static int
nvd_bio_submit(struct nvd_disk *ndisk, struct bio *bp)
{
int err;
bp->bio_driver1 = NULL;
atomic_add_int(&ndisk->cur_depth, 1);
err = nvme_ns_bio_process(ndisk->ns, bp, nvd_done);
if (err) {
atomic_add_int(&ndisk->cur_depth, -1);
if (__predict_false(bp->bio_flags & BIO_ORDERED))
atomic_add_int(&ndisk->ordered_in_flight, -1);
bp->bio_error = err;
bp->bio_flags |= BIO_ERROR;
bp->bio_resid = bp->bio_bcount;
biodone(bp);
return (-1);
}
return (0);
}
/* Main flash handling task. */
static void
opalflash_task(void *arg)
{
struct opalflash_softc *sc;
struct bio *bp;
device_t dev;
sc = arg;
for (;;) {
dev = sc->sc_dev;
OPALFLASH_LOCK(sc);
do {
bp = bioq_first(&sc->sc_bio_queue);
if (bp == NULL)
msleep(sc, &sc->sc_mtx, PRIBIO, "opalflash", 0);
} while (bp == NULL);
bioq_remove(&sc->sc_bio_queue, bp);
OPALFLASH_UNLOCK(sc);
switch (bp->bio_cmd) {
case BIO_DELETE:
bp->bio_error = opalflash_erase(sc, bp->bio_offset,
bp->bio_bcount);
break;
case BIO_READ:
bp->bio_error = opalflash_read(sc, bp->bio_offset,
bp->bio_data, bp->bio_bcount);
break;
case BIO_WRITE:
bp->bio_error = opalflash_write(sc, bp->bio_offset,
bp->bio_data, bp->bio_bcount);
break;
default:
bp->bio_error = EINVAL;
}
biodone(bp);
}
}
void
puffs_parkdone_asyncbiowrite(struct puffs_mount *pmp,
struct puffs_req *preq, void *arg)
{
struct puffs_vnmsg_write *write_msg = (void *)preq;
struct buf *bp = arg;
DPRINTF(("%s\n", __func__));
bp->b_error = checkerr(pmp, preq->preq_rv, __func__);
if (bp->b_error == 0) {
if (write_msg->pvnr_resid > bp->b_bcount) {
puffs_senderr(pmp, PUFFS_ERR_WRITE, E2BIG,
"resid grew", preq->preq_cookie);
bp->b_error = E2BIG;
} else {
bp->b_resid = write_msg->pvnr_resid;
}
}
biodone(bp);
}
/*
* Memory file system I/O.
*
* Trivial on the HP since buffer has already been mapped into KVA space.
*/
void
mfs_doio(struct mfsnode *mfsp, struct buf *bp)
{
caddr_t base;
long offset = bp->b_blkno << DEV_BSHIFT;
int s;
if (bp->b_bcount > mfsp->mfs_size - offset)
bp->b_bcount = mfsp->mfs_size - offset;
base = mfsp->mfs_baseoff + offset;
if (bp->b_flags & B_READ)
bp->b_error = copyin(base, bp->b_data, bp->b_bcount);
else
bp->b_error = copyout(bp->b_data, base, bp->b_bcount);
if (bp->b_error)
bp->b_flags |= B_ERROR;
else
bp->b_resid = 0;
s = splbio();
biodone(bp);
splx(s);
}
static void
dk_done1(struct dk_softc *dksc, struct buf *bp, bool lock)
{
struct disk *dk = &dksc->sc_dkdev;
if (bp->b_error != 0) {
struct cfdriver *cd = device_cfdriver(dksc->sc_dev);
diskerr(bp, cd->cd_name, "error", LOG_PRINTF, 0,
dk->dk_label);
printf("\n");
}
if (lock)
mutex_enter(&dksc->sc_iolock);
disk_unbusy(dk, bp->b_bcount - bp->b_resid, (bp->b_flags & B_READ));
if (lock)
mutex_exit(&dksc->sc_iolock);
rnd_add_uint32(&dksc->sc_rnd_source, bp->b_rawblkno);
biodone(bp);
}
static void
destroy_geom_disk(struct nvd_disk *ndisk)
{
struct bio *bp;
taskqueue_free(ndisk->tq);
disk_destroy(ndisk->disk);
mtx_lock(&ndisk->bioqlock);
for (;;) {
bp = bioq_takefirst(&ndisk->bioq);
if (bp == NULL)
break;
bp->bio_error = EIO;
bp->bio_flags |= BIO_ERROR;
bp->bio_resid = bp->bio_bcount;
biodone(bp);
}
mtx_unlock(&ndisk->bioqlock);
mtx_destroy(&ndisk->bioqlock);
}
/* I/O on subdisk completed */
void
sdio_done(struct bio *bio)
{
struct sdbuf *sbp;
get_mplock();
sbp = (struct sdbuf *) bio->bio_buf;
if (sbp->b.b_flags & B_ERROR) { /* had an error */
sbp->bio->bio_buf->b_flags |= B_ERROR; /* propagate upwards */
sbp->bio->bio_buf->b_error = sbp->b.b_error;
}
#ifdef VINUMDEBUG
if (debug & DEBUG_LASTREQS)
logrq(loginfo_sdiodone, (union rqinfou)bio, bio);
#endif
sbp->bio->bio_buf->b_resid = sbp->b.b_resid; /* copy the resid field */
/* Now update the statistics */
if (sbp->b.b_cmd == BUF_CMD_READ) { /* read operation */
DRIVE[sbp->driveno].reads++;
DRIVE[sbp->driveno].bytes_read += sbp->b.b_bcount;
SD[sbp->sdno].reads++;
SD[sbp->sdno].bytes_read += sbp->b.b_bcount;
} else { /* write operation */
DRIVE[sbp->driveno].writes++;
DRIVE[sbp->driveno].bytes_written += sbp->b.b_bcount;
SD[sbp->sdno].writes++;
SD[sbp->sdno].bytes_written += sbp->b.b_bcount;
}
biodone_sync(bio);
biodone(sbp->bio); /* complete the caller's I/O */
BUF_UNLOCK(&sbp->b);
uninitbufbio(&sbp->b);
Free(sbp);
rel_mplock();
}
static void
mfi_disk_strategy(struct bio *bio)
{
struct mfi_disk *sc;
struct mfi_softc *controller;
sc = bio->bio_disk->d_drv1;
if (sc == NULL) {
bio->bio_error = EINVAL;
bio->bio_flags |= BIO_ERROR;
bio->bio_resid = bio->bio_bcount;
biodone(bio);
return;
}
controller = sc->ld_controller;
bio->bio_driver1 = (void *)(uintptr_t)sc->ld_id;
mtx_lock(&controller->mfi_io_lock);
mfi_enqueue_bio(controller, bio);
mfi_startio(controller);
mtx_unlock(&controller->mfi_io_lock);
return;
}
void
rdstrategy(struct buf *bp)
{
struct rdsoftc *rd;
struct hdcsoftc *sc;
struct disklabel *lp;
int s;
if ((rd = device_lookup_private(&rd_cd, DISKUNIT(bp->b_dev))) == NULL) {
bp->b_error = ENXIO;
goto done;
}
sc = rd->sc_hdc;
lp = rd->sc_disk.dk_label;
if ((bounds_check_with_label(&rd->sc_disk, bp, 1)) <= 0)
goto done;
if (bp->b_bcount == 0)
goto done;
bp->b_rawblkno =
bp->b_blkno + lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
bp->b_cylinder = bp->b_rawblkno / lp->d_secpercyl;
s = splbio();
BUFQ_PUT(sc->sc_q, bp);
if (inq == 0) {
inq = 1;
vsbus_dma_start(&sc->sc_vd);
}
splx(s);
return;
done: biodone(bp);
}
/*
* Read/write routine for a buffer. Finds the proper unit, range checks
* arguments, and schedules the transfer. Does not wait for the transfer
* to complete. Multi-page transfers are supported. All I/O requests must
* be a multiple of a sector in length.
*/
static void
idad_strategy(struct bio *bp)
{
struct idad_softc *drv;
int s;
drv = bp->bio_disk->d_drv1;
if (drv == NULL) {
bp->bio_error = EINVAL;
goto bad;
}
/*
* software write protect check
*/
if (drv->flags & DRV_WRITEPROT && (bp->bio_cmd == BIO_WRITE)) {
bp->bio_error = EROFS;
goto bad;
}
bp->bio_driver1 = drv;
s = splbio();
ida_submit_buf(drv->controller, bp);
splx(s);
return;
bad:
bp->bio_flags |= BIO_ERROR;
/*
* Correctly set the buf to indicate a completed transfer
*/
bp->bio_resid = bp->bio_bcount;
biodone(bp);
return;
}
void
destroy_geom_disk(struct nand_chip *chip)
{
struct bio *bp;
taskqueue_free(chip->tq);
disk_destroy(chip->ndisk);
disk_destroy(chip->rdisk);
mtx_lock(&chip->qlock);
for (;;) {
bp = bioq_takefirst(&chip->bioq);
if (bp == NULL)
break;
bp->bio_error = EIO;
bp->bio_flags |= BIO_ERROR;
bp->bio_resid = bp->bio_bcount;
biodone(bp);
}
mtx_unlock(&chip->qlock);
mtx_destroy(&chip->qlock);
}
void
fss_strategy(struct buf *bp)
{
const bool write = ((bp->b_flags & B_READ) != B_READ);
struct fss_softc *sc = device_lookup_private(&fss_cd, minor(bp->b_dev));
mutex_enter(&sc->sc_slock);
if (write || !FSS_ISVALID(sc)) {
mutex_exit(&sc->sc_slock);
bp->b_error = (write ? EROFS : ENXIO);
bp->b_resid = bp->b_bcount;
biodone(bp);
return;
}
bp->b_rawblkno = bp->b_blkno;
bufq_put(sc->sc_bufq, bp);
cv_signal(&sc->sc_work_cv);
mutex_exit(&sc->sc_slock);
}
/*
* Calculate the logical to physical mapping if not done already,
* then call the device strategy routine.
*/
int
ntfs_strategy(void *v)
{
struct vop_strategy_args /* {
struct vnode *a_vp;
struct buf *a_bp;
} */ *ap = v;
struct buf *bp = ap->a_bp;
struct vnode *vp = ap->a_vp;
struct fnode *fp = VTOF(vp);
struct ntnode *ip = FTONT(fp);
struct ntfsmount *ntmp = ip->i_mp;
int error;
dprintf(("ntfs_strategy: blkno: %d, lblkno: %d\n",
(u_int32_t)bp->b_blkno,
(u_int32_t)bp->b_lblkno));
dprintf(("strategy: bcount: %u flags: 0x%x\n",
(u_int32_t)bp->b_bcount,bp->b_flags));
if (bp->b_flags & B_READ) {
u_int32_t toread;
if (ntfs_cntob(bp->b_blkno) >= fp->f_size) {
clrbuf(bp);
error = 0;
} else {
toread = MIN(bp->b_bcount,
fp->f_size - ntfs_cntob(bp->b_blkno));
dprintf(("ntfs_strategy: toread: %d, fsize: %d\n",
toread,(u_int32_t)fp->f_size));
error = ntfs_readattr(ntmp, ip, fp->f_attrtype,
fp->f_attrname, ntfs_cntob(bp->b_blkno),
toread, bp->b_data, NULL);
if (error) {
printf("ntfs_strategy: ntfs_readattr failed\n");
bp->b_error = error;
}
memset((char *)bp->b_data + toread, 0,
bp->b_bcount - toread);
}
} else {
size_t tmp;
u_int32_t towrite;
if (ntfs_cntob(bp->b_blkno) + bp->b_bcount >= fp->f_size) {
printf("ntfs_strategy: CAN'T EXTEND FILE\n");
bp->b_error = error = EFBIG;
} else {
towrite = MIN(bp->b_bcount,
fp->f_size - ntfs_cntob(bp->b_blkno));
dprintf(("ntfs_strategy: towrite: %d, fsize: %d\n",
towrite,(u_int32_t)fp->f_size));
error = ntfs_writeattr_plain(ntmp, ip, fp->f_attrtype,
fp->f_attrname, ntfs_cntob(bp->b_blkno),towrite,
bp->b_data, &tmp, NULL);
if (error) {
printf("ntfs_strategy: ntfs_writeattr fail\n");
bp->b_error = error;
}
}
}
biodone(bp);
return (error);
}
/*
* Strategy routine called from dm_strategy.
*/
static int
dm_target_stripe_strategy(dm_table_entry_t *table_en, struct buf *bp)
{
dm_target_stripe_config_t *tsc;
struct bio *bio = &bp->b_bio1;
struct buf *nestbuf;
uint64_t blkno, blkoff;
uint64_t stripe, blknr;
uint32_t stripe_off, stripe_rest, num_blks, issue_blks;
int devnr;
tsc = table_en->target_config;
if (tsc == NULL)
return 0;
/* calculate extent of request */
KKASSERT(bp->b_resid % DEV_BSIZE == 0);
switch(bp->b_cmd) {
case BUF_CMD_READ:
case BUF_CMD_WRITE:
case BUF_CMD_FREEBLKS:
/*
* Loop through to individual operations
*/
blkno = bp->b_bio1.bio_offset / DEV_BSIZE;
blkoff = 0;
num_blks = bp->b_resid / DEV_BSIZE;
nestiobuf_init(bio);
while (num_blks > 0) {
/* blockno to strip piece nr */
stripe = blkno / tsc->stripe_chunksize;
stripe_off = blkno % tsc->stripe_chunksize;
/* where we are inside the strip */
devnr = stripe % tsc->stripe_num;
blknr = stripe / tsc->stripe_num;
/* how much is left before we hit a boundary */
stripe_rest = tsc->stripe_chunksize - stripe_off;
/* issue this piece on stripe `stripe' */
issue_blks = MIN(stripe_rest, num_blks);
nestbuf = getpbuf(NULL);
nestbuf->b_flags |= bio->bio_buf->b_flags & B_HASBOGUS;
nestiobuf_add(bio, nestbuf, blkoff,
issue_blks * DEV_BSIZE, NULL);
/* I need number of bytes. */
nestbuf->b_bio1.bio_offset =
blknr * tsc->stripe_chunksize + stripe_off;
nestbuf->b_bio1.bio_offset +=
tsc->stripe_devs[devnr].offset;
nestbuf->b_bio1.bio_offset *= DEV_BSIZE;
vn_strategy(tsc->stripe_devs[devnr].pdev->pdev_vnode,
&nestbuf->b_bio1);
blkno += issue_blks;
blkoff += issue_blks * DEV_BSIZE;
num_blks -= issue_blks;
}
nestiobuf_start(bio);
break;
case BUF_CMD_FLUSH:
nestiobuf_init(bio);
for (devnr = 0; devnr < tsc->stripe_num; ++devnr) {
nestbuf = getpbuf(NULL);
nestbuf->b_flags |= bio->bio_buf->b_flags & B_HASBOGUS;
nestiobuf_add(bio, nestbuf, 0, 0, NULL);
nestbuf->b_bio1.bio_offset = 0;
vn_strategy(tsc->stripe_devs[devnr].pdev->pdev_vnode,
&nestbuf->b_bio1);
}
nestiobuf_start(bio);
break;
default:
bp->b_flags |= B_ERROR;
bp->b_error = EIO;
biodone(bio);
break;
}
return 0;
}
/* Pseudo strategy function
* Called by scsipi_do_ioctl() via physio/physstrat if there is to
* be data transfered, and directly if there is no data transfer.
*
* Should I reorganize this so it returns to physio instead
* of sleeping in scsiio_scsipi_cmd? Is there any advantage, other
* than avoiding the probable duplicate wakeup in iodone? [PD]
*
* No, seems ok to me... [JRE]
* (I don't see any duplicate wakeups)
*
* Can't be used with block devices or raw_read/raw_write directly
* from the cdevsw/bdevsw tables because they couldn't have added
* the screq structure. [JRE]
*/
static void
scsistrategy(struct buf *bp)
{
struct scsi_ioctl *si;
scsireq_t *screq;
struct scsipi_periph *periph;
int error;
int flags = 0;
si = si_find(bp);
if (si == NULL) {
printf("scsistrategy: "
"No matching ioctl request found in queue\n");
error = EINVAL;
goto done;
}
screq = &si->si_screq;
periph = si->si_periph;
SC_DEBUG(periph, SCSIPI_DB2, ("user_strategy\n"));
/*
* We're in trouble if physio tried to break up the transfer.
*/
if (bp->b_bcount != screq->datalen) {
scsipi_printaddr(periph);
printf("physio split the request.. cannot proceed\n");
error = EIO;
goto done;
}
if (screq->timeout == 0) {
error = EINVAL;
goto done;
}
if (screq->cmdlen > sizeof(struct scsipi_generic)) {
scsipi_printaddr(periph);
printf("cmdlen too big\n");
error = EFAULT;
goto done;
}
if ((screq->flags & SCCMD_READ) && screq->datalen > 0)
flags |= XS_CTL_DATA_IN;
if ((screq->flags & SCCMD_WRITE) && screq->datalen > 0)
flags |= XS_CTL_DATA_OUT;
if (screq->flags & SCCMD_TARGET)
flags |= XS_CTL_TARGET;
if (screq->flags & SCCMD_ESCAPE)
flags |= XS_CTL_ESCAPE;
error = scsipi_command(periph, (void *)screq->cmd, screq->cmdlen,
(void *)bp->b_data, screq->datalen,
0, /* user must do the retries *//* ignored */
screq->timeout, bp, flags | XS_CTL_USERCMD);
done:
if (error)
bp->b_resid = bp->b_bcount;
bp->b_error = error;
biodone(bp);
return;
}
static void
isf_task(void *arg)
{
struct isf_softc *sc = arg;
struct bio *bp;
int ss = sc->isf_disk->d_sectorsize;
int error, i;
for (;;) {
ISF_LOCK(sc);
do {
bp = bioq_first(&sc->isf_bioq);
if (bp == NULL) {
if (sc->isf_doomed)
kproc_exit(0);
else
ISF_SLEEP(sc, sc, 0);
}
} while (bp == NULL);
bioq_remove(&sc->isf_bioq, bp);
error = 0;
switch (bp->bio_cmd) {
case BIO_READ:
isf_read(sc, bp->bio_pblkno * ss, bp->bio_data,
bp->bio_bcount);
break;
case BIO_WRITE:
/*
* In principle one could suspend the in-progress
* erase, process any pending writes to other
* blocks and then proceed, but that seems
* overly complex for the likely usage modes.
*/
if (sc->isf_erasing) {
error = EBUSY;
break;
}
/*
* Read in the block we want to write and check that
* we're only setting bits to 0. If an erase would
* be required return an I/O error.
*/
isf_read(sc, bp->bio_pblkno * ss, sc->isf_rbuf,
bp->bio_bcount);
for (i = 0; i < bp->bio_bcount / 2; i++)
if ((sc->isf_rbuf[i] &
((uint16_t *)bp->bio_data)[i]) !=
((uint16_t *)bp->bio_data)[i]) {
device_printf(sc->isf_dev, "write"
" requires erase at 0x%08jx\n",
bp->bio_pblkno * ss);
error = EIO;
break;
}
if (error != 0)
break;
error = isf_write(sc, bp->bio_pblkno * ss,
bp->bio_data, bp->bio_bcount);
break;
default:
panic("%s: unsupported I/O operation %d", __func__,
bp->bio_cmd);
}
if (error == 0)
biodone(bp);
else
biofinish(bp, NULL, error);
ISF_UNLOCK(sc);
}
}
/*
* Calculate the logical to physical mapping if not done already,
* then call the device strategy routine.
*/
int
ufs_strategy(void *v)
{
struct vop_strategy_args /* {
struct vnode *a_vp;
struct buf *a_bp;
} */ *ap = v;
struct buf *bp;
struct vnode *vp;
struct inode *ip;
struct mount *mp;
int error;
bp = ap->a_bp;
vp = ap->a_vp;
ip = VTOI(vp);
if (vp->v_type == VBLK || vp->v_type == VCHR)
panic("ufs_strategy: spec");
KASSERT(bp->b_bcount != 0);
if (bp->b_blkno == bp->b_lblkno) {
error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno,
NULL);
if (error) {
bp->b_error = error;
biodone(bp);
return (error);
}
if (bp->b_blkno == -1) /* no valid data */
clrbuf(bp);
}
if (bp->b_blkno < 0) { /* block is not on disk */
biodone(bp);
return (0);
}
vp = ip->i_devvp;
error = VOP_STRATEGY(vp, bp);
if (error)
return error;
if (!BUF_ISREAD(bp))
return 0;
mp = wapbl_vptomp(vp);
if (mp == NULL || mp->mnt_wapbl_replay == NULL ||
!WAPBL_REPLAY_ISOPEN(mp) ||
!WAPBL_REPLAY_CAN_READ(mp, bp->b_blkno, bp->b_bcount))
return 0;
error = biowait(bp);
if (error)
return error;
error = WAPBL_REPLAY_READ(mp, bp->b_data, bp->b_blkno, bp->b_bcount);
if (error) {
mutex_enter(&bufcache_lock);
SET(bp->b_cflags, BC_INVAL);
mutex_exit(&bufcache_lock);
}
return error;
}
static void
ptdone(struct cam_periph *periph, union ccb *done_ccb)
{
struct pt_softc *softc;
struct ccb_scsiio *csio;
softc = (struct pt_softc *)periph->softc;
csio = &done_ccb->csio;
switch (csio->ccb_h.ccb_state) {
case PT_CCB_BUFFER_IO:
case PT_CCB_BUFFER_IO_UA:
{
struct buf *bp;
struct bio *bio;
bio = (struct bio *)done_ccb->ccb_h.ccb_bio;
bp = bio->bio_buf;
if ((done_ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
int error;
int sf;
if ((csio->ccb_h.ccb_state & PT_CCB_RETRY_UA) != 0)
sf = SF_RETRY_UA;
else
sf = 0;
error = pterror(done_ccb, CAM_RETRY_SELTO, sf);
if (error == ERESTART) {
/*
* A retry was scheuled, so
* just return.
*/
return;
}
if (error != 0) {
struct buf *q_bp;
struct bio *q_bio;
if (error == ENXIO) {
/*
* Catastrophic error. Mark our device
* as invalid.
*/
xpt_print(periph->path,
"Invalidating device\n");
softc->flags |= PT_FLAG_DEVICE_INVALID;
}
/*
* return all queued I/O with EIO, so that
* the client can retry these I/Os in the
* proper order should it attempt to recover.
*/
while ((q_bio = bioq_takefirst(&softc->bio_queue)) != NULL) {
q_bp = q_bio->bio_buf;
q_bp->b_resid = q_bp->b_bcount;
q_bp->b_error = EIO;
q_bp->b_flags |= B_ERROR;
biodone(q_bio);
}
bp->b_error = error;
bp->b_resid = bp->b_bcount;
bp->b_flags |= B_ERROR;
} else {
bp->b_resid = csio->resid;
bp->b_error = 0;
if (bp->b_resid != 0) {
/* Short transfer ??? */
bp->b_flags |= B_ERROR;
}
}
if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0)
cam_release_devq(done_ccb->ccb_h.path,
/*relsim_flags*/0,
/*reduction*/0,
/*timeout*/0,
/*getcount_only*/0);
} else {
bp->b_resid = csio->resid;
if (bp->b_resid != 0)
bp->b_flags |= B_ERROR;
}
/*
* Block out any asyncronous callbacks
* while we touch the pending ccb list.
*/
LIST_REMOVE(&done_ccb->ccb_h, periph_links.le);
devstat_end_transaction_buf(&softc->device_stats, bp);
biodone(bio);
break;
}
case PT_CCB_WAITING:
/* Caller will release the CCB */
wakeup(&done_ccb->ccb_h.cbfcnp);
return;
}
xpt_release_ccb(done_ccb);
}
void
bmdstrategy(struct buf *bp)
{
int unit = BMD_UNIT(bp->b_dev);
struct bmd_softc *sc;
int offset, disksize, resid;
int page, pg_offset, pg_resid;
void *data;
if (unit >= bmd_cd.cd_ndevs) {
bp->b_error = ENXIO;
goto done;
}
sc = device_lookup_private(&bmd_cd, BMD_UNIT(bp->b_dev));
if (sc == NULL) {
bp->b_error = ENXIO;
goto done;
}
DPRINTF(("bmdstrategy: %s blkno %d bcount %ld:",
(bp->b_flags & B_READ) ? "read " : "write",
bp->b_blkno, bp->b_bcount));
bp->b_resid = bp->b_bcount;
offset = (bp->b_blkno << DEV_BSHIFT);
disksize = sc->sc_maxpage * BMD_PAGESIZE;
if (offset >= disksize) {
/* EOF if read, EIO if write */
if (bp->b_flags & B_READ)
goto done;
bp->b_error = EIO;
goto done;
}
resid = bp->b_resid;
if (resid > disksize - offset)
resid = disksize - offset;
data = bp->b_data;
do {
page = offset / BMD_PAGESIZE;
pg_offset = offset % BMD_PAGESIZE;
/* length */
pg_resid = MIN(resid, BMD_PAGESIZE - pg_offset);
/* switch bank page */
bus_space_write_1(sc->sc_iot, sc->sc_ioh, BMD_PAGE, page);
/* XXX we should use DMA transfer? */
if ((bp->b_flags & B_READ)) {
bus_space_read_region_1(sc->sc_iot, sc->sc_bank,
pg_offset, data, pg_resid);
} else {
bus_space_write_region_1(sc->sc_iot, sc->sc_bank,
pg_offset, data, pg_resid);
}
data = (char *)data + pg_resid;
offset += pg_resid;
resid -= pg_resid;
bp->b_resid -= pg_resid;
} while (resid > 0);
DPRINTF(("\n"));
done:
biodone(bp);
}
static void
mcdstrategy(struct bio *bp)
{
struct mcd_softc *sc;
int s;
sc = (struct mcd_softc *)bp->bio_dev->si_drv1;
/* test validity */
/*MCD_TRACE("strategy: buf=0x%lx, unit=%ld, block#=%ld bcount=%ld\n",
bp,unit,bp->bio_blkno,bp->bio_bcount);*/
if (bp->bio_blkno < 0) {
device_printf(sc->dev, "strategy failure: blkno = %ld, bcount = %ld\n",
(long)bp->bio_blkno, bp->bio_bcount);
bp->bio_error = EINVAL;
bp->bio_flags |= BIO_ERROR;
goto bad;
}
/* if device invalidated (e.g. media change, door open), error */
if (!(sc->data.flags & MCDVALID)) {
device_printf(sc->dev, "media changed\n");
bp->bio_error = EIO;
goto bad;
}
/* read only */
if (!(bp->bio_cmd == BIO_READ)) {
bp->bio_error = EROFS;
goto bad;
}
/* no data to read */
if (bp->bio_bcount == 0)
goto done;
if (!(sc->data.flags & MCDTOC)) {
bp->bio_error = EIO;
goto bad;
}
bp->bio_pblkno = bp->bio_blkno;
bp->bio_resid = 0;
/* queue it */
s = splbio();
bioqdisksort(&sc->data.head, bp);
splx(s);
/* now check whether we can perform processing */
mcd_start(sc);
return;
bad:
bp->bio_flags |= BIO_ERROR;
done:
bp->bio_resid = bp->bio_bcount;
biodone(bp);
return;
}
int
zvol_strategy(buf_t *bp)
{
zvol_state_t *zv = ddi_get_soft_state(zvol_state, getminor(bp->b_edev));
uint64_t off, volsize;
size_t size, resid;
char *addr;
objset_t *os;
int error = 0;
int sync;
int reading;
int txg_sync_needed = B_FALSE;
if (zv == NULL) {
bioerror(bp, ENXIO);
biodone(bp);
return (0);
}
if (getminor(bp->b_edev) == 0) {
bioerror(bp, EINVAL);
biodone(bp);
return (0);
}
if (zv->zv_readonly && !(bp->b_flags & B_READ)) {
bioerror(bp, EROFS);
biodone(bp);
return (0);
}
off = ldbtob(bp->b_blkno);
volsize = zv->zv_volsize;
os = zv->zv_objset;
ASSERT(os != NULL);
sync = !(bp->b_flags & B_ASYNC) && !(zil_disable);
bp_mapin(bp);
addr = bp->b_un.b_addr;
resid = bp->b_bcount;
/*
* There must be no buffer changes when doing a dmu_sync() because
* we can't change the data whilst calculating the checksum.
* A better approach than a per zvol rwlock would be to lock ranges.
*/
reading = bp->b_flags & B_READ;
if (reading || resid <= zvol_immediate_write_sz)
rw_enter(&zv->zv_dslock, RW_READER);
else
rw_enter(&zv->zv_dslock, RW_WRITER);
while (resid != 0 && off < volsize) {
size = MIN(resid, 1UL << 20); /* cap at 1MB per tx */
if (size > volsize - off) /* don't write past the end */
size = volsize - off;
if (reading) {
error = dmu_read(os, ZVOL_OBJ, off, size, addr);
} else {
dmu_tx_t *tx = dmu_tx_create(os);
dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
} else {
dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
if (sync) {
/* use the ZIL to commit this write */
if (zvol_log_write(zv, tx, off, size,
addr) != 0) {
txg_sync_needed = B_TRUE;
}
}
dmu_tx_commit(tx);
}
}
if (error)
break;
off += size;
addr += size;
resid -= size;
}
rw_exit(&zv->zv_dslock);
if ((bp->b_resid = resid) == bp->b_bcount)
bioerror(bp, off > volsize ? EINVAL : error);
biodone(bp);
if (sync) {
if (txg_sync_needed)
txg_wait_synced(dmu_objset_pool(os), 0);
else
zil_commit(zv->zv_zilog, UINT64_MAX, 0);
}
return (0);
}
void
marustrategy(struct buf *bp)
{
struct maru_softc *sc;
struct disklabel *lp;
struct partition *pp;
int len;
int err = ENXIO;
m_u64 offset;
DB("marustrategy(%p)\n", bp);
maru_printbuf(bp);
DB("ms:1\n");
sc = &maru_softc[maruunit(bp->b_dev)];
if (num_maru<1 ||
maruunit(bp->b_dev) >= num_maru ||
!(sc->sc_flags&MUF_INITED) ||
!sc->sc_kapi)
{
err:
DB("ms:2\n");
maru_berror(bp, err);
DB("ms:3\n");
return;
}
DB("ms:4\n");
len = bp->b_bcount;
bp->b_resid = len;
if (len<1)
{
DB("ms:5\n");
biodone(bp);
DB("ms:6\n");
return;
}
DB("ms:6.1\n");
offset = dbtob(bp->b_blkno);
lp = sc->sc_dkdev.dk_label;
/* the transfer must be a whole number of blocks */
if (len % lp->d_secsize != 0)
{
maru_berror(bp, EINVAL);
return;
}
/*
* Do bounds checking and adjust transfer. If there's an error,
* the bounds check will flag that for us.
*/
DB("ms:6.2\n");
if (DISKPART(bp->b_dev) != RAW_PART &&
bounds_check_with_label(bp, lp, sc->sc_flags&MUF_WLABEL) <= 0)
{
biodone(bp);
return;
}
/*
* Translate the partition-relative block number to an absolute.
*/
DB("ms:6.3\n");
if (DISKPART(bp->b_dev) != RAW_PART)
{
pp = &sc->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
offset += pp->p_offset * lp->d_secsize;
}
if (bp->b_flags & B_READ)
{
struct maru_message *msg;
DB("ms:7\n");
msg = malloc(sizeof *msg, M_DEVBUF, M_NOWAIT);
if (!msg)
goto err;
msg->mm_flags = MARU_READ_REQ;
DB("ms:8\n");
msg->mm_id = maru_acquire_token(sc, bp);
msg->mm_len = len;
msg->mm_offset = offset;
DB("ms:9\n");
if ((err = sc->sc_kapi->ka_inject(sc->sc_kapi, msg, sizeof *msg)))
{
DB("ms:10\n");
free(msg, M_DEVBUF);
goto err;
}
DB("ms:11\n");
sc->sc_reading++;
return;
}
else /* B_WRITE */
{
struct maru_message *msg;
DB("ms:13\n");
msg = malloc(sizeof *msg, M_DEVBUF, M_NOWAIT);
if (!msg)
goto err;
msg->mm_flags = MARU_WRITE;
msg->mm_id = maru_acquire_token(sc, bp);
msg->mm_len = len;
msg->mm_offset = offset;
DB("ms:14\n");
if ((err = sc->sc_kapi->ka_inject(sc->sc_kapi, msg, sizeof(msg)+msg->mm_len)))
{
DB("ms:15\n");
free(msg, M_DEVBUF);
goto err;
}
DB("ms:16\n");
sc->sc_writing++;
return;
}
DB("ms:17\n");
}
/*
* Actually translate the requested transfer into one the physical driver can
* understand. The transfer is described by a buf and will include only one
* physical transfer.
*/
void
cdstrategy(struct buf *bp)
{
struct cd_softc *cd;
int s;
if ((cd = cdlookup(DISKUNIT(bp->b_dev))) == NULL) {
bp->b_error = ENXIO;
goto bad;
}
SC_DEBUG(cd->sc_link, SDEV_DB2, ("cdstrategy: %ld bytes @ blk %d\n",
bp->b_bcount, bp->b_blkno));
/*
* If the device has been made invalid, error out
* maybe the media changed, or no media loaded
*/
if ((cd->sc_link->flags & SDEV_MEDIA_LOADED) == 0) {
bp->b_error = EIO;
goto bad;
}
/*
* The transfer must be a whole number of blocks.
*/
if ((bp->b_bcount % cd->sc_dk.dk_label->d_secsize) != 0) {
bp->b_error = EINVAL;
goto bad;
}
/*
* If it's a null transfer, return immediately
*/
if (bp->b_bcount == 0)
goto done;
/*
* Do bounds checking, adjust transfer. if error, process.
* If end of partition, just return.
*/
if (bounds_check_with_label(bp, cd->sc_dk.dk_label,
(cd->flags & (CDF_WLABEL|CDF_LABELLING)) != 0) <= 0)
goto done;
s = splbio();
/*
* Place it in the queue of disk activities for this disk
*/
disksort(&cd->buf_queue, bp);
/*
* Tell the device to get going on the transfer if it's
* not doing anything, otherwise just wait for completion
*/
cdstart(cd);
device_unref(&cd->sc_dev);
splx(s);
return;
bad:
bp->b_flags |= B_ERROR;
done:
/*
* Correctly set the buf to indicate a completed xfer
*/
bp->b_resid = bp->b_bcount;
s = splbio();
biodone(bp);
splx(s);
if (cd != NULL)
device_unref(&cd->sc_dev);
}
int
puffs_doio(struct vnode *vp, struct bio *bio, struct thread *td)
{
struct buf *bp = bio->bio_buf;
struct ucred *cred;
struct uio *uiop;
struct uio uio;
struct iovec io;
size_t n;
int error = 0;
if (td != NULL && td->td_proc != NULL)
cred = td->td_proc->p_ucred;
else
cred = proc0.p_ucred;
uiop = &uio;
uiop->uio_iov = &io;
uiop->uio_iovcnt = 1;
uiop->uio_segflg = UIO_SYSSPACE;
uiop->uio_td = td;
/*
* clear B_ERROR and B_INVAL state prior to initiating the I/O. We
* do this here so we do not have to do it in all the code that
* calls us.
*/
bp->b_flags &= ~(B_ERROR | B_INVAL);
KASSERT(bp->b_cmd != BUF_CMD_DONE,
("puffs_doio: bp %p already marked done!", bp));
if (bp->b_cmd == BUF_CMD_READ) {
io.iov_len = uiop->uio_resid = (size_t)bp->b_bcount;
io.iov_base = bp->b_data;
uiop->uio_rw = UIO_READ;
uiop->uio_offset = bio->bio_offset;
error = puffs_directread(vp, uiop, 0, cred);
if (error == 0 && uiop->uio_resid) {
n = (size_t)bp->b_bcount - uiop->uio_resid;
bzero(bp->b_data + n, bp->b_bcount - n);
uiop->uio_resid = 0;
}
if (error) {
bp->b_flags |= B_ERROR;
bp->b_error = error;
}
bp->b_resid = uiop->uio_resid;
} else {
KKASSERT(bp->b_cmd == BUF_CMD_WRITE);
if (bio->bio_offset + bp->b_dirtyend > puffs_meta_getsize(vp))
bp->b_dirtyend = puffs_meta_getsize(vp) -
bio->bio_offset;
if (bp->b_dirtyend > bp->b_dirtyoff) {
io.iov_len = uiop->uio_resid = bp->b_dirtyend
- bp->b_dirtyoff;
uiop->uio_offset = bio->bio_offset + bp->b_dirtyoff;
io.iov_base = (char *)bp->b_data + bp->b_dirtyoff;
uiop->uio_rw = UIO_WRITE;
error = puffs_directwrite(vp, uiop, 0, cred);
if (error == EINTR
|| (!error && (bp->b_flags & B_NEEDCOMMIT))) {
crit_enter();
bp->b_flags &= ~(B_INVAL|B_NOCACHE);
if ((bp->b_flags & B_PAGING) == 0)
bdirty(bp);
if (error)
bp->b_flags |= B_EINTR;
crit_exit();
} else {
if (error) {
bp->b_flags |= B_ERROR;
bp->b_error = error;
}
bp->b_dirtyoff = bp->b_dirtyend = 0;
}
bp->b_resid = uiop->uio_resid;
} else {
bp->b_resid = 0;
}
}
biodone(bio);
KKASSERT(bp->b_cmd == BUF_CMD_DONE);
if (bp->b_flags & B_EINTR)
return (EINTR);
if (bp->b_flags & B_ERROR)
return (bp->b_error ? bp->b_error : EIO);
return (0);
}
/*
* cdstart looks to see if there is a buf waiting for the device
* and that the device is not already busy. If both are true,
* It deques the buf and creates a scsi command to perform the
* transfer in the buf. The transfer request will call scsi_done
* on completion, which will in turn call this routine again
* so that the next queued transfer is performed.
* The bufs are queued by the strategy routine (cdstrategy)
*
* This routine is also called after other non-queued requests
* have been made of the scsi driver, to ensure that the queue
* continues to be drained.
*
* must be called at the correct (highish) spl level
* cdstart() is called at splbio from cdstrategy, cdrestart and scsi_done
*/
void
cdstart(void *v)
{
struct cd_softc *cd = v;
struct scsi_link *sc_link = cd->sc_link;
struct buf *bp = 0;
struct buf *dp;
struct scsi_rw_big cmd_big;
struct scsi_rw cmd_small;
struct scsi_generic *cmdp;
int blkno, nblks, cmdlen, error;
struct partition *p;
splassert(IPL_BIO);
SC_DEBUG(sc_link, SDEV_DB2, ("cdstart\n"));
/*
* Check if the device has room for another command
*/
while (sc_link->openings > 0) {
/*
* there is excess capacity, but a special waits
* It'll need the adapter as soon as we clear out of the
* way and let it run (user level wait).
*/
if (sc_link->flags & SDEV_WAITING) {
sc_link->flags &= ~SDEV_WAITING;
wakeup((caddr_t)sc_link);
return;
}
/*
* See if there is a buf with work for us to do..
*/
dp = &cd->buf_queue;
if ((bp = dp->b_actf) == NULL) /* yes, an assign */
return;
dp->b_actf = bp->b_actf;
/*
* If the device has become invalid, abort all the
* reads and writes until all files have been closed and
* re-opened
*/
if ((sc_link->flags & SDEV_MEDIA_LOADED) == 0) {
bp->b_error = EIO;
bp->b_flags |= B_ERROR;
bp->b_resid = bp->b_bcount;
biodone(bp);
continue;
}
/*
* We have a buf, now we should make a command
*
* First, translate the block to absolute and put it in terms
* of the logical blocksize of the device.
*/
blkno =
bp->b_blkno / (cd->sc_dk.dk_label->d_secsize / DEV_BSIZE);
p = &cd->sc_dk.dk_label->d_partitions[DISKPART(bp->b_dev)];
blkno += DL_GETPOFFSET(p);
nblks = howmany(bp->b_bcount, cd->sc_dk.dk_label->d_secsize);
/*
* Fill out the scsi command. If the transfer will
* fit in a "small" cdb, use it.
*/
if (!(sc_link->flags & SDEV_ATAPI) &&
!(sc_link->quirks & SDEV_ONLYBIG) &&
((blkno & 0x1fffff) == blkno) &&
((nblks & 0xff) == nblks)) {
/*
* We can fit in a small cdb.
*/
bzero(&cmd_small, sizeof(cmd_small));
cmd_small.opcode = (bp->b_flags & B_READ) ?
READ_COMMAND : WRITE_COMMAND;
_lto3b(blkno, cmd_small.addr);
cmd_small.length = nblks & 0xff;
cmdlen = sizeof(cmd_small);
cmdp = (struct scsi_generic *)&cmd_small;
} else {
/*
* Need a large cdb.
*/
bzero(&cmd_big, sizeof(cmd_big));
cmd_big.opcode = (bp->b_flags & B_READ) ?
READ_BIG : WRITE_BIG;
_lto4b(blkno, cmd_big.addr);
_lto2b(nblks, cmd_big.length);
cmdlen = sizeof(cmd_big);
cmdp = (struct scsi_generic *)&cmd_big;
}
/* Instrumentation. */
disk_busy(&cd->sc_dk);
/*
* Call the routine that chats with the adapter.
* Note: we cannot sleep as we may be an interrupt
*/
error = scsi_scsi_cmd(sc_link, cmdp, cmdlen,
(u_char *) bp->b_data, bp->b_bcount, SCSI_RETRIES, 30000,
bp, SCSI_NOSLEEP | ((bp->b_flags & B_READ) ? SCSI_DATA_IN :
SCSI_DATA_OUT));
switch (error) {
case 0:
timeout_del(&cd->sc_timeout);
break;
case EAGAIN:
/*
* The device can't start another i/o. Try again later.
*/
dp->b_actf = bp;
disk_unbusy(&cd->sc_dk, 0, 0);
timeout_add(&cd->sc_timeout, 1);
return;
default:
disk_unbusy(&cd->sc_dk, 0, 0);
printf("%s: not queued, error %d\n",
cd->sc_dev.dv_xname, error);
break;
}
}
}
/*
* Mark I/O complete on a buffer, release it if I/O is asynchronous,
* and wake up anyone waiting for it.
*/
void
iodone(struct buf *bp)
{
ASSERT(SEMA_HELD(&bp->b_sem));
(void) biodone(bp);
}
static void
htif_blk_task(void *arg)
{
struct htif_blk_request req __aligned(HTIF_ALIGN);
struct htif_blk_softc *sc;
uint64_t req_paddr;
struct bio *bp;
uint64_t paddr;
uint64_t resp;
uint64_t cmd;
int i;
sc = (struct htif_blk_softc *)arg;
while (1) {
HTIF_BLK_LOCK(sc);
do {
bp = bioq_takefirst(&sc->bio_queue);
if (bp == NULL)
msleep(sc, &sc->sc_mtx, PRIBIO, "jobqueue", 0);
} while (bp == NULL);
HTIF_BLK_UNLOCK(sc);
if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
HTIF_BLK_LOCK(sc);
rmb();
req.offset = (bp->bio_pblkno * sc->disk->d_sectorsize);
req.size = bp->bio_bcount;
paddr = vtophys(bp->bio_data);
KASSERT(paddr != 0, ("paddr is 0"));
req.addr = paddr;
sc->curtag++;
req.tag = sc->curtag;
cmd = sc->index;
cmd <<= HTIF_DEV_ID_SHIFT;
if (bp->bio_cmd == BIO_READ)
cmd |= (HTIF_CMD_READ << HTIF_CMD_SHIFT);
else
cmd |= (HTIF_CMD_WRITE << HTIF_CMD_SHIFT);
req_paddr = vtophys(&req);
KASSERT(req_paddr != 0, ("req_paddr is 0"));
cmd |= req_paddr;
sc->cmd_done = 0;
resp = htif_command(cmd);
htif_blk_intr(sc, resp);
/* Wait for interrupt */
i = 0;
while (sc->cmd_done == 0) {
msleep(&sc->intr_chan, &sc->sc_mtx, PRIBIO, "intr", hz/2);
if (i++ > 2) {
/* TODO: try to re-issue operation on timeout ? */
bp->bio_error = EIO;
bp->bio_flags |= BIO_ERROR;
disk_err(bp, "hard error", -1, 1);
break;
}
}
HTIF_BLK_UNLOCK(sc);
biodone(bp);
} else {
printf("unknown op %d\n", bp->bio_cmd);
}
}
}
void
sd_buf_done(struct scsi_xfer *xs)
{
struct sd_softc *sc = xs->sc_link->device_softc;
struct buf *bp = xs->cookie;
int error, s;
switch (xs->error) {
case XS_NOERROR:
bp->b_error = 0;
bp->b_resid = xs->resid;
break;
case XS_NO_CCB:
/* The adapter is busy, requeue the buf and try it later. */
disk_unbusy(&sc->sc_dk, bp->b_bcount - xs->resid,
bp->b_flags & B_READ);
bufq_requeue(&sc->sc_bufq, bp);
scsi_xs_put(xs);
SET(sc->flags, SDF_WAITING);
timeout_add(&sc->sc_timeout, 1);
return;
case XS_SENSE:
case XS_SHORTSENSE:
#ifdef SCSIDEBUG
scsi_sense_print_debug(xs);
#endif
error = sd_interpret_sense(xs);
if (error == 0) {
bp->b_error = 0;
bp->b_resid = xs->resid;
break;
}
if (error != ERESTART) {
bp->b_error = error;
xs->retries = 0;
}
goto retry;
case XS_BUSY:
if (xs->retries) {
if (scsi_delay(xs, 1) != ERESTART)
xs->retries = 0;
}
goto retry;
case XS_TIMEOUT:
retry:
if (xs->retries--) {
scsi_xs_exec(xs);
return;
}
/* FALLTHROUGH */
default:
if (bp->b_error == 0)
bp->b_error = EIO;
bp->b_flags |= B_ERROR;
bp->b_resid = bp->b_bcount;
break;
}
disk_unbusy(&sc->sc_dk, bp->b_bcount - xs->resid,
bp->b_flags & B_READ);
s = splbio();
biodone(bp);
splx(s);
scsi_xs_put(xs);
}
static void
mcd_doread(struct mcd_softc *sc, int state, struct mcd_mbx *mbxin)
{
struct mcd_mbx *mbx;
struct bio *bp;
int rm, i, k;
struct mcd_read2 rbuf;
int blknum;
caddr_t addr;
mbx = (state!=MCD_S_BEGIN) ? sc->ch_mbxsave : mbxin;
bp = mbx->bp;
loop:
switch (state) {
case MCD_S_BEGIN:
mbx = sc->ch_mbxsave = mbxin;
case MCD_S_BEGIN1:
retry_status:
/* get status */
MCD_WRITE(sc, MCD_REG_COMMAND, MCD_CMDGETSTAT);
mbx->count = RDELAY_WAITSTAT;
sc->ch_state = MCD_S_WAITSTAT;
sc->ch = timeout(mcd_timeout, (caddr_t)sc, hz/100); /* XXX */
return;
case MCD_S_WAITSTAT:
sc->ch_state = MCD_S_WAITSTAT;
untimeout(mcd_timeout,(caddr_t)sc, sc->ch);
if (mbx->count-- >= 0) {
if (MCD_READ(sc, MCD_FLAGS) & MFL_STATUS_NOT_AVAIL) {
sc->ch_state = MCD_S_WAITSTAT;
timeout(mcd_timeout, (caddr_t)sc, hz/100); /* XXX */
return;
}
sc->data.status = MCD_READ(sc, MCD_REG_STATUS) & 0xFF;
if (sc->data.status & MCD_ST_CMDCHECK)
goto retry_status;
if (mcd_setflags(sc) < 0)
goto changed;
MCD_TRACE("got WAITSTAT delay=%d\n",
RDELAY_WAITSTAT-mbx->count);
/* reject, if audio active */
if (sc->data.status & MCDAUDIOBSY) {
device_printf(sc->dev, "audio is active\n");
goto readerr;
}
retry_mode:
/* to check for raw/cooked mode */
if (sc->data.flags & MCDREADRAW) {
rm = MCD_MD_RAW;
mbx->sz = MCDRBLK;
} else {
rm = MCD_MD_COOKED;
mbx->sz = sc->data.blksize;
}
if (rm == sc->data.curr_mode)
goto modedone;
mbx->count = RDELAY_WAITMODE;
sc->data.curr_mode = MCD_MD_UNKNOWN;
mbx->mode = rm;
MCD_WRITE(sc, MCD_REG_COMMAND, MCD_CMDSETMODE);
MCD_WRITE(sc, MCD_REG_COMMAND, rm);
sc->ch_state = MCD_S_WAITMODE;
sc->ch = timeout(mcd_timeout, (caddr_t)sc, hz/100); /* XXX */
return;
} else {
device_printf(sc->dev, "timeout getstatus\n");
goto readerr;
}
case MCD_S_WAITMODE:
sc->ch_state = MCD_S_WAITMODE;
untimeout(mcd_timeout, (caddr_t)sc, sc->ch);
if (mbx->count-- < 0) {
device_printf(sc->dev, "timeout set mode\n");
goto readerr;
}
if (MCD_READ(sc, MCD_FLAGS) & MFL_STATUS_NOT_AVAIL) {
sc->ch_state = MCD_S_WAITMODE;
sc->ch = timeout(mcd_timeout, (caddr_t)sc, hz/100);
return;
}
sc->data.status = MCD_READ(sc, MCD_REG_STATUS) & 0xFF;
if (sc->data.status & MCD_ST_CMDCHECK) {
sc->data.curr_mode = MCD_MD_UNKNOWN;
goto retry_mode;
}
if (mcd_setflags(sc) < 0)
goto changed;
sc->data.curr_mode = mbx->mode;
MCD_TRACE("got WAITMODE delay=%d\n",
RDELAY_WAITMODE-mbx->count);
modedone:
/* for first block */
mbx->nblk = (bp->bio_bcount + (mbx->sz-1)) / mbx->sz;
mbx->skip = 0;
nextblock:
blknum = (bp->bio_blkno / (mbx->sz/DEV_BSIZE))
+ mbx->skip/mbx->sz;
MCD_TRACE("mcd_doread: read blknum=%d for bp=%p\n",
blknum, bp);
/* build parameter block */
hsg2msf(blknum,rbuf.start_msf);
retry_read:
/* send the read command */
critical_enter();
MCD_WRITE(sc, MCD_REG_COMMAND, sc->data.read_command);
MCD_WRITE(sc, MCD_REG_COMMAND, rbuf.start_msf[0]);
MCD_WRITE(sc, MCD_REG_COMMAND, rbuf.start_msf[1]);
MCD_WRITE(sc, MCD_REG_COMMAND, rbuf.start_msf[2]);
MCD_WRITE(sc, MCD_REG_COMMAND, 0);
MCD_WRITE(sc, MCD_REG_COMMAND, 0);
MCD_WRITE(sc, MCD_REG_COMMAND, 1);
critical_exit();
/* Spin briefly (<= 2ms) to avoid missing next block */
for (i = 0; i < 20; i++) {
k = MCD_READ(sc, MCD_FLAGS);
if (!(k & MFL_DATA_NOT_AVAIL))
goto got_it;
DELAY(100);
}
mbx->count = RDELAY_WAITREAD;
sc->ch_state = MCD_S_WAITREAD;
sc->ch = timeout(mcd_timeout, (caddr_t)sc, hz/100); /* XXX */
return;
case MCD_S_WAITREAD:
sc->ch_state = MCD_S_WAITREAD;
untimeout(mcd_timeout, (caddr_t)sc, sc->ch);
if (mbx->count-- > 0) {
k = MCD_READ(sc, MCD_FLAGS);
if (!(k & MFL_DATA_NOT_AVAIL)) { /* XXX */
MCD_TRACE("got data delay=%d\n",
RDELAY_WAITREAD-mbx->count);
got_it:
/* data is ready */
addr = bp->bio_data + mbx->skip;
MCD_WRITE(sc, MCD_REG_CTL2,0x04); /* XXX */
for (i=0; i<mbx->sz; i++)
*addr++ = MCD_READ(sc, MCD_REG_RDATA);
MCD_WRITE(sc, MCD_REG_CTL2,0x0c); /* XXX */
k = MCD_READ(sc, MCD_FLAGS);
/* If we still have some junk, read it too */
if (!(k & MFL_DATA_NOT_AVAIL)) {
MCD_WRITE(sc, MCD_REG_CTL2, 0x04); /* XXX */
(void)MCD_READ(sc, MCD_REG_RDATA);
(void)MCD_READ(sc, MCD_REG_RDATA);
MCD_WRITE(sc, MCD_REG_CTL2, 0x0c); /* XXX */
}
if (--mbx->nblk > 0) {
mbx->skip += mbx->sz;
goto nextblock;
}
/* return buffer */
bp->bio_resid = 0;
biodone(bp);
sc->data.flags &= ~(MCDMBXBSY|MCDREADRAW);
mcd_start(sc);
return;
}
if (!(k & MFL_STATUS_NOT_AVAIL)) {
sc->data.status = MCD_READ(sc, MCD_REG_STATUS) & 0xFF;
if (sc->data.status & MCD_ST_CMDCHECK)
goto retry_read;
if (mcd_setflags(sc) < 0)
goto changed;
}
sc->ch_state = MCD_S_WAITREAD;
sc->ch = timeout(mcd_timeout, (caddr_t)sc, hz/100); /* XXX */
return;
} else {
device_printf(sc->dev, "timeout read data\n");
goto readerr;
}
}
readerr:
if (mbx->retry-- > 0) {
device_printf(sc->dev, "retrying\n");
state = MCD_S_BEGIN1;
goto loop;
}
harderr:
/* invalidate the buffer */
bp->bio_flags |= BIO_ERROR;
bp->bio_resid = bp->bio_bcount;
biodone(bp);
sc->data.flags &= ~(MCDMBXBSY|MCDREADRAW);
mcd_start(sc);
return;
changed:
device_printf(sc->dev, "media changed\n");
goto harderr;
#ifdef NOTDEF
device_printf(sc->dev, "unit timeout, resetting\n");
MCD_WRITE(sc, MCD_REG_RESET, MCD_CMDRESET);
DELAY(300000);
(void)mcd_getstat(sc, 1);
(void)mcd_getstat(sc, 1);
/*sc->data.status &= ~MCDDSKCHNG; */
sc->data.debug = 1; /* preventive set debug mode */
#endif
}
/*
* define the low-level requests needed to perform
* a high-level I/O operation for a specific plex
* 'plexno'.
*
* Return 0 if all subdisks involved in the
* request are up, 1 if some subdisks are not up,
* and -1 if the request is at least partially
* outside the bounds of the subdisks.
*
* Modify the pointer *diskstart to point to the
* end address. On read, return on the first bad
* subdisk, so that the caller
* (build_read_request) can try alternatives.
*
* On entry to this routine, the prq structures
* are not assigned. The assignment is performed
* by expandrq(). Strictly speaking, the elements
* rqe->sdno of all entries should be set to -1,
* since 0 (from bzero) is a valid subdisk number.
* We avoid this problem by initializing the ones
* we use, and not looking at the others (index >=
* prq->requests).
*/
enum requeststatus
bre5(struct request *rq,
int plexno,
daddr_t * diskaddr,
daddr_t diskend)
{
struct metrics m; /* most of the information */
struct sd *sd;
struct plex *plex;
struct buf *bp; /* user's bp */
struct rqgroup *rqg; /* the request group that we will create */
struct rqelement *rqe; /* point to this request information */
int rsectors; /* sectors remaining in this stripe */
int mysdno; /* another sd index in loops */
int rqno; /* request number */
rqg = NULL; /* shut up, damn compiler */
m.diskstart = *diskaddr; /* start of transfer */
bp = rq->bp; /* buffer pointer */
plex = &PLEX[plexno]; /* point to the plex */
while (*diskaddr < diskend) { /* until we get it all sorted out */
if (*diskaddr >= plex->length) /* beyond the end of the plex */
return REQUEST_EOF; /* can't continue */
m.badsdno = -1; /* no bad subdisk yet */
/* Part A: Define the request */
/*
* First, calculate some sizes:
* The offset of the start address from
* the start of the stripe.
*/
m.stripeoffset = *diskaddr % (plex->stripesize * (plex->subdisks - 1));
/*
* The plex-relative address of the
* start of the stripe.
*/
m.stripebase = *diskaddr - m.stripeoffset;
/* subdisk containing the parity stripe */
if (plex->organization == plex_raid5)
m.psdno = plex->subdisks - 1
- (*diskaddr / (plex->stripesize * (plex->subdisks - 1)))
% plex->subdisks;
else /* RAID-4 */
m.psdno = plex->subdisks - 1;
/*
* The number of the subdisk in which
* the start is located.
*/
m.firstsdno = m.stripeoffset / plex->stripesize;
if (m.firstsdno >= m.psdno) /* at or past parity sd */
m.firstsdno++; /* increment it */
/*
* The offset from the beginning of
* the stripe on this subdisk.
*/
m.initoffset = m.stripeoffset % plex->stripesize;
/* The offset of the stripe start relative to this subdisk */
m.sdbase = m.stripebase / (plex->subdisks - 1);
m.useroffset = *diskaddr - m.diskstart; /* The offset of the start in the user buffer */
/*
* The number of sectors to transfer in the
* current (first) subdisk.
*/
m.initlen = min(diskend - *diskaddr, /* the amount remaining to transfer */
plex->stripesize - m.initoffset); /* and the amount left in this block */
/*
* The number of sectors to transfer in this stripe
* is the minumum of the amount remaining to transfer
* and the amount left in this stripe.
*/
m.stripesectors = min(diskend - *diskaddr,
plex->stripesize * (plex->subdisks - 1) - m.stripeoffset);
/* The number of data subdisks involved in this request */
m.sdcount = (m.stripesectors + m.initoffset + plex->stripesize - 1) / plex->stripesize;
/* Part B: decide what kind of transfer this will be.
* start and end addresses of the transfer in
* the current block.
*
* There are a number of different kinds of
* transfer, each of which relates to a
* specific subdisk:
*
* 1. Normal read. All participating subdisks
* are up, and the transfer can be made
* directly to the user buffer. The bounds
* of the transfer are described by
* m.dataoffset and m.datalen. We have
* already calculated m.initoffset and
* m.initlen, which define the parameters
* for the first data block.
*
* 2. Recovery read. One participating
* subdisk is down. To recover data, all
* the other subdisks, including the parity
* subdisk, must be read. The data is
* recovered by exclusive-oring all the
* other blocks. The bounds of the
* transfer are described by m.groupoffset
* and m.grouplen.
*
* 3. A read request may request reading both
* available data (normal read) and
* non-available data (recovery read).
* This can be a problem if the address
* ranges of the two reads do not coincide:
* in this case, the normal read needs to
* be extended to cover the address range
* of the recovery read, and must thus be
* performed out of malloced memory.
*
* 4. Normal write. All the participating
* subdisks are up. The bounds of the
* transfer are described by m.dataoffset
* and m.datalen. Since these values
* differ for each block, we calculate the
* bounds for the parity block
* independently as the maximum of the
* individual blocks and store these values
* in m.writeoffset and m.writelen. This
* write proceeds in four phases:
*
* i. Read the old contents of each block
* and the parity block.
* ii. ``Remove'' the old contents from
* the parity block with exclusive or.
* iii. ``Insert'' the new contents of the
* block in the parity block, again
* with exclusive or.
*
* iv. Write the new contents of the data
* blocks and the parity block. The data
* block transfers can be made directly from
* the user buffer.
*
* 5. Degraded write where the data block is
* not available. The bounds of the
* transfer are described by m.groupoffset
* and m.grouplen. This requires the
* following steps:
*
* i. Read in all the other data blocks,
* excluding the parity block.
*
* ii. Recreate the parity block from the
* other data blocks and the data to be
* written.
*
* iii. Write the parity block.
*
* 6. Parityless write, a write where the
* parity block is not available. This is
* in fact the simplest: just write the
* data blocks. This can proceed directly
* from the user buffer. The bounds of the
* transfer are described by m.dataoffset
* and m.datalen.
*
* 7. Combination of degraded data block write
* and normal write. In this case the
* address ranges of the reads may also
* need to be extended to cover all
* participating blocks.
*
* All requests in a group transfer transfer
* the same address range relative to their
* subdisk. The individual transfers may
* vary, but since our group of requests is
* all in a single slice, we can define a
* range in which they all fall.
*
* In the following code section, we determine
* which kind of transfer we will perform. If
* there is a group transfer, we also decide
* its bounds relative to the subdisks. At
* the end, we have the following values:
*
* m.flags indicates the kinds of transfers
* we will perform.
* m.initoffset indicates the offset of the
* beginning of any data operation relative
* to the beginning of the stripe base.
* m.initlen specifies the length of any data
* operation.
* m.dataoffset contains the same value as
* m.initoffset.
* m.datalen contains the same value as
* m.initlen. Initially dataoffset and
* datalen describe the parameters for the
* first data block; while building the data
* block requests, they are updated for each
* block.
* m.groupoffset indicates the offset of any
* group operation relative to the beginning
* of the stripe base.
* m.grouplen specifies the length of any
* group operation.
* m.writeoffset indicates the offset of a
* normal write relative to the beginning of
* the stripe base. This value differs from
* m.dataoffset in that it applies to the
* entire operation, and not just the first
* block.
* m.writelen specifies the total span of a
* normal write operation. writeoffset and
* writelen are used to define the parity
* block.
*/
m.groupoffset = 0; /* assume no group... */
m.grouplen = 0; /* until we know we have one */
m.writeoffset = m.initoffset; /* start offset of transfer */
m.writelen = 0; /* nothing to write yet */
m.flags = 0; /* no flags yet */
rsectors = m.stripesectors; /* remaining sectors to examine */
m.dataoffset = m.initoffset; /* start at the beginning of the transfer */
m.datalen = m.initlen;
if (m.sdcount > 1) {
plex->multiblock++; /* more than one block for the request */
/*
* If we have two transfers that don't overlap,
* (one at the end of the first block, the other
* at the beginning of the second block),
* it's cheaper to split them.
*/
if (rsectors < plex->stripesize) {
m.sdcount = 1; /* just one subdisk */
m.stripesectors = m.initlen; /* and just this many sectors */
rsectors = m.initlen; /* and in the loop counter */
}
}
if (SD[plex->sdnos[m.psdno]].state < sd_reborn) /* is our parity subdisk down? */
m.badsdno = m.psdno; /* note that it's down */
if (bp->b_flags & B_READ) { /* read operation */
for (mysdno = m.firstsdno; rsectors > 0; mysdno++) {
if (mysdno == m.psdno) /* ignore parity on read */
mysdno++;
if (mysdno == plex->subdisks) /* wraparound */
mysdno = 0;
if (mysdno == m.psdno) /* parity, */
mysdno++; /* we've given already */
if (SD[plex->sdnos[mysdno]].state < sd_reborn) { /* got a bad subdisk, */
if (m.badsdno >= 0) /* we had one already, */
return REQUEST_DOWN; /* we can't take a second */
m.badsdno = mysdno; /* got the first */
m.groupoffset = m.dataoffset; /* define the bounds */
m.grouplen = m.datalen;
m.flags |= XFR_RECOVERY_READ; /* we need recovery */
plex->recovered_reads++; /* count another one */
} else
m.flags |= XFR_NORMAL_READ; /* normal read */
/* Update the pointers for the next block */
m.dataoffset = 0; /* back to the start of the stripe */
rsectors -= m.datalen; /* remaining sectors to examine */
m.datalen = min(rsectors, plex->stripesize); /* amount that will fit in this block */
}
} else { /* write operation */
for (mysdno = m.firstsdno; rsectors > 0; mysdno++) {
if (mysdno == m.psdno) /* parity stripe, we've dealt with that */
mysdno++;
if (mysdno == plex->subdisks) /* wraparound */
mysdno = 0;
if (mysdno == m.psdno) /* parity, */
mysdno++; /* we've given already */
sd = &SD[plex->sdnos[mysdno]];
if (sd->state != sd_up) {
enum requeststatus s;
s = checksdstate(sd, rq, *diskaddr, diskend); /* do we need to change state? */
if (s && (m.badsdno >= 0)) { /* second bad disk, */
int sdno;
/*
* If the parity disk is down, there's
* no recovery. We make all involved
* subdisks stale. Otherwise, we
* should be able to recover, but it's
* like pulling teeth. Fix it later.
*/
for (sdno = 0; sdno < m.sdcount; sdno++) {
struct sd *sd = &SD[plex->sdnos[sdno]];
if (sd->state >= sd_reborn) /* sort of up, */
set_sd_state(sd->sdno, sd_stale, setstate_force); /* make it stale */
}
return s; /* and crap out */
}
m.badsdno = mysdno; /* note which one is bad */
m.flags |= XFR_DEGRADED_WRITE; /* we need recovery */
plex->degraded_writes++; /* count another one */
m.groupoffset = m.dataoffset; /* define the bounds */
m.grouplen = m.datalen;
} else {
m.flags |= XFR_NORMAL_WRITE; /* normal write operation */
if (m.writeoffset > m.dataoffset) { /* move write operation lower */
m.writelen = max(m.writeoffset + m.writelen,
m.dataoffset + m.datalen)
- m.dataoffset;
m.writeoffset = m.dataoffset;
} else
m.writelen = max(m.writeoffset + m.writelen,
m.dataoffset + m.datalen)
- m.writeoffset;
}
/* Update the pointers for the next block */
m.dataoffset = 0; /* back to the start of the stripe */
rsectors -= m.datalen; /* remaining sectors to examine */
m.datalen = min(rsectors, plex->stripesize); /* amount that will fit in this block */
}
if (m.badsdno == m.psdno) { /* got a bad parity block, */
struct sd *psd = &SD[plex->sdnos[m.psdno]];
if (psd->state == sd_down)
set_sd_state(psd->sdno, sd_obsolete, setstate_force); /* it's obsolete now */
else if (psd->state == sd_crashed)
set_sd_state(psd->sdno, sd_stale, setstate_force); /* it's stale now */
m.flags &= ~XFR_NORMAL_WRITE; /* this write isn't normal, */
m.flags |= XFR_PARITYLESS_WRITE; /* it's parityless */
plex->parityless_writes++; /* count another one */
}
}
/* reset the initial transfer values */
m.dataoffset = m.initoffset; /* start at the beginning of the transfer */
m.datalen = m.initlen;
/* decide how many requests we need */
if (m.flags & (XFR_RECOVERY_READ | XFR_DEGRADED_WRITE))
/* doing a recovery read or degraded write, */
m.rqcount = plex->subdisks; /* all subdisks */
else if (m.flags & XFR_NORMAL_WRITE) /* normal write, */
m.rqcount = m.sdcount + 1; /* all data blocks and the parity block */
else /* parityless write or normal read */
m.rqcount = m.sdcount; /* just the data blocks */
/* Part C: build the requests */
rqg = allocrqg(rq, m.rqcount); /* get a request group */
if (rqg == NULL) { /* malloc failed */
bp->b_flags |= B_ERROR;
bp->b_error = ENOMEM;
biodone(bp);
return REQUEST_ENOMEM;
}
rqg->plexno = plexno;
rqg->flags = m.flags;
rqno = 0; /* index in the request group */
/* 1: PARITY BLOCK */
/*
* Are we performing an operation which requires parity? In that case,
* work out the parameters and define the parity block.
* XFR_PARITYOP is XFR_NORMAL_WRITE | XFR_RECOVERY_READ | XFR_DEGRADED_WRITE
*/
if (m.flags & XFR_PARITYOP) { /* need parity */
rqe = &rqg->rqe[rqno]; /* point to element */
sd = &SD[plex->sdnos[m.psdno]]; /* the subdisk in question */
rqe->rqg = rqg; /* point back to group */
rqe->flags = (m.flags | XFR_PARITY_BLOCK | XFR_MALLOCED) /* always malloc parity block */
&~(XFR_NORMAL_READ | XFR_PARITYLESS_WRITE); /* transfer flags without data op stuf */
setrqebounds(rqe, &m); /* set up the bounds of the transfer */
rqe->sdno = sd->sdno; /* subdisk number */
rqe->driveno = sd->driveno;
if (build_rq_buffer(rqe, plex)) /* build the buffer */
return REQUEST_ENOMEM; /* can't do it */
rqe->b.b_flags |= B_READ; /* we must read first */
m.sdcount++; /* adjust the subdisk count */
rqno++; /* and point to the next request */
}
/*
* 2: DATA BLOCKS
* Now build up requests for the blocks required
* for individual transfers
*/
for (mysdno = m.firstsdno; rqno < m.sdcount; mysdno++, rqno++) {
if (mysdno == m.psdno) /* parity, */
mysdno++; /* we've given already */
if (mysdno == plex->subdisks) /* got to the end, */
mysdno = 0; /* wrap around */
if (mysdno == m.psdno) /* parity, */
mysdno++; /* we've given already */
rqe = &rqg->rqe[rqno]; /* point to element */
sd = &SD[plex->sdnos[mysdno]]; /* the subdisk in question */
rqe->rqg = rqg; /* point to group */
if (m.flags & XFR_NEEDS_MALLOC) /* we need a malloced buffer first */
rqe->flags = m.flags | XFR_DATA_BLOCK | XFR_MALLOCED; /* transfer flags */
else
rqe->flags = m.flags | XFR_DATA_BLOCK; /* transfer flags */
if (mysdno == m.badsdno) { /* this is the bad subdisk */
rqg->badsdno = rqno; /* note which one */
rqe->flags |= XFR_BAD_SUBDISK; /* note that it's dead */
/*
* we can't read or write from/to it,
* but we don't need to malloc
*/
rqe->flags &= ~(XFR_MALLOCED | XFR_NORMAL_READ | XFR_NORMAL_WRITE);
}
setrqebounds(rqe, &m); /* set up the bounds of the transfer */
rqe->useroffset = m.useroffset; /* offset in user buffer */
rqe->sdno = sd->sdno; /* subdisk number */
rqe->driveno = sd->driveno;
if (build_rq_buffer(rqe, plex)) /* build the buffer */
return REQUEST_ENOMEM; /* can't do it */
if ((m.flags & XFR_PARITYOP) /* parity operation, */
&&((m.flags & XFR_BAD_SUBDISK) == 0)) /* and not the bad subdisk, */
rqe->b.b_flags |= B_READ; /* we must read first */
/* Now update pointers for the next block */
*diskaddr += m.datalen; /* skip past what we've done */
m.stripesectors -= m.datalen; /* deduct from what's left */
m.useroffset += m.datalen; /* and move on in the user buffer */
m.datalen = min(m.stripesectors, plex->stripesize); /* and recalculate */
m.dataoffset = 0; /* start at the beginning of next block */
}
/*
* 3: REMAINING BLOCKS FOR RECOVERY
* Finally, if we have a recovery operation, build
* up transfers for the other subdisks. Follow the
* subdisks around until we get to where we started.
* These requests use only the group parameters.
*/
if ((rqno < m.rqcount) /* haven't done them all already */
&&(m.flags & (XFR_RECOVERY_READ | XFR_DEGRADED_WRITE))) {
for (; rqno < m.rqcount; rqno++, mysdno++) {
if (mysdno == m.psdno) /* parity, */
mysdno++; /* we've given already */
if (mysdno == plex->subdisks) /* got to the end, */
mysdno = 0; /* wrap around */
if (mysdno == m.psdno) /* parity, */
mysdno++; /* we've given already */
rqe = &rqg->rqe[rqno]; /* point to element */
sd = &SD[plex->sdnos[mysdno]]; /* the subdisk in question */
rqe->rqg = rqg; /* point to group */
rqe->sdoffset = m.sdbase + m.groupoffset; /* start of transfer */
rqe->dataoffset = 0; /* for tidiness' sake */
rqe->groupoffset = 0; /* group starts at the beginining */
rqe->datalen = 0;
rqe->grouplen = m.grouplen;
rqe->buflen = m.grouplen;
rqe->flags = (m.flags | XFR_MALLOCED) /* transfer flags without data op stuf */
&~XFR_DATAOP;
rqe->sdno = sd->sdno; /* subdisk number */
rqe->driveno = sd->driveno;
if (build_rq_buffer(rqe, plex)) /* build the buffer */
return REQUEST_ENOMEM; /* can't do it */
rqe->b.b_flags |= B_READ; /* we must read first */
}
}
/*
* We need to lock the address range before
* doing anything. We don't have to be
* performing a recovery operation: somebody
* else could be doing so, and the results could
* influence us. Note the fact here, we'll perform
* the lock in launch_requests.
*/
rqg->lockbase = m.stripebase;
if (*diskaddr < diskend) /* didn't finish the request on this stripe */
plex->multistripe++; /* count another one */
}
return REQUEST_OK;
}
