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         /* An empty flush is queued as P_BARRIER,
  92          * which will cause it to complete "successfully",
  93          * even if the local disk flush failed.
  94          *
  95          * Just like "real" requests, empty flushes (blkdev_issue_flush()) will
  96          * only see an error if neither local nor remote data is reachable. */
  97         QUEUE_AS_DRBD_BARRIER,
  98
  99         SEND_CANCELED,
 100         SEND_FAILED,
 101         HANDED_OVER_TO_NETWORK,
 102         OOS_HANDED_TO_NETWORK,
 103         CONNECTION_LOST_WHILE_PENDING,
 104         READ_RETRY_REMOTE_CANCELED,
 105         RECV_ACKED_BY_PEER,
 106         WRITE_ACKED_BY_PEER,
 107         WRITE_ACKED_BY_PEER_AND_SIS, /* and set_in_sync */
 108         CONFLICT_RESOLVED,
 109         POSTPONE_WRITE,
 110         NEG_ACKED,
 111         BARRIER_ACKED, /* in protocol A and B */
 112         DATA_RECEIVED, /* (remote read) */
 113
 114         READ_COMPLETED_WITH_ERROR,
 115         READ_AHEAD_COMPLETED_WITH_ERROR,
 116         WRITE_COMPLETED_WITH_ERROR,
 117         ABORT_DISK_IO,
 118         COMPLETED_OK,
 119         RESEND,
 120         FAIL_FROZEN_DISK_IO,
 121         RESTART_FROZEN_DISK_IO,
 122         NOTHING,
 123 };
 124
 125 /* encoding of request states for now.  we don't actually need that many bits.
 126  * we don't need to do atomic bit operations either, since most of the time we
 127  * need to look at the connection state and/or manipulate some lists at the
 128  * same time, so we should hold the request lock anyways.
 129  */
 130 enum drbd_req_state_bits {
 131         /* 3210
 132          * 0000: no local possible
 133          * 0001: to be submitted
 134          *    UNUSED, we could map: 011: submitted, completion still pending
 135          * 0110: completed ok
 136          * 0010: completed with error
 137          * 1001: Aborted (before completion)
 138          * 1x10: Aborted and completed -> free
 139          */
 140         __RQ_LOCAL_PENDING,
 141         __RQ_LOCAL_COMPLETED,
 142         __RQ_LOCAL_OK,
 143         __RQ_LOCAL_ABORTED,
 144
 145         /* 87654
 146          * 00000: no network possible
 147          * 00001: to be send
 148          * 00011: to be send, on worker queue
 149          * 00101: sent, expecting recv_ack (B) or write_ack (C)
 150          * 11101: sent,
 151          *        recv_ack (B) or implicit "ack" (A),
 152          *        still waiting for the barrier ack.
 153          *        master_bio may already be completed and invalidated.
 154          * 11100: write acked (C),
 155          *        data received (for remote read, any protocol)
 156          *        or finally the barrier ack has arrived (B,A)...
 157          *        request can be freed
 158          * 01100: neg-acked (write, protocol C)
 159          *        or neg-d-acked (read, any protocol)
 160          *        or killed from the transfer log
 161          *        during cleanup after connection loss
 162          *        request can be freed
 163          * 01000: canceled or send failed...
 164          *        request can be freed
 165          */
 166
 167         /* if "SENT" is not set, yet, this can still fail or be canceled.
 168          * if "SENT" is set already, we still wait for an Ack packet.
 169          * when cleared, the master_bio may be completed.
 170          * in (B,A) the request object may still linger on the transaction log
 171          * until the corresponding barrier ack comes in */
 172         __RQ_NET_PENDING,
 173
 174         /* If it is QUEUED, and it is a WRITE, it is also registered in the
 175          * transfer log. Currently we need this flag to avoid conflicts between
 176          * worker canceling the request and tl_clear_barrier killing it from
 177          * transfer log.  We should restructure the code so this conflict does
 178          * no longer occur. */
 179         __RQ_NET_QUEUED,
 180
 181         /* well, actually only "handed over to the network stack".
 182          *
 183          * TODO can potentially be dropped because of the similar meaning
 184          * of RQ_NET_SENT and ~RQ_NET_QUEUED.
 185          * however it is not exactly the same. before we drop it
 186          * we must ensure that we can tell a request with network part
 187          * from a request without, regardless of what happens to it. */
 188         __RQ_NET_SENT,
 189
 190         /* when set, the request may be freed (if RQ_NET_QUEUED is clear).
 191          * basically this means the corresponding P_BARRIER_ACK was received */
 192         __RQ_NET_DONE,
 193
 194         /* whether or not we know (C) or pretend (B,A) that the write
 195          * was successfully written on the peer.
 196          */
 197         __RQ_NET_OK,
 198
 199         /* peer called drbd_set_in_sync() for this write */
 200         __RQ_NET_SIS,
 201
 202         /* keep this last, its for the RQ_NET_MASK */
 203         __RQ_NET_MAX,
 204
 205         /* Set when this is a write, clear for a read */
 206         __RQ_WRITE,
 207
 208         /* Should call drbd_al_complete_io() for this request... */
 209         __RQ_IN_ACT_LOG,
 210
 211         /* The peer has sent a retry ACK */
 212         __RQ_POSTPONED,
 213
 214         /* would have been completed,
 215          * but was not, because of drbd_suspended() */
 216         __RQ_COMPLETION_SUSP,
 217
 218         /* We expect a receive ACK (wire proto B) */
 219         __RQ_EXP_RECEIVE_ACK,
 220
 221         /* We expect a write ACK (wite proto C) */
 222         __RQ_EXP_WRITE_ACK,
 223
 224         /* waiting for a barrier ack, did an extra kref_get */
 225         __RQ_EXP_BARR_ACK,
 226 };
 227
 228 #define RQ_LOCAL_PENDING   (1UL << __RQ_LOCAL_PENDING)
 229 #define RQ_LOCAL_COMPLETED (1UL << __RQ_LOCAL_COMPLETED)
 230 #define RQ_LOCAL_OK        (1UL << __RQ_LOCAL_OK)
 231 #define RQ_LOCAL_ABORTED   (1UL << __RQ_LOCAL_ABORTED)
 232
 233 #define RQ_LOCAL_MASK      ((RQ_LOCAL_ABORTED << 1)-1)
 234
 235 #define RQ_NET_PENDING     (1UL << __RQ_NET_PENDING)
 236 #define RQ_NET_QUEUED      (1UL << __RQ_NET_QUEUED)
 237 #define RQ_NET_SENT        (1UL << __RQ_NET_SENT)
 238 #define RQ_NET_DONE        (1UL << __RQ_NET_DONE)
 239 #define RQ_NET_OK          (1UL << __RQ_NET_OK)
 240 #define RQ_NET_SIS         (1UL << __RQ_NET_SIS)
 241
 242 /* 0x1f8 */
 243 #define RQ_NET_MASK        (((1UL << __RQ_NET_MAX)-1) & ~RQ_LOCAL_MASK)
 244
 245 #define RQ_WRITE           (1UL << __RQ_WRITE)
 246 #define RQ_IN_ACT_LOG      (1UL << __RQ_IN_ACT_LOG)
 247 #define RQ_POSTPONED       (1UL << __RQ_POSTPONED)
 248 #define RQ_COMPLETION_SUSP (1UL << __RQ_COMPLETION_SUSP)
 249 #define RQ_EXP_RECEIVE_ACK (1UL << __RQ_EXP_RECEIVE_ACK)
 250 #define RQ_EXP_WRITE_ACK   (1UL << __RQ_EXP_WRITE_ACK)
 251 #define RQ_EXP_BARR_ACK    (1UL << __RQ_EXP_BARR_ACK)
 252
 253 /* For waking up the frozen transfer log mod_req() has to return if the request
 254    should be counted in the epoch object*/
 255 #define MR_WRITE       1
 256 #define MR_READ        2
 257
 258 static inline void drbd_req_make_private_bio(struct drbd_request *req, struct bio *bio_src)
 259 {
 260         struct bio *bio;
 261         bio = bio_clone(bio_src, GFP_NOIO); /* XXX cannot fail?? */
 262
 263         req->private_bio = bio;
 264
 265         bio->bi_private  = req;
 266         bio->bi_end_io   = drbd_request_endio;
 267         bio->bi_next     = NULL;
 268 }
 269
 270 /* Short lived temporary struct on the stack.
 271  * We could squirrel the error to be returned into
 272  * bio->bi_size, or similar. But that would be too ugly. */
 273 struct bio_and_error {
 274         struct bio *bio;
 275         int error;
 276 };
 277
 278 extern void start_new_tl_epoch(struct drbd_tconn *tconn);
 279 extern void drbd_req_destroy(struct kref *kref);
 280 extern void _req_may_be_done(struct drbd_request *req,
 281                 struct bio_and_error *m);
 282 extern int __req_mod(struct drbd_request *req, enum drbd_req_event what,
 283                 struct bio_and_error *m);
 284 extern void complete_master_bio(struct drbd_conf *mdev,
 285                 struct bio_and_error *m);
 286 extern void request_timer_fn(unsigned long data);
 287 extern void tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what);
 288 extern void _tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what);
 289
 290 /* this is in drbd_main.c */
 291 extern void drbd_restart_request(struct drbd_request *req);
 292
 293 /* use this if you don't want to deal with calling complete_master_bio()
 294  * outside the spinlock, e.g. when walking some list on cleanup. */
 295 static inline int _req_mod(struct drbd_request *req, enum drbd_req_event what)
 296 {
 297         struct drbd_conf *mdev = req->w.mdev;
 298         struct bio_and_error m;
 299         int rv;
 300
 301         /* __req_mod possibly frees req, do not touch req after that! */
 302         rv = __req_mod(req, what, &m);
 303         if (m.bio)
 304                 complete_master_bio(mdev, &m);
 305
 306         return rv;
 307 }
 308
 309 /* completion of master bio is outside of our spinlock.
 310  * We still may or may not be inside some irqs disabled section
 311  * of the lower level driver completion callback, so we need to
 312  * spin_lock_irqsave here. */
 313 static inline int req_mod(struct drbd_request *req,
 314                 enum drbd_req_event what)
 315 {
 316         unsigned long flags;
 317         struct drbd_conf *mdev = req->w.mdev;
 318         struct bio_and_error m;
 319         int rv;
 320
 321         spin_lock_irqsave(&mdev->tconn->req_lock, flags);
 322         rv = __req_mod(req, what, &m);
 323         spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
 324
 325         if (m.bio)
 326                 complete_master_bio(mdev, &m);
 327
 328         return rv;
 329 }
 330
 331 static inline bool drbd_should_do_remote(union drbd_dev_state s)
 332 {
 333         return s.pdsk == D_UP_TO_DATE ||
 334                 (s.pdsk >= D_INCONSISTENT &&
 335                  s.conn >= C_WF_BITMAP_T &&
 336                  s.conn < C_AHEAD);
 337         /* Before proto 96 that was >= CONNECTED instead of >= C_WF_BITMAP_T.
 338            That is equivalent since before 96 IO was frozen in the C_WF_BITMAP*
 339            states. */
 340 }
 341 static inline bool drbd_should_send_out_of_sync(union drbd_dev_state s)
 342 {
 343         return s.conn == C_AHEAD || s.conn == C_WF_BITMAP_S;
 344         /* pdsk = D_INCONSISTENT as a consequence. Protocol 96 check not necessary
 345            since we enter state C_AHEAD only if proto >= 96 */
 346 }
 347
 348 #endif