Page MenuHomeFreeBSD
Files F166462827

No OneTemporary
Actions

View Options

Index: stable/8/sys/amd64/include/xen
===================================================================
--- stable/8/sys/amd64/include/xen (revision 231331)
+++ stable/8/sys/amd64/include/xen (revision 231332)
Property changes on: stable/8/sys/amd64/include/xen
___________________________________________________________________
Modified: svn:mergeinfo
## -0,0 +0,1 ##
Merged /head/sys/amd64/include/xen:r230605
Index: stable/8/sys/boot
===================================================================
--- stable/8/sys/boot (revision 231331)
+++ stable/8/sys/boot (revision 231332)
Property changes on: stable/8/sys/boot
___________________________________________________________________
Modified: svn:mergeinfo
## -0,0 +0,1 ##
Merged /head/sys/boot:r230605
Index: stable/8/sys/cddl/contrib/opensolaris
===================================================================
--- stable/8/sys/cddl/contrib/opensolaris (revision 231331)
+++ stable/8/sys/cddl/contrib/opensolaris (revision 231332)
Property changes on: stable/8/sys/cddl/contrib/opensolaris
___________________________________________________________________
Modified: svn:mergeinfo
## -0,0 +0,1 ##
Merged /head/sys/cddl/contrib/opensolaris:r230605
Index: stable/8/sys/contrib/dev/acpica
===================================================================
--- stable/8/sys/contrib/dev/acpica (revision 231331)
+++ stable/8/sys/contrib/dev/acpica (revision 231332)
Property changes on: stable/8/sys/contrib/dev/acpica
___________________________________________________________________
Modified: svn:mergeinfo
## -0,0 +0,1 ##
Merged /head/sys/contrib/dev/acpica:r230605
Index: stable/8/sys/contrib/pf
===================================================================
--- stable/8/sys/contrib/pf (revision 231331)
+++ stable/8/sys/contrib/pf (revision 231332)
Property changes on: stable/8/sys/contrib/pf
___________________________________________________________________
Modified: svn:mergeinfo
## -0,0 +0,1 ##
Merged /head/sys/contrib/pf:r230605
Index: stable/8/sys/dev/e1000
===================================================================
--- stable/8/sys/dev/e1000 (revision 231331)
+++ stable/8/sys/dev/e1000 (revision 231332)
Property changes on: stable/8/sys/dev/e1000
___________________________________________________________________
Modified: svn:mergeinfo
## -0,0 +0,1 ##
Merged /head/sys/dev/e1000:r230605
Index: stable/8/sys/fs/nfsclient/nfs_clbio.c
===================================================================
--- stable/8/sys/fs/nfsclient/nfs_clbio.c (revision 231331)
+++ stable/8/sys/fs/nfsclient/nfs_clbio.c (revision 231332)
@@ -1,1809 +1,1805 @@
/*-
* Copyright (c) 1989, 1993
* The Regents of the University of California. All rights reserved.
*
* This code is derived from software contributed to Berkeley by
* Rick Macklem at The University of Guelph.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* @(#)nfs_bio.c 8.9 (Berkeley) 3/30/95
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bio.h>
#include <sys/buf.h>
#include <sys/kernel.h>
#include <sys/mount.h>
#include <sys/vmmeter.h>
#include <sys/vnode.h>
#include <vm/vm.h>
#include <vm/vm_extern.h>
#include <vm/vm_page.h>
#include <vm/vm_object.h>
#include <vm/vm_pager.h>
#include <vm/vnode_pager.h>
#include <fs/nfs/nfsport.h>
#include <fs/nfsclient/nfsmount.h>
#include <fs/nfsclient/nfs.h>
#include <fs/nfsclient/nfsnode.h>
extern int newnfs_directio_allow_mmap;
extern struct nfsstats newnfsstats;
extern struct mtx ncl_iod_mutex;
extern int ncl_numasync;
extern enum nfsiod_state ncl_iodwant[NFS_MAXASYNCDAEMON];
extern struct nfsmount *ncl_iodmount[NFS_MAXASYNCDAEMON];
extern int newnfs_directio_enable;
int ncl_pbuf_freecnt = -1; /* start out unlimited */
static struct buf *nfs_getcacheblk(struct vnode *vp, daddr_t bn, int size,
struct thread *td);
static int nfs_directio_write(struct vnode *vp, struct uio *uiop,
struct ucred *cred, int ioflag);
/*
* Vnode op for VM getpages.
*/
int
ncl_getpages(struct vop_getpages_args *ap)
{
int i, error, nextoff, size, toff, count, npages;
struct uio uio;
struct iovec iov;
vm_offset_t kva;
struct buf *bp;
struct vnode *vp;
struct thread *td;
struct ucred *cred;
struct nfsmount *nmp;
vm_object_t object;
vm_page_t *pages;
struct nfsnode *np;
vp = ap->a_vp;
np = VTONFS(vp);
td = curthread; /* XXX */
cred = curthread->td_ucred; /* XXX */
nmp = VFSTONFS(vp->v_mount);
pages = ap->a_m;
count = ap->a_count;
if ((object = vp->v_object) == NULL) {
ncl_printf("nfs_getpages: called with non-merged cache vnode??\n");
return (VM_PAGER_ERROR);
}
if (newnfs_directio_enable && !newnfs_directio_allow_mmap) {
mtx_lock(&np->n_mtx);
if ((np->n_flag & NNONCACHE) && (vp->v_type == VREG)) {
mtx_unlock(&np->n_mtx);
ncl_printf("nfs_getpages: called on non-cacheable vnode??\n");
return (VM_PAGER_ERROR);
} else
mtx_unlock(&np->n_mtx);
}
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 &&
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
mtx_unlock(&nmp->nm_mtx);
/* We'll never get here for v4, because we always have fsinfo */
(void)ncl_fsinfo(nmp, vp, cred, td);
} else
mtx_unlock(&nmp->nm_mtx);
npages = btoc(count);
/*
* If the requested page is partially valid, just return it and
* allow the pager to zero-out the blanks. Partially valid pages
* can only occur at the file EOF.
*/
VM_OBJECT_LOCK(object);
if (pages[ap->a_reqpage]->valid != 0) {
vm_page_lock_queues();
for (i = 0; i < npages; ++i) {
if (i != ap->a_reqpage)
vm_page_free(pages[i]);
}
vm_page_unlock_queues();
VM_OBJECT_UNLOCK(object);
return (0);
}
VM_OBJECT_UNLOCK(object);
/*
* We use only the kva address for the buffer, but this is extremely
* convienient and fast.
*/
bp = getpbuf(&ncl_pbuf_freecnt);
kva = (vm_offset_t) bp->b_data;
pmap_qenter(kva, pages, npages);
PCPU_INC(cnt.v_vnodein);
PCPU_ADD(cnt.v_vnodepgsin, npages);
iov.iov_base = (caddr_t) kva;
iov.iov_len = count;
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
uio.uio_resid = count;
uio.uio_segflg = UIO_SYSSPACE;
uio.uio_rw = UIO_READ;
uio.uio_td = td;
error = ncl_readrpc(vp, &uio, cred);
pmap_qremove(kva, npages);
relpbuf(bp, &ncl_pbuf_freecnt);
if (error && (uio.uio_resid == count)) {
ncl_printf("nfs_getpages: error %d\n", error);
VM_OBJECT_LOCK(object);
vm_page_lock_queues();
for (i = 0; i < npages; ++i) {
if (i != ap->a_reqpage)
vm_page_free(pages[i]);
}
vm_page_unlock_queues();
VM_OBJECT_UNLOCK(object);
return (VM_PAGER_ERROR);
}
/*
* Calculate the number of bytes read and validate only that number
* of bytes. Note that due to pending writes, size may be 0. This
* does not mean that the remaining data is invalid!
*/
size = count - uio.uio_resid;
VM_OBJECT_LOCK(object);
vm_page_lock_queues();
for (i = 0, toff = 0; i < npages; i++, toff = nextoff) {
vm_page_t m;
nextoff = toff + PAGE_SIZE;
m = pages[i];
if (nextoff <= size) {
/*
* Read operation filled an entire page
*/
m->valid = VM_PAGE_BITS_ALL;
KASSERT(m->dirty == 0,
("nfs_getpages: page %p is dirty", m));
} else if (size > toff) {
/*
* Read operation filled a partial page.
*/
m->valid = 0;
vm_page_set_valid(m, 0, size - toff);
KASSERT(m->dirty == 0,
("nfs_getpages: page %p is dirty", m));
} else {
/*
* Read operation was short. If no error occured
* we may have hit a zero-fill section. We simply
* leave valid set to 0.
*/
;
}
if (i != ap->a_reqpage) {
/*
* Whether or not to leave the page activated is up in
* the air, but we should put the page on a page queue
* somewhere (it already is in the object). Result:
* It appears that emperical results show that
* deactivating pages is best.
*/
/*
* Just in case someone was asking for this page we
* now tell them that it is ok to use.
*/
if (!error) {
if (m->oflags & VPO_WANTED)
vm_page_activate(m);
else
vm_page_deactivate(m);
vm_page_wakeup(m);
} else {
vm_page_free(m);
}
}
}
vm_page_unlock_queues();
VM_OBJECT_UNLOCK(object);
return (0);
}
/*
* Vnode op for VM putpages.
*/
int
ncl_putpages(struct vop_putpages_args *ap)
{
struct uio uio;
struct iovec iov;
vm_offset_t kva;
struct buf *bp;
int iomode, must_commit, i, error, npages, count;
off_t offset;
int *rtvals;
struct vnode *vp;
struct thread *td;
struct ucred *cred;
struct nfsmount *nmp;
struct nfsnode *np;
vm_page_t *pages;
vp = ap->a_vp;
np = VTONFS(vp);
td = curthread; /* XXX */
cred = curthread->td_ucred; /* XXX */
nmp = VFSTONFS(vp->v_mount);
pages = ap->a_m;
count = ap->a_count;
rtvals = ap->a_rtvals;
npages = btoc(count);
offset = IDX_TO_OFF(pages[0]->pindex);
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 &&
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
mtx_unlock(&nmp->nm_mtx);
(void)ncl_fsinfo(nmp, vp, cred, td);
} else
mtx_unlock(&nmp->nm_mtx);
mtx_lock(&np->n_mtx);
if (newnfs_directio_enable && !newnfs_directio_allow_mmap &&
(np->n_flag & NNONCACHE) && (vp->v_type == VREG)) {
mtx_unlock(&np->n_mtx);
ncl_printf("ncl_putpages: called on noncache-able vnode??\n");
mtx_lock(&np->n_mtx);
}
for (i = 0; i < npages; i++)
rtvals[i] = VM_PAGER_ERROR;
/*
* When putting pages, do not extend file past EOF.
*/
if (offset + count > np->n_size) {
count = np->n_size - offset;
if (count < 0)
count = 0;
}
mtx_unlock(&np->n_mtx);
/*
* We use only the kva address for the buffer, but this is extremely
* convienient and fast.
*/
bp = getpbuf(&ncl_pbuf_freecnt);
kva = (vm_offset_t) bp->b_data;
pmap_qenter(kva, pages, npages);
PCPU_INC(cnt.v_vnodeout);
PCPU_ADD(cnt.v_vnodepgsout, count);
iov.iov_base = (caddr_t) kva;
iov.iov_len = count;
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = offset;
uio.uio_resid = count;
uio.uio_segflg = UIO_SYSSPACE;
uio.uio_rw = UIO_WRITE;
uio.uio_td = td;
if ((ap->a_sync & VM_PAGER_PUT_SYNC) == 0)
iomode = NFSWRITE_UNSTABLE;
else
iomode = NFSWRITE_FILESYNC;
error = ncl_writerpc(vp, &uio, cred, &iomode, &must_commit, 0);
pmap_qremove(kva, npages);
relpbuf(bp, &ncl_pbuf_freecnt);
vnode_pager_undirty_pages(pages, rtvals, count - uio.uio_resid);
if (must_commit)
ncl_clearcommit(vp->v_mount);
return rtvals[0];
}
/*
* For nfs, cache consistency can only be maintained approximately.
* Although RFC1094 does not specify the criteria, the following is
* believed to be compatible with the reference port.
* For nfs:
* If the file's modify time on the server has changed since the
* last read rpc or you have written to the file,
* you may have lost data cache consistency with the
* server, so flush all of the file's data out of the cache.
* Then force a getattr rpc to ensure that you have up to date
* attributes.
* NB: This implies that cache data can be read when up to
* NFS_ATTRTIMEO seconds out of date. If you find that you need current
* attributes this could be forced by setting n_attrstamp to 0 before
* the VOP_GETATTR() call.
*/
static inline int
nfs_bioread_check_cons(struct vnode *vp, struct thread *td, struct ucred *cred)
{
int error = 0;
struct vattr vattr;
struct nfsnode *np = VTONFS(vp);
int old_lock;
/*
* Grab the exclusive lock before checking whether the cache is
* consistent.
* XXX - We can make this cheaper later (by acquiring cheaper locks).
* But for now, this suffices.
*/
old_lock = ncl_upgrade_vnlock(vp);
if (vp->v_iflag & VI_DOOMED) {
ncl_downgrade_vnlock(vp, old_lock);
return (EBADF);
}
mtx_lock(&np->n_mtx);
if (np->n_flag & NMODIFIED) {
mtx_unlock(&np->n_mtx);
if (vp->v_type != VREG) {
if (vp->v_type != VDIR)
panic("nfs: bioread, not dir");
ncl_invaldir(vp);
error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
if (error)
goto out;
}
np->n_attrstamp = 0;
error = VOP_GETATTR(vp, &vattr, cred);
if (error)
goto out;
mtx_lock(&np->n_mtx);
np->n_mtime = vattr.va_mtime;
mtx_unlock(&np->n_mtx);
} else {
mtx_unlock(&np->n_mtx);
error = VOP_GETATTR(vp, &vattr, cred);
if (error)
return (error);
mtx_lock(&np->n_mtx);
if ((np->n_flag & NSIZECHANGED)
|| (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime))) {
mtx_unlock(&np->n_mtx);
if (vp->v_type == VDIR)
ncl_invaldir(vp);
error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
if (error)
goto out;
mtx_lock(&np->n_mtx);
np->n_mtime = vattr.va_mtime;
np->n_flag &= ~NSIZECHANGED;
}
mtx_unlock(&np->n_mtx);
}
out:
ncl_downgrade_vnlock(vp, old_lock);
return error;
}
/*
* Vnode op for read using bio
*/
int
ncl_bioread(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *cred)
{
struct nfsnode *np = VTONFS(vp);
int biosize, i;
struct buf *bp, *rabp;
struct thread *td;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
daddr_t lbn, rabn;
int bcount;
int seqcount;
int nra, error = 0, n = 0, on = 0;
off_t tmp_off;
KASSERT(uio->uio_rw == UIO_READ, ("ncl_read mode"));
if (uio->uio_resid == 0)
return (0);
if (uio->uio_offset < 0) /* XXX VDIR cookies can be negative */
return (EINVAL);
td = uio->uio_td;
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 &&
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
mtx_unlock(&nmp->nm_mtx);
(void)ncl_fsinfo(nmp, vp, cred, td);
mtx_lock(&nmp->nm_mtx);
}
if (nmp->nm_rsize == 0 || nmp->nm_readdirsize == 0)
(void) newnfs_iosize(nmp);
tmp_off = uio->uio_offset + uio->uio_resid;
if (vp->v_type != VDIR &&
(tmp_off > nmp->nm_maxfilesize || tmp_off < uio->uio_offset)) {
mtx_unlock(&nmp->nm_mtx);
return (EFBIG);
}
mtx_unlock(&nmp->nm_mtx);
if (newnfs_directio_enable && (ioflag & IO_DIRECT) && (vp->v_type == VREG))
/* No caching/ no readaheads. Just read data into the user buffer */
return ncl_readrpc(vp, uio, cred);
- biosize = vp->v_mount->mnt_stat.f_iosize;
+ biosize = vp->v_bufobj.bo_bsize;
seqcount = (int)((off_t)(ioflag >> IO_SEQSHIFT) * biosize / BKVASIZE);
error = nfs_bioread_check_cons(vp, td, cred);
if (error)
return error;
do {
u_quad_t nsize;
mtx_lock(&np->n_mtx);
nsize = np->n_size;
mtx_unlock(&np->n_mtx);
switch (vp->v_type) {
case VREG:
NFSINCRGLOBAL(newnfsstats.biocache_reads);
lbn = uio->uio_offset / biosize;
on = uio->uio_offset & (biosize - 1);
/*
* Start the read ahead(s), as required.
*/
if (nmp->nm_readahead > 0) {
for (nra = 0; nra < nmp->nm_readahead && nra < seqcount &&
(off_t)(lbn + 1 + nra) * biosize < nsize; nra++) {
rabn = lbn + 1 + nra;
if (incore(&vp->v_bufobj, rabn) == NULL) {
rabp = nfs_getcacheblk(vp, rabn, biosize, td);
if (!rabp) {
error = newnfs_sigintr(nmp, td);
return (error ? error : EINTR);
}
if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) {
rabp->b_flags |= B_ASYNC;
rabp->b_iocmd = BIO_READ;
vfs_busy_pages(rabp, 0);
if (ncl_asyncio(nmp, rabp, cred, td)) {
rabp->b_flags |= B_INVAL;
rabp->b_ioflags |= BIO_ERROR;
vfs_unbusy_pages(rabp);
brelse(rabp);
break;
}
} else {
brelse(rabp);
}
}
}
}
/* Note that bcount is *not* DEV_BSIZE aligned. */
bcount = biosize;
if ((off_t)lbn * biosize >= nsize) {
bcount = 0;
} else if ((off_t)(lbn + 1) * biosize > nsize) {
bcount = nsize - (off_t)lbn * biosize;
}
bp = nfs_getcacheblk(vp, lbn, bcount, td);
if (!bp) {
error = newnfs_sigintr(nmp, td);
return (error ? error : EINTR);
}
/*
* If B_CACHE is not set, we must issue the read. If this
* fails, we return an error.
*/
if ((bp->b_flags & B_CACHE) == 0) {
bp->b_iocmd = BIO_READ;
vfs_busy_pages(bp, 0);
error = ncl_doio(vp, bp, cred, td, 0);
if (error) {
brelse(bp);
return (error);
}
}
/*
* on is the offset into the current bp. Figure out how many
* bytes we can copy out of the bp. Note that bcount is
* NOT DEV_BSIZE aligned.
*
* Then figure out how many bytes we can copy into the uio.
*/
n = 0;
if (on < bcount)
n = min((unsigned)(bcount - on), uio->uio_resid);
break;
case VLNK:
NFSINCRGLOBAL(newnfsstats.biocache_readlinks);
bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, td);
if (!bp) {
error = newnfs_sigintr(nmp, td);
return (error ? error : EINTR);
}
if ((bp->b_flags & B_CACHE) == 0) {
bp->b_iocmd = BIO_READ;
vfs_busy_pages(bp, 0);
error = ncl_doio(vp, bp, cred, td, 0);
if (error) {
bp->b_ioflags |= BIO_ERROR;
brelse(bp);
return (error);
}
}
n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid);
on = 0;
break;
case VDIR:
NFSINCRGLOBAL(newnfsstats.biocache_readdirs);
if (np->n_direofoffset
&& uio->uio_offset >= np->n_direofoffset) {
return (0);
}
lbn = (uoff_t)uio->uio_offset / NFS_DIRBLKSIZ;
on = uio->uio_offset & (NFS_DIRBLKSIZ - 1);
bp = nfs_getcacheblk(vp, lbn, NFS_DIRBLKSIZ, td);
if (!bp) {
error = newnfs_sigintr(nmp, td);
return (error ? error : EINTR);
}
if ((bp->b_flags & B_CACHE) == 0) {
bp->b_iocmd = BIO_READ;
vfs_busy_pages(bp, 0);
error = ncl_doio(vp, bp, cred, td, 0);
if (error) {
brelse(bp);
}
while (error == NFSERR_BAD_COOKIE) {
ncl_invaldir(vp);
error = ncl_vinvalbuf(vp, 0, td, 1);
/*
* Yuck! The directory has been modified on the
* server. The only way to get the block is by
* reading from the beginning to get all the
* offset cookies.
*
* Leave the last bp intact unless there is an error.
* Loop back up to the while if the error is another
* NFSERR_BAD_COOKIE (double yuch!).
*/
for (i = 0; i <= lbn && !error; i++) {
if (np->n_direofoffset
&& (i * NFS_DIRBLKSIZ) >= np->n_direofoffset)
return (0);
bp = nfs_getcacheblk(vp, i, NFS_DIRBLKSIZ, td);
if (!bp) {
error = newnfs_sigintr(nmp, td);
return (error ? error : EINTR);
}
if ((bp->b_flags & B_CACHE) == 0) {
bp->b_iocmd = BIO_READ;
vfs_busy_pages(bp, 0);
error = ncl_doio(vp, bp, cred, td, 0);
/*
* no error + B_INVAL == directory EOF,
* use the block.
*/
if (error == 0 && (bp->b_flags & B_INVAL))
break;
}
/*
* An error will throw away the block and the
* for loop will break out. If no error and this
* is not the block we want, we throw away the
* block and go for the next one via the for loop.
*/
if (error || i < lbn)
brelse(bp);
}
}
/*
* The above while is repeated if we hit another cookie
* error. If we hit an error and it wasn't a cookie error,
* we give up.
*/
if (error)
return (error);
}
/*
* If not eof and read aheads are enabled, start one.
* (You need the current block first, so that you have the
* directory offset cookie of the next block.)
*/
if (nmp->nm_readahead > 0 &&
(bp->b_flags & B_INVAL) == 0 &&
(np->n_direofoffset == 0 ||
(lbn + 1) * NFS_DIRBLKSIZ < np->n_direofoffset) &&
incore(&vp->v_bufobj, lbn + 1) == NULL) {
rabp = nfs_getcacheblk(vp, lbn + 1, NFS_DIRBLKSIZ, td);
if (rabp) {
if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) {
rabp->b_flags |= B_ASYNC;
rabp->b_iocmd = BIO_READ;
vfs_busy_pages(rabp, 0);
if (ncl_asyncio(nmp, rabp, cred, td)) {
rabp->b_flags |= B_INVAL;
rabp->b_ioflags |= BIO_ERROR;
vfs_unbusy_pages(rabp);
brelse(rabp);
}
} else {
brelse(rabp);
}
}
}
/*
* Unlike VREG files, whos buffer size ( bp->b_bcount ) is
* chopped for the EOF condition, we cannot tell how large
* NFS directories are going to be until we hit EOF. So
* an NFS directory buffer is *not* chopped to its EOF. Now,
* it just so happens that b_resid will effectively chop it
* to EOF. *BUT* this information is lost if the buffer goes
* away and is reconstituted into a B_CACHE state ( due to
* being VMIO ) later. So we keep track of the directory eof
* in np->n_direofoffset and chop it off as an extra step
* right here.
*/
n = lmin(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid - on);
if (np->n_direofoffset && n > np->n_direofoffset - uio->uio_offset)
n = np->n_direofoffset - uio->uio_offset;
break;
default:
ncl_printf(" ncl_bioread: type %x unexpected\n", vp->v_type);
bp = NULL;
break;
};
if (n > 0) {
error = uiomove(bp->b_data + on, (int)n, uio);
}
if (vp->v_type == VLNK)
n = 0;
if (bp != NULL)
brelse(bp);
} while (error == 0 && uio->uio_resid > 0 && n > 0);
return (error);
}
/*
* The NFS write path cannot handle iovecs with len > 1. So we need to
* break up iovecs accordingly (restricting them to wsize).
* For the SYNC case, we can do this with 1 copy (user buffer -> mbuf).
* For the ASYNC case, 2 copies are needed. The first a copy from the
* user buffer to a staging buffer and then a second copy from the staging
* buffer to mbufs. This can be optimized by copying from the user buffer
* directly into mbufs and passing the chain down, but that requires a
* fair amount of re-working of the relevant codepaths (and can be done
* later).
*/
static int
nfs_directio_write(vp, uiop, cred, ioflag)
struct vnode *vp;
struct uio *uiop;
struct ucred *cred;
int ioflag;
{
int error;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
struct thread *td = uiop->uio_td;
int size;
int wsize;
mtx_lock(&nmp->nm_mtx);
wsize = nmp->nm_wsize;
mtx_unlock(&nmp->nm_mtx);
if (ioflag & IO_SYNC) {
int iomode, must_commit;
struct uio uio;
struct iovec iov;
do_sync:
while (uiop->uio_resid > 0) {
size = min(uiop->uio_resid, wsize);
size = min(uiop->uio_iov->iov_len, size);
iov.iov_base = uiop->uio_iov->iov_base;
iov.iov_len = size;
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = uiop->uio_offset;
uio.uio_resid = size;
uio.uio_segflg = UIO_USERSPACE;
uio.uio_rw = UIO_WRITE;
uio.uio_td = td;
iomode = NFSWRITE_FILESYNC;
error = ncl_writerpc(vp, &uio, cred, &iomode,
&must_commit, 0);
KASSERT((must_commit == 0),
("ncl_directio_write: Did not commit write"));
if (error)
return (error);
uiop->uio_offset += size;
uiop->uio_resid -= size;
if (uiop->uio_iov->iov_len <= size) {
uiop->uio_iovcnt--;
uiop->uio_iov++;
} else {
uiop->uio_iov->iov_base =
(char *)uiop->uio_iov->iov_base + size;
uiop->uio_iov->iov_len -= size;
}
}
} else {
struct uio *t_uio;
struct iovec *t_iov;
struct buf *bp;
/*
* Break up the write into blocksize chunks and hand these
* over to nfsiod's for write back.
* Unfortunately, this incurs a copy of the data. Since
* the user could modify the buffer before the write is
* initiated.
*
* The obvious optimization here is that one of the 2 copies
* in the async write path can be eliminated by copying the
* data here directly into mbufs and passing the mbuf chain
* down. But that will require a fair amount of re-working
* of the code and can be done if there's enough interest
* in NFS directio access.
*/
while (uiop->uio_resid > 0) {
size = min(uiop->uio_resid, wsize);
size = min(uiop->uio_iov->iov_len, size);
bp = getpbuf(&ncl_pbuf_freecnt);
t_uio = malloc(sizeof(struct uio), M_NFSDIRECTIO, M_WAITOK);
t_iov = malloc(sizeof(struct iovec), M_NFSDIRECTIO, M_WAITOK);
t_iov->iov_base = malloc(size, M_NFSDIRECTIO, M_WAITOK);
t_iov->iov_len = size;
t_uio->uio_iov = t_iov;
t_uio->uio_iovcnt = 1;
t_uio->uio_offset = uiop->uio_offset;
t_uio->uio_resid = size;
t_uio->uio_segflg = UIO_SYSSPACE;
t_uio->uio_rw = UIO_WRITE;
t_uio->uio_td = td;
bcopy(uiop->uio_iov->iov_base, t_iov->iov_base, size);
bp->b_flags |= B_DIRECT;
bp->b_iocmd = BIO_WRITE;
if (cred != NOCRED) {
crhold(cred);
bp->b_wcred = cred;
} else
bp->b_wcred = NOCRED;
bp->b_caller1 = (void *)t_uio;
bp->b_vp = vp;
error = ncl_asyncio(nmp, bp, NOCRED, td);
if (error) {
free(t_iov->iov_base, M_NFSDIRECTIO);
free(t_iov, M_NFSDIRECTIO);
free(t_uio, M_NFSDIRECTIO);
bp->b_vp = NULL;
relpbuf(bp, &ncl_pbuf_freecnt);
if (error == EINTR)
return (error);
goto do_sync;
}
uiop->uio_offset += size;
uiop->uio_resid -= size;
if (uiop->uio_iov->iov_len <= size) {
uiop->uio_iovcnt--;
uiop->uio_iov++;
} else {
uiop->uio_iov->iov_base =
(char *)uiop->uio_iov->iov_base + size;
uiop->uio_iov->iov_len -= size;
}
}
}
return (0);
}
/*
* Vnode op for write using bio
*/
int
ncl_write(struct vop_write_args *ap)
{
int biosize;
struct uio *uio = ap->a_uio;
struct thread *td = uio->uio_td;
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
struct ucred *cred = ap->a_cred;
int ioflag = ap->a_ioflag;
struct buf *bp;
struct vattr vattr;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
daddr_t lbn;
int bcount;
int n, on, error = 0;
off_t tmp_off;
KASSERT(uio->uio_rw == UIO_WRITE, ("ncl_write mode"));
KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
("ncl_write proc"));
if (vp->v_type != VREG)
return (EIO);
mtx_lock(&np->n_mtx);
if (np->n_flag & NWRITEERR) {
np->n_flag &= ~NWRITEERR;
mtx_unlock(&np->n_mtx);
return (np->n_error);
} else
mtx_unlock(&np->n_mtx);
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 &&
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
mtx_unlock(&nmp->nm_mtx);
(void)ncl_fsinfo(nmp, vp, cred, td);
mtx_lock(&nmp->nm_mtx);
}
if (nmp->nm_wsize == 0)
(void) newnfs_iosize(nmp);
mtx_unlock(&nmp->nm_mtx);
/*
* Synchronously flush pending buffers if we are in synchronous
* mode or if we are appending.
*/
if (ioflag & (IO_APPEND | IO_SYNC)) {
mtx_lock(&np->n_mtx);
if (np->n_flag & NMODIFIED) {
mtx_unlock(&np->n_mtx);
#ifdef notyet /* Needs matching nonblock semantics elsewhere, too. */
/*
* Require non-blocking, synchronous writes to
* dirty files to inform the program it needs
* to fsync(2) explicitly.
*/
if (ioflag & IO_NDELAY)
return (EAGAIN);
#endif
flush_and_restart:
np->n_attrstamp = 0;
error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
if (error)
return (error);
} else
mtx_unlock(&np->n_mtx);
}
/*
* If IO_APPEND then load uio_offset. We restart here if we cannot
* get the append lock.
*/
if (ioflag & IO_APPEND) {
np->n_attrstamp = 0;
error = VOP_GETATTR(vp, &vattr, cred);
if (error)
return (error);
mtx_lock(&np->n_mtx);
uio->uio_offset = np->n_size;
mtx_unlock(&np->n_mtx);
}
if (uio->uio_offset < 0)
return (EINVAL);
tmp_off = uio->uio_offset + uio->uio_resid;
if (tmp_off > nmp->nm_maxfilesize || tmp_off < uio->uio_offset)
return (EFBIG);
if (uio->uio_resid == 0)
return (0);
if (newnfs_directio_enable && (ioflag & IO_DIRECT) && vp->v_type == VREG)
return nfs_directio_write(vp, uio, cred, ioflag);
/*
* Maybe this should be above the vnode op call, but so long as
* file servers have no limits, i don't think it matters
*/
if (vn_rlimit_fsize(vp, uio, td))
return (EFBIG);
- biosize = vp->v_mount->mnt_stat.f_iosize;
+ biosize = vp->v_bufobj.bo_bsize;
/*
* Find all of this file's B_NEEDCOMMIT buffers. If our writes
* would exceed the local maximum per-file write commit size when
* combined with those, we must decide whether to flush,
* go synchronous, or return error. We don't bother checking
* IO_UNIT -- we just make all writes atomic anyway, as there's
* no point optimizing for something that really won't ever happen.
*/
if (!(ioflag & IO_SYNC)) {
int nflag;
mtx_lock(&np->n_mtx);
nflag = np->n_flag;
mtx_unlock(&np->n_mtx);
int needrestart = 0;
if (nmp->nm_wcommitsize < uio->uio_resid) {
/*
* If this request could not possibly be completed
* without exceeding the maximum outstanding write
* commit size, see if we can convert it into a
* synchronous write operation.
*/
if (ioflag & IO_NDELAY)
return (EAGAIN);
ioflag |= IO_SYNC;
if (nflag & NMODIFIED)
needrestart = 1;
} else if (nflag & NMODIFIED) {
int wouldcommit = 0;
BO_LOCK(&vp->v_bufobj);
if (vp->v_bufobj.bo_dirty.bv_cnt != 0) {
TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd,
b_bobufs) {
if (bp->b_flags & B_NEEDCOMMIT)
wouldcommit += bp->b_bcount;
}
}
BO_UNLOCK(&vp->v_bufobj);
/*
* Since we're not operating synchronously and
* bypassing the buffer cache, we are in a commit
* and holding all of these buffers whether
* transmitted or not. If not limited, this
* will lead to the buffer cache deadlocking,
* as no one else can flush our uncommitted buffers.
*/
wouldcommit += uio->uio_resid;
/*
* If we would initially exceed the maximum
* outstanding write commit size, flush and restart.
*/
if (wouldcommit > nmp->nm_wcommitsize)
needrestart = 1;
}
if (needrestart)
goto flush_and_restart;
}
do {
NFSINCRGLOBAL(newnfsstats.biocache_writes);
lbn = uio->uio_offset / biosize;
on = uio->uio_offset & (biosize-1);
n = min((unsigned)(biosize - on), uio->uio_resid);
again:
/*
* Handle direct append and file extension cases, calculate
* unaligned buffer size.
*/
mtx_lock(&np->n_mtx);
if (uio->uio_offset == np->n_size && n) {
mtx_unlock(&np->n_mtx);
/*
* Get the buffer (in its pre-append state to maintain
* B_CACHE if it was previously set). Resize the
* nfsnode after we have locked the buffer to prevent
* readers from reading garbage.
*/
bcount = on;
bp = nfs_getcacheblk(vp, lbn, bcount, td);
if (bp != NULL) {
long save;
mtx_lock(&np->n_mtx);
np->n_size = uio->uio_offset + n;
np->n_flag |= NMODIFIED;
vnode_pager_setsize(vp, np->n_size);
mtx_unlock(&np->n_mtx);
save = bp->b_flags & B_CACHE;
bcount += n;
allocbuf(bp, bcount);
bp->b_flags |= save;
}
} else {
/*
* Obtain the locked cache block first, and then
* adjust the file's size as appropriate.
*/
bcount = on + n;
if ((off_t)lbn * biosize + bcount < np->n_size) {
if ((off_t)(lbn + 1) * biosize < np->n_size)
bcount = biosize;
else
bcount = np->n_size - (off_t)lbn * biosize;
}
mtx_unlock(&np->n_mtx);
bp = nfs_getcacheblk(vp, lbn, bcount, td);
mtx_lock(&np->n_mtx);
if (uio->uio_offset + n > np->n_size) {
np->n_size = uio->uio_offset + n;
np->n_flag |= NMODIFIED;
vnode_pager_setsize(vp, np->n_size);
}
mtx_unlock(&np->n_mtx);
}
if (!bp) {
error = newnfs_sigintr(nmp, td);
if (!error)
error = EINTR;
break;
}
/*
* Issue a READ if B_CACHE is not set. In special-append
* mode, B_CACHE is based on the buffer prior to the write
* op and is typically set, avoiding the read. If a read
* is required in special append mode, the server will
* probably send us a short-read since we extended the file
* on our end, resulting in b_resid == 0 and, thusly,
* B_CACHE getting set.
*
* We can also avoid issuing the read if the write covers
* the entire buffer. We have to make sure the buffer state
* is reasonable in this case since we will not be initiating
* I/O. See the comments in kern/vfs_bio.c's getblk() for
* more information.
*
* B_CACHE may also be set due to the buffer being cached
* normally.
*/
if (on == 0 && n == bcount) {
bp->b_flags |= B_CACHE;
bp->b_flags &= ~B_INVAL;
bp->b_ioflags &= ~BIO_ERROR;
}
if ((bp->b_flags & B_CACHE) == 0) {
bp->b_iocmd = BIO_READ;
vfs_busy_pages(bp, 0);
error = ncl_doio(vp, bp, cred, td, 0);
if (error) {
brelse(bp);
break;
}
}
if (bp->b_wcred == NOCRED)
bp->b_wcred = crhold(cred);
mtx_lock(&np->n_mtx);
np->n_flag |= NMODIFIED;
mtx_unlock(&np->n_mtx);
/*
* If dirtyend exceeds file size, chop it down. This should
* not normally occur but there is an append race where it
* might occur XXX, so we log it.
*
* If the chopping creates a reverse-indexed or degenerate
* situation with dirtyoff/end, we 0 both of them.
*/
if (bp->b_dirtyend > bcount) {
ncl_printf("NFS append race @%lx:%d\n",
(long)bp->b_blkno * DEV_BSIZE,
bp->b_dirtyend - bcount);
bp->b_dirtyend = bcount;
}
if (bp->b_dirtyoff >= bp->b_dirtyend)
bp->b_dirtyoff = bp->b_dirtyend = 0;
/*
* If the new write will leave a contiguous dirty
* area, just update the b_dirtyoff and b_dirtyend,
* otherwise force a write rpc of the old dirty area.
*
* While it is possible to merge discontiguous writes due to
* our having a B_CACHE buffer ( and thus valid read data
* for the hole), we don't because it could lead to
* significant cache coherency problems with multiple clients,
* especially if locking is implemented later on.
*
* as an optimization we could theoretically maintain
* a linked list of discontinuous areas, but we would still
* have to commit them separately so there isn't much
* advantage to it except perhaps a bit of asynchronization.
*/
if (bp->b_dirtyend > 0 &&
(on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) {
if (bwrite(bp) == EINTR) {
error = EINTR;
break;
}
goto again;
}
error = uiomove((char *)bp->b_data + on, n, uio);
/*
* Since this block is being modified, it must be written
* again and not just committed. Since write clustering does
* not work for the stage 1 data write, only the stage 2
* commit rpc, we have to clear B_CLUSTEROK as well.
*/
bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
if (error) {
bp->b_ioflags |= BIO_ERROR;
brelse(bp);
break;
}
/*
* Only update dirtyoff/dirtyend if not a degenerate
* condition.
*/
if (n) {
if (bp->b_dirtyend > 0) {
bp->b_dirtyoff = min(on, bp->b_dirtyoff);
bp->b_dirtyend = max((on + n), bp->b_dirtyend);
} else {
bp->b_dirtyoff = on;
bp->b_dirtyend = on + n;
}
vfs_bio_set_valid(bp, on, n);
}
/*
* If IO_SYNC do bwrite().
*
* IO_INVAL appears to be unused. The idea appears to be
* to turn off caching in this case. Very odd. XXX
*/
if ((ioflag & IO_SYNC)) {
if (ioflag & IO_INVAL)
bp->b_flags |= B_NOCACHE;
error = bwrite(bp);
if (error)
break;
} else if ((n + on) == biosize) {
bp->b_flags |= B_ASYNC;
(void) ncl_writebp(bp, 0, NULL);
} else {
bdwrite(bp);
}
} while (uio->uio_resid > 0 && n > 0);
return (error);
}
/*
* Get an nfs cache block.
*
* Allocate a new one if the block isn't currently in the cache
* and return the block marked busy. If the calling process is
* interrupted by a signal for an interruptible mount point, return
* NULL.
*
* The caller must carefully deal with the possible B_INVAL state of
* the buffer. ncl_doio() clears B_INVAL (and ncl_asyncio() clears it
* indirectly), so synchronous reads can be issued without worrying about
* the B_INVAL state. We have to be a little more careful when dealing
* with writes (see comments in nfs_write()) when extending a file past
* its EOF.
*/
static struct buf *
nfs_getcacheblk(struct vnode *vp, daddr_t bn, int size, struct thread *td)
{
struct buf *bp;
struct mount *mp;
struct nfsmount *nmp;
mp = vp->v_mount;
nmp = VFSTONFS(mp);
if (nmp->nm_flag & NFSMNT_INT) {
sigset_t oldset;
newnfs_set_sigmask(td, &oldset);
bp = getblk(vp, bn, size, NFS_PCATCH, 0, 0);
newnfs_restore_sigmask(td, &oldset);
while (bp == NULL) {
if (newnfs_sigintr(nmp, td))
return (NULL);
bp = getblk(vp, bn, size, 0, 2 * hz, 0);
}
} else {
bp = getblk(vp, bn, size, 0, 0, 0);
}
- if (vp->v_type == VREG) {
- int biosize;
-
- biosize = mp->mnt_stat.f_iosize;
- bp->b_blkno = bn * (biosize / DEV_BSIZE);
- }
+ if (vp->v_type == VREG)
+ bp->b_blkno = bn * (vp->v_bufobj.bo_bsize / DEV_BSIZE);
return (bp);
}
/*
* Flush and invalidate all dirty buffers. If another process is already
* doing the flush, just wait for completion.
*/
int
ncl_vinvalbuf(struct vnode *vp, int flags, struct thread *td, int intrflg)
{
struct nfsnode *np = VTONFS(vp);
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
int error = 0, slpflag, slptimeo;
int old_lock = 0;
ASSERT_VOP_LOCKED(vp, "ncl_vinvalbuf");
if ((nmp->nm_flag & NFSMNT_INT) == 0)
intrflg = 0;
if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF))
intrflg = 1;
if (intrflg) {
slpflag = NFS_PCATCH;
slptimeo = 2 * hz;
} else {
slpflag = 0;
slptimeo = 0;
}
old_lock = ncl_upgrade_vnlock(vp);
if (vp->v_iflag & VI_DOOMED) {
/*
* Since vgonel() uses the generic vinvalbuf() to flush
* dirty buffers and it does not call this function, it
* is safe to just return OK when VI_DOOMED is set.
*/
ncl_downgrade_vnlock(vp, old_lock);
return (0);
}
/*
* Now, flush as required.
*/
if ((flags & V_SAVE) && (vp->v_bufobj.bo_object != NULL)) {
VM_OBJECT_LOCK(vp->v_bufobj.bo_object);
vm_object_page_clean(vp->v_bufobj.bo_object, 0, 0, OBJPC_SYNC);
VM_OBJECT_UNLOCK(vp->v_bufobj.bo_object);
/*
* If the page clean was interrupted, fail the invalidation.
* Not doing so, we run the risk of losing dirty pages in the
* vinvalbuf() call below.
*/
if (intrflg && (error = newnfs_sigintr(nmp, td)))
goto out;
}
error = vinvalbuf(vp, flags, slpflag, 0);
while (error) {
if (intrflg && (error = newnfs_sigintr(nmp, td)))
goto out;
error = vinvalbuf(vp, flags, 0, slptimeo);
}
mtx_lock(&np->n_mtx);
if (np->n_directio_asyncwr == 0)
np->n_flag &= ~NMODIFIED;
mtx_unlock(&np->n_mtx);
out:
ncl_downgrade_vnlock(vp, old_lock);
return error;
}
/*
* Initiate asynchronous I/O. Return an error if no nfsiods are available.
* This is mainly to avoid queueing async I/O requests when the nfsiods
* are all hung on a dead server.
*
* Note: ncl_asyncio() does not clear (BIO_ERROR|B_INVAL) but when the bp
* is eventually dequeued by the async daemon, ncl_doio() *will*.
*/
int
ncl_asyncio(struct nfsmount *nmp, struct buf *bp, struct ucred *cred, struct thread *td)
{
int iod;
int gotiod;
int slpflag = 0;
int slptimeo = 0;
int error, error2;
/*
* Commits are usually short and sweet so lets save some cpu and
* leave the async daemons for more important rpc's (such as reads
* and writes).
*/
mtx_lock(&ncl_iod_mutex);
if (bp->b_iocmd == BIO_WRITE && (bp->b_flags & B_NEEDCOMMIT) &&
(nmp->nm_bufqiods > ncl_numasync / 2)) {
mtx_unlock(&ncl_iod_mutex);
return(EIO);
}
again:
if (nmp->nm_flag & NFSMNT_INT)
slpflag = NFS_PCATCH;
gotiod = FALSE;
/*
* Find a free iod to process this request.
*/
for (iod = 0; iod < ncl_numasync; iod++)
if (ncl_iodwant[iod] == NFSIOD_AVAILABLE) {
gotiod = TRUE;
break;
}
/*
* Try to create one if none are free.
*/
if (!gotiod)
ncl_nfsiodnew();
else {
/*
* Found one, so wake it up and tell it which
* mount to process.
*/
NFS_DPF(ASYNCIO, ("ncl_asyncio: waking iod %d for mount %p\n",
iod, nmp));
ncl_iodwant[iod] = NFSIOD_NOT_AVAILABLE;
ncl_iodmount[iod] = nmp;
nmp->nm_bufqiods++;
wakeup(&ncl_iodwant[iod]);
}
/*
* If none are free, we may already have an iod working on this mount
* point. If so, it will process our request.
*/
if (!gotiod) {
if (nmp->nm_bufqiods > 0) {
NFS_DPF(ASYNCIO,
("ncl_asyncio: %d iods are already processing mount %p\n",
nmp->nm_bufqiods, nmp));
gotiod = TRUE;
}
}
/*
* If we have an iod which can process the request, then queue
* the buffer.
*/
if (gotiod) {
/*
* Ensure that the queue never grows too large. We still want
* to asynchronize so we block rather then return EIO.
*/
while (nmp->nm_bufqlen >= 2*ncl_numasync) {
NFS_DPF(ASYNCIO,
("ncl_asyncio: waiting for mount %p queue to drain\n", nmp));
nmp->nm_bufqwant = TRUE;
error = newnfs_msleep(td, &nmp->nm_bufq,
&ncl_iod_mutex, slpflag | PRIBIO, "nfsaio",
slptimeo);
if (error) {
error2 = newnfs_sigintr(nmp, td);
if (error2) {
mtx_unlock(&ncl_iod_mutex);
return (error2);
}
if (slpflag == NFS_PCATCH) {
slpflag = 0;
slptimeo = 2 * hz;
}
}
/*
* We might have lost our iod while sleeping,
* so check and loop if nescessary.
*/
goto again;
}
/* We might have lost our nfsiod */
if (nmp->nm_bufqiods == 0) {
NFS_DPF(ASYNCIO,
("ncl_asyncio: no iods after mount %p queue was drained, looping\n", nmp));
goto again;
}
if (bp->b_iocmd == BIO_READ) {
if (bp->b_rcred == NOCRED && cred != NOCRED)
bp->b_rcred = crhold(cred);
} else {
if (bp->b_wcred == NOCRED && cred != NOCRED)
bp->b_wcred = crhold(cred);
}
if (bp->b_flags & B_REMFREE)
bremfreef(bp);
BUF_KERNPROC(bp);
TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist);
nmp->nm_bufqlen++;
if ((bp->b_flags & B_DIRECT) && bp->b_iocmd == BIO_WRITE) {
mtx_lock(&(VTONFS(bp->b_vp))->n_mtx);
VTONFS(bp->b_vp)->n_flag |= NMODIFIED;
VTONFS(bp->b_vp)->n_directio_asyncwr++;
mtx_unlock(&(VTONFS(bp->b_vp))->n_mtx);
}
mtx_unlock(&ncl_iod_mutex);
return (0);
}
mtx_unlock(&ncl_iod_mutex);
/*
* All the iods are busy on other mounts, so return EIO to
* force the caller to process the i/o synchronously.
*/
NFS_DPF(ASYNCIO, ("ncl_asyncio: no iods available, i/o is synchronous\n"));
return (EIO);
}
void
ncl_doio_directwrite(struct buf *bp)
{
int iomode, must_commit;
struct uio *uiop = (struct uio *)bp->b_caller1;
char *iov_base = uiop->uio_iov->iov_base;
iomode = NFSWRITE_FILESYNC;
uiop->uio_td = NULL; /* NULL since we're in nfsiod */
ncl_writerpc(bp->b_vp, uiop, bp->b_wcred, &iomode, &must_commit, 0);
KASSERT((must_commit == 0), ("ncl_doio_directwrite: Did not commit write"));
free(iov_base, M_NFSDIRECTIO);
free(uiop->uio_iov, M_NFSDIRECTIO);
free(uiop, M_NFSDIRECTIO);
if ((bp->b_flags & B_DIRECT) && bp->b_iocmd == BIO_WRITE) {
struct nfsnode *np = VTONFS(bp->b_vp);
mtx_lock(&np->n_mtx);
np->n_directio_asyncwr--;
if (np->n_directio_asyncwr == 0) {
np->n_flag &= ~NMODIFIED;
if ((np->n_flag & NFSYNCWAIT)) {
np->n_flag &= ~NFSYNCWAIT;
wakeup((caddr_t)&np->n_directio_asyncwr);
}
}
mtx_unlock(&np->n_mtx);
}
bp->b_vp = NULL;
relpbuf(bp, &ncl_pbuf_freecnt);
}
/*
* Do an I/O operation to/from a cache block. This may be called
* synchronously or from an nfsiod.
*/
int
ncl_doio(struct vnode *vp, struct buf *bp, struct ucred *cr, struct thread *td,
int called_from_strategy)
{
struct uio *uiop;
struct nfsnode *np;
struct nfsmount *nmp;
int error = 0, iomode, must_commit = 0;
struct uio uio;
struct iovec io;
struct proc *p = td ? td->td_proc : NULL;
uint8_t iocmd;
np = VTONFS(vp);
nmp = VFSTONFS(vp->v_mount);
uiop = &uio;
uiop->uio_iov = &io;
uiop->uio_iovcnt = 1;
uiop->uio_segflg = UIO_SYSSPACE;
uiop->uio_td = td;
/*
* clear BIO_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_INVAL;
bp->b_ioflags &= ~BIO_ERROR;
KASSERT(!(bp->b_flags & B_DONE), ("ncl_doio: bp %p already marked done", bp));
iocmd = bp->b_iocmd;
if (iocmd == BIO_READ) {
io.iov_len = uiop->uio_resid = bp->b_bcount;
io.iov_base = bp->b_data;
uiop->uio_rw = UIO_READ;
switch (vp->v_type) {
case VREG:
uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE;
NFSINCRGLOBAL(newnfsstats.read_bios);
error = ncl_readrpc(vp, uiop, cr);
if (!error) {
if (uiop->uio_resid) {
/*
* If we had a short read with no error, we must have
* hit a file hole. We should zero-fill the remainder.
* This can also occur if the server hits the file EOF.
*
* Holes used to be able to occur due to pending
* writes, but that is not possible any longer.
*/
int nread = bp->b_bcount - uiop->uio_resid;
int left = uiop->uio_resid;
if (left > 0)
bzero((char *)bp->b_data + nread, left);
uiop->uio_resid = 0;
}
}
/* ASSERT_VOP_LOCKED(vp, "ncl_doio"); */
if (p && (vp->v_vflag & VV_TEXT)) {
mtx_lock(&np->n_mtx);
if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &np->n_vattr.na_mtime)) {
mtx_unlock(&np->n_mtx);
PROC_LOCK(p);
killproc(p, "text file modification");
PROC_UNLOCK(p);
} else
mtx_unlock(&np->n_mtx);
}
break;
case VLNK:
uiop->uio_offset = (off_t)0;
NFSINCRGLOBAL(newnfsstats.readlink_bios);
error = ncl_readlinkrpc(vp, uiop, cr);
break;
case VDIR:
NFSINCRGLOBAL(newnfsstats.readdir_bios);
uiop->uio_offset = ((u_quad_t)bp->b_lblkno) * NFS_DIRBLKSIZ;
if ((nmp->nm_flag & NFSMNT_RDIRPLUS) != 0) {
error = ncl_readdirplusrpc(vp, uiop, cr, td);
if (error == NFSERR_NOTSUPP)
nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
}
if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0)
error = ncl_readdirrpc(vp, uiop, cr, td);
/*
* end-of-directory sets B_INVAL but does not generate an
* error.
*/
if (error == 0 && uiop->uio_resid == bp->b_bcount)
bp->b_flags |= B_INVAL;
break;
default:
ncl_printf("ncl_doio: type %x unexpected\n", vp->v_type);
break;
};
if (error) {
bp->b_ioflags |= BIO_ERROR;
bp->b_error = error;
}
} else {
/*
* If we only need to commit, try to commit
*/
if (bp->b_flags & B_NEEDCOMMIT) {
int retv;
off_t off;
off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff;
retv = ncl_commit(vp, off, bp->b_dirtyend-bp->b_dirtyoff,
bp->b_wcred, td);
if (retv == 0) {
bp->b_dirtyoff = bp->b_dirtyend = 0;
bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
bp->b_resid = 0;
bufdone(bp);
return (0);
}
if (retv == NFSERR_STALEWRITEVERF) {
ncl_clearcommit(vp->v_mount);
}
}
/*
* Setup for actual write
*/
mtx_lock(&np->n_mtx);
if ((off_t)bp->b_blkno * DEV_BSIZE + bp->b_dirtyend > np->n_size)
bp->b_dirtyend = np->n_size - (off_t)bp->b_blkno * DEV_BSIZE;
mtx_unlock(&np->n_mtx);
if (bp->b_dirtyend > bp->b_dirtyoff) {
io.iov_len = uiop->uio_resid = bp->b_dirtyend
- bp->b_dirtyoff;
uiop->uio_offset = (off_t)bp->b_blkno * DEV_BSIZE
+ bp->b_dirtyoff;
io.iov_base = (char *)bp->b_data + bp->b_dirtyoff;
uiop->uio_rw = UIO_WRITE;
NFSINCRGLOBAL(newnfsstats.write_bios);
if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE | B_CLUSTER)) == B_ASYNC)
iomode = NFSWRITE_UNSTABLE;
else
iomode = NFSWRITE_FILESYNC;
error = ncl_writerpc(vp, uiop, cr, &iomode, &must_commit,
called_from_strategy);
/*
* When setting B_NEEDCOMMIT also set B_CLUSTEROK to try
* to cluster the buffers needing commit. This will allow
* the system to submit a single commit rpc for the whole
* cluster. We can do this even if the buffer is not 100%
* dirty (relative to the NFS blocksize), so we optimize the
* append-to-file-case.
*
* (when clearing B_NEEDCOMMIT, B_CLUSTEROK must also be
* cleared because write clustering only works for commit
* rpc's, not for the data portion of the write).
*/
if (!error && iomode == NFSWRITE_UNSTABLE) {
bp->b_flags |= B_NEEDCOMMIT;
if (bp->b_dirtyoff == 0
&& bp->b_dirtyend == bp->b_bcount)
bp->b_flags |= B_CLUSTEROK;
} else {
bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
}
/*
* For an interrupted write, the buffer is still valid
* and the write hasn't been pushed to the server yet,
* so we can't set BIO_ERROR and report the interruption
* by setting B_EINTR. For the B_ASYNC case, B_EINTR
* is not relevant, so the rpc attempt is essentially
* a noop. For the case of a V3 write rpc not being
* committed to stable storage, the block is still
* dirty and requires either a commit rpc or another
* write rpc with iomode == NFSV3WRITE_FILESYNC before
* the block is reused. This is indicated by setting
* the B_DELWRI and B_NEEDCOMMIT flags.
*
* EIO is returned by ncl_writerpc() to indicate a recoverable
* write error and is handled as above, except that
* B_EINTR isn't set. One cause of this is a stale stateid
* error for the RPC that indicates recovery is required,
* when called with called_from_strategy != 0.
*
* If the buffer is marked B_PAGING, it does not reside on
* the vp's paging queues so we cannot call bdirty(). The
* bp in this case is not an NFS cache block so we should
* be safe. XXX
*
* The logic below breaks up errors into recoverable and
* unrecoverable. For the former, we clear B_INVAL|B_NOCACHE
* and keep the buffer around for potential write retries.
* For the latter (eg ESTALE), we toss the buffer away (B_INVAL)
* and save the error in the nfsnode. This is less than ideal
* but necessary. Keeping such buffers around could potentially
* cause buffer exhaustion eventually (they can never be written
* out, so will get constantly be re-dirtied). It also causes
* all sorts of vfs panics. For non-recoverable write errors,
* also invalidate the attrcache, so we'll be forced to go over
* the wire for this object, returning an error to user on next
* call (most of the time).
*/
if (error == EINTR || error == EIO || error == ETIMEDOUT
|| (!error && (bp->b_flags & B_NEEDCOMMIT))) {
int s;
s = splbio();
bp->b_flags &= ~(B_INVAL|B_NOCACHE);
if ((bp->b_flags & B_PAGING) == 0) {
bdirty(bp);
bp->b_flags &= ~B_DONE;
}
if ((error == EINTR || error == ETIMEDOUT) &&
(bp->b_flags & B_ASYNC) == 0)
bp->b_flags |= B_EINTR;
splx(s);
} else {
if (error) {
bp->b_ioflags |= BIO_ERROR;
bp->b_flags |= B_INVAL;
bp->b_error = np->n_error = error;
mtx_lock(&np->n_mtx);
np->n_flag |= NWRITEERR;
np->n_attrstamp = 0;
mtx_unlock(&np->n_mtx);
}
bp->b_dirtyoff = bp->b_dirtyend = 0;
}
} else {
bp->b_resid = 0;
bufdone(bp);
return (0);
}
}
bp->b_resid = uiop->uio_resid;
if (must_commit)
ncl_clearcommit(vp->v_mount);
bufdone(bp);
return (error);
}
/*
* Used to aid in handling ftruncate() operations on the NFS client side.
* Truncation creates a number of special problems for NFS. We have to
* throw away VM pages and buffer cache buffers that are beyond EOF, and
* we have to properly handle VM pages or (potentially dirty) buffers
* that straddle the truncation point.
*/
int
ncl_meta_setsize(struct vnode *vp, struct ucred *cred, struct thread *td, u_quad_t nsize)
{
struct nfsnode *np = VTONFS(vp);
u_quad_t tsize;
- int biosize = vp->v_mount->mnt_stat.f_iosize;
+ int biosize = vp->v_bufobj.bo_bsize;
int error = 0;
mtx_lock(&np->n_mtx);
tsize = np->n_size;
np->n_size = nsize;
mtx_unlock(&np->n_mtx);
if (nsize < tsize) {
struct buf *bp;
daddr_t lbn;
int bufsize;
/*
* vtruncbuf() doesn't get the buffer overlapping the
* truncation point. We may have a B_DELWRI and/or B_CACHE
* buffer that now needs to be truncated.
*/
error = vtruncbuf(vp, cred, td, nsize, biosize);
lbn = nsize / biosize;
bufsize = nsize & (biosize - 1);
bp = nfs_getcacheblk(vp, lbn, bufsize, td);
if (!bp)
return EINTR;
if (bp->b_dirtyoff > bp->b_bcount)
bp->b_dirtyoff = bp->b_bcount;
if (bp->b_dirtyend > bp->b_bcount)
bp->b_dirtyend = bp->b_bcount;
bp->b_flags |= B_RELBUF; /* don't leave garbage around */
brelse(bp);
} else {
vnode_pager_setsize(vp, nsize);
}
return(error);
}
Index: stable/8/sys/fs/nfsclient/nfs_clnode.c
===================================================================
--- stable/8/sys/fs/nfsclient/nfs_clnode.c (revision 231331)
+++ stable/8/sys/fs/nfsclient/nfs_clnode.c (revision 231332)
@@ -1,323 +1,324 @@
/*-
* Copyright (c) 1989, 1993
* The Regents of the University of California. All rights reserved.
*
* This code is derived from software contributed to Berkeley by
* Rick Macklem at The University of Guelph.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* from nfs_node.c 8.6 (Berkeley) 5/22/95
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/fcntl.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mount.h>
#include <sys/namei.h>
#include <sys/proc.h>
#include <sys/socket.h>
#include <sys/sysctl.h>
#include <sys/taskqueue.h>
#include <sys/vnode.h>
#include <vm/uma.h>
#include <fs/nfs/nfsport.h>
#include <fs/nfsclient/nfsnode.h>
#include <fs/nfsclient/nfsmount.h>
#include <fs/nfsclient/nfs.h>
#include <nfs/nfs_lock.h>
extern struct vop_vector newnfs_vnodeops;
extern struct buf_ops buf_ops_newnfs;
MALLOC_DECLARE(M_NEWNFSREQ);
uma_zone_t newnfsnode_zone;
static void nfs_freesillyrename(void *arg, __unused int pending);
void
ncl_nhinit(void)
{
newnfsnode_zone = uma_zcreate("NCLNODE", sizeof(struct nfsnode), NULL,
NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
}
void
ncl_nhuninit(void)
{
uma_zdestroy(newnfsnode_zone);
}
/*
* ONLY USED FOR THE ROOT DIRECTORY. nfscl_nget() does the rest. If this
* function is going to be used to get Regular Files, code must be added
* to fill in the "struct nfsv4node".
* Look up a vnode/nfsnode by file handle.
* Callers must check for mount points!!
* In all cases, a pointer to a
* nfsnode structure is returned.
*/
int
ncl_nget(struct mount *mntp, u_int8_t *fhp, int fhsize, struct nfsnode **npp,
int lkflags)
{
struct thread *td = curthread; /* XXX */
struct nfsnode *np;
struct vnode *vp;
struct vnode *nvp;
int error;
u_int hash;
struct nfsmount *nmp;
struct nfsfh *nfhp;
nmp = VFSTONFS(mntp);
*npp = NULL;
hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
M_NFSFH, M_WAITOK);
bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
nfhp->nfh_len = fhsize;
error = vfs_hash_get(mntp, hash, lkflags,
td, &nvp, newnfs_vncmpf, nfhp);
FREE(nfhp, M_NFSFH);
if (error)
return (error);
if (nvp != NULL) {
*npp = VTONFS(nvp);
return (0);
}
/*
* Allocate before getnewvnode since doing so afterward
* might cause a bogus v_data pointer to get dereferenced
* elsewhere if zalloc should block.
*/
np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
error = getnewvnode("newnfs", mntp, &newnfs_vnodeops, &nvp);
if (error) {
uma_zfree(newnfsnode_zone, np);
return (error);
}
vp = nvp;
+ KASSERT(vp->v_bufobj.bo_bsize != 0, ("ncl_nget: bo_bsize == 0"));
vp->v_bufobj.bo_ops = &buf_ops_newnfs;
vp->v_data = np;
np->n_vnode = vp;
/*
* Initialize the mutex even if the vnode is going to be a loser.
* This simplifies the logic in reclaim, which can then unconditionally
* destroy the mutex (in the case of the loser, or if hash_insert
* happened to return an error no special casing is needed).
*/
mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
/*
* NFS supports recursive and shared locking.
*/
VN_LOCK_AREC(vp);
VN_LOCK_ASHARE(vp);
/*
* Are we getting the root? If so, make sure the vnode flags
* are correct
*/
if ((fhsize == nmp->nm_fhsize) &&
!bcmp(fhp, nmp->nm_fh, fhsize)) {
if (vp->v_type == VNON)
vp->v_type = VDIR;
vp->v_vflag |= VV_ROOT;
}
MALLOC(np->n_fhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
M_NFSFH, M_WAITOK);
bcopy(fhp, np->n_fhp->nfh_fh, fhsize);
np->n_fhp->nfh_len = fhsize;
lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
error = insmntque(vp, mntp);
if (error != 0) {
*npp = NULL;
FREE((caddr_t)np->n_fhp, M_NFSFH);
mtx_destroy(&np->n_mtx);
uma_zfree(newnfsnode_zone, np);
return (error);
}
error = vfs_hash_insert(vp, hash, lkflags,
td, &nvp, newnfs_vncmpf, np->n_fhp);
if (error)
return (error);
if (nvp != NULL) {
*npp = VTONFS(nvp);
/* vfs_hash_insert() vput()'s the losing vnode */
return (0);
}
*npp = np;
return (0);
}
/*
* Do the vrele(sp->s_dvp) as a separate task in order to avoid a
* deadlock because of a LOR when vrele() locks the directory vnode.
*/
static void
nfs_freesillyrename(void *arg, __unused int pending)
{
struct sillyrename *sp;
sp = arg;
vrele(sp->s_dvp);
free(sp, M_NEWNFSREQ);
}
int
ncl_inactive(struct vop_inactive_args *ap)
{
struct nfsnode *np;
struct sillyrename *sp;
struct vnode *vp = ap->a_vp;
np = VTONFS(vp);
if (NFS_ISV4(vp) && vp->v_type == VREG) {
/*
* Since mmap()'d files do I/O after VOP_CLOSE(), the NFSv4
* Close operations are delayed until now. Any dirty buffers
* must be flushed before the close, so that the stateid is
* available for the writes.
*/
(void) ncl_flush(vp, MNT_WAIT, NULL, ap->a_td, 1, 0);
(void) nfsrpc_close(vp, 1, ap->a_td);
}
mtx_lock(&np->n_mtx);
if (vp->v_type != VDIR) {
sp = np->n_sillyrename;
np->n_sillyrename = NULL;
} else
sp = NULL;
if (sp) {
mtx_unlock(&np->n_mtx);
(void) ncl_vinvalbuf(vp, 0, ap->a_td, 1);
/*
* Remove the silly file that was rename'd earlier
*/
ncl_removeit(sp, vp);
crfree(sp->s_cred);
TASK_INIT(&sp->s_task, 0, nfs_freesillyrename, sp);
taskqueue_enqueue(taskqueue_thread, &sp->s_task);
mtx_lock(&np->n_mtx);
}
np->n_flag &= NMODIFIED;
mtx_unlock(&np->n_mtx);
return (0);
}
/*
* Reclaim an nfsnode so that it can be used for other purposes.
*/
int
ncl_reclaim(struct vop_reclaim_args *ap)
{
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
struct nfsdmap *dp, *dp2;
if (NFS_ISV4(vp) && vp->v_type == VREG)
/*
* Since mmap()'d files do I/O after VOP_CLOSE(), the NFSv4
* Close operations are delayed until ncl_inactive().
* However, since VOP_INACTIVE() is not guaranteed to be
* called, we need to do it again here.
*/
(void) nfsrpc_close(vp, 1, ap->a_td);
/*
* If the NLM is running, give it a chance to abort pending
* locks.
*/
if (nfs_reclaim_p != NULL)
nfs_reclaim_p(ap);
/*
* Destroy the vm object and flush associated pages.
*/
vnode_destroy_vobject(vp);
vfs_hash_remove(vp);
/*
* Call nfscl_reclaimnode() to save attributes in the delegation,
* as required.
*/
if (vp->v_type == VREG)
nfscl_reclaimnode(vp);
/*
* Free up any directory cookie structures and
* large file handle structures that might be associated with
* this nfs node.
*/
if (vp->v_type == VDIR) {
dp = LIST_FIRST(&np->n_cookies);
while (dp) {
dp2 = dp;
dp = LIST_NEXT(dp, ndm_list);
FREE((caddr_t)dp2, M_NFSDIROFF);
}
}
FREE((caddr_t)np->n_fhp, M_NFSFH);
if (np->n_v4 != NULL)
FREE((caddr_t)np->n_v4, M_NFSV4NODE);
mtx_destroy(&np->n_mtx);
uma_zfree(newnfsnode_zone, vp->v_data);
vp->v_data = NULL;
return (0);
}
/*
* Invalidate both the access and attribute caches for this vnode.
*/
void
ncl_invalcaches(struct vnode *vp)
{
struct nfsnode *np = VTONFS(vp);
int i;
mtx_lock(&np->n_mtx);
for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
np->n_accesscache[i].stamp = 0;
np->n_attrstamp = 0;
mtx_unlock(&np->n_mtx);
}
Index: stable/8/sys/fs/nfsclient/nfs_clport.c
===================================================================
--- stable/8/sys/fs/nfsclient/nfs_clport.c (revision 231331)
+++ stable/8/sys/fs/nfsclient/nfs_clport.c (revision 231332)
@@ -1,1301 +1,1302 @@
/*-
* Copyright (c) 1989, 1993
* The Regents of the University of California. All rights reserved.
*
* This code is derived from software contributed to Berkeley by
* Rick Macklem at The University of Guelph.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_inet6.h"
/*
* generally, I don't like #includes inside .h files, but it seems to
* be the easiest way to handle the port.
*/
#include <sys/hash.h>
#include <fs/nfs/nfsport.h>
#include <netinet/if_ether.h>
#include <net/if_types.h>
extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
extern struct vop_vector newnfs_vnodeops;
extern struct vop_vector newnfs_fifoops;
extern uma_zone_t newnfsnode_zone;
extern struct buf_ops buf_ops_newnfs;
extern int ncl_pbuf_freecnt;
extern short nfsv4_cbport;
extern int nfscl_enablecallb;
extern int nfs_numnfscbd;
extern int nfscl_inited;
struct mtx nfs_clstate_mutex;
struct mtx ncl_iod_mutex;
NFSDLOCKMUTEX;
extern void (*ncl_call_invalcaches)(struct vnode *);
/*
* Comparison function for vfs_hash functions.
*/
int
newnfs_vncmpf(struct vnode *vp, void *arg)
{
struct nfsfh *nfhp = (struct nfsfh *)arg;
struct nfsnode *np = VTONFS(vp);
if (np->n_fhp->nfh_len != nfhp->nfh_len ||
NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
return (1);
return (0);
}
/*
* Look up a vnode/nfsnode by file handle.
* Callers must check for mount points!!
* In all cases, a pointer to a
* nfsnode structure is returned.
* This variant takes a "struct nfsfh *" as second argument and uses
* that structure up, either by hanging off the nfsnode or FREEing it.
*/
int
nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
struct componentname *cnp, struct thread *td, struct nfsnode **npp,
void *stuff, int lkflags)
{
struct nfsnode *np, *dnp;
struct vnode *vp, *nvp;
struct nfsv4node *newd, *oldd;
int error;
u_int hash;
struct nfsmount *nmp;
nmp = VFSTONFS(mntp);
dnp = VTONFS(dvp);
*npp = NULL;
hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
error = vfs_hash_get(mntp, hash, lkflags,
td, &nvp, newnfs_vncmpf, nfhp);
if (error == 0 && nvp != NULL) {
/*
* I believe there is a slight chance that vgonel() could
* get called on this vnode between when NFSVOPLOCK() drops
* the VI_LOCK() and vget() acquires it again, so that it
* hasn't yet had v_usecount incremented. If this were to
* happen, the VI_DOOMED flag would be set, so check for
* that here. Since we now have the v_usecount incremented,
* we should be ok until we vrele() it, if the VI_DOOMED
* flag isn't set now.
*/
VI_LOCK(nvp);
if ((nvp->v_iflag & VI_DOOMED)) {
VI_UNLOCK(nvp);
vrele(nvp);
error = ENOENT;
} else {
VI_UNLOCK(nvp);
}
}
if (error) {
FREE((caddr_t)nfhp, M_NFSFH);
return (error);
}
if (nvp != NULL) {
np = VTONFS(nvp);
/*
* For NFSv4, check to see if it is the same name and
* replace the name, if it is different.
*/
oldd = newd = NULL;
if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
nvp->v_type == VREG &&
(np->n_v4->n4_namelen != cnp->cn_namelen ||
NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
cnp->cn_namelen) ||
dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
dnp->n_fhp->nfh_len))) {
MALLOC(newd, struct nfsv4node *,
sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
+ cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
NFSLOCKNODE(np);
if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
&& (np->n_v4->n4_namelen != cnp->cn_namelen ||
NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
cnp->cn_namelen) ||
dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
dnp->n_fhp->nfh_len))) {
oldd = np->n_v4;
np->n_v4 = newd;
newd = NULL;
np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
np->n_v4->n4_namelen = cnp->cn_namelen;
NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
dnp->n_fhp->nfh_len);
NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
cnp->cn_namelen);
}
NFSUNLOCKNODE(np);
}
if (newd != NULL)
FREE((caddr_t)newd, M_NFSV4NODE);
if (oldd != NULL)
FREE((caddr_t)oldd, M_NFSV4NODE);
*npp = np;
FREE((caddr_t)nfhp, M_NFSFH);
return (0);
}
/*
* Allocate before getnewvnode since doing so afterward
* might cause a bogus v_data pointer to get dereferenced
* elsewhere if zalloc should block.
*/
np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
error = getnewvnode("newnfs", mntp, &newnfs_vnodeops, &nvp);
if (error) {
uma_zfree(newnfsnode_zone, np);
FREE((caddr_t)nfhp, M_NFSFH);
return (error);
}
vp = nvp;
+ KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
vp->v_bufobj.bo_ops = &buf_ops_newnfs;
vp->v_data = np;
np->n_vnode = vp;
/*
* Initialize the mutex even if the vnode is going to be a loser.
* This simplifies the logic in reclaim, which can then unconditionally
* destroy the mutex (in the case of the loser, or if hash_insert
* happened to return an error no special casing is needed).
*/
mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
/*
* Are we getting the root? If so, make sure the vnode flags
* are correct
*/
if ((nfhp->nfh_len == nmp->nm_fhsize) &&
!bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
if (vp->v_type == VNON)
vp->v_type = VDIR;
vp->v_vflag |= VV_ROOT;
}
np->n_fhp = nfhp;
/*
* For NFSv4, we have to attach the directory file handle and
* file name, so that Open Ops can be done later.
*/
if (nmp->nm_flag & NFSMNT_NFSV4) {
MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
+ dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
M_WAITOK);
np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
np->n_v4->n4_namelen = cnp->cn_namelen;
NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
dnp->n_fhp->nfh_len);
NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
cnp->cn_namelen);
} else {
np->n_v4 = NULL;
}
/*
* NFS supports recursive and shared locking.
*/
VN_LOCK_AREC(vp);
VN_LOCK_ASHARE(vp);
lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
error = insmntque(vp, mntp);
if (error != 0) {
*npp = NULL;
mtx_destroy(&np->n_mtx);
FREE((caddr_t)nfhp, M_NFSFH);
if (np->n_v4 != NULL)
FREE((caddr_t)np->n_v4, M_NFSV4NODE);
uma_zfree(newnfsnode_zone, np);
return (error);
}
error = vfs_hash_insert(vp, hash, lkflags,
td, &nvp, newnfs_vncmpf, nfhp);
if (error)
return (error);
if (nvp != NULL) {
*npp = VTONFS(nvp);
/* vfs_hash_insert() vput()'s the losing vnode */
return (0);
}
*npp = np;
return (0);
}
/*
* Anothe variant of nfs_nget(). This one is only used by reopen. It
* takes almost the same args as nfs_nget(), but only succeeds if an entry
* exists in the cache. (Since files should already be "open" with a
* vnode ref cnt on the node when reopen calls this, it should always
* succeed.)
* Also, don't get a vnode lock, since it may already be locked by some
* other process that is handling it. This is ok, since all other threads
* on the client are blocked by the nfsc_lock being exclusively held by the
* caller of this function.
*/
int
nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
struct thread *td, struct nfsnode **npp)
{
struct vnode *nvp;
u_int hash;
struct nfsfh *nfhp;
int error;
*npp = NULL;
/* For forced dismounts, just return error. */
if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
return (EINTR);
MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
M_NFSFH, M_WAITOK);
bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
nfhp->nfh_len = fhsize;
hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
/*
* First, try to get the vnode locked, but don't block for the lock.
*/
error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
newnfs_vncmpf, nfhp);
if (error == 0 && nvp != NULL) {
NFSVOPUNLOCK(nvp, 0);
} else if (error == EBUSY) {
/*
* The LK_EXCLOTHER lock type tells nfs_lock1() to not try
* and lock the vnode, but just get a v_usecount on it.
* LK_NOWAIT is set so that when vget() returns ENOENT,
* vfs_hash_get() fails instead of looping.
* If this succeeds, it is safe so long as a vflush() with
* FORCECLOSE has not been done. Since the Renew thread is
* stopped and the MNTK_UNMOUNTF flag is set before doing
* a vflush() with FORCECLOSE, we should be ok here.
*/
if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
error = EINTR;
else
error = vfs_hash_get(mntp, hash,
(LK_EXCLOTHER | LK_NOWAIT), td, &nvp,
newnfs_vncmpf, nfhp);
}
FREE(nfhp, M_NFSFH);
if (error)
return (error);
if (nvp != NULL) {
*npp = VTONFS(nvp);
return (0);
}
return (EINVAL);
}
/*
* Load the attribute cache (that lives in the nfsnode entry) with
* the attributes of the second argument and
* Iff vaper not NULL
* copy the attributes to *vaper
* Similar to nfs_loadattrcache(), except the attributes are passed in
* instead of being parsed out of the mbuf list.
*/
int
nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
void *stuff, int writeattr, int dontshrink)
{
struct vnode *vp = *vpp;
struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
struct nfsnode *np;
struct nfsmount *nmp;
struct timespec mtime_save;
/*
* If v_type == VNON it is a new node, so fill in the v_type,
* n_mtime fields. Check to see if it represents a special
* device, and if so, check for a possible alias. Once the
* correct vnode has been obtained, fill in the rest of the
* information.
*/
np = VTONFS(vp);
NFSLOCKNODE(np);
if (vp->v_type != nvap->va_type) {
vp->v_type = nvap->va_type;
if (vp->v_type == VFIFO)
vp->v_op = &newnfs_fifoops;
np->n_mtime = nvap->va_mtime;
}
nmp = VFSTONFS(vp->v_mount);
vap = &np->n_vattr.na_vattr;
mtime_save = vap->va_mtime;
if (writeattr) {
np->n_vattr.na_filerev = nap->na_filerev;
np->n_vattr.na_size = nap->na_size;
np->n_vattr.na_mtime = nap->na_mtime;
np->n_vattr.na_ctime = nap->na_ctime;
np->n_vattr.na_fsid = nap->na_fsid;
np->n_vattr.na_mode = nap->na_mode;
} else {
NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
sizeof (struct nfsvattr));
}
/*
* For NFSv4, if the node's fsid is not equal to the mount point's
* fsid, return the low order 32bits of the node's fsid. This
* allows getcwd(3) to work. There is a chance that the fsid might
* be the same as a local fs, but since this is in an NFS mount
* point, I don't think that will cause any problems?
*/
if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
(nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
/*
* va_fsid needs to be set to some value derived from
* np->n_vattr.na_filesid that is not equal
* vp->v_mount->mnt_stat.f_fsid[0], so that it changes
* from the value used for the top level server volume
* in the mounted subtree.
*/
if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
(uint32_t)np->n_vattr.na_filesid[0])
vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
else
vap->va_fsid = (uint32_t)hash32_buf(
np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
} else
vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
np->n_attrstamp = time_second;
if (vap->va_size != np->n_size) {
if (vap->va_type == VREG) {
if (dontshrink && vap->va_size < np->n_size) {
/*
* We've been told not to shrink the file;
* zero np->n_attrstamp to indicate that
* the attributes are stale.
*/
vap->va_size = np->n_size;
np->n_attrstamp = 0;
} else if (np->n_flag & NMODIFIED) {
/*
* We've modified the file: Use the larger
* of our size, and the server's size.
*/
if (vap->va_size < np->n_size) {
vap->va_size = np->n_size;
} else {
np->n_size = vap->va_size;
np->n_flag |= NSIZECHANGED;
}
} else {
np->n_size = vap->va_size;
np->n_flag |= NSIZECHANGED;
}
vnode_pager_setsize(vp, np->n_size);
} else {
np->n_size = vap->va_size;
}
}
/*
* The following checks are added to prevent a race between (say)
* a READDIR+ and a WRITE.
* READDIR+, WRITE requests sent out.
* READDIR+ resp, WRITE resp received on client.
* However, the WRITE resp was handled before the READDIR+ resp
* causing the post op attrs from the write to be loaded first
* and the attrs from the READDIR+ to be loaded later. If this
* happens, we have stale attrs loaded into the attrcache.
* We detect this by for the mtime moving back. We invalidate the
* attrcache when this happens.
*/
if (timespeccmp(&mtime_save, &vap->va_mtime, >))
/* Size changed or mtime went backwards */
np->n_attrstamp = 0;
if (vaper != NULL) {
NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
if (np->n_flag & NCHG) {
if (np->n_flag & NACC)
vaper->va_atime = np->n_atim;
if (np->n_flag & NUPD)
vaper->va_mtime = np->n_mtim;
}
}
NFSUNLOCKNODE(np);
return (0);
}
/*
* Fill in the client id name. For these bytes:
* 1 - they must be unique
* 2 - they should be persistent across client reboots
* 1 is more critical than 2
* Use the mount point's unique id plus either the uuid or, if that
* isn't set, random junk.
*/
void
nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
{
int uuidlen;
/*
* First, put in the 64bit mount point identifier.
*/
if (idlen >= sizeof (u_int64_t)) {
NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
cp += sizeof (u_int64_t);
idlen -= sizeof (u_int64_t);
}
/*
* If uuid is non-zero length, use it.
*/
uuidlen = strlen(uuid);
if (uuidlen > 0 && idlen >= uuidlen) {
NFSBCOPY(uuid, cp, uuidlen);
cp += uuidlen;
idlen -= uuidlen;
}
/*
* This only normally happens if the uuid isn't set.
*/
while (idlen > 0) {
*cp++ = (u_int8_t)(arc4random() % 256);
idlen--;
}
}
/*
* Fill in a lock owner name. For now, pid + the process's creation time.
*/
void
nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
{
union {
u_int32_t lval;
u_int8_t cval[4];
} tl;
struct proc *p;
if (id == NULL) {
printf("NULL id\n");
bzero(cp, NFSV4CL_LOCKNAMELEN);
return;
}
if ((flags & F_POSIX) != 0) {
p = (struct proc *)id;
tl.lval = p->p_pid;
*cp++ = tl.cval[0];
*cp++ = tl.cval[1];
*cp++ = tl.cval[2];
*cp++ = tl.cval[3];
tl.lval = p->p_stats->p_start.tv_sec;
*cp++ = tl.cval[0];
*cp++ = tl.cval[1];
*cp++ = tl.cval[2];
*cp++ = tl.cval[3];
tl.lval = p->p_stats->p_start.tv_usec;
*cp++ = tl.cval[0];
*cp++ = tl.cval[1];
*cp++ = tl.cval[2];
*cp = tl.cval[3];
} else if ((flags & F_FLOCK) != 0) {
bcopy(&id, cp, sizeof(id));
bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
} else {
printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
bzero(cp, NFSV4CL_LOCKNAMELEN);
}
}
/*
* Find the parent process for the thread passed in as an argument.
* If none exists, return NULL, otherwise return a thread for the parent.
* (Can be any of the threads, since it is only used for td->td_proc.)
*/
NFSPROC_T *
nfscl_getparent(struct thread *td)
{
struct proc *p;
struct thread *ptd;
if (td == NULL)
return (NULL);
p = td->td_proc;
if (p->p_pid == 0)
return (NULL);
p = p->p_pptr;
if (p == NULL)
return (NULL);
ptd = TAILQ_FIRST(&p->p_threads);
return (ptd);
}
/*
* Start up the renew kernel thread.
*/
static void
start_nfscl(void *arg)
{
struct nfsclclient *clp;
struct thread *td;
clp = (struct nfsclclient *)arg;
td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
nfscl_renewthread(clp, td);
kproc_exit(0);
}
void
nfscl_start_renewthread(struct nfsclclient *clp)
{
kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
"nfscl");
}
/*
* Handle wcc_data.
* For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
* as the first Op after PutFH.
* (For NFSv4, the postop attributes are after the Op, so they can't be
* parsed here. A separate call to nfscl_postop_attr() is required.)
*/
int
nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
{
u_int32_t *tl;
struct nfsnode *np = VTONFS(vp);
struct nfsvattr nfsva;
int error = 0;
if (wccflagp != NULL)
*wccflagp = 0;
if (nd->nd_flag & ND_NFSV3) {
*flagp = 0;
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
if (*tl == newnfs_true) {
NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
if (wccflagp != NULL) {
mtx_lock(&np->n_mtx);
*wccflagp = (np->n_mtime.tv_sec ==
fxdr_unsigned(u_int32_t, *(tl + 2)) &&
np->n_mtime.tv_nsec ==
fxdr_unsigned(u_int32_t, *(tl + 3)));
mtx_unlock(&np->n_mtx);
}
}
error = nfscl_postop_attr(nd, nap, flagp, stuff);
} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
== (ND_NFSV4 | ND_V4WCCATTR)) {
error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
NULL, NULL, NULL, NULL, NULL);
if (error)
return (error);
/*
* Get rid of Op# and status for next op.
*/
NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
if (*++tl)
nd->nd_flag |= ND_NOMOREDATA;
if (wccflagp != NULL &&
nfsva.na_vattr.va_mtime.tv_sec != 0) {
mtx_lock(&np->n_mtx);
*wccflagp = (np->n_mtime.tv_sec ==
nfsva.na_vattr.va_mtime.tv_sec &&
np->n_mtime.tv_nsec ==
nfsva.na_vattr.va_mtime.tv_sec);
mtx_unlock(&np->n_mtx);
}
}
nfsmout:
return (error);
}
/*
* Get postop attributes.
*/
int
nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
void *stuff)
{
u_int32_t *tl;
int error = 0;
*retp = 0;
if (nd->nd_flag & ND_NOMOREDATA)
return (error);
if (nd->nd_flag & ND_NFSV3) {
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
*retp = fxdr_unsigned(int, *tl);
} else if (nd->nd_flag & ND_NFSV4) {
/*
* For NFSv4, the postop attr are at the end, so no point
* in looking if nd_repstat != 0.
*/
if (!nd->nd_repstat) {
NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
if (*(tl + 1))
/* should never happen since nd_repstat != 0 */
nd->nd_flag |= ND_NOMOREDATA;
else
*retp = 1;
}
} else if (!nd->nd_repstat) {
/* For NFSv2, the attributes are here iff nd_repstat == 0 */
*retp = 1;
}
if (*retp) {
error = nfsm_loadattr(nd, nap);
if (error)
*retp = 0;
}
nfsmout:
return (error);
}
/*
* Fill in the setable attributes. The full argument indicates whether
* to fill in them all or just mode and time.
*/
void
nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
struct vnode *vp, int flags, u_int32_t rdev)
{
u_int32_t *tl;
struct nfsv2_sattr *sp;
nfsattrbit_t attrbits;
struct timeval curtime;
switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
case ND_NFSV2:
NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
if (vap->va_mode == (mode_t)VNOVAL)
sp->sa_mode = newnfs_xdrneg1;
else
sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
if (vap->va_uid == (uid_t)VNOVAL)
sp->sa_uid = newnfs_xdrneg1;
else
sp->sa_uid = txdr_unsigned(vap->va_uid);
if (vap->va_gid == (gid_t)VNOVAL)
sp->sa_gid = newnfs_xdrneg1;
else
sp->sa_gid = txdr_unsigned(vap->va_gid);
if (flags & NFSSATTR_SIZE0)
sp->sa_size = 0;
else if (flags & NFSSATTR_SIZENEG1)
sp->sa_size = newnfs_xdrneg1;
else if (flags & NFSSATTR_SIZERDEV)
sp->sa_size = txdr_unsigned(rdev);
else
sp->sa_size = txdr_unsigned(vap->va_size);
txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
break;
case ND_NFSV3:
getmicrotime(&curtime);
if (vap->va_mode != (mode_t)VNOVAL) {
NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
*tl++ = newnfs_true;
*tl = txdr_unsigned(vap->va_mode);
} else {
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = newnfs_false;
}
if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
*tl++ = newnfs_true;
*tl = txdr_unsigned(vap->va_uid);
} else {
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = newnfs_false;
}
if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
*tl++ = newnfs_true;
*tl = txdr_unsigned(vap->va_gid);
} else {
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = newnfs_false;
}
if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
*tl++ = newnfs_true;
txdr_hyper(vap->va_size, tl);
} else {
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = newnfs_false;
}
if (vap->va_atime.tv_sec != VNOVAL) {
if (vap->va_atime.tv_sec != curtime.tv_sec) {
NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
txdr_nfsv3time(&vap->va_atime, tl);
} else {
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
}
} else {
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
}
if (vap->va_mtime.tv_sec != VNOVAL) {
if (vap->va_mtime.tv_sec != curtime.tv_sec) {
NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
txdr_nfsv3time(&vap->va_mtime, tl);
} else {
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
}
} else {
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
}
break;
case ND_NFSV4:
NFSZERO_ATTRBIT(&attrbits);
if (vap->va_mode != (mode_t)VNOVAL)
NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
if (vap->va_atime.tv_sec != VNOVAL)
NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
if (vap->va_mtime.tv_sec != VNOVAL)
NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
(void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
&attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
break;
};
}
/*
* nfscl_request() - mostly a wrapper for newnfs_request().
*/
int
nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
struct ucred *cred, void *stuff)
{
int ret, vers;
struct nfsmount *nmp;
nmp = VFSTONFS(vp->v_mount);
if (nd->nd_flag & ND_NFSV4)
vers = NFS_VER4;
else if (nd->nd_flag & ND_NFSV3)
vers = NFS_VER3;
else
vers = NFS_VER2;
ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
NFS_PROG, vers, NULL, 1, NULL);
return (ret);
}
/*
* fill in this bsden's variant of statfs using nfsstatfs.
*/
void
nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
{
struct statfs *sbp = (struct statfs *)statfs;
if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
sbp->f_bsize = NFS_FABLKSIZE;
sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
/*
* Although sf_abytes is uint64_t and f_bavail is int64_t,
* the value after dividing by NFS_FABLKSIZE is small
* enough that it will fit in 63bits, so it is ok to
* assign it to f_bavail without fear that it will become
* negative.
*/
sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
sbp->f_files = sfp->sf_tfiles;
/* Since f_ffree is int64_t, clip it to 63bits. */
if (sfp->sf_ffiles > INT64_MAX)
sbp->f_ffree = INT64_MAX;
else
sbp->f_ffree = sfp->sf_ffiles;
} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
/*
* The type casts to (int32_t) ensure that this code is
* compatible with the old NFS client, in that it will
* propagate bit31 to the high order bits. This may or may
* not be correct for NFSv2, but since it is a legacy
* environment, I'd rather retain backwards compatibility.
*/
sbp->f_bsize = (int32_t)sfp->sf_bsize;
sbp->f_blocks = (int32_t)sfp->sf_blocks;
sbp->f_bfree = (int32_t)sfp->sf_bfree;
sbp->f_bavail = (int32_t)sfp->sf_bavail;
sbp->f_files = 0;
sbp->f_ffree = 0;
}
}
/*
* Use the fsinfo stuff to update the mount point.
*/
void
nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
{
if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
fsp->fs_wtpref >= NFS_FABLKSIZE)
nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
~(NFS_FABLKSIZE - 1);
if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
if (nmp->nm_wsize == 0)
nmp->nm_wsize = fsp->fs_wtmax;
}
if (nmp->nm_wsize < NFS_FABLKSIZE)
nmp->nm_wsize = NFS_FABLKSIZE;
if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
fsp->fs_rtpref >= NFS_FABLKSIZE)
nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
~(NFS_FABLKSIZE - 1);
if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
if (nmp->nm_rsize == 0)
nmp->nm_rsize = fsp->fs_rtmax;
}
if (nmp->nm_rsize < NFS_FABLKSIZE)
nmp->nm_rsize = NFS_FABLKSIZE;
if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
&& fsp->fs_dtpref >= NFS_DIRBLKSIZ)
nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
~(NFS_DIRBLKSIZ - 1);
if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
if (nmp->nm_readdirsize == 0)
nmp->nm_readdirsize = fsp->fs_rtmax;
}
if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
nmp->nm_readdirsize = NFS_DIRBLKSIZ;
if (fsp->fs_maxfilesize > 0 &&
fsp->fs_maxfilesize < nmp->nm_maxfilesize)
nmp->nm_maxfilesize = fsp->fs_maxfilesize;
nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
nmp->nm_state |= NFSSTA_GOTFSINFO;
}
/*
* Get a pointer to my IP addrress and return it.
* Return NULL if you can't find one.
*/
u_int8_t *
nfscl_getmyip(struct nfsmount *nmp, int *isinet6p)
{
struct sockaddr_in sad, *sin;
struct rtentry *rt;
u_int8_t *retp = NULL;
static struct in_addr laddr;
*isinet6p = 0;
/*
* Loop up a route for the destination address.
*/
if (nmp->nm_nam->sa_family == AF_INET) {
bzero(&sad, sizeof (sad));
sin = (struct sockaddr_in *)nmp->nm_nam;
sad.sin_family = AF_INET;
sad.sin_len = sizeof (struct sockaddr_in);
sad.sin_addr.s_addr = sin->sin_addr.s_addr;
CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
rt = rtalloc1((struct sockaddr *)&sad, 0, 0UL);
if (rt != NULL) {
if (rt->rt_ifp != NULL &&
rt->rt_ifa != NULL &&
((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
rt->rt_ifa->ifa_addr->sa_family == AF_INET) {
sin = (struct sockaddr_in *)
rt->rt_ifa->ifa_addr;
laddr.s_addr = sin->sin_addr.s_addr;
retp = (u_int8_t *)&laddr;
}
RTFREE_LOCKED(rt);
}
CURVNET_RESTORE();
#ifdef INET6
} else if (nmp->nm_nam->sa_family == AF_INET6) {
struct sockaddr_in6 sad6, *sin6;
static struct in6_addr laddr6;
bzero(&sad6, sizeof (sad6));
sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
sad6.sin6_family = AF_INET6;
sad6.sin6_len = sizeof (struct sockaddr_in6);
sad6.sin6_addr = sin6->sin6_addr;
CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
rt = rtalloc1((struct sockaddr *)&sad6, 0, 0UL);
if (rt != NULL) {
if (rt->rt_ifp != NULL &&
rt->rt_ifa != NULL &&
((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
rt->rt_ifa->ifa_addr->sa_family == AF_INET6) {
sin6 = (struct sockaddr_in6 *)
rt->rt_ifa->ifa_addr;
laddr6 = sin6->sin6_addr;
retp = (u_int8_t *)&laddr6;
*isinet6p = 1;
}
RTFREE_LOCKED(rt);
}
CURVNET_RESTORE();
#endif
}
return (retp);
}
/*
* Copy NFS uid, gids from the cred structure.
*/
void
newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
{
int i;
KASSERT(cr->cr_ngroups >= 0,
("newnfs_copyincred: negative cr_ngroups"));
nfscr->nfsc_uid = cr->cr_uid;
nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
for (i = 0; i < nfscr->nfsc_ngroups; i++)
nfscr->nfsc_groups[i] = cr->cr_groups[i];
}
/*
* Do any client specific initialization.
*/
void
nfscl_init(void)
{
static int inited = 0;
if (inited)
return;
inited = 1;
nfscl_inited = 1;
ncl_pbuf_freecnt = nswbuf / 2 + 1;
}
/*
* Check each of the attributes to be set, to ensure they aren't already
* the correct value. Disable setting ones already correct.
*/
int
nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
{
if (vap->va_mode != (mode_t)VNOVAL) {
if (vap->va_mode == nvap->na_mode)
vap->va_mode = (mode_t)VNOVAL;
}
if (vap->va_uid != (uid_t)VNOVAL) {
if (vap->va_uid == nvap->na_uid)
vap->va_uid = (uid_t)VNOVAL;
}
if (vap->va_gid != (gid_t)VNOVAL) {
if (vap->va_gid == nvap->na_gid)
vap->va_gid = (gid_t)VNOVAL;
}
if (vap->va_size != VNOVAL) {
if (vap->va_size == nvap->na_size)
vap->va_size = VNOVAL;
}
/*
* We are normally called with only a partially initialized
* VAP. Since the NFSv3 spec says that server may use the
* file attributes to store the verifier, the spec requires
* us to do a SETATTR RPC. FreeBSD servers store the verifier
* in atime, but we can't really assume that all servers will
* so we ensure that our SETATTR sets both atime and mtime.
*/
if (vap->va_mtime.tv_sec == VNOVAL)
vfs_timestamp(&vap->va_mtime);
if (vap->va_atime.tv_sec == VNOVAL)
vap->va_atime = vap->va_mtime;
return (1);
}
/*
* Map nfsv4 errors to errno.h errors.
* The uid and gid arguments are only used for NFSERR_BADOWNER and that
* error should only be returned for the Open, Create and Setattr Ops.
* As such, most calls can just pass in 0 for those arguments.
*/
APPLESTATIC int
nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
{
struct proc *p;
if (error < 10000)
return (error);
if (td != NULL)
p = td->td_proc;
else
p = NULL;
switch (error) {
case NFSERR_BADOWNER:
tprintf(p, LOG_INFO,
"No name and/or group mapping for uid,gid:(%d,%d)\n",
uid, gid);
return (EPERM);
case NFSERR_STALECLIENTID:
case NFSERR_STALESTATEID:
case NFSERR_EXPIRED:
case NFSERR_BADSTATEID:
printf("nfsv4 recover err returned %d\n", error);
return (EIO);
case NFSERR_BADHANDLE:
case NFSERR_SERVERFAULT:
case NFSERR_BADTYPE:
case NFSERR_FHEXPIRED:
case NFSERR_RESOURCE:
case NFSERR_MOVED:
case NFSERR_NOFILEHANDLE:
case NFSERR_MINORVERMISMATCH:
case NFSERR_OLDSTATEID:
case NFSERR_BADSEQID:
case NFSERR_LEASEMOVED:
case NFSERR_RECLAIMBAD:
case NFSERR_BADXDR:
case NFSERR_BADCHAR:
case NFSERR_BADNAME:
case NFSERR_OPILLEGAL:
printf("nfsv4 client/server protocol prob err=%d\n",
error);
return (EIO);
default:
tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
return (EIO);
};
}
/*
* Locate a process by number; return only "live" processes -- i.e., neither
* zombies nor newly born but incompletely initialized processes. By not
* returning processes in the PRS_NEW state, we allow callers to avoid
* testing for that condition to avoid dereferencing p_ucred, et al.
* Identical to pfind() in kern_proc.c, except it assume the list is
* already locked.
*/
static struct proc *
pfind_locked(pid_t pid)
{
struct proc *p;
LIST_FOREACH(p, PIDHASH(pid), p_hash)
if (p->p_pid == pid) {
PROC_LOCK(p);
if (p->p_state == PRS_NEW) {
PROC_UNLOCK(p);
p = NULL;
}
break;
}
return (p);
}
/*
* Check to see if the process for this owner exists. Return 1 if it doesn't
* and 0 otherwise.
*/
int
nfscl_procdoesntexist(u_int8_t *own)
{
union {
u_int32_t lval;
u_int8_t cval[4];
} tl;
struct proc *p;
pid_t pid;
int ret = 0;
tl.cval[0] = *own++;
tl.cval[1] = *own++;
tl.cval[2] = *own++;
tl.cval[3] = *own++;
pid = tl.lval;
p = pfind_locked(pid);
if (p == NULL)
return (1);
if (p->p_stats == NULL) {
PROC_UNLOCK(p);
return (0);
}
tl.cval[0] = *own++;
tl.cval[1] = *own++;
tl.cval[2] = *own++;
tl.cval[3] = *own++;
if (tl.lval != p->p_stats->p_start.tv_sec) {
ret = 1;
} else {
tl.cval[0] = *own++;
tl.cval[1] = *own++;
tl.cval[2] = *own++;
tl.cval[3] = *own;
if (tl.lval != p->p_stats->p_start.tv_usec)
ret = 1;
}
PROC_UNLOCK(p);
return (ret);
}
/*
* - nfs pseudo system call for the client
*/
/*
* MPSAFE
*/
static int
nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
{
struct file *fp;
struct nfscbd_args nfscbdarg;
struct nfsd_nfscbd_args nfscbdarg2;
int error;
if (uap->flag & NFSSVC_CBADDSOCK) {
error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
if (error)
return (error);
if ((error = fget(td, nfscbdarg.sock, &fp)) != 0) {
return (error);
}
if (fp->f_type != DTYPE_SOCKET) {
fdrop(fp, td);
return (EPERM);
}
error = nfscbd_addsock(fp);
fdrop(fp, td);
if (!error && nfscl_enablecallb == 0) {
nfsv4_cbport = nfscbdarg.port;
nfscl_enablecallb = 1;
}
} else if (uap->flag & NFSSVC_NFSCBD) {
if (uap->argp == NULL)
return (EINVAL);
error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
sizeof(nfscbdarg2));
if (error)
return (error);
error = nfscbd_nfsd(td, &nfscbdarg2);
} else {
error = EINVAL;
}
return (error);
}
extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
/*
* Called once to initialize data structures...
*/
static int
nfscl_modevent(module_t mod, int type, void *data)
{
int error = 0;
static int loaded = 0;
switch (type) {
case MOD_LOAD:
if (loaded)
return (0);
newnfs_portinit();
mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL,
MTX_DEF);
mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
nfscl_init();
NFSD_LOCK();
nfsrvd_cbinit(0);
NFSD_UNLOCK();
ncl_call_invalcaches = ncl_invalcaches;
nfsd_call_nfscl = nfssvc_nfscl;
loaded = 1;
break;
case MOD_UNLOAD:
if (nfs_numnfscbd != 0) {
error = EBUSY;
break;
}
/*
* XXX: Unloading of nfscl module is unsupported.
*/
#if 0
ncl_call_invalcaches = NULL;
nfsd_call_nfscl = NULL;
/* and get rid of the mutexes */
mtx_destroy(&nfs_clstate_mutex);
mtx_destroy(&ncl_iod_mutex);
loaded = 0;
break;
#else
/* FALLTHROUGH */
#endif
default:
error = EOPNOTSUPP;
break;
}
return error;
}
static moduledata_t nfscl_mod = {
"nfscl",
nfscl_modevent,
NULL,
};
DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
/* So that loader and kldload(2) can find us, wherever we are.. */
MODULE_VERSION(nfscl, 1);
MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
Index: stable/8/sys/nfsclient/nfs_bio.c
===================================================================
--- stable/8/sys/nfsclient/nfs_bio.c (revision 231331)
+++ stable/8/sys/nfsclient/nfs_bio.c (revision 231332)
@@ -1,1798 +1,1794 @@
/*-
* Copyright (c) 1989, 1993
* The Regents of the University of California. All rights reserved.
*
* This code is derived from software contributed to Berkeley by
* Rick Macklem at The University of Guelph.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* @(#)nfs_bio.c 8.9 (Berkeley) 3/30/95
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_kdtrace.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bio.h>
#include <sys/buf.h>
#include <sys/kernel.h>
#include <sys/mbuf.h>
#include <sys/mount.h>
#include <sys/proc.h>
#include <sys/vmmeter.h>
#include <sys/vnode.h>
#include <vm/vm.h>
#include <vm/vm_extern.h>
#include <vm/vm_page.h>
#include <vm/vm_object.h>
#include <vm/vm_pager.h>
#include <vm/vnode_pager.h>
#include <nfs/nfsproto.h>
#include <nfsclient/nfs.h>
#include <nfsclient/nfsmount.h>
#include <nfsclient/nfsnode.h>
#include <nfsclient/nfs_kdtrace.h>
static struct buf *nfs_getcacheblk(struct vnode *vp, daddr_t bn, int size,
struct thread *td);
static int nfs_directio_write(struct vnode *vp, struct uio *uiop,
struct ucred *cred, int ioflag);
extern int nfs_directio_enable;
extern int nfs_directio_allow_mmap;
/*
* Vnode op for VM getpages.
*/
int
nfs_getpages(struct vop_getpages_args *ap)
{
int i, error, nextoff, size, toff, count, npages;
struct uio uio;
struct iovec iov;
vm_offset_t kva;
struct buf *bp;
struct vnode *vp;
struct thread *td;
struct ucred *cred;
struct nfsmount *nmp;
vm_object_t object;
vm_page_t *pages;
struct nfsnode *np;
vp = ap->a_vp;
np = VTONFS(vp);
td = curthread; /* XXX */
cred = curthread->td_ucred; /* XXX */
nmp = VFSTONFS(vp->v_mount);
pages = ap->a_m;
count = ap->a_count;
if ((object = vp->v_object) == NULL) {
nfs_printf("nfs_getpages: called with non-merged cache vnode??\n");
return (VM_PAGER_ERROR);
}
if (nfs_directio_enable && !nfs_directio_allow_mmap) {
mtx_lock(&np->n_mtx);
if ((np->n_flag & NNONCACHE) && (vp->v_type == VREG)) {
mtx_unlock(&np->n_mtx);
nfs_printf("nfs_getpages: called on non-cacheable vnode??\n");
return (VM_PAGER_ERROR);
} else
mtx_unlock(&np->n_mtx);
}
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 &&
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
mtx_unlock(&nmp->nm_mtx);
/* We'll never get here for v4, because we always have fsinfo */
(void)nfs_fsinfo(nmp, vp, cred, td);
} else
mtx_unlock(&nmp->nm_mtx);
npages = btoc(count);
/*
* If the requested page is partially valid, just return it and
* allow the pager to zero-out the blanks. Partially valid pages
* can only occur at the file EOF.
*/
VM_OBJECT_LOCK(object);
if (pages[ap->a_reqpage]->valid != 0) {
vm_page_lock_queues();
for (i = 0; i < npages; ++i) {
if (i != ap->a_reqpage)
vm_page_free(pages[i]);
}
vm_page_unlock_queues();
VM_OBJECT_UNLOCK(object);
return (0);
}
VM_OBJECT_UNLOCK(object);
/*
* We use only the kva address for the buffer, but this is extremely
* convienient and fast.
*/
bp = getpbuf(&nfs_pbuf_freecnt);
kva = (vm_offset_t) bp->b_data;
pmap_qenter(kva, pages, npages);
PCPU_INC(cnt.v_vnodein);
PCPU_ADD(cnt.v_vnodepgsin, npages);
iov.iov_base = (caddr_t) kva;
iov.iov_len = count;
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
uio.uio_resid = count;
uio.uio_segflg = UIO_SYSSPACE;
uio.uio_rw = UIO_READ;
uio.uio_td = td;
error = (nmp->nm_rpcops->nr_readrpc)(vp, &uio, cred);
pmap_qremove(kva, npages);
relpbuf(bp, &nfs_pbuf_freecnt);
if (error && (uio.uio_resid == count)) {
nfs_printf("nfs_getpages: error %d\n", error);
VM_OBJECT_LOCK(object);
vm_page_lock_queues();
for (i = 0; i < npages; ++i) {
if (i != ap->a_reqpage)
vm_page_free(pages[i]);
}
vm_page_unlock_queues();
VM_OBJECT_UNLOCK(object);
return (VM_PAGER_ERROR);
}
/*
* Calculate the number of bytes read and validate only that number
* of bytes. Note that due to pending writes, size may be 0. This
* does not mean that the remaining data is invalid!
*/
size = count - uio.uio_resid;
VM_OBJECT_LOCK(object);
vm_page_lock_queues();
for (i = 0, toff = 0; i < npages; i++, toff = nextoff) {
vm_page_t m;
nextoff = toff + PAGE_SIZE;
m = pages[i];
if (nextoff <= size) {
/*
* Read operation filled an entire page
*/
m->valid = VM_PAGE_BITS_ALL;
KASSERT(m->dirty == 0,
("nfs_getpages: page %p is dirty", m));
} else if (size > toff) {
/*
* Read operation filled a partial page.
*/
m->valid = 0;
vm_page_set_valid(m, 0, size - toff);
KASSERT(m->dirty == 0,
("nfs_getpages: page %p is dirty", m));
} else {
/*
* Read operation was short. If no error occured
* we may have hit a zero-fill section. We simply
* leave valid set to 0.
*/
;
}
if (i != ap->a_reqpage) {
/*
* Whether or not to leave the page activated is up in
* the air, but we should put the page on a page queue
* somewhere (it already is in the object). Result:
* It appears that emperical results show that
* deactivating pages is best.
*/
/*
* Just in case someone was asking for this page we
* now tell them that it is ok to use.
*/
if (!error) {
if (m->oflags & VPO_WANTED)
vm_page_activate(m);
else
vm_page_deactivate(m);
vm_page_wakeup(m);
} else {
vm_page_free(m);
}
}
}
vm_page_unlock_queues();
VM_OBJECT_UNLOCK(object);
return (0);
}
/*
* Vnode op for VM putpages.
*/
int
nfs_putpages(struct vop_putpages_args *ap)
{
struct uio uio;
struct iovec iov;
vm_offset_t kva;
struct buf *bp;
int iomode, must_commit, i, error, npages, count;
off_t offset;
int *rtvals;
struct vnode *vp;
struct thread *td;
struct ucred *cred;
struct nfsmount *nmp;
struct nfsnode *np;
vm_page_t *pages;
vp = ap->a_vp;
np = VTONFS(vp);
td = curthread; /* XXX */
cred = curthread->td_ucred; /* XXX */
nmp = VFSTONFS(vp->v_mount);
pages = ap->a_m;
count = ap->a_count;
rtvals = ap->a_rtvals;
npages = btoc(count);
offset = IDX_TO_OFF(pages[0]->pindex);
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 &&
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
mtx_unlock(&nmp->nm_mtx);
(void)nfs_fsinfo(nmp, vp, cred, td);
} else
mtx_unlock(&nmp->nm_mtx);
mtx_lock(&np->n_mtx);
if (nfs_directio_enable && !nfs_directio_allow_mmap &&
(np->n_flag & NNONCACHE) && (vp->v_type == VREG)) {
mtx_unlock(&np->n_mtx);
nfs_printf("nfs_putpages: called on noncache-able vnode??\n");
mtx_lock(&np->n_mtx);
}
for (i = 0; i < npages; i++)
rtvals[i] = VM_PAGER_ERROR;
/*
* When putting pages, do not extend file past EOF.
*/
if (offset + count > np->n_size) {
count = np->n_size - offset;
if (count < 0)
count = 0;
}
mtx_unlock(&np->n_mtx);
/*
* We use only the kva address for the buffer, but this is extremely
* convienient and fast.
*/
bp = getpbuf(&nfs_pbuf_freecnt);
kva = (vm_offset_t) bp->b_data;
pmap_qenter(kva, pages, npages);
PCPU_INC(cnt.v_vnodeout);
PCPU_ADD(cnt.v_vnodepgsout, count);
iov.iov_base = (caddr_t) kva;
iov.iov_len = count;
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = offset;
uio.uio_resid = count;
uio.uio_segflg = UIO_SYSSPACE;
uio.uio_rw = UIO_WRITE;
uio.uio_td = td;
if ((ap->a_sync & VM_PAGER_PUT_SYNC) == 0)
iomode = NFSV3WRITE_UNSTABLE;
else
iomode = NFSV3WRITE_FILESYNC;
error = (nmp->nm_rpcops->nr_writerpc)(vp, &uio, cred, &iomode, &must_commit);
pmap_qremove(kva, npages);
relpbuf(bp, &nfs_pbuf_freecnt);
if (!error) {
vnode_pager_undirty_pages(pages, rtvals, count - uio.uio_resid);
if (must_commit) {
nfs_clearcommit(vp->v_mount);
}
}
return rtvals[0];
}
/*
* For nfs, cache consistency can only be maintained approximately.
* Although RFC1094 does not specify the criteria, the following is
* believed to be compatible with the reference port.
* For nfs:
* If the file's modify time on the server has changed since the
* last read rpc or you have written to the file,
* you may have lost data cache consistency with the
* server, so flush all of the file's data out of the cache.
* Then force a getattr rpc to ensure that you have up to date
* attributes.
* NB: This implies that cache data can be read when up to
* NFS_ATTRTIMEO seconds out of date. If you find that you need current
* attributes this could be forced by setting n_attrstamp to 0 before
* the VOP_GETATTR() call.
*/
static inline int
nfs_bioread_check_cons(struct vnode *vp, struct thread *td, struct ucred *cred)
{
int error = 0;
struct vattr vattr;
struct nfsnode *np = VTONFS(vp);
int old_lock;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
/*
* Grab the exclusive lock before checking whether the cache is
* consistent.
* XXX - We can make this cheaper later (by acquiring cheaper locks).
* But for now, this suffices.
*/
old_lock = nfs_upgrade_vnlock(vp);
if (vp->v_iflag & VI_DOOMED) {
nfs_downgrade_vnlock(vp, old_lock);
return (EBADF);
}
mtx_lock(&np->n_mtx);
if (np->n_flag & NMODIFIED) {
mtx_unlock(&np->n_mtx);
if (vp->v_type != VREG) {
if (vp->v_type != VDIR)
panic("nfs: bioread, not dir");
(nmp->nm_rpcops->nr_invaldir)(vp);
error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
if (error)
goto out;
}
np->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
error = VOP_GETATTR(vp, &vattr, cred);
if (error)
goto out;
mtx_lock(&np->n_mtx);
np->n_mtime = vattr.va_mtime;
mtx_unlock(&np->n_mtx);
} else {
mtx_unlock(&np->n_mtx);
error = VOP_GETATTR(vp, &vattr, cred);
if (error)
return (error);
mtx_lock(&np->n_mtx);
if ((np->n_flag & NSIZECHANGED)
|| (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime))) {
mtx_unlock(&np->n_mtx);
if (vp->v_type == VDIR)
(nmp->nm_rpcops->nr_invaldir)(vp);
error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
if (error)
goto out;
mtx_lock(&np->n_mtx);
np->n_mtime = vattr.va_mtime;
np->n_flag &= ~NSIZECHANGED;
}
mtx_unlock(&np->n_mtx);
}
out:
nfs_downgrade_vnlock(vp, old_lock);
return error;
}
/*
* Vnode op for read using bio
*/
int
nfs_bioread(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *cred)
{
struct nfsnode *np = VTONFS(vp);
int biosize, i;
struct buf *bp, *rabp;
struct thread *td;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
daddr_t lbn, rabn;
off_t end;
int bcount;
int seqcount;
int nra, error = 0, n = 0, on = 0;
KASSERT(uio->uio_rw == UIO_READ, ("nfs_read mode"));
if (uio->uio_resid == 0)
return (0);
if (uio->uio_offset < 0) /* XXX VDIR cookies can be negative */
return (EINVAL);
td = uio->uio_td;
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 &&
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
mtx_unlock(&nmp->nm_mtx);
(void)nfs_fsinfo(nmp, vp, cred, td);
} else
mtx_unlock(&nmp->nm_mtx);
end = uio->uio_offset + uio->uio_resid;
if (vp->v_type != VDIR &&
(end > nmp->nm_maxfilesize || end < uio->uio_offset))
return (EFBIG);
if (nfs_directio_enable && (ioflag & IO_DIRECT) && (vp->v_type == VREG))
/* No caching/ no readaheads. Just read data into the user buffer */
return nfs_readrpc(vp, uio, cred);
- biosize = vp->v_mount->mnt_stat.f_iosize;
+ biosize = vp->v_bufobj.bo_bsize;
seqcount = (int)((off_t)(ioflag >> IO_SEQSHIFT) * biosize / BKVASIZE);
error = nfs_bioread_check_cons(vp, td, cred);
if (error)
return error;
do {
u_quad_t nsize;
mtx_lock(&np->n_mtx);
nsize = np->n_size;
mtx_unlock(&np->n_mtx);
switch (vp->v_type) {
case VREG:
nfsstats.biocache_reads++;
lbn = uio->uio_offset / biosize;
on = uio->uio_offset & (biosize - 1);
/*
* Start the read ahead(s), as required.
*/
if (nmp->nm_readahead > 0) {
for (nra = 0; nra < nmp->nm_readahead && nra < seqcount &&
(off_t)(lbn + 1 + nra) * biosize < nsize; nra++) {
rabn = lbn + 1 + nra;
if (incore(&vp->v_bufobj, rabn) == NULL) {
rabp = nfs_getcacheblk(vp, rabn, biosize, td);
if (!rabp) {
error = nfs_sigintr(nmp, td);
return (error ? error : EINTR);
}
if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) {
rabp->b_flags |= B_ASYNC;
rabp->b_iocmd = BIO_READ;
vfs_busy_pages(rabp, 0);
if (nfs_asyncio(nmp, rabp, cred, td)) {
rabp->b_flags |= B_INVAL;
rabp->b_ioflags |= BIO_ERROR;
vfs_unbusy_pages(rabp);
brelse(rabp);
break;
}
} else {
brelse(rabp);
}
}
}
}
/* Note that bcount is *not* DEV_BSIZE aligned. */
bcount = biosize;
if ((off_t)lbn * biosize >= nsize) {
bcount = 0;
} else if ((off_t)(lbn + 1) * biosize > nsize) {
bcount = nsize - (off_t)lbn * biosize;
}
bp = nfs_getcacheblk(vp, lbn, bcount, td);
if (!bp) {
error = nfs_sigintr(nmp, td);
return (error ? error : EINTR);
}
/*
* If B_CACHE is not set, we must issue the read. If this
* fails, we return an error.
*/
if ((bp->b_flags & B_CACHE) == 0) {
bp->b_iocmd = BIO_READ;
vfs_busy_pages(bp, 0);
error = nfs_doio(vp, bp, cred, td);
if (error) {
brelse(bp);
return (error);
}
}
/*
* on is the offset into the current bp. Figure out how many
* bytes we can copy out of the bp. Note that bcount is
* NOT DEV_BSIZE aligned.
*
* Then figure out how many bytes we can copy into the uio.
*/
n = 0;
if (on < bcount)
n = min((unsigned)(bcount - on), uio->uio_resid);
break;
case VLNK:
nfsstats.biocache_readlinks++;
bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, td);
if (!bp) {
error = nfs_sigintr(nmp, td);
return (error ? error : EINTR);
}
if ((bp->b_flags & B_CACHE) == 0) {
bp->b_iocmd = BIO_READ;
vfs_busy_pages(bp, 0);
error = nfs_doio(vp, bp, cred, td);
if (error) {
bp->b_ioflags |= BIO_ERROR;
brelse(bp);
return (error);
}
}
n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid);
on = 0;
break;
case VDIR:
nfsstats.biocache_readdirs++;
if (np->n_direofoffset
&& uio->uio_offset >= np->n_direofoffset) {
return (0);
}
lbn = (uoff_t)uio->uio_offset / NFS_DIRBLKSIZ;
on = uio->uio_offset & (NFS_DIRBLKSIZ - 1);
bp = nfs_getcacheblk(vp, lbn, NFS_DIRBLKSIZ, td);
if (!bp) {
error = nfs_sigintr(nmp, td);
return (error ? error : EINTR);
}
if ((bp->b_flags & B_CACHE) == 0) {
bp->b_iocmd = BIO_READ;
vfs_busy_pages(bp, 0);
error = nfs_doio(vp, bp, cred, td);
if (error) {
brelse(bp);
}
while (error == NFSERR_BAD_COOKIE) {
(nmp->nm_rpcops->nr_invaldir)(vp);
error = nfs_vinvalbuf(vp, 0, td, 1);
/*
* Yuck! The directory has been modified on the
* server. The only way to get the block is by
* reading from the beginning to get all the
* offset cookies.
*
* Leave the last bp intact unless there is an error.
* Loop back up to the while if the error is another
* NFSERR_BAD_COOKIE (double yuch!).
*/
for (i = 0; i <= lbn && !error; i++) {
if (np->n_direofoffset
&& (i * NFS_DIRBLKSIZ) >= np->n_direofoffset)
return (0);
bp = nfs_getcacheblk(vp, i, NFS_DIRBLKSIZ, td);
if (!bp) {
error = nfs_sigintr(nmp, td);
return (error ? error : EINTR);
}
if ((bp->b_flags & B_CACHE) == 0) {
bp->b_iocmd = BIO_READ;
vfs_busy_pages(bp, 0);
error = nfs_doio(vp, bp, cred, td);
/*
* no error + B_INVAL == directory EOF,
* use the block.
*/
if (error == 0 && (bp->b_flags & B_INVAL))
break;
}
/*
* An error will throw away the block and the
* for loop will break out. If no error and this
* is not the block we want, we throw away the
* block and go for the next one via the for loop.
*/
if (error || i < lbn)
brelse(bp);
}
}
/*
* The above while is repeated if we hit another cookie
* error. If we hit an error and it wasn't a cookie error,
* we give up.
*/
if (error)
return (error);
}
/*
* If not eof and read aheads are enabled, start one.
* (You need the current block first, so that you have the
* directory offset cookie of the next block.)
*/
if (nmp->nm_readahead > 0 &&
(bp->b_flags & B_INVAL) == 0 &&
(np->n_direofoffset == 0 ||
(lbn + 1) * NFS_DIRBLKSIZ < np->n_direofoffset) &&
incore(&vp->v_bufobj, lbn + 1) == NULL) {
rabp = nfs_getcacheblk(vp, lbn + 1, NFS_DIRBLKSIZ, td);
if (rabp) {
if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) {
rabp->b_flags |= B_ASYNC;
rabp->b_iocmd = BIO_READ;
vfs_busy_pages(rabp, 0);
if (nfs_asyncio(nmp, rabp, cred, td)) {
rabp->b_flags |= B_INVAL;
rabp->b_ioflags |= BIO_ERROR;
vfs_unbusy_pages(rabp);
brelse(rabp);
}
} else {
brelse(rabp);
}
}
}
/*
* Unlike VREG files, whos buffer size ( bp->b_bcount ) is
* chopped for the EOF condition, we cannot tell how large
* NFS directories are going to be until we hit EOF. So
* an NFS directory buffer is *not* chopped to its EOF. Now,
* it just so happens that b_resid will effectively chop it
* to EOF. *BUT* this information is lost if the buffer goes
* away and is reconstituted into a B_CACHE state ( due to
* being VMIO ) later. So we keep track of the directory eof
* in np->n_direofoffset and chop it off as an extra step
* right here.
*/
n = lmin(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid - on);
if (np->n_direofoffset && n > np->n_direofoffset - uio->uio_offset)
n = np->n_direofoffset - uio->uio_offset;
break;
default:
nfs_printf(" nfs_bioread: type %x unexpected\n", vp->v_type);
bp = NULL;
break;
};
if (n > 0) {
error = uiomove(bp->b_data + on, (int)n, uio);
}
if (vp->v_type == VLNK)
n = 0;
if (bp != NULL)
brelse(bp);
} while (error == 0 && uio->uio_resid > 0 && n > 0);
return (error);
}
/*
* The NFS write path cannot handle iovecs with len > 1. So we need to
* break up iovecs accordingly (restricting them to wsize).
* For the SYNC case, we can do this with 1 copy (user buffer -> mbuf).
* For the ASYNC case, 2 copies are needed. The first a copy from the
* user buffer to a staging buffer and then a second copy from the staging
* buffer to mbufs. This can be optimized by copying from the user buffer
* directly into mbufs and passing the chain down, but that requires a
* fair amount of re-working of the relevant codepaths (and can be done
* later).
*/
static int
nfs_directio_write(vp, uiop, cred, ioflag)
struct vnode *vp;
struct uio *uiop;
struct ucred *cred;
int ioflag;
{
int error;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
struct thread *td = uiop->uio_td;
int size;
int wsize;
mtx_lock(&nmp->nm_mtx);
wsize = nmp->nm_wsize;
mtx_unlock(&nmp->nm_mtx);
if (ioflag & IO_SYNC) {
int iomode, must_commit;
struct uio uio;
struct iovec iov;
do_sync:
while (uiop->uio_resid > 0) {
size = min(uiop->uio_resid, wsize);
size = min(uiop->uio_iov->iov_len, size);
iov.iov_base = uiop->uio_iov->iov_base;
iov.iov_len = size;
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = uiop->uio_offset;
uio.uio_resid = size;
uio.uio_segflg = UIO_USERSPACE;
uio.uio_rw = UIO_WRITE;
uio.uio_td = td;
iomode = NFSV3WRITE_FILESYNC;
error = (nmp->nm_rpcops->nr_writerpc)(vp, &uio, cred,
&iomode, &must_commit);
KASSERT((must_commit == 0),
("nfs_directio_write: Did not commit write"));
if (error)
return (error);
uiop->uio_offset += size;
uiop->uio_resid -= size;
if (uiop->uio_iov->iov_len <= size) {
uiop->uio_iovcnt--;
uiop->uio_iov++;
} else {
uiop->uio_iov->iov_base =
(char *)uiop->uio_iov->iov_base + size;
uiop->uio_iov->iov_len -= size;
}
}
} else {
struct uio *t_uio;
struct iovec *t_iov;
struct buf *bp;
/*
* Break up the write into blocksize chunks and hand these
* over to nfsiod's for write back.
* Unfortunately, this incurs a copy of the data. Since
* the user could modify the buffer before the write is
* initiated.
*
* The obvious optimization here is that one of the 2 copies
* in the async write path can be eliminated by copying the
* data here directly into mbufs and passing the mbuf chain
* down. But that will require a fair amount of re-working
* of the code and can be done if there's enough interest
* in NFS directio access.
*/
while (uiop->uio_resid > 0) {
size = min(uiop->uio_resid, wsize);
size = min(uiop->uio_iov->iov_len, size);
bp = getpbuf(&nfs_pbuf_freecnt);
t_uio = malloc(sizeof(struct uio), M_NFSDIRECTIO, M_WAITOK);
t_iov = malloc(sizeof(struct iovec), M_NFSDIRECTIO, M_WAITOK);
t_iov->iov_base = malloc(size, M_NFSDIRECTIO, M_WAITOK);
t_iov->iov_len = size;
t_uio->uio_iov = t_iov;
t_uio->uio_iovcnt = 1;
t_uio->uio_offset = uiop->uio_offset;
t_uio->uio_resid = size;
t_uio->uio_segflg = UIO_SYSSPACE;
t_uio->uio_rw = UIO_WRITE;
t_uio->uio_td = td;
bcopy(uiop->uio_iov->iov_base, t_iov->iov_base, size);
bp->b_flags |= B_DIRECT;
bp->b_iocmd = BIO_WRITE;
if (cred != NOCRED) {
crhold(cred);
bp->b_wcred = cred;
} else
bp->b_wcred = NOCRED;
bp->b_caller1 = (void *)t_uio;
bp->b_vp = vp;
error = nfs_asyncio(nmp, bp, NOCRED, td);
if (error) {
free(t_iov->iov_base, M_NFSDIRECTIO);
free(t_iov, M_NFSDIRECTIO);
free(t_uio, M_NFSDIRECTIO);
bp->b_vp = NULL;
relpbuf(bp, &nfs_pbuf_freecnt);
if (error == EINTR)
return (error);
goto do_sync;
}
uiop->uio_offset += size;
uiop->uio_resid -= size;
if (uiop->uio_iov->iov_len <= size) {
uiop->uio_iovcnt--;
uiop->uio_iov++;
} else {
uiop->uio_iov->iov_base =
(char *)uiop->uio_iov->iov_base + size;
uiop->uio_iov->iov_len -= size;
}
}
}
return (0);
}
/*
* Vnode op for write using bio
*/
int
nfs_write(struct vop_write_args *ap)
{
int biosize;
struct uio *uio = ap->a_uio;
struct thread *td = uio->uio_td;
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
struct ucred *cred = ap->a_cred;
int ioflag = ap->a_ioflag;
struct buf *bp;
struct vattr vattr;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
daddr_t lbn;
off_t end;
int bcount;
int n, on, error = 0;
KASSERT(uio->uio_rw == UIO_WRITE, ("nfs_write mode"));
KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread,
("nfs_write proc"));
if (vp->v_type != VREG)
return (EIO);
mtx_lock(&np->n_mtx);
if (np->n_flag & NWRITEERR) {
np->n_flag &= ~NWRITEERR;
mtx_unlock(&np->n_mtx);
return (np->n_error);
} else
mtx_unlock(&np->n_mtx);
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 &&
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
mtx_unlock(&nmp->nm_mtx);
(void)nfs_fsinfo(nmp, vp, cred, td);
} else
mtx_unlock(&nmp->nm_mtx);
/*
* Synchronously flush pending buffers if we are in synchronous
* mode or if we are appending.
*/
if (ioflag & (IO_APPEND | IO_SYNC)) {
mtx_lock(&np->n_mtx);
if (np->n_flag & NMODIFIED) {
mtx_unlock(&np->n_mtx);
#ifdef notyet /* Needs matching nonblock semantics elsewhere, too. */
/*
* Require non-blocking, synchronous writes to
* dirty files to inform the program it needs
* to fsync(2) explicitly.
*/
if (ioflag & IO_NDELAY)
return (EAGAIN);
#endif
flush_and_restart:
np->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
if (error)
return (error);
} else
mtx_unlock(&np->n_mtx);
}
/*
* If IO_APPEND then load uio_offset. We restart here if we cannot
* get the append lock.
*/
if (ioflag & IO_APPEND) {
np->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
error = VOP_GETATTR(vp, &vattr, cred);
if (error)
return (error);
mtx_lock(&np->n_mtx);
uio->uio_offset = np->n_size;
mtx_unlock(&np->n_mtx);
}
if (uio->uio_offset < 0)
return (EINVAL);
end = uio->uio_offset + uio->uio_resid;
if (end > nmp->nm_maxfilesize || end < uio->uio_offset)
return (EFBIG);
if (uio->uio_resid == 0)
return (0);
if (nfs_directio_enable && (ioflag & IO_DIRECT) && vp->v_type == VREG)
return nfs_directio_write(vp, uio, cred, ioflag);
/*
* Maybe this should be above the vnode op call, but so long as
* file servers have no limits, i don't think it matters
*/
if (vn_rlimit_fsize(vp, uio, td))
return (EFBIG);
- biosize = vp->v_mount->mnt_stat.f_iosize;
+ biosize = vp->v_bufobj.bo_bsize;
/*
* Find all of this file's B_NEEDCOMMIT buffers. If our writes
* would exceed the local maximum per-file write commit size when
* combined with those, we must decide whether to flush,
* go synchronous, or return error. We don't bother checking
* IO_UNIT -- we just make all writes atomic anyway, as there's
* no point optimizing for something that really won't ever happen.
*/
if (!(ioflag & IO_SYNC)) {
int nflag;
mtx_lock(&np->n_mtx);
nflag = np->n_flag;
mtx_unlock(&np->n_mtx);
int needrestart = 0;
if (nmp->nm_wcommitsize < uio->uio_resid) {
/*
* If this request could not possibly be completed
* without exceeding the maximum outstanding write
* commit size, see if we can convert it into a
* synchronous write operation.
*/
if (ioflag & IO_NDELAY)
return (EAGAIN);
ioflag |= IO_SYNC;
if (nflag & NMODIFIED)
needrestart = 1;
} else if (nflag & NMODIFIED) {
int wouldcommit = 0;
BO_LOCK(&vp->v_bufobj);
if (vp->v_bufobj.bo_dirty.bv_cnt != 0) {
TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd,
b_bobufs) {
if (bp->b_flags & B_NEEDCOMMIT)
wouldcommit += bp->b_bcount;
}
}
BO_UNLOCK(&vp->v_bufobj);
/*
* Since we're not operating synchronously and
* bypassing the buffer cache, we are in a commit
* and holding all of these buffers whether
* transmitted or not. If not limited, this
* will lead to the buffer cache deadlocking,
* as no one else can flush our uncommitted buffers.
*/
wouldcommit += uio->uio_resid;
/*
* If we would initially exceed the maximum
* outstanding write commit size, flush and restart.
*/
if (wouldcommit > nmp->nm_wcommitsize)
needrestart = 1;
}
if (needrestart)
goto flush_and_restart;
}
do {
nfsstats.biocache_writes++;
lbn = uio->uio_offset / biosize;
on = uio->uio_offset & (biosize-1);
n = min((unsigned)(biosize - on), uio->uio_resid);
again:
/*
* Handle direct append and file extension cases, calculate
* unaligned buffer size.
*/
mtx_lock(&np->n_mtx);
if (uio->uio_offset == np->n_size && n) {
mtx_unlock(&np->n_mtx);
/*
* Get the buffer (in its pre-append state to maintain
* B_CACHE if it was previously set). Resize the
* nfsnode after we have locked the buffer to prevent
* readers from reading garbage.
*/
bcount = on;
bp = nfs_getcacheblk(vp, lbn, bcount, td);
if (bp != NULL) {
long save;
mtx_lock(&np->n_mtx);
np->n_size = uio->uio_offset + n;
np->n_flag |= NMODIFIED;
vnode_pager_setsize(vp, np->n_size);
mtx_unlock(&np->n_mtx);
save = bp->b_flags & B_CACHE;
bcount += n;
allocbuf(bp, bcount);
bp->b_flags |= save;
}
} else {
/*
* Obtain the locked cache block first, and then
* adjust the file's size as appropriate.
*/
bcount = on + n;
if ((off_t)lbn * biosize + bcount < np->n_size) {
if ((off_t)(lbn + 1) * biosize < np->n_size)
bcount = biosize;
else
bcount = np->n_size - (off_t)lbn * biosize;
}
mtx_unlock(&np->n_mtx);
bp = nfs_getcacheblk(vp, lbn, bcount, td);
mtx_lock(&np->n_mtx);
if (uio->uio_offset + n > np->n_size) {
np->n_size = uio->uio_offset + n;
np->n_flag |= NMODIFIED;
vnode_pager_setsize(vp, np->n_size);
}
mtx_unlock(&np->n_mtx);
}
if (!bp) {
error = nfs_sigintr(nmp, td);
if (!error)
error = EINTR;
break;
}
/*
* Issue a READ if B_CACHE is not set. In special-append
* mode, B_CACHE is based on the buffer prior to the write
* op and is typically set, avoiding the read. If a read
* is required in special append mode, the server will
* probably send us a short-read since we extended the file
* on our end, resulting in b_resid == 0 and, thusly,
* B_CACHE getting set.
*
* We can also avoid issuing the read if the write covers
* the entire buffer. We have to make sure the buffer state
* is reasonable in this case since we will not be initiating
* I/O. See the comments in kern/vfs_bio.c's getblk() for
* more information.
*
* B_CACHE may also be set due to the buffer being cached
* normally.
*/
if (on == 0 && n == bcount) {
bp->b_flags |= B_CACHE;
bp->b_flags &= ~B_INVAL;
bp->b_ioflags &= ~BIO_ERROR;
}
if ((bp->b_flags & B_CACHE) == 0) {
bp->b_iocmd = BIO_READ;
vfs_busy_pages(bp, 0);
error = nfs_doio(vp, bp, cred, td);
if (error) {
brelse(bp);
break;
}
}
if (bp->b_wcred == NOCRED)
bp->b_wcred = crhold(cred);
mtx_lock(&np->n_mtx);
np->n_flag |= NMODIFIED;
mtx_unlock(&np->n_mtx);
/*
* If dirtyend exceeds file size, chop it down. This should
* not normally occur but there is an append race where it
* might occur XXX, so we log it.
*
* If the chopping creates a reverse-indexed or degenerate
* situation with dirtyoff/end, we 0 both of them.
*/
if (bp->b_dirtyend > bcount) {
nfs_printf("NFS append race @%lx:%d\n",
(long)bp->b_blkno * DEV_BSIZE,
bp->b_dirtyend - bcount);
bp->b_dirtyend = bcount;
}
if (bp->b_dirtyoff >= bp->b_dirtyend)
bp->b_dirtyoff = bp->b_dirtyend = 0;
/*
* If the new write will leave a contiguous dirty
* area, just update the b_dirtyoff and b_dirtyend,
* otherwise force a write rpc of the old dirty area.
*
* While it is possible to merge discontiguous writes due to
* our having a B_CACHE buffer ( and thus valid read data
* for the hole), we don't because it could lead to
* significant cache coherency problems with multiple clients,
* especially if locking is implemented later on.
*
* as an optimization we could theoretically maintain
* a linked list of discontinuous areas, but we would still
* have to commit them separately so there isn't much
* advantage to it except perhaps a bit of asynchronization.
*/
if (bp->b_dirtyend > 0 &&
(on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) {
if (bwrite(bp) == EINTR) {
error = EINTR;
break;
}
goto again;
}
error = uiomove((char *)bp->b_data + on, n, uio);
/*
* Since this block is being modified, it must be written
* again and not just committed. Since write clustering does
* not work for the stage 1 data write, only the stage 2
* commit rpc, we have to clear B_CLUSTEROK as well.
*/
bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
if (error) {
bp->b_ioflags |= BIO_ERROR;
brelse(bp);
break;
}
/*
* Only update dirtyoff/dirtyend if not a degenerate
* condition.
*/
if (n) {
if (bp->b_dirtyend > 0) {
bp->b_dirtyoff = min(on, bp->b_dirtyoff);
bp->b_dirtyend = max((on + n), bp->b_dirtyend);
} else {
bp->b_dirtyoff = on;
bp->b_dirtyend = on + n;
}
vfs_bio_set_valid(bp, on, n);
}
/*
* If IO_SYNC do bwrite().
*
* IO_INVAL appears to be unused. The idea appears to be
* to turn off caching in this case. Very odd. XXX
*/
if ((ioflag & IO_SYNC)) {
if (ioflag & IO_INVAL)
bp->b_flags |= B_NOCACHE;
error = bwrite(bp);
if (error)
break;
} else if ((n + on) == biosize) {
bp->b_flags |= B_ASYNC;
(void) (nmp->nm_rpcops->nr_writebp)(bp, 0, NULL);
} else {
bdwrite(bp);
}
} while (uio->uio_resid > 0 && n > 0);
return (error);
}
/*
* Get an nfs cache block.
*
* Allocate a new one if the block isn't currently in the cache
* and return the block marked busy. If the calling process is
* interrupted by a signal for an interruptible mount point, return
* NULL.
*
* The caller must carefully deal with the possible B_INVAL state of
* the buffer. nfs_doio() clears B_INVAL (and nfs_asyncio() clears it
* indirectly), so synchronous reads can be issued without worrying about
* the B_INVAL state. We have to be a little more careful when dealing
* with writes (see comments in nfs_write()) when extending a file past
* its EOF.
*/
static struct buf *
nfs_getcacheblk(struct vnode *vp, daddr_t bn, int size, struct thread *td)
{
struct buf *bp;
struct mount *mp;
struct nfsmount *nmp;
mp = vp->v_mount;
nmp = VFSTONFS(mp);
if (nmp->nm_flag & NFSMNT_INT) {
sigset_t oldset;
nfs_set_sigmask(td, &oldset);
bp = getblk(vp, bn, size, NFS_PCATCH, 0, 0);
nfs_restore_sigmask(td, &oldset);
while (bp == NULL) {
if (nfs_sigintr(nmp, td))
return (NULL);
bp = getblk(vp, bn, size, 0, 2 * hz, 0);
}
} else {
bp = getblk(vp, bn, size, 0, 0, 0);
}
- if (vp->v_type == VREG) {
- int biosize;
-
- biosize = mp->mnt_stat.f_iosize;
- bp->b_blkno = bn * (biosize / DEV_BSIZE);
- }
+ if (vp->v_type == VREG)
+ bp->b_blkno = bn * (vp->v_bufobj.bo_bsize / DEV_BSIZE);
return (bp);
}
/*
* Flush and invalidate all dirty buffers. If another process is already
* doing the flush, just wait for completion.
*/
int
nfs_vinvalbuf(struct vnode *vp, int flags, struct thread *td, int intrflg)
{
struct nfsnode *np = VTONFS(vp);
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
int error = 0, slpflag, slptimeo;
int old_lock = 0;
ASSERT_VOP_LOCKED(vp, "nfs_vinvalbuf");
if ((nmp->nm_flag & NFSMNT_INT) == 0)
intrflg = 0;
if (intrflg) {
slpflag = NFS_PCATCH;
slptimeo = 2 * hz;
} else {
slpflag = 0;
slptimeo = 0;
}
old_lock = nfs_upgrade_vnlock(vp);
if (vp->v_iflag & VI_DOOMED) {
/*
* Since vgonel() uses the generic vinvalbuf() to flush
* dirty buffers and it does not call this function, it
* is safe to just return OK when VI_DOOMED is set.
*/
nfs_downgrade_vnlock(vp, old_lock);
return (0);
}
/*
* Now, flush as required.
*/
if ((flags & V_SAVE) && (vp->v_bufobj.bo_object != NULL)) {
VM_OBJECT_LOCK(vp->v_bufobj.bo_object);
vm_object_page_clean(vp->v_bufobj.bo_object, 0, 0, OBJPC_SYNC);
VM_OBJECT_UNLOCK(vp->v_bufobj.bo_object);
/*
* If the page clean was interrupted, fail the invalidation.
* Not doing so, we run the risk of losing dirty pages in the
* vinvalbuf() call below.
*/
if (intrflg && (error = nfs_sigintr(nmp, td)))
goto out;
}
error = vinvalbuf(vp, flags, slpflag, 0);
while (error) {
if (intrflg && (error = nfs_sigintr(nmp, td)))
goto out;
error = vinvalbuf(vp, flags, 0, slptimeo);
}
mtx_lock(&np->n_mtx);
if (np->n_directio_asyncwr == 0)
np->n_flag &= ~NMODIFIED;
mtx_unlock(&np->n_mtx);
out:
nfs_downgrade_vnlock(vp, old_lock);
return error;
}
/*
* Initiate asynchronous I/O. Return an error if no nfsiods are available.
* This is mainly to avoid queueing async I/O requests when the nfsiods
* are all hung on a dead server.
*
* Note: nfs_asyncio() does not clear (BIO_ERROR|B_INVAL) but when the bp
* is eventually dequeued by the async daemon, nfs_doio() *will*.
*/
int
nfs_asyncio(struct nfsmount *nmp, struct buf *bp, struct ucred *cred, struct thread *td)
{
int iod;
int gotiod;
int slpflag = 0;
int slptimeo = 0;
int error, error2;
/*
* Commits are usually short and sweet so lets save some cpu and
* leave the async daemons for more important rpc's (such as reads
* and writes).
*/
mtx_lock(&nfs_iod_mtx);
if (bp->b_iocmd == BIO_WRITE && (bp->b_flags & B_NEEDCOMMIT) &&
(nmp->nm_bufqiods > nfs_numasync / 2)) {
mtx_unlock(&nfs_iod_mtx);
return(EIO);
}
again:
if (nmp->nm_flag & NFSMNT_INT)
slpflag = NFS_PCATCH;
gotiod = FALSE;
/*
* Find a free iod to process this request.
*/
for (iod = 0; iod < nfs_numasync; iod++)
if (nfs_iodwant[iod] == NFSIOD_AVAILABLE) {
gotiod = TRUE;
break;
}
/*
* Try to create one if none are free.
*/
if (!gotiod)
nfs_nfsiodnew();
else {
/*
* Found one, so wake it up and tell it which
* mount to process.
*/
NFS_DPF(ASYNCIO, ("nfs_asyncio: waking iod %d for mount %p\n",
iod, nmp));
nfs_iodwant[iod] = NFSIOD_NOT_AVAILABLE;
nfs_iodmount[iod] = nmp;
nmp->nm_bufqiods++;
wakeup(&nfs_iodwant[iod]);
}
/*
* If none are free, we may already have an iod working on this mount
* point. If so, it will process our request.
*/
if (!gotiod) {
if (nmp->nm_bufqiods > 0) {
NFS_DPF(ASYNCIO,
("nfs_asyncio: %d iods are already processing mount %p\n",
nmp->nm_bufqiods, nmp));
gotiod = TRUE;
}
}
/*
* If we have an iod which can process the request, then queue
* the buffer.
*/
if (gotiod) {
/*
* Ensure that the queue never grows too large. We still want
* to asynchronize so we block rather then return EIO.
*/
while (nmp->nm_bufqlen >= 2 * nfs_numasync) {
NFS_DPF(ASYNCIO,
("nfs_asyncio: waiting for mount %p queue to drain\n", nmp));
nmp->nm_bufqwant = TRUE;
error = nfs_msleep(td, &nmp->nm_bufq, &nfs_iod_mtx,
slpflag | PRIBIO,
"nfsaio", slptimeo);
if (error) {
error2 = nfs_sigintr(nmp, td);
if (error2) {
mtx_unlock(&nfs_iod_mtx);
return (error2);
}
if (slpflag == NFS_PCATCH) {
slpflag = 0;
slptimeo = 2 * hz;
}
}
/*
* We might have lost our iod while sleeping,
* so check and loop if nescessary.
*/
goto again;
}
/* We might have lost our nfsiod */
if (nmp->nm_bufqiods == 0) {
NFS_DPF(ASYNCIO,
("nfs_asyncio: no iods after mount %p queue was drained, looping\n", nmp));
goto again;
}
if (bp->b_iocmd == BIO_READ) {
if (bp->b_rcred == NOCRED && cred != NOCRED)
bp->b_rcred = crhold(cred);
} else {
if (bp->b_wcred == NOCRED && cred != NOCRED)
bp->b_wcred = crhold(cred);
}
if (bp->b_flags & B_REMFREE)
bremfreef(bp);
BUF_KERNPROC(bp);
TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist);
nmp->nm_bufqlen++;
if ((bp->b_flags & B_DIRECT) && bp->b_iocmd == BIO_WRITE) {
mtx_lock(&(VTONFS(bp->b_vp))->n_mtx);
VTONFS(bp->b_vp)->n_flag |= NMODIFIED;
VTONFS(bp->b_vp)->n_directio_asyncwr++;
mtx_unlock(&(VTONFS(bp->b_vp))->n_mtx);
}
mtx_unlock(&nfs_iod_mtx);
return (0);
}
mtx_unlock(&nfs_iod_mtx);
/*
* All the iods are busy on other mounts, so return EIO to
* force the caller to process the i/o synchronously.
*/
NFS_DPF(ASYNCIO, ("nfs_asyncio: no iods available, i/o is synchronous\n"));
return (EIO);
}
void
nfs_doio_directwrite(struct buf *bp)
{
int iomode, must_commit;
struct uio *uiop = (struct uio *)bp->b_caller1;
char *iov_base = uiop->uio_iov->iov_base;
struct nfsmount *nmp = VFSTONFS(bp->b_vp->v_mount);
iomode = NFSV3WRITE_FILESYNC;
uiop->uio_td = NULL; /* NULL since we're in nfsiod */
(nmp->nm_rpcops->nr_writerpc)(bp->b_vp, uiop, bp->b_wcred, &iomode, &must_commit);
KASSERT((must_commit == 0), ("nfs_doio_directwrite: Did not commit write"));
free(iov_base, M_NFSDIRECTIO);
free(uiop->uio_iov, M_NFSDIRECTIO);
free(uiop, M_NFSDIRECTIO);
if ((bp->b_flags & B_DIRECT) && bp->b_iocmd == BIO_WRITE) {
struct nfsnode *np = VTONFS(bp->b_vp);
mtx_lock(&np->n_mtx);
np->n_directio_asyncwr--;
if (np->n_directio_asyncwr == 0) {
VTONFS(bp->b_vp)->n_flag &= ~NMODIFIED;
if ((np->n_flag & NFSYNCWAIT)) {
np->n_flag &= ~NFSYNCWAIT;
wakeup((caddr_t)&np->n_directio_asyncwr);
}
}
mtx_unlock(&np->n_mtx);
}
bp->b_vp = NULL;
relpbuf(bp, &nfs_pbuf_freecnt);
}
/*
* Do an I/O operation to/from a cache block. This may be called
* synchronously or from an nfsiod.
*/
int
nfs_doio(struct vnode *vp, struct buf *bp, struct ucred *cr, struct thread *td)
{
struct uio *uiop;
struct nfsnode *np;
struct nfsmount *nmp;
int error = 0, iomode, must_commit = 0;
struct uio uio;
struct iovec io;
struct proc *p = td ? td->td_proc : NULL;
uint8_t iocmd;
np = VTONFS(vp);
nmp = VFSTONFS(vp->v_mount);
uiop = &uio;
uiop->uio_iov = &io;
uiop->uio_iovcnt = 1;
uiop->uio_segflg = UIO_SYSSPACE;
uiop->uio_td = td;
/*
* clear BIO_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_INVAL;
bp->b_ioflags &= ~BIO_ERROR;
KASSERT(!(bp->b_flags & B_DONE), ("nfs_doio: bp %p already marked done", bp));
iocmd = bp->b_iocmd;
if (iocmd == BIO_READ) {
io.iov_len = uiop->uio_resid = bp->b_bcount;
io.iov_base = bp->b_data;
uiop->uio_rw = UIO_READ;
switch (vp->v_type) {
case VREG:
uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE;
nfsstats.read_bios++;
error = (nmp->nm_rpcops->nr_readrpc)(vp, uiop, cr);
if (!error) {
if (uiop->uio_resid) {
/*
* If we had a short read with no error, we must have
* hit a file hole. We should zero-fill the remainder.
* This can also occur if the server hits the file EOF.
*
* Holes used to be able to occur due to pending
* writes, but that is not possible any longer.
*/
int nread = bp->b_bcount - uiop->uio_resid;
int left = uiop->uio_resid;
if (left > 0)
bzero((char *)bp->b_data + nread, left);
uiop->uio_resid = 0;
}
}
/* ASSERT_VOP_LOCKED(vp, "nfs_doio"); */
if (p && (vp->v_vflag & VV_TEXT)) {
mtx_lock(&np->n_mtx);
if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &np->n_vattr.va_mtime)) {
mtx_unlock(&np->n_mtx);
PROC_LOCK(p);
killproc(p, "text file modification");
PROC_UNLOCK(p);
} else
mtx_unlock(&np->n_mtx);
}
break;
case VLNK:
uiop->uio_offset = (off_t)0;
nfsstats.readlink_bios++;
error = (nmp->nm_rpcops->nr_readlinkrpc)(vp, uiop, cr);
break;
case VDIR:
nfsstats.readdir_bios++;
uiop->uio_offset = ((u_quad_t)bp->b_lblkno) * NFS_DIRBLKSIZ;
if ((nmp->nm_flag & NFSMNT_RDIRPLUS) != 0) {
error = nfs_readdirplusrpc(vp, uiop, cr);
if (error == NFSERR_NOTSUPP)
nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
}
if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0)
error = nfs_readdirrpc(vp, uiop, cr);
/*
* end-of-directory sets B_INVAL but does not generate an
* error.
*/
if (error == 0 && uiop->uio_resid == bp->b_bcount)
bp->b_flags |= B_INVAL;
break;
default:
nfs_printf("nfs_doio: type %x unexpected\n", vp->v_type);
break;
};
if (error) {
bp->b_ioflags |= BIO_ERROR;
bp->b_error = error;
}
} else {
/*
* If we only need to commit, try to commit
*/
if (bp->b_flags & B_NEEDCOMMIT) {
int retv;
off_t off;
off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff;
retv = (nmp->nm_rpcops->nr_commit)(
vp, off, bp->b_dirtyend-bp->b_dirtyoff,
bp->b_wcred, td);
if (retv == 0) {
bp->b_dirtyoff = bp->b_dirtyend = 0;
bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
bp->b_resid = 0;
bufdone(bp);
return (0);
}
if (retv == NFSERR_STALEWRITEVERF) {
nfs_clearcommit(vp->v_mount);
}
}
/*
* Setup for actual write
*/
mtx_lock(&np->n_mtx);
if ((off_t)bp->b_blkno * DEV_BSIZE + bp->b_dirtyend > np->n_size)
bp->b_dirtyend = np->n_size - (off_t)bp->b_blkno * DEV_BSIZE;
mtx_unlock(&np->n_mtx);
if (bp->b_dirtyend > bp->b_dirtyoff) {
io.iov_len = uiop->uio_resid = bp->b_dirtyend
- bp->b_dirtyoff;
uiop->uio_offset = (off_t)bp->b_blkno * DEV_BSIZE
+ bp->b_dirtyoff;
io.iov_base = (char *)bp->b_data + bp->b_dirtyoff;
uiop->uio_rw = UIO_WRITE;
nfsstats.write_bios++;
if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE | B_CLUSTER)) == B_ASYNC)
iomode = NFSV3WRITE_UNSTABLE;
else
iomode = NFSV3WRITE_FILESYNC;
error = (nmp->nm_rpcops->nr_writerpc)(vp, uiop, cr, &iomode, &must_commit);
/*
* When setting B_NEEDCOMMIT also set B_CLUSTEROK to try
* to cluster the buffers needing commit. This will allow
* the system to submit a single commit rpc for the whole
* cluster. We can do this even if the buffer is not 100%
* dirty (relative to the NFS blocksize), so we optimize the
* append-to-file-case.
*
* (when clearing B_NEEDCOMMIT, B_CLUSTEROK must also be
* cleared because write clustering only works for commit
* rpc's, not for the data portion of the write).
*/
if (!error && iomode == NFSV3WRITE_UNSTABLE) {
bp->b_flags |= B_NEEDCOMMIT;
if (bp->b_dirtyoff == 0
&& bp->b_dirtyend == bp->b_bcount)
bp->b_flags |= B_CLUSTEROK;
} else {
bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
}
/*
* For an interrupted write, the buffer is still valid
* and the write hasn't been pushed to the server yet,
* so we can't set BIO_ERROR and report the interruption
* by setting B_EINTR. For the B_ASYNC case, B_EINTR
* is not relevant, so the rpc attempt is essentially
* a noop. For the case of a V3 write rpc not being
* committed to stable storage, the block is still
* dirty and requires either a commit rpc or another
* write rpc with iomode == NFSV3WRITE_FILESYNC before
* the block is reused. This is indicated by setting
* the B_DELWRI and B_NEEDCOMMIT flags.
*
* If the buffer is marked B_PAGING, it does not reside on
* the vp's paging queues so we cannot call bdirty(). The
* bp in this case is not an NFS cache block so we should
* be safe. XXX
*
* The logic below breaks up errors into recoverable and
* unrecoverable. For the former, we clear B_INVAL|B_NOCACHE
* and keep the buffer around for potential write retries.
* For the latter (eg ESTALE), we toss the buffer away (B_INVAL)
* and save the error in the nfsnode. This is less than ideal
* but necessary. Keeping such buffers around could potentially
* cause buffer exhaustion eventually (they can never be written
* out, so will get constantly be re-dirtied). It also causes
* all sorts of vfs panics. For non-recoverable write errors,
* also invalidate the attrcache, so we'll be forced to go over
* the wire for this object, returning an error to user on next
* call (most of the time).
*/
if (error == EINTR || error == EIO || error == ETIMEDOUT
|| (!error && (bp->b_flags & B_NEEDCOMMIT))) {
int s;
s = splbio();
bp->b_flags &= ~(B_INVAL|B_NOCACHE);
if ((bp->b_flags & B_PAGING) == 0) {
bdirty(bp);
bp->b_flags &= ~B_DONE;
}
if (error && (bp->b_flags & B_ASYNC) == 0)
bp->b_flags |= B_EINTR;
splx(s);
} else {
if (error) {
bp->b_ioflags |= BIO_ERROR;
bp->b_flags |= B_INVAL;
bp->b_error = np->n_error = error;
mtx_lock(&np->n_mtx);
np->n_flag |= NWRITEERR;
np->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
mtx_unlock(&np->n_mtx);
}
bp->b_dirtyoff = bp->b_dirtyend = 0;
}
} else {
bp->b_resid = 0;
bufdone(bp);
return (0);
}
}
bp->b_resid = uiop->uio_resid;
if (must_commit)
nfs_clearcommit(vp->v_mount);
bufdone(bp);
return (error);
}
/*
* Used to aid in handling ftruncate() operations on the NFS client side.
* Truncation creates a number of special problems for NFS. We have to
* throw away VM pages and buffer cache buffers that are beyond EOF, and
* we have to properly handle VM pages or (potentially dirty) buffers
* that straddle the truncation point.
*/
int
nfs_meta_setsize(struct vnode *vp, struct ucred *cred, struct thread *td, u_quad_t nsize)
{
struct nfsnode *np = VTONFS(vp);
u_quad_t tsize;
- int biosize = vp->v_mount->mnt_stat.f_iosize;
+ int biosize = vp->v_bufobj.bo_bsize;
int error = 0;
mtx_lock(&np->n_mtx);
tsize = np->n_size;
np->n_size = nsize;
mtx_unlock(&np->n_mtx);
if (nsize < tsize) {
struct buf *bp;
daddr_t lbn;
int bufsize;
/*
* vtruncbuf() doesn't get the buffer overlapping the
* truncation point. We may have a B_DELWRI and/or B_CACHE
* buffer that now needs to be truncated.
*/
error = vtruncbuf(vp, cred, td, nsize, biosize);
lbn = nsize / biosize;
bufsize = nsize & (biosize - 1);
bp = nfs_getcacheblk(vp, lbn, bufsize, td);
if (!bp)
return EINTR;
if (bp->b_dirtyoff > bp->b_bcount)
bp->b_dirtyoff = bp->b_bcount;
if (bp->b_dirtyend > bp->b_bcount)
bp->b_dirtyend = bp->b_bcount;
bp->b_flags |= B_RELBUF; /* don't leave garbage around */
brelse(bp);
} else {
vnode_pager_setsize(vp, nsize);
}
return(error);
}
Index: stable/8/sys
===================================================================
--- stable/8/sys (revision 231331)
+++ stable/8/sys (revision 231332)
Property changes on: stable/8/sys
___________________________________________________________________
Modified: svn:mergeinfo
## -0,0 +0,1 ##
Merged /head/sys:r230605

File Metadata

Mime Type
text/x-c
Expires
Sat, Aug 15, 5:34 PM (2 d)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
36744496
Default Alt Text
(151 KB)

Event Timeline