xref: /openbmc/linux/drivers/block/drbd/drbd_req.h (revision 261a9af6)
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 
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,
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.
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 {
80 	created,
81 	to_be_send,
82 	to_be_submitted,
83 
84 	/* XXX yes, now I am inconsistent...
85 	 * these are not "events" but "actions"
86 	 * oh, well... */
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 */
100 	conflict_discarded_by_peer,
101 	neg_acked,
102 	barrier_acked, /* in protocol A and B */
103 	data_received, /* (remote read) */
104 
105 	read_completed_with_error,
106 	read_ahead_completed_with_error,
107 	write_completed_with_error,
108 	completed_ok,
109 	resend,
110 	fail_frozen_disk_io,
111 	restart_frozen_disk_io,
112 	nothing, /* for tracing only */
113 };
114 
115 /* encoding of request states for now.  we don't actually need that many bits.
116  * we don't need to do atomic bit operations either, since most of the time we
117  * need to look at the connection state and/or manipulate some lists at the
118  * same time, so we should hold the request lock anyways.
119  */
120 enum drbd_req_state_bits {
121 	/* 210
122 	 * 000: no local possible
123 	 * 001: to be submitted
124 	 *    UNUSED, we could map: 011: submitted, completion still pending
125 	 * 110: completed ok
126 	 * 010: completed with error
127 	 */
128 	__RQ_LOCAL_PENDING,
129 	__RQ_LOCAL_COMPLETED,
130 	__RQ_LOCAL_OK,
131 
132 	/* 76543
133 	 * 00000: no network possible
134 	 * 00001: to be send
135 	 * 00011: to be send, on worker queue
136 	 * 00101: sent, expecting recv_ack (B) or write_ack (C)
137 	 * 11101: sent,
138 	 *        recv_ack (B) or implicit "ack" (A),
139 	 *        still waiting for the barrier ack.
140 	 *        master_bio may already be completed and invalidated.
141 	 * 11100: write_acked (C),
142 	 *        data_received (for remote read, any protocol)
143 	 *        or finally the barrier ack has arrived (B,A)...
144 	 *        request can be freed
145 	 * 01100: neg-acked (write, protocol C)
146 	 *        or neg-d-acked (read, any protocol)
147 	 *        or killed from the transfer log
148 	 *        during cleanup after connection loss
149 	 *        request can be freed
150 	 * 01000: canceled or send failed...
151 	 *        request can be freed
152 	 */
153 
154 	/* if "SENT" is not set, yet, this can still fail or be canceled.
155 	 * if "SENT" is set already, we still wait for an Ack packet.
156 	 * when cleared, the master_bio may be completed.
157 	 * in (B,A) the request object may still linger on the transaction log
158 	 * until the corresponding barrier ack comes in */
159 	__RQ_NET_PENDING,
160 
161 	/* If it is QUEUED, and it is a WRITE, it is also registered in the
162 	 * transfer log. Currently we need this flag to avoid conflicts between
163 	 * worker canceling the request and tl_clear_barrier killing it from
164 	 * transfer log.  We should restructure the code so this conflict does
165 	 * no longer occur. */
166 	__RQ_NET_QUEUED,
167 
168 	/* well, actually only "handed over to the network stack".
169 	 *
170 	 * TODO can potentially be dropped because of the similar meaning
171 	 * of RQ_NET_SENT and ~RQ_NET_QUEUED.
172 	 * however it is not exactly the same. before we drop it
173 	 * we must ensure that we can tell a request with network part
174 	 * from a request without, regardless of what happens to it. */
175 	__RQ_NET_SENT,
176 
177 	/* when set, the request may be freed (if RQ_NET_QUEUED is clear).
178 	 * basically this means the corresponding P_BARRIER_ACK was received */
179 	__RQ_NET_DONE,
180 
181 	/* whether or not we know (C) or pretend (B,A) that the write
182 	 * was successfully written on the peer.
183 	 */
184 	__RQ_NET_OK,
185 
186 	/* peer called drbd_set_in_sync() for this write */
187 	__RQ_NET_SIS,
188 
189 	/* keep this last, its for the RQ_NET_MASK */
190 	__RQ_NET_MAX,
191 
192 	/* Set when this is a write, clear for a read */
193 	__RQ_WRITE,
194 
195 	/* Should call drbd_al_complete_io() for this request... */
196 	__RQ_IN_ACT_LOG,
197 };
198 
199 #define RQ_LOCAL_PENDING   (1UL << __RQ_LOCAL_PENDING)
200 #define RQ_LOCAL_COMPLETED (1UL << __RQ_LOCAL_COMPLETED)
201 #define RQ_LOCAL_OK        (1UL << __RQ_LOCAL_OK)
202 
203 #define RQ_LOCAL_MASK      ((RQ_LOCAL_OK << 1)-1) /* 0x07 */
204 
205 #define RQ_NET_PENDING     (1UL << __RQ_NET_PENDING)
206 #define RQ_NET_QUEUED      (1UL << __RQ_NET_QUEUED)
207 #define RQ_NET_SENT        (1UL << __RQ_NET_SENT)
208 #define RQ_NET_DONE        (1UL << __RQ_NET_DONE)
209 #define RQ_NET_OK          (1UL << __RQ_NET_OK)
210 #define RQ_NET_SIS         (1UL << __RQ_NET_SIS)
211 
212 /* 0x1f8 */
213 #define RQ_NET_MASK        (((1UL << __RQ_NET_MAX)-1) & ~RQ_LOCAL_MASK)
214 
215 #define RQ_WRITE           (1UL << __RQ_WRITE)
216 #define RQ_IN_ACT_LOG      (1UL << __RQ_IN_ACT_LOG)
217 
218 /* For waking up the frozen transfer log mod_req() has to return if the request
219    should be counted in the epoch object*/
220 #define MR_WRITE_SHIFT 0
221 #define MR_WRITE       (1 << MR_WRITE_SHIFT)
222 #define MR_READ_SHIFT  1
223 #define MR_READ        (1 << MR_READ_SHIFT)
224 
225 /* epoch entries */
226 static inline
227 struct hlist_head *ee_hash_slot(struct drbd_conf *mdev, sector_t sector)
228 {
229 	BUG_ON(mdev->ee_hash_s == 0);
230 	return mdev->ee_hash +
231 		((unsigned int)(sector>>HT_SHIFT) % mdev->ee_hash_s);
232 }
233 
234 /* transfer log (drbd_request objects) */
235 static inline
236 struct hlist_head *tl_hash_slot(struct drbd_conf *mdev, sector_t sector)
237 {
238 	BUG_ON(mdev->tl_hash_s == 0);
239 	return mdev->tl_hash +
240 		((unsigned int)(sector>>HT_SHIFT) % mdev->tl_hash_s);
241 }
242 
243 /* application reads (drbd_request objects) */
244 static struct hlist_head *ar_hash_slot(struct drbd_conf *mdev, sector_t sector)
245 {
246 	return mdev->app_reads_hash
247 		+ ((unsigned int)(sector) % APP_R_HSIZE);
248 }
249 
250 /* when we receive the answer for a read request,
251  * verify that we actually know about it */
252 static inline struct drbd_request *_ar_id_to_req(struct drbd_conf *mdev,
253 	u64 id, sector_t sector)
254 {
255 	struct hlist_head *slot = ar_hash_slot(mdev, sector);
256 	struct hlist_node *n;
257 	struct drbd_request *req;
258 
259 	hlist_for_each_entry(req, n, slot, collision) {
260 		if ((unsigned long)req == (unsigned long)id) {
261 			D_ASSERT(req->sector == sector);
262 			return req;
263 		}
264 	}
265 	return NULL;
266 }
267 
268 static inline void drbd_req_make_private_bio(struct drbd_request *req, struct bio *bio_src)
269 {
270 	struct bio *bio;
271 	bio = bio_clone(bio_src, GFP_NOIO); /* XXX cannot fail?? */
272 
273 	req->private_bio = bio;
274 
275 	bio->bi_private  = req;
276 	bio->bi_end_io   = drbd_endio_pri;
277 	bio->bi_next     = NULL;
278 }
279 
280 static inline struct drbd_request *drbd_req_new(struct drbd_conf *mdev,
281 	struct bio *bio_src)
282 {
283 	struct drbd_request *req =
284 		mempool_alloc(drbd_request_mempool, GFP_NOIO);
285 	if (likely(req)) {
286 		drbd_req_make_private_bio(req, bio_src);
287 
288 		req->rq_state    = bio_data_dir(bio_src) == WRITE ? RQ_WRITE : 0;
289 		req->mdev        = mdev;
290 		req->master_bio  = bio_src;
291 		req->epoch       = 0;
292 		req->sector      = bio_src->bi_sector;
293 		req->size        = bio_src->bi_size;
294 		INIT_HLIST_NODE(&req->collision);
295 		INIT_LIST_HEAD(&req->tl_requests);
296 		INIT_LIST_HEAD(&req->w.list);
297 	}
298 	return req;
299 }
300 
301 static inline void drbd_req_free(struct drbd_request *req)
302 {
303 	mempool_free(req, drbd_request_mempool);
304 }
305 
306 static inline int overlaps(sector_t s1, int l1, sector_t s2, int l2)
307 {
308 	return !((s1 + (l1>>9) <= s2) || (s1 >= s2 + (l2>>9)));
309 }
310 
311 /* Short lived temporary struct on the stack.
312  * We could squirrel the error to be returned into
313  * bio->bi_size, or similar. But that would be too ugly. */
314 struct bio_and_error {
315 	struct bio *bio;
316 	int error;
317 };
318 
319 extern void _req_may_be_done(struct drbd_request *req,
320 		struct bio_and_error *m);
321 extern int __req_mod(struct drbd_request *req, enum drbd_req_event what,
322 		struct bio_and_error *m);
323 extern void complete_master_bio(struct drbd_conf *mdev,
324 		struct bio_and_error *m);
325 extern void request_timer_fn(unsigned long data);
326 extern void tl_restart(struct drbd_conf *mdev, enum drbd_req_event what);
327 
328 /* use this if you don't want to deal with calling complete_master_bio()
329  * outside the spinlock, e.g. when walking some list on cleanup. */
330 static inline int _req_mod(struct drbd_request *req, enum drbd_req_event what)
331 {
332 	struct drbd_conf *mdev = req->mdev;
333 	struct bio_and_error m;
334 	int rv;
335 
336 	/* __req_mod possibly frees req, do not touch req after that! */
337 	rv = __req_mod(req, what, &m);
338 	if (m.bio)
339 		complete_master_bio(mdev, &m);
340 
341 	return rv;
342 }
343 
344 /* completion of master bio is outside of our spinlock.
345  * We still may or may not be inside some irqs disabled section
346  * of the lower level driver completion callback, so we need to
347  * spin_lock_irqsave here. */
348 static inline int req_mod(struct drbd_request *req,
349 		enum drbd_req_event what)
350 {
351 	unsigned long flags;
352 	struct drbd_conf *mdev = req->mdev;
353 	struct bio_and_error m;
354 	int rv;
355 
356 	spin_lock_irqsave(&mdev->req_lock, flags);
357 	rv = __req_mod(req, what, &m);
358 	spin_unlock_irqrestore(&mdev->req_lock, flags);
359 
360 	if (m.bio)
361 		complete_master_bio(mdev, &m);
362 
363 	return rv;
364 }
365 
366 static inline bool drbd_should_do_remote(union drbd_state s)
367 {
368 	return s.pdsk == D_UP_TO_DATE ||
369 		(s.pdsk >= D_INCONSISTENT &&
370 		 s.conn >= C_WF_BITMAP_T &&
371 		 s.conn < C_AHEAD);
372 	/* Before proto 96 that was >= CONNECTED instead of >= C_WF_BITMAP_T.
373 	   That is equivalent since before 96 IO was frozen in the C_WF_BITMAP*
374 	   states. */
375 }
376 static inline bool drbd_should_send_oos(union drbd_state s)
377 {
378 	return s.conn == C_AHEAD || s.conn == C_WF_BITMAP_S;
379 	/* pdsk = D_INCONSISTENT as a consequence. Protocol 96 check not necessary
380 	   since we enter state C_AHEAD only if proto >= 96 */
381 }
382 
383 #endif
384