/*
* 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.
* 3. 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
*/
/*
* 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.
*
* 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* nfs_attrtimeo seconds out of date.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.
*/
if (vp->v_type != VLNK) {
error = nfs_flushstalebuf(vp, cred, l,
NFS_FLUSHSTALEBUF_MYWRITE);
if (error)
return error;
}
nfs_delayedtruncate(vp);
if (np->n_size <= uio->uio_offset) {
break;
}
bytelen =
MIN(np->n_size - uio->uio_offset, uio->uio_resid);
error = ubc_uiomove(&vp->v_uobj, uio, bytelen, advice,
UBC_READ | UBC_PARTIALOK | UBC_VNODE_FLAGS(vp));
if (error) {
/*
* XXXkludge
* the file has been truncated on the server.
* there isn't much we can do.
*/
if (uio->uio_offset >= np->n_size) {
/* end of file */
error = 0;
} else {
break;
}
}
}
break;
case VLNK:
nfsstats.biocache_readlinks++;
bp = nfs_getcacheblk(vp, (daddr_t)0, MAXPATHLEN, l);
if (!bp)
return (EINTR);
if ((bp->b_oflags & BO_DONE) == 0) {
bp->b_flags |= B_READ;
error = nfs_doio(bp);
if (error) {
brelse(bp, 0);
return (error);
}
}
n = MIN(uio->uio_resid, MAXPATHLEN - bp->b_resid);
got_buf = 1;
on = 0;
break;
case VDIR:
diragain:
nfsstats.biocache_readdirs++;
ndp = nfs_searchdircache(vp, uio->uio_offset,
(nmp->nm_flag & NFSMNT_XLATECOOKIE), 0);
if (!ndp) {
/*
* We've been handed a cookie that is not
* in the cache. If we're not translating
* 32 <-> 64, it may be a value that was
* flushed out of the cache because it grew
* too big. Let the server judge if it's
* valid or not. In the translation case,
* we have no way of validating this value,
* so punt.
*/
if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
return (EINVAL);
ndp = nfs_enterdircache(vp, uio->uio_offset,
uio->uio_offset, 0, 0);
}
bp = nfs_getcacheblk(vp, NFSDC_BLKNO(ndp), NFS_DIRBLKSIZ, l);
if (!bp)
return (EINTR);
if ((bp->b_oflags & BO_DONE) == 0) {
bp->b_flags |= B_READ;
bp->b_dcookie = ndp->dc_blkcookie;
error = nfs_doio(bp);
if (error) {
/*
* Yuck! The directory has been modified on the
* server. Punt and let the userland code
* deal with it.
*/
nfs_putdircache(np, ndp);
brelse(bp, 0);
/*
* nfs_request maps NFSERR_BAD_COOKIE to EINVAL.
*/
if (error == EINVAL) { /* NFSERR_BAD_COOKIE */
nfs_invaldircache(vp, 0);
nfs_vinvalbuf(vp, 0, cred, l, 1);
}
return (error);
}
}
/*
* Just return if we hit EOF right away with this
* block. Always check here, because direofoffset
* may have been set by an nfsiod since the last
* check.
*
* also, empty block implies EOF.
*/
/*
* If the entry number was bigger than the number of
* entries in the block, or the cookie of the previous
* entry doesn't match, the directory cache is
* stale. Flush it and try again (i.e. go to
* the server).
*/
if (dp >= edp || (struct dirent *)_DIRENT_NEXT(dp) > edp ||
(en > 0 && NFS_GETCOOKIE(pdp) != ndp->dc_cookie)) {
#ifdef DEBUG
printf("invalid cache: %p %p %p off %jx %jx\n",
pdp, dp, edp,
(uintmax_t)uio->uio_offset,
(uintmax_t)NFS_GETCOOKIE(pdp));
#endif
nfs_putdircache(np, ndp);
brelse(bp, 0);
nfs_invaldircache(vp, 0);
nfs_vinvalbuf(vp, 0, cred, l, 0);
goto diragain;
}
on = (char *)dp - (char *)bp->b_data;
/*
* Cache all entries that may be exported to the
* user, as they may be thrown back at us. The
* NFSBIO_CACHECOOKIES flag indicates that all
* entries are being 'exported', so cache them all.
*/
if (en == 0 && pdp == dp) {
dp = _DIRENT_NEXT(dp);
enn++;
}
if (uio->uio_resid < (bp->b_bcount - bp->b_resid - on)) {
n = uio->uio_resid;
enough = 1;
} else
n = bp->b_bcount - bp->b_resid - on;
ep = (struct dirent *)(void *)((char *)bp->b_data + on + n);
/*
* Find last complete entry to copy, caching entries
* (if requested) as we go.
*/
/*
* If the last requested entry was not the last in the
* buffer (happens if NFS_DIRFRAGSIZ < NFS_DIRBLKSIZ),
* cache the cookie of the last requested one, and
* set of the offset to it.
*/
/*
* Always cache the entry for the next block,
* so that readaheads can use it.
*/
nndp = nfs_enterdircache(vp, bp->b_dcookie, bp->b_dcookie, 0,0);
if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
if (curoff == bp->b_dcookie) {
NFS_STASHCOOKIE32(pdp, nndp->dc_cookie32);
curoff = nndp->dc_cookie32;
}
}
n = (char *)_DIRENT_NEXT(pdp) - ((char *)bp->b_data + on);
/*
* 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 (nfs_numasync > 0 && nmp->nm_readahead > 0 &&
!NFS_EOFVALID(np)) {
rabp = nfs_getcacheblk(vp, NFSDC_BLKNO(nndp),
NFS_DIRBLKSIZ, l);
if (rabp) {
if ((rabp->b_oflags & (BO_DONE | BO_DELWRI)) == 0) {
rabp->b_dcookie = nndp->dc_cookie;
rabp->b_flags |= (B_READ | B_ASYNC);
if (nfs_asyncio(rabp)) {
brelse(rabp, BC_INVAL);
}
} else
brelse(rabp, 0);
}
}
nfs_putdircache(np, nndp);
got_buf = 1;
break;
default:
printf(" nfsbioread: type %x unexpected\n",vp->v_type);
break;
}
if (n > 0) {
if (!baddr)
baddr = bp->b_data;
error = uiomove((char *)baddr + on, (int)n, uio);
}
switch (vp->v_type) {
case VREG:
break;
case VLNK:
n = 0;
break;
case VDIR:
uio->uio_offset = curoff;
if (enough)
n = 0;
break;
default:
printf(" nfsbioread: type %x unexpected\n",vp->v_type);
}
if (got_buf)
brelse(bp, 0);
} while (error == 0 && uio->uio_resid > 0 && n > 0);
return (error);
}
/*
* Vnode op for write using bio
*/
int
nfs_write(void *v)
{
struct vop_write_args /* {
struct vnode *a_vp;
struct uio *a_uio;
int a_ioflag;
kauth_cred_t a_cred;
} */ *ap = v;
struct uio *uio = ap->a_uio;
struct lwp *l = curlwp;
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
kauth_cred_t cred = ap->a_cred;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
voff_t oldoff, origoff;
vsize_t bytelen;
int error = 0;
int ioflag = ap->a_ioflag;
#ifdef DIAGNOSTIC
if (uio->uio_rw != UIO_WRITE)
panic("nfs_write mode");
#endif
if (vp->v_type != VREG)
return (EIO);
if (np->n_flag & NWRITEERR) {
np->n_flag &= ~NWRITEERR;
return (np->n_error);
}
#ifndef NFS_V2_ONLY
if ((nmp->nm_flag & NFSMNT_NFSV3) &&
!(nmp->nm_iflag & NFSMNT_GOTFSINFO))
(void)nfs_fsinfo(nmp, vp, cred, l);
#endif
if (ioflag & IO_APPEND) {
NFS_INVALIDATE_ATTRCACHE(np);
error = nfs_flushstalebuf(vp, cred, l,
NFS_FLUSHSTALEBUF_MYWRITE);
if (error)
return (error);
uio->uio_offset = np->n_size;
/*
* This is already checked above VOP_WRITE, but recheck
* the append case here to make sure our idea of the
* file size is as fresh as possible.
*/
if (uio->uio_offset + uio->uio_resid >
l->l_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
mutex_enter(&proc_lock);
psignal(l->l_proc, SIGXFSZ);
mutex_exit(&proc_lock);
return (EFBIG);
}
}
if (uio->uio_offset < 0)
return (EINVAL);
if ((uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize)
return (EFBIG);
if (uio->uio_resid == 0)
return (0);
origoff = uio->uio_offset;
do {
bool overwrite; /* if we are overwriting whole pages */
u_quad_t oldsize;
oldoff = uio->uio_offset;
bytelen = uio->uio_resid;
/*
* 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.
*/
struct buf *
nfs_getcacheblk(struct vnode *vp, daddr_t bn, int size, struct lwp *l)
{
struct buf *bp;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
if (nmp->nm_flag & NFSMNT_INT) {
bp = getblk(vp, bn, size, PCATCH, 0);
while (bp == NULL) {
if (nfs_sigintr(nmp, NULL, l))
return (NULL);
bp = getblk(vp, bn, size, 0, 2 * hz);
}
} else
bp = getblk(vp, bn, size, 0, 0);
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, kauth_cred_t cred,
struct lwp *l, int intrflg)
{
struct nfsnode *np = VTONFS(vp);
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
int error = 0, allerror = 0, slptimeo;
bool catch_p;
if ((nmp->nm_flag & NFSMNT_INT) == 0)
intrflg = 0;
if (intrflg) {
catch_p = true;
slptimeo = 2 * hz;
} else {
catch_p = false;
if (nmp->nm_flag & NFSMNT_SOFT)
slptimeo = nmp->nm_retry * nmp->nm_timeo;
else
slptimeo = 0;
}
/*
* First wait for any other process doing a flush to complete.
*/
mutex_enter(vp->v_interlock);
while (np->n_flag & NFLUSHINPROG) {
np->n_flag |= NFLUSHWANT;
error = mtsleep(&np->n_flag, PRIBIO + 2, "nfsvinval",
slptimeo, vp->v_interlock);
if (error && intrflg && nfs_sigintr(nmp, NULL, l)) {
mutex_exit(vp->v_interlock);
return EINTR;
}
}
/*
* Now, flush as required.
*/
np->n_flag |= NFLUSHINPROG;
mutex_exit(vp->v_interlock);
error = vinvalbuf(vp, flags, cred, l, catch_p, 0);
while (error) {
if (allerror == 0)
allerror = error;
if (intrflg && nfs_sigintr(nmp, NULL, l)) {
error = EINTR;
break;
}
error = vinvalbuf(vp, flags, cred, l, 0, slptimeo);
}
mutex_enter(vp->v_interlock);
if (allerror != 0) {
/*
* Keep error from vinvalbuf so fsync/close will know.
*/
np->n_error = allerror;
np->n_flag |= NWRITEERR;
}
if (error == 0)
np->n_flag &= ~NMODIFIED;
np->n_flag &= ~NFLUSHINPROG;
if (np->n_flag & NFLUSHWANT) {
np->n_flag &= ~NFLUSHWANT;
wakeup(&np->n_flag);
}
mutex_exit(vp->v_interlock);
return error;
}
/*
* nfs_flushstalebuf: flush cache if it's stale.
*
* => caller shouldn't own any pages or buffers which belong to the vnode.
*/
int
nfs_flushstalebuf(struct vnode *vp, kauth_cred_t cred, struct lwp *l,
int flags)
{
struct nfsnode *np = VTONFS(vp);
struct vattr vattr;
int 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.
*/
int
nfs_asyncio(struct buf *bp)
{
struct nfs_iod *iod;
struct nfsmount *nmp;
int slptimeo = 0, error;
bool catch_p = false;
if (nfs_numasync == 0)
return (EIO);
nmp = VFSTONFS(bp->b_vp->v_mount);
if (nmp->nm_flag & NFSMNT_SOFT)
slptimeo = nmp->nm_retry * nmp->nm_timeo;
if (nmp->nm_iflag & NFSMNT_DISMNTFORCE)
slptimeo = hz;
again:
if (nmp->nm_flag & NFSMNT_INT)
catch_p = true;
/*
* Find a free iod to process this request.
*/
mutex_enter(&nfs_iodlist_lock);
iod = LIST_FIRST(&nfs_iodlist_idle);
if (iod) {
/*
* Found one, so wake it up and tell it which
* mount to process.
*/
LIST_REMOVE(iod, nid_idle);
mutex_enter(&iod->nid_lock);
mutex_exit(&nfs_iodlist_lock);
KASSERT(iod->nid_mount == NULL);
iod->nid_mount = nmp;
cv_signal(&iod->nid_cv);
mutex_enter(&nmp->nm_lock);
mutex_exit(&iod->nid_lock);
nmp->nm_bufqiods++;
if (nmp->nm_bufqlen < 2 * nmp->nm_bufqiods) {
cv_broadcast(&nmp->nm_aiocv);
}
} else {
mutex_exit(&nfs_iodlist_lock);
mutex_enter(&nmp->nm_lock);
}
KASSERT(mutex_owned(&nmp->nm_lock));
/*
* If we have an iod which can process the request, then queue
* the buffer. However, even if we have an iod, do not initiate
* queue cleaning if curproc is the pageout daemon. if the NFS mount
* is via local loopback, we may put curproc (pagedaemon) to sleep
* waiting for the writes to complete. But the server (ourself)
* may block the write, waiting for its (ie., our) pagedaemon
* to produce clean pages to handle the write: deadlock.
* XXX: start non-loopback mounts straight away? If "lots free",
* let pagedaemon start loopback writes anyway?
*/
if (nmp->nm_bufqiods > 0) {
/*
* Ensure that the queue never grows too large.
*/
if (curlwp == uvm.pagedaemon_lwp) {
/* Enqueue for later, to avoid free-page deadlock */
} else while (nmp->nm_bufqlen >= 2 * nmp->nm_bufqiods) {
if (catch_p) {
error = cv_timedwait_sig(&nmp->nm_aiocv,
&nmp->nm_lock, slptimeo);
} else {
error = cv_timedwait(&nmp->nm_aiocv,
&nmp->nm_lock, slptimeo);
}
if (error) {
if (error == EWOULDBLOCK &&
nmp->nm_flag & NFSMNT_SOFT) {
mutex_exit(&nmp->nm_lock);
bp->b_error = EIO;
return (EIO);
}
uiop->uio_rw = UIO_READ;
switch (vp->v_type) {
case VREG:
nfsstats.read_bios++;
error = nfs_readrpc(vp, uiop);
if (!error && uiop->uio_resid) {
int diff, len;
/*
* If uio_resid > 0, there is a hole in the file and
* no writes after the hole have been pushed to
* the server yet or the file has been truncated
* on the server.
* Just zero fill the rest of the valid area.
*/
for (i = 0; i < npages; i++) {
pgs[i] = uvm_pageratop((vaddr_t)bp->b_data + (i << PAGE_SHIFT));
if (pgs[i]->uobject == uobj &&
pgs[i]->offset == uiop->uio_offset + (i << PAGE_SHIFT)) {
KASSERT(pgs[i]->flags & PG_BUSY);
/*
* this page belongs to our object.
*/
rw_enter(uobj->vmobjlock, RW_WRITER);
/*
* write out the page stably if it's about to
* be released because we can't resend it
* on the server crash.
*
* XXX assuming PG_RELEASE|PG_PAGEOUT won't be
* changed until unbusy the page.
*/
if (pgs[i]->flags & (PG_RELEASED|PG_PAGEOUT))
iomode = NFSV3WRITE_FILESYNC;
/*
* if we met a page which hasn't been sent yet,
* we need do WRITE RPC.
*/
if ((pgs[i]->flags & PG_NEEDCOMMIT) == 0)
needcommit = false;
rw_exit(uobj->vmobjlock);
} else {
iomode = NFSV3WRITE_FILESYNC;
needcommit = false;
}
}
if (!needcommit && iomode == NFSV3WRITE_UNSTABLE) {
rw_enter(uobj->vmobjlock, RW_WRITER);
for (i = 0; i < npages; i++) {
pgs[i]->flags |= PG_NEEDCOMMIT | PG_RDONLY;
pmap_page_protect(pgs[i], VM_PROT_READ);
}
rw_exit(uobj->vmobjlock);
pageprotected = true; /* pages can't be modified during i/o. */
} else
pageprotected = false;
/*
* Send the data to the server if necessary,
* otherwise just send a commit rpc.
*/
#ifndef NFS_V2_ONLY
if (needcommit) {
/*
* If the buffer is in the range that we already committed,
* there's nothing to do.
*
* If it's in the range that we need to commit, push the
* whole range at once, otherwise only push the buffer.
* In both these cases, acquire the commit lock to avoid
* other processes modifying the range.
*/
off = uiop->uio_offset;
cnt = bp->b_bcount;
mutex_enter(&np->n_commitlock);
if (!nfs_in_committed_range(vp, off, bp->b_bcount)) {
bool pushedrange;
if (nfs_in_tobecommitted_range(vp, off, bp->b_bcount)) {
pushedrange = true;
off = np->n_pushlo;
cnt = np->n_pushhi - np->n_pushlo;
} else {
pushedrange = false;
}
error = nfs_commit(vp, off, cnt, curlwp);
if (error == 0) {
if (pushedrange) {
nfs_merge_commit_ranges(vp);
} else {
nfs_add_committed_range(vp, off, cnt);
}
}
} else {
error = 0;
}
mutex_exit(&np->n_commitlock);
rw_exit(&nmp->nm_writeverflock);
if (!error) {
/*
* pages are now on stable storage.
*/
uiop->uio_resid = 0;
rw_enter(uobj->vmobjlock, RW_WRITER);
for (i = 0; i < npages; i++) {
pgs[i]->flags &= ~(PG_NEEDCOMMIT | PG_RDONLY);
}
rw_exit(uobj->vmobjlock);
goto out;
} else if (error == NFSERR_STALEWRITEVERF) {
nfs_clearcommit(vp->v_mount);
goto again;
}
if (error) {
bp->b_error = np->n_error = error;
np->n_flag |= NWRITEERR;
}
goto out;
}
#endif
off = uiop->uio_offset;
cnt = bp->b_bcount;
uiop->uio_rw = UIO_WRITE;
nfsstats.write_bios++;
error = nfs_writerpc(vp, uiop, &iomode, pageprotected, &stalewriteverf);
#ifndef NFS_V2_ONLY
if (!error && iomode == NFSV3WRITE_UNSTABLE) {
/*
* we need to commit pages later.
*/
mutex_enter(&np->n_commitlock);
nfs_add_tobecommitted_range(vp, off, cnt);
/*
* if there can be too many uncommitted pages, commit them now.
*/
if (np->n_pushhi - np->n_pushlo > nfs_commitsize) {
off = np->n_pushlo;
cnt = nfs_commitsize >> 1;
error = nfs_commit(vp, off, cnt, curlwp);
if (!error) {
nfs_add_committed_range(vp, off, cnt);
nfs_del_tobecommitted_range(vp, off, cnt);
}
if (error == NFSERR_STALEWRITEVERF) {
stalewriteverf = true;
error = 0; /* it isn't a real error */
}
} else {
/*
* re-dirty pages so that they will be passed
* to us later again.
*/
rw_enter(uobj->vmobjlock, RW_WRITER);
for (i = 0; i < npages; i++) {
uvm_pagemarkdirty(pgs[i],
UVM_PAGE_STATUS_DIRTY);
}
rw_exit(uobj->vmobjlock);
}
mutex_exit(&np->n_commitlock);
} else
#endif
if (!error) {
/*
* pages are now on stable storage.
*/
mutex_enter(&np->n_commitlock);
nfs_del_committed_range(vp, off, cnt);
mutex_exit(&np->n_commitlock);
rw_enter(uobj->vmobjlock, RW_WRITER);
for (i = 0; i < npages; i++) {
pgs[i]->flags &= ~(PG_NEEDCOMMIT | PG_RDONLY);
}
rw_exit(uobj->vmobjlock);
} else {
/*
* we got an error.
*/
bp->b_error = np->n_error = error;
np->n_flag |= NWRITEERR;
}
rw_exit(&nmp->nm_writeverflock);
if (stalewriteverf) {
nfs_clearcommit(vp->v_mount);
}
#ifndef NFS_V2_ONLY
out:
#endif
if (pgs != spgs)
kmem_free(pgs, sizeof(*pgs) * npages);
return error;
}
/*
* nfs_doio for B_PHYS.
*/
static int
nfs_doio_phys(struct buf *bp, struct uio *uiop)
{
struct vnode *vp = bp->b_vp;
int error;
/*
* Do an I/O operation to/from a cache block. This may be called
* synchronously or from an nfsiod.
*/
int
nfs_doio(struct buf *bp)
{
int error;
struct uio uio;
struct uio *uiop = &uio;
struct iovec io;
UVMHIST_FUNC("nfs_doio"); UVMHIST_CALLED(ubchist);
/*
* XXX NFS wants to modify the pages below and that can't be done
* with a read lock. We can't upgrade the lock here because it
* would screw up UVM fault processing. Have NFS take the I/O
* path.
*/
if (locked && rw_lock_op(uobj->vmobjlock) == RW_READER) {
*ap->a_count = 0;
ap->a_m[ap->a_centeridx] = NULL;
return EBUSY;
}
/*
* If we are not locked we are not really using opgs,
* so just initialize it
*/
if (!locked || npages < __arraycount(spgs))
opgs = spgs;
else {
if ((opgs = kmem_alloc(npages * sizeof(*opgs), KM_NOSLEEP)) ==
NULL)
return ENOMEM;
}
/*
* call the genfs code to get the pages. `pgs' may be NULL
* when doing read-ahead.
*/
pgs = ap->a_m;
if (write && locked && v3) {
KASSERT(pgs != NULL);
#ifdef DEBUG
/*
* If PGO_LOCKED is set, real pages shouldn't exists
* in the array.
*/
for (i = 0; i < npages; i++)
KDASSERT(pgs[i] == NULL || pgs[i] == PGO_DONTCARE);
#endif
memcpy(opgs, pgs, npages * sizeof(struct vm_pages *));
}
error = genfs_getpages(v);
if (error)
goto out;
/*
* for read faults where the nfs node is not yet marked NMODIFIED,
* set PG_RDONLY on the pages so that we come back here if someone
* tries to modify later via the mapping that will be entered for
* this fault.
*/
if (!write && (np->n_flag & NMODIFIED) == 0 && pgs != NULL) {
if (!locked) {
rw_enter(uobj->vmobjlock, RW_WRITER);
}
for (i = 0; i < npages; i++) {
pg = pgs[i];
if (pg == NULL || pg == PGO_DONTCARE) {
continue;
}
pg->flags |= PG_RDONLY;
}
if (!locked) {
rw_exit(uobj->vmobjlock);
}
}
if (!write)
goto out;
/*
* this is a write fault, update the commit info.
*/
origoffset = ap->a_offset;
len = npages << PAGE_SHIFT;
if (v3) {
if (!locked) {
mutex_enter(&np->n_commitlock);
} else {
if (!mutex_tryenter(&np->n_commitlock)) {
/*
* tell the caller that there are no pages
* available and put back original pgs array.
*/
