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
  34 /* The request callbacks will be called in irq context by the IDE drivers,
  35    and in Softirqs/Tasklets/BH context by the SCSI drivers,
  36    and by the receiver and worker in kernel-thread context.
  37    Try to get the locking right :) */
  38
  39 /*
  40  * Objects of type struct drbd_request do only exist on a R_PRIMARY node, and are
  41  * associated with IO requests originating from the block layer above us.
  42  *
  43  * There are quite a few things that may happen to a drbd request
  44  * during its lifetime.
  45  *
  46  *  It will be created.
  47  *  It will be marked with the intention to be
  48  *    submitted to local disk and/or
  49  *    send via the network.
  50  *
  51  *  It has to be placed on the transfer log and other housekeeping lists,
  52  *  In case we have a network connection.
  53  *
  54  *  It may be identified as a concurrent (write) request
  55  *    and be handled accordingly.
  56  *
  57  *  It may me handed over to the local disk subsystem.
  58  *  It may be completed by the local disk subsystem,
  59  *    either successfully or with io-error.
  60  *  In case it is a READ request, and it failed locally,
  61  *    it may be retried remotely.
  62  *
  63  *  It may be queued for sending.
  64  *  It may be handed over to the network stack,
  65  *    which may fail.
  66  *  It may be acknowledged by the "peer" according to the wire_protocol in use.
  67  *    this may be a negative ack.
  68  *  It may receive a faked ack when the network connection is lost and the
  69  *  transfer log is cleaned up.
  70  *  Sending may be canceled due to network connection loss.
  71  *  When it finally has outlived its time,
  72  *    corresponding dirty bits in the resync-bitmap may be cleared or set,
  73  *    it will be destroyed,
  74  *    and completion will be signalled to the originator,
  75  *      with or without "success".
  76  */
  77
  78 enum drbd_req_event {
  79         CREATED,
  80         TO_BE_SENT,
  81         TO_BE_SUBMITTED,
  82
  83         /* XXX yes, now I am inconsistent...
  84          * these are not "events" but "actions"
  85          * oh, well... */
  86         QUEUE_FOR_NET_WRITE,
  87         QUEUE_FOR_NET_READ,
  88         QUEUE_FOR_SEND_OOS,
  89
  90         /* An empty flush is queued as P_BARRIER,
  91          * which will cause it to complete "successfully",
  92          * even if the local disk flush failed.
  93          *
  94          * Just like "real" requests, empty flushes (blkdev_issue_flush()) will
  95          * only see an error if neither local nor remote data is reachable. */
  96         QUEUE_AS_DRBD_BARRIER,
  97
  98         SEND_CANCELED,
  99         SEND_FAILED,
 100         HANDED_OVER_TO_NETWORK,
 101         OOS_HANDED_TO_NETWORK,
 102         CONNECTION_LOST_WHILE_PENDING,
 103         READ_RETRY_REMOTE_CANCELED,
 104         RECV_ACKED_BY_PEER,
 105         WRITE_ACKED_BY_PEER,
 106         WRITE_ACKED_BY_PEER_AND_SIS, /* and set_in_sync */
 107         CONFLICT_RESOLVED,
 108         POSTPONE_WRITE,
 109         NEG_ACKED,
 110         BARRIER_ACKED, /* in protocol A and B */
 111         DATA_RECEIVED, /* (remote read) */
 112
 113         COMPLETED_OK,
 114         READ_COMPLETED_WITH_ERROR,
 115         READ_AHEAD_COMPLETED_WITH_ERROR,
 116         WRITE_COMPLETED_WITH_ERROR,
 117         DISCARD_COMPLETED_NOTSUPP,
 118         DISCARD_COMPLETED_WITH_ERROR,
 119
 120         ABORT_DISK_IO,
 121         RESEND,
 122         FAIL_FROZEN_DISK_IO,
 123         RESTART_FROZEN_DISK_IO,
 124         NOTHING,
 125 };
 126
 127 /* encoding of request states for now.  we don't actually need that many bits.
 128  * we don't need to do atomic bit operations either, since most of the time we
 129  * need to look at the connection state and/or manipulate some lists at the
 130  * same time, so we should hold the request lock anyways.
 131  */
 132 enum drbd_req_state_bits {
 133         /* 3210
 134          * 0000: no local possible
 135          * 0001: to be submitted
 136          *    UNUSED, we could map: 011: submitted, completion still pending
 137          * 0110: completed ok
 138          * 0010: completed with error
 139          * 1001: Aborted (before completion)
 140          * 1x10: Aborted and completed -> free
 141          */
 142         __RQ_LOCAL_PENDING,
 143         __RQ_LOCAL_COMPLETED,
 144         __RQ_LOCAL_OK,
 145         __RQ_LOCAL_ABORTED,
 146
 147         /* 87654
 148          * 00000: no network possible
 149          * 00001: to be send
 150          * 00011: to be send, on worker queue
 151          * 00101: sent, expecting recv_ack (B) or write_ack (C)
 152          * 11101: sent,
 153          *        recv_ack (B) or implicit "ack" (A),
 154          *        still waiting for the barrier ack.
 155          *        master_bio may already be completed and invalidated.
 156          * 11100: write acked (C),
 157          *        data received (for remote read, any protocol)
 158          *        or finally the barrier ack has arrived (B,A)...
 159          *        request can be freed
 160          * 01100: neg-acked (write, protocol C)
 161          *        or neg-d-acked (read, any protocol)
 162          *        or killed from the transfer log
 163          *        during cleanup after connection loss
 164          *        request can be freed
 165          * 01000: canceled or send failed...
 166          *        request can be freed
 167          */
 168
 169         /* if "SENT" is not set, yet, this can still fail or be canceled.
 170          * if "SENT" is set already, we still wait for an Ack packet.
 171          * when cleared, the master_bio may be completed.
 172          * in (B,A) the request object may still linger on the transaction log
 173          * until the corresponding barrier ack comes in */
 174         __RQ_NET_PENDING,
 175
 176         /* If it is QUEUED, and it is a WRITE, it is also registered in the
 177          * transfer log. Currently we need this flag to avoid conflicts between
 178          * worker canceling the request and tl_clear_barrier killing it from
 179          * transfer log.  We should restructure the code so this conflict does
 180          * no longer occur. */
 181         __RQ_NET_QUEUED,
 182
 183         /* well, actually only "handed over to the network stack".
 184          *
 185          * TODO can potentially be dropped because of the similar meaning
 186          * of RQ_NET_SENT and ~RQ_NET_QUEUED.
 187          * however it is not exactly the same. before we drop it
 188          * we must ensure that we can tell a request with network part
 189          * from a request without, regardless of what happens to it. */
 190         __RQ_NET_SENT,
 191
 192         /* when set, the request may be freed (if RQ_NET_QUEUED is clear).
 193          * basically this means the corresponding P_BARRIER_ACK was received */
 194         __RQ_NET_DONE,
 195
 196         /* whether or not we know (C) or pretend (B,A) that the write
 197          * was successfully written on the peer.
 198          */
 199         __RQ_NET_OK,
 200
 201         /* peer called drbd_set_in_sync() for this write */
 202         __RQ_NET_SIS,
 203
 204         /* keep this last, its for the RQ_NET_MASK */
 205         __RQ_NET_MAX,
 206
 207         /* Set when this is a write, clear for a read */
 208         __RQ_WRITE,
 209
 210         /* Should call drbd_al_complete_io() for this request... */
 211         __RQ_IN_ACT_LOG,
 212
 213         /* The peer has sent a retry ACK */
 214         __RQ_POSTPONED,
 215
 216         /* would have been completed,
 217          * but was not, because of drbd_suspended() */
 218         __RQ_COMPLETION_SUSP,
 219
 220         /* We expect a receive ACK (wire proto B) */
 221         __RQ_EXP_RECEIVE_ACK,
 222
 223         /* We expect a write ACK (wite proto C) */
 224         __RQ_EXP_WRITE_ACK,
 225
 226         /* waiting for a barrier ack, did an extra kref_get */
 227         __RQ_EXP_BARR_ACK,
 228 };
 229
 230 #define RQ_LOCAL_PENDING   (1UL << __RQ_LOCAL_PENDING)
 231 #define RQ_LOCAL_COMPLETED (1UL << __RQ_LOCAL_COMPLETED)
 232 #define RQ_LOCAL_OK        (1UL << __RQ_LOCAL_OK)
 233 #define RQ_LOCAL_ABORTED   (1UL << __RQ_LOCAL_ABORTED)
 234
 235 #define RQ_LOCAL_MASK      ((RQ_LOCAL_ABORTED << 1)-1)
 236
 237 #define RQ_NET_PENDING     (1UL << __RQ_NET_PENDING)
 238 #define RQ_NET_QUEUED      (1UL << __RQ_NET_QUEUED)
 239 #define RQ_NET_SENT        (1UL << __RQ_NET_SENT)
 240 #define RQ_NET_DONE        (1UL << __RQ_NET_DONE)
 241 #define RQ_NET_OK          (1UL << __RQ_NET_OK)
 242 #define RQ_NET_SIS         (1UL << __RQ_NET_SIS)
 243
 244 /* 0x1f8 */
 245 #define RQ_NET_MASK        (((1UL << __RQ_NET_MAX)-1) & ~RQ_LOCAL_MASK)
 246
 247 #define RQ_WRITE           (1UL << __RQ_WRITE)
 248 #define RQ_IN_ACT_LOG      (1UL << __RQ_IN_ACT_LOG)
 249 #define RQ_POSTPONED       (1UL << __RQ_POSTPONED)
 250 #define RQ_COMPLETION_SUSP (1UL << __RQ_COMPLETION_SUSP)
 251 #define RQ_EXP_RECEIVE_ACK (1UL << __RQ_EXP_RECEIVE_ACK)
 252 #define RQ_EXP_WRITE_ACK   (1UL << __RQ_EXP_WRITE_ACK)
 253 #define RQ_EXP_BARR_ACK    (1UL << __RQ_EXP_BARR_ACK)
 254
 255 /* For waking up the frozen transfer log mod_req() has to return if the request
 256    should be counted in the epoch object*/
 257 #define MR_WRITE       1
 258 #define MR_READ        2
 259
 260 static inline void drbd_req_make_private_bio(struct drbd_request *req, struct bio *bio_src)
 261 {
 262         struct bio *bio;
 263         bio = bio_clone(bio_src, GFP_NOIO); /* XXX cannot fail?? */
 264
 265         req->private_bio = bio;
 266
 267         bio->bi_private  = req;
 268         bio->bi_end_io   = drbd_request_endio;
 269         bio->bi_next     = NULL;
 270 }
 271
 272 /* Short lived temporary struct on the stack.
 273  * We could squirrel the error to be returned into
 274  * bio->bi_iter.bi_size, or similar. But that would be too ugly. */
 275 struct bio_and_error {
 276         struct bio *bio;
 277         int error;
 278 };
 279
 280 extern void start_new_tl_epoch(struct drbd_connection *connection);
 281 extern void drbd_req_destroy(struct kref *kref);
 282 extern void _req_may_be_done(struct drbd_request *req,
 283                 struct bio_and_error *m);
 284 extern int __req_mod(struct drbd_request *req, enum drbd_req_event what,
 285                 struct bio_and_error *m);
 286 extern void complete_master_bio(struct drbd_device *device,
 287                 struct bio_and_error *m);
 288 extern void request_timer_fn(unsigned long data);
 289 extern void tl_restart(struct drbd_connection *connection, enum drbd_req_event what);
 290 extern void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what);
 291 extern void tl_abort_disk_io(struct drbd_device *device);
 292
 293 /* this is in drbd_main.c */
 294 extern void drbd_restart_request(struct drbd_request *req);
 295
 296 /* use this if you don't want to deal with calling complete_master_bio()
 297  * outside the spinlock, e.g. when walking some list on cleanup. */
 298 static inline int _req_mod(struct drbd_request *req, enum drbd_req_event what)
 299 {
 300         struct drbd_device *device = req->device;
 301         struct bio_and_error m;
 302         int rv;
 303
 304         /* __req_mod possibly frees req, do not touch req after that! */
 305         rv = __req_mod(req, what, &m);
 306         if (m.bio)
 307                 complete_master_bio(device, &m);
 308
 309         return rv;
 310 }
 311
 312 /* completion of master bio is outside of our spinlock.
 313  * We still may or may not be inside some irqs disabled section
 314  * of the lower level driver completion callback, so we need to
 315  * spin_lock_irqsave here. */
 316 static inline int req_mod(struct drbd_request *req,
 317                 enum drbd_req_event what)
 318 {
 319         unsigned long flags;
 320         struct drbd_device *device = req->device;
 321         struct bio_and_error m;
 322         int rv;
 323
 324         spin_lock_irqsave(&device->resource->req_lock, flags);
 325         rv = __req_mod(req, what, &m);
 326         spin_unlock_irqrestore(&device->resource->req_lock, flags);
 327
 328         if (m.bio)
 329                 complete_master_bio(device, &m);
 330
 331         return rv;
 332 }
 333
 334 static inline bool drbd_should_do_remote(union drbd_dev_state s)
 335 {
 336         return s.pdsk == D_UP_TO_DATE ||
 337                 (s.pdsk >= D_INCONSISTENT &&
 338                  s.conn >= C_WF_BITMAP_T &&
 339                  s.conn < C_AHEAD);
 340         /* Before proto 96 that was >= CONNECTED instead of >= C_WF_BITMAP_T.
 341            That is equivalent since before 96 IO was frozen in the C_WF_BITMAP*
 342            states. */
 343 }
 344 static inline bool drbd_should_send_out_of_sync(union drbd_dev_state s)
 345 {
 346         return s.conn == C_AHEAD || s.conn == C_WF_BITMAP_S;
 347         /* pdsk = D_INCONSISTENT as a consequence. Protocol 96 check not necessary
 348            since we enter state C_AHEAD only if proto >= 96 */
 349 }
 350
 351 #endif