sys/dev/vinum/vinuminterrupt.c

   1 /* vinuminterrupt.c: bottom half of the driver */
   2
   3 /*-
   4  * Copyright (c) 1997, 1998, 1999
   5  *      Nan Yang Computer Services Limited.  All rights reserved.
   6  *
   7  *  Parts copyright (c) 1997, 1998 Cybernet Corporation, NetMAX project.
   8  *
   9  *  Written by Greg Lehey
  10  *
  11  *  This software is distributed under the so-called ``Berkeley
  12  *  License'':
  13  *
  14  * Redistribution and use in source and binary forms, with or without
  15  * modification, are permitted provided that the following conditions
  16  * are met:
  17  * 1. Redistributions of source code must retain the above copyright
  18  *    notice, this list of conditions and the following disclaimer.
  19  * 2. Redistributions in binary form must reproduce the above copyright
  20  *    notice, this list of conditions and the following disclaimer in the
  21  *    documentation and/or other materials provided with the distribution.
  22  * 3. All advertising materials mentioning features or use of this software
  23  *    must display the following acknowledgement:
  24  *      This product includes software developed by Nan Yang Computer
  25  *      Services Limited.
  26  * 4. Neither the name of the Company nor the names of its contributors
  27  *    may be used to endorse or promote products derived from this software
  28  *    without specific prior written permission.
  29  *
  30  * This software is provided ``as is'', and any express or implied
  31  * warranties, including, but not limited to, the implied warranties of
  32  * merchantability and fitness for a particular purpose are disclaimed.
  33  * In no event shall the company or contributors be liable for any
  34  * direct, indirect, incidental, special, exemplary, or consequential
  35  * damages (including, but not limited to, procurement of substitute
  36  * goods or services; loss of use, data, or profits; or business
  37  * interruption) however caused and on any theory of liability, whether
  38  * in contract, strict liability, or tort (including negligence or
  39  * otherwise) arising in any way out of the use of this software, even if
  40  * advised of the possibility of such damage.
  41  *
  42  * $Id: vinuminterrupt.c,v 1.14 2001/05/23 23:03:37 grog Exp grog $
  43  * $FreeBSD$
  44  */
  45
  46 #include <dev/vinum/vinumhdr.h>
  47 #include <dev/vinum/request.h>
  48 #include <sys/resourcevar.h>
  49
  50 void complete_raid5_write(struct rqelement *);
  51 void complete_rqe(struct buf *bp);
  52 void sdio_done(struct buf *bp);
  53
  54 /*
  55  * Take a completed buffer, transfer the data back if
  56  * it's a read, and complete the high-level request
  57  * if this is the last subrequest.
  58  *
  59  * The bp parameter is in fact a struct rqelement, which
  60  * includes a couple of extras at the end.
  61  */
  62 void
  63 complete_rqe(struct buf *bp)
  64 {
  65     struct rqelement *rqe;
  66     struct request *rq;
  67     struct rqgroup *rqg;
  68     struct buf *ubp;                                        /* user buffer */
  69     struct drive *drive;
  70     struct sd *sd;
  71     char *gravity;                                          /* for error messages */
  72
  73     rqe = (struct rqelement *) bp;                          /* point to the element element that completed */
  74     rqg = rqe->rqg;                                         /* and the request group */
  75     rq = rqg->rq;                                           /* and the complete request */
  76     ubp = rq->bp;                                           /* user buffer */
  77
  78 #ifdef VINUMDEBUG
  79     if (debug & DEBUG_LASTREQS)
  80         logrq(loginfo_iodone, (union rqinfou) rqe, ubp);
  81 #endif
  82     drive = &DRIVE[rqe->driveno];
  83     drive->active--;                                        /* one less outstanding I/O on this drive */
  84     vinum_conf.active--;                                    /* one less outstanding I/O globally */
  85     if ((drive->active == (DRIVE_MAXACTIVE - 1))            /* we were at the drive limit */
  86     ||(vinum_conf.active == VINUM_MAXACTIVE))               /* or the global limit */
  87         wakeup(&launch_requests);                           /* let another one at it */
  88     if ((bp->b_io.bio_flags & BIO_ERROR) != 0) {            /* transfer in error */
  89         gravity = "";
  90         sd = &SD[rqe->sdno];
  91
  92         if (bp->b_error != 0)                               /* did it return a number? */
  93             rq->error = bp->b_error;                        /* yes, put it in. */
  94         else if (rq->error == 0)                            /* no: do we have one already? */
  95             rq->error = EIO;                                /* no: catchall "I/O error" */
  96         sd->lasterror = rq->error;
  97         if (bp->b_iocmd == BIO_READ) {                      /* read operation */
  98             if ((rq->error == ENXIO) || (sd->flags & VF_RETRYERRORS) == 0) {
  99                 gravity = " fatal";
 100                 set_sd_state(rqe->sdno, sd_crashed, setstate_force); /* subdisk is crashed */
 101             }
 102             log(LOG_ERR,
 103                 "%s:%s read error, block %lld for %ld bytes\n",
 104                 gravity,
 105                 sd->name,
 106                 bp->b_blkno,
 107                 bp->b_bcount);
 108         } else {                                            /* write operation */
 109             if ((rq->error == ENXIO) || (sd->flags & VF_RETRYERRORS) == 0) {
 110                 gravity = "fatal ";
 111                 set_sd_state(rqe->sdno, sd_stale, setstate_force); /* subdisk is stale */
 112             }
 113             log(LOG_ERR,
 114                 "%s:%s write error, block %lld for %ld bytes\n",
 115                 gravity,
 116                 sd->name,
 117                 bp->b_blkno,
 118                 bp->b_bcount);
 119         }
 120         log(LOG_ERR,
 121             "%s: user buffer block %lld for %ld bytes\n",
 122             sd->name,
 123             ubp->b_blkno,
 124             ubp->b_bcount);
 125         if (rq->error == ENXIO) {                           /* the drive's down too */
 126             log(LOG_ERR,
 127                 "%s: fatal drive I/O error, block %lld for %ld bytes\n",
 128                 DRIVE[rqe->driveno].label.name,
 129                 bp->b_blkno,
 130                 bp->b_bcount);
 131             DRIVE[rqe->driveno].lasterror = rq->error;
 132             set_drive_state(rqe->driveno,                   /* take the drive down */
 133                 drive_down,
 134                 setstate_force);
 135         }
 136     }
 137     /* Now update the statistics */
 138     if (bp->b_iocmd == BIO_READ) {                          /* read operation */
 139         DRIVE[rqe->driveno].reads++;
 140         DRIVE[rqe->driveno].bytes_read += bp->b_bcount;
 141         SD[rqe->sdno].reads++;
 142         SD[rqe->sdno].bytes_read += bp->b_bcount;
 143         PLEX[rqe->rqg->plexno].reads++;
 144         PLEX[rqe->rqg->plexno].bytes_read += bp->b_bcount;
 145         if (PLEX[rqe->rqg->plexno].volno >= 0) {            /* volume I/O, not plex */
 146             VOL[PLEX[rqe->rqg->plexno].volno].reads++;
 147             VOL[PLEX[rqe->rqg->plexno].volno].bytes_read += bp->b_bcount;
 148         }
 149     } else {                                                /* write operation */
 150         DRIVE[rqe->driveno].writes++;
 151         DRIVE[rqe->driveno].bytes_written += bp->b_bcount;
 152         SD[rqe->sdno].writes++;
 153         SD[rqe->sdno].bytes_written += bp->b_bcount;
 154         PLEX[rqe->rqg->plexno].writes++;
 155         PLEX[rqe->rqg->plexno].bytes_written += bp->b_bcount;
 156         if (PLEX[rqe->rqg->plexno].volno >= 0) {            /* volume I/O, not plex */
 157             VOL[PLEX[rqe->rqg->plexno].volno].writes++;
 158             VOL[PLEX[rqe->rqg->plexno].volno].bytes_written += bp->b_bcount;
 159         }
 160     }
 161     if (rqg->flags & XFR_RECOVERY_READ) {                   /* recovery read, */
 162         int *sdata;                                         /* source */
 163         int *data;                                          /* and group data */
 164         int length;                                         /* and count involved */
 165         int count;                                          /* loop counter */
 166         struct rqelement *urqe = &rqg->rqe[rqg->badsdno];   /* rqe of the bad subdisk */
 167
 168         /* XOR destination is the user data */
 169         sdata = (int *) &rqe->b.b_data[rqe->groupoffset << DEV_BSHIFT]; /* old data contents */
 170         data = (int *) &urqe->b.b_data[urqe->groupoffset << DEV_BSHIFT]; /* destination */
 171         length = urqe->grouplen * (DEV_BSIZE / sizeof(int)); /* and number of ints */
 172
 173         for (count = 0; count < length; count++)
 174             data[count] ^= sdata[count];
 175
 176         /*
 177          * In a normal read, we will normally read directly
 178          * into the user buffer.  This doesn't work if
 179          * we're also doing a recovery, so we have to
 180          * copy it
 181          */
 182         if (rqe->flags & XFR_NORMAL_READ) {                 /* normal read as well, */
 183             char *src = &rqe->b.b_data[rqe->dataoffset << DEV_BSHIFT]; /* read data is here */
 184             char *dst;
 185
 186             dst = (char *) ubp->b_data + (rqe->useroffset << DEV_BSHIFT); /* where to put it in user buffer */
 187             length = rqe->datalen << DEV_BSHIFT;            /* and count involved */
 188             bcopy(src, dst, length);                        /* move it */
 189         }
 190     } else if ((rqg->flags & (XFR_NORMAL_WRITE | XFR_DEGRADED_WRITE)) /* RAID 4/5 group write operation  */
 191     &&(rqg->active == 1))                                   /* and this is the last active request */
 192         complete_raid5_write(rqe);
 193     /*
 194      * This is the earliest place where we can be
 195      * sure that the request has really finished,
 196      * since complete_raid5_write can issue new
 197      * requests.
 198      */
 199     rqg->active--;                                          /* this request now finished */
 200     if (rqg->active == 0) {                                 /* request group finished, */
 201         rq->active--;                                       /* one less */
 202         if (rqg->lock) {                                    /* got a lock? */
 203             unlockrange(rqg->plexno, rqg->lock);            /* yes, free it */
 204             rqg->lock = 0;
 205         }
 206     }
 207     if (rq->active == 0) {                                  /* request finished, */
 208 #ifdef VINUMDEBUG
 209         if (debug & DEBUG_RESID) {
 210             if (ubp->b_resid != 0)                          /* still something to transfer? */
 211                 Debugger("resid");
 212         }
 213 #endif
 214
 215         if (rq->error) {                                    /* did we have an error? */
 216             if (rq->isplex) {                               /* plex operation, */
 217                 ubp->b_io.bio_flags |= BIO_ERROR;           /* yes, propagate to user */
 218                 ubp->b_error = rq->error;
 219             } else                                          /* try to recover */
 220                 queue_daemon_request(daemonrq_ioerror, (union daemoninfo) rq); /* let the daemon complete */
 221         } else {
 222             ubp->b_resid = 0;                               /* completed our transfer */
 223             if (rq->isplex == 0)                            /* volume request, */
 224                 VOL[rq->volplex.volno].active--;            /* another request finished */
 225             if (rq->flags & XFR_COPYBUF) {
 226                 Free(ubp->b_data);
 227                 ubp->b_data = rq->save_data;
 228             }
 229             bufdone(ubp);                                   /* top level buffer completed */
 230             freerq(rq);                                     /* return the request storage */
 231         }
 232     }
 233 }
 234
 235 /* Free a request block and anything hanging off it */
 236 void
 237 freerq(struct request *rq)
 238 {
 239     struct rqgroup *rqg;
 240     struct rqgroup *nrqg;                                   /* next in chain */
 241     int rqno;
 242
 243     for (rqg = rq->rqg; rqg != NULL; rqg = nrqg) {          /* through the whole request chain */
 244         if (rqg->lock)                                      /* got a lock? */
 245             unlockrange(rqg->plexno, rqg->lock);            /* yes, free it */
 246         for (rqno = 0; rqno < rqg->count; rqno++) {
 247             if ((rqg->rqe[rqno].flags & XFR_MALLOCED)       /* data buffer was malloced, */
 248             &&rqg->rqe[rqno].b.b_data)                      /* and the allocation succeeded */
 249                 Free(rqg->rqe[rqno].b.b_data);              /* free it */
 250             if (rqg->rqe[rqno].flags & XFR_BUFLOCKED) {     /* locked this buffer, */
 251                 BUF_UNLOCK(&rqg->rqe[rqno].b);              /* unlock it again */
 252                 BUF_LOCKFREE(&rqg->rqe[rqno].b);
 253             }
 254         }
 255         nrqg = rqg->next;                                   /* note the next one */
 256         Free(rqg);                                          /* and free this one */
 257     }
 258     Free(rq);                                               /* free the request itself */
 259 }
 260
 261 /* I/O on subdisk completed */
 262 void
 263 sdio_done(struct buf *bp)
 264 {
 265     struct sdbuf *sbp;
 266
 267     sbp = (struct sdbuf *) bp;
 268     if (sbp->b.b_io.bio_flags & BIO_ERROR) {                /* had an error */
 269         sbp->bp->b_io.bio_flags |= BIO_ERROR;               /* propagate upwards */
 270         sbp->bp->b_error = sbp->b.b_error;
 271     }
 272 #ifdef VINUMDEBUG
 273     if (debug & DEBUG_LASTREQS)
 274         logrq(loginfo_sdiodone, (union rqinfou) bp, bp);
 275 #endif
 276     sbp->bp->b_resid = sbp->b.b_resid;                      /* copy the resid field */
 277     /* Now update the statistics */
 278     if (bp->b_iocmd == BIO_READ) {                          /* read operation */
 279         DRIVE[sbp->driveno].reads++;
 280         DRIVE[sbp->driveno].bytes_read += sbp->b.b_bcount;
 281         SD[sbp->sdno].reads++;
 282         SD[sbp->sdno].bytes_read += sbp->b.b_bcount;
 283     } else {                                                /* write operation */
 284         DRIVE[sbp->driveno].writes++;
 285         DRIVE[sbp->driveno].bytes_written += sbp->b.b_bcount;
 286         SD[sbp->sdno].writes++;
 287         SD[sbp->sdno].bytes_written += sbp->b.b_bcount;
 288     }
 289     bufdone(sbp->bp);                                       /* complete the caller's I/O */
 290     BUF_UNLOCK(&sbp->b);
 291     BUF_LOCKFREE(&sbp->b);
 292     Free(sbp);
 293 }
 294
 295 /* Start the second phase of a RAID-4 or RAID-5 group write operation. */
 296 void
 297 complete_raid5_write(struct rqelement *rqe)
 298 {
 299     int *sdata;                                             /* source */
 300     int *pdata;                                             /* and parity block data */
 301     int length;                                             /* and count involved */
 302     int count;                                              /* loop counter */
 303     int rqno;                                               /* request index */
 304     int rqoffset;                                           /* offset of request data from parity data */
 305     struct buf *ubp;                                        /* user buffer header */
 306     struct request *rq;                                     /* pointer to our request */
 307     struct rqgroup *rqg;                                    /* and to the request group */
 308     struct rqelement *prqe;                                 /* point to the parity block */
 309     struct drive *drive;                                    /* drive to access */
 310
 311     rqg = rqe->rqg;                                         /* and to our request group */
 312     rq = rqg->rq;                                           /* point to our request */
 313     ubp = rq->bp;                                           /* user's buffer header */
 314     prqe = &rqg->rqe[0];                                    /* point to the parity block */
 315
 316     /*
 317      * If we get to this function, we have normal or
 318      * degraded writes, or a combination of both.  We do
 319      * the same thing in each case: we perform an
 320      * exclusive or to the parity block.  The only
 321      * difference is the origin of the data and the
 322      * address range.
 323      */
 324     if (rqe->flags & XFR_DEGRADED_WRITE) {                  /* do the degraded write stuff */
 325         pdata = (int *) (&prqe->b.b_data[(prqe->groupoffset) << DEV_BSHIFT]); /* parity data pointer */
 326         bzero(pdata, prqe->grouplen << DEV_BSHIFT);         /* start with nothing in the parity block */
 327
 328         /* Now get what data we need from each block */
 329         for (rqno = 1; rqno < rqg->count; rqno++) {         /* for all the data blocks */
 330             rqe = &rqg->rqe[rqno];                          /* this request */
 331             sdata = (int *) (&rqe->b.b_data[rqe->groupoffset << DEV_BSHIFT]); /* old data */
 332             length = rqe->grouplen << (DEV_BSHIFT - 2);     /* and count involved */
 333
 334             /*
 335              * Add the data block to the parity block.  Before
 336              * we started the request, we zeroed the parity
 337              * block, so the result of adding all the other
 338              * blocks and the block we want to write will be
 339              * the correct parity block.
 340              */
 341             for (count = 0; count < length; count++)
 342                 pdata[count] ^= sdata[count];
 343             if ((rqe->flags & XFR_MALLOCED)                 /* the buffer was malloced, */
 344             &&((rqg->flags & XFR_NORMAL_WRITE) == 0)) {     /* and we have no normal write, */
 345                 Free(rqe->b.b_data);                        /* free it now */
 346                 rqe->flags &= ~XFR_MALLOCED;
 347             }
 348         }
 349     }
 350     if (rqg->flags & XFR_NORMAL_WRITE) {                    /* do normal write stuff */
 351         /* Get what data we need from each block */
 352         for (rqno = 1; rqno < rqg->count; rqno++) {         /* for all the data blocks */
 353             rqe = &rqg->rqe[rqno];                          /* this request */
 354             if ((rqe->flags & (XFR_DATA_BLOCK | XFR_BAD_SUBDISK | XFR_NORMAL_WRITE))
 355                 == (XFR_DATA_BLOCK | XFR_NORMAL_WRITE)) {   /* good data block to write */
 356                 sdata = (int *) &rqe->b.b_data[rqe->dataoffset << DEV_BSHIFT]; /* old data contents */
 357                 rqoffset = rqe->dataoffset + rqe->sdoffset - prqe->sdoffset; /* corresponding parity block offset */
 358                 pdata = (int *) (&prqe->b.b_data[rqoffset << DEV_BSHIFT]); /* parity data pointer */
 359                 length = rqe->datalen * (DEV_BSIZE / sizeof(int)); /* and number of ints */
 360
 361                 /*
 362                  * "remove" the old data block
 363                  * from the parity block
 364                  */
 365                 if ((pdata < ((int *) prqe->b.b_data))
 366                     || (&pdata[length] > ((int *) (prqe->b.b_data + prqe->b.b_bcount)))
 367                     || (sdata < ((int *) rqe->b.b_data))
 368                     || (&sdata[length] > ((int *) (rqe->b.b_data + rqe->b.b_bcount))))
 369                     panic("complete_raid5_write: bounds overflow");
 370                 for (count = 0; count < length; count++)
 371                     pdata[count] ^= sdata[count];
 372
 373                 /* "add" the new data block */
 374                 sdata = (int *) (&ubp->b_data[rqe->useroffset << DEV_BSHIFT]); /* new data */
 375                 if ((sdata < ((int *) ubp->b_data))
 376                     || (&sdata[length] > ((int *) (ubp->b_data + ubp->b_bcount))))
 377                     panic("complete_raid5_write: bounds overflow");
 378                 for (count = 0; count < length; count++)
 379                     pdata[count] ^= sdata[count];
 380
 381                 /* Free the malloced buffer */
 382                 if (rqe->flags & XFR_MALLOCED) {            /* the buffer was malloced, */
 383                     Free(rqe->b.b_data);                    /* free it */
 384                     rqe->flags &= ~XFR_MALLOCED;
 385                 } else
 386                     panic("complete_raid5_write: malloc conflict");
 387
 388                 if ((rqe->b.b_iocmd == BIO_READ)            /* this was a read */
 389                 &&((rqe->flags & XFR_BAD_SUBDISK) == 0)) {  /* and we can write this block */
 390                     rqe->b.b_flags &= ~B_DONE;              /* start a new request */
 391                     rqe->b.b_iocmd = BIO_WRITE;             /* we're writing now */
 392                     rqe->b.b_iodone = complete_rqe;         /* call us here when done */
 393                     rqe->flags &= ~XFR_PARITYOP;            /* reset flags that brought us here */
 394                     rqe->b.b_data = &ubp->b_data[rqe->useroffset << DEV_BSHIFT]; /* point to the user data */
 395                     rqe->b.b_bcount = rqe->datalen << DEV_BSHIFT; /* length to write */
 396                     rqe->b.b_bufsize = rqe->b.b_bcount;     /* don't claim more */
 397                     rqe->b.b_resid = rqe->b.b_bcount;       /* nothing transferred */
 398                     rqe->b.b_blkno += rqe->dataoffset;      /* point to the correct block */
 399                     rqg->active++;                          /* another active request */
 400                     drive = &DRIVE[rqe->driveno];           /* drive to access */
 401
 402                                                             /* We can't sleep here, so we just increment the counters. */
 403                     drive->active++;
 404                     if (drive->active >= drive->maxactive)
 405                         drive->maxactive = drive->active;
 406                     vinum_conf.active++;
 407                     if (vinum_conf.active >= vinum_conf.maxactive)
 408                         vinum_conf.maxactive = vinum_conf.active;
 409 #ifdef VINUMDEBUG
 410                     if (debug & DEBUG_ADDRESSES)
 411                         log(LOG_DEBUG,
 412                             "  %s dev %d.%d, sd %d, offset 0x%x, devoffset 0x%llx, length %ld\n",
 413                             rqe->b.b_iocmd == BIO_READ ? "Read" : "Write",
 414                             major(rqe->b.b_dev),
 415                             minor(rqe->b.b_dev),
 416                             rqe->sdno,
 417                             (u_int) (rqe->b.b_blkno - SD[rqe->sdno].driveoffset),
 418                             rqe->b.b_blkno,
 419                             rqe->b.b_bcount);
 420                     if (debug & DEBUG_LASTREQS)
 421                         logrq(loginfo_raid5_data, (union rqinfou) rqe, ubp);
 422 #endif
 423                     DEV_STRATEGY(&rqe->b, 0);
 424                 }
 425             }
 426         }
 427     }
 428     /* Finally, write the parity block */
 429     rqe = &rqg->rqe[0];
 430     rqe->b.b_flags &= ~B_DONE;                              /* we're not done */
 431     rqe->b.b_iocmd = BIO_WRITE;                             /* we're writing now */
 432     rqe->b.b_iodone = complete_rqe;                         /* call us here when done */
 433     rqg->flags &= ~XFR_PARITYOP;                            /* reset flags that brought us here */
 434     rqe->b.b_bcount = rqe->buflen << DEV_BSHIFT;            /* length to write */
 435     rqe->b.b_bufsize = rqe->b.b_bcount;                     /* don't claim we have more */
 436     rqe->b.b_resid = rqe->b.b_bcount;                       /* nothing transferred */
 437     rqg->active++;                                          /* another active request */
 438     drive = &DRIVE[rqe->driveno];                           /* drive to access */
 439
 440     /* We can't sleep here, so we just increment the counters. */
 441     drive->active++;
 442     if (drive->active >= drive->maxactive)
 443         drive->maxactive = drive->active;
 444     vinum_conf.active++;
 445     if (vinum_conf.active >= vinum_conf.maxactive)
 446         vinum_conf.maxactive = vinum_conf.active;
 447
 448 #ifdef VINUMDEBUG
 449     if (debug & DEBUG_ADDRESSES)
 450         log(LOG_DEBUG,
 451            "  %s dev %d.%d, sd %d, offset 0x%x, devoffset 0x%llx, length %ld\n",
 452             rqe->b.b_iocmd == BIO_READ ? "Read" : "Write",
 453             major(rqe->b.b_dev),
 454             minor(rqe->b.b_dev),
 455             rqe->sdno,
 456             (u_int) (rqe->b.b_blkno - SD[rqe->sdno].driveoffset),
 457             rqe->b.b_blkno,
 458             rqe->b.b_bcount);
 459     if (debug & DEBUG_LASTREQS)
 460         logrq(loginfo_raid5_parity, (union rqinfou) rqe, ubp);
 461 #endif
 462     DEV_STRATEGY(&rqe->b, 0);
 463 }
 464
 465 /* Local Variables: */
 466 /* fill-column: 50 */
 467 /* End: */