Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_req.h | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2006-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 2006-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
8 | Copyright (C) 2006-2008, Philipp Reisner <philipp.reisner@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 | #ifndef _DRBD_REQ_H | |
26 | #define _DRBD_REQ_H | |
27 | ||
b411b363 PR |
28 | #include <linux/module.h> |
29 | ||
30 | #include <linux/slab.h> | |
31 | #include <linux/drbd.h> | |
32 | #include "drbd_int.h" | |
33 | #include "drbd_wrappers.h" | |
34 | ||
35 | /* The request callbacks will be called in irq context by the IDE drivers, | |
36 | and in Softirqs/Tasklets/BH context by the SCSI drivers, | |
37 | and by the receiver and worker in kernel-thread context. | |
38 | Try to get the locking right :) */ | |
39 | ||
40 | /* | |
41 | * Objects of type struct drbd_request do only exist on a R_PRIMARY node, and are | |
42 | * associated with IO requests originating from the block layer above us. | |
43 | * | |
44 | * There are quite a few things that may happen to a drbd request | |
45 | * during its lifetime. | |
46 | * | |
47 | * It will be created. | |
48 | * It will be marked with the intention to be | |
49 | * submitted to local disk and/or | |
50 | * send via the network. | |
51 | * | |
52 | * It has to be placed on the transfer log and other housekeeping lists, | |
53 | * In case we have a network connection. | |
54 | * | |
55 | * It may be identified as a concurrent (write) request | |
56 | * and be handled accordingly. | |
57 | * | |
58 | * It may me handed over to the local disk subsystem. | |
59 | * It may be completed by the local disk subsystem, | |
3ad2f3fb | 60 | * either successfully or with io-error. |
b411b363 PR |
61 | * In case it is a READ request, and it failed locally, |
62 | * it may be retried remotely. | |
63 | * | |
64 | * It may be queued for sending. | |
65 | * It may be handed over to the network stack, | |
66 | * which may fail. | |
67 | * It may be acknowledged by the "peer" according to the wire_protocol in use. | |
68 | * this may be a negative ack. | |
69 | * It may receive a faked ack when the network connection is lost and the | |
70 | * transfer log is cleaned up. | |
71 | * Sending may be canceled due to network connection loss. | |
72 | * When it finally has outlived its time, | |
73 | * corresponding dirty bits in the resync-bitmap may be cleared or set, | |
74 | * it will be destroyed, | |
75 | * and completion will be signalled to the originator, | |
76 | * with or without "success". | |
77 | */ | |
78 | ||
79 | enum drbd_req_event { | |
8554df1c AG |
80 | CREATED, |
81 | TO_BE_SENT, | |
82 | TO_BE_SUBMITTED, | |
b411b363 PR |
83 | |
84 | /* XXX yes, now I am inconsistent... | |
73a01a18 | 85 | * these are not "events" but "actions" |
b411b363 | 86 | * oh, well... */ |
8554df1c AG |
87 | QUEUE_FOR_NET_WRITE, |
88 | QUEUE_FOR_NET_READ, | |
89 | QUEUE_FOR_SEND_OOS, | |
90 | ||
91 | SEND_CANCELED, | |
92 | SEND_FAILED, | |
93 | HANDED_OVER_TO_NETWORK, | |
94 | OOS_HANDED_TO_NETWORK, | |
95 | CONNECTION_LOST_WHILE_PENDING, | |
96 | READ_RETRY_REMOTE_CANCELED, | |
97 | RECV_ACKED_BY_PEER, | |
98 | WRITE_ACKED_BY_PEER, | |
99 | WRITE_ACKED_BY_PEER_AND_SIS, /* and set_in_sync */ | |
7be8da07 AG |
100 | DISCARD_WRITE, |
101 | POSTPONE_WRITE, | |
8554df1c AG |
102 | NEG_ACKED, |
103 | BARRIER_ACKED, /* in protocol A and B */ | |
104 | DATA_RECEIVED, /* (remote read) */ | |
105 | ||
106 | READ_COMPLETED_WITH_ERROR, | |
107 | READ_AHEAD_COMPLETED_WITH_ERROR, | |
108 | WRITE_COMPLETED_WITH_ERROR, | |
cdfda633 | 109 | ABORT_DISK_IO, |
8554df1c AG |
110 | COMPLETED_OK, |
111 | RESEND, | |
112 | FAIL_FROZEN_DISK_IO, | |
113 | RESTART_FROZEN_DISK_IO, | |
114 | NOTHING, | |
b411b363 PR |
115 | }; |
116 | ||
117 | /* encoding of request states for now. we don't actually need that many bits. | |
118 | * we don't need to do atomic bit operations either, since most of the time we | |
119 | * need to look at the connection state and/or manipulate some lists at the | |
120 | * same time, so we should hold the request lock anyways. | |
121 | */ | |
122 | enum drbd_req_state_bits { | |
cdfda633 PR |
123 | /* 3210 |
124 | * 0000: no local possible | |
125 | * 0001: to be submitted | |
b411b363 | 126 | * UNUSED, we could map: 011: submitted, completion still pending |
cdfda633 PR |
127 | * 0110: completed ok |
128 | * 0010: completed with error | |
129 | * 1001: Aborted (before completion) | |
130 | * 1x10: Aborted and completed -> free | |
b411b363 PR |
131 | */ |
132 | __RQ_LOCAL_PENDING, | |
133 | __RQ_LOCAL_COMPLETED, | |
134 | __RQ_LOCAL_OK, | |
cdfda633 | 135 | __RQ_LOCAL_ABORTED, |
b411b363 | 136 | |
cdfda633 | 137 | /* 87654 |
b411b363 PR |
138 | * 00000: no network possible |
139 | * 00001: to be send | |
140 | * 00011: to be send, on worker queue | |
141 | * 00101: sent, expecting recv_ack (B) or write_ack (C) | |
142 | * 11101: sent, | |
143 | * recv_ack (B) or implicit "ack" (A), | |
144 | * still waiting for the barrier ack. | |
145 | * master_bio may already be completed and invalidated. | |
8554df1c AG |
146 | * 11100: write acked (C), |
147 | * data received (for remote read, any protocol) | |
b411b363 PR |
148 | * or finally the barrier ack has arrived (B,A)... |
149 | * request can be freed | |
150 | * 01100: neg-acked (write, protocol C) | |
151 | * or neg-d-acked (read, any protocol) | |
152 | * or killed from the transfer log | |
153 | * during cleanup after connection loss | |
154 | * request can be freed | |
155 | * 01000: canceled or send failed... | |
156 | * request can be freed | |
157 | */ | |
158 | ||
159 | /* if "SENT" is not set, yet, this can still fail or be canceled. | |
160 | * if "SENT" is set already, we still wait for an Ack packet. | |
161 | * when cleared, the master_bio may be completed. | |
162 | * in (B,A) the request object may still linger on the transaction log | |
163 | * until the corresponding barrier ack comes in */ | |
164 | __RQ_NET_PENDING, | |
165 | ||
166 | /* If it is QUEUED, and it is a WRITE, it is also registered in the | |
167 | * transfer log. Currently we need this flag to avoid conflicts between | |
168 | * worker canceling the request and tl_clear_barrier killing it from | |
169 | * transfer log. We should restructure the code so this conflict does | |
170 | * no longer occur. */ | |
171 | __RQ_NET_QUEUED, | |
172 | ||
173 | /* well, actually only "handed over to the network stack". | |
174 | * | |
175 | * TODO can potentially be dropped because of the similar meaning | |
176 | * of RQ_NET_SENT and ~RQ_NET_QUEUED. | |
177 | * however it is not exactly the same. before we drop it | |
178 | * we must ensure that we can tell a request with network part | |
179 | * from a request without, regardless of what happens to it. */ | |
180 | __RQ_NET_SENT, | |
181 | ||
182 | /* when set, the request may be freed (if RQ_NET_QUEUED is clear). | |
183 | * basically this means the corresponding P_BARRIER_ACK was received */ | |
184 | __RQ_NET_DONE, | |
185 | ||
186 | /* whether or not we know (C) or pretend (B,A) that the write | |
187 | * was successfully written on the peer. | |
188 | */ | |
189 | __RQ_NET_OK, | |
190 | ||
191 | /* peer called drbd_set_in_sync() for this write */ | |
192 | __RQ_NET_SIS, | |
193 | ||
194 | /* keep this last, its for the RQ_NET_MASK */ | |
195 | __RQ_NET_MAX, | |
288f422e PR |
196 | |
197 | /* Set when this is a write, clear for a read */ | |
198 | __RQ_WRITE, | |
0778286a PR |
199 | |
200 | /* Should call drbd_al_complete_io() for this request... */ | |
201 | __RQ_IN_ACT_LOG, | |
7be8da07 AG |
202 | |
203 | /* The peer has sent a retry ACK */ | |
204 | __RQ_POSTPONED, | |
303d1448 PR |
205 | |
206 | /* We expect a receive ACK (wire proto B) */ | |
207 | __RQ_EXP_RECEIVE_ACK, | |
208 | ||
209 | /* We expect a write ACK (wite proto C) */ | |
210 | __RQ_EXP_WRITE_ACK, | |
b411b363 PR |
211 | }; |
212 | ||
213 | #define RQ_LOCAL_PENDING (1UL << __RQ_LOCAL_PENDING) | |
214 | #define RQ_LOCAL_COMPLETED (1UL << __RQ_LOCAL_COMPLETED) | |
215 | #define RQ_LOCAL_OK (1UL << __RQ_LOCAL_OK) | |
cdfda633 | 216 | #define RQ_LOCAL_ABORTED (1UL << __RQ_LOCAL_ABORTED) |
b411b363 | 217 | |
cdfda633 | 218 | #define RQ_LOCAL_MASK ((RQ_LOCAL_ABORTED << 1)-1) |
b411b363 PR |
219 | |
220 | #define RQ_NET_PENDING (1UL << __RQ_NET_PENDING) | |
221 | #define RQ_NET_QUEUED (1UL << __RQ_NET_QUEUED) | |
222 | #define RQ_NET_SENT (1UL << __RQ_NET_SENT) | |
223 | #define RQ_NET_DONE (1UL << __RQ_NET_DONE) | |
224 | #define RQ_NET_OK (1UL << __RQ_NET_OK) | |
225 | #define RQ_NET_SIS (1UL << __RQ_NET_SIS) | |
226 | ||
227 | /* 0x1f8 */ | |
228 | #define RQ_NET_MASK (((1UL << __RQ_NET_MAX)-1) & ~RQ_LOCAL_MASK) | |
229 | ||
288f422e | 230 | #define RQ_WRITE (1UL << __RQ_WRITE) |
0778286a | 231 | #define RQ_IN_ACT_LOG (1UL << __RQ_IN_ACT_LOG) |
7be8da07 | 232 | #define RQ_POSTPONED (1UL << __RQ_POSTPONED) |
303d1448 PR |
233 | #define RQ_EXP_RECEIVE_ACK (1UL << __RQ_EXP_RECEIVE_ACK) |
234 | #define RQ_EXP_WRITE_ACK (1UL << __RQ_EXP_WRITE_ACK) | |
288f422e | 235 | |
11b58e73 PR |
236 | /* For waking up the frozen transfer log mod_req() has to return if the request |
237 | should be counted in the epoch object*/ | |
f497609e AG |
238 | #define MR_WRITE 1 |
239 | #define MR_READ 2 | |
11b58e73 | 240 | |
5ba82308 PR |
241 | static inline void drbd_req_make_private_bio(struct drbd_request *req, struct bio *bio_src) |
242 | { | |
243 | struct bio *bio; | |
244 | bio = bio_clone(bio_src, GFP_NOIO); /* XXX cannot fail?? */ | |
245 | ||
246 | req->private_bio = bio; | |
247 | ||
248 | bio->bi_private = req; | |
fcefa62e | 249 | bio->bi_end_io = drbd_request_endio; |
5ba82308 PR |
250 | bio->bi_next = NULL; |
251 | } | |
252 | ||
b411b363 PR |
253 | /* Short lived temporary struct on the stack. |
254 | * We could squirrel the error to be returned into | |
255 | * bio->bi_size, or similar. But that would be too ugly. */ | |
256 | struct bio_and_error { | |
257 | struct bio *bio; | |
258 | int error; | |
259 | }; | |
260 | ||
261 | extern void _req_may_be_done(struct drbd_request *req, | |
262 | struct bio_and_error *m); | |
2a80699f | 263 | extern int __req_mod(struct drbd_request *req, enum drbd_req_event what, |
b411b363 PR |
264 | struct bio_and_error *m); |
265 | extern void complete_master_bio(struct drbd_conf *mdev, | |
266 | struct bio_and_error *m); | |
7fde2be9 | 267 | extern void request_timer_fn(unsigned long data); |
2f5cdd0b PR |
268 | extern void tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what); |
269 | extern void _tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what); | |
b411b363 PR |
270 | |
271 | /* use this if you don't want to deal with calling complete_master_bio() | |
272 | * outside the spinlock, e.g. when walking some list on cleanup. */ | |
2a80699f | 273 | static inline int _req_mod(struct drbd_request *req, enum drbd_req_event what) |
b411b363 | 274 | { |
a21e9298 | 275 | struct drbd_conf *mdev = req->w.mdev; |
b411b363 | 276 | struct bio_and_error m; |
2a80699f | 277 | int rv; |
b411b363 PR |
278 | |
279 | /* __req_mod possibly frees req, do not touch req after that! */ | |
2a80699f | 280 | rv = __req_mod(req, what, &m); |
b411b363 PR |
281 | if (m.bio) |
282 | complete_master_bio(mdev, &m); | |
2a80699f PR |
283 | |
284 | return rv; | |
b411b363 PR |
285 | } |
286 | ||
759fbdfb PR |
287 | /* completion of master bio is outside of our spinlock. |
288 | * We still may or may not be inside some irqs disabled section | |
289 | * of the lower level driver completion callback, so we need to | |
290 | * spin_lock_irqsave here. */ | |
2a80699f | 291 | static inline int req_mod(struct drbd_request *req, |
b411b363 PR |
292 | enum drbd_req_event what) |
293 | { | |
759fbdfb | 294 | unsigned long flags; |
a21e9298 | 295 | struct drbd_conf *mdev = req->w.mdev; |
b411b363 | 296 | struct bio_and_error m; |
2a80699f PR |
297 | int rv; |
298 | ||
87eeee41 | 299 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
2a80699f | 300 | rv = __req_mod(req, what, &m); |
87eeee41 | 301 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
b411b363 PR |
302 | |
303 | if (m.bio) | |
304 | complete_master_bio(mdev, &m); | |
2a80699f PR |
305 | |
306 | return rv; | |
b411b363 | 307 | } |
6a35c45f | 308 | |
da9fbc27 | 309 | static inline bool drbd_should_do_remote(union drbd_dev_state s) |
6a35c45f PR |
310 | { |
311 | return s.pdsk == D_UP_TO_DATE || | |
312 | (s.pdsk >= D_INCONSISTENT && | |
313 | s.conn >= C_WF_BITMAP_T && | |
314 | s.conn < C_AHEAD); | |
315 | /* Before proto 96 that was >= CONNECTED instead of >= C_WF_BITMAP_T. | |
316 | That is equivalent since before 96 IO was frozen in the C_WF_BITMAP* | |
317 | states. */ | |
318 | } | |
da9fbc27 | 319 | static inline bool drbd_should_send_out_of_sync(union drbd_dev_state s) |
6a35c45f PR |
320 | { |
321 | return s.conn == C_AHEAD || s.conn == C_WF_BITMAP_S; | |
322 | /* pdsk = D_INCONSISTENT as a consequence. Protocol 96 check not necessary | |
323 | since we enter state C_AHEAD only if proto >= 96 */ | |
324 | } | |
325 | ||
b411b363 | 326 | #endif |