4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
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>.
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)
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.
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.
28 #include <linux/module.h>
30 #include <linux/slab.h>
31 #include <linux/drbd.h>
33 #include "drbd_wrappers.h"
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 :) */
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.
44 * There are quite a few things that may happen to a drbd request
45 * during its lifetime.
48 * It will be marked with the intention to be
49 * submitted to local disk and/or
50 * send via the network.
52 * It has to be placed on the transfer log and other housekeeping lists,
53 * In case we have a network connection.
55 * It may be identified as a concurrent (write) request
56 * and be handled accordingly.
58 * It may me handed over to the local disk subsystem.
59 * It may be completed by the local disk subsystem,
60 * either successfully or with io-error.
61 * In case it is a READ request, and it failed locally,
62 * it may be retried remotely.
64 * It may be queued for sending.
65 * It may be handed over to the network stack,
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".
84 /* XXX yes, now I am inconsistent...
85 * these two are not "events" but "actions"
92 handed_over_to_network
,
93 connection_lost_while_pending
,
94 read_retry_remote_canceled
,
97 write_acked_by_peer_and_sis
, /* and set_in_sync */
98 conflict_discarded_by_peer
,
100 barrier_acked
, /* in protocol A and B */
101 data_received
, /* (remote read) */
103 read_completed_with_error
,
104 read_ahead_completed_with_error
,
105 write_completed_with_error
,
107 nothing
, /* for tracing only */
110 /* encoding of request states for now. we don't actually need that many bits.
111 * we don't need to do atomic bit operations either, since most of the time we
112 * need to look at the connection state and/or manipulate some lists at the
113 * same time, so we should hold the request lock anyways.
115 enum drbd_req_state_bits
{
117 * 000: no local possible
118 * 001: to be submitted
119 * UNUSED, we could map: 011: submitted, completion still pending
121 * 010: completed with error
124 __RQ_LOCAL_COMPLETED
,
128 * 00000: no network possible
130 * 00011: to be send, on worker queue
131 * 00101: sent, expecting recv_ack (B) or write_ack (C)
133 * recv_ack (B) or implicit "ack" (A),
134 * still waiting for the barrier ack.
135 * master_bio may already be completed and invalidated.
136 * 11100: write_acked (C),
137 * data_received (for remote read, any protocol)
138 * or finally the barrier ack has arrived (B,A)...
139 * request can be freed
140 * 01100: neg-acked (write, protocol C)
141 * or neg-d-acked (read, any protocol)
142 * or killed from the transfer log
143 * during cleanup after connection loss
144 * request can be freed
145 * 01000: canceled or send failed...
146 * request can be freed
149 /* if "SENT" is not set, yet, this can still fail or be canceled.
150 * if "SENT" is set already, we still wait for an Ack packet.
151 * when cleared, the master_bio may be completed.
152 * in (B,A) the request object may still linger on the transaction log
153 * until the corresponding barrier ack comes in */
156 /* If it is QUEUED, and it is a WRITE, it is also registered in the
157 * transfer log. Currently we need this flag to avoid conflicts between
158 * worker canceling the request and tl_clear_barrier killing it from
159 * transfer log. We should restructure the code so this conflict does
160 * no longer occur. */
163 /* well, actually only "handed over to the network stack".
165 * TODO can potentially be dropped because of the similar meaning
166 * of RQ_NET_SENT and ~RQ_NET_QUEUED.
167 * however it is not exactly the same. before we drop it
168 * we must ensure that we can tell a request with network part
169 * from a request without, regardless of what happens to it. */
172 /* when set, the request may be freed (if RQ_NET_QUEUED is clear).
173 * basically this means the corresponding P_BARRIER_ACK was received */
176 /* whether or not we know (C) or pretend (B,A) that the write
177 * was successfully written on the peer.
181 /* peer called drbd_set_in_sync() for this write */
184 /* keep this last, its for the RQ_NET_MASK */
188 #define RQ_LOCAL_PENDING (1UL << __RQ_LOCAL_PENDING)
189 #define RQ_LOCAL_COMPLETED (1UL << __RQ_LOCAL_COMPLETED)
190 #define RQ_LOCAL_OK (1UL << __RQ_LOCAL_OK)
192 #define RQ_LOCAL_MASK ((RQ_LOCAL_OK << 1)-1) /* 0x07 */
194 #define RQ_NET_PENDING (1UL << __RQ_NET_PENDING)
195 #define RQ_NET_QUEUED (1UL << __RQ_NET_QUEUED)
196 #define RQ_NET_SENT (1UL << __RQ_NET_SENT)
197 #define RQ_NET_DONE (1UL << __RQ_NET_DONE)
198 #define RQ_NET_OK (1UL << __RQ_NET_OK)
199 #define RQ_NET_SIS (1UL << __RQ_NET_SIS)
202 #define RQ_NET_MASK (((1UL << __RQ_NET_MAX)-1) & ~RQ_LOCAL_MASK)
206 struct hlist_head
*ee_hash_slot(struct drbd_conf
*mdev
, sector_t sector
)
208 BUG_ON(mdev
->ee_hash_s
== 0);
209 return mdev
->ee_hash
+
210 ((unsigned int)(sector
>>HT_SHIFT
) % mdev
->ee_hash_s
);
213 /* transfer log (drbd_request objects) */
215 struct hlist_head
*tl_hash_slot(struct drbd_conf
*mdev
, sector_t sector
)
217 BUG_ON(mdev
->tl_hash_s
== 0);
218 return mdev
->tl_hash
+
219 ((unsigned int)(sector
>>HT_SHIFT
) % mdev
->tl_hash_s
);
222 /* application reads (drbd_request objects) */
223 static struct hlist_head
*ar_hash_slot(struct drbd_conf
*mdev
, sector_t sector
)
225 return mdev
->app_reads_hash
226 + ((unsigned int)(sector
) % APP_R_HSIZE
);
229 /* when we receive the answer for a read request,
230 * verify that we actually know about it */
231 static inline struct drbd_request
*_ar_id_to_req(struct drbd_conf
*mdev
,
232 u64 id
, sector_t sector
)
234 struct hlist_head
*slot
= ar_hash_slot(mdev
, sector
);
235 struct hlist_node
*n
;
236 struct drbd_request
*req
;
238 hlist_for_each_entry(req
, n
, slot
, colision
) {
239 if ((unsigned long)req
== (unsigned long)id
) {
240 D_ASSERT(req
->sector
== sector
);
247 static inline struct drbd_request
*drbd_req_new(struct drbd_conf
*mdev
,
251 struct drbd_request
*req
=
252 mempool_alloc(drbd_request_mempool
, GFP_NOIO
);
254 bio
= bio_clone(bio_src
, GFP_NOIO
); /* XXX cannot fail?? */
258 req
->master_bio
= bio_src
;
259 req
->private_bio
= bio
;
261 req
->sector
= bio
->bi_sector
;
262 req
->size
= bio
->bi_size
;
263 req
->start_time
= jiffies
;
264 INIT_HLIST_NODE(&req
->colision
);
265 INIT_LIST_HEAD(&req
->tl_requests
);
266 INIT_LIST_HEAD(&req
->w
.list
);
268 bio
->bi_private
= req
;
269 bio
->bi_end_io
= drbd_endio_pri
;
275 static inline void drbd_req_free(struct drbd_request
*req
)
277 mempool_free(req
, drbd_request_mempool
);
280 static inline int overlaps(sector_t s1
, int l1
, sector_t s2
, int l2
)
282 return !((s1
+ (l1
>>9) <= s2
) || (s1
>= s2
+ (l2
>>9)));
285 /* Short lived temporary struct on the stack.
286 * We could squirrel the error to be returned into
287 * bio->bi_size, or similar. But that would be too ugly. */
288 struct bio_and_error
{
293 extern void _req_may_be_done(struct drbd_request
*req
,
294 struct bio_and_error
*m
);
295 extern void __req_mod(struct drbd_request
*req
, enum drbd_req_event what
,
296 struct bio_and_error
*m
);
297 extern void complete_master_bio(struct drbd_conf
*mdev
,
298 struct bio_and_error
*m
);
300 /* use this if you don't want to deal with calling complete_master_bio()
301 * outside the spinlock, e.g. when walking some list on cleanup. */
302 static inline void _req_mod(struct drbd_request
*req
, enum drbd_req_event what
)
304 struct drbd_conf
*mdev
= req
->mdev
;
305 struct bio_and_error m
;
307 /* __req_mod possibly frees req, do not touch req after that! */
308 __req_mod(req
, what
, &m
);
310 complete_master_bio(mdev
, &m
);
313 /* completion of master bio is outside of spinlock.
314 * If you need it irqsave, do it your self! */
315 static inline void req_mod(struct drbd_request
*req
,
316 enum drbd_req_event what
)
318 struct drbd_conf
*mdev
= req
->mdev
;
319 struct bio_and_error m
;
320 spin_lock_irq(&mdev
->req_lock
);
321 __req_mod(req
, what
, &m
);
322 spin_unlock_irq(&mdev
->req_lock
);
325 complete_master_bio(mdev
, &m
);