/*
* Recover a failed I/O operation.
*
* The correct way to do this is to examine the request and determine
* how to recover each individual failure. In the case of a write,
* this could be as simple as doing nothing: the defective drives may
* already be down, and there may be nothing else to do. In case of
* a read, it will be necessary to retry if there are alternative
* copies of the data.
*
* The easy way (here) is just to reissue the request. This will take
* a little longer, but nothing like as long as the failure will have
* taken.
*
*/
void
recover_io(struct request *rq)
{
/*
* This should read:
*
* vinumstrategy(rq->bp);
*
* Negotiate with phk to get it fixed.
*/
BUF_STRATEGY(rq->bp, 0); /* reissue the command */
}
int
physio(dev_t dev, struct uio *uio, int ioflag)
{
int i;
int error;
int spl;
caddr_t sa;
off_t blockno;
u_int iolen;
struct buf *bp;
/* Keep the process UPAGES from being swapped. XXX: why ? */
PHOLD(curproc);
bp = getpbuf(NULL);
sa = bp->b_data;
error = bp->b_error = 0;
/* XXX: sanity check */
if(dev->si_iosize_max < PAGE_SIZE) {
printf("WARNING: %s si_iosize_max=%d, using DFLTPHYS.\n",
devtoname(dev), dev->si_iosize_max);
dev->si_iosize_max = DFLTPHYS;
}
for (i = 0; i < uio->uio_iovcnt; i++) {
while (uio->uio_iov[i].iov_len) {
if (uio->uio_rw == UIO_READ)
bp->b_flags = B_PHYS | B_CALL | B_READ;
else
bp->b_flags = B_PHYS | B_CALL | B_WRITE;
bp->b_dev = dev;
bp->b_iodone = physwakeup;
bp->b_data = uio->uio_iov[i].iov_base;
bp->b_bcount = uio->uio_iov[i].iov_len;
bp->b_offset = uio->uio_offset;
bp->b_saveaddr = sa;
/* Don't exceed drivers iosize limit */
if (bp->b_bcount > dev->si_iosize_max)
bp->b_bcount = dev->si_iosize_max;
/*
* Make sure the pbuf can map the request
* XXX: The pbuf has kvasize = MAXPHYS so a request
* XXX: larger than MAXPHYS - PAGE_SIZE must be
* XXX: page aligned or it will be fragmented.
*/
iolen = ((vm_offset_t) bp->b_data) & PAGE_MASK;
if ((bp->b_bcount + iolen) > bp->b_kvasize) {
bp->b_bcount = bp->b_kvasize;
if (iolen != 0)
bp->b_bcount -= PAGE_SIZE;
}
bp->b_bufsize = bp->b_bcount;
blockno = bp->b_offset >> DEV_BSHIFT;
if ((daddr_t)blockno != blockno) {
error = EINVAL; /* blockno overflow */
goto doerror;
}
bp->b_blkno = blockno;
if (uio->uio_segflg == UIO_USERSPACE) {
if (!useracc(bp->b_data, bp->b_bufsize,
bp->b_flags & B_READ ?
VM_PROT_WRITE : VM_PROT_READ)) {
error = EFAULT;
goto doerror;
}
vmapbuf(bp);
}
BUF_STRATEGY(bp, 0);
spl = splbio();
while ((bp->b_flags & B_DONE) == 0)
tsleep((caddr_t)bp, PRIBIO, "physstr", 0);
splx(spl);
if (uio->uio_segflg == UIO_USERSPACE)
vunmapbuf(bp);
iolen = bp->b_bcount - bp->b_resid;
if (iolen == 0 && !(bp->b_flags & B_ERROR))
goto doerror; /* EOF */
uio->uio_iov[i].iov_len -= iolen;
uio->uio_iov[i].iov_base += iolen;
uio->uio_resid -= iolen;
uio->uio_offset += iolen;
if( bp->b_flags & B_ERROR) {
error = bp->b_error;
goto doerror;
}
}
}
doerror:
relpbuf(bp, NULL);
PRELE(curproc);
return (error);
}
/*
* Call the low-level strategy routines to
* perform the requests in a struct request
*/
int
launch_requests(struct request *rq, int reviveok)
{
int s;
struct rqgroup *rqg;
int rqno; /* loop index */
struct rqelement *rqe; /* current element */
struct drive *drive;
int rcount; /* request count */
/*
* First find out whether we're reviving, and the
* request contains a conflict. If so, we hang
* the request off plex->waitlist of the first
* plex we find which is reviving
*/
if ((rq->flags & XFR_REVIVECONFLICT) /* possible revive conflict */
&&(!reviveok)) { /* and we don't want to do it now, */
struct sd *sd;
struct request *waitlist; /* point to the waitlist */
sd = &SD[rq->sdno];
if (sd->waitlist != NULL) { /* something there already, */
waitlist = sd->waitlist;
while (waitlist->next != NULL) /* find the end */
waitlist = waitlist->next;
waitlist->next = rq; /* hook our request there */
} else
sd->waitlist = rq; /* hook our request at the front */
#if VINUMDEBUG
if (debug & DEBUG_REVIVECONFLICT)
log(LOG_DEBUG,
"Revive conflict sd %d: %p\n%s dev %d.%d, offset 0x%x, length %ld\n",
rq->sdno,
rq,
rq->bp->b_flags & B_READ ? "Read" : "Write",
major(rq->bp->b_dev),
minor(rq->bp->b_dev),
rq->bp->b_blkno,
rq->bp->b_bcount);
#endif
return 0; /* and get out of here */
}
rq->active = 0; /* nothing yet */
#if VINUMDEBUG
if (debug & DEBUG_ADDRESSES)
log(LOG_DEBUG,
"Request: %p\n%s dev %d.%d, offset 0x%x, length %ld\n",
rq,
rq->bp->b_flags & B_READ ? "Read" : "Write",
major(rq->bp->b_dev),
minor(rq->bp->b_dev),
rq->bp->b_blkno,
rq->bp->b_bcount);
vinum_conf.lastrq = rq;
vinum_conf.lastbuf = rq->bp;
if (debug & DEBUG_LASTREQS)
logrq(loginfo_user_bpl, (union rqinfou) rq->bp, rq->bp);
#endif
/*
* With the division of labour below (first count the requests, then
* issue them), it's possible that we don't need this splbio()
* protection. But I'll try that some other time.
*/
s = splbio();
for (rqg = rq->rqg; rqg != NULL; rqg = rqg->next) { /* through the whole request chain */
rqg->active = rqg->count; /* they're all active */
for (rqno = 0; rqno < rqg->count; rqno++) {
rqe = &rqg->rqe[rqno];
if (rqe->flags & XFR_BAD_SUBDISK) /* this subdisk is bad, */
rqg->active--; /* one less active request */
}
if (rqg->active) /* we have at least one active request, */
rq->active++; /* one more active request group */
}
/* Now fire off the requests */
for (rqg = rq->rqg; rqg != NULL;) { /* through the whole request chain */
if (rqg->lockbase >= 0) /* this rqg needs a lock first */
rqg->lock = lockrange(rqg->lockbase, rqg->rq->bp, &PLEX[rqg->plexno]);
rcount = rqg->count;
for (rqno = 0; rqno < rcount;) {
rqe = &rqg->rqe[rqno];
/*
* Point to next rqg before the bottom end
* changes the structures.
*/
if (++rqno >= rcount)
rqg = rqg->next;
if ((rqe->flags & XFR_BAD_SUBDISK) == 0) { /* this subdisk is good, */
drive = &DRIVE[rqe->driveno]; /* look at drive */
drive->active++;
if (drive->active >= drive->maxactive)
drive->maxactive = drive->active;
vinum_conf.active++;
if (vinum_conf.active >= vinum_conf.maxactive)
vinum_conf.maxactive = vinum_conf.active;
#if VINUMDEBUG
if (debug & DEBUG_ADDRESSES)
log(LOG_DEBUG,
" %s dev %d.%d, sd %d, offset 0x%x, devoffset 0x%x, length %ld\n",
rqe->b.b_flags & B_READ ? "Read" : "Write",
major(rqe->b.b_dev),
minor(rqe->b.b_dev),
rqe->sdno,
(u_int) (rqe->b.b_blkno - SD[rqe->sdno].driveoffset),
rqe->b.b_blkno,
rqe->b.b_bcount);
if (debug & DEBUG_LASTREQS)
logrq(loginfo_rqe, (union rqinfou) rqe, rq->bp);
#endif
/* fire off the request */
BUF_STRATEGY(&rqe->b, 0);
}
}
}
splx(s);
return 0;
}
/* Perform I/O on a subdisk */
void
sdio(struct buf *bp)
{
int s; /* spl */
struct sd *sd;
struct sdbuf *sbp;
daddr_t endoffset;
struct drive *drive;
#if VINUMDEBUG
if (debug & DEBUG_LASTREQS)
logrq(loginfo_sdio, (union rqinfou) bp, bp);
#endif
sd = &SD[Sdno(bp->b_dev)]; /* point to the subdisk */
drive = &DRIVE[sd->driveno];
if (drive->state != drive_up) {
if (sd->state >= sd_crashed) {
if (bp->b_flags & B_READ) /* reading, */
set_sd_state(sd->sdno, sd_crashed, setstate_force);
else
set_sd_state(sd->sdno, sd_stale, setstate_force);
}
bp->b_flags |= B_ERROR;
bp->b_error = EIO;
biodone(bp);
return;
}
/*
* We allow access to any kind of subdisk as long as we can expect
* to get the I/O performed.
*/
if (sd->state < sd_empty) { /* nothing to talk to, */
bp->b_flags |= B_ERROR;
bp->b_error = EIO;
biodone(bp);
return;
}
/* Get a buffer */
sbp = (struct sdbuf *) Malloc(sizeof(struct sdbuf));
if (sbp == NULL) {
bp->b_flags |= B_ERROR;
bp->b_error = ENOMEM;
biodone(bp);
return;
}
bzero(sbp, sizeof(struct sdbuf)); /* start with nothing */
sbp->b.b_flags = bp->b_flags | B_CALL; /* inform us when it's done */
sbp->b.b_bufsize = bp->b_bufsize; /* buffer size */
sbp->b.b_bcount = bp->b_bcount; /* number of bytes to transfer */
sbp->b.b_resid = bp->b_resid; /* and amount waiting */
sbp->b.b_dev = DRIVE[sd->driveno].dev; /* device */
sbp->b.b_data = bp->b_data; /* data buffer */
sbp->b.b_blkno = bp->b_blkno + sd->driveoffset;
sbp->b.b_iodone = sdio_done; /* come here on completion */
BUF_LOCKINIT(&sbp->b); /* get a lock for the buffer */
BUF_LOCK(&sbp->b, LK_EXCLUSIVE); /* and lock it */
sbp->bp = bp; /* note the address of the original header */
sbp->sdno = sd->sdno; /* note for statistics */
sbp->driveno = sd->driveno;
endoffset = bp->b_blkno + sbp->b.b_bcount / DEV_BSIZE; /* final sector offset */
if (endoffset > sd->sectors) { /* beyond the end */
sbp->b.b_bcount -= (endoffset - sd->sectors) * DEV_BSIZE; /* trim */
if (sbp->b.b_bcount <= 0) { /* nothing to transfer */
bp->b_resid = bp->b_bcount; /* nothing transferred */
biodone(bp);
Free(sbp);
return;
}
}
#if VINUMDEBUG
if (debug & DEBUG_ADDRESSES)
log(LOG_DEBUG,
" %s dev %d.%d, sd %d, offset 0x%x, devoffset 0x%x, length %ld\n",
sbp->b.b_flags & B_READ ? "Read" : "Write",
major(sbp->b.b_dev),
minor(sbp->b.b_dev),
sbp->sdno,
(u_int) (sbp->b.b_blkno - SD[sbp->sdno].driveoffset),
(int) sbp->b.b_blkno,
sbp->b.b_bcount);
#endif
s = splbio();
#if VINUMDEBUG
if (debug & DEBUG_LASTREQS)
logrq(loginfo_sdiol, (union rqinfou) &sbp->b, &sbp->b);
#endif
BUF_STRATEGY(&sbp->b, 0);
splx(s);
}
/*
* Write disk label back to device after modification.
*/
static int
l32_writedisklabel(cdev_t dev, struct diskslices *ssp, struct diskslice *sp,
disklabel_t lpx)
{
struct disklabel32 *lp;
struct disklabel32 *dlp;
struct buf *bp;
const char *msg;
int error = 0;
lp = lpx.lab32;
if (lp->d_partitions[RAW_PART].p_offset != 0)
return (EXDEV); /* not quite right */
bp = geteblk((int)lp->d_secsize);
bp->b_bio1.bio_offset = (off_t)LABELSECTOR32 * lp->d_secsize;
bp->b_bio1.bio_done = biodone_sync;
bp->b_bio1.bio_flags |= BIO_SYNC;
bp->b_bcount = lp->d_secsize;
#if 1
/*
* We read the label first to see if it's there,
* in which case we will put ours at the same offset into the block..
* (I think this is stupid [Julian])
* Note that you can't write a label out over a corrupted label!
* (also stupid.. how do you write the first one? by raw writes?)
*/
bp->b_flags &= ~B_INVAL;
bp->b_cmd = BUF_CMD_READ;
KKASSERT(dkpart(dev) == WHOLE_SLICE_PART);
dev_dstrategy(dev, &bp->b_bio1);
error = biowait(&bp->b_bio1, "labrd");
if (error)
goto done;
for (dlp = (struct disklabel32 *)bp->b_data;
dlp <= (struct disklabel32 *)
((char *)bp->b_data + lp->d_secsize - sizeof(*dlp));
dlp = (struct disklabel32 *)((char *)dlp + sizeof(long))) {
if (dlp->d_magic == DISKMAGIC32 &&
dlp->d_magic2 == DISKMAGIC32 && dkcksum32(dlp) == 0) {
*dlp = *lp;
lpx.lab32 = dlp;
msg = l32_fixlabel(NULL, sp, lpx, TRUE);
if (msg) {
error = EINVAL;
} else {
bp->b_cmd = BUF_CMD_WRITE;
bp->b_bio1.bio_done = biodone_sync;
bp->b_bio1.bio_flags |= BIO_SYNC;
KKASSERT(dkpart(dev) == WHOLE_SLICE_PART);
dev_dstrategy(dev, &bp->b_bio1);
error = biowait(&bp->b_bio1, "labwr");
}
goto done;
}
}
error = ESRCH;
done:
#else
bzero(bp->b_data, lp->d_secsize);
dlp = (struct disklabel32 *)bp->b_data;
*dlp = *lp;
bp->b_flags &= ~B_INVAL;
bp->b_cmd = BUF_CMD_WRITE;
bp->b_bio1.bio_done = biodone_sync;
bp->b_bio1.bio_flags |= BIO_SYNC;
BUF_STRATEGY(bp, 1);
error = biowait(&bp->b_bio1, "labwr");
#endif
bp->b_flags |= B_INVAL | B_AGE;
brelse(bp);
return (error);
}
