/*
* Abort a request: free resources and complete the
* user request with the specified error
*/
int
abortrequest(struct request *rq, int error)
{
struct buf *bp = rq->bp; /* user buffer */
bp->b_flags |= B_ERROR;
bp->b_error = error;
freerq(rq); /* free everything we're doing */
biodone(bp);
return error; /* and give up */
}
/*
* Take a completed buffer, transfer the data back if
* it's a read, and complete the high-level request
* if this is the last subrequest.
*
* The bp parameter is in fact a struct rqelement, which
* includes a couple of extras at the end.
*/
void
complete_rqe(struct bio *bio)
{
union daemoninfo di;
struct buf *bp = bio->bio_buf;
struct rqelement *rqe;
struct request *rq;
struct rqgroup *rqg;
struct bio *ubio; /* user buffer */
struct drive *drive;
struct sd *sd;
char *gravity; /* for error messages */
get_mplock();
rqe = (struct rqelement *) bp; /* point to the element that completed */
rqg = rqe->rqg; /* and the request group */
rq = rqg->rq; /* and the complete request */
ubio = rq->bio; /* user buffer */
#ifdef VINUMDEBUG
if (debug & DEBUG_LASTREQS)
logrq(loginfo_iodone, (union rqinfou) rqe, ubio);
#endif
drive = &DRIVE[rqe->driveno];
drive->active--; /* one less outstanding I/O on this drive */
vinum_conf.active--; /* one less outstanding I/O globally */
if ((drive->active == (DRIVE_MAXACTIVE - 1)) /* we were at the drive limit */
||(vinum_conf.active == VINUM_MAXACTIVE)) /* or the global limit */
wakeup(&launch_requests); /* let another one at it */
if ((bp->b_flags & B_ERROR) != 0) { /* transfer in error */
gravity = "";
sd = &SD[rqe->sdno];
if (bp->b_error != 0) /* did it return a number? */
rq->error = bp->b_error; /* yes, put it in. */
else if (rq->error == 0) /* no: do we have one already? */
rq->error = EIO; /* no: catchall "I/O error" */
sd->lasterror = rq->error;
if (bp->b_cmd == BUF_CMD_READ) {
if ((rq->error == ENXIO) || (sd->flags & VF_RETRYERRORS) == 0) {
gravity = " fatal";
set_sd_state(rqe->sdno, sd_crashed, setstate_force); /* subdisk is crashed */
}
log(LOG_ERR,
"%s:%s read error, offset %lld for %d bytes\n",
gravity,
sd->name,
(long long)bio->bio_offset,
bp->b_bcount);
} else { /* write operation */
if ((rq->error == ENXIO) || (sd->flags & VF_RETRYERRORS) == 0) {
gravity = "fatal ";
set_sd_state(rqe->sdno, sd_stale, setstate_force); /* subdisk is stale */
}
log(LOG_ERR,
"%s:%s write error, offset %lld for %d bytes\n",
gravity,
sd->name,
(long long)bio->bio_offset,
bp->b_bcount);
}
log(LOG_ERR,
"%s: user buffer offset %lld for %d bytes\n",
sd->name,
(long long)ubio->bio_offset,
ubio->bio_buf->b_bcount);
if (rq->error == ENXIO) { /* the drive's down too */
log(LOG_ERR,
"%s: fatal drive I/O error, offset %lld for %d bytes\n",
DRIVE[rqe->driveno].label.name,
(long long)bio->bio_offset,
bp->b_bcount);
DRIVE[rqe->driveno].lasterror = rq->error;
set_drive_state(rqe->driveno, /* take the drive down */
drive_down,
setstate_force);
}
}
/* Now update the statistics */
if (bp->b_cmd == BUF_CMD_READ) { /* read operation */
DRIVE[rqe->driveno].reads++;
DRIVE[rqe->driveno].bytes_read += bp->b_bcount;
SD[rqe->sdno].reads++;
SD[rqe->sdno].bytes_read += bp->b_bcount;
PLEX[rqe->rqg->plexno].reads++;
PLEX[rqe->rqg->plexno].bytes_read += bp->b_bcount;
if (PLEX[rqe->rqg->plexno].volno >= 0) { /* volume I/O, not plex */
VOL[PLEX[rqe->rqg->plexno].volno].reads++;
VOL[PLEX[rqe->rqg->plexno].volno].bytes_read += bp->b_bcount;
}
} else { /* write operation */
DRIVE[rqe->driveno].writes++;
DRIVE[rqe->driveno].bytes_written += bp->b_bcount;
SD[rqe->sdno].writes++;
SD[rqe->sdno].bytes_written += bp->b_bcount;
PLEX[rqe->rqg->plexno].writes++;
PLEX[rqe->rqg->plexno].bytes_written += bp->b_bcount;
if (PLEX[rqe->rqg->plexno].volno >= 0) { /* volume I/O, not plex */
VOL[PLEX[rqe->rqg->plexno].volno].writes++;
VOL[PLEX[rqe->rqg->plexno].volno].bytes_written += bp->b_bcount;
}
}
if (rqg->flags & XFR_RECOVERY_READ) { /* recovery read, */
int *sdata; /* source */
int *data; /* and group data */
int length; /* and count involved */
int count; /* loop counter */
struct rqelement *urqe = &rqg->rqe[rqg->badsdno]; /* rqe of the bad subdisk */
/* XOR destination is the user data */
sdata = (int *) &rqe->b.b_data[rqe->groupoffset << DEV_BSHIFT]; /* old data contents */
data = (int *) &urqe->b.b_data[urqe->groupoffset << DEV_BSHIFT]; /* destination */
length = urqe->grouplen * (DEV_BSIZE / sizeof(int)); /* and number of ints */
for (count = 0; count < length; count++)
data[count] ^= sdata[count];
/*
* In a normal read, we will normally read directly
* into the user buffer. This doesn't work if
* we're also doing a recovery, so we have to
* copy it
*/
if (rqe->flags & XFR_NORMAL_READ) { /* normal read as well, */
char *src = &rqe->b.b_data[rqe->dataoffset << DEV_BSHIFT]; /* read data is here */
char *dst;
dst = (char *) ubio->bio_buf->b_data + (rqe->useroffset << DEV_BSHIFT); /* where to put it in user buffer */
length = rqe->datalen << DEV_BSHIFT; /* and count involved */
bcopy(src, dst, length); /* move it */
}
} else if ((rqg->flags & (XFR_NORMAL_WRITE | XFR_DEGRADED_WRITE)) /* RAID 4/5 group write operation */
&&(rqg->active == 1)) /* and this is the last active request */
complete_raid5_write(rqe);
/*
* This is the earliest place where we can be
* sure that the request has really finished,
* since complete_raid5_write can issue new
* requests.
*/
rqg->active--; /* this request now finished */
if (rqg->active == 0) { /* request group finished, */
rq->active--; /* one less */
if (rqg->lock) { /* got a lock? */
unlockrange(rqg->plexno, rqg->lock); /* yes, free it */
rqg->lock = 0;
}
}
if (rq->active == 0) { /* request finished, */
#ifdef VINUMDEBUG
if (debug & DEBUG_RESID) {
if (ubio->bio_buf->b_resid != 0) /* still something to transfer? */
Debugger("resid");
}
#endif
if (rq->error) { /* did we have an error? */
if (rq->isplex) { /* plex operation, */
ubio->bio_buf->b_flags |= B_ERROR; /* yes, propagate to user */
ubio->bio_buf->b_error = rq->error;
} else { /* try to recover */
di.rq = rq;
queue_daemon_request(daemonrq_ioerror, di); /* let the daemon complete */
}
} else {
ubio->bio_buf->b_resid = 0; /* completed our transfer */
if (rq->isplex == 0) /* volume request, */
VOL[rq->volplex.volno].active--; /* another request finished */
biodone(ubio); /* top level buffer completed */
freerq(rq); /* return the request storage */
}
}
rel_mplock();
}
/*
* Start a transfer. Return -1 on error,
* 0 if OK, 1 if we need to retry.
* Parameter reviveok is set when doing
* transfers for revives: it allows transfers to
* be started immediately when a revive is in
* progress. During revive, normal transfers
* are queued if they share address space with
* a currently active revive operation.
*/
int
vinumstart(struct buf *bp, int reviveok)
{
int plexno;
int maxplex; /* maximum number of plexes to handle */
struct volume *vol;
struct request *rq; /* build up our request here */
enum requeststatus status;
#if VINUMDEBUG
if (debug & DEBUG_LASTREQS)
logrq(loginfo_user_bp, (union rqinfou) bp, bp);
#endif
if ((bp->b_bcount % DEV_BSIZE) != 0) { /* bad length */
bp->b_error = EINVAL; /* invalid size */
bp->b_flags |= B_ERROR;
biodone(bp);
return -1;
}
rq = (struct request *) Malloc(sizeof(struct request)); /* allocate a request struct */
if (rq == NULL) { /* can't do it */
bp->b_error = ENOMEM; /* can't get memory */
bp->b_flags |= B_ERROR;
biodone(bp);
return -1;
}
bzero(rq, sizeof(struct request));
/*
* Note the volume ID. This can be NULL, which
* the request building functions use as an
* indication for single plex I/O
*/
rq->bp = bp; /* and the user buffer struct */
if (DEVTYPE(bp->b_dev) == VINUM_VOLUME_TYPE) { /* it's a volume, */
rq->volplex.volno = Volno(bp->b_dev); /* get the volume number */
vol = &VOL[rq->volplex.volno]; /* and point to it */
vol->active++; /* one more active request */
maxplex = vol->plexes; /* consider all its plexes */
} else {
vol = NULL; /* no volume */
rq->volplex.plexno = Plexno(bp->b_dev); /* point to the plex */
rq->isplex = 1; /* note that it's a plex */
maxplex = 1; /* just the one plex */
}
if (bp->b_flags & B_READ) {
/*
* This is a read request. Decide
* which plex to read from.
*
* There's a potential race condition here,
* since we're not locked, and we could end
* up multiply incrementing the round-robin
* counter. This doesn't have any serious
* effects, however.
*/
if (vol != NULL) {
vol->reads++;
plexno = vol->preferred_plex; /* get the plex to use */
if (plexno < 0) { /* round robin */
plexno = vol->last_plex_read;
vol->last_plex_read++;
if (vol->last_plex_read >= vol->plexes) /* got the the end? */
vol->last_plex_read = 0; /* wrap around */
}
status = build_read_request(rq, plexno); /* build a request */
} else {
daddr_t diskaddr = bp->b_blkno; /* start offset of transfer */
status = bre(rq, /* build a request list */
rq->volplex.plexno,
&diskaddr,
diskaddr + (bp->b_bcount / DEV_BSIZE));
}
if ((status > REQUEST_RECOVERED) /* can't satisfy it */
||(bp->b_flags & B_DONE)) { /* XXX shouldn't get this without bad status */
if (status == REQUEST_DOWN) { /* not enough subdisks */
bp->b_error = EIO; /* I/O error */
bp->b_flags |= B_ERROR;
}
biodone(bp);
freerq(rq);
return -1;
}
return launch_requests(rq, reviveok); /* now start the requests if we can */
} else
/*
* This is a write operation. We write to all plexes. If this is
* a RAID-4 or RAID-5 plex, we must also update the parity stripe.
*/
{
if (vol != NULL) {
vol->writes++;
status = build_write_request(rq); /* Not all the subdisks are up */
} else { /* plex I/O */
daddr_t diskstart;
diskstart = bp->b_blkno; /* start offset of transfer */
status = bre(rq,
Plexno(bp->b_dev),
&diskstart,
bp->b_blkno + (bp->b_bcount / DEV_BSIZE)); /* build requests for the plex */
}
if ((status > REQUEST_RECOVERED) /* can't satisfy it */
||(bp->b_flags & B_DONE)) { /* XXX shouldn't get this without bad status */
if (status == REQUEST_DOWN) { /* not enough subdisks */
bp->b_error = EIO; /* I/O error */
bp->b_flags |= B_ERROR;
}
if ((bp->b_flags & B_DONE) == 0)
biodone(bp);
freerq(rq);
return -1;
}
return launch_requests(rq, reviveok); /* now start the requests if we can */
}
}