; | |
639 | ||
b411b363 PR |
640 | if (req->rq_state & RQ_NET_PENDING) { |
641 | /* barrier came in before all requests have been acked. | |
642 | * this is bad, because if the connection is lost now, | |
643 | * we won't be able to clean them up... */ | |
8554df1c | 644 | dev_err(DEV, "FIXME (BARRIER_ACKED but pending)\n"); |
87eeee41 | 645 | list_move(&req->tl_requests, &mdev->tconn->out_of_sequence_requests); |
b411b363 | 646 | } |
e636db5b LE |
647 | if ((req->rq_state & RQ_NET_MASK) != 0) { |
648 | req->rq_state |= RQ_NET_DONE; | |
89e58e75 PR |
649 | if (mdev->tconn->net_conf->wire_protocol == DRBD_PROT_A) |
650 | atomic_sub(req->i.size>>9, &mdev->ap_in_flight); | |
e636db5b | 651 | } |
cfa03415 | 652 | _req_may_be_done(req, m); /* Allowed while state.susp */ |
b411b363 PR |
653 | break; |
654 | ||
8554df1c | 655 | case DATA_RECEIVED: |
b411b363 PR |
656 | D_ASSERT(req->rq_state & RQ_NET_PENDING); |
657 | dec_ap_pending(mdev); | |
658 | req->rq_state &= ~RQ_NET_PENDING; | |
659 | req->rq_state |= (RQ_NET_OK|RQ_NET_DONE); | |
cfa03415 | 660 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
661 | break; |
662 | }; | |
2a80699f PR |
663 | |
664 | return rv; | |
b411b363 PR |
665 | } |
666 | ||
667 | /* we may do a local read if: | |
668 | * - we are consistent (of course), | |
669 | * - or we are generally inconsistent, | |
670 | * BUT we are still/already IN SYNC for this area. | |
671 | * since size may be bigger than BM_BLOCK_SIZE, | |
672 | * we may need to check several bits. | |
673 | */ | |
0da34df0 | 674 | static bool drbd_may_do_local_read(struct drbd_conf *mdev, sector_t sector, int size) |
b411b363 PR |
675 | { |
676 | unsigned long sbnr, ebnr; | |
677 | sector_t esector, nr_sectors; | |
678 | ||
679 | if (mdev->state.disk == D_UP_TO_DATE) | |
0da34df0 | 680 | return true; |
8c387def | 681 | if (mdev->state.disk != D_INCONSISTENT) |
0da34df0 | 682 | return false; |
b411b363 | 683 | esector = sector + (size >> 9) - 1; |
8ca9844f | 684 | nr_sectors = drbd_get_capacity(mdev->this_bdev); |
b411b363 PR |
685 | D_ASSERT(sector < nr_sectors); |
686 | D_ASSERT(esector < nr_sectors); | |
687 | ||
688 | sbnr = BM_SECT_TO_BIT(sector); | |
689 | ebnr = BM_SECT_TO_BIT(esector); | |
690 | ||
0da34df0 | 691 | return drbd_bm_count_bits(mdev, sbnr, ebnr) == 0; |
b411b363 PR |
692 | } |
693 | ||
6024fece AG |
694 | /* |
695 | * complete_conflicting_writes - wait for any conflicting write requests | |
696 | * | |
697 | * The write_requests tree contains all active write requests which we | |
698 | * currently know about. Wait for any requests to complete which conflict with | |
699 | * the new one. | |
700 | */ | |
701 | static int complete_conflicting_writes(struct drbd_conf *mdev, | |
702 | sector_t sector, int size) | |
703 | { | |
704 | for(;;) { | |
6024fece | 705 | struct drbd_interval *i; |
7be8da07 | 706 | int err; |
6024fece AG |
707 | |
708 | i = drbd_find_overlap(&mdev->write_requests, sector, size); | |
709 | if (!i) | |
710 | return 0; | |
7be8da07 AG |
711 | err = drbd_wait_misc(mdev, i); |
712 | if (err) | |
713 | return err; | |
6024fece AG |
714 | } |
715 | } | |
716 | ||
7be8da07 | 717 | int __drbd_make_request(struct drbd_conf *mdev, struct bio *bio, unsigned long start_time) |
b411b363 PR |
718 | { |
719 | const int rw = bio_rw(bio); | |
720 | const int size = bio->bi_size; | |
721 | const sector_t sector = bio->bi_sector; | |
722 | struct drbd_tl_epoch *b = NULL; | |
723 | struct drbd_request *req; | |
73a01a18 | 724 | int local, remote, send_oos = 0; |
6024fece | 725 | int err; |
9a25a04c | 726 | int ret = 0; |
b411b363 PR |
727 | |
728 | /* allocate outside of all locks; */ | |
729 | req = drbd_req_new(mdev, bio); | |
730 | if (!req) { | |
731 | dec_ap_bio(mdev); | |
732 | /* only pass the error to the upper layers. | |
733 | * if user cannot handle io errors, that's not our business. */ | |
734 | dev_err(DEV, "could not kmalloc() req\n"); | |
735 | bio_endio(bio, -ENOMEM); | |
736 | return 0; | |
737 | } | |
aeda1cd6 | 738 | req->start_time = start_time; |
b411b363 | 739 | |
b411b363 PR |
740 | local = get_ldev(mdev); |
741 | if (!local) { | |
742 | bio_put(req->private_bio); /* or we get a bio leak */ | |
743 | req->private_bio = NULL; | |
744 | } | |
745 | if (rw == WRITE) { | |
746 | remote = 1; | |
747 | } else { | |
748 | /* READ || READA */ | |
749 | if (local) { | |
750 | if (!drbd_may_do_local_read(mdev, sector, size)) { | |
751 | /* we could kick the syncer to | |
752 | * sync this extent asap, wait for | |
753 | * it, then continue locally. | |
754 | * Or just issue the request remotely. | |
755 | */ | |
756 | local = 0; | |
757 | bio_put(req->private_bio); | |
758 | req->private_bio = NULL; | |
759 | put_ldev(mdev); | |
760 | } | |
761 | } | |
762 | remote = !local && mdev->state.pdsk >= D_UP_TO_DATE; | |
763 | } | |
764 | ||
765 | /* If we have a disk, but a READA request is mapped to remote, | |
766 | * we are R_PRIMARY, D_INCONSISTENT, SyncTarget. | |
767 | * Just fail that READA request right here. | |
768 | * | |
769 | * THINK: maybe fail all READA when not local? | |
770 | * or make this configurable... | |
771 | * if network is slow, READA won't do any good. | |
772 | */ | |
773 | if (rw == READA && mdev->state.disk >= D_INCONSISTENT && !local) { | |
774 | err = -EWOULDBLOCK; | |
775 | goto fail_and_free_req; | |
776 | } | |
777 | ||
778 | /* For WRITES going to the local disk, grab a reference on the target | |
779 | * extent. This waits for any resync activity in the corresponding | |
780 | * resync extent to finish, and, if necessary, pulls in the target | |
781 | * extent into the activity log, which involves further disk io because | |
782 | * of transactional on-disk meta data updates. */ | |
0778286a PR |
783 | if (rw == WRITE && local && !test_bit(AL_SUSPENDED, &mdev->flags)) { |
784 | req->rq_state |= RQ_IN_ACT_LOG; | |
181286ad | 785 | drbd_al_begin_io(mdev, &req->i); |
0778286a | 786 | } |
b411b363 | 787 | |
6a35c45f | 788 | remote = remote && drbd_should_do_remote(mdev->state); |
8f7bed77 | 789 | send_oos = rw == WRITE && drbd_should_send_out_of_sync(mdev->state); |
3719094e | 790 | D_ASSERT(!(remote && send_oos)); |
b411b363 | 791 | |
2aebfabb | 792 | if (!(local || remote) && !drbd_suspended(mdev)) { |
fb2c7a10 LE |
793 | if (__ratelimit(&drbd_ratelimit_state)) |
794 | dev_err(DEV, "IO ERROR: neither local nor remote disk\n"); | |
6024fece | 795 | err = -EIO; |
b411b363 PR |
796 | goto fail_free_complete; |
797 | } | |
798 | ||
799 | /* For WRITE request, we have to make sure that we have an | |
800 | * unused_spare_tle, in case we need to start a new epoch. | |
801 | * I try to be smart and avoid to pre-allocate always "just in case", | |
802 | * but there is a race between testing the bit and pointer outside the | |
803 | * spinlock, and grabbing the spinlock. | |
804 | * if we lost that race, we retry. */ | |
73a01a18 | 805 | if (rw == WRITE && (remote || send_oos) && |
87eeee41 | 806 | mdev->tconn->unused_spare_tle == NULL && |
b411b363 PR |
807 | test_bit(CREATE_BARRIER, &mdev->flags)) { |
808 | allocate_barrier: | |
809 | b = kmalloc(sizeof(struct drbd_tl_epoch), GFP_NOIO); | |
810 | if (!b) { | |
811 | dev_err(DEV, "Failed to alloc barrier.\n"); | |
812 | err = -ENOMEM; | |
813 | goto fail_free_complete; | |
814 | } | |
815 | } | |
816 | ||
817 | /* GOOD, everything prepared, grab the spin_lock */ | |
87eeee41 | 818 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 819 | |
6024fece AG |
820 | if (rw == WRITE) { |
821 | err = complete_conflicting_writes(mdev, sector, size); | |
822 | if (err) { | |
7be8da07 AG |
823 | if (err != -ERESTARTSYS) |
824 | _conn_request_state(mdev->tconn, | |
825 | NS(conn, C_TIMEOUT), | |
826 | CS_HARD); | |
6024fece | 827 | spin_unlock_irq(&mdev->tconn->req_lock); |
7be8da07 | 828 | err = -EIO; |
6024fece AG |
829 | goto fail_free_complete; |
830 | } | |
831 | } | |
832 | ||
2aebfabb | 833 | if (drbd_suspended(mdev)) { |
9a25a04c PR |
834 | /* If we got suspended, use the retry mechanism of |
835 | generic_make_request() to restart processing of this | |
2f58dcfc | 836 | bio. In the next call to drbd_make_request |
9a25a04c PR |
837 | we sleep in inc_ap_bio() */ |
838 | ret = 1; | |
87eeee41 | 839 | spin_unlock_irq(&mdev->tconn->req_lock); |
9a25a04c PR |
840 | goto fail_free_complete; |
841 | } | |
842 | ||
73a01a18 | 843 | if (remote || send_oos) { |
6a35c45f | 844 | remote = drbd_should_do_remote(mdev->state); |
8f7bed77 | 845 | send_oos = rw == WRITE && drbd_should_send_out_of_sync(mdev->state); |
3719094e | 846 | D_ASSERT(!(remote && send_oos)); |
73a01a18 PR |
847 | |
848 | if (!(remote || send_oos)) | |
b411b363 PR |
849 | dev_warn(DEV, "lost connection while grabbing the req_lock!\n"); |
850 | if (!(local || remote)) { | |
851 | dev_err(DEV, "IO ERROR: neither local nor remote disk\n"); | |
87eeee41 | 852 | spin_unlock_irq(&mdev->tconn->req_lock); |
6024fece | 853 | err = -EIO; |
b411b363 PR |
854 | goto fail_free_complete; |
855 | } | |
856 | } | |
857 | ||
87eeee41 PR |
858 | if (b && mdev->tconn->unused_spare_tle == NULL) { |
859 | mdev->tconn->unused_spare_tle = b; | |
b411b363 PR |
860 | b = NULL; |
861 | } | |
73a01a18 | 862 | if (rw == WRITE && (remote || send_oos) && |
87eeee41 | 863 | mdev->tconn->unused_spare_tle == NULL && |
b411b363 PR |
864 | test_bit(CREATE_BARRIER, &mdev->flags)) { |
865 | /* someone closed the current epoch | |
866 | * while we were grabbing the spinlock */ | |
87eeee41 | 867 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
868 | goto allocate_barrier; |
869 | } | |
870 | ||
871 | ||
872 | /* Update disk stats */ | |
873 | _drbd_start_io_acct(mdev, req, bio); | |
874 | ||
875 | /* _maybe_start_new_epoch(mdev); | |
876 | * If we need to generate a write barrier packet, we have to add the | |
877 | * new epoch (barrier) object, and queue the barrier packet for sending, | |
878 | * and queue the req's data after it _within the same lock_, otherwise | |
879 | * we have race conditions were the reorder domains could be mixed up. | |
880 | * | |
881 | * Even read requests may start a new epoch and queue the corresponding | |
882 | * barrier packet. To get the write ordering right, we only have to | |
883 | * make sure that, if this is a write request and it triggered a | |
884 | * barrier packet, this request is queued within the same spinlock. */ | |
87eeee41 | 885 | if ((remote || send_oos) && mdev->tconn->unused_spare_tle && |
b411b363 | 886 | test_and_clear_bit(CREATE_BARRIER, &mdev->flags)) { |
2f5cdd0b | 887 | _tl_add_barrier(mdev->tconn, mdev->tconn->unused_spare_tle); |
87eeee41 | 888 | mdev->tconn->unused_spare_tle = NULL; |
b411b363 PR |
889 | } else { |
890 | D_ASSERT(!(remote && rw == WRITE && | |
891 | test_bit(CREATE_BARRIER, &mdev->flags))); | |
892 | } | |
893 | ||
894 | /* NOTE | |
895 | * Actually, 'local' may be wrong here already, since we may have failed | |
896 | * to write to the meta data, and may become wrong anytime because of | |
897 | * local io-error for some other request, which would lead to us | |
898 | * "detaching" the local disk. | |
899 | * | |
900 | * 'remote' may become wrong any time because the network could fail. | |
901 | * | |
902 | * This is a harmless race condition, though, since it is handled | |
903 | * correctly at the appropriate places; so it just defers the failure | |
904 | * of the respective operation. | |
905 | */ | |
906 | ||
907 | /* mark them early for readability. | |
908 | * this just sets some state flags. */ | |
909 | if (remote) | |
8554df1c | 910 | _req_mod(req, TO_BE_SENT); |
b411b363 | 911 | if (local) |
8554df1c | 912 | _req_mod(req, TO_BE_SUBMITTED); |
b411b363 | 913 | |
87eeee41 | 914 | list_add_tail(&req->tl_requests, &mdev->tconn->newest_tle->requests); |
288f422e | 915 | |
b411b363 PR |
916 | /* NOTE remote first: to get the concurrent write detection right, |
917 | * we must register the request before start of local IO. */ | |
918 | if (remote) { | |
919 | /* either WRITE and C_CONNECTED, | |
920 | * or READ, and no local disk, | |
921 | * or READ, but not in sync. | |
922 | */ | |
923 | _req_mod(req, (rw == WRITE) | |
8554df1c AG |
924 | ? QUEUE_FOR_NET_WRITE |
925 | : QUEUE_FOR_NET_READ); | |
b411b363 | 926 | } |
73a01a18 | 927 | if (send_oos && drbd_set_out_of_sync(mdev, sector, size)) |
8554df1c | 928 | _req_mod(req, QUEUE_FOR_SEND_OOS); |
67531718 | 929 | |
73a01a18 | 930 | if (remote && |
31890f4a | 931 | mdev->tconn->net_conf->on_congestion != OC_BLOCK && mdev->tconn->agreed_pro_version >= 96) { |
67531718 PR |
932 | int congested = 0; |
933 | ||
89e58e75 PR |
934 | if (mdev->tconn->net_conf->cong_fill && |
935 | atomic_read(&mdev->ap_in_flight) >= mdev->tconn->net_conf->cong_fill) { | |
67531718 PR |
936 | dev_info(DEV, "Congestion-fill threshold reached\n"); |
937 | congested = 1; | |
938 | } | |
939 | ||
89e58e75 | 940 | if (mdev->act_log->used >= mdev->tconn->net_conf->cong_extents) { |
67531718 PR |
941 | dev_info(DEV, "Congestion-extents threshold reached\n"); |
942 | congested = 1; | |
943 | } | |
944 | ||
71c78cfb | 945 | if (congested) { |
039312b6 | 946 | queue_barrier(mdev); /* last barrier, after mirrored writes */ |
73a01a18 | 947 | |
89e58e75 | 948 | if (mdev->tconn->net_conf->on_congestion == OC_PULL_AHEAD) |
67531718 | 949 | _drbd_set_state(_NS(mdev, conn, C_AHEAD), 0, NULL); |
89e58e75 | 950 | else /*mdev->tconn->net_conf->on_congestion == OC_DISCONNECT */ |
67531718 PR |
951 | _drbd_set_state(_NS(mdev, conn, C_DISCONNECTING), 0, NULL); |
952 | } | |
953 | } | |
954 | ||
87eeee41 | 955 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
956 | kfree(b); /* if someone else has beaten us to it... */ |
957 | ||
958 | if (local) { | |
959 | req->private_bio->bi_bdev = mdev->ldev->backing_bdev; | |
960 | ||
6719fb03 LE |
961 | /* State may have changed since we grabbed our reference on the |
962 | * mdev->ldev member. Double check, and short-circuit to endio. | |
963 | * In case the last activity log transaction failed to get on | |
964 | * stable storage, and this is a WRITE, we may not even submit | |
965 | * this bio. */ | |
966 | if (get_ldev(mdev)) { | |
0cf9d27e AG |
967 | if (drbd_insert_fault(mdev, rw == WRITE ? DRBD_FAULT_DT_WR |
968 | : rw == READ ? DRBD_FAULT_DT_RD | |
969 | : DRBD_FAULT_DT_RA)) | |
6719fb03 LE |
970 | bio_endio(req->private_bio, -EIO); |
971 | else | |
972 | generic_make_request(req->private_bio); | |
973 | put_ldev(mdev); | |
974 | } else | |
b411b363 | 975 | bio_endio(req->private_bio, -EIO); |
b411b363 PR |
976 | } |
977 | ||
b411b363 PR |
978 | return 0; |
979 | ||
980 | fail_free_complete: | |
76727f68 | 981 | if (req->rq_state & RQ_IN_ACT_LOG) |
181286ad | 982 | drbd_al_complete_io(mdev, &req->i); |
b411b363 PR |
983 | fail_and_free_req: |
984 | if (local) { | |
985 | bio_put(req->private_bio); | |
986 | req->private_bio = NULL; | |
987 | put_ldev(mdev); | |
988 | } | |
9a25a04c PR |
989 | if (!ret) |
990 | bio_endio(bio, err); | |
991 | ||
b411b363 PR |
992 | drbd_req_free(req); |
993 | dec_ap_bio(mdev); | |
994 | kfree(b); | |
995 | ||
9a25a04c | 996 | return ret; |
b411b363 PR |
997 | } |
998 | ||
2f58dcfc | 999 | int drbd_make_request(struct request_queue *q, struct bio *bio) |
b411b363 | 1000 | { |
b411b363 | 1001 | struct drbd_conf *mdev = (struct drbd_conf *) q->queuedata; |
aeda1cd6 | 1002 | unsigned long start_time; |
b411b363 | 1003 | |
aeda1cd6 PR |
1004 | start_time = jiffies; |
1005 | ||
b411b363 PR |
1006 | /* |
1007 | * what we "blindly" assume: | |
1008 | */ | |
1009 | D_ASSERT(bio->bi_size > 0); | |
c670a398 | 1010 | D_ASSERT(IS_ALIGNED(bio->bi_size, 512)); |
b411b363 | 1011 | |
23361cf3 LE |
1012 | inc_ap_bio(mdev); |
1013 | return __drbd_make_request(mdev, bio, start_time); | |
b411b363 PR |
1014 | } |
1015 | ||
23361cf3 LE |
1016 | /* This is called by bio_add_page(). |
1017 | * | |
1018 | * q->max_hw_sectors and other global limits are already enforced there. | |
b411b363 | 1019 | * |
23361cf3 LE |
1020 | * We need to call down to our lower level device, |
1021 | * in case it has special restrictions. | |
1022 | * | |
1023 | * We also may need to enforce configured max-bio-bvecs limits. | |
b411b363 PR |
1024 | * |
1025 | * As long as the BIO is empty we have to allow at least one bvec, | |
23361cf3 | 1026 | * regardless of size and offset, so no need to ask lower levels. |
b411b363 PR |
1027 | */ |
1028 | int drbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, struct bio_vec *bvec) | |
1029 | { | |
1030 | struct drbd_conf *mdev = (struct drbd_conf *) q->queuedata; | |
b411b363 | 1031 | unsigned int bio_size = bvm->bi_size; |
23361cf3 LE |
1032 | int limit = DRBD_MAX_BIO_SIZE; |
1033 | int backing_limit; | |
1034 | ||
1035 | if (bio_size && get_ldev(mdev)) { | |
b411b363 PR |
1036 | struct request_queue * const b = |
1037 | mdev->ldev->backing_bdev->bd_disk->queue; | |
a1c88d0d | 1038 | if (b->merge_bvec_fn) { |
b411b363 PR |
1039 | backing_limit = b->merge_bvec_fn(b, bvm, bvec); |
1040 | limit = min(limit, backing_limit); | |
1041 | } | |
1042 | put_ldev(mdev); | |
1043 | } | |
1044 | return limit; | |
1045 | } | |
7fde2be9 PR |
1046 | |
1047 | void request_timer_fn(unsigned long data) | |
1048 | { | |
1049 | struct drbd_conf *mdev = (struct drbd_conf *) data; | |
8b924f1d | 1050 | struct drbd_tconn *tconn = mdev->tconn; |
7fde2be9 PR |
1051 | struct drbd_request *req; /* oldest request */ |
1052 | struct list_head *le; | |
1053 | unsigned long et = 0; /* effective timeout = ko_count * timeout */ | |
1054 | ||
8b924f1d PR |
1055 | if (get_net_conf(tconn)) { |
1056 | et = tconn->net_conf->timeout*HZ/10 * tconn->net_conf->ko_count; | |
1057 | put_net_conf(tconn); | |
7fde2be9 PR |
1058 | } |
1059 | if (!et || mdev->state.conn < C_WF_REPORT_PARAMS) | |
1060 | return; /* Recurring timer stopped */ | |
1061 | ||
8b924f1d PR |
1062 | spin_lock_irq(&tconn->req_lock); |
1063 | le = &tconn->oldest_tle->requests; | |
7fde2be9 | 1064 | if (list_empty(le)) { |
8b924f1d | 1065 | spin_unlock_irq(&tconn->req_lock); |
7fde2be9 PR |
1066 | mod_timer(&mdev->request_timer, jiffies + et); |
1067 | return; | |
1068 | } | |
1069 | ||
1070 | le = le->prev; | |
1071 | req = list_entry(le, struct drbd_request, tl_requests); | |
1072 | if (time_is_before_eq_jiffies(req->start_time + et)) { | |
1073 | if (req->rq_state & RQ_NET_PENDING) { | |
1074 | dev_warn(DEV, "Remote failed to finish a request within ko-count * timeout\n"); | |
1075 | _drbd_set_state(_NS(mdev, conn, C_TIMEOUT), CS_VERBOSE, NULL); | |
1076 | } else { | |
1077 | dev_warn(DEV, "Local backing block device frozen?\n"); | |
1078 | mod_timer(&mdev->request_timer, jiffies + et); | |
1079 | } | |
1080 | } else { | |
1081 | mod_timer(&mdev->request_timer, req->start_time + et); | |
1082 | } | |
1083 | ||
8b924f1d | 1084 | spin_unlock_irq(&tconn->req_lock); |
7fde2be9 | 1085 | } |