Print this page
7367 blkdev: support block size larger than 512
Reviewed by: Hans Rosenfeld <hans.rosenfeld@nexenta.com>
Reviewed by: Robert Mustacchi <rm@joyent.com>
Reviewed by: Garrett D'Amore <garrett@damore.org>
@@ -21,10 +21,11 @@
/*
* Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright 2012 Garrett D'Amore <garrett@damore.org>. All rights reserved.
* Copyright 2012 Alexey Zaytsev <alexey.zaytsev@gmail.com> All rights reserved.
* Copyright 2016 Nexenta Systems, Inc. All rights reserved.
+ * Copyright 2017 The MathWorks, Inc. All rights reserved.
*/
#include <sys/types.h>
#include <sys/ksynch.h>
#include <sys/kmem.h>
@@ -169,10 +170,11 @@
static void bd_submit(bd_t *, bd_xfer_impl_t *);
static void bd_runq_exit(bd_xfer_impl_t *, int);
static void bd_update_state(bd_t *);
static int bd_check_state(bd_t *, enum dkio_state *);
static int bd_flush_write_cache(bd_t *, struct dk_callback *);
+static int bd_check_uio(dev_t, struct uio *);
struct cmlb_tg_ops bd_tg_ops = {
TG_DK_OPS_VERSION_1,
bd_tg_rdwr,
bd_tg_getinfo,
@@ -751,11 +753,11 @@
ASSERT(bp);
xi->i_bp = bp;
xi->i_func = func;
- xi->i_blkno = bp->b_lblkno;
+ xi->i_blkno = bp->b_lblkno >> (bd->d_blkshift - DEV_BSHIFT);
if (bp->b_bcount == 0) {
xi->i_len = 0;
xi->i_nblks = 0;
xi->i_kaddr = NULL;
@@ -816,11 +818,11 @@
if ((ddi_dma_numwin(xi->i_dmah, &xi->i_num_win) !=
DDI_SUCCESS) ||
(ddi_dma_getwin(xi->i_dmah, 0, &xi->i_offset,
&len, &xi->i_dmac, &xi->i_ndmac) !=
DDI_SUCCESS) ||
- (P2PHASE(len, shift) != 0)) {
+ (P2PHASE(len, (1U << shift)) != 0)) {
(void) ddi_dma_unbind_handle(xi->i_dmah);
rv = EFAULT;
goto done;
}
xi->i_len = len;
@@ -1033,52 +1035,58 @@
diskaddr_t psize;
bd_t *bd;
bd_xfer_impl_t *xi;
buf_t *bp;
int rv;
+ uint32_t shift;
+ daddr_t d_blkno;
+ int d_nblk;
rw_enter(&bd_lock, RW_READER);
part = BDPART(dev);
inst = BDINST(dev);
if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
rw_exit(&bd_lock);
return (ENXIO);
}
+ shift = bd->d_blkshift;
+ d_blkno = blkno >> (shift - DEV_BSHIFT);
+ d_nblk = nblk >> (shift - DEV_BSHIFT);
/*
* do cmlb, but do it synchronously unless we already have the
* partition (which we probably should.)
*/
if (cmlb_partinfo(bd->d_cmlbh, part, &psize, &pstart, NULL, NULL,
(void *)1)) {
rw_exit(&bd_lock);
return (ENXIO);
}
- if ((blkno + nblk) > psize) {
+ if ((d_blkno + d_nblk) > psize) {
rw_exit(&bd_lock);
return (EINVAL);
}
bp = getrbuf(KM_NOSLEEP);
if (bp == NULL) {
rw_exit(&bd_lock);
return (ENOMEM);
}
- bp->b_bcount = nblk << bd->d_blkshift;
+ bp->b_bcount = nblk << DEV_BSHIFT;
bp->b_resid = bp->b_bcount;
bp->b_lblkno = blkno;
bp->b_un.b_addr = caddr;
xi = bd_xfer_alloc(bd, bp, bd->d_ops.o_write, KM_NOSLEEP);
if (xi == NULL) {
rw_exit(&bd_lock);
freerbuf(bp);
return (ENOMEM);
}
- xi->i_blkno = blkno + pstart;
+ xi->i_blkno = d_blkno + pstart;
xi->i_flags = BD_XFER_POLL;
bd_submit(bd, xi);
rw_exit(&bd_lock);
/*
@@ -1111,34 +1119,69 @@
if (bp->b_bcount > bd->d_maxxfer)
bp->b_bcount = bd->d_maxxfer;
}
static int
+bd_check_uio(dev_t dev, struct uio *uio)
+{
+ bd_t *bd;
+ uint32_t shift;
+
+ if ((bd = ddi_get_soft_state(bd_state, BDINST(dev))) == NULL) {
+ return (ENXIO);
+ }
+
+ shift = bd->d_blkshift;
+ if ((P2PHASE(uio->uio_loffset, (1U << shift)) != 0) ||
+ (P2PHASE(uio->uio_iov->iov_len, (1U << shift)) != 0)) {
+ return (EINVAL);
+ }
+
+ return (0);
+}
+
+static int
bd_read(dev_t dev, struct uio *uio, cred_t *credp)
{
_NOTE(ARGUNUSED(credp));
+ int ret = bd_check_uio(dev, uio);
+ if (ret != 0) {
+ return (ret);
+ }
return (physio(bd_strategy, NULL, dev, B_READ, bd_minphys, uio));
}
static int
bd_write(dev_t dev, struct uio *uio, cred_t *credp)
{
_NOTE(ARGUNUSED(credp));
+ int ret = bd_check_uio(dev, uio);
+ if (ret != 0) {
+ return (ret);
+ }
return (physio(bd_strategy, NULL, dev, B_WRITE, bd_minphys, uio));
}
static int
bd_aread(dev_t dev, struct aio_req *aio, cred_t *credp)
{
_NOTE(ARGUNUSED(credp));
+ int ret = bd_check_uio(dev, aio->aio_uio);
+ if (ret != 0) {
+ return (ret);
+ }
return (aphysio(bd_strategy, anocancel, dev, B_READ, bd_minphys, aio));
}
static int
bd_awrite(dev_t dev, struct aio_req *aio, cred_t *credp)
{
_NOTE(ARGUNUSED(credp));
+ int ret = bd_check_uio(dev, aio->aio_uio);
+ if (ret != 0) {
+ return (ret);
+ }
return (aphysio(bd_strategy, anocancel, dev, B_WRITE, bd_minphys, aio));
}
static int
bd_strategy(struct buf *bp)
@@ -1150,10 +1193,11 @@
diskaddr_t p_nblks;
diskaddr_t b_nblks;
bd_xfer_impl_t *xi;
uint32_t shift;
int (*func)(void *, bd_xfer_t *);
+ diskaddr_t lblkno;
part = BDPART(bp->b_edev);
inst = BDINST(bp->b_edev);
ASSERT(bp);
@@ -1172,25 +1216,26 @@
biodone(bp);
return (0);
}
shift = bd->d_blkshift;
-
- if ((P2PHASE(bp->b_bcount, (1U << shift)) != 0) ||
- (bp->b_lblkno > p_nblks)) {
- bioerror(bp, ENXIO);
+ lblkno = bp->b_lblkno >> (shift - DEV_BSHIFT);
+ if ((P2PHASE(bp->b_lblkno, (1U << (shift - DEV_BSHIFT))) != 0) ||
+ (P2PHASE(bp->b_bcount, (1U << shift)) != 0) ||
+ (lblkno > p_nblks)) {
+ bioerror(bp, EINVAL);
biodone(bp);
return (0);
}
b_nblks = bp->b_bcount >> shift;
- if ((bp->b_lblkno == p_nblks) || (bp->b_bcount == 0)) {
+ if ((lblkno == p_nblks) || (bp->b_bcount == 0)) {
biodone(bp);
return (0);
}
- if ((b_nblks + bp->b_lblkno) > p_nblks) {
- bp->b_resid = ((bp->b_lblkno + b_nblks - p_nblks) << shift);
+ if ((b_nblks + lblkno) > p_nblks) {
+ bp->b_resid = ((lblkno + b_nblks - p_nblks) << shift);
bp->b_bcount -= bp->b_resid;
} else {
bp->b_resid = 0;
}
func = (bp->b_flags & B_READ) ? bd->d_ops.o_read : bd->d_ops.o_write;
@@ -1202,11 +1247,11 @@
if (xi == NULL) {
/* bd_request_alloc will have done bioerror */
biodone(bp);
return (0);
}
- xi->i_blkno = bp->b_lblkno + p_lba;
+ xi->i_blkno = lblkno + p_lba;
bd_submit(bd, xi);
return (0);
}
@@ -1903,11 +1948,11 @@
len = min(bp->b_bcount - xi->i_offset, bd->d_maxxfer);
}
if ((rv != DDI_SUCCESS) ||
- (P2PHASE(len, (1U << xi->i_blkshift) != 0))) {
+ (P2PHASE(len, (1U << xi->i_blkshift)) != 0)) {
bd_runq_exit(xi, EFAULT);
bp->b_resid += xi->i_resid;
bd_xfer_free(xi);
bioerror(bp, EFAULT);