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 are not "events" but "actions"
93 HANDED_OVER_TO_NETWORK,
94 OOS_HANDED_TO_NETWORK,
95 CONNECTION_LOST_WHILE_PENDING,
96 READ_RETRY_REMOTE_CANCELED,
99 WRITE_ACKED_BY_PEER_AND_SIS, /* and set_in_sync */
103 BARRIER_ACKED, /* in protocol A and B */
104 DATA_RECEIVED, /* (remote read) */
106 READ_COMPLETED_WITH_ERROR,
107 READ_AHEAD_COMPLETED_WITH_ERROR,
108 WRITE_COMPLETED_WITH_ERROR,
112 RESTART_FROZEN_DISK_IO,
116 /* encoding of request states for now. we don't actually need that many bits.
117 * we don't need to do atomic bit operations either, since most of the time we
118 * need to look at the connection state and/or manipulate some lists at the
119 * same time, so we should hold the request lock anyways.
121 enum drbd_req_state_bits {
123 * 000: no local possible
124 * 001: to be submitted
125 * UNUSED, we could map: 011: submitted, completion still pending
127 * 010: completed with error
130 __RQ_LOCAL_COMPLETED,
134 * 00000: no network possible
136 * 00011: to be send, on worker queue
137 * 00101: sent, expecting recv_ack (B) or write_ack (C)
139 * recv_ack (B) or implicit "ack" (A),
140 * still waiting for the barrier ack.
141 * master_bio may already be completed and invalidated.
142 * 11100: write acked (C),
143 * data received (for remote read, any protocol)
144 * or finally the barrier ack has arrived (B,A)...
145 * request can be freed
146 * 01100: neg-acked (write, protocol C)
147 * or neg-d-acked (read, any protocol)
148 * or killed from the transfer log
149 * during cleanup after connection loss
150 * request can be freed
151 * 01000: canceled or send failed...
152 * request can be freed
155 /* if "SENT" is not set, yet, this can still fail or be canceled.
156 * if "SENT" is set already, we still wait for an Ack packet.
157 * when cleared, the master_bio may be completed.
158 * in (B,A) the request object may still linger on the transaction log
159 * until the corresponding barrier ack comes in */
162 /* If it is QUEUED, and it is a WRITE, it is also registered in the
163 * transfer log. Currently we need this flag to avoid conflicts between
164 * worker canceling the request and tl_clear_barrier killing it from
165 * transfer log. We should restructure the code so this conflict does
166 * no longer occur. */
169 /* well, actually only "handed over to the network stack".
171 * TODO can potentially be dropped because of the similar meaning
172 * of RQ_NET_SENT and ~RQ_NET_QUEUED.
173 * however it is not exactly the same. before we drop it
174 * we must ensure that we can tell a request with network part
175 * from a request without, regardless of what happens to it. */
178 /* when set, the request may be freed (if RQ_NET_QUEUED is clear).
179 * basically this means the corresponding P_BARRIER_ACK was received */
182 /* whether or not we know (C) or pretend (B,A) that the write
183 * was successfully written on the peer.
187 /* peer called drbd_set_in_sync() for this write */
190 /* keep this last, its for the RQ_NET_MASK */
193 /* Set when this is a write, clear for a read */
196 /* Should call drbd_al_complete_io() for this request... */
199 /* The peer has sent a retry ACK */
202 /* We expect a receive ACK (wire proto B) */
203 __RQ_EXP_RECEIVE_ACK,
205 /* We expect a write ACK (wite proto C) */
209 #define RQ_LOCAL_PENDING (1UL << __RQ_LOCAL_PENDING)
210 #define RQ_LOCAL_COMPLETED (1UL << __RQ_LOCAL_COMPLETED)
211 #define RQ_LOCAL_OK (1UL << __RQ_LOCAL_OK)
213 #define RQ_LOCAL_MASK ((RQ_LOCAL_OK << 1)-1) /* 0x07 */
215 #define RQ_NET_PENDING (1UL << __RQ_NET_PENDING)
216 #define RQ_NET_QUEUED (1UL << __RQ_NET_QUEUED)
217 #define RQ_NET_SENT (1UL << __RQ_NET_SENT)
218 #define RQ_NET_DONE (1UL << __RQ_NET_DONE)
219 #define RQ_NET_OK (1UL << __RQ_NET_OK)
220 #define RQ_NET_SIS (1UL << __RQ_NET_SIS)
223 #define RQ_NET_MASK (((1UL << __RQ_NET_MAX)-1) & ~RQ_LOCAL_MASK)
225 #define RQ_WRITE (1UL << __RQ_WRITE)
226 #define RQ_IN_ACT_LOG (1UL << __RQ_IN_ACT_LOG)
227 #define RQ_POSTPONED (1UL << __RQ_POSTPONED)
228 #define RQ_EXP_RECEIVE_ACK (1UL << __RQ_EXP_RECEIVE_ACK)
229 #define RQ_EXP_WRITE_ACK (1UL << __RQ_EXP_WRITE_ACK)
231 /* For waking up the frozen transfer log mod_req() has to return if the request
232 should be counted in the epoch object*/
233 #define MR_WRITE_SHIFT 0
234 #define MR_WRITE (1 << MR_WRITE_SHIFT)
235 #define MR_READ_SHIFT 1
236 #define MR_READ (1 << MR_READ_SHIFT)
238 static inline void drbd_req_make_private_bio(struct drbd_request *req, struct bio *bio_src)
241 bio = bio_clone(bio_src, GFP_NOIO); /* XXX cannot fail?? */
243 req->private_bio = bio;
245 bio->bi_private = req;
246 bio->bi_end_io = drbd_request_endio;
250 /* Short lived temporary struct on the stack.
251 * We could squirrel the error to be returned into
252 * bio->bi_size, or similar. But that would be too ugly. */
253 struct bio_and_error {
258 extern void _req_may_be_done(struct drbd_request *req,
259 struct bio_and_error *m);
260 extern int __req_mod(struct drbd_request *req, enum drbd_req_event what,
261 struct bio_and_error *m);
262 extern void complete_master_bio(struct drbd_conf *mdev,
263 struct bio_and_error *m);
264 extern void request_timer_fn(unsigned long data);
265 extern void tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what);
266 extern void _tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what);
268 /* use this if you don't want to deal with calling complete_master_bio()
269 * outside the spinlock, e.g. when walking some list on cleanup. */
270 static inline int _req_mod(struct drbd_request *req, enum drbd_req_event what)
272 struct drbd_conf *mdev = req->w.mdev;
273 struct bio_and_error m;
276 /* __req_mod possibly frees req, do not touch req after that! */
277 rv = __req_mod(req, what, &m);
279 complete_master_bio(mdev, &m);
284 /* completion of master bio is outside of our spinlock.
285 * We still may or may not be inside some irqs disabled section
286 * of the lower level driver completion callback, so we need to
287 * spin_lock_irqsave here. */
288 static inline int req_mod(struct drbd_request *req,
289 enum drbd_req_event what)
292 struct drbd_conf *mdev = req->w.mdev;
293 struct bio_and_error m;
296 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
297 rv = __req_mod(req, what, &m);
298 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
301 complete_master_bio(mdev, &m);
306 static inline bool drbd_should_do_remote(union drbd_dev_state s)
308 return s.pdsk == D_UP_TO_DATE ||
309 (s.pdsk >= D_INCONSISTENT &&
310 s.conn >= C_WF_BITMAP_T &&
312 /* Before proto 96 that was >= CONNECTED instead of >= C_WF_BITMAP_T.
313 That is equivalent since before 96 IO was frozen in the C_WF_BITMAP*
316 static inline bool drbd_should_send_out_of_sync(union drbd_dev_state s)
318 return s.conn == C_AHEAD || s.conn == C_WF_BITMAP_S;
319 /* pdsk = D_INCONSISTENT as a consequence. Protocol 96 check not necessary
320 since we enter state C_AHEAD only if proto >= 96 */