]>
Commit | Line | Data |
---|---|---|
c6ae4c04 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
b411b363 PR |
2 | /* |
3 | drbd_req.c | |
4 | ||
5 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
6 | ||
7 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
8 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
9 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
10 | ||
b411b363 PR |
11 | |
12 | */ | |
13 | ||
b411b363 PR |
14 | #include <linux/module.h> |
15 | ||
16 | #include <linux/slab.h> | |
17 | #include <linux/drbd.h> | |
18 | #include "drbd_int.h" | |
b411b363 PR |
19 | #include "drbd_req.h" |
20 | ||
21 | ||
b30ab791 | 22 | static bool drbd_may_do_local_read(struct drbd_device *device, sector_t sector, int size); |
57bcb6cf | 23 | |
9104d31a | 24 | static struct drbd_request *drbd_req_new(struct drbd_device *device, struct bio *bio_src) |
9e204cdd AG |
25 | { |
26 | struct drbd_request *req; | |
27 | ||
0892fac8 | 28 | req = mempool_alloc(&drbd_request_mempool, GFP_NOIO); |
9e204cdd AG |
29 | if (!req) |
30 | return NULL; | |
23fe8f8b | 31 | memset(req, 0, sizeof(*req)); |
9e204cdd | 32 | |
ae7153f1 CH |
33 | req->private_bio = bio_clone_fast(bio_src, GFP_NOIO, &drbd_io_bio_set); |
34 | req->private_bio->bi_private = req; | |
35 | req->private_bio->bi_end_io = drbd_request_endio; | |
36 | ||
9104d31a LE |
37 | req->rq_state = (bio_data_dir(bio_src) == WRITE ? RQ_WRITE : 0) |
38 | | (bio_op(bio_src) == REQ_OP_WRITE_SAME ? RQ_WSAME : 0) | |
f31e583a | 39 | | (bio_op(bio_src) == REQ_OP_WRITE_ZEROES ? RQ_ZEROES : 0) |
9104d31a LE |
40 | | (bio_op(bio_src) == REQ_OP_DISCARD ? RQ_UNMAP : 0); |
41 | req->device = device; | |
42 | req->master_bio = bio_src; | |
43 | req->epoch = 0; | |
53840641 | 44 | |
9e204cdd | 45 | drbd_clear_interval(&req->i); |
4f024f37 KO |
46 | req->i.sector = bio_src->bi_iter.bi_sector; |
47 | req->i.size = bio_src->bi_iter.bi_size; | |
5e472264 | 48 | req->i.local = true; |
53840641 AG |
49 | req->i.waiting = false; |
50 | ||
9e204cdd AG |
51 | INIT_LIST_HEAD(&req->tl_requests); |
52 | INIT_LIST_HEAD(&req->w.list); | |
844a6ae7 LE |
53 | INIT_LIST_HEAD(&req->req_pending_master_completion); |
54 | INIT_LIST_HEAD(&req->req_pending_local); | |
9e204cdd | 55 | |
a0d856df | 56 | /* one reference to be put by __drbd_make_request */ |
b406777e | 57 | atomic_set(&req->completion_ref, 1); |
a0d856df | 58 | /* one kref as long as completion_ref > 0 */ |
b406777e | 59 | kref_init(&req->kref); |
9e204cdd AG |
60 | return req; |
61 | } | |
62 | ||
08d0dabf LE |
63 | static void drbd_remove_request_interval(struct rb_root *root, |
64 | struct drbd_request *req) | |
65 | { | |
66 | struct drbd_device *device = req->device; | |
67 | struct drbd_interval *i = &req->i; | |
68 | ||
69 | drbd_remove_interval(root, i); | |
70 | ||
71 | /* Wake up any processes waiting for this request to complete. */ | |
72 | if (i->waiting) | |
73 | wake_up(&device->misc_wait); | |
74 | } | |
75 | ||
9a278a79 | 76 | void drbd_req_destroy(struct kref *kref) |
b411b363 | 77 | { |
b406777e | 78 | struct drbd_request *req = container_of(kref, struct drbd_request, kref); |
84b8c06b | 79 | struct drbd_device *device = req->device; |
a0d856df LE |
80 | const unsigned s = req->rq_state; |
81 | ||
82 | if ((req->master_bio && !(s & RQ_POSTPONED)) || | |
83 | atomic_read(&req->completion_ref) || | |
84 | (s & RQ_LOCAL_PENDING) || | |
85 | ((s & RQ_NET_MASK) && !(s & RQ_NET_DONE))) { | |
d0180171 | 86 | drbd_err(device, "drbd_req_destroy: Logic BUG rq_state = 0x%x, completion_ref = %d\n", |
a0d856df LE |
87 | s, atomic_read(&req->completion_ref)); |
88 | return; | |
89 | } | |
288f422e | 90 | |
844a6ae7 LE |
91 | /* If called from mod_rq_state (expected normal case) or |
92 | * drbd_send_and_submit (the less likely normal path), this holds the | |
93 | * req_lock, and req->tl_requests will typicaly be on ->transfer_log, | |
94 | * though it may be still empty (never added to the transfer log). | |
95 | * | |
96 | * If called from do_retry(), we do NOT hold the req_lock, but we are | |
97 | * still allowed to unconditionally list_del(&req->tl_requests), | |
98 | * because it will be on a local on-stack list only. */ | |
2312f0b3 | 99 | list_del_init(&req->tl_requests); |
288f422e | 100 | |
08d0dabf LE |
101 | /* finally remove the request from the conflict detection |
102 | * respective block_id verification interval tree. */ | |
103 | if (!drbd_interval_empty(&req->i)) { | |
104 | struct rb_root *root; | |
105 | ||
106 | if (s & RQ_WRITE) | |
107 | root = &device->write_requests; | |
108 | else | |
109 | root = &device->read_requests; | |
110 | drbd_remove_request_interval(root, req); | |
111 | } else if (s & (RQ_NET_MASK & ~RQ_NET_DONE) && req->i.size != 0) | |
112 | drbd_err(device, "drbd_req_destroy: Logic BUG: interval empty, but: rq_state=0x%x, sect=%llu, size=%u\n", | |
113 | s, (unsigned long long)req->i.sector, req->i.size); | |
114 | ||
b411b363 PR |
115 | /* if it was a write, we may have to set the corresponding |
116 | * bit(s) out-of-sync first. If it had a local part, we need to | |
117 | * release the reference to the activity log. */ | |
b406777e | 118 | if (s & RQ_WRITE) { |
b411b363 PR |
119 | /* Set out-of-sync unless both OK flags are set |
120 | * (local only or remote failed). | |
121 | * Other places where we set out-of-sync: | |
122 | * READ with local io-error */ | |
b411b363 | 123 | |
70f17b6b LE |
124 | /* There is a special case: |
125 | * we may notice late that IO was suspended, | |
126 | * and postpone, or schedule for retry, a write, | |
127 | * before it even was submitted or sent. | |
128 | * In that case we do not want to touch the bitmap at all. | |
129 | */ | |
130 | if ((s & (RQ_POSTPONED|RQ_LOCAL_MASK|RQ_NET_MASK)) != RQ_POSTPONED) { | |
d7644018 | 131 | if (!(s & RQ_NET_OK) || !(s & RQ_LOCAL_OK)) |
b30ab791 | 132 | drbd_set_out_of_sync(device, req->i.sector, req->i.size); |
b411b363 | 133 | |
d7644018 | 134 | if ((s & RQ_NET_OK) && (s & RQ_LOCAL_OK) && (s & RQ_NET_SIS)) |
b30ab791 | 135 | drbd_set_in_sync(device, req->i.sector, req->i.size); |
d7644018 | 136 | } |
b411b363 PR |
137 | |
138 | /* one might be tempted to move the drbd_al_complete_io | |
fcefa62e | 139 | * to the local io completion callback drbd_request_endio. |
b411b363 PR |
140 | * but, if this was a mirror write, we may only |
141 | * drbd_al_complete_io after this is RQ_NET_DONE, | |
142 | * otherwise the extent could be dropped from the al | |
143 | * before it has actually been written on the peer. | |
144 | * if we crash before our peer knows about the request, | |
145 | * but after the extent has been dropped from the al, | |
146 | * we would forget to resync the corresponding extent. | |
147 | */ | |
76590cd1 | 148 | if (s & RQ_IN_ACT_LOG) { |
b30ab791 AG |
149 | if (get_ldev_if_state(device, D_FAILED)) { |
150 | drbd_al_complete_io(device, &req->i); | |
151 | put_ldev(device); | |
b411b363 | 152 | } else if (__ratelimit(&drbd_ratelimit_state)) { |
d0180171 | 153 | drbd_warn(device, "Should have called drbd_al_complete_io(, %llu, %u), " |
181286ad LE |
154 | "but my Disk seems to have failed :(\n", |
155 | (unsigned long long) req->i.sector, req->i.size); | |
b411b363 PR |
156 | } |
157 | } | |
158 | } | |
159 | ||
0892fac8 | 160 | mempool_free(req, &drbd_request_mempool); |
b411b363 PR |
161 | } |
162 | ||
bde89a9e AG |
163 | static void wake_all_senders(struct drbd_connection *connection) |
164 | { | |
165 | wake_up(&connection->sender_work.q_wait); | |
b411b363 PR |
166 | } |
167 | ||
b6dd1a89 | 168 | /* must hold resource->req_lock */ |
bde89a9e | 169 | void start_new_tl_epoch(struct drbd_connection *connection) |
b411b363 | 170 | { |
99b4d8fe | 171 | /* no point closing an epoch, if it is empty, anyways. */ |
bde89a9e | 172 | if (connection->current_tle_writes == 0) |
99b4d8fe | 173 | return; |
b411b363 | 174 | |
bde89a9e AG |
175 | connection->current_tle_writes = 0; |
176 | atomic_inc(&connection->current_tle_nr); | |
177 | wake_all_senders(connection); | |
b411b363 PR |
178 | } |
179 | ||
b30ab791 | 180 | void complete_master_bio(struct drbd_device *device, |
b411b363 PR |
181 | struct bio_and_error *m) |
182 | { | |
cb53cf66 LE |
183 | if (unlikely(m->error)) |
184 | m->bio->bi_status = errno_to_blk_status(m->error); | |
4246a0b6 | 185 | bio_endio(m->bio); |
b30ab791 | 186 | dec_ap_bio(device); |
b411b363 PR |
187 | } |
188 | ||
53840641 | 189 | |
b411b363 PR |
190 | /* Helper for __req_mod(). |
191 | * Set m->bio to the master bio, if it is fit to be completed, | |
192 | * or leave it alone (it is initialized to NULL in __req_mod), | |
193 | * if it has already been completed, or cannot be completed yet. | |
194 | * If m->bio is set, the error status to be returned is placed in m->error. | |
195 | */ | |
6870ca6d | 196 | static |
a0d856df | 197 | void drbd_req_complete(struct drbd_request *req, struct bio_and_error *m) |
b411b363 | 198 | { |
a0d856df | 199 | const unsigned s = req->rq_state; |
84b8c06b | 200 | struct drbd_device *device = req->device; |
a0d856df | 201 | int error, ok; |
b411b363 | 202 | |
b411b363 PR |
203 | /* we must not complete the master bio, while it is |
204 | * still being processed by _drbd_send_zc_bio (drbd_send_dblock) | |
205 | * not yet acknowledged by the peer | |
206 | * not yet completed by the local io subsystem | |
207 | * these flags may get cleared in any order by | |
208 | * the worker, | |
209 | * the receiver, | |
210 | * the bio_endio completion callbacks. | |
211 | */ | |
a0d856df LE |
212 | if ((s & RQ_LOCAL_PENDING && !(s & RQ_LOCAL_ABORTED)) || |
213 | (s & RQ_NET_QUEUED) || (s & RQ_NET_PENDING) || | |
214 | (s & RQ_COMPLETION_SUSP)) { | |
d0180171 | 215 | drbd_err(device, "drbd_req_complete: Logic BUG rq_state = 0x%x\n", s); |
b411b363 | 216 | return; |
a0d856df LE |
217 | } |
218 | ||
219 | if (!req->master_bio) { | |
d0180171 | 220 | drbd_err(device, "drbd_req_complete: Logic BUG, master_bio == NULL!\n"); |
b411b363 | 221 | return; |
a0d856df | 222 | } |
b411b363 | 223 | |
a0d856df LE |
224 | /* |
225 | * figure out whether to report success or failure. | |
226 | * | |
227 | * report success when at least one of the operations succeeded. | |
228 | * or, to put the other way, | |
229 | * only report failure, when both operations failed. | |
230 | * | |
231 | * what to do about the failures is handled elsewhere. | |
232 | * what we need to do here is just: complete the master_bio. | |
233 | * | |
234 | * local completion error, if any, has been stored as ERR_PTR | |
235 | * in private_bio within drbd_request_endio. | |
236 | */ | |
237 | ok = (s & RQ_LOCAL_OK) || (s & RQ_NET_OK); | |
238 | error = PTR_ERR(req->private_bio); | |
b411b363 | 239 | |
a0d856df LE |
240 | /* Before we can signal completion to the upper layers, |
241 | * we may need to close the current transfer log epoch. | |
242 | * We are within the request lock, so we can simply compare | |
243 | * the request epoch number with the current transfer log | |
244 | * epoch number. If they match, increase the current_tle_nr, | |
245 | * and reset the transfer log epoch write_cnt. | |
246 | */ | |
70246286 | 247 | if (op_is_write(bio_op(req->master_bio)) && |
a6b32bc3 AG |
248 | req->epoch == atomic_read(&first_peer_device(device)->connection->current_tle_nr)) |
249 | start_new_tl_epoch(first_peer_device(device)->connection); | |
b411b363 | 250 | |
a0d856df | 251 | /* Update disk stats */ |
24d69293 | 252 | bio_end_io_acct(req->master_bio, req->start_jif); |
b411b363 | 253 | |
a0d856df LE |
254 | /* If READ failed, |
255 | * have it be pushed back to the retry work queue, | |
256 | * so it will re-enter __drbd_make_request(), | |
257 | * and be re-assigned to a suitable local or remote path, | |
258 | * or failed if we do not have access to good data anymore. | |
259 | * | |
260 | * Unless it was failed early by __drbd_make_request(), | |
261 | * because no path was available, in which case | |
262 | * it was not even added to the transfer_log. | |
263 | * | |
70246286 | 264 | * read-ahead may fail, and will not be retried. |
a0d856df LE |
265 | * |
266 | * WRITE should have used all available paths already. | |
267 | */ | |
70246286 CH |
268 | if (!ok && |
269 | bio_op(req->master_bio) == REQ_OP_READ && | |
1eff9d32 | 270 | !(req->master_bio->bi_opf & REQ_RAHEAD) && |
70246286 | 271 | !list_empty(&req->tl_requests)) |
a0d856df | 272 | req->rq_state |= RQ_POSTPONED; |
b411b363 | 273 | |
a0d856df | 274 | if (!(req->rq_state & RQ_POSTPONED)) { |
b411b363 PR |
275 | m->error = ok ? 0 : (error ?: -EIO); |
276 | m->bio = req->master_bio; | |
277 | req->master_bio = NULL; | |
08d0dabf LE |
278 | /* We leave it in the tree, to be able to verify later |
279 | * write-acks in protocol != C during resync. | |
280 | * But we mark it as "complete", so it won't be counted as | |
281 | * conflict in a multi-primary setup. */ | |
282 | req->i.completed = true; | |
b411b363 | 283 | } |
08d0dabf LE |
284 | |
285 | if (req->i.waiting) | |
286 | wake_up(&device->misc_wait); | |
844a6ae7 LE |
287 | |
288 | /* Either we are about to complete to upper layers, | |
289 | * or we will restart this request. | |
290 | * In either case, the request object will be destroyed soon, | |
291 | * so better remove it from all lists. */ | |
292 | list_del_init(&req->req_pending_master_completion); | |
b411b363 | 293 | } |
b411b363 | 294 | |
844a6ae7 | 295 | /* still holds resource->req_lock */ |
a00ebd1c | 296 | static void drbd_req_put_completion_ref(struct drbd_request *req, struct bio_and_error *m, int put) |
cfa03415 | 297 | { |
84b8c06b | 298 | struct drbd_device *device = req->device; |
0b0ba1ef | 299 | D_ASSERT(device, m || (req->rq_state & RQ_POSTPONED)); |
a0d856df | 300 | |
a00ebd1c LE |
301 | if (!put) |
302 | return; | |
303 | ||
a0d856df | 304 | if (!atomic_sub_and_test(put, &req->completion_ref)) |
a00ebd1c | 305 | return; |
2b4dd36f | 306 | |
a0d856df | 307 | drbd_req_complete(req, m); |
9a278a79 | 308 | |
a00ebd1c LE |
309 | /* local completion may still come in later, |
310 | * we need to keep the req object around. */ | |
311 | if (req->rq_state & RQ_LOCAL_ABORTED) | |
312 | return; | |
313 | ||
9a278a79 LE |
314 | if (req->rq_state & RQ_POSTPONED) { |
315 | /* don't destroy the req object just yet, | |
316 | * but queue it for retry */ | |
317 | drbd_restart_request(req); | |
a00ebd1c | 318 | return; |
b411b363 | 319 | } |
9a278a79 | 320 | |
a00ebd1c | 321 | kref_put(&req->kref, drbd_req_destroy); |
b411b363 PR |
322 | } |
323 | ||
7753a4c1 LE |
324 | static void set_if_null_req_next(struct drbd_peer_device *peer_device, struct drbd_request *req) |
325 | { | |
326 | struct drbd_connection *connection = peer_device ? peer_device->connection : NULL; | |
327 | if (!connection) | |
328 | return; | |
329 | if (connection->req_next == NULL) | |
330 | connection->req_next = req; | |
331 | } | |
332 | ||
333 | static void advance_conn_req_next(struct drbd_peer_device *peer_device, struct drbd_request *req) | |
334 | { | |
335 | struct drbd_connection *connection = peer_device ? peer_device->connection : NULL; | |
336 | if (!connection) | |
337 | return; | |
338 | if (connection->req_next != req) | |
339 | return; | |
340 | list_for_each_entry_continue(req, &connection->transfer_log, tl_requests) { | |
341 | const unsigned s = req->rq_state; | |
342 | if (s & RQ_NET_QUEUED) | |
343 | break; | |
344 | } | |
345 | if (&req->tl_requests == &connection->transfer_log) | |
346 | req = NULL; | |
347 | connection->req_next = req; | |
348 | } | |
349 | ||
350 | static void set_if_null_req_ack_pending(struct drbd_peer_device *peer_device, struct drbd_request *req) | |
351 | { | |
352 | struct drbd_connection *connection = peer_device ? peer_device->connection : NULL; | |
353 | if (!connection) | |
354 | return; | |
355 | if (connection->req_ack_pending == NULL) | |
356 | connection->req_ack_pending = req; | |
357 | } | |
358 | ||
359 | static void advance_conn_req_ack_pending(struct drbd_peer_device *peer_device, struct drbd_request *req) | |
360 | { | |
361 | struct drbd_connection *connection = peer_device ? peer_device->connection : NULL; | |
362 | if (!connection) | |
363 | return; | |
364 | if (connection->req_ack_pending != req) | |
365 | return; | |
366 | list_for_each_entry_continue(req, &connection->transfer_log, tl_requests) { | |
367 | const unsigned s = req->rq_state; | |
368 | if ((s & RQ_NET_SENT) && (s & RQ_NET_PENDING)) | |
369 | break; | |
370 | } | |
371 | if (&req->tl_requests == &connection->transfer_log) | |
372 | req = NULL; | |
373 | connection->req_ack_pending = req; | |
374 | } | |
375 | ||
376 | static void set_if_null_req_not_net_done(struct drbd_peer_device *peer_device, struct drbd_request *req) | |
377 | { | |
378 | struct drbd_connection *connection = peer_device ? peer_device->connection : NULL; | |
379 | if (!connection) | |
380 | return; | |
381 | if (connection->req_not_net_done == NULL) | |
382 | connection->req_not_net_done = req; | |
383 | } | |
384 | ||
385 | static void advance_conn_req_not_net_done(struct drbd_peer_device *peer_device, struct drbd_request *req) | |
386 | { | |
387 | struct drbd_connection *connection = peer_device ? peer_device->connection : NULL; | |
388 | if (!connection) | |
389 | return; | |
390 | if (connection->req_not_net_done != req) | |
391 | return; | |
392 | list_for_each_entry_continue(req, &connection->transfer_log, tl_requests) { | |
393 | const unsigned s = req->rq_state; | |
394 | if ((s & RQ_NET_SENT) && !(s & RQ_NET_DONE)) | |
395 | break; | |
396 | } | |
397 | if (&req->tl_requests == &connection->transfer_log) | |
398 | req = NULL; | |
399 | connection->req_not_net_done = req; | |
400 | } | |
401 | ||
a0d856df LE |
402 | /* I'd like this to be the only place that manipulates |
403 | * req->completion_ref and req->kref. */ | |
404 | static void mod_rq_state(struct drbd_request *req, struct bio_and_error *m, | |
405 | int clear, int set) | |
cfa03415 | 406 | { |
84b8c06b | 407 | struct drbd_device *device = req->device; |
7753a4c1 | 408 | struct drbd_peer_device *peer_device = first_peer_device(device); |
a0d856df LE |
409 | unsigned s = req->rq_state; |
410 | int c_put = 0; | |
cfa03415 | 411 | |
b30ab791 | 412 | if (drbd_suspended(device) && !((s | clear) & RQ_COMPLETION_SUSP)) |
5af2e8ce | 413 | set |= RQ_COMPLETION_SUSP; |
cfa03415 | 414 | |
a0d856df | 415 | /* apply */ |
b411b363 | 416 | |
a0d856df LE |
417 | req->rq_state &= ~clear; |
418 | req->rq_state |= set; | |
b411b363 | 419 | |
a0d856df LE |
420 | /* no change? */ |
421 | if (req->rq_state == s) | |
422 | return; | |
b411b363 | 423 | |
a0d856df LE |
424 | /* intent: get references */ |
425 | ||
bdfafc4f PZ |
426 | kref_get(&req->kref); |
427 | ||
a0d856df LE |
428 | if (!(s & RQ_LOCAL_PENDING) && (set & RQ_LOCAL_PENDING)) |
429 | atomic_inc(&req->completion_ref); | |
430 | ||
431 | if (!(s & RQ_NET_PENDING) && (set & RQ_NET_PENDING)) { | |
b30ab791 | 432 | inc_ap_pending(device); |
a0d856df | 433 | atomic_inc(&req->completion_ref); |
b411b363 PR |
434 | } |
435 | ||
e5f891b2 | 436 | if (!(s & RQ_NET_QUEUED) && (set & RQ_NET_QUEUED)) { |
a0d856df | 437 | atomic_inc(&req->completion_ref); |
7753a4c1 | 438 | set_if_null_req_next(peer_device, req); |
e5f891b2 | 439 | } |
a0d856df LE |
440 | |
441 | if (!(s & RQ_EXP_BARR_ACK) && (set & RQ_EXP_BARR_ACK)) | |
442 | kref_get(&req->kref); /* wait for the DONE */ | |
443 | ||
e5f891b2 | 444 | if (!(s & RQ_NET_SENT) && (set & RQ_NET_SENT)) { |
668700b4 | 445 | /* potentially already completed in the ack_receiver thread */ |
7753a4c1 | 446 | if (!(s & RQ_NET_DONE)) { |
e5f891b2 | 447 | atomic_add(req->i.size >> 9, &device->ap_in_flight); |
7753a4c1 LE |
448 | set_if_null_req_not_net_done(peer_device, req); |
449 | } | |
f85d9f2d | 450 | if (req->rq_state & RQ_NET_PENDING) |
7753a4c1 | 451 | set_if_null_req_ack_pending(peer_device, req); |
e5f891b2 | 452 | } |
a0d856df | 453 | |
5af2e8ce PR |
454 | if (!(s & RQ_COMPLETION_SUSP) && (set & RQ_COMPLETION_SUSP)) |
455 | atomic_inc(&req->completion_ref); | |
456 | ||
a0d856df LE |
457 | /* progress: put references */ |
458 | ||
459 | if ((s & RQ_COMPLETION_SUSP) && (clear & RQ_COMPLETION_SUSP)) | |
460 | ++c_put; | |
461 | ||
462 | if (!(s & RQ_LOCAL_ABORTED) && (set & RQ_LOCAL_ABORTED)) { | |
0b0ba1ef | 463 | D_ASSERT(device, req->rq_state & RQ_LOCAL_PENDING); |
a0d856df | 464 | ++c_put; |
b411b363 | 465 | } |
b411b363 | 466 | |
a0d856df LE |
467 | if ((s & RQ_LOCAL_PENDING) && (clear & RQ_LOCAL_PENDING)) { |
468 | if (req->rq_state & RQ_LOCAL_ABORTED) | |
bdfafc4f | 469 | kref_put(&req->kref, drbd_req_destroy); |
a0d856df LE |
470 | else |
471 | ++c_put; | |
844a6ae7 | 472 | list_del_init(&req->req_pending_local); |
a0d856df | 473 | } |
b411b363 | 474 | |
a0d856df | 475 | if ((s & RQ_NET_PENDING) && (clear & RQ_NET_PENDING)) { |
b30ab791 | 476 | dec_ap_pending(device); |
a0d856df | 477 | ++c_put; |
e5f891b2 | 478 | req->acked_jif = jiffies; |
7753a4c1 | 479 | advance_conn_req_ack_pending(peer_device, req); |
a0d856df LE |
480 | } |
481 | ||
7753a4c1 | 482 | if ((s & RQ_NET_QUEUED) && (clear & RQ_NET_QUEUED)) { |
a0d856df | 483 | ++c_put; |
7753a4c1 LE |
484 | advance_conn_req_next(peer_device, req); |
485 | } | |
a0d856df | 486 | |
e5f891b2 LE |
487 | if (!(s & RQ_NET_DONE) && (set & RQ_NET_DONE)) { |
488 | if (s & RQ_NET_SENT) | |
b30ab791 | 489 | atomic_sub(req->i.size >> 9, &device->ap_in_flight); |
e5f891b2 | 490 | if (s & RQ_EXP_BARR_ACK) |
bdfafc4f | 491 | kref_put(&req->kref, drbd_req_destroy); |
e5f891b2 | 492 | req->net_done_jif = jiffies; |
7753a4c1 LE |
493 | |
494 | /* in ahead/behind mode, or just in case, | |
495 | * before we finally destroy this request, | |
496 | * the caching pointers must not reference it anymore */ | |
497 | advance_conn_req_next(peer_device, req); | |
498 | advance_conn_req_ack_pending(peer_device, req); | |
499 | advance_conn_req_not_net_done(peer_device, req); | |
a0d856df LE |
500 | } |
501 | ||
502 | /* potentially complete and destroy */ | |
503 | ||
a0d856df LE |
504 | /* If we made progress, retry conflicting peer requests, if any. */ |
505 | if (req->i.waiting) | |
b30ab791 | 506 | wake_up(&device->misc_wait); |
a0d856df | 507 | |
a00ebd1c LE |
508 | drbd_req_put_completion_ref(req, m, c_put); |
509 | kref_put(&req->kref, drbd_req_destroy); | |
b411b363 PR |
510 | } |
511 | ||
b30ab791 | 512 | static void drbd_report_io_error(struct drbd_device *device, struct drbd_request *req) |
ccae7868 LE |
513 | { |
514 | char b[BDEVNAME_SIZE]; | |
515 | ||
42839f65 | 516 | if (!__ratelimit(&drbd_ratelimit_state)) |
ccae7868 LE |
517 | return; |
518 | ||
d0180171 | 519 | drbd_warn(device, "local %s IO error sector %llu+%u on %s\n", |
ccae7868 | 520 | (req->rq_state & RQ_WRITE) ? "WRITE" : "READ", |
42839f65 LE |
521 | (unsigned long long)req->i.sector, |
522 | req->i.size >> 9, | |
b30ab791 | 523 | bdevname(device->ldev->backing_bdev, b)); |
ccae7868 LE |
524 | } |
525 | ||
e5f891b2 LE |
526 | /* Helper for HANDED_OVER_TO_NETWORK. |
527 | * Is this a protocol A write (neither WRITE_ACK nor RECEIVE_ACK expected)? | |
528 | * Is it also still "PENDING"? | |
529 | * --> If so, clear PENDING and set NET_OK below. | |
530 | * If it is a protocol A write, but not RQ_PENDING anymore, neg-ack was faster | |
531 | * (and we must not set RQ_NET_OK) */ | |
532 | static inline bool is_pending_write_protocol_A(struct drbd_request *req) | |
533 | { | |
534 | return (req->rq_state & | |
535 | (RQ_WRITE|RQ_NET_PENDING|RQ_EXP_WRITE_ACK|RQ_EXP_RECEIVE_ACK)) | |
536 | == (RQ_WRITE|RQ_NET_PENDING); | |
537 | } | |
538 | ||
b411b363 PR |
539 | /* obviously this could be coded as many single functions |
540 | * instead of one huge switch, | |
541 | * or by putting the code directly in the respective locations | |
542 | * (as it has been before). | |
543 | * | |
544 | * but having it this way | |
545 | * enforces that it is all in this one place, where it is easier to audit, | |
546 | * it makes it obvious that whatever "event" "happens" to a request should | |
547 | * happen "atomically" within the req_lock, | |
548 | * and it enforces that we have to think in a very structured manner | |
549 | * about the "events" that may happen to a request during its life time ... | |
550 | */ | |
2a80699f | 551 | int __req_mod(struct drbd_request *req, enum drbd_req_event what, |
b411b363 PR |
552 | struct bio_and_error *m) |
553 | { | |
44a4d551 LE |
554 | struct drbd_device *const device = req->device; |
555 | struct drbd_peer_device *const peer_device = first_peer_device(device); | |
556 | struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL; | |
44ed167d | 557 | struct net_conf *nc; |
303d1448 | 558 | int p, rv = 0; |
7be8da07 AG |
559 | |
560 | if (m) | |
561 | m->bio = NULL; | |
b411b363 | 562 | |
b411b363 PR |
563 | switch (what) { |
564 | default: | |
d0180171 | 565 | drbd_err(device, "LOGIC BUG in %s:%u\n", __FILE__ , __LINE__); |
b411b363 PR |
566 | break; |
567 | ||
568 | /* does not happen... | |
569 | * initialization done in drbd_req_new | |
8554df1c | 570 | case CREATED: |
b411b363 PR |
571 | break; |
572 | */ | |
573 | ||
8554df1c | 574 | case TO_BE_SENT: /* via network */ |
7be8da07 | 575 | /* reached via __drbd_make_request |
b411b363 | 576 | * and from w_read_retry_remote */ |
0b0ba1ef | 577 | D_ASSERT(device, !(req->rq_state & RQ_NET_MASK)); |
44ed167d | 578 | rcu_read_lock(); |
44a4d551 | 579 | nc = rcu_dereference(connection->net_conf); |
44ed167d PR |
580 | p = nc->wire_protocol; |
581 | rcu_read_unlock(); | |
303d1448 PR |
582 | req->rq_state |= |
583 | p == DRBD_PROT_C ? RQ_EXP_WRITE_ACK : | |
584 | p == DRBD_PROT_B ? RQ_EXP_RECEIVE_ACK : 0; | |
a0d856df | 585 | mod_rq_state(req, m, 0, RQ_NET_PENDING); |
b411b363 PR |
586 | break; |
587 | ||
8554df1c | 588 | case TO_BE_SUBMITTED: /* locally */ |
7be8da07 | 589 | /* reached via __drbd_make_request */ |
0b0ba1ef | 590 | D_ASSERT(device, !(req->rq_state & RQ_LOCAL_MASK)); |
a0d856df | 591 | mod_rq_state(req, m, 0, RQ_LOCAL_PENDING); |
b411b363 PR |
592 | break; |
593 | ||
8554df1c | 594 | case COMPLETED_OK: |
2b4dd36f | 595 | if (req->rq_state & RQ_WRITE) |
b30ab791 | 596 | device->writ_cnt += req->i.size >> 9; |
b411b363 | 597 | else |
b30ab791 | 598 | device->read_cnt += req->i.size >> 9; |
b411b363 | 599 | |
a0d856df LE |
600 | mod_rq_state(req, m, RQ_LOCAL_PENDING, |
601 | RQ_LOCAL_COMPLETED|RQ_LOCAL_OK); | |
b411b363 PR |
602 | break; |
603 | ||
cdfda633 | 604 | case ABORT_DISK_IO: |
a0d856df | 605 | mod_rq_state(req, m, 0, RQ_LOCAL_ABORTED); |
2b4dd36f PR |
606 | break; |
607 | ||
edc9f5eb | 608 | case WRITE_COMPLETED_WITH_ERROR: |
b30ab791 AG |
609 | drbd_report_io_error(device, req); |
610 | __drbd_chk_io_error(device, DRBD_WRITE_ERROR); | |
edc9f5eb | 611 | mod_rq_state(req, m, RQ_LOCAL_PENDING, RQ_LOCAL_COMPLETED); |
b411b363 PR |
612 | break; |
613 | ||
8554df1c | 614 | case READ_COMPLETED_WITH_ERROR: |
b30ab791 AG |
615 | drbd_set_out_of_sync(device, req->i.sector, req->i.size); |
616 | drbd_report_io_error(device, req); | |
617 | __drbd_chk_io_error(device, DRBD_READ_ERROR); | |
df561f66 | 618 | fallthrough; |
a0d856df | 619 | case READ_AHEAD_COMPLETED_WITH_ERROR: |
70246286 | 620 | /* it is legal to fail read-ahead, no __drbd_chk_io_error in that case. */ |
a0d856df | 621 | mod_rq_state(req, m, RQ_LOCAL_PENDING, RQ_LOCAL_COMPLETED); |
2f632aeb LE |
622 | break; |
623 | ||
624 | case DISCARD_COMPLETED_NOTSUPP: | |
625 | case DISCARD_COMPLETED_WITH_ERROR: | |
626 | /* I'd rather not detach from local disk just because it | |
9305455a | 627 | * failed a REQ_OP_DISCARD. */ |
2f632aeb | 628 | mod_rq_state(req, m, RQ_LOCAL_PENDING, RQ_LOCAL_COMPLETED); |
4439c400 | 629 | break; |
b411b363 | 630 | |
8554df1c | 631 | case QUEUE_FOR_NET_READ: |
70246286 | 632 | /* READ, and |
b411b363 PR |
633 | * no local disk, |
634 | * or target area marked as invalid, | |
635 | * or just got an io-error. */ | |
7be8da07 | 636 | /* from __drbd_make_request |
b411b363 PR |
637 | * or from bio_endio during read io-error recovery */ |
638 | ||
6870ca6d LE |
639 | /* So we can verify the handle in the answer packet. |
640 | * Corresponding drbd_remove_request_interval is in | |
a0d856df | 641 | * drbd_req_complete() */ |
0b0ba1ef | 642 | D_ASSERT(device, drbd_interval_empty(&req->i)); |
b30ab791 | 643 | drbd_insert_interval(&device->read_requests, &req->i); |
b411b363 | 644 | |
b30ab791 | 645 | set_bit(UNPLUG_REMOTE, &device->flags); |
b411b363 | 646 | |
0b0ba1ef AG |
647 | D_ASSERT(device, req->rq_state & RQ_NET_PENDING); |
648 | D_ASSERT(device, (req->rq_state & RQ_LOCAL_MASK) == 0); | |
a0d856df | 649 | mod_rq_state(req, m, 0, RQ_NET_QUEUED); |
4439c400 | 650 | req->w.cb = w_send_read_req; |
44a4d551 | 651 | drbd_queue_work(&connection->sender_work, |
84b8c06b | 652 | &req->w); |
b411b363 PR |
653 | break; |
654 | ||
8554df1c | 655 | case QUEUE_FOR_NET_WRITE: |
b411b363 | 656 | /* assert something? */ |
7be8da07 | 657 | /* from __drbd_make_request only */ |
b411b363 | 658 | |
6870ca6d | 659 | /* Corresponding drbd_remove_request_interval is in |
a0d856df | 660 | * drbd_req_complete() */ |
0b0ba1ef | 661 | D_ASSERT(device, drbd_interval_empty(&req->i)); |
b30ab791 | 662 | drbd_insert_interval(&device->write_requests, &req->i); |
b411b363 PR |
663 | |
664 | /* NOTE | |
665 | * In case the req ended up on the transfer log before being | |
666 | * queued on the worker, it could lead to this request being | |
667 | * missed during cleanup after connection loss. | |
668 | * So we have to do both operations here, | |
669 | * within the same lock that protects the transfer log. | |
670 | * | |
671 | * _req_add_to_epoch(req); this has to be after the | |
672 | * _maybe_start_new_epoch(req); which happened in | |
7be8da07 | 673 | * __drbd_make_request, because we now may set the bit |
b411b363 PR |
674 | * again ourselves to close the current epoch. |
675 | * | |
676 | * Add req to the (now) current epoch (barrier). */ | |
677 | ||
83c38830 LE |
678 | /* otherwise we may lose an unplug, which may cause some remote |
679 | * io-scheduler timeout to expire, increasing maximum latency, | |
680 | * hurting performance. */ | |
b30ab791 | 681 | set_bit(UNPLUG_REMOTE, &device->flags); |
b411b363 PR |
682 | |
683 | /* queue work item to send data */ | |
0b0ba1ef | 684 | D_ASSERT(device, req->rq_state & RQ_NET_PENDING); |
a0d856df | 685 | mod_rq_state(req, m, 0, RQ_NET_QUEUED|RQ_EXP_BARR_ACK); |
b411b363 | 686 | req->w.cb = w_send_dblock; |
44a4d551 | 687 | drbd_queue_work(&connection->sender_work, |
84b8c06b | 688 | &req->w); |
b411b363 PR |
689 | |
690 | /* close the epoch, in case it outgrew the limit */ | |
44ed167d | 691 | rcu_read_lock(); |
44a4d551 | 692 | nc = rcu_dereference(connection->net_conf); |
44ed167d PR |
693 | p = nc->max_epoch_size; |
694 | rcu_read_unlock(); | |
44a4d551 LE |
695 | if (connection->current_tle_writes >= p) |
696 | start_new_tl_epoch(connection); | |
b411b363 PR |
697 | |
698 | break; | |
699 | ||
8554df1c | 700 | case QUEUE_FOR_SEND_OOS: |
a0d856df | 701 | mod_rq_state(req, m, 0, RQ_NET_QUEUED); |
8f7bed77 | 702 | req->w.cb = w_send_out_of_sync; |
44a4d551 | 703 | drbd_queue_work(&connection->sender_work, |
84b8c06b | 704 | &req->w); |
73a01a18 PR |
705 | break; |
706 | ||
ea9d6729 | 707 | case READ_RETRY_REMOTE_CANCELED: |
8554df1c | 708 | case SEND_CANCELED: |
8554df1c | 709 | case SEND_FAILED: |
b411b363 PR |
710 | /* real cleanup will be done from tl_clear. just update flags |
711 | * so it is no longer marked as on the worker queue */ | |
a0d856df | 712 | mod_rq_state(req, m, RQ_NET_QUEUED, 0); |
b411b363 PR |
713 | break; |
714 | ||
8554df1c | 715 | case HANDED_OVER_TO_NETWORK: |
b411b363 | 716 | /* assert something? */ |
e5f891b2 | 717 | if (is_pending_write_protocol_A(req)) |
b411b363 PR |
718 | /* this is what is dangerous about protocol A: |
719 | * pretend it was successfully written on the peer. */ | |
e5f891b2 LE |
720 | mod_rq_state(req, m, RQ_NET_QUEUED|RQ_NET_PENDING, |
721 | RQ_NET_SENT|RQ_NET_OK); | |
722 | else | |
723 | mod_rq_state(req, m, RQ_NET_QUEUED, RQ_NET_SENT); | |
724 | /* It is still not yet RQ_NET_DONE until the | |
725 | * corresponding epoch barrier got acked as well, | |
726 | * so we know what to dirty on connection loss. */ | |
6d49e101 LE |
727 | break; |
728 | ||
27a434fe | 729 | case OOS_HANDED_TO_NETWORK: |
6d49e101 LE |
730 | /* Was not set PENDING, no longer QUEUED, so is now DONE |
731 | * as far as this connection is concerned. */ | |
a0d856df | 732 | mod_rq_state(req, m, RQ_NET_QUEUED, RQ_NET_DONE); |
b411b363 PR |
733 | break; |
734 | ||
8554df1c | 735 | case CONNECTION_LOST_WHILE_PENDING: |
b411b363 | 736 | /* transfer log cleanup after connection loss */ |
a0d856df LE |
737 | mod_rq_state(req, m, |
738 | RQ_NET_OK|RQ_NET_PENDING|RQ_COMPLETION_SUSP, | |
739 | RQ_NET_DONE); | |
b411b363 PR |
740 | break; |
741 | ||
d4dabbe2 LE |
742 | case CONFLICT_RESOLVED: |
743 | /* for superseded conflicting writes of multiple primaries, | |
b411b363 | 744 | * there is no need to keep anything in the tl, potential |
934722a2 LE |
745 | * node crashes are covered by the activity log. |
746 | * | |
747 | * If this request had been marked as RQ_POSTPONED before, | |
d4dabbe2 | 748 | * it will actually not be completed, but "restarted", |
934722a2 | 749 | * resubmitted from the retry worker context. */ |
0b0ba1ef AG |
750 | D_ASSERT(device, req->rq_state & RQ_NET_PENDING); |
751 | D_ASSERT(device, req->rq_state & RQ_EXP_WRITE_ACK); | |
934722a2 LE |
752 | mod_rq_state(req, m, RQ_NET_PENDING, RQ_NET_DONE|RQ_NET_OK); |
753 | break; | |
754 | ||
0afd569a | 755 | case WRITE_ACKED_BY_PEER_AND_SIS: |
934722a2 | 756 | req->rq_state |= RQ_NET_SIS; |
6327c911 | 757 | fallthrough; |
8554df1c | 758 | case WRITE_ACKED_BY_PEER: |
08d0dabf LE |
759 | /* Normal operation protocol C: successfully written on peer. |
760 | * During resync, even in protocol != C, | |
761 | * we requested an explicit write ack anyways. | |
762 | * Which means we cannot even assert anything here. | |
d64957c9 | 763 | * Nothing more to do here. |
b411b363 | 764 | * We want to keep the tl in place for all protocols, to cater |
d64957c9 | 765 | * for volatile write-back caches on lower level devices. */ |
303d1448 | 766 | goto ack_common; |
8554df1c | 767 | case RECV_ACKED_BY_PEER: |
0b0ba1ef | 768 | D_ASSERT(device, req->rq_state & RQ_EXP_RECEIVE_ACK); |
b411b363 | 769 | /* protocol B; pretends to be successfully written on peer. |
8554df1c | 770 | * see also notes above in HANDED_OVER_TO_NETWORK about |
b411b363 | 771 | * protocol != C */ |
303d1448 | 772 | ack_common: |
a0d856df | 773 | mod_rq_state(req, m, RQ_NET_PENDING, RQ_NET_OK); |
b411b363 PR |
774 | break; |
775 | ||
7be8da07 | 776 | case POSTPONE_WRITE: |
0b0ba1ef | 777 | D_ASSERT(device, req->rq_state & RQ_EXP_WRITE_ACK); |
303d1448 | 778 | /* If this node has already detected the write conflict, the |
7be8da07 AG |
779 | * worker will be waiting on misc_wait. Wake it up once this |
780 | * request has completed locally. | |
781 | */ | |
0b0ba1ef | 782 | D_ASSERT(device, req->rq_state & RQ_NET_PENDING); |
7be8da07 | 783 | req->rq_state |= RQ_POSTPONED; |
a0d856df | 784 | if (req->i.waiting) |
b30ab791 | 785 | wake_up(&device->misc_wait); |
a0d856df LE |
786 | /* Do not clear RQ_NET_PENDING. This request will make further |
787 | * progress via restart_conflicting_writes() or | |
788 | * fail_postponed_requests(). Hopefully. */ | |
7be8da07 | 789 | break; |
b411b363 | 790 | |
8554df1c | 791 | case NEG_ACKED: |
46e21bba | 792 | mod_rq_state(req, m, RQ_NET_OK|RQ_NET_PENDING, 0); |
b411b363 PR |
793 | break; |
794 | ||
8554df1c | 795 | case FAIL_FROZEN_DISK_IO: |
265be2d0 PR |
796 | if (!(req->rq_state & RQ_LOCAL_COMPLETED)) |
797 | break; | |
a0d856df | 798 | mod_rq_state(req, m, RQ_COMPLETION_SUSP, 0); |
265be2d0 PR |
799 | break; |
800 | ||
8554df1c | 801 | case RESTART_FROZEN_DISK_IO: |
265be2d0 PR |
802 | if (!(req->rq_state & RQ_LOCAL_COMPLETED)) |
803 | break; | |
804 | ||
a0d856df LE |
805 | mod_rq_state(req, m, |
806 | RQ_COMPLETION_SUSP|RQ_LOCAL_COMPLETED, | |
807 | RQ_LOCAL_PENDING); | |
265be2d0 PR |
808 | |
809 | rv = MR_READ; | |
810 | if (bio_data_dir(req->master_bio) == WRITE) | |
811 | rv = MR_WRITE; | |
812 | ||
b30ab791 | 813 | get_ldev(device); /* always succeeds in this call path */ |
265be2d0 | 814 | req->w.cb = w_restart_disk_io; |
44a4d551 | 815 | drbd_queue_work(&connection->sender_work, |
84b8c06b | 816 | &req->w); |
265be2d0 PR |
817 | break; |
818 | ||
8554df1c | 819 | case RESEND: |
509fc019 PR |
820 | /* Simply complete (local only) READs. */ |
821 | if (!(req->rq_state & RQ_WRITE) && !req->w.cb) { | |
8a0bab2a | 822 | mod_rq_state(req, m, RQ_COMPLETION_SUSP, 0); |
509fc019 PR |
823 | break; |
824 | } | |
825 | ||
11b58e73 | 826 | /* If RQ_NET_OK is already set, we got a P_WRITE_ACK or P_RECV_ACK |
a0d856df LE |
827 | before the connection loss (B&C only); only P_BARRIER_ACK |
828 | (or the local completion?) was missing when we suspended. | |
6870ca6d LE |
829 | Throwing them out of the TL here by pretending we got a BARRIER_ACK. |
830 | During connection handshake, we ensure that the peer was not rebooted. */ | |
11b58e73 | 831 | if (!(req->rq_state & RQ_NET_OK)) { |
84b8c06b | 832 | /* FIXME could this possibly be a req->dw.cb == w_send_out_of_sync? |
a0d856df LE |
833 | * in that case we must not set RQ_NET_PENDING. */ |
834 | ||
835 | mod_rq_state(req, m, RQ_COMPLETION_SUSP, RQ_NET_QUEUED|RQ_NET_PENDING); | |
11b58e73 | 836 | if (req->w.cb) { |
44a4d551 LE |
837 | /* w.cb expected to be w_send_dblock, or w_send_read_req */ |
838 | drbd_queue_work(&connection->sender_work, | |
84b8c06b | 839 | &req->w); |
11b58e73 | 840 | rv = req->rq_state & RQ_WRITE ? MR_WRITE : MR_READ; |
a0d856df | 841 | } /* else: FIXME can this happen? */ |
11b58e73 PR |
842 | break; |
843 | } | |
df561f66 | 844 | fallthrough; /* to BARRIER_ACKED */ |
11b58e73 | 845 | |
8554df1c | 846 | case BARRIER_ACKED: |
a0d856df | 847 | /* barrier ack for READ requests does not make sense */ |
288f422e PR |
848 | if (!(req->rq_state & RQ_WRITE)) |
849 | break; | |
850 | ||
b411b363 | 851 | if (req->rq_state & RQ_NET_PENDING) { |
a209b4ae | 852 | /* barrier came in before all requests were acked. |
b411b363 PR |
853 | * this is bad, because if the connection is lost now, |
854 | * we won't be able to clean them up... */ | |
d0180171 | 855 | drbd_err(device, "FIXME (BARRIER_ACKED but pending)\n"); |
b411b363 | 856 | } |
a0d856df LE |
857 | /* Allowed to complete requests, even while suspended. |
858 | * As this is called for all requests within a matching epoch, | |
859 | * we need to filter, and only set RQ_NET_DONE for those that | |
860 | * have actually been on the wire. */ | |
861 | mod_rq_state(req, m, RQ_COMPLETION_SUSP, | |
862 | (req->rq_state & RQ_NET_MASK) ? RQ_NET_DONE : 0); | |
b411b363 PR |
863 | break; |
864 | ||
8554df1c | 865 | case DATA_RECEIVED: |
0b0ba1ef | 866 | D_ASSERT(device, req->rq_state & RQ_NET_PENDING); |
a0d856df | 867 | mod_rq_state(req, m, RQ_NET_PENDING, RQ_NET_OK|RQ_NET_DONE); |
b411b363 | 868 | break; |
7074e4a7 LE |
869 | |
870 | case QUEUE_AS_DRBD_BARRIER: | |
44a4d551 | 871 | start_new_tl_epoch(connection); |
7074e4a7 LE |
872 | mod_rq_state(req, m, 0, RQ_NET_OK|RQ_NET_DONE); |
873 | break; | |
ec45a263 | 874 | } |
2a80699f PR |
875 | |
876 | return rv; | |
b411b363 PR |
877 | } |
878 | ||
879 | /* we may do a local read if: | |
880 | * - we are consistent (of course), | |
881 | * - or we are generally inconsistent, | |
882 | * BUT we are still/already IN SYNC for this area. | |
883 | * since size may be bigger than BM_BLOCK_SIZE, | |
884 | * we may need to check several bits. | |
885 | */ | |
b30ab791 | 886 | static bool drbd_may_do_local_read(struct drbd_device *device, sector_t sector, int size) |
b411b363 PR |
887 | { |
888 | unsigned long sbnr, ebnr; | |
889 | sector_t esector, nr_sectors; | |
890 | ||
b30ab791 | 891 | if (device->state.disk == D_UP_TO_DATE) |
0da34df0 | 892 | return true; |
b30ab791 | 893 | if (device->state.disk != D_INCONSISTENT) |
0da34df0 | 894 | return false; |
b411b363 | 895 | esector = sector + (size >> 9) - 1; |
155bd9d1 | 896 | nr_sectors = get_capacity(device->vdisk); |
0b0ba1ef AG |
897 | D_ASSERT(device, sector < nr_sectors); |
898 | D_ASSERT(device, esector < nr_sectors); | |
b411b363 PR |
899 | |
900 | sbnr = BM_SECT_TO_BIT(sector); | |
901 | ebnr = BM_SECT_TO_BIT(esector); | |
902 | ||
b30ab791 | 903 | return drbd_bm_count_bits(device, sbnr, ebnr) == 0; |
b411b363 PR |
904 | } |
905 | ||
b30ab791 | 906 | static bool remote_due_to_read_balancing(struct drbd_device *device, sector_t sector, |
5da9c836 | 907 | enum drbd_read_balancing rbm) |
380207d0 | 908 | { |
d60de03a | 909 | int stripe_shift; |
380207d0 | 910 | |
380207d0 PR |
911 | switch (rbm) { |
912 | case RB_CONGESTED_REMOTE: | |
edb0872f CH |
913 | return bdi_read_congested( |
914 | device->ldev->backing_bdev->bd_disk->bdi); | |
380207d0 | 915 | case RB_LEAST_PENDING: |
b30ab791 AG |
916 | return atomic_read(&device->local_cnt) > |
917 | atomic_read(&device->ap_pending_cnt) + atomic_read(&device->rs_pending_cnt); | |
d60de03a PR |
918 | case RB_32K_STRIPING: /* stripe_shift = 15 */ |
919 | case RB_64K_STRIPING: | |
920 | case RB_128K_STRIPING: | |
921 | case RB_256K_STRIPING: | |
922 | case RB_512K_STRIPING: | |
923 | case RB_1M_STRIPING: /* stripe_shift = 20 */ | |
924 | stripe_shift = (rbm - RB_32K_STRIPING + 15); | |
925 | return (sector >> (stripe_shift - 9)) & 1; | |
380207d0 | 926 | case RB_ROUND_ROBIN: |
b30ab791 | 927 | return test_and_change_bit(READ_BALANCE_RR, &device->flags); |
380207d0 PR |
928 | case RB_PREFER_REMOTE: |
929 | return true; | |
930 | case RB_PREFER_LOCAL: | |
931 | default: | |
932 | return false; | |
933 | } | |
934 | } | |
935 | ||
6024fece AG |
936 | /* |
937 | * complete_conflicting_writes - wait for any conflicting write requests | |
938 | * | |
939 | * The write_requests tree contains all active write requests which we | |
940 | * currently know about. Wait for any requests to complete which conflict with | |
941 | * the new one. | |
648e46b5 LE |
942 | * |
943 | * Only way out: remove the conflicting intervals from the tree. | |
6024fece | 944 | */ |
648e46b5 | 945 | static void complete_conflicting_writes(struct drbd_request *req) |
6024fece | 946 | { |
648e46b5 | 947 | DEFINE_WAIT(wait); |
84b8c06b | 948 | struct drbd_device *device = req->device; |
648e46b5 LE |
949 | struct drbd_interval *i; |
950 | sector_t sector = req->i.sector; | |
951 | int size = req->i.size; | |
952 | ||
648e46b5 | 953 | for (;;) { |
1b228c98 LE |
954 | drbd_for_each_overlap(i, &device->write_requests, sector, size) { |
955 | /* Ignore, if already completed to upper layers. */ | |
956 | if (i->completed) | |
957 | continue; | |
958 | /* Handle the first found overlap. After the schedule | |
959 | * we have to restart the tree walk. */ | |
648e46b5 | 960 | break; |
1b228c98 LE |
961 | } |
962 | if (!i) /* if any */ | |
963 | break; | |
964 | ||
648e46b5 | 965 | /* Indicate to wake up device->misc_wait on progress. */ |
1b228c98 | 966 | prepare_to_wait(&device->misc_wait, &wait, TASK_UNINTERRUPTIBLE); |
648e46b5 | 967 | i->waiting = true; |
0500813f | 968 | spin_unlock_irq(&device->resource->req_lock); |
648e46b5 | 969 | schedule(); |
0500813f | 970 | spin_lock_irq(&device->resource->req_lock); |
6024fece | 971 | } |
b30ab791 | 972 | finish_wait(&device->misc_wait, &wait); |
b411b363 PR |
973 | } |
974 | ||
7e5fec31 | 975 | /* called within req_lock */ |
b30ab791 | 976 | static void maybe_pull_ahead(struct drbd_device *device) |
0d5934e3 | 977 | { |
a6b32bc3 | 978 | struct drbd_connection *connection = first_peer_device(device)->connection; |
5da9c836 LE |
979 | struct net_conf *nc; |
980 | bool congested = false; | |
981 | enum drbd_on_congestion on_congestion; | |
982 | ||
607f25e5 | 983 | rcu_read_lock(); |
bde89a9e | 984 | nc = rcu_dereference(connection->net_conf); |
5da9c836 | 985 | on_congestion = nc ? nc->on_congestion : OC_BLOCK; |
607f25e5 | 986 | rcu_read_unlock(); |
5da9c836 | 987 | if (on_congestion == OC_BLOCK || |
bde89a9e | 988 | connection->agreed_pro_version < 96) |
3b9ef85e | 989 | return; |
0d5934e3 | 990 | |
0c066bc3 LE |
991 | if (on_congestion == OC_PULL_AHEAD && device->state.conn == C_AHEAD) |
992 | return; /* nothing to do ... */ | |
993 | ||
0d5934e3 LE |
994 | /* If I don't even have good local storage, we can not reasonably try |
995 | * to pull ahead of the peer. We also need the local reference to make | |
b30ab791 | 996 | * sure device->act_log is there. |
0d5934e3 | 997 | */ |
b30ab791 | 998 | if (!get_ldev_if_state(device, D_UP_TO_DATE)) |
0d5934e3 LE |
999 | return; |
1000 | ||
5da9c836 | 1001 | if (nc->cong_fill && |
b30ab791 | 1002 | atomic_read(&device->ap_in_flight) >= nc->cong_fill) { |
d0180171 | 1003 | drbd_info(device, "Congestion-fill threshold reached\n"); |
5da9c836 | 1004 | congested = true; |
0d5934e3 LE |
1005 | } |
1006 | ||
b30ab791 | 1007 | if (device->act_log->used >= nc->cong_extents) { |
d0180171 | 1008 | drbd_info(device, "Congestion-extents threshold reached\n"); |
5da9c836 | 1009 | congested = true; |
0d5934e3 LE |
1010 | } |
1011 | ||
1012 | if (congested) { | |
99b4d8fe | 1013 | /* start a new epoch for non-mirrored writes */ |
a6b32bc3 | 1014 | start_new_tl_epoch(first_peer_device(device)->connection); |
0d5934e3 | 1015 | |
5da9c836 | 1016 | if (on_congestion == OC_PULL_AHEAD) |
b30ab791 | 1017 | _drbd_set_state(_NS(device, conn, C_AHEAD), 0, NULL); |
5da9c836 | 1018 | else /*nc->on_congestion == OC_DISCONNECT */ |
b30ab791 | 1019 | _drbd_set_state(_NS(device, conn, C_DISCONNECTING), 0, NULL); |
0d5934e3 | 1020 | } |
b30ab791 | 1021 | put_ldev(device); |
0d5934e3 LE |
1022 | } |
1023 | ||
5da9c836 LE |
1024 | /* If this returns false, and req->private_bio is still set, |
1025 | * this should be submitted locally. | |
1026 | * | |
1027 | * If it returns false, but req->private_bio is not set, | |
1028 | * we do not have access to good data :( | |
1029 | * | |
1030 | * Otherwise, this destroys req->private_bio, if any, | |
1031 | * and returns true. | |
1032 | */ | |
1033 | static bool do_remote_read(struct drbd_request *req) | |
1034 | { | |
84b8c06b | 1035 | struct drbd_device *device = req->device; |
5da9c836 LE |
1036 | enum drbd_read_balancing rbm; |
1037 | ||
1038 | if (req->private_bio) { | |
b30ab791 | 1039 | if (!drbd_may_do_local_read(device, |
5da9c836 LE |
1040 | req->i.sector, req->i.size)) { |
1041 | bio_put(req->private_bio); | |
1042 | req->private_bio = NULL; | |
b30ab791 | 1043 | put_ldev(device); |
5da9c836 LE |
1044 | } |
1045 | } | |
1046 | ||
b30ab791 | 1047 | if (device->state.pdsk != D_UP_TO_DATE) |
5da9c836 LE |
1048 | return false; |
1049 | ||
a0d856df LE |
1050 | if (req->private_bio == NULL) |
1051 | return true; | |
1052 | ||
5da9c836 LE |
1053 | /* TODO: improve read balancing decisions, take into account drbd |
1054 | * protocol, pending requests etc. */ | |
1055 | ||
1056 | rcu_read_lock(); | |
b30ab791 | 1057 | rbm = rcu_dereference(device->ldev->disk_conf)->read_balancing; |
5da9c836 LE |
1058 | rcu_read_unlock(); |
1059 | ||
1060 | if (rbm == RB_PREFER_LOCAL && req->private_bio) | |
1061 | return false; /* submit locally */ | |
1062 | ||
b30ab791 | 1063 | if (remote_due_to_read_balancing(device, req->i.sector, rbm)) { |
5da9c836 LE |
1064 | if (req->private_bio) { |
1065 | bio_put(req->private_bio); | |
1066 | req->private_bio = NULL; | |
b30ab791 | 1067 | put_ldev(device); |
5da9c836 LE |
1068 | } |
1069 | return true; | |
1070 | } | |
1071 | ||
1072 | return false; | |
1073 | } | |
1074 | ||
2e9ffde6 AG |
1075 | bool drbd_should_do_remote(union drbd_dev_state s) |
1076 | { | |
1077 | return s.pdsk == D_UP_TO_DATE || | |
1078 | (s.pdsk >= D_INCONSISTENT && | |
1079 | s.conn >= C_WF_BITMAP_T && | |
1080 | s.conn < C_AHEAD); | |
1081 | /* Before proto 96 that was >= CONNECTED instead of >= C_WF_BITMAP_T. | |
1082 | That is equivalent since before 96 IO was frozen in the C_WF_BITMAP* | |
1083 | states. */ | |
1084 | } | |
1085 | ||
1086 | static bool drbd_should_send_out_of_sync(union drbd_dev_state s) | |
1087 | { | |
1088 | return s.conn == C_AHEAD || s.conn == C_WF_BITMAP_S; | |
1089 | /* pdsk = D_INCONSISTENT as a consequence. Protocol 96 check not necessary | |
1090 | since we enter state C_AHEAD only if proto >= 96 */ | |
1091 | } | |
1092 | ||
5da9c836 LE |
1093 | /* returns number of connections (== 1, for drbd 8.4) |
1094 | * expected to actually write this data, | |
1095 | * which does NOT include those that we are L_AHEAD for. */ | |
1096 | static int drbd_process_write_request(struct drbd_request *req) | |
1097 | { | |
84b8c06b | 1098 | struct drbd_device *device = req->device; |
5da9c836 LE |
1099 | int remote, send_oos; |
1100 | ||
b30ab791 AG |
1101 | remote = drbd_should_do_remote(device->state); |
1102 | send_oos = drbd_should_send_out_of_sync(device->state); | |
5da9c836 | 1103 | |
519b6d3e LE |
1104 | /* Need to replicate writes. Unless it is an empty flush, |
1105 | * which is better mapped to a DRBD P_BARRIER packet, | |
1106 | * also for drbd wire protocol compatibility reasons. | |
1107 | * If this was a flush, just start a new epoch. | |
1108 | * Unless the current epoch was empty anyways, or we are not currently | |
1109 | * replicating, in which case there is no point. */ | |
1110 | if (unlikely(req->i.size == 0)) { | |
1111 | /* The only size==0 bios we expect are empty flushes. */ | |
1eff9d32 | 1112 | D_ASSERT(device, req->master_bio->bi_opf & REQ_PREFLUSH); |
99b4d8fe | 1113 | if (remote) |
7074e4a7 LE |
1114 | _req_mod(req, QUEUE_AS_DRBD_BARRIER); |
1115 | return remote; | |
519b6d3e LE |
1116 | } |
1117 | ||
5da9c836 LE |
1118 | if (!remote && !send_oos) |
1119 | return 0; | |
1120 | ||
0b0ba1ef | 1121 | D_ASSERT(device, !(remote && send_oos)); |
5da9c836 LE |
1122 | |
1123 | if (remote) { | |
1124 | _req_mod(req, TO_BE_SENT); | |
1125 | _req_mod(req, QUEUE_FOR_NET_WRITE); | |
b30ab791 | 1126 | } else if (drbd_set_out_of_sync(device, req->i.sector, req->i.size)) |
5da9c836 LE |
1127 | _req_mod(req, QUEUE_FOR_SEND_OOS); |
1128 | ||
1129 | return remote; | |
1130 | } | |
1131 | ||
f31e583a | 1132 | static void drbd_process_discard_or_zeroes_req(struct drbd_request *req, int flags) |
7435e901 | 1133 | { |
f31e583a LE |
1134 | int err = drbd_issue_discard_or_zero_out(req->device, |
1135 | req->i.sector, req->i.size >> 9, flags); | |
1136 | if (err) | |
4e4cbee9 | 1137 | req->private_bio->bi_status = BLK_STS_IOERR; |
7435e901 LE |
1138 | bio_endio(req->private_bio); |
1139 | } | |
1140 | ||
5da9c836 LE |
1141 | static void |
1142 | drbd_submit_req_private_bio(struct drbd_request *req) | |
1143 | { | |
84b8c06b | 1144 | struct drbd_device *device = req->device; |
5da9c836 | 1145 | struct bio *bio = req->private_bio; |
70246286 CH |
1146 | unsigned int type; |
1147 | ||
1148 | if (bio_op(bio) != REQ_OP_READ) | |
1149 | type = DRBD_FAULT_DT_WR; | |
1eff9d32 | 1150 | else if (bio->bi_opf & REQ_RAHEAD) |
70246286 CH |
1151 | type = DRBD_FAULT_DT_RA; |
1152 | else | |
1153 | type = DRBD_FAULT_DT_RD; | |
5da9c836 | 1154 | |
74d46992 | 1155 | bio_set_dev(bio, device->ldev->backing_bdev); |
5da9c836 LE |
1156 | |
1157 | /* State may have changed since we grabbed our reference on the | |
1158 | * ->ldev member. Double check, and short-circuit to endio. | |
1159 | * In case the last activity log transaction failed to get on | |
1160 | * stable storage, and this is a WRITE, we may not even submit | |
1161 | * this bio. */ | |
b30ab791 | 1162 | if (get_ldev(device)) { |
70246286 | 1163 | if (drbd_insert_fault(device, type)) |
4246a0b6 | 1164 | bio_io_error(bio); |
f31e583a LE |
1165 | else if (bio_op(bio) == REQ_OP_WRITE_ZEROES) |
1166 | drbd_process_discard_or_zeroes_req(req, EE_ZEROOUT | | |
1167 | ((bio->bi_opf & REQ_NOUNMAP) ? 0 : EE_TRIM)); | |
1168 | else if (bio_op(bio) == REQ_OP_DISCARD) | |
1169 | drbd_process_discard_or_zeroes_req(req, EE_TRIM); | |
5da9c836 | 1170 | else |
ed00aabd | 1171 | submit_bio_noacct(bio); |
b30ab791 | 1172 | put_ldev(device); |
5da9c836 | 1173 | } else |
4246a0b6 | 1174 | bio_io_error(bio); |
5da9c836 LE |
1175 | } |
1176 | ||
b30ab791 | 1177 | static void drbd_queue_write(struct drbd_device *device, struct drbd_request *req) |
779b3fe4 | 1178 | { |
844a6ae7 | 1179 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 1180 | list_add_tail(&req->tl_requests, &device->submit.writes); |
844a6ae7 LE |
1181 | list_add_tail(&req->req_pending_master_completion, |
1182 | &device->pending_master_completion[1 /* WRITE */]); | |
1183 | spin_unlock_irq(&device->resource->req_lock); | |
b30ab791 | 1184 | queue_work(device->submit.wq, &device->submit.worker); |
f5b90b6b LE |
1185 | /* do_submit() may sleep internally on al_wait, too */ |
1186 | wake_up(&device->al_wait); | |
779b3fe4 LE |
1187 | } |
1188 | ||
6d9febe2 LE |
1189 | /* returns the new drbd_request pointer, if the caller is expected to |
1190 | * drbd_send_and_submit() it (to save latency), or NULL if we queued the | |
1191 | * request on the submitter thread. | |
1192 | * Returns ERR_PTR(-ENOMEM) if we cannot allocate a drbd_request. | |
1193 | */ | |
01cd2636 | 1194 | static struct drbd_request * |
370276ba | 1195 | drbd_request_prepare(struct drbd_device *device, struct bio *bio) |
b411b363 | 1196 | { |
6d9febe2 | 1197 | const int rw = bio_data_dir(bio); |
b411b363 | 1198 | struct drbd_request *req; |
b411b363 PR |
1199 | |
1200 | /* allocate outside of all locks; */ | |
b30ab791 | 1201 | req = drbd_req_new(device, bio); |
b411b363 | 1202 | if (!req) { |
b30ab791 | 1203 | dec_ap_bio(device); |
b411b363 PR |
1204 | /* only pass the error to the upper layers. |
1205 | * if user cannot handle io errors, that's not our business. */ | |
d0180171 | 1206 | drbd_err(device, "could not kmalloc() req\n"); |
4e4cbee9 | 1207 | bio->bi_status = BLK_STS_RESOURCE; |
4246a0b6 | 1208 | bio_endio(bio); |
6d9febe2 | 1209 | return ERR_PTR(-ENOMEM); |
b411b363 | 1210 | } |
24d69293 CH |
1211 | |
1212 | /* Update disk stats */ | |
1213 | req->start_jif = bio_start_io_acct(req->master_bio); | |
b411b363 | 1214 | |
b30ab791 | 1215 | if (!get_ldev(device)) { |
5da9c836 | 1216 | bio_put(req->private_bio); |
b411b363 PR |
1217 | req->private_bio = NULL; |
1218 | } | |
b411b363 | 1219 | |
7435e901 | 1220 | /* process discards always from our submitter thread */ |
fad2d4ef BVA |
1221 | if (bio_op(bio) == REQ_OP_WRITE_ZEROES || |
1222 | bio_op(bio) == REQ_OP_DISCARD) | |
7435e901 LE |
1223 | goto queue_for_submitter_thread; |
1224 | ||
519b6d3e | 1225 | if (rw == WRITE && req->private_bio && req->i.size |
b30ab791 | 1226 | && !test_bit(AL_SUSPENDED, &device->flags)) { |
7435e901 LE |
1227 | if (!drbd_al_begin_io_fastpath(device, &req->i)) |
1228 | goto queue_for_submitter_thread; | |
0778286a | 1229 | req->rq_state |= RQ_IN_ACT_LOG; |
e5f891b2 | 1230 | req->in_actlog_jif = jiffies; |
0778286a | 1231 | } |
6d9febe2 | 1232 | return req; |
7435e901 LE |
1233 | |
1234 | queue_for_submitter_thread: | |
1235 | atomic_inc(&device->ap_actlog_cnt); | |
1236 | drbd_queue_write(device, req); | |
1237 | return NULL; | |
6d9febe2 LE |
1238 | } |
1239 | ||
0ead5cca LE |
1240 | /* Require at least one path to current data. |
1241 | * We don't want to allow writes on C_STANDALONE D_INCONSISTENT: | |
1242 | * We would not allow to read what was written, | |
1243 | * we would not have bumped the data generation uuids, | |
1244 | * we would cause data divergence for all the wrong reasons. | |
1245 | * | |
1246 | * If we don't see at least one D_UP_TO_DATE, we will fail this request, | |
1247 | * which either returns EIO, or, if OND_SUSPEND_IO is set, suspends IO, | |
1248 | * and queues for retry later. | |
1249 | */ | |
1250 | static bool may_do_writes(struct drbd_device *device) | |
1251 | { | |
1252 | const union drbd_dev_state s = device->state; | |
1253 | return s.disk == D_UP_TO_DATE || s.pdsk == D_UP_TO_DATE; | |
1254 | } | |
1255 | ||
c51a0ef3 LE |
1256 | struct drbd_plug_cb { |
1257 | struct blk_plug_cb cb; | |
1258 | struct drbd_request *most_recent_req; | |
1259 | /* do we need more? */ | |
1260 | }; | |
1261 | ||
1262 | static void drbd_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1263 | { | |
1264 | struct drbd_plug_cb *plug = container_of(cb, struct drbd_plug_cb, cb); | |
1265 | struct drbd_resource *resource = plug->cb.data; | |
1266 | struct drbd_request *req = plug->most_recent_req; | |
1267 | ||
de6978be | 1268 | kfree(cb); |
c51a0ef3 LE |
1269 | if (!req) |
1270 | return; | |
1271 | ||
1272 | spin_lock_irq(&resource->req_lock); | |
1273 | /* In case the sender did not process it yet, raise the flag to | |
1274 | * have it followed with P_UNPLUG_REMOTE just after. */ | |
1275 | req->rq_state |= RQ_UNPLUG; | |
1276 | /* but also queue a generic unplug */ | |
1277 | drbd_queue_unplug(req->device); | |
c51a0ef3 | 1278 | kref_put(&req->kref, drbd_req_destroy); |
de6978be | 1279 | spin_unlock_irq(&resource->req_lock); |
c51a0ef3 LE |
1280 | } |
1281 | ||
1282 | static struct drbd_plug_cb* drbd_check_plugged(struct drbd_resource *resource) | |
1283 | { | |
1284 | /* A lot of text to say | |
1285 | * return (struct drbd_plug_cb*)blk_check_plugged(); */ | |
1286 | struct drbd_plug_cb *plug; | |
1287 | struct blk_plug_cb *cb = blk_check_plugged(drbd_unplug, resource, sizeof(*plug)); | |
1288 | ||
1289 | if (cb) | |
1290 | plug = container_of(cb, struct drbd_plug_cb, cb); | |
1291 | else | |
1292 | plug = NULL; | |
1293 | return plug; | |
1294 | } | |
1295 | ||
1296 | static void drbd_update_plug(struct drbd_plug_cb *plug, struct drbd_request *req) | |
1297 | { | |
1298 | struct drbd_request *tmp = plug->most_recent_req; | |
1299 | /* Will be sent to some peer. | |
1300 | * Remember to tag it with UNPLUG_REMOTE on unplug */ | |
1301 | kref_get(&req->kref); | |
1302 | plug->most_recent_req = req; | |
1303 | if (tmp) | |
1304 | kref_put(&tmp->kref, drbd_req_destroy); | |
1305 | } | |
1306 | ||
b30ab791 | 1307 | static void drbd_send_and_submit(struct drbd_device *device, struct drbd_request *req) |
6d9febe2 | 1308 | { |
35b5ed5b | 1309 | struct drbd_resource *resource = device->resource; |
70246286 | 1310 | const int rw = bio_data_dir(req->master_bio); |
6d9febe2 LE |
1311 | struct bio_and_error m = { NULL, }; |
1312 | bool no_remote = false; | |
35b5ed5b | 1313 | bool submit_private_bio = false; |
6d9febe2 | 1314 | |
35b5ed5b | 1315 | spin_lock_irq(&resource->req_lock); |
6024fece | 1316 | if (rw == WRITE) { |
648e46b5 LE |
1317 | /* This may temporarily give up the req_lock, |
1318 | * but will re-aquire it before it returns here. | |
1319 | * Needs to be before the check on drbd_suspended() */ | |
1320 | complete_conflicting_writes(req); | |
607f25e5 LE |
1321 | /* no more giving up req_lock from now on! */ |
1322 | ||
1323 | /* check for congestion, and potentially stop sending | |
1324 | * full data updates, but start sending "dirty bits" only. */ | |
b30ab791 | 1325 | maybe_pull_ahead(device); |
b411b363 PR |
1326 | } |
1327 | ||
9a25a04c | 1328 | |
b30ab791 | 1329 | if (drbd_suspended(device)) { |
5da9c836 LE |
1330 | /* push back and retry: */ |
1331 | req->rq_state |= RQ_POSTPONED; | |
1332 | if (req->private_bio) { | |
1333 | bio_put(req->private_bio); | |
1334 | req->private_bio = NULL; | |
b30ab791 | 1335 | put_ldev(device); |
b411b363 | 1336 | } |
5da9c836 | 1337 | goto out; |
b411b363 PR |
1338 | } |
1339 | ||
70246286 | 1340 | /* We fail READ early, if we can not serve it. |
5da9c836 | 1341 | * We must do this before req is registered on any lists. |
a0d856df | 1342 | * Otherwise, drbd_req_complete() will queue failed READ for retry. */ |
5da9c836 LE |
1343 | if (rw != WRITE) { |
1344 | if (!do_remote_read(req) && !req->private_bio) | |
1345 | goto nodata; | |
b411b363 PR |
1346 | } |
1347 | ||
b6dd1a89 | 1348 | /* which transfer log epoch does this belong to? */ |
a6b32bc3 | 1349 | req->epoch = atomic_read(&first_peer_device(device)->connection->current_tle_nr); |
288f422e | 1350 | |
227f052f LE |
1351 | /* no point in adding empty flushes to the transfer log, |
1352 | * they are mapped to drbd barriers already. */ | |
99b4d8fe LE |
1353 | if (likely(req->i.size!=0)) { |
1354 | if (rw == WRITE) | |
a6b32bc3 | 1355 | first_peer_device(device)->connection->current_tle_writes++; |
288f422e | 1356 | |
a6b32bc3 | 1357 | list_add_tail(&req->tl_requests, &first_peer_device(device)->connection->transfer_log); |
b411b363 | 1358 | } |
67531718 | 1359 | |
5da9c836 | 1360 | if (rw == WRITE) { |
0ead5cca LE |
1361 | if (req->private_bio && !may_do_writes(device)) { |
1362 | bio_put(req->private_bio); | |
1363 | req->private_bio = NULL; | |
1364 | put_ldev(device); | |
1365 | goto nodata; | |
1366 | } | |
5da9c836 LE |
1367 | if (!drbd_process_write_request(req)) |
1368 | no_remote = true; | |
1369 | } else { | |
1370 | /* We either have a private_bio, or we can read from remote. | |
1371 | * Otherwise we had done the goto nodata above. */ | |
1372 | if (req->private_bio == NULL) { | |
1373 | _req_mod(req, TO_BE_SENT); | |
1374 | _req_mod(req, QUEUE_FOR_NET_READ); | |
6719fb03 | 1375 | } else |
5da9c836 | 1376 | no_remote = true; |
b411b363 PR |
1377 | } |
1378 | ||
de6978be LE |
1379 | if (no_remote == false) { |
1380 | struct drbd_plug_cb *plug = drbd_check_plugged(resource); | |
1381 | if (plug) | |
1382 | drbd_update_plug(plug, req); | |
1383 | } | |
c51a0ef3 | 1384 | |
844a6ae7 LE |
1385 | /* If it took the fast path in drbd_request_prepare, add it here. |
1386 | * The slow path has added it already. */ | |
1387 | if (list_empty(&req->req_pending_master_completion)) | |
1388 | list_add_tail(&req->req_pending_master_completion, | |
1389 | &device->pending_master_completion[rw == WRITE]); | |
5da9c836 LE |
1390 | if (req->private_bio) { |
1391 | /* needs to be marked within the same spinlock */ | |
05cbbb39 | 1392 | req->pre_submit_jif = jiffies; |
844a6ae7 LE |
1393 | list_add_tail(&req->req_pending_local, |
1394 | &device->pending_completion[rw == WRITE]); | |
5da9c836 LE |
1395 | _req_mod(req, TO_BE_SUBMITTED); |
1396 | /* but we need to give up the spinlock to submit */ | |
35b5ed5b | 1397 | submit_private_bio = true; |
5da9c836 LE |
1398 | } else if (no_remote) { |
1399 | nodata: | |
1400 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 1401 | drbd_err(device, "IO ERROR: neither local nor remote data, sector %llu+%u\n", |
42839f65 | 1402 | (unsigned long long)req->i.sector, req->i.size >> 9); |
5da9c836 | 1403 | /* A write may have been queued for send_oos, however. |
a0d856df | 1404 | * So we can not simply free it, we must go through drbd_req_put_completion_ref() */ |
b411b363 | 1405 | } |
b411b363 | 1406 | |
5da9c836 | 1407 | out: |
a00ebd1c | 1408 | drbd_req_put_completion_ref(req, &m, 1); |
35b5ed5b LE |
1409 | spin_unlock_irq(&resource->req_lock); |
1410 | ||
1411 | /* Even though above is a kref_put(), this is safe. | |
1412 | * As long as we still need to submit our private bio, | |
1413 | * we hold a completion ref, and the request cannot disappear. | |
1414 | * If however this request did not even have a private bio to submit | |
1415 | * (e.g. remote read), req may already be invalid now. | |
1416 | * That's why we cannot check on req->private_bio. */ | |
1417 | if (submit_private_bio) | |
1418 | drbd_submit_req_private_bio(req); | |
5da9c836 | 1419 | if (m.bio) |
b30ab791 | 1420 | complete_master_bio(device, &m); |
6d9febe2 LE |
1421 | } |
1422 | ||
370276ba | 1423 | void __drbd_make_request(struct drbd_device *device, struct bio *bio) |
6d9febe2 | 1424 | { |
370276ba | 1425 | struct drbd_request *req = drbd_request_prepare(device, bio); |
6d9febe2 LE |
1426 | if (IS_ERR_OR_NULL(req)) |
1427 | return; | |
b30ab791 | 1428 | drbd_send_and_submit(device, req); |
b411b363 PR |
1429 | } |
1430 | ||
b30ab791 | 1431 | static void submit_fast_path(struct drbd_device *device, struct list_head *incoming) |
113fef9e | 1432 | { |
de6978be | 1433 | struct blk_plug plug; |
08a1ddab | 1434 | struct drbd_request *req, *tmp; |
de6978be LE |
1435 | |
1436 | blk_start_plug(&plug); | |
08a1ddab LE |
1437 | list_for_each_entry_safe(req, tmp, incoming, tl_requests) { |
1438 | const int rw = bio_data_dir(req->master_bio); | |
113fef9e | 1439 | |
08a1ddab LE |
1440 | if (rw == WRITE /* rw != WRITE should not even end up here! */ |
1441 | && req->private_bio && req->i.size | |
b30ab791 AG |
1442 | && !test_bit(AL_SUSPENDED, &device->flags)) { |
1443 | if (!drbd_al_begin_io_fastpath(device, &req->i)) | |
08a1ddab LE |
1444 | continue; |
1445 | ||
1446 | req->rq_state |= RQ_IN_ACT_LOG; | |
e5f891b2 | 1447 | req->in_actlog_jif = jiffies; |
ad3fee79 | 1448 | atomic_dec(&device->ap_actlog_cnt); |
08a1ddab LE |
1449 | } |
1450 | ||
1451 | list_del_init(&req->tl_requests); | |
b30ab791 | 1452 | drbd_send_and_submit(device, req); |
113fef9e | 1453 | } |
de6978be | 1454 | blk_finish_plug(&plug); |
113fef9e LE |
1455 | } |
1456 | ||
b30ab791 | 1457 | static bool prepare_al_transaction_nonblock(struct drbd_device *device, |
08a1ddab | 1458 | struct list_head *incoming, |
f5b90b6b LE |
1459 | struct list_head *pending, |
1460 | struct list_head *later) | |
08a1ddab | 1461 | { |
9da10e8d | 1462 | struct drbd_request *req; |
08a1ddab LE |
1463 | int wake = 0; |
1464 | int err; | |
1465 | ||
b30ab791 | 1466 | spin_lock_irq(&device->al_lock); |
9da10e8d | 1467 | while ((req = list_first_entry_or_null(incoming, struct drbd_request, tl_requests))) { |
b30ab791 | 1468 | err = drbd_al_begin_io_nonblock(device, &req->i); |
f5b90b6b LE |
1469 | if (err == -ENOBUFS) |
1470 | break; | |
08a1ddab LE |
1471 | if (err == -EBUSY) |
1472 | wake = 1; | |
1473 | if (err) | |
f5b90b6b LE |
1474 | list_move_tail(&req->tl_requests, later); |
1475 | else | |
1476 | list_move_tail(&req->tl_requests, pending); | |
08a1ddab | 1477 | } |
b30ab791 | 1478 | spin_unlock_irq(&device->al_lock); |
08a1ddab | 1479 | if (wake) |
b30ab791 | 1480 | wake_up(&device->al_wait); |
08a1ddab LE |
1481 | return !list_empty(pending); |
1482 | } | |
113fef9e | 1483 | |
de6978be | 1484 | static void send_and_submit_pending(struct drbd_device *device, struct list_head *pending) |
f5b90b6b | 1485 | { |
de6978be | 1486 | struct blk_plug plug; |
9da10e8d | 1487 | struct drbd_request *req; |
f5b90b6b | 1488 | |
de6978be | 1489 | blk_start_plug(&plug); |
9da10e8d | 1490 | while ((req = list_first_entry_or_null(pending, struct drbd_request, tl_requests))) { |
f5b90b6b LE |
1491 | req->rq_state |= RQ_IN_ACT_LOG; |
1492 | req->in_actlog_jif = jiffies; | |
1493 | atomic_dec(&device->ap_actlog_cnt); | |
1494 | list_del_init(&req->tl_requests); | |
1495 | drbd_send_and_submit(device, req); | |
1496 | } | |
de6978be | 1497 | blk_finish_plug(&plug); |
f5b90b6b LE |
1498 | } |
1499 | ||
113fef9e LE |
1500 | void do_submit(struct work_struct *ws) |
1501 | { | |
b30ab791 | 1502 | struct drbd_device *device = container_of(ws, struct drbd_device, submit.worker); |
f5b90b6b LE |
1503 | LIST_HEAD(incoming); /* from drbd_make_request() */ |
1504 | LIST_HEAD(pending); /* to be submitted after next AL-transaction commit */ | |
1505 | LIST_HEAD(busy); /* blocked by resync requests */ | |
1506 | ||
1507 | /* grab new incoming requests */ | |
1508 | spin_lock_irq(&device->resource->req_lock); | |
1509 | list_splice_tail_init(&device->submit.writes, &incoming); | |
1510 | spin_unlock_irq(&device->resource->req_lock); | |
113fef9e | 1511 | |
08a1ddab | 1512 | for (;;) { |
f5b90b6b | 1513 | DEFINE_WAIT(wait); |
113fef9e | 1514 | |
f5b90b6b LE |
1515 | /* move used-to-be-busy back to front of incoming */ |
1516 | list_splice_init(&busy, &incoming); | |
b30ab791 | 1517 | submit_fast_path(device, &incoming); |
08a1ddab LE |
1518 | if (list_empty(&incoming)) |
1519 | break; | |
1520 | ||
45ad07b3 | 1521 | for (;;) { |
f5b90b6b LE |
1522 | prepare_to_wait(&device->al_wait, &wait, TASK_UNINTERRUPTIBLE); |
1523 | ||
1524 | list_splice_init(&busy, &incoming); | |
1525 | prepare_al_transaction_nonblock(device, &incoming, &pending, &busy); | |
1526 | if (!list_empty(&pending)) | |
1527 | break; | |
1528 | ||
1529 | schedule(); | |
1530 | ||
1531 | /* If all currently "hot" activity log extents are kept busy by | |
1532 | * incoming requests, we still must not totally starve new | |
1533 | * requests to "cold" extents. | |
1534 | * Something left on &incoming means there had not been | |
1535 | * enough update slots available, and the activity log | |
1536 | * has been marked as "starving". | |
1537 | * | |
1538 | * Try again now, without looking for new requests, | |
1539 | * effectively blocking all new requests until we made | |
1540 | * at least _some_ progress with what we currently have. | |
1541 | */ | |
1542 | if (!list_empty(&incoming)) | |
1543 | continue; | |
1544 | ||
1545 | /* Nothing moved to pending, but nothing left | |
1546 | * on incoming: all moved to busy! | |
1547 | * Grab new and iterate. */ | |
1548 | spin_lock_irq(&device->resource->req_lock); | |
1549 | list_splice_tail_init(&device->submit.writes, &incoming); | |
1550 | spin_unlock_irq(&device->resource->req_lock); | |
1551 | } | |
1552 | finish_wait(&device->al_wait, &wait); | |
1553 | ||
1554 | /* If the transaction was full, before all incoming requests | |
1555 | * had been processed, skip ahead to commit, and iterate | |
1556 | * without splicing in more incoming requests from upper layers. | |
1557 | * | |
1558 | * Else, if all incoming have been processed, | |
1559 | * they have become either "pending" (to be submitted after | |
1560 | * next transaction commit) or "busy" (blocked by resync). | |
1561 | * | |
1562 | * Maybe more was queued, while we prepared the transaction? | |
1563 | * Try to stuff those into this transaction as well. | |
1564 | * Be strictly non-blocking here, | |
1565 | * we already have something to commit. | |
1566 | * | |
1567 | * Commit if we don't make any more progres. | |
1568 | */ | |
1569 | ||
1570 | while (list_empty(&incoming)) { | |
45ad07b3 LE |
1571 | LIST_HEAD(more_pending); |
1572 | LIST_HEAD(more_incoming); | |
1573 | bool made_progress; | |
1574 | ||
1575 | /* It is ok to look outside the lock, | |
1576 | * it's only an optimization anyways */ | |
b30ab791 | 1577 | if (list_empty(&device->submit.writes)) |
45ad07b3 LE |
1578 | break; |
1579 | ||
844a6ae7 | 1580 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 1581 | list_splice_tail_init(&device->submit.writes, &more_incoming); |
844a6ae7 | 1582 | spin_unlock_irq(&device->resource->req_lock); |
45ad07b3 LE |
1583 | |
1584 | if (list_empty(&more_incoming)) | |
1585 | break; | |
1586 | ||
f5b90b6b | 1587 | made_progress = prepare_al_transaction_nonblock(device, &more_incoming, &more_pending, &busy); |
45ad07b3 LE |
1588 | |
1589 | list_splice_tail_init(&more_pending, &pending); | |
1590 | list_splice_tail_init(&more_incoming, &incoming); | |
45ad07b3 LE |
1591 | if (!made_progress) |
1592 | break; | |
1593 | } | |
08a1ddab | 1594 | |
f5b90b6b LE |
1595 | drbd_al_begin_io_commit(device); |
1596 | send_and_submit_pending(device, &pending); | |
113fef9e LE |
1597 | } |
1598 | } | |
1599 | ||
c62b37d9 | 1600 | blk_qc_t drbd_submit_bio(struct bio *bio) |
b411b363 | 1601 | { |
309dca30 | 1602 | struct drbd_device *device = bio->bi_bdev->bd_disk->private_data; |
b411b363 | 1603 | |
f695ca38 | 1604 | blk_queue_split(&bio); |
54efd50b | 1605 | |
b411b363 PR |
1606 | /* |
1607 | * what we "blindly" assume: | |
1608 | */ | |
0b0ba1ef | 1609 | D_ASSERT(device, IS_ALIGNED(bio->bi_iter.bi_size, 512)); |
b411b363 | 1610 | |
b30ab791 | 1611 | inc_ap_bio(device); |
370276ba | 1612 | __drbd_make_request(device, bio); |
dece1635 | 1613 | return BLK_QC_T_NONE; |
b411b363 PR |
1614 | } |
1615 | ||
84d34f2f LE |
1616 | static bool net_timeout_reached(struct drbd_request *net_req, |
1617 | struct drbd_connection *connection, | |
1618 | unsigned long now, unsigned long ent, | |
1619 | unsigned int ko_count, unsigned int timeout) | |
1620 | { | |
1621 | struct drbd_device *device = net_req->device; | |
1622 | ||
1623 | if (!time_after(now, net_req->pre_send_jif + ent)) | |
1624 | return false; | |
1625 | ||
1626 | if (time_in_range(now, connection->last_reconnect_jif, connection->last_reconnect_jif + ent)) | |
1627 | return false; | |
1628 | ||
1629 | if (net_req->rq_state & RQ_NET_PENDING) { | |
1630 | drbd_warn(device, "Remote failed to finish a request within %ums > ko-count (%u) * timeout (%u * 0.1s)\n", | |
1631 | jiffies_to_msecs(now - net_req->pre_send_jif), ko_count, timeout); | |
1632 | return true; | |
1633 | } | |
1634 | ||
1635 | /* We received an ACK already (or are using protocol A), | |
1636 | * but are waiting for the epoch closing barrier ack. | |
1637 | * Check if we sent the barrier already. We should not blame the peer | |
1638 | * for being unresponsive, if we did not even ask it yet. */ | |
1639 | if (net_req->epoch == connection->send.current_epoch_nr) { | |
1640 | drbd_warn(device, | |
1641 | "We did not send a P_BARRIER for %ums > ko-count (%u) * timeout (%u * 0.1s); drbd kernel thread blocked?\n", | |
1642 | jiffies_to_msecs(now - net_req->pre_send_jif), ko_count, timeout); | |
1643 | return false; | |
1644 | } | |
1645 | ||
1646 | /* Worst case: we may have been blocked for whatever reason, then | |
1647 | * suddenly are able to send a lot of requests (and epoch separating | |
1648 | * barriers) in quick succession. | |
1649 | * The timestamp of the net_req may be much too old and not correspond | |
1650 | * to the sending time of the relevant unack'ed barrier packet, so | |
1651 | * would trigger a spurious timeout. The latest barrier packet may | |
1652 | * have a too recent timestamp to trigger the timeout, potentially miss | |
1653 | * a timeout. Right now we don't have a place to conveniently store | |
1654 | * these timestamps. | |
1655 | * But in this particular situation, the application requests are still | |
1656 | * completed to upper layers, DRBD should still "feel" responsive. | |
1657 | * No need yet to kill this connection, it may still recover. | |
1658 | * If not, eventually we will have queued enough into the network for | |
1659 | * us to block. From that point of view, the timestamp of the last sent | |
1660 | * barrier packet is relevant enough. | |
1661 | */ | |
1662 | if (time_after(now, connection->send.last_sent_barrier_jif + ent)) { | |
1663 | drbd_warn(device, "Remote failed to answer a P_BARRIER (sent at %lu jif; now=%lu jif) within %ums > ko-count (%u) * timeout (%u * 0.1s)\n", | |
1664 | connection->send.last_sent_barrier_jif, now, | |
1665 | jiffies_to_msecs(now - connection->send.last_sent_barrier_jif), ko_count, timeout); | |
1666 | return true; | |
1667 | } | |
1668 | return false; | |
1669 | } | |
1670 | ||
1671 | /* A request is considered timed out, if | |
1672 | * - we have some effective timeout from the configuration, | |
1673 | * with some state restrictions applied, | |
1674 | * - the oldest request is waiting for a response from the network | |
1675 | * resp. the local disk, | |
1676 | * - the oldest request is in fact older than the effective timeout, | |
1677 | * - the connection was established (resp. disk was attached) | |
1678 | * for longer than the timeout already. | |
1679 | * Note that for 32bit jiffies and very stable connections/disks, | |
1680 | * we may have a wrap around, which is catched by | |
1681 | * !time_in_range(now, last_..._jif, last_..._jif + timeout). | |
1682 | * | |
1683 | * Side effect: once per 32bit wrap-around interval, which means every | |
1684 | * ~198 days with 250 HZ, we have a window where the timeout would need | |
1685 | * to expire twice (worst case) to become effective. Good enough. | |
1686 | */ | |
1687 | ||
2bccef39 | 1688 | void request_timer_fn(struct timer_list *t) |
7fde2be9 | 1689 | { |
2bccef39 | 1690 | struct drbd_device *device = from_timer(device, t, request_timer); |
a6b32bc3 | 1691 | struct drbd_connection *connection = first_peer_device(device)->connection; |
7753a4c1 | 1692 | struct drbd_request *req_read, *req_write, *req_peer; /* oldest request */ |
44ed167d | 1693 | struct net_conf *nc; |
7753a4c1 | 1694 | unsigned long oldest_submit_jif; |
dfa8bedb | 1695 | unsigned long ent = 0, dt = 0, et, nt; /* effective timeout = ko_count * timeout */ |
ba280c09 | 1696 | unsigned long now; |
84d34f2f | 1697 | unsigned int ko_count = 0, timeout = 0; |
7fde2be9 | 1698 | |
44ed167d | 1699 | rcu_read_lock(); |
bde89a9e | 1700 | nc = rcu_dereference(connection->net_conf); |
84d34f2f LE |
1701 | if (nc && device->state.conn >= C_WF_REPORT_PARAMS) { |
1702 | ko_count = nc->ko_count; | |
1703 | timeout = nc->timeout; | |
1704 | } | |
cdfda633 | 1705 | |
b30ab791 AG |
1706 | if (get_ldev(device)) { /* implicit state.disk >= D_INCONSISTENT */ |
1707 | dt = rcu_dereference(device->ldev->disk_conf)->disk_timeout * HZ / 10; | |
1708 | put_ldev(device); | |
dfa8bedb | 1709 | } |
44ed167d | 1710 | rcu_read_unlock(); |
7fde2be9 | 1711 | |
84d34f2f LE |
1712 | |
1713 | ent = timeout * HZ/10 * ko_count; | |
dfa8bedb PR |
1714 | et = min_not_zero(dt, ent); |
1715 | ||
ba280c09 | 1716 | if (!et) |
7fde2be9 PR |
1717 | return; /* Recurring timer stopped */ |
1718 | ||
ba280c09 | 1719 | now = jiffies; |
7753a4c1 | 1720 | nt = now + et; |
ba280c09 | 1721 | |
0500813f | 1722 | spin_lock_irq(&device->resource->req_lock); |
7753a4c1 LE |
1723 | req_read = list_first_entry_or_null(&device->pending_completion[0], struct drbd_request, req_pending_local); |
1724 | req_write = list_first_entry_or_null(&device->pending_completion[1], struct drbd_request, req_pending_local); | |
84d34f2f | 1725 | |
7753a4c1 | 1726 | /* maybe the oldest request waiting for the peer is in fact still |
84d34f2f LE |
1727 | * blocking in tcp sendmsg. That's ok, though, that's handled via the |
1728 | * socket send timeout, requesting a ping, and bumping ko-count in | |
1729 | * we_should_drop_the_connection(). | |
1730 | */ | |
1731 | ||
1732 | /* check the oldest request we did successfully sent, | |
1733 | * but which is still waiting for an ACK. */ | |
1734 | req_peer = connection->req_ack_pending; | |
1735 | ||
1736 | /* if we don't have such request (e.g. protocoll A) | |
1737 | * check the oldest requests which is still waiting on its epoch | |
1738 | * closing barrier ack. */ | |
1739 | if (!req_peer) | |
1740 | req_peer = connection->req_not_net_done; | |
7753a4c1 LE |
1741 | |
1742 | /* evaluate the oldest peer request only in one timer! */ | |
1743 | if (req_peer && req_peer->device != device) | |
1744 | req_peer = NULL; | |
1745 | ||
1746 | /* do we have something to evaluate? */ | |
1747 | if (req_peer == NULL && req_write == NULL && req_read == NULL) | |
1748 | goto out; | |
1749 | ||
1750 | oldest_submit_jif = | |
1751 | (req_write && req_read) | |
1752 | ? ( time_before(req_write->pre_submit_jif, req_read->pre_submit_jif) | |
1753 | ? req_write->pre_submit_jif : req_read->pre_submit_jif ) | |
1754 | : req_write ? req_write->pre_submit_jif | |
1755 | : req_read ? req_read->pre_submit_jif : now; | |
7fde2be9 | 1756 | |
84d34f2f | 1757 | if (ent && req_peer && net_timeout_reached(req_peer, connection, now, ent, ko_count, timeout)) |
9581f97a | 1758 | _conn_request_state(connection, NS(conn, C_TIMEOUT), CS_VERBOSE | CS_HARD); |
84d34f2f | 1759 | |
7753a4c1 LE |
1760 | if (dt && oldest_submit_jif != now && |
1761 | time_after(now, oldest_submit_jif + dt) && | |
b30ab791 | 1762 | !time_in_range(now, device->last_reattach_jif, device->last_reattach_jif + dt)) { |
d0180171 | 1763 | drbd_warn(device, "Local backing device failed to meet the disk-timeout\n"); |
b30ab791 | 1764 | __drbd_chk_io_error(device, DRBD_FORCE_DETACH); |
dfa8bedb | 1765 | } |
08535466 LE |
1766 | |
1767 | /* Reschedule timer for the nearest not already expired timeout. | |
1768 | * Fallback to now + min(effective network timeout, disk timeout). */ | |
7753a4c1 LE |
1769 | ent = (ent && req_peer && time_before(now, req_peer->pre_send_jif + ent)) |
1770 | ? req_peer->pre_send_jif + ent : now + et; | |
1771 | dt = (dt && oldest_submit_jif != now && time_before(now, oldest_submit_jif + dt)) | |
1772 | ? oldest_submit_jif + dt : now + et; | |
08535466 | 1773 | nt = time_before(ent, dt) ? ent : dt; |
7753a4c1 | 1774 | out: |
8d4ba3f0 | 1775 | spin_unlock_irq(&device->resource->req_lock); |
b30ab791 | 1776 | mod_timer(&device->request_timer, nt); |
7fde2be9 | 1777 | } |