]>
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 | ||
59 | static void _req_is_done(struct drbd_conf *mdev, struct drbd_request *req, const int rw) | |
60 | { | |
61 | const unsigned long s = req->rq_state; | |
288f422e PR |
62 | |
63 | /* remove it from the transfer log. | |
64 | * well, only if it had been there in the first | |
65 | * place... if it had not (local only or conflicting | |
66 | * and never sent), it should still be "empty" as | |
67 | * initialized in drbd_req_new(), so we can list_del() it | |
68 | * here unconditionally */ | |
69 | list_del(&req->tl_requests); | |
70 | ||
b411b363 PR |
71 | /* if it was a write, we may have to set the corresponding |
72 | * bit(s) out-of-sync first. If it had a local part, we need to | |
73 | * release the reference to the activity log. */ | |
74 | if (rw == WRITE) { | |
b411b363 PR |
75 | /* Set out-of-sync unless both OK flags are set |
76 | * (local only or remote failed). | |
77 | * Other places where we set out-of-sync: | |
78 | * READ with local io-error */ | |
79 | if (!(s & RQ_NET_OK) || !(s & RQ_LOCAL_OK)) | |
80 | drbd_set_out_of_sync(mdev, req->sector, req->size); | |
81 | ||
82 | if ((s & RQ_NET_OK) && (s & RQ_LOCAL_OK) && (s & RQ_NET_SIS)) | |
83 | drbd_set_in_sync(mdev, req->sector, req->size); | |
84 | ||
85 | /* one might be tempted to move the drbd_al_complete_io | |
86 | * to the local io completion callback drbd_endio_pri. | |
87 | * but, if this was a mirror write, we may only | |
88 | * drbd_al_complete_io after this is RQ_NET_DONE, | |
89 | * otherwise the extent could be dropped from the al | |
90 | * before it has actually been written on the peer. | |
91 | * if we crash before our peer knows about the request, | |
92 | * but after the extent has been dropped from the al, | |
93 | * we would forget to resync the corresponding extent. | |
94 | */ | |
95 | if (s & RQ_LOCAL_MASK) { | |
96 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
0778286a PR |
97 | if (s & RQ_IN_ACT_LOG) |
98 | drbd_al_complete_io(mdev, req->sector); | |
b411b363 PR |
99 | put_ldev(mdev); |
100 | } else if (__ratelimit(&drbd_ratelimit_state)) { | |
101 | dev_warn(DEV, "Should have called drbd_al_complete_io(, %llu), " | |
102 | "but my Disk seems to have failed :(\n", | |
103 | (unsigned long long) req->sector); | |
104 | } | |
105 | } | |
106 | } | |
107 | ||
32fa7e91 | 108 | drbd_req_free(req); |
b411b363 PR |
109 | } |
110 | ||
111 | static void queue_barrier(struct drbd_conf *mdev) | |
112 | { | |
113 | struct drbd_tl_epoch *b; | |
114 | ||
115 | /* We are within the req_lock. Once we queued the barrier for sending, | |
116 | * we set the CREATE_BARRIER bit. It is cleared as soon as a new | |
117 | * barrier/epoch object is added. This is the only place this bit is | |
118 | * set. It indicates that the barrier for this epoch is already queued, | |
119 | * and no new epoch has been created yet. */ | |
120 | if (test_bit(CREATE_BARRIER, &mdev->flags)) | |
121 | return; | |
122 | ||
123 | b = mdev->newest_tle; | |
124 | b->w.cb = w_send_barrier; | |
125 | /* inc_ap_pending done here, so we won't | |
126 | * get imbalanced on connection loss. | |
127 | * dec_ap_pending will be done in got_BarrierAck | |
128 | * or (on connection loss) in tl_clear. */ | |
129 | inc_ap_pending(mdev); | |
130 | drbd_queue_work(&mdev->data.work, &b->w); | |
131 | set_bit(CREATE_BARRIER, &mdev->flags); | |
132 | } | |
133 | ||
134 | static void _about_to_complete_local_write(struct drbd_conf *mdev, | |
135 | struct drbd_request *req) | |
136 | { | |
137 | const unsigned long s = req->rq_state; | |
138 | struct drbd_request *i; | |
139 | struct drbd_epoch_entry *e; | |
140 | struct hlist_node *n; | |
141 | struct hlist_head *slot; | |
142 | ||
143 | /* before we can signal completion to the upper layers, | |
144 | * we may need to close the current epoch */ | |
145 | if (mdev->state.conn >= C_CONNECTED && | |
146 | req->epoch == mdev->newest_tle->br_number) | |
147 | queue_barrier(mdev); | |
148 | ||
149 | /* we need to do the conflict detection stuff, | |
150 | * if we have the ee_hash (two_primaries) and | |
151 | * this has been on the network */ | |
152 | if ((s & RQ_NET_DONE) && mdev->ee_hash != NULL) { | |
153 | const sector_t sector = req->sector; | |
154 | const int size = req->size; | |
155 | ||
156 | /* ASSERT: | |
157 | * there must be no conflicting requests, since | |
158 | * they must have been failed on the spot */ | |
159 | #define OVERLAPS overlaps(sector, size, i->sector, i->size) | |
160 | slot = tl_hash_slot(mdev, sector); | |
161 | hlist_for_each_entry(i, n, slot, colision) { | |
162 | if (OVERLAPS) { | |
163 | dev_alert(DEV, "LOGIC BUG: completed: %p %llus +%u; " | |
164 | "other: %p %llus +%u\n", | |
165 | req, (unsigned long long)sector, size, | |
166 | i, (unsigned long long)i->sector, i->size); | |
167 | } | |
168 | } | |
169 | ||
170 | /* maybe "wake" those conflicting epoch entries | |
171 | * that wait for this request to finish. | |
172 | * | |
173 | * currently, there can be only _one_ such ee | |
174 | * (well, or some more, which would be pending | |
175 | * P_DISCARD_ACK not yet sent by the asender...), | |
176 | * since we block the receiver thread upon the | |
177 | * first conflict detection, which will wait on | |
178 | * misc_wait. maybe we want to assert that? | |
179 | * | |
180 | * anyways, if we found one, | |
181 | * we just have to do a wake_up. */ | |
182 | #undef OVERLAPS | |
183 | #define OVERLAPS overlaps(sector, size, e->sector, e->size) | |
184 | slot = ee_hash_slot(mdev, req->sector); | |
185 | hlist_for_each_entry(e, n, slot, colision) { | |
186 | if (OVERLAPS) { | |
187 | wake_up(&mdev->misc_wait); | |
188 | break; | |
189 | } | |
190 | } | |
191 | } | |
192 | #undef OVERLAPS | |
193 | } | |
194 | ||
195 | void complete_master_bio(struct drbd_conf *mdev, | |
196 | struct bio_and_error *m) | |
197 | { | |
b411b363 PR |
198 | bio_endio(m->bio, m->error); |
199 | dec_ap_bio(mdev); | |
200 | } | |
201 | ||
202 | /* Helper for __req_mod(). | |
203 | * Set m->bio to the master bio, if it is fit to be completed, | |
204 | * or leave it alone (it is initialized to NULL in __req_mod), | |
205 | * if it has already been completed, or cannot be completed yet. | |
206 | * If m->bio is set, the error status to be returned is placed in m->error. | |
207 | */ | |
208 | void _req_may_be_done(struct drbd_request *req, struct bio_and_error *m) | |
209 | { | |
210 | const unsigned long s = req->rq_state; | |
211 | struct drbd_conf *mdev = req->mdev; | |
212 | /* only WRITES may end up here without a master bio (on barrier ack) */ | |
213 | int rw = req->master_bio ? bio_data_dir(req->master_bio) : WRITE; | |
214 | ||
b411b363 PR |
215 | /* we must not complete the master bio, while it is |
216 | * still being processed by _drbd_send_zc_bio (drbd_send_dblock) | |
217 | * not yet acknowledged by the peer | |
218 | * not yet completed by the local io subsystem | |
219 | * these flags may get cleared in any order by | |
220 | * the worker, | |
221 | * the receiver, | |
222 | * the bio_endio completion callbacks. | |
223 | */ | |
224 | if (s & RQ_NET_QUEUED) | |
225 | return; | |
226 | if (s & RQ_NET_PENDING) | |
227 | return; | |
228 | if (s & RQ_LOCAL_PENDING) | |
229 | return; | |
230 | ||
231 | if (req->master_bio) { | |
232 | /* this is data_received (remote read) | |
233 | * or protocol C P_WRITE_ACK | |
234 | * or protocol B P_RECV_ACK | |
235 | * or protocol A "handed_over_to_network" (SendAck) | |
236 | * or canceled or failed, | |
237 | * or killed from the transfer log due to connection loss. | |
238 | */ | |
239 | ||
240 | /* | |
241 | * figure out whether to report success or failure. | |
242 | * | |
243 | * report success when at least one of the operations succeeded. | |
244 | * or, to put the other way, | |
245 | * only report failure, when both operations failed. | |
246 | * | |
247 | * what to do about the failures is handled elsewhere. | |
248 | * what we need to do here is just: complete the master_bio. | |
249 | * | |
250 | * local completion error, if any, has been stored as ERR_PTR | |
251 | * in private_bio within drbd_endio_pri. | |
252 | */ | |
253 | int ok = (s & RQ_LOCAL_OK) || (s & RQ_NET_OK); | |
254 | int error = PTR_ERR(req->private_bio); | |
255 | ||
256 | /* remove the request from the conflict detection | |
257 | * respective block_id verification hash */ | |
258 | if (!hlist_unhashed(&req->colision)) | |
259 | hlist_del(&req->colision); | |
260 | else | |
261 | D_ASSERT((s & RQ_NET_MASK) == 0); | |
262 | ||
263 | /* for writes we need to do some extra housekeeping */ | |
264 | if (rw == WRITE) | |
265 | _about_to_complete_local_write(mdev, req); | |
266 | ||
267 | /* Update disk stats */ | |
268 | _drbd_end_io_acct(mdev, req); | |
269 | ||
270 | m->error = ok ? 0 : (error ?: -EIO); | |
271 | m->bio = req->master_bio; | |
272 | req->master_bio = NULL; | |
273 | } | |
274 | ||
275 | if ((s & RQ_NET_MASK) == 0 || (s & RQ_NET_DONE)) { | |
276 | /* this is disconnected (local only) operation, | |
277 | * or protocol C P_WRITE_ACK, | |
278 | * or protocol A or B P_BARRIER_ACK, | |
279 | * or killed from the transfer log due to connection loss. */ | |
280 | _req_is_done(mdev, req, rw); | |
281 | } | |
282 | /* else: network part and not DONE yet. that is | |
283 | * protocol A or B, barrier ack still pending... */ | |
284 | } | |
285 | ||
cfa03415 PR |
286 | static void _req_may_be_done_not_susp(struct drbd_request *req, struct bio_and_error *m) |
287 | { | |
288 | struct drbd_conf *mdev = req->mdev; | |
289 | ||
290 | if (!mdev->state.susp) | |
291 | _req_may_be_done(req, m); | |
292 | } | |
293 | ||
b411b363 PR |
294 | /* |
295 | * checks whether there was an overlapping request | |
296 | * or ee already registered. | |
297 | * | |
298 | * if so, return 1, in which case this request is completed on the spot, | |
299 | * without ever being submitted or send. | |
300 | * | |
301 | * return 0 if it is ok to submit this request. | |
302 | * | |
303 | * NOTE: | |
304 | * paranoia: assume something above us is broken, and issues different write | |
305 | * requests for the same block simultaneously... | |
306 | * | |
307 | * To ensure these won't be reordered differently on both nodes, resulting in | |
308 | * diverging data sets, we discard the later one(s). Not that this is supposed | |
309 | * to happen, but this is the rationale why we also have to check for | |
310 | * conflicting requests with local origin, and why we have to do so regardless | |
311 | * of whether we allowed multiple primaries. | |
312 | * | |
313 | * BTW, in case we only have one primary, the ee_hash is empty anyways, and the | |
314 | * second hlist_for_each_entry becomes a noop. This is even simpler than to | |
315 | * grab a reference on the net_conf, and check for the two_primaries flag... | |
316 | */ | |
317 | static int _req_conflicts(struct drbd_request *req) | |
318 | { | |
319 | struct drbd_conf *mdev = req->mdev; | |
320 | const sector_t sector = req->sector; | |
321 | const int size = req->size; | |
322 | struct drbd_request *i; | |
323 | struct drbd_epoch_entry *e; | |
324 | struct hlist_node *n; | |
325 | struct hlist_head *slot; | |
326 | ||
327 | D_ASSERT(hlist_unhashed(&req->colision)); | |
328 | ||
329 | if (!get_net_conf(mdev)) | |
330 | return 0; | |
331 | ||
332 | /* BUG_ON */ | |
333 | ERR_IF (mdev->tl_hash_s == 0) | |
334 | goto out_no_conflict; | |
335 | BUG_ON(mdev->tl_hash == NULL); | |
336 | ||
337 | #define OVERLAPS overlaps(i->sector, i->size, sector, size) | |
338 | slot = tl_hash_slot(mdev, sector); | |
339 | hlist_for_each_entry(i, n, slot, colision) { | |
340 | if (OVERLAPS) { | |
341 | dev_alert(DEV, "%s[%u] Concurrent local write detected! " | |
342 | "[DISCARD L] new: %llus +%u; " | |
343 | "pending: %llus +%u\n", | |
344 | current->comm, current->pid, | |
345 | (unsigned long long)sector, size, | |
346 | (unsigned long long)i->sector, i->size); | |
347 | goto out_conflict; | |
348 | } | |
349 | } | |
350 | ||
351 | if (mdev->ee_hash_s) { | |
352 | /* now, check for overlapping requests with remote origin */ | |
353 | BUG_ON(mdev->ee_hash == NULL); | |
354 | #undef OVERLAPS | |
355 | #define OVERLAPS overlaps(e->sector, e->size, sector, size) | |
356 | slot = ee_hash_slot(mdev, sector); | |
357 | hlist_for_each_entry(e, n, slot, colision) { | |
358 | if (OVERLAPS) { | |
359 | dev_alert(DEV, "%s[%u] Concurrent remote write detected!" | |
360 | " [DISCARD L] new: %llus +%u; " | |
361 | "pending: %llus +%u\n", | |
362 | current->comm, current->pid, | |
363 | (unsigned long long)sector, size, | |
364 | (unsigned long long)e->sector, e->size); | |
365 | goto out_conflict; | |
366 | } | |
367 | } | |
368 | } | |
369 | #undef OVERLAPS | |
370 | ||
371 | out_no_conflict: | |
372 | /* this is like it should be, and what we expected. | |
373 | * our users do behave after all... */ | |
374 | put_net_conf(mdev); | |
375 | return 0; | |
376 | ||
377 | out_conflict: | |
378 | put_net_conf(mdev); | |
379 | return 1; | |
380 | } | |
381 | ||
382 | /* obviously this could be coded as many single functions | |
383 | * instead of one huge switch, | |
384 | * or by putting the code directly in the respective locations | |
385 | * (as it has been before). | |
386 | * | |
387 | * but having it this way | |
388 | * enforces that it is all in this one place, where it is easier to audit, | |
389 | * it makes it obvious that whatever "event" "happens" to a request should | |
390 | * happen "atomically" within the req_lock, | |
391 | * and it enforces that we have to think in a very structured manner | |
392 | * about the "events" that may happen to a request during its life time ... | |
393 | */ | |
2a80699f | 394 | int __req_mod(struct drbd_request *req, enum drbd_req_event what, |
b411b363 PR |
395 | struct bio_and_error *m) |
396 | { | |
397 | struct drbd_conf *mdev = req->mdev; | |
2a80699f | 398 | int rv = 0; |
b411b363 PR |
399 | m->bio = NULL; |
400 | ||
b411b363 PR |
401 | switch (what) { |
402 | default: | |
403 | dev_err(DEV, "LOGIC BUG in %s:%u\n", __FILE__ , __LINE__); | |
404 | break; | |
405 | ||
406 | /* does not happen... | |
407 | * initialization done in drbd_req_new | |
408 | case created: | |
409 | break; | |
410 | */ | |
411 | ||
412 | case to_be_send: /* via network */ | |
413 | /* reached via drbd_make_request_common | |
414 | * and from w_read_retry_remote */ | |
415 | D_ASSERT(!(req->rq_state & RQ_NET_MASK)); | |
416 | req->rq_state |= RQ_NET_PENDING; | |
417 | inc_ap_pending(mdev); | |
418 | break; | |
419 | ||
420 | case to_be_submitted: /* locally */ | |
421 | /* reached via drbd_make_request_common */ | |
422 | D_ASSERT(!(req->rq_state & RQ_LOCAL_MASK)); | |
423 | req->rq_state |= RQ_LOCAL_PENDING; | |
424 | break; | |
425 | ||
426 | case completed_ok: | |
427 | if (bio_data_dir(req->master_bio) == WRITE) | |
428 | mdev->writ_cnt += req->size>>9; | |
429 | else | |
430 | mdev->read_cnt += req->size>>9; | |
431 | ||
432 | req->rq_state |= (RQ_LOCAL_COMPLETED|RQ_LOCAL_OK); | |
433 | req->rq_state &= ~RQ_LOCAL_PENDING; | |
434 | ||
cfa03415 | 435 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
436 | put_ldev(mdev); |
437 | break; | |
438 | ||
439 | case write_completed_with_error: | |
440 | req->rq_state |= RQ_LOCAL_COMPLETED; | |
441 | req->rq_state &= ~RQ_LOCAL_PENDING; | |
442 | ||
b411b363 | 443 | __drbd_chk_io_error(mdev, FALSE); |
cfa03415 | 444 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
445 | put_ldev(mdev); |
446 | break; | |
447 | ||
448 | case read_ahead_completed_with_error: | |
449 | /* it is legal to fail READA */ | |
450 | req->rq_state |= RQ_LOCAL_COMPLETED; | |
451 | req->rq_state &= ~RQ_LOCAL_PENDING; | |
cfa03415 | 452 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
453 | put_ldev(mdev); |
454 | break; | |
455 | ||
456 | case read_completed_with_error: | |
457 | drbd_set_out_of_sync(mdev, req->sector, req->size); | |
458 | ||
459 | req->rq_state |= RQ_LOCAL_COMPLETED; | |
460 | req->rq_state &= ~RQ_LOCAL_PENDING; | |
461 | ||
b411b363 | 462 | D_ASSERT(!(req->rq_state & RQ_NET_MASK)); |
b411b363 PR |
463 | |
464 | __drbd_chk_io_error(mdev, FALSE); | |
465 | put_ldev(mdev); | |
b411b363 | 466 | |
d255e5ff LE |
467 | /* no point in retrying if there is no good remote data, |
468 | * or we have no connection. */ | |
469 | if (mdev->state.pdsk != D_UP_TO_DATE) { | |
cfa03415 | 470 | _req_may_be_done_not_susp(req, m); |
d255e5ff LE |
471 | break; |
472 | } | |
473 | ||
474 | /* _req_mod(req,to_be_send); oops, recursion... */ | |
475 | req->rq_state |= RQ_NET_PENDING; | |
476 | inc_ap_pending(mdev); | |
b411b363 PR |
477 | /* fall through: _req_mod(req,queue_for_net_read); */ |
478 | ||
479 | case queue_for_net_read: | |
480 | /* READ or READA, and | |
481 | * no local disk, | |
482 | * or target area marked as invalid, | |
483 | * or just got an io-error. */ | |
484 | /* from drbd_make_request_common | |
485 | * or from bio_endio during read io-error recovery */ | |
486 | ||
487 | /* so we can verify the handle in the answer packet | |
488 | * corresponding hlist_del is in _req_may_be_done() */ | |
489 | hlist_add_head(&req->colision, ar_hash_slot(mdev, req->sector)); | |
490 | ||
83c38830 | 491 | set_bit(UNPLUG_REMOTE, &mdev->flags); |
b411b363 PR |
492 | |
493 | D_ASSERT(req->rq_state & RQ_NET_PENDING); | |
494 | req->rq_state |= RQ_NET_QUEUED; | |
495 | req->w.cb = (req->rq_state & RQ_LOCAL_MASK) | |
496 | ? w_read_retry_remote | |
497 | : w_send_read_req; | |
498 | drbd_queue_work(&mdev->data.work, &req->w); | |
499 | break; | |
500 | ||
501 | case queue_for_net_write: | |
502 | /* assert something? */ | |
503 | /* from drbd_make_request_common only */ | |
504 | ||
505 | hlist_add_head(&req->colision, tl_hash_slot(mdev, req->sector)); | |
506 | /* corresponding hlist_del is in _req_may_be_done() */ | |
507 | ||
508 | /* NOTE | |
509 | * In case the req ended up on the transfer log before being | |
510 | * queued on the worker, it could lead to this request being | |
511 | * missed during cleanup after connection loss. | |
512 | * So we have to do both operations here, | |
513 | * within the same lock that protects the transfer log. | |
514 | * | |
515 | * _req_add_to_epoch(req); this has to be after the | |
516 | * _maybe_start_new_epoch(req); which happened in | |
517 | * drbd_make_request_common, because we now may set the bit | |
518 | * again ourselves to close the current epoch. | |
519 | * | |
520 | * Add req to the (now) current epoch (barrier). */ | |
521 | ||
83c38830 LE |
522 | /* otherwise we may lose an unplug, which may cause some remote |
523 | * io-scheduler timeout to expire, increasing maximum latency, | |
524 | * hurting performance. */ | |
525 | set_bit(UNPLUG_REMOTE, &mdev->flags); | |
526 | ||
b411b363 PR |
527 | /* see drbd_make_request_common, |
528 | * just after it grabs the req_lock */ | |
529 | D_ASSERT(test_bit(CREATE_BARRIER, &mdev->flags) == 0); | |
530 | ||
531 | req->epoch = mdev->newest_tle->br_number; | |
b411b363 PR |
532 | |
533 | /* increment size of current epoch */ | |
7e602c0a | 534 | mdev->newest_tle->n_writes++; |
b411b363 PR |
535 | |
536 | /* queue work item to send data */ | |
537 | D_ASSERT(req->rq_state & RQ_NET_PENDING); | |
538 | req->rq_state |= RQ_NET_QUEUED; | |
539 | req->w.cb = w_send_dblock; | |
540 | drbd_queue_work(&mdev->data.work, &req->w); | |
541 | ||
542 | /* close the epoch, in case it outgrew the limit */ | |
7e602c0a | 543 | if (mdev->newest_tle->n_writes >= mdev->net_conf->max_epoch_size) |
b411b363 PR |
544 | queue_barrier(mdev); |
545 | ||
546 | break; | |
547 | ||
548 | case send_canceled: | |
549 | /* treat it the same */ | |
550 | case send_failed: | |
551 | /* real cleanup will be done from tl_clear. just update flags | |
552 | * so it is no longer marked as on the worker queue */ | |
553 | req->rq_state &= ~RQ_NET_QUEUED; | |
554 | /* if we did it right, tl_clear should be scheduled only after | |
555 | * this, so this should not be necessary! */ | |
cfa03415 | 556 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
557 | break; |
558 | ||
559 | case handed_over_to_network: | |
560 | /* assert something? */ | |
561 | if (bio_data_dir(req->master_bio) == WRITE && | |
562 | mdev->net_conf->wire_protocol == DRBD_PROT_A) { | |
563 | /* this is what is dangerous about protocol A: | |
564 | * pretend it was successfully written on the peer. */ | |
565 | if (req->rq_state & RQ_NET_PENDING) { | |
566 | dec_ap_pending(mdev); | |
567 | req->rq_state &= ~RQ_NET_PENDING; | |
568 | req->rq_state |= RQ_NET_OK; | |
569 | } /* else: neg-ack was faster... */ | |
570 | /* it is still not yet RQ_NET_DONE until the | |
571 | * corresponding epoch barrier got acked as well, | |
572 | * so we know what to dirty on connection loss */ | |
573 | } | |
574 | req->rq_state &= ~RQ_NET_QUEUED; | |
575 | req->rq_state |= RQ_NET_SENT; | |
576 | /* because _drbd_send_zc_bio could sleep, and may want to | |
577 | * dereference the bio even after the "write_acked_by_peer" and | |
578 | * "completed_ok" events came in, once we return from | |
579 | * _drbd_send_zc_bio (drbd_send_dblock), we have to check | |
580 | * whether it is done already, and end it. */ | |
cfa03415 | 581 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
582 | break; |
583 | ||
d255e5ff LE |
584 | case read_retry_remote_canceled: |
585 | req->rq_state &= ~RQ_NET_QUEUED; | |
586 | /* fall through, in case we raced with drbd_disconnect */ | |
b411b363 PR |
587 | case connection_lost_while_pending: |
588 | /* transfer log cleanup after connection loss */ | |
589 | /* assert something? */ | |
590 | if (req->rq_state & RQ_NET_PENDING) | |
591 | dec_ap_pending(mdev); | |
592 | req->rq_state &= ~(RQ_NET_OK|RQ_NET_PENDING); | |
593 | req->rq_state |= RQ_NET_DONE; | |
594 | /* if it is still queued, we may not complete it here. | |
595 | * it will be canceled soon. */ | |
596 | if (!(req->rq_state & RQ_NET_QUEUED)) | |
cfa03415 | 597 | _req_may_be_done(req, m); /* Allowed while state.susp */ |
b411b363 PR |
598 | break; |
599 | ||
600 | case write_acked_by_peer_and_sis: | |
601 | req->rq_state |= RQ_NET_SIS; | |
602 | case conflict_discarded_by_peer: | |
603 | /* for discarded conflicting writes of multiple primaries, | |
604 | * there is no need to keep anything in the tl, potential | |
605 | * node crashes are covered by the activity log. */ | |
606 | if (what == conflict_discarded_by_peer) | |
607 | dev_alert(DEV, "Got DiscardAck packet %llus +%u!" | |
608 | " DRBD is not a random data generator!\n", | |
609 | (unsigned long long)req->sector, req->size); | |
610 | req->rq_state |= RQ_NET_DONE; | |
611 | /* fall through */ | |
612 | case write_acked_by_peer: | |
613 | /* protocol C; successfully written on peer. | |
614 | * Nothing to do here. | |
615 | * We want to keep the tl in place for all protocols, to cater | |
616 | * for volatile write-back caches on lower level devices. | |
617 | * | |
618 | * A barrier request is expected to have forced all prior | |
619 | * requests onto stable storage, so completion of a barrier | |
620 | * request could set NET_DONE right here, and not wait for the | |
621 | * P_BARRIER_ACK, but that is an unnecessary optimization. */ | |
622 | ||
623 | /* this makes it effectively the same as for: */ | |
624 | case recv_acked_by_peer: | |
625 | /* protocol B; pretends to be successfully written on peer. | |
626 | * see also notes above in handed_over_to_network about | |
627 | * protocol != C */ | |
628 | req->rq_state |= RQ_NET_OK; | |
629 | D_ASSERT(req->rq_state & RQ_NET_PENDING); | |
630 | dec_ap_pending(mdev); | |
631 | req->rq_state &= ~RQ_NET_PENDING; | |
cfa03415 | 632 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
633 | break; |
634 | ||
635 | case neg_acked: | |
636 | /* assert something? */ | |
637 | if (req->rq_state & RQ_NET_PENDING) | |
638 | dec_ap_pending(mdev); | |
639 | req->rq_state &= ~(RQ_NET_OK|RQ_NET_PENDING); | |
640 | ||
641 | req->rq_state |= RQ_NET_DONE; | |
cfa03415 | 642 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
643 | /* else: done by handed_over_to_network */ |
644 | break; | |
645 | ||
265be2d0 PR |
646 | case fail_frozen_disk_io: |
647 | if (!(req->rq_state & RQ_LOCAL_COMPLETED)) | |
648 | break; | |
649 | ||
cfa03415 | 650 | _req_may_be_done(req, m); /* Allowed while state.susp */ |
265be2d0 PR |
651 | break; |
652 | ||
653 | case restart_frozen_disk_io: | |
654 | if (!(req->rq_state & RQ_LOCAL_COMPLETED)) | |
655 | break; | |
656 | ||
657 | req->rq_state &= ~RQ_LOCAL_COMPLETED; | |
658 | ||
659 | rv = MR_READ; | |
660 | if (bio_data_dir(req->master_bio) == WRITE) | |
661 | rv = MR_WRITE; | |
662 | ||
663 | get_ldev(mdev); | |
664 | req->w.cb = w_restart_disk_io; | |
665 | drbd_queue_work(&mdev->data.work, &req->w); | |
666 | break; | |
667 | ||
11b58e73 PR |
668 | case resend: |
669 | /* If RQ_NET_OK is already set, we got a P_WRITE_ACK or P_RECV_ACK | |
47ff2d0a | 670 | before the connection loss (B&C only); only P_BARRIER_ACK was missing. |
11b58e73 | 671 | Trowing them out of the TL here by pretending we got a BARRIER_ACK |
481c6f50 | 672 | We ensure that the peer was not rebooted */ |
11b58e73 PR |
673 | if (!(req->rq_state & RQ_NET_OK)) { |
674 | if (req->w.cb) { | |
675 | drbd_queue_work(&mdev->data.work, &req->w); | |
676 | rv = req->rq_state & RQ_WRITE ? MR_WRITE : MR_READ; | |
677 | } | |
678 | break; | |
679 | } | |
680 | /* else, fall through to barrier_acked */ | |
681 | ||
b411b363 | 682 | case barrier_acked: |
288f422e PR |
683 | if (!(req->rq_state & RQ_WRITE)) |
684 | break; | |
685 | ||
b411b363 PR |
686 | if (req->rq_state & RQ_NET_PENDING) { |
687 | /* barrier came in before all requests have been acked. | |
688 | * this is bad, because if the connection is lost now, | |
689 | * we won't be able to clean them up... */ | |
690 | dev_err(DEV, "FIXME (barrier_acked but pending)\n"); | |
b411b363 PR |
691 | list_move(&req->tl_requests, &mdev->out_of_sequence_requests); |
692 | } | |
693 | D_ASSERT(req->rq_state & RQ_NET_SENT); | |
694 | req->rq_state |= RQ_NET_DONE; | |
cfa03415 | 695 | _req_may_be_done(req, m); /* Allowed while state.susp */ |
b411b363 PR |
696 | break; |
697 | ||
698 | case data_received: | |
699 | D_ASSERT(req->rq_state & RQ_NET_PENDING); | |
700 | dec_ap_pending(mdev); | |
701 | req->rq_state &= ~RQ_NET_PENDING; | |
702 | req->rq_state |= (RQ_NET_OK|RQ_NET_DONE); | |
cfa03415 | 703 | _req_may_be_done_not_susp(req, m); |
b411b363 PR |
704 | break; |
705 | }; | |
2a80699f PR |
706 | |
707 | return rv; | |
b411b363 PR |
708 | } |
709 | ||
710 | /* we may do a local read if: | |
711 | * - we are consistent (of course), | |
712 | * - or we are generally inconsistent, | |
713 | * BUT we are still/already IN SYNC for this area. | |
714 | * since size may be bigger than BM_BLOCK_SIZE, | |
715 | * we may need to check several bits. | |
716 | */ | |
717 | static int drbd_may_do_local_read(struct drbd_conf *mdev, sector_t sector, int size) | |
718 | { | |
719 | unsigned long sbnr, ebnr; | |
720 | sector_t esector, nr_sectors; | |
721 | ||
722 | if (mdev->state.disk == D_UP_TO_DATE) | |
723 | return 1; | |
724 | if (mdev->state.disk >= D_OUTDATED) | |
725 | return 0; | |
726 | if (mdev->state.disk < D_INCONSISTENT) | |
727 | return 0; | |
728 | /* state.disk == D_INCONSISTENT We will have a look at the BitMap */ | |
729 | nr_sectors = drbd_get_capacity(mdev->this_bdev); | |
730 | esector = sector + (size >> 9) - 1; | |
731 | ||
732 | D_ASSERT(sector < nr_sectors); | |
733 | D_ASSERT(esector < nr_sectors); | |
734 | ||
735 | sbnr = BM_SECT_TO_BIT(sector); | |
736 | ebnr = BM_SECT_TO_BIT(esector); | |
737 | ||
738 | return 0 == drbd_bm_count_bits(mdev, sbnr, ebnr); | |
739 | } | |
740 | ||
741 | static int drbd_make_request_common(struct drbd_conf *mdev, struct bio *bio) | |
742 | { | |
743 | const int rw = bio_rw(bio); | |
744 | const int size = bio->bi_size; | |
745 | const sector_t sector = bio->bi_sector; | |
746 | struct drbd_tl_epoch *b = NULL; | |
747 | struct drbd_request *req; | |
748 | int local, remote; | |
749 | int err = -EIO; | |
9a25a04c | 750 | int ret = 0; |
b411b363 PR |
751 | |
752 | /* allocate outside of all locks; */ | |
753 | req = drbd_req_new(mdev, bio); | |
754 | if (!req) { | |
755 | dec_ap_bio(mdev); | |
756 | /* only pass the error to the upper layers. | |
757 | * if user cannot handle io errors, that's not our business. */ | |
758 | dev_err(DEV, "could not kmalloc() req\n"); | |
759 | bio_endio(bio, -ENOMEM); | |
760 | return 0; | |
761 | } | |
762 | ||
b411b363 PR |
763 | local = get_ldev(mdev); |
764 | if (!local) { | |
765 | bio_put(req->private_bio); /* or we get a bio leak */ | |
766 | req->private_bio = NULL; | |
767 | } | |
768 | if (rw == WRITE) { | |
769 | remote = 1; | |
770 | } else { | |
771 | /* READ || READA */ | |
772 | if (local) { | |
773 | if (!drbd_may_do_local_read(mdev, sector, size)) { | |
774 | /* we could kick the syncer to | |
775 | * sync this extent asap, wait for | |
776 | * it, then continue locally. | |
777 | * Or just issue the request remotely. | |
778 | */ | |
779 | local = 0; | |
780 | bio_put(req->private_bio); | |
781 | req->private_bio = NULL; | |
782 | put_ldev(mdev); | |
783 | } | |
784 | } | |
785 | remote = !local && mdev->state.pdsk >= D_UP_TO_DATE; | |
786 | } | |
787 | ||
788 | /* If we have a disk, but a READA request is mapped to remote, | |
789 | * we are R_PRIMARY, D_INCONSISTENT, SyncTarget. | |
790 | * Just fail that READA request right here. | |
791 | * | |
792 | * THINK: maybe fail all READA when not local? | |
793 | * or make this configurable... | |
794 | * if network is slow, READA won't do any good. | |
795 | */ | |
796 | if (rw == READA && mdev->state.disk >= D_INCONSISTENT && !local) { | |
797 | err = -EWOULDBLOCK; | |
798 | goto fail_and_free_req; | |
799 | } | |
800 | ||
801 | /* For WRITES going to the local disk, grab a reference on the target | |
802 | * extent. This waits for any resync activity in the corresponding | |
803 | * resync extent to finish, and, if necessary, pulls in the target | |
804 | * extent into the activity log, which involves further disk io because | |
805 | * of transactional on-disk meta data updates. */ | |
0778286a PR |
806 | if (rw == WRITE && local && !test_bit(AL_SUSPENDED, &mdev->flags)) { |
807 | req->rq_state |= RQ_IN_ACT_LOG; | |
b411b363 | 808 | drbd_al_begin_io(mdev, sector); |
0778286a | 809 | } |
b411b363 PR |
810 | |
811 | remote = remote && (mdev->state.pdsk == D_UP_TO_DATE || | |
812 | (mdev->state.pdsk == D_INCONSISTENT && | |
813 | mdev->state.conn >= C_CONNECTED)); | |
814 | ||
9a25a04c | 815 | if (!(local || remote) && !mdev->state.susp) { |
b411b363 PR |
816 | dev_err(DEV, "IO ERROR: neither local nor remote disk\n"); |
817 | goto fail_free_complete; | |
818 | } | |
819 | ||
820 | /* For WRITE request, we have to make sure that we have an | |
821 | * unused_spare_tle, in case we need to start a new epoch. | |
822 | * I try to be smart and avoid to pre-allocate always "just in case", | |
823 | * but there is a race between testing the bit and pointer outside the | |
824 | * spinlock, and grabbing the spinlock. | |
825 | * if we lost that race, we retry. */ | |
826 | if (rw == WRITE && remote && | |
827 | mdev->unused_spare_tle == NULL && | |
828 | test_bit(CREATE_BARRIER, &mdev->flags)) { | |
829 | allocate_barrier: | |
830 | b = kmalloc(sizeof(struct drbd_tl_epoch), GFP_NOIO); | |
831 | if (!b) { | |
832 | dev_err(DEV, "Failed to alloc barrier.\n"); | |
833 | err = -ENOMEM; | |
834 | goto fail_free_complete; | |
835 | } | |
836 | } | |
837 | ||
838 | /* GOOD, everything prepared, grab the spin_lock */ | |
839 | spin_lock_irq(&mdev->req_lock); | |
840 | ||
9a25a04c PR |
841 | if (mdev->state.susp) { |
842 | /* If we got suspended, use the retry mechanism of | |
843 | generic_make_request() to restart processing of this | |
844 | bio. In the next call to drbd_make_request_26 | |
845 | we sleep in inc_ap_bio() */ | |
846 | ret = 1; | |
847 | spin_unlock_irq(&mdev->req_lock); | |
848 | goto fail_free_complete; | |
849 | } | |
850 | ||
b411b363 PR |
851 | if (remote) { |
852 | remote = (mdev->state.pdsk == D_UP_TO_DATE || | |
853 | (mdev->state.pdsk == D_INCONSISTENT && | |
854 | mdev->state.conn >= C_CONNECTED)); | |
855 | if (!remote) | |
856 | dev_warn(DEV, "lost connection while grabbing the req_lock!\n"); | |
857 | if (!(local || remote)) { | |
858 | dev_err(DEV, "IO ERROR: neither local nor remote disk\n"); | |
859 | spin_unlock_irq(&mdev->req_lock); | |
860 | goto fail_free_complete; | |
861 | } | |
862 | } | |
863 | ||
864 | if (b && mdev->unused_spare_tle == NULL) { | |
865 | mdev->unused_spare_tle = b; | |
866 | b = NULL; | |
867 | } | |
868 | if (rw == WRITE && remote && | |
869 | mdev->unused_spare_tle == NULL && | |
870 | test_bit(CREATE_BARRIER, &mdev->flags)) { | |
871 | /* someone closed the current epoch | |
872 | * while we were grabbing the spinlock */ | |
873 | spin_unlock_irq(&mdev->req_lock); | |
874 | goto allocate_barrier; | |
875 | } | |
876 | ||
877 | ||
878 | /* Update disk stats */ | |
879 | _drbd_start_io_acct(mdev, req, bio); | |
880 | ||
881 | /* _maybe_start_new_epoch(mdev); | |
882 | * If we need to generate a write barrier packet, we have to add the | |
883 | * new epoch (barrier) object, and queue the barrier packet for sending, | |
884 | * and queue the req's data after it _within the same lock_, otherwise | |
885 | * we have race conditions were the reorder domains could be mixed up. | |
886 | * | |
887 | * Even read requests may start a new epoch and queue the corresponding | |
888 | * barrier packet. To get the write ordering right, we only have to | |
889 | * make sure that, if this is a write request and it triggered a | |
890 | * barrier packet, this request is queued within the same spinlock. */ | |
891 | if (remote && mdev->unused_spare_tle && | |
892 | test_and_clear_bit(CREATE_BARRIER, &mdev->flags)) { | |
893 | _tl_add_barrier(mdev, mdev->unused_spare_tle); | |
894 | mdev->unused_spare_tle = NULL; | |
895 | } else { | |
896 | D_ASSERT(!(remote && rw == WRITE && | |
897 | test_bit(CREATE_BARRIER, &mdev->flags))); | |
898 | } | |
899 | ||
900 | /* NOTE | |
901 | * Actually, 'local' may be wrong here already, since we may have failed | |
902 | * to write to the meta data, and may become wrong anytime because of | |
903 | * local io-error for some other request, which would lead to us | |
904 | * "detaching" the local disk. | |
905 | * | |
906 | * 'remote' may become wrong any time because the network could fail. | |
907 | * | |
908 | * This is a harmless race condition, though, since it is handled | |
909 | * correctly at the appropriate places; so it just defers the failure | |
910 | * of the respective operation. | |
911 | */ | |
912 | ||
913 | /* mark them early for readability. | |
914 | * this just sets some state flags. */ | |
915 | if (remote) | |
916 | _req_mod(req, to_be_send); | |
917 | if (local) | |
918 | _req_mod(req, to_be_submitted); | |
919 | ||
920 | /* check this request on the collision detection hash tables. | |
921 | * if we have a conflict, just complete it here. | |
922 | * THINK do we want to check reads, too? (I don't think so...) */ | |
d28fd092 LE |
923 | if (rw == WRITE && _req_conflicts(req)) |
924 | goto fail_conflicting; | |
288f422e PR |
925 | |
926 | list_add_tail(&req->tl_requests, &mdev->newest_tle->requests); | |
927 | ||
b411b363 PR |
928 | /* NOTE remote first: to get the concurrent write detection right, |
929 | * we must register the request before start of local IO. */ | |
930 | if (remote) { | |
931 | /* either WRITE and C_CONNECTED, | |
932 | * or READ, and no local disk, | |
933 | * or READ, but not in sync. | |
934 | */ | |
935 | _req_mod(req, (rw == WRITE) | |
936 | ? queue_for_net_write | |
937 | : queue_for_net_read); | |
938 | } | |
939 | spin_unlock_irq(&mdev->req_lock); | |
940 | kfree(b); /* if someone else has beaten us to it... */ | |
941 | ||
942 | if (local) { | |
943 | req->private_bio->bi_bdev = mdev->ldev->backing_bdev; | |
944 | ||
b411b363 PR |
945 | if (FAULT_ACTIVE(mdev, rw == WRITE ? DRBD_FAULT_DT_WR |
946 | : rw == READ ? DRBD_FAULT_DT_RD | |
947 | : DRBD_FAULT_DT_RA)) | |
948 | bio_endio(req->private_bio, -EIO); | |
949 | else | |
950 | generic_make_request(req->private_bio); | |
951 | } | |
952 | ||
953 | /* we need to plug ALWAYS since we possibly need to kick lo_dev. | |
954 | * we plug after submit, so we won't miss an unplug event */ | |
955 | drbd_plug_device(mdev); | |
956 | ||
957 | return 0; | |
958 | ||
d28fd092 LE |
959 | fail_conflicting: |
960 | /* this is a conflicting request. | |
961 | * even though it may have been only _partially_ | |
962 | * overlapping with one of the currently pending requests, | |
963 | * without even submitting or sending it, we will | |
964 | * pretend that it was successfully served right now. | |
965 | */ | |
966 | _drbd_end_io_acct(mdev, req); | |
967 | spin_unlock_irq(&mdev->req_lock); | |
968 | if (remote) | |
969 | dec_ap_pending(mdev); | |
970 | /* THINK: do we want to fail it (-EIO), or pretend success? | |
971 | * this pretends success. */ | |
972 | err = 0; | |
973 | ||
b411b363 PR |
974 | fail_free_complete: |
975 | if (rw == WRITE && local) | |
976 | drbd_al_complete_io(mdev, sector); | |
977 | fail_and_free_req: | |
978 | if (local) { | |
979 | bio_put(req->private_bio); | |
980 | req->private_bio = NULL; | |
981 | put_ldev(mdev); | |
982 | } | |
9a25a04c PR |
983 | if (!ret) |
984 | bio_endio(bio, err); | |
985 | ||
b411b363 PR |
986 | drbd_req_free(req); |
987 | dec_ap_bio(mdev); | |
988 | kfree(b); | |
989 | ||
9a25a04c | 990 | return ret; |
b411b363 PR |
991 | } |
992 | ||
993 | /* helper function for drbd_make_request | |
994 | * if we can determine just by the mdev (state) that this request will fail, | |
995 | * return 1 | |
996 | * otherwise return 0 | |
997 | */ | |
998 | static int drbd_fail_request_early(struct drbd_conf *mdev, int is_write) | |
999 | { | |
b411b363 PR |
1000 | if (mdev->state.role != R_PRIMARY && |
1001 | (!allow_oos || is_write)) { | |
1002 | if (__ratelimit(&drbd_ratelimit_state)) { | |
1003 | dev_err(DEV, "Process %s[%u] tried to %s; " | |
1004 | "since we are not in Primary state, " | |
1005 | "we cannot allow this\n", | |
1006 | current->comm, current->pid, | |
1007 | is_write ? "WRITE" : "READ"); | |
1008 | } | |
1009 | return 1; | |
1010 | } | |
1011 | ||
b411b363 PR |
1012 | return 0; |
1013 | } | |
1014 | ||
1015 | int drbd_make_request_26(struct request_queue *q, struct bio *bio) | |
1016 | { | |
1017 | unsigned int s_enr, e_enr; | |
1018 | struct drbd_conf *mdev = (struct drbd_conf *) q->queuedata; | |
1019 | ||
1020 | if (drbd_fail_request_early(mdev, bio_data_dir(bio) & WRITE)) { | |
1021 | bio_endio(bio, -EPERM); | |
1022 | return 0; | |
1023 | } | |
1024 | ||
1025 | /* Reject barrier requests if we know the underlying device does | |
1026 | * not support them. | |
1027 | * XXX: Need to get this info from peer as well some how so we | |
1028 | * XXX: reject if EITHER side/data/metadata area does not support them. | |
1029 | * | |
1030 | * because of those XXX, this is not yet enabled, | |
1031 | * i.e. in drbd_init_set_defaults we set the NO_BARRIER_SUPP bit. | |
1032 | */ | |
7b6d91da | 1033 | if (unlikely(bio->bi_rw & REQ_HARDBARRIER) && test_bit(NO_BARRIER_SUPP, &mdev->flags)) { |
b411b363 PR |
1034 | /* dev_warn(DEV, "Rejecting barrier request as underlying device does not support\n"); */ |
1035 | bio_endio(bio, -EOPNOTSUPP); | |
1036 | return 0; | |
1037 | } | |
1038 | ||
1039 | /* | |
1040 | * what we "blindly" assume: | |
1041 | */ | |
1042 | D_ASSERT(bio->bi_size > 0); | |
1043 | D_ASSERT((bio->bi_size & 0x1ff) == 0); | |
1044 | D_ASSERT(bio->bi_idx == 0); | |
1045 | ||
1046 | /* to make some things easier, force alignment of requests within the | |
1047 | * granularity of our hash tables */ | |
1048 | s_enr = bio->bi_sector >> HT_SHIFT; | |
1049 | e_enr = (bio->bi_sector+(bio->bi_size>>9)-1) >> HT_SHIFT; | |
1050 | ||
1051 | if (likely(s_enr == e_enr)) { | |
1052 | inc_ap_bio(mdev, 1); | |
1053 | return drbd_make_request_common(mdev, bio); | |
1054 | } | |
1055 | ||
1056 | /* can this bio be split generically? | |
1057 | * Maybe add our own split-arbitrary-bios function. */ | |
1058 | if (bio->bi_vcnt != 1 || bio->bi_idx != 0 || bio->bi_size > DRBD_MAX_SEGMENT_SIZE) { | |
1059 | /* rather error out here than BUG in bio_split */ | |
1060 | dev_err(DEV, "bio would need to, but cannot, be split: " | |
1061 | "(vcnt=%u,idx=%u,size=%u,sector=%llu)\n", | |
1062 | bio->bi_vcnt, bio->bi_idx, bio->bi_size, | |
1063 | (unsigned long long)bio->bi_sector); | |
1064 | bio_endio(bio, -EINVAL); | |
1065 | } else { | |
1066 | /* This bio crosses some boundary, so we have to split it. */ | |
1067 | struct bio_pair *bp; | |
1068 | /* works for the "do not cross hash slot boundaries" case | |
1069 | * e.g. sector 262269, size 4096 | |
1070 | * s_enr = 262269 >> 6 = 4097 | |
1071 | * e_enr = (262269+8-1) >> 6 = 4098 | |
1072 | * HT_SHIFT = 6 | |
1073 | * sps = 64, mask = 63 | |
1074 | * first_sectors = 64 - (262269 & 63) = 3 | |
1075 | */ | |
1076 | const sector_t sect = bio->bi_sector; | |
1077 | const int sps = 1 << HT_SHIFT; /* sectors per slot */ | |
1078 | const int mask = sps - 1; | |
1079 | const sector_t first_sectors = sps - (sect & mask); | |
1080 | bp = bio_split(bio, | |
1081 | #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28) | |
1082 | bio_split_pool, | |
1083 | #endif | |
1084 | first_sectors); | |
1085 | ||
1086 | /* we need to get a "reference count" (ap_bio_cnt) | |
1087 | * to avoid races with the disconnect/reconnect/suspend code. | |
9a25a04c | 1088 | * In case we need to split the bio here, we need to get three references |
b411b363 PR |
1089 | * atomically, otherwise we might deadlock when trying to submit the |
1090 | * second one! */ | |
9a25a04c | 1091 | inc_ap_bio(mdev, 3); |
b411b363 PR |
1092 | |
1093 | D_ASSERT(e_enr == s_enr + 1); | |
1094 | ||
9a25a04c PR |
1095 | while (drbd_make_request_common(mdev, &bp->bio1)) |
1096 | inc_ap_bio(mdev, 1); | |
1097 | ||
1098 | while (drbd_make_request_common(mdev, &bp->bio2)) | |
1099 | inc_ap_bio(mdev, 1); | |
1100 | ||
1101 | dec_ap_bio(mdev); | |
1102 | ||
b411b363 PR |
1103 | bio_pair_release(bp); |
1104 | } | |
1105 | return 0; | |
1106 | } | |
1107 | ||
1108 | /* This is called by bio_add_page(). With this function we reduce | |
1109 | * the number of BIOs that span over multiple DRBD_MAX_SEGMENT_SIZEs | |
1110 | * units (was AL_EXTENTs). | |
1111 | * | |
1112 | * we do the calculation within the lower 32bit of the byte offsets, | |
1113 | * since we don't care for actual offset, but only check whether it | |
1114 | * would cross "activity log extent" boundaries. | |
1115 | * | |
1116 | * As long as the BIO is empty we have to allow at least one bvec, | |
1117 | * regardless of size and offset. so the resulting bio may still | |
1118 | * cross extent boundaries. those are dealt with (bio_split) in | |
1119 | * drbd_make_request_26. | |
1120 | */ | |
1121 | int drbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, struct bio_vec *bvec) | |
1122 | { | |
1123 | struct drbd_conf *mdev = (struct drbd_conf *) q->queuedata; | |
1124 | unsigned int bio_offset = | |
1125 | (unsigned int)bvm->bi_sector << 9; /* 32 bit */ | |
1126 | unsigned int bio_size = bvm->bi_size; | |
1127 | int limit, backing_limit; | |
1128 | ||
1129 | limit = DRBD_MAX_SEGMENT_SIZE | |
1130 | - ((bio_offset & (DRBD_MAX_SEGMENT_SIZE-1)) + bio_size); | |
1131 | if (limit < 0) | |
1132 | limit = 0; | |
1133 | if (bio_size == 0) { | |
1134 | if (limit <= bvec->bv_len) | |
1135 | limit = bvec->bv_len; | |
1136 | } else if (limit && get_ldev(mdev)) { | |
1137 | struct request_queue * const b = | |
1138 | mdev->ldev->backing_bdev->bd_disk->queue; | |
a1c88d0d | 1139 | if (b->merge_bvec_fn) { |
b411b363 PR |
1140 | backing_limit = b->merge_bvec_fn(b, bvm, bvec); |
1141 | limit = min(limit, backing_limit); | |
1142 | } | |
1143 | put_ldev(mdev); | |
1144 | } | |
1145 | return limit; | |
1146 | } |