drivers/block/drbd/drbd_req.h

   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_SENT,
  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         DISCARD_WRITE,
 101         POSTPONE_WRITE,
 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,
 109         ABORT_DISK_IO,
 110         COMPLETED_OK,
 111         RESEND,
 112         FAIL_FROZEN_DISK_IO,
 113         RESTART_FROZEN_DISK_IO,
 114         NOTHING,
 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 {
 123         /* 3210
 124          * 0000: no local possible
 125          * 0001: to be submitted
 126          *    UNUSED, we could map: 011: submitted, completion still pending
 127          * 0110: completed ok
 128          * 0010: completed with error
 129          * 1001: Aborted (before completion)
 130          * 1x10: Aborted and completed -> free
 131          */
 132         __RQ_LOCAL_PENDING,
 133         __RQ_LOCAL_COMPLETED,
 134         __RQ_LOCAL_OK,
 135         __RQ_LOCAL_ABORTED,
 136
 137         /* 87654
 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.
 146          * 11100: write acked (C),
 147          *        data received (for remote read, any protocol)
 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,
 196
 197         /* Set when this is a write, clear for a read */
 198         __RQ_WRITE,
 199
 200         /* Should call drbd_al_complete_io() for this request... */
 201         __RQ_IN_ACT_LOG,
 202
 203         /* The peer has sent a retry ACK */
 204         __RQ_POSTPONED,
 205
 206         /* would have been completed,
 207          * but was not, because of drbd_suspended() */
 208         __RQ_COMPLETION_SUSP,
 209
 210         /* We expect a receive ACK (wire proto B) */
 211         __RQ_EXP_RECEIVE_ACK,
 212
 213         /* We expect a write ACK (wite proto C) */
 214         __RQ_EXP_WRITE_ACK,
 215
 216         /* waiting for a barrier ack, did an extra kref_get */
 217         __RQ_EXP_BARR_ACK,
 218 };
 219
 220 #define RQ_LOCAL_PENDING   (1UL << __RQ_LOCAL_PENDING)
 221 #define RQ_LOCAL_COMPLETED (1UL << __RQ_LOCAL_COMPLETED)
 222 #define RQ_LOCAL_OK        (1UL << __RQ_LOCAL_OK)
 223 #define RQ_LOCAL_ABORTED   (1UL << __RQ_LOCAL_ABORTED)
 224
 225 #define RQ_LOCAL_MASK      ((RQ_LOCAL_ABORTED << 1)-1)
 226
 227 #define RQ_NET_PENDING     (1UL << __RQ_NET_PENDING)
 228 #define RQ_NET_QUEUED      (1UL << __RQ_NET_QUEUED)
 229 #define RQ_NET_SENT        (1UL << __RQ_NET_SENT)
 230 #define RQ_NET_DONE        (1UL << __RQ_NET_DONE)
 231 #define RQ_NET_OK          (1UL << __RQ_NET_OK)
 232 #define RQ_NET_SIS         (1UL << __RQ_NET_SIS)
 233
 234 /* 0x1f8 */
 235 #define RQ_NET_MASK        (((1UL << __RQ_NET_MAX)-1) & ~RQ_LOCAL_MASK)
 236
 237 #define RQ_WRITE           (1UL << __RQ_WRITE)
 238 #define RQ_IN_ACT_LOG      (1UL << __RQ_IN_ACT_LOG)
 239 #define RQ_POSTPONED       (1UL << __RQ_POSTPONED)
 240 #define RQ_COMPLETION_SUSP (1UL << __RQ_COMPLETION_SUSP)
 241 #define RQ_EXP_RECEIVE_ACK (1UL << __RQ_EXP_RECEIVE_ACK)
 242 #define RQ_EXP_WRITE_ACK   (1UL << __RQ_EXP_WRITE_ACK)
 243 #define RQ_EXP_BARR_ACK    (1UL << __RQ_EXP_BARR_ACK)
 244
 245 /* For waking up the frozen transfer log mod_req() has to return if the request
 246    should be counted in the epoch object*/
 247 #define MR_WRITE       1
 248 #define MR_READ        2
 249
 250 static inline void drbd_req_make_private_bio(struct drbd_request *req, struct bio *bio_src)
 251 {
 252         struct bio *bio;
 253         bio = bio_clone(bio_src, GFP_NOIO); /* XXX cannot fail?? */
 254
 255         req->private_bio = bio;
 256
 257         bio->bi_private  = req;
 258         bio->bi_end_io   = drbd_request_endio;
 259         bio->bi_next     = NULL;
 260 }
 261
 262 /* Short lived temporary struct on the stack.
 263  * We could squirrel the error to be returned into
 264  * bio->bi_size, or similar. But that would be too ugly. */
 265 struct bio_and_error {
 266         struct bio *bio;
 267         int error;
 268 };
 269
 270 extern void drbd_req_destroy(struct kref *kref);
 271 extern void _req_may_be_done(struct drbd_request *req,
 272                 struct bio_and_error *m);
 273 extern int __req_mod(struct drbd_request *req, enum drbd_req_event what,
 274                 struct bio_and_error *m);
 275 extern void complete_master_bio(struct drbd_conf *mdev,
 276                 struct bio_and_error *m);
 277 extern void request_timer_fn(unsigned long data);
 278 extern void tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what);
 279 extern void _tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what);
 280
 281 /* this is in drbd_main.c */
 282 extern void drbd_restart_request(struct drbd_request *req);
 283
 284 /* use this if you don't want to deal with calling complete_master_bio()
 285  * outside the spinlock, e.g. when walking some list on cleanup. */
 286 static inline int _req_mod(struct drbd_request *req, enum drbd_req_event what)
 287 {
 288         struct drbd_conf *mdev = req->w.mdev;
 289         struct bio_and_error m;
 290         int rv;
 291
 292         /* __req_mod possibly frees req, do not touch req after that! */
 293         rv = __req_mod(req, what, &m);
 294         if (m.bio)
 295                 complete_master_bio(mdev, &m);
 296
 297         return rv;
 298 }
 299
 300 /* completion of master bio is outside of our spinlock.
 301  * We still may or may not be inside some irqs disabled section
 302  * of the lower level driver completion callback, so we need to
 303  * spin_lock_irqsave here. */
 304 static inline int req_mod(struct drbd_request *req,
 305                 enum drbd_req_event what)
 306 {
 307         unsigned long flags;
 308         struct drbd_conf *mdev = req->w.mdev;
 309         struct bio_and_error m;
 310         int rv;
 311
 312         spin_lock_irqsave(&mdev->tconn->req_lock, flags);
 313         rv = __req_mod(req, what, &m);
 314         spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
 315
 316         if (m.bio)
 317                 complete_master_bio(mdev, &m);
 318
 319         return rv;
 320 }
 321
 322 static inline bool drbd_should_do_remote(union drbd_dev_state s)
 323 {
 324         return s.pdsk == D_UP_TO_DATE ||
 325                 (s.pdsk >= D_INCONSISTENT &&
 326                  s.conn >= C_WF_BITMAP_T &&
 327                  s.conn < C_AHEAD);
 328         /* Before proto 96 that was >= CONNECTED instead of >= C_WF_BITMAP_T.
 329            That is equivalent since before 96 IO was frozen in the C_WF_BITMAP*
 330            states. */
 331 }
 332 static inline bool drbd_should_send_out_of_sync(union drbd_dev_state s)
 333 {
 334         return s.conn == C_AHEAD || s.conn == C_WF_BITMAP_S;
 335         /* pdsk = D_INCONSISTENT as a consequence. Protocol 96 check not necessary
 336            since we enter state C_AHEAD only if proto >= 96 */
 337 }
 338
 339 #endif