-
/*
* Copyright (c) 2000-2002 Apple Computer, Inc. All rights reserved.
*
* @APPLE_LICENSE_HEADER_START@
*
- * The contents of this file constitute Original Code as defined in and
- * are subject to the Apple Public Source License Version 1.1 (the
- * "License"). You may not use this file except in compliance with the
- * License. Please obtain a copy of the License at
- * http://www.apple.com/publicsource and read it before using this file.
+ * Copyright (c) 1999-2003 Apple Computer, Inc. All Rights Reserved.
+ *
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this
+ * file.
*
- * This Original Code and all software distributed under the License are
- * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
- * License for the specific language governing rights and limitations
- * under the License.
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
*
* @APPLE_LICENSE_HEADER_END@
*/
#define CL_NOZERO 0x80
#define CL_PAGEIN 0x100
#define CL_DEV_MEMORY 0x200
+#define CL_PRESERVE 0x400
+
+
+struct clios {
+ u_int io_completed; /* amount of io that has currently completed */
+ u_int io_issued; /* amount of io that was successfully issued */
+ int io_error; /* error code of first error encountered */
+ int io_wanted; /* someone is sleeping waiting for a change in state */
+};
+
static void cluster_zero(upl_t upl, vm_offset_t upl_offset,
int size, struct buf *bp);
static int cluster_nocopy_write(struct vnode *vp, struct uio *uio,
off_t newEOF, int devblocksize, int flags);
static int cluster_phys_read(struct vnode *vp, struct uio *uio,
- off_t filesize);
-static int cluster_phys_write(struct vnode *vp, struct uio *uio);
+ off_t filesize, int devblocksize, int flags);
+static int cluster_phys_write(struct vnode *vp, struct uio *uio,
+ off_t newEOF, int devblocksize, int flags);
+static int cluster_align_phys_io(struct vnode *vp, struct uio *uio,
+ addr64_t usr_paddr, int xsize, int devblocksize, int flags);
static int cluster_push_x(struct vnode *vp, off_t EOF, daddr_t first, daddr_t last, int can_delay);
static int cluster_try_push(struct vnode *vp, off_t newEOF, int can_delay, int push_all);
struct buf *cbp_next;
struct buf *real_bp;
struct vnode *vp;
+ struct clios *iostate;
int commit_size;
int pg_offset;
real_bp = cbp->b_real_bp;
vp = cbp->b_vp;
zero_offset= cbp->b_validend;
+ iostate = (struct clios *)cbp->b_iostate;
while (cbp) {
- if (cbp->b_vectorcount > 1)
- _FREE(cbp->b_vectorlist, M_SEGMENT);
-
if ((cbp->b_flags & B_ERROR) && error == 0)
error = cbp->b_error;
cbp = cbp_next;
}
+ if (zero_offset)
+ cluster_zero(upl, zero_offset, PAGE_SIZE - (zero_offset & PAGE_MASK), real_bp);
+
if ((vp->v_flag & VTHROTTLED) && (vp->v_numoutput <= (ASYNC_THROTTLE / 3))) {
vp->v_flag &= ~VTHROTTLED;
wakeup((caddr_t)&vp->v_numoutput);
}
- if (zero_offset)
- cluster_zero(upl, zero_offset, PAGE_SIZE - (zero_offset & PAGE_MASK), real_bp);
+ if (iostate) {
+ /*
+ * someone has issued multiple I/Os asynchrounsly
+ * and is waiting for them to complete (streaming)
+ */
+ if (error && iostate->io_error == 0)
+ iostate->io_error = error;
+ iostate->io_completed += total_size;
+
+ if (iostate->io_wanted) {
+ /*
+ * someone is waiting for the state of
+ * this io stream to change
+ */
+ iostate->io_wanted = 0;
+ wakeup((caddr_t)&iostate->io_wanted);
+ }
+ }
if ((b_flags & B_NEED_IODONE) && real_bp) {
if (error) {
real_bp->b_flags |= B_ERROR;
error = EIO;
if (b_flags & B_COMMIT_UPL) {
- pg_offset = upl_offset & PAGE_MASK;
+ pg_offset = upl_offset & PAGE_MASK;
commit_size = (((pg_offset + total_size) + (PAGE_SIZE - 1)) / PAGE_SIZE) * PAGE_SIZE;
- if (error || (b_flags & B_NOCACHE)) {
+ if (error || (b_flags & B_NOCACHE) || ((b_flags & B_PHYS) && !(b_flags & B_READ))) {
int upl_abort_code;
- if ((b_flags & B_PAGEOUT) && (error != ENXIO)) /* transient error */
+ if (b_flags & B_PHYS)
+ upl_abort_code = UPL_ABORT_FREE_ON_EMPTY;
+ else if ((b_flags & B_PAGEOUT) && (error != ENXIO)) /* transient error */
upl_abort_code = UPL_ABORT_FREE_ON_EMPTY;
else if (b_flags & B_PGIN)
upl_abort_code = UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR;
} else {
int upl_commit_flags = UPL_COMMIT_FREE_ON_EMPTY;
- if ( !(b_flags & B_PAGEOUT))
+ if (b_flags & B_PHYS)
+ upl_commit_flags |= UPL_COMMIT_SET_DIRTY;
+ else if ( !(b_flags & B_PAGEOUT))
upl_commit_flags |= UPL_COMMIT_CLEAR_DIRTY;
if (b_flags & B_AGE)
upl_commit_flags |= UPL_COMMIT_INACTIVATE;
}
static int
-cluster_io(vp, upl, upl_offset, f_offset, non_rounded_size, devblocksize, flags, real_bp)
+cluster_io(vp, upl, upl_offset, f_offset, non_rounded_size, devblocksize, flags, real_bp, iostate)
struct vnode *vp;
upl_t upl;
vm_offset_t upl_offset;
int devblocksize;
int flags;
struct buf *real_bp;
+ struct clios *iostate;
{
struct buf *cbp;
- struct iovec *iovp;
- u_int size;
+ u_int size;
+ u_int io_size;
int io_flags;
int error = 0;
int retval = 0;
u_int max_vectors;
int priv;
int zero_offset = 0;
+ u_int first_lblkno;
if (flags & CL_READ) {
io_flags = (B_VECTORLIST | B_READ);
}
pl = ubc_upl_pageinfo(upl);
- if (flags & CL_ASYNC)
- io_flags |= (B_CALL | B_ASYNC);
if (flags & CL_AGE)
io_flags |= B_AGE;
if (flags & CL_DUMP)
io_flags |= B_NOCACHE;
if (flags & CL_PAGEIN)
io_flags |= B_PGIN;
+ if (flags & CL_PAGEOUT)
+ io_flags |= B_PAGEOUT;
+ if (flags & CL_COMMIT)
+ io_flags |= B_COMMIT_UPL;
+ if (flags & CL_PRESERVE)
+ io_flags |= B_PHYS;
if (devblocksize)
size = (non_rounded_size + (devblocksize - 1)) & ~(devblocksize - 1);
zero_offset = upl_offset + non_rounded_size;
}
while (size) {
- size_t io_size;
- int vsize;
- int i;
+ int i;
int pl_index;
int pg_resid;
int num_contig;
else
io_size = size;
- if (error = VOP_CMAP(vp, f_offset, io_size, &blkno, &io_size, NULL)) {
+ if (error = VOP_CMAP(vp, f_offset, io_size, &blkno, (size_t *)&io_size, NULL)) {
if (error == EOPNOTSUPP)
panic("VOP_CMAP Unimplemented");
break;
real_bp->b_blkno = blkno;
}
- if (pg_count > 1) {
- if (pg_count > max_vectors) {
- io_size -= (pg_count - max_vectors) * PAGE_SIZE;
+ if (pg_count > max_vectors) {
+ io_size -= (pg_count - max_vectors) * PAGE_SIZE;
- if (io_size < 0) {
- io_size = PAGE_SIZE - pg_offset;
- pg_count = 1;
- } else
- pg_count = max_vectors;
- }
- /*
- * we need to allocate space for the vector list
- */
- if (pg_count > 1) {
- iovp = (struct iovec *)_MALLOC(sizeof(struct iovec) * pg_count,
- M_SEGMENT, M_NOWAIT);
-
- if (iovp == (struct iovec *) 0) {
- /*
- * if the allocation fails, then throttle down to a single page
- */
- io_size = PAGE_SIZE - pg_offset;
- pg_count = 1;
- }
- }
+ if (io_size < 0) {
+ io_size = PAGE_SIZE - pg_offset;
+ pg_count = 1;
+ } else
+ pg_count = max_vectors;
}
/* Throttle the speculative IO */
cbp = alloc_io_buf(vp, priv);
- if (pg_count == 1)
- /*
- * we use the io vector that's reserved in the buffer header
- * this insures we can always issue an I/O even in a low memory
- * condition that prevents the _MALLOC from succeeding... this
- * is necessary to prevent deadlocks with the pager
- */
- iovp = (struct iovec *)(&cbp->b_vects[0]);
-
- cbp->b_vectorlist = (void *)iovp;
- cbp->b_vectorcount = pg_count;
-
- if (flags & CL_DEV_MEMORY) {
-
- iovp->iov_len = io_size;
- iovp->iov_base = (caddr_t)upl_phys_page(pl, 0);
- if (iovp->iov_base == (caddr_t) 0) {
- free_io_buf(cbp);
- error = EINVAL;
- } else
- iovp->iov_base += upl_offset;
- } else {
-
- for (i = 0, vsize = io_size; i < pg_count; i++, iovp++) {
- int psize;
-
- psize = PAGE_SIZE - pg_offset;
-
- if (psize > vsize)
- psize = vsize;
-
- iovp->iov_len = psize;
- iovp->iov_base = (caddr_t)upl_phys_page(pl, pl_index + i);
-
- if (iovp->iov_base == (caddr_t) 0) {
- if (pg_count > 1)
- _FREE(cbp->b_vectorlist, M_SEGMENT);
- free_io_buf(cbp);
-
- error = EINVAL;
- break;
- }
- iovp->iov_base += pg_offset;
- pg_offset = 0;
-
- if (flags & CL_PAGEOUT) {
+ if (flags & CL_PAGEOUT) {
+ for (i = 0; i < pg_count; i++) {
int s;
struct buf *bp;
}
splx(s);
}
- vsize -= psize;
- }
}
- if (error)
- break;
-
- if (flags & CL_ASYNC)
- cbp->b_iodone = (void *)cluster_iodone;
+ if (flags & CL_ASYNC) {
+ cbp->b_flags |= (B_CALL | B_ASYNC);
+ cbp->b_iodone = (void *)cluster_iodone;
+ }
cbp->b_flags |= io_flags;
cbp->b_lblkno = lblkno;
cbp->b_uploffset = upl_offset;
cbp->b_trans_next = (struct buf *)0;
+ if (cbp->b_iostate = (void *)iostate)
+ /*
+ * caller wants to track the state of this
+ * io... bump the amount issued against this stream
+ */
+ iostate->io_issued += io_size;
+
if (flags & CL_READ)
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 26)) | DBG_FUNC_NONE,
cbp->b_lblkno, cbp->b_blkno, upl_offset, io_size, 0);
* then go ahead and issue the I/O
*/
start_io:
- if (flags & CL_COMMIT)
- cbp_head->b_flags |= B_COMMIT_UPL;
- if (flags & CL_PAGEOUT)
- cbp_head->b_flags |= B_PAGEOUT;
- if (flags & CL_PAGEIN)
- cbp_head->b_flags |= B_PGIN;
-
if (real_bp) {
cbp_head->b_flags |= B_NEED_IODONE;
cbp_head->b_real_bp = real_bp;
if (error) {
int abort_size;
+ io_size = 0;
+
for (cbp = cbp_head; cbp;) {
struct buf * cbp_next;
- if (cbp->b_vectorcount > 1)
- _FREE(cbp->b_vectorlist, M_SEGMENT);
upl_offset -= cbp->b_bcount;
size += cbp->b_bcount;
+ io_size += cbp->b_bcount;
cbp_next = cbp->b_trans_next;
free_io_buf(cbp);
cbp = cbp_next;
}
+ if (iostate) {
+ /*
+ * update the error condition for this stream
+ * since we never really issued the io
+ * just go ahead and adjust it back
+ */
+ if (iostate->io_error == 0)
+ iostate->io_error = error;
+ iostate->io_issued -= io_size;
+
+ if (iostate->io_wanted) {
+ /*
+ * someone is waiting for the state of
+ * this io stream to change
+ */
+ iostate->io_wanted = 0;
+ wakeup((caddr_t)&iostate->io_wanted);
+ }
+ }
pg_offset = upl_offset & PAGE_MASK;
abort_size = ((size + pg_offset + (PAGE_SIZE - 1)) / PAGE_SIZE) * PAGE_SIZE;
if (flags & CL_COMMIT) {
int upl_abort_code;
- if ((flags & CL_PAGEOUT) && (error != ENXIO)) /* transient error */
+ if (flags & CL_PRESERVE)
+ upl_abort_code = UPL_ABORT_FREE_ON_EMPTY;
+ else if ((flags & CL_PAGEOUT) && (error != ENXIO)) /* transient error */
upl_abort_code = UPL_ABORT_FREE_ON_EMPTY;
else if (flags & CL_PAGEIN)
- upl_abort_code = UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR;
+ upl_abort_code = UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR;
else
upl_abort_code = UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_DUMP_PAGES;
}
return (cluster_io(vp, upl, upl_offset, f_offset, io_size, devblocksize,
- local_flags, (struct buf *)0));
+ local_flags, (struct buf *)0, (struct clios *)0));
}
int
size - (upl_offset + rounded_size), UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR);
retval = cluster_io(vp, upl, upl_offset, f_offset, io_size, devblocksize,
- local_flags | CL_READ | CL_PAGEIN, (struct buf *)0);
+ local_flags | CL_READ | CL_PAGEIN, (struct buf *)0, (struct clios *)0);
if (retval == 0) {
int b_lblkno;
f_offset = ubc_blktooff(bp->b_vp, bp->b_lblkno);
- return (cluster_io(bp->b_vp, bp->b_pagelist, 0, f_offset, bp->b_bcount, 0, flags, bp));
+ return (cluster_io(bp->b_vp, bp->b_pagelist, 0, f_offset, bp->b_bcount, 0, flags, bp, (struct clios *)0));
}
int
int retval = 0;
- if ((!uio) || (uio->uio_segflg != UIO_USERSPACE) || (!(vp->v_flag & VNOCACHE_DATA)))
+ if ( (!(vp->v_flag & VNOCACHE_DATA)) || (!uio) || (uio->uio_segflg != UIO_USERSPACE))
{
retval = cluster_write_x(vp, uio, oldEOF, newEOF, headOff, tailOff, devblocksize, flags);
return(retval);
if (upl_flags & UPL_PHYS_CONTIG)
{
- /*
- * since the interface to the IOKit below us uses physical block #'s and
- * block counts to specify the I/O, we can't handle anything that isn't
- * devblocksize aligned
- */
- if ((uio->uio_offset & (devblocksize - 1)) || (uio->uio_resid & (devblocksize - 1)))
- return(EINVAL);
-
if (flags & IO_HEADZEROFILL)
{
flags &= ~IO_HEADZEROFILL;
return(retval);
}
- retval = cluster_phys_write(vp, uio);
+ retval = cluster_phys_write(vp, uio, newEOF, devblocksize, flags);
if (uio->uio_resid == 0 && (flags & IO_TAILZEROFILL))
{
return(retval);
}
+
static int
cluster_nocopy_write(vp, uio, newEOF, devblocksize, flags)
struct vnode *vp;
vm_offset_t upl_offset;
off_t max_io_size;
int io_size;
+ int io_flag;
int upl_size;
int upl_needed_size;
int pages_in_pl;
kern_return_t kret;
struct iovec *iov;
int i;
+ int first = 1;
int force_data_sync;
int error = 0;
+ struct clios iostate;
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 75)) | DBG_FUNC_START,
(int)uio->uio_offset, (int)uio->uio_resid,
* -- the resid is a page multiple
* -- the resid will not exceed iov_len
*/
+ cluster_try_push(vp, newEOF, 0, 1);
+
+ iostate.io_completed = 0;
+ iostate.io_issued = 0;
+ iostate.io_error = 0;
+ iostate.io_wanted = 0;
iov = uio->uio_iov;
while (uio->uio_resid && uio->uio_offset < newEOF && error == 0) {
- io_size = uio->uio_resid;
-
- if (io_size > (MAX_UPL_TRANSFER * PAGE_SIZE))
- io_size = MAX_UPL_TRANSFER * PAGE_SIZE;
+ io_size = uio->uio_resid;
- upl_offset = (vm_offset_t)iov->iov_base & PAGE_MASK_64;
- upl_needed_size = (upl_offset + io_size + (PAGE_SIZE -1)) & ~PAGE_MASK;
+ if (io_size > (MAX_UPL_TRANSFER * PAGE_SIZE))
+ io_size = MAX_UPL_TRANSFER * PAGE_SIZE;
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 76)) | DBG_FUNC_START,
- (int)upl_offset, upl_needed_size, (int)iov->iov_base, io_size, 0);
-
- for (force_data_sync = 0; force_data_sync < 3; force_data_sync++)
- {
- pages_in_pl = 0;
- upl_size = upl_needed_size;
- upl_flags = UPL_FILE_IO | UPL_COPYOUT_FROM | UPL_NO_SYNC |
- UPL_CLEAN_IN_PLACE | UPL_SET_INTERNAL;
-
- kret = vm_map_get_upl(current_map(),
- (vm_offset_t)iov->iov_base & ~PAGE_MASK,
- &upl_size,
- &upl,
- NULL,
- &pages_in_pl,
- &upl_flags,
- force_data_sync);
+ if (first) {
+ if (io_size > (MAX_UPL_TRANSFER * PAGE_SIZE) / 4)
+ io_size = (MAX_UPL_TRANSFER * PAGE_SIZE) / 8;
+ first = 0;
+ }
+ upl_offset = (vm_offset_t)iov->iov_base & PAGE_MASK_64;
+ upl_needed_size = (upl_offset + io_size + (PAGE_SIZE -1)) & ~PAGE_MASK;
+
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 76)) | DBG_FUNC_START,
+ (int)upl_offset, upl_needed_size, (int)iov->iov_base, io_size, 0);
+
+ for (force_data_sync = 0; force_data_sync < 3; force_data_sync++) {
+ pages_in_pl = 0;
+ upl_size = upl_needed_size;
+ upl_flags = UPL_FILE_IO | UPL_COPYOUT_FROM | UPL_NO_SYNC |
+ UPL_CLEAN_IN_PLACE | UPL_SET_INTERNAL;
+
+ kret = vm_map_get_upl(current_map(),
+ (vm_offset_t)iov->iov_base & ~PAGE_MASK,
+ &upl_size,
+ &upl,
+ NULL,
+ &pages_in_pl,
+ &upl_flags,
+ force_data_sync);
+
+ if (kret != KERN_SUCCESS) {
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 76)) | DBG_FUNC_END,
+ 0, 0, 0, kret, 0);
- if (kret != KERN_SUCCESS)
- {
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 76)) | DBG_FUNC_END,
- 0, 0, 0, kret, 0);
+ /*
+ * cluster_nocopy_write: failed to get pagelist
+ *
+ * we may have already spun some portion of this request
+ * off as async requests... we need to wait for the I/O
+ * to complete before returning
+ */
+ goto wait_for_writes;
+ }
+ pl = UPL_GET_INTERNAL_PAGE_LIST(upl);
+ pages_in_pl = upl_size / PAGE_SIZE;
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 75)) | DBG_FUNC_END,
- (int)uio->uio_offset, (int)uio->uio_resid, kret, 1, 0);
+ for (i = 0; i < pages_in_pl; i++) {
+ if (!upl_valid_page(pl, i))
+ break;
+ }
+ if (i == pages_in_pl)
+ break;
- /* cluster_nocopy_write: failed to get pagelist */
- /* do not return kret here */
- return(0);
+ /*
+ * didn't get all the pages back that we
+ * needed... release this upl and try again
+ */
+ ubc_upl_abort_range(upl, (upl_offset & ~PAGE_MASK), upl_size,
+ UPL_ABORT_FREE_ON_EMPTY);
}
+ if (force_data_sync >= 3) {
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 76)) | DBG_FUNC_END,
+ i, pages_in_pl, upl_size, kret, 0);
- pl = UPL_GET_INTERNAL_PAGE_LIST(upl);
- pages_in_pl = upl_size / PAGE_SIZE;
-
- for(i=0; i < pages_in_pl; i++)
- {
- if (!upl_valid_page(pl, i))
- break;
+ /*
+ * for some reason, we couldn't acquire a hold on all
+ * the pages needed in the user's address space
+ *
+ * we may have already spun some portion of this request
+ * off as async requests... we need to wait for the I/O
+ * to complete before returning
+ */
+ goto wait_for_writes;
}
- if (i == pages_in_pl)
- break;
-
- ubc_upl_abort_range(upl, (upl_offset & ~PAGE_MASK), upl_size,
- UPL_ABORT_FREE_ON_EMPTY);
- }
-
- if (force_data_sync >= 3)
- {
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 76)) | DBG_FUNC_END,
- i, pages_in_pl, upl_size, kret, 0);
-
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 75)) | DBG_FUNC_END,
- (int)uio->uio_offset, (int)uio->uio_resid, kret, 2, 0);
- return(0);
- }
-
- /*
- * Consider the possibility that upl_size wasn't satisfied.
- */
- if (upl_size != upl_needed_size)
- io_size = (upl_size - (int)upl_offset) & ~PAGE_MASK;
-
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 76)) | DBG_FUNC_END,
- (int)upl_offset, upl_size, (int)iov->iov_base, io_size, 0);
+ /*
+ * Consider the possibility that upl_size wasn't satisfied.
+ */
+ if (upl_size != upl_needed_size)
+ io_size = (upl_size - (int)upl_offset) & ~PAGE_MASK;
- if (io_size == 0)
- {
- ubc_upl_abort_range(upl, (upl_offset & ~PAGE_MASK), upl_size,
- UPL_ABORT_FREE_ON_EMPTY);
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 75)) | DBG_FUNC_END,
- (int)uio->uio_offset, uio->uio_resid, 0, 3, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 76)) | DBG_FUNC_END,
+ (int)upl_offset, upl_size, (int)iov->iov_base, io_size, 0);
- return(0);
- }
+ if (io_size == 0) {
+ ubc_upl_abort_range(upl, (upl_offset & ~PAGE_MASK), upl_size,
+ UPL_ABORT_FREE_ON_EMPTY);
- /*
- * Now look for pages already in the cache
- * and throw them away.
- */
+ /*
+ * we may have already spun some portion of this request
+ * off as async requests... we need to wait for the I/O
+ * to complete before returning
+ */
+ goto wait_for_writes;
+ }
+ /*
+ * Now look for pages already in the cache
+ * and throw them away.
+ */
- upl_f_offset = uio->uio_offset; /* this is page aligned in the file */
- max_io_size = io_size;
+ upl_f_offset = uio->uio_offset; /* this is page aligned in the file */
+ max_io_size = io_size;
- while (max_io_size) {
+ while (max_io_size) {
+ /*
+ * Flag UPL_POP_DUMP says if the page is found
+ * in the page cache it must be thrown away.
+ */
+ ubc_page_op(vp,
+ upl_f_offset,
+ UPL_POP_SET | UPL_POP_BUSY | UPL_POP_DUMP,
+ 0, 0);
+ max_io_size -= PAGE_SIZE_64;
+ upl_f_offset += PAGE_SIZE_64;
+ }
+ /*
+ * we want push out these writes asynchronously so that we can overlap
+ * the preparation of the next I/O
+ * if there are already too many outstanding writes
+ * wait until some complete before issuing the next
+ */
+ while ((iostate.io_issued - iostate.io_completed) > (2 * MAX_UPL_TRANSFER * PAGE_SIZE)) {
+ iostate.io_wanted = 1;
+ tsleep((caddr_t)&iostate.io_wanted, PRIBIO + 1, "cluster_nocopy_write", 0);
+ }
+ if (iostate.io_error) {
+ /*
+ * one of the earlier writes we issued ran into a hard error
+ * don't issue any more writes, cleanup the UPL
+ * that was just created but not used, then
+ * go wait for all writes that are part of this stream
+ * to complete before returning the error to the caller
+ */
+ ubc_upl_abort_range(upl, (upl_offset & ~PAGE_MASK), upl_size,
+ UPL_ABORT_FREE_ON_EMPTY);
- /*
- * Flag UPL_POP_DUMP says if the page is found
- * in the page cache it must be thrown away.
- */
- ubc_page_op(vp,
- upl_f_offset,
- UPL_POP_SET | UPL_POP_BUSY | UPL_POP_DUMP,
- 0, 0);
- max_io_size -= PAGE_SIZE;
- upl_f_offset += PAGE_SIZE;
- }
+ goto wait_for_writes;
+ }
+ io_flag = CL_ASYNC | CL_PRESERVE | CL_COMMIT;
- /*
- * issue a synchronous write to cluster_io
- */
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 77)) | DBG_FUNC_START,
+ (int)upl_offset, (int)uio->uio_offset, io_size, io_flag, 0);
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 77)) | DBG_FUNC_START,
- (int)upl_offset, (int)uio->uio_offset, io_size, 0, 0);
+ error = cluster_io(vp, upl, upl_offset, uio->uio_offset,
+ io_size, devblocksize, io_flag, (struct buf *)0, &iostate);
- error = cluster_io(vp, upl, upl_offset, uio->uio_offset,
- io_size, devblocksize, 0, (struct buf *)0);
+ iov->iov_len -= io_size;
+ iov->iov_base += io_size;
+ uio->uio_resid -= io_size;
+ uio->uio_offset += io_size;
- if (error == 0) {
- /*
- * The cluster_io write completed successfully,
- * update the uio structure.
- */
- iov->iov_base += io_size;
- iov->iov_len -= io_size;
- uio->uio_resid -= io_size;
- uio->uio_offset += io_size;
- }
- /*
- * always 'commit' the I/O via the abort primitive whether the I/O
- * succeeded cleanly or not... this is necessary to insure that
- * we preserve the state of the DIRTY flag on the pages used to
- * provide the data for the I/O... the state of this flag SHOULD
- * NOT be changed by a write
- */
- ubc_upl_abort_range(upl, (upl_offset & ~PAGE_MASK), upl_size,
- UPL_ABORT_FREE_ON_EMPTY);
-
-
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 77)) | DBG_FUNC_END,
- (int)upl_offset, (int)uio->uio_offset, (int)uio->uio_resid, error, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 77)) | DBG_FUNC_END,
+ (int)upl_offset, (int)uio->uio_offset, (int)uio->uio_resid, error, 0);
} /* end while */
+wait_for_writes:
+ /*
+ * make sure all async writes issued as part of this stream
+ * have completed before we return
+ */
+ while (iostate.io_issued != iostate.io_completed) {
+ iostate.io_wanted = 1;
+ tsleep((caddr_t)&iostate.io_wanted, PRIBIO + 1, "cluster_nocopy_write", 0);
+ }
+ if (iostate.io_error)
+ error = iostate.io_error;
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 75)) | DBG_FUNC_END,
(int)uio->uio_offset, (int)uio->uio_resid, error, 4, 0);
return (error);
}
+
static int
-cluster_phys_write(vp, uio)
+cluster_phys_write(vp, uio, newEOF, devblocksize, flags)
struct vnode *vp;
struct uio *uio;
+ off_t newEOF;
+ int devblocksize;
+ int flags;
{
+ upl_page_info_t *pl;
+ addr64_t src_paddr;
upl_t upl;
vm_offset_t upl_offset;
+ int tail_size;
int io_size;
int upl_size;
int upl_needed_size;
* -- the resid will not exceed iov_len
* -- the vector target address is physcially contiguous
*/
+ cluster_try_push(vp, newEOF, 0, 1);
iov = uio->uio_iov;
io_size = iov->iov_len;
(vm_offset_t)iov->iov_base & ~PAGE_MASK,
&upl_size, &upl, NULL, &pages_in_pl, &upl_flags, 0);
- if (kret != KERN_SUCCESS)
- {
- /* cluster_phys_write: failed to get pagelist */
- /* note: return kret here */
+ if (kret != KERN_SUCCESS) {
+ /*
+ * cluster_phys_write: failed to get pagelist
+ * note: return kret here
+ */
return(EINVAL);
- }
-
+ }
/*
* Consider the possibility that upl_size wasn't satisfied.
* This is a failure in the physical memory case.
*/
- if (upl_size < upl_needed_size)
- {
- kernel_upl_abort_range(upl, 0, upl_size, UPL_ABORT_FREE_ON_EMPTY);
- return(EINVAL);
- }
+ if (upl_size < upl_needed_size) {
+ kernel_upl_abort_range(upl, 0, upl_size, UPL_ABORT_FREE_ON_EMPTY);
+ return(EINVAL);
+ }
+ pl = ubc_upl_pageinfo(upl);
- /*
- * issue a synchronous write to cluster_io
- */
+ src_paddr = (((addr64_t)(int)upl_phys_page(pl, 0)) << 12) + ((addr64_t)iov->iov_base & PAGE_MASK);
+
+ while (((uio->uio_offset & (devblocksize - 1)) || io_size < devblocksize) && io_size) {
+ int head_size;
+
+ head_size = devblocksize - (int)(uio->uio_offset & (devblocksize - 1));
+
+ if (head_size > io_size)
+ head_size = io_size;
+
+ error = cluster_align_phys_io(vp, uio, src_paddr, head_size, devblocksize, 0);
- error = cluster_io(vp, upl, upl_offset, uio->uio_offset,
- io_size, 0, CL_DEV_MEMORY, (struct buf *)0);
+ if (error) {
+ ubc_upl_abort_range(upl, 0, upl_size, UPL_ABORT_FREE_ON_EMPTY);
+
+ return(EINVAL);
+ }
+ upl_offset += head_size;
+ src_paddr += head_size;
+ io_size -= head_size;
+ }
+ tail_size = io_size & (devblocksize - 1);
+ io_size -= tail_size;
+ if (io_size) {
+ /*
+ * issue a synchronous write to cluster_io
+ */
+ error = cluster_io(vp, upl, upl_offset, uio->uio_offset,
+ io_size, 0, CL_DEV_MEMORY, (struct buf *)0, (struct clios *)0);
+ }
if (error == 0) {
- /*
- * The cluster_io write completed successfully,
- * update the uio structure and commit.
- */
-
- ubc_upl_commit_range(upl, 0, upl_size, UPL_COMMIT_FREE_ON_EMPTY);
-
- iov->iov_base += io_size;
- iov->iov_len -= io_size;
- uio->uio_resid -= io_size;
- uio->uio_offset += io_size;
+ /*
+ * The cluster_io write completed successfully,
+ * update the uio structure
+ */
+ uio->uio_resid -= io_size;
+ iov->iov_len -= io_size;
+ iov->iov_base += io_size;
+ uio->uio_offset += io_size;
+ src_paddr += io_size;
+
+ if (tail_size)
+ error = cluster_align_phys_io(vp, uio, src_paddr, tail_size, devblocksize, 0);
}
- else
- ubc_upl_abort_range(upl, 0, upl_size, UPL_ABORT_FREE_ON_EMPTY);
+ /*
+ * just release our hold on the physically contiguous
+ * region without changing any state
+ */
+ ubc_upl_abort_range(upl, 0, upl_size, UPL_ABORT_FREE_ON_EMPTY);
return (error);
}
+
static int
cluster_write_x(vp, uio, oldEOF, newEOF, headOff, tailOff, devblocksize, flags)
struct vnode *vp;
read_size = newEOF - upl_f_offset;
retval = cluster_io(vp, upl, 0, upl_f_offset, read_size, devblocksize,
- CL_READ, (struct buf *)0);
+ CL_READ, (struct buf *)0, (struct clios *)0);
if (retval) {
/*
* we had an error during the read which causes us to abort
read_size = newEOF - (upl_f_offset + upl_offset);
retval = cluster_io(vp, upl, upl_offset, upl_f_offset + upl_offset, read_size, devblocksize,
- CL_READ, (struct buf *)0);
+ CL_READ, (struct buf *)0, (struct clios *)0);
if (retval) {
/*
* we had an error during the read which causes us to abort
else
can_delay = 1;
- if (cluster_try_push(vp, newEOF, can_delay, 0) == 0) {
+ if (cluster_try_push(vp, newEOF, 0, 0) == 0) {
vp->v_flag |= VHASDIRTY;
goto delay_io;
}
if (last_blkno > vp->v_lastw)
vp->v_lastw = last_blkno;
- ubc_upl_commit_range(upl, 0, upl_size, UPL_COMMIT_SET_DIRTY | UPL_COMMIT_INACTIVATE | UPL_COMMIT_FREE_ON_EMPTY);
+ ubc_upl_commit_range(upl, 0, upl_size, UPL_COMMIT_SET_DIRTY | UPL_COMMIT_INACTIVATE | UPL_COMMIT_FREE_ON_EMPTY);
continue;
issue_io:
/*
tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "cluster_write", 0);
}
retval = cluster_io(vp, upl, 0, upl_f_offset, io_size, devblocksize,
- io_flags, (struct buf *)0);
+ io_flags, (struct buf *)0, (struct clios *)0);
}
}
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 40)) | DBG_FUNC_END,
if (upl_flags & UPL_PHYS_CONTIG)
{
- retval = cluster_phys_read(vp, uio, filesize);
+ retval = cluster_phys_read(vp, uio, filesize, devblocksize, flags);
}
else if (uio->uio_resid < 4 * PAGE_SIZE)
{
return(retval);
}
+
static int
cluster_read_x(vp, uio, filesize, devblocksize, flags)
struct vnode *vp;
(int)uio->uio_offset, io_size, uio->uio_resid, 0, 0);
while (io_size && retval == 0) {
- int xsize;
- vm_offset_t paddr;
+ int xsize;
+ ppnum_t paddr;
if (ubc_page_op(vp,
upl_f_offset,
if (xsize > io_size)
xsize = io_size;
- retval = uiomove((caddr_t)(paddr + start_offset), xsize, uio);
+ retval = uiomove64((addr64_t)(((addr64_t)paddr << 12) + start_offset), xsize, uio);
ubc_page_op(vp, upl_f_offset,
UPL_POP_CLR | UPL_POP_BUSY, 0, 0);
*/
error = cluster_io(vp, upl, upl_offset, upl_f_offset + upl_offset,
- io_size, devblocksize, CL_READ, (struct buf *)0);
+ io_size, devblocksize, CL_READ, (struct buf *)0, (struct clios *)0);
}
if (error == 0) {
/*
while (val_size && retval == 0) {
int csize;
int i;
- caddr_t paddr;
+ addr64_t paddr;
i = offset / PAGE_SIZE;
csize = min(PAGE_SIZE - start_offset, val_size);
- paddr = (caddr_t)upl_phys_page(pl, i) + start_offset;
+ paddr = ((addr64_t)upl_phys_page(pl, i) << 12) + start_offset;
- retval = uiomove(paddr, csize, uio);
+ retval = uiomove64(paddr, csize, uio);
val_size -= csize;
offset += csize;
return (retval);
}
+
static int
cluster_nocopy_read(vp, uio, filesize, devblocksize, flags)
struct vnode *vp;
int upl_size;
int upl_needed_size;
int pages_in_pl;
- vm_offset_t paddr;
+ ppnum_t paddr;
int upl_flags;
kern_return_t kret;
int segflg;
struct iovec *iov;
int i;
int force_data_sync;
- int error = 0;
int retval = 0;
+ int first = 1;
+ struct clios iostate;
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 70)) | DBG_FUNC_START,
(int)uio->uio_offset, uio->uio_resid, (int)filesize, devblocksize, 0);
* -- the resid will not exceed iov_len
*/
+ iostate.io_completed = 0;
+ iostate.io_issued = 0;
+ iostate.io_error = 0;
+ iostate.io_wanted = 0;
+
iov = uio->uio_iov;
+
while (uio->uio_resid && uio->uio_offset < filesize && retval == 0) {
- max_io_size = filesize - uio->uio_offset;
+ max_io_size = filesize - uio->uio_offset;
- if (max_io_size < (off_t)((unsigned int)uio->uio_resid))
- io_size = max_io_size;
- else
- io_size = uio->uio_resid;
+ if (max_io_size < (off_t)((unsigned int)uio->uio_resid))
+ io_size = max_io_size;
+ else
+ io_size = uio->uio_resid;
- /*
- * We don't come into this routine unless
- * UIO_USERSPACE is set.
- */
- segflg = uio->uio_segflg;
+ /*
+ * We don't come into this routine unless
+ * UIO_USERSPACE is set.
+ */
+ segflg = uio->uio_segflg;
- uio->uio_segflg = UIO_PHYS_USERSPACE;
+ uio->uio_segflg = UIO_PHYS_USERSPACE;
- /*
- * First look for pages already in the cache
- * and move them to user space.
- */
- while (io_size && (retval == 0)) {
- upl_f_offset = uio->uio_offset;
+ /*
+ * First look for pages already in the cache
+ * and move them to user space.
+ */
+ while (io_size && (retval == 0)) {
+ upl_f_offset = uio->uio_offset;
- /*
- * If this call fails, it means the page is not
- * in the page cache.
- */
- if (ubc_page_op(vp, upl_f_offset,
- UPL_POP_SET | UPL_POP_BUSY, &paddr, 0) != KERN_SUCCESS)
- break;
+ /*
+ * If this call fails, it means the page is not
+ * in the page cache.
+ */
+ if (ubc_page_op(vp, upl_f_offset,
+ UPL_POP_SET | UPL_POP_BUSY, &paddr, 0) != KERN_SUCCESS)
+ break;
- retval = uiomove((caddr_t)(paddr), PAGE_SIZE, uio);
+ retval = uiomove64((addr64_t)paddr << 12, PAGE_SIZE, uio);
- ubc_page_op(vp, upl_f_offset,
- UPL_POP_CLR | UPL_POP_BUSY, 0, 0);
+ ubc_page_op(vp, upl_f_offset,
+ UPL_POP_CLR | UPL_POP_BUSY, 0, 0);
- io_size -= PAGE_SIZE;
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 71)) | DBG_FUNC_NONE,
- (int)uio->uio_offset, io_size, uio->uio_resid, 0, 0);
- }
-
- uio->uio_segflg = segflg;
+ io_size -= PAGE_SIZE;
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 71)) | DBG_FUNC_NONE,
+ (int)uio->uio_offset, io_size, uio->uio_resid, 0, 0);
+ }
+ uio->uio_segflg = segflg;
- if (retval)
- {
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 70)) | DBG_FUNC_END,
- (int)uio->uio_offset, uio->uio_resid, 2, retval, 0);
- return(retval);
- }
-
- /* If we are already finished with this read, then return */
- if (io_size == 0)
- {
-
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 70)) | DBG_FUNC_END,
- (int)uio->uio_offset, uio->uio_resid, 3, io_size, 0);
- return(0);
- }
-
- max_io_size = io_size;
- if (max_io_size > (MAX_UPL_TRANSFER * PAGE_SIZE))
- max_io_size = MAX_UPL_TRANSFER * PAGE_SIZE;
-
- start_upl_f_offset = uio->uio_offset; /* this is page aligned in the file */
- upl_f_offset = start_upl_f_offset;
- io_size = 0;
-
- while(io_size < max_io_size)
- {
-
- if(ubc_page_op(vp, upl_f_offset,
- UPL_POP_SET | UPL_POP_BUSY, &paddr, 0) == KERN_SUCCESS)
- {
- ubc_page_op(vp, upl_f_offset,
- UPL_POP_CLR | UPL_POP_BUSY, 0, 0);
- break;
- }
-
- /*
- * Build up the io request parameters.
- */
-
- io_size += PAGE_SIZE;
- upl_f_offset += PAGE_SIZE;
+ if (retval) {
+ /*
+ * we may have already spun some portion of this request
+ * off as async requests... we need to wait for the I/O
+ * to complete before returning
+ */
+ goto wait_for_reads;
}
+ /*
+ * If we are already finished with this read, then return
+ */
+ if (io_size == 0) {
+ /*
+ * we may have already spun some portion of this request
+ * off as async requests... we need to wait for the I/O
+ * to complete before returning
+ */
+ goto wait_for_reads;
+ }
+ max_io_size = io_size;
+
+ if (max_io_size > (MAX_UPL_TRANSFER * PAGE_SIZE))
+ max_io_size = MAX_UPL_TRANSFER * PAGE_SIZE;
+ if (first) {
+ if (max_io_size > (MAX_UPL_TRANSFER * PAGE_SIZE) / 4)
+ max_io_size = (MAX_UPL_TRANSFER * PAGE_SIZE) / 8;
+ first = 0;
+ }
+ start_upl_f_offset = uio->uio_offset; /* this is page aligned in the file */
+ upl_f_offset = start_upl_f_offset;
+ io_size = 0;
+
+ while (io_size < max_io_size) {
+ if (ubc_page_op(vp, upl_f_offset,
+ UPL_POP_SET | UPL_POP_BUSY, &paddr, 0) == KERN_SUCCESS) {
+ ubc_page_op(vp, upl_f_offset,
+ UPL_POP_CLR | UPL_POP_BUSY, 0, 0);
+ break;
+ }
+ /*
+ * Build up the io request parameters.
+ */
+ io_size += PAGE_SIZE_64;
+ upl_f_offset += PAGE_SIZE_64;
+ }
+ if (io_size == 0)
+ /*
+ * we may have already spun some portion of this request
+ * off as async requests... we need to wait for the I/O
+ * to complete before returning
+ */
+ goto wait_for_reads;
- if (io_size == 0)
- return(retval);
-
- upl_offset = (vm_offset_t)iov->iov_base & PAGE_MASK_64;
- upl_needed_size = (upl_offset + io_size + (PAGE_SIZE -1)) & ~PAGE_MASK;
+ upl_offset = (vm_offset_t)iov->iov_base & PAGE_MASK_64;
+ upl_needed_size = (upl_offset + io_size + (PAGE_SIZE -1)) & ~PAGE_MASK;
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 72)) | DBG_FUNC_START,
- (int)upl_offset, upl_needed_size, (int)iov->iov_base, io_size, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 72)) | DBG_FUNC_START,
+ (int)upl_offset, upl_needed_size, (int)iov->iov_base, io_size, 0);
- for (force_data_sync = 0; force_data_sync < 3; force_data_sync++)
- {
- pages_in_pl = 0;
- upl_size = upl_needed_size;
- upl_flags = UPL_FILE_IO | UPL_NO_SYNC | UPL_CLEAN_IN_PLACE | UPL_SET_INTERNAL;
+ for (force_data_sync = 0; force_data_sync < 3; force_data_sync++) {
+ pages_in_pl = 0;
+ upl_size = upl_needed_size;
+ upl_flags = UPL_FILE_IO | UPL_NO_SYNC | UPL_CLEAN_IN_PLACE | UPL_SET_INTERNAL;
- kret = vm_map_get_upl(current_map(),
- (vm_offset_t)iov->iov_base & ~PAGE_MASK,
- &upl_size, &upl, NULL, &pages_in_pl, &upl_flags, force_data_sync);
+ kret = vm_map_get_upl(current_map(),
+ (vm_offset_t)iov->iov_base & ~PAGE_MASK,
+ &upl_size, &upl, NULL, &pages_in_pl, &upl_flags, force_data_sync);
- if (kret != KERN_SUCCESS)
- {
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 72)) | DBG_FUNC_END,
- (int)upl_offset, upl_size, io_size, kret, 0);
+ if (kret != KERN_SUCCESS) {
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 72)) | DBG_FUNC_END,
+ (int)upl_offset, upl_size, io_size, kret, 0);
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 70)) | DBG_FUNC_END,
- (int)uio->uio_offset, uio->uio_resid, 4, retval, 0);
-
- /* cluster_nocopy_read: failed to get pagelist */
- /* do not return kret here */
- return(retval);
- }
+ /*
+ * cluster_nocopy_read: failed to get pagelist
+ *
+ * we may have already spun some portion of this request
+ * off as async requests... we need to wait for the I/O
+ * to complete before returning
+ */
+ goto wait_for_reads;
+ }
+ pages_in_pl = upl_size / PAGE_SIZE;
+ pl = UPL_GET_INTERNAL_PAGE_LIST(upl);
- pages_in_pl = upl_size / PAGE_SIZE;
- pl = UPL_GET_INTERNAL_PAGE_LIST(upl);
+ for (i = 0; i < pages_in_pl; i++) {
+ if (!upl_valid_page(pl, i))
+ break;
+ }
+ if (i == pages_in_pl)
+ break;
- for(i=0; i < pages_in_pl; i++)
- {
- if (!upl_valid_page(pl, i))
- break;
+ ubc_upl_abort_range(upl, (upl_offset & ~PAGE_MASK), upl_size,
+ UPL_ABORT_FREE_ON_EMPTY);
}
- if (i == pages_in_pl)
- break;
-
- ubc_upl_abort_range(upl, (upl_offset & ~PAGE_MASK), upl_size,
- UPL_ABORT_FREE_ON_EMPTY);
- }
-
- if (force_data_sync >= 3)
- {
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 72)) | DBG_FUNC_END,
- (int)upl_offset, upl_size, io_size, kret, 0);
+ if (force_data_sync >= 3) {
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 72)) | DBG_FUNC_END,
+ (int)upl_offset, upl_size, io_size, kret, 0);
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 70)) | DBG_FUNC_END,
- (int)uio->uio_offset, uio->uio_resid, 5, retval, 0);
- return(retval);
- }
- /*
- * Consider the possibility that upl_size wasn't satisfied.
- */
- if (upl_size != upl_needed_size)
- io_size = (upl_size - (int)upl_offset) & ~PAGE_MASK;
-
- if (io_size == 0)
- {
- ubc_upl_abort_range(upl, (upl_offset & ~PAGE_MASK), upl_size,
- UPL_ABORT_FREE_ON_EMPTY);
- return(retval);
- }
-
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 72)) | DBG_FUNC_END,
- (int)upl_offset, upl_size, io_size, kret, 0);
+ goto wait_for_reads;
+ }
+ /*
+ * Consider the possibility that upl_size wasn't satisfied.
+ */
+ if (upl_size != upl_needed_size)
+ io_size = (upl_size - (int)upl_offset) & ~PAGE_MASK;
- /*
- * issue a synchronous read to cluster_io
- */
+ if (io_size == 0) {
+ ubc_upl_abort_range(upl, (upl_offset & ~PAGE_MASK), upl_size,
+ UPL_ABORT_FREE_ON_EMPTY);
+ goto wait_for_reads;
+ }
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 72)) | DBG_FUNC_END,
+ (int)upl_offset, upl_size, io_size, kret, 0);
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 73)) | DBG_FUNC_START,
- (int)upl, (int)upl_offset, (int)start_upl_f_offset, io_size, 0);
+ /*
+ * request asynchronously so that we can overlap
+ * the preparation of the next I/O
+ * if there are already too many outstanding reads
+ * wait until some have completed before issuing the next read
+ */
+ while ((iostate.io_issued - iostate.io_completed) > (2 * MAX_UPL_TRANSFER * PAGE_SIZE)) {
+ iostate.io_wanted = 1;
+ tsleep((caddr_t)&iostate.io_wanted, PRIBIO + 1, "cluster_nocopy_read", 0);
+ }
+ if (iostate.io_error) {
+ /*
+ * one of the earlier reads we issued ran into a hard error
+ * don't issue any more reads, cleanup the UPL
+ * that was just created but not used, then
+ * go wait for any other reads to complete before
+ * returning the error to the caller
+ */
+ ubc_upl_abort_range(upl, (upl_offset & ~PAGE_MASK), upl_size,
+ UPL_ABORT_FREE_ON_EMPTY);
- error = cluster_io(vp, upl, upl_offset, start_upl_f_offset,
- io_size, devblocksize, CL_READ| CL_NOZERO, (struct buf *)0);
+ goto wait_for_reads;
+ }
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 73)) | DBG_FUNC_START,
+ (int)upl, (int)upl_offset, (int)start_upl_f_offset, io_size, 0);
- if (error == 0) {
- /*
- * The cluster_io read completed successfully,
- * update the uio structure and commit.
- */
+ retval = cluster_io(vp, upl, upl_offset, start_upl_f_offset,
+ io_size, devblocksize,
+ CL_PRESERVE | CL_COMMIT | CL_READ | CL_ASYNC | CL_NOZERO,
+ (struct buf *)0, &iostate);
- ubc_upl_commit_range(upl, (upl_offset & ~PAGE_MASK), upl_size,
- UPL_COMMIT_SET_DIRTY | UPL_COMMIT_FREE_ON_EMPTY);
-
- iov->iov_base += io_size;
- iov->iov_len -= io_size;
- uio->uio_resid -= io_size;
- uio->uio_offset += io_size;
- }
- else {
- ubc_upl_abort_range(upl, (upl_offset & ~PAGE_MASK), upl_size,
- UPL_ABORT_FREE_ON_EMPTY);
- }
-
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 73)) | DBG_FUNC_END,
- (int)upl, (int)uio->uio_offset, (int)uio->uio_resid, error, 0);
+ /*
+ * update the uio structure
+ */
+ iov->iov_base += io_size;
+ iov->iov_len -= io_size;
+ uio->uio_resid -= io_size;
+ uio->uio_offset += io_size;
- if (retval == 0)
- retval = error;
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 73)) | DBG_FUNC_END,
+ (int)upl, (int)uio->uio_offset, (int)uio->uio_resid, retval, 0);
} /* end while */
+wait_for_reads:
+ /*
+ * make sure all async reads that are part of this stream
+ * have completed before we return
+ */
+ while (iostate.io_issued != iostate.io_completed) {
+ iostate.io_wanted = 1;
+ tsleep((caddr_t)&iostate.io_wanted, PRIBIO + 1, "cluster_nocopy_read", 0);
+ }
+ if (iostate.io_error)
+ retval = iostate.io_error;
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 70)) | DBG_FUNC_END,
(int)uio->uio_offset, (int)uio->uio_resid, 6, retval, 0);
static int
-cluster_phys_read(vp, uio, filesize)
+cluster_phys_read(vp, uio, filesize, devblocksize, flags)
struct vnode *vp;
struct uio *uio;
off_t filesize;
+ int devblocksize;
+ int flags;
{
+ upl_page_info_t *pl;
upl_t upl;
vm_offset_t upl_offset;
+ addr64_t dst_paddr;
off_t max_size;
int io_size;
+ int tail_size;
int upl_size;
int upl_needed_size;
int pages_in_pl;
int upl_flags;
kern_return_t kret;
struct iovec *iov;
+ struct clios iostate;
int error;
/*
max_size = filesize - uio->uio_offset;
- if (max_size < (off_t)((unsigned int)iov->iov_len))
- io_size = max_size;
+ if (max_size > (off_t)((unsigned int)iov->iov_len))
+ io_size = iov->iov_len;
else
- io_size = iov->iov_len;
+ io_size = max_size;
upl_offset = (vm_offset_t)iov->iov_base & PAGE_MASK_64;
upl_needed_size = upl_offset + io_size;
+ error = 0;
pages_in_pl = 0;
upl_size = upl_needed_size;
upl_flags = UPL_FILE_IO | UPL_NO_SYNC | UPL_CLEAN_IN_PLACE | UPL_SET_INTERNAL;
(vm_offset_t)iov->iov_base & ~PAGE_MASK,
&upl_size, &upl, NULL, &pages_in_pl, &upl_flags, 0);
- if (kret != KERN_SUCCESS)
- {
- /* cluster_phys_read: failed to get pagelist */
- return(EINVAL);
- }
+ if (kret != KERN_SUCCESS) {
+ /*
+ * cluster_phys_read: failed to get pagelist
+ */
+ return(EINVAL);
+ }
+ if (upl_size < upl_needed_size) {
+ /*
+ * The upl_size wasn't satisfied.
+ */
+ ubc_upl_abort_range(upl, 0, upl_size, UPL_ABORT_FREE_ON_EMPTY);
+
+ return(EINVAL);
+ }
+ pl = ubc_upl_pageinfo(upl);
+
+ dst_paddr = (((addr64_t)(int)upl_phys_page(pl, 0)) << 12) + ((addr64_t)iov->iov_base & PAGE_MASK);
+
+ while (((uio->uio_offset & (devblocksize - 1)) || io_size < devblocksize) && io_size) {
+ int head_size;
+ head_size = devblocksize - (int)(uio->uio_offset & (devblocksize - 1));
+
+ if (head_size > io_size)
+ head_size = io_size;
+
+ error = cluster_align_phys_io(vp, uio, dst_paddr, head_size, devblocksize, CL_READ);
+
+ if (error) {
+ ubc_upl_abort_range(upl, 0, upl_size, UPL_ABORT_FREE_ON_EMPTY);
+
+ return(EINVAL);
+ }
+ upl_offset += head_size;
+ dst_paddr += head_size;
+ io_size -= head_size;
+ }
+ tail_size = io_size & (devblocksize - 1);
+ io_size -= tail_size;
+
+ iostate.io_completed = 0;
+ iostate.io_issued = 0;
+ iostate.io_error = 0;
+ iostate.io_wanted = 0;
+
+ while (io_size && error == 0) {
+ int xsize;
+
+ if (io_size > (MAX_UPL_TRANSFER * PAGE_SIZE))
+ xsize = MAX_UPL_TRANSFER * PAGE_SIZE;
+ else
+ xsize = io_size;
+ /*
+ * request asynchronously so that we can overlap
+ * the preparation of the next I/O... we'll do
+ * the commit after all the I/O has completed
+ * since its all issued against the same UPL
+ * if there are already too many outstanding reads
+ * wait until some have completed before issuing the next
+ */
+ while ((iostate.io_issued - iostate.io_completed) > (2 * MAX_UPL_TRANSFER * PAGE_SIZE)) {
+ iostate.io_wanted = 1;
+ tsleep((caddr_t)&iostate.io_wanted, PRIBIO + 1, "cluster_phys_read", 0);
+ }
+
+ error = cluster_io(vp, upl, upl_offset, uio->uio_offset, xsize, 0,
+ CL_READ | CL_NOZERO | CL_DEV_MEMORY | CL_ASYNC,
+ (struct buf *)0, &iostate);
+ /*
+ * The cluster_io read was issued successfully,
+ * update the uio structure
+ */
+ if (error == 0) {
+ uio->uio_resid -= xsize;
+ iov->iov_len -= xsize;
+ iov->iov_base += xsize;
+ uio->uio_offset += xsize;
+ dst_paddr += xsize;
+ upl_offset += xsize;
+ io_size -= xsize;
+ }
+ }
/*
- * Consider the possibility that upl_size wasn't satisfied.
+ * make sure all async reads that are part of this stream
+ * have completed before we proceed
*/
- if (upl_size < upl_needed_size)
- {
- ubc_upl_abort_range(upl, 0, upl_size, UPL_ABORT_FREE_ON_EMPTY);
- return(EINVAL);
- }
+ while (iostate.io_issued != iostate.io_completed) {
+ iostate.io_wanted = 1;
+ tsleep((caddr_t)&iostate.io_wanted, PRIBIO + 1, "cluster_phys_read", 0);
+ }
+ if (iostate.io_error) {
+ error = iostate.io_error;
+ }
+ if (error == 0 && tail_size)
+ error = cluster_align_phys_io(vp, uio, dst_paddr, tail_size, devblocksize, CL_READ);
/*
- * issue a synchronous read to cluster_io
+ * just release our hold on the physically contiguous
+ * region without changing any state
*/
-
- error = cluster_io(vp, upl, upl_offset, uio->uio_offset,
- io_size, 0, CL_READ| CL_NOZERO | CL_DEV_MEMORY, (struct buf *)0);
-
- if (error == 0)
- {
- /*
- * The cluster_io read completed successfully,
- * update the uio structure and commit.
- */
-
- ubc_upl_commit_range(upl, 0, upl_size, UPL_COMMIT_FREE_ON_EMPTY);
-
- iov->iov_base += io_size;
- iov->iov_len -= io_size;
- uio->uio_resid -= io_size;
- uio->uio_offset += io_size;
- }
- else
- ubc_upl_abort_range(upl, 0, upl_size, UPL_ABORT_FREE_ON_EMPTY);
+ ubc_upl_abort_range(upl, 0, upl_size, UPL_ABORT_FREE_ON_EMPTY);
return (error);
}
+
/*
* generate advisory I/O's in the largest chunks possible
* the completed pages will be released into the VM cache
* issue an asynchronous read to cluster_io
*/
retval = cluster_io(vp, upl, upl_offset, upl_f_offset + upl_offset, io_size, devblocksize,
- CL_ASYNC | CL_READ | CL_COMMIT | CL_AGE, (struct buf *)0);
+ CL_ASYNC | CL_READ | CL_COMMIT | CL_AGE, (struct buf *)0, (struct clios *)0);
issued_io = 1;
}
vp->v_flag |= VTHROTTLED;
tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "cluster_push", 0);
}
- cluster_io(vp, upl, upl_offset, upl_f_offset + upl_offset, io_size, vp->v_ciosiz, io_flags, (struct buf *)0);
+ cluster_io(vp, upl, upl_offset, upl_f_offset + upl_offset, io_size, vp->v_ciosiz, io_flags, (struct buf *)0, (struct clios *)0);
size -= io_size;
}
return(1);
}
+
+
+
+static int
+cluster_align_phys_io(struct vnode *vp, struct uio *uio, addr64_t usr_paddr, int xsize, int devblocksize, int flags)
+{
+ struct iovec *iov;
+ upl_page_info_t *pl;
+ upl_t upl;
+ addr64_t ubc_paddr;
+ kern_return_t kret;
+ int error = 0;
+
+ iov = uio->uio_iov;
+
+ kret = ubc_create_upl(vp,
+ uio->uio_offset & ~PAGE_MASK_64,
+ PAGE_SIZE,
+ &upl,
+ &pl,
+ UPL_FLAGS_NONE);
+
+ if (kret != KERN_SUCCESS)
+ return(EINVAL);
+
+ if (!upl_valid_page(pl, 0)) {
+ /*
+ * issue a synchronous read to cluster_io
+ */
+ error = cluster_io(vp, upl, 0, uio->uio_offset & ~PAGE_MASK_64, PAGE_SIZE, devblocksize,
+ CL_READ, (struct buf *)0, (struct clios *)0);
+ if (error) {
+ ubc_upl_abort_range(upl, 0, PAGE_SIZE, UPL_ABORT_DUMP_PAGES | UPL_ABORT_FREE_ON_EMPTY);
+
+ return(error);
+ }
+ }
+ ubc_paddr = ((addr64_t)upl_phys_page(pl, 0) << 12) + (addr64_t)(uio->uio_offset & PAGE_MASK_64);
+
+/*
+ * NOTE: There is no prototype for the following in BSD. It, and the definitions
+ * of the defines for cppvPsrc, cppvPsnk, cppvFsnk, and cppvFsrc will be found in
+ * osfmk/ppc/mappings.h. They are not included here because there appears to be no
+ * way to do so without exporting them to kexts as well.
+ */
+ if (flags & CL_READ)
+// copypv(ubc_paddr, usr_paddr, xsize, cppvPsrc | cppvPsnk | cppvFsnk); /* Copy physical to physical and flush the destination */
+ copypv(ubc_paddr, usr_paddr, xsize, 2 | 1 | 4); /* Copy physical to physical and flush the destination */
+ else
+// copypv(ubc_paddr, usr_paddr, xsize, cppvPsrc | cppvPsnk | cppvFsrc); /* Copy physical to physical and flush the source */
+ copypv(ubc_paddr, usr_paddr, xsize, 2 | 1 | 8); /* Copy physical to physical and flush the source */
+
+ if ( !(flags & CL_READ) || upl_dirty_page(pl, 0)) {
+ /*
+ * issue a synchronous write to cluster_io
+ */
+ error = cluster_io(vp, upl, 0, uio->uio_offset & ~PAGE_MASK_64, PAGE_SIZE, devblocksize,
+ 0, (struct buf *)0, (struct clios *)0);
+ }
+ if (error == 0) {
+ uio->uio_offset += xsize;
+ iov->iov_base += xsize;
+ iov->iov_len -= xsize;
+ uio->uio_resid -= xsize;
+ }
+ ubc_upl_abort_range(upl, 0, PAGE_SIZE, UPL_ABORT_DUMP_PAGES | UPL_ABORT_FREE_ON_EMPTY);
+
+ return (error);
+}
projects / apple / xnu.git / blobdiff