freebsd-dev/sys/nfsclient/nfs_vnops.c

3123 lines
81 KiB
C
Raw Normal View History

1994-05-24 10:09:53 +00:00
/*
* 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. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by the University of
* California, Berkeley and its contributors.
* 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_vnops.c 8.16 (Berkeley) 5/27/95
1994-05-24 10:09:53 +00:00
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
1994-05-24 10:09:53 +00:00
/*
* vnode op calls for Sun NFS version 2 and 3
1994-05-24 10:09:53 +00:00
*/
#include "opt_inet.h"
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/systm.h>
#include <sys/resourcevar.h>
#include <sys/proc.h>
1994-05-24 10:09:53 +00:00
#include <sys/mount.h>
#include <sys/bio.h>
1994-05-24 10:09:53 +00:00
#include <sys/buf.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/namei.h>
#include <sys/socket.h>
1994-05-24 10:09:53 +00:00
#include <sys/vnode.h>
#include <sys/dirent.h>
#include <sys/fcntl.h>
#include <sys/lockf.h>
#include <sys/stat.h>
#include <sys/sysctl.h>
1994-05-24 10:09:53 +00:00
#include <vm/vm.h>
#include <vm/vm_extern.h>
1994-05-24 10:09:53 +00:00
#include <fs/fifofs/fifo.h>
1994-05-24 10:09:53 +00:00
#include <nfs/rpcv2.h>
#include <nfs/nfsproto.h>
#include <nfsclient/nfs.h>
#include <nfsclient/nfsnode.h>
#include <nfsclient/nfsmount.h>
#include <nfsclient/nfs_lock.h>
1994-05-24 10:09:53 +00:00
#include <nfs/xdr_subs.h>
#include <nfsclient/nfsm_subs.h>
1994-05-24 10:09:53 +00:00
#include <net/if.h>
#include <netinet/in.h>
#include <netinet/in_var.h>
1994-05-24 10:09:53 +00:00
/* Defs */
#define TRUE 1
#define FALSE 0
/*
* Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
* calls are not in getblk() and brelse() so that they would not be necessary
* here.
*/
#ifndef B_VMIO
#define vfs_busy_pages(bp, f)
#endif
static int nfsspec_read(struct vop_read_args *);
static int nfsspec_write(struct vop_write_args *);
static int nfsfifo_read(struct vop_read_args *);
static int nfsfifo_write(struct vop_write_args *);
static int nfsspec_close(struct vop_close_args *);
static int nfsfifo_close(struct vop_close_args *);
static int nfs_flush(struct vnode *, struct ucred *, int, struct thread *,
int);
static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
struct thread *);
static int nfs_lookup(struct vop_lookup_args *);
static int nfs_create(struct vop_create_args *);
static int nfs_mknod(struct vop_mknod_args *);
static int nfs_open(struct vop_open_args *);
static int nfs_close(struct vop_close_args *);
static int nfs_access(struct vop_access_args *);
static int nfs_getattr(struct vop_getattr_args *);
static int nfs_setattr(struct vop_setattr_args *);
static int nfs_read(struct vop_read_args *);
static int nfs_fsync(struct vop_fsync_args *);
static int nfs_remove(struct vop_remove_args *);
static int nfs_link(struct vop_link_args *);
static int nfs_rename(struct vop_rename_args *);
static int nfs_mkdir(struct vop_mkdir_args *);
static int nfs_rmdir(struct vop_rmdir_args *);
static int nfs_symlink(struct vop_symlink_args *);
static int nfs_readdir(struct vop_readdir_args *);
static int nfs_strategy(struct vop_strategy_args *);
static int nfs_lookitup(struct vnode *, const char *, int,
struct ucred *, struct thread *, struct nfsnode **);
static int nfs_sillyrename(struct vnode *, struct vnode *,
struct componentname *);
static int nfsspec_access(struct vop_access_args *);
static int nfs_readlink(struct vop_readlink_args *);
static int nfs_print(struct vop_print_args *);
static int nfs_advlock(struct vop_advlock_args *);
1994-05-24 10:09:53 +00:00
/*
* Global vfs data structures for nfs
*/
vop_t **nfsv2_vnodeop_p;
static struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
{ &vop_default_desc, (vop_t *) vop_defaultop },
{ &vop_access_desc, (vop_t *) nfs_access },
{ &vop_advlock_desc, (vop_t *) nfs_advlock },
{ &vop_close_desc, (vop_t *) nfs_close },
{ &vop_create_desc, (vop_t *) nfs_create },
{ &vop_fsync_desc, (vop_t *) nfs_fsync },
{ &vop_getattr_desc, (vop_t *) nfs_getattr },
{ &vop_getpages_desc, (vop_t *) nfs_getpages },
{ &vop_putpages_desc, (vop_t *) nfs_putpages },
{ &vop_inactive_desc, (vop_t *) nfs_inactive },
{ &vop_islocked_desc, (vop_t *) vop_stdislocked },
{ &vop_lease_desc, (vop_t *) vop_null },
{ &vop_link_desc, (vop_t *) nfs_link },
{ &vop_lock_desc, (vop_t *) vop_sharedlock },
{ &vop_lookup_desc, (vop_t *) nfs_lookup },
{ &vop_mkdir_desc, (vop_t *) nfs_mkdir },
{ &vop_mknod_desc, (vop_t *) nfs_mknod },
{ &vop_open_desc, (vop_t *) nfs_open },
{ &vop_print_desc, (vop_t *) nfs_print },
{ &vop_read_desc, (vop_t *) nfs_read },
{ &vop_readdir_desc, (vop_t *) nfs_readdir },
{ &vop_readlink_desc, (vop_t *) nfs_readlink },
{ &vop_reclaim_desc, (vop_t *) nfs_reclaim },
{ &vop_remove_desc, (vop_t *) nfs_remove },
{ &vop_rename_desc, (vop_t *) nfs_rename },
{ &vop_rmdir_desc, (vop_t *) nfs_rmdir },
{ &vop_setattr_desc, (vop_t *) nfs_setattr },
{ &vop_strategy_desc, (vop_t *) nfs_strategy },
{ &vop_symlink_desc, (vop_t *) nfs_symlink },
{ &vop_unlock_desc, (vop_t *) vop_stdunlock },
{ &vop_write_desc, (vop_t *) nfs_write },
{ NULL, NULL }
1994-05-24 10:09:53 +00:00
};
static struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
1994-05-24 10:09:53 +00:00
{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
VNODEOP_SET(nfsv2_vnodeop_opv_desc);
1994-05-24 10:09:53 +00:00
/*
* Special device vnode ops
*/
vop_t **spec_nfsv2nodeop_p;
static struct vnodeopv_entry_desc nfsv2_specop_entries[] = {
{ &vop_default_desc, (vop_t *) spec_vnoperate },
{ &vop_access_desc, (vop_t *) nfsspec_access },
{ &vop_close_desc, (vop_t *) nfsspec_close },
{ &vop_fsync_desc, (vop_t *) nfs_fsync },
{ &vop_getattr_desc, (vop_t *) nfs_getattr },
{ &vop_islocked_desc, (vop_t *) vop_stdislocked },
{ &vop_inactive_desc, (vop_t *) nfs_inactive },
{ &vop_lock_desc, (vop_t *) vop_sharedlock },
{ &vop_print_desc, (vop_t *) nfs_print },
{ &vop_read_desc, (vop_t *) nfsspec_read },
{ &vop_reclaim_desc, (vop_t *) nfs_reclaim },
{ &vop_setattr_desc, (vop_t *) nfs_setattr },
{ &vop_unlock_desc, (vop_t *) vop_stdunlock },
{ &vop_write_desc, (vop_t *) nfsspec_write },
{ NULL, NULL }
1994-05-24 10:09:53 +00:00
};
static struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
{ &spec_nfsv2nodeop_p, nfsv2_specop_entries };
VNODEOP_SET(spec_nfsv2nodeop_opv_desc);
1994-05-24 10:09:53 +00:00
vop_t **fifo_nfsv2nodeop_p;
static struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = {
{ &vop_default_desc, (vop_t *) fifo_vnoperate },
{ &vop_access_desc, (vop_t *) nfsspec_access },
{ &vop_close_desc, (vop_t *) nfsfifo_close },
{ &vop_fsync_desc, (vop_t *) nfs_fsync },
{ &vop_getattr_desc, (vop_t *) nfs_getattr },
{ &vop_inactive_desc, (vop_t *) nfs_inactive },
{ &vop_islocked_desc, (vop_t *) vop_stdislocked },
{ &vop_lock_desc, (vop_t *) vop_sharedlock },
{ &vop_print_desc, (vop_t *) nfs_print },
{ &vop_read_desc, (vop_t *) nfsfifo_read },
{ &vop_reclaim_desc, (vop_t *) nfs_reclaim },
{ &vop_setattr_desc, (vop_t *) nfs_setattr },
{ &vop_unlock_desc, (vop_t *) vop_stdunlock },
{ &vop_write_desc, (vop_t *) nfsfifo_write },
{ NULL, NULL }
1994-05-24 10:09:53 +00:00
};
static struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
{ &fifo_nfsv2nodeop_p, nfsv2_fifoop_entries };
VNODEOP_SET(fifo_nfsv2nodeop_opv_desc);
1994-05-24 10:09:53 +00:00
static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
struct componentname *cnp, struct vattr *vap);
static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
struct ucred *cred, struct thread *td);
static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
int fnamelen, struct vnode *tdvp,
const char *tnameptr, int tnamelen,
struct ucred *cred, struct thread *td);
static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
struct sillyrename *sp);
1994-05-24 10:09:53 +00:00
/*
* Global variables
*/
struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
int nfs_numasync = 0;
1994-05-24 10:09:53 +00:00
#define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
SYSCTL_DECL(_vfs_nfs);
static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
&nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
static int nfsv3_commit_on_close = 0;
SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
&nfsv3_commit_on_close, 0, "write+commit on close, else only write");
#if 0
SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
&nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
&nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
#endif
#define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
| NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
| NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
static int
nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
struct ucred *cred)
{
const int v3 = 1;
u_int32_t *tl;
int error = 0, attrflag;
struct mbuf *mreq, *mrep, *md, *mb;
caddr_t bpos, dpos;
u_int32_t rmode;
struct nfsnode *np = VTONFS(vp);
nfsstats.rpccnt[NFSPROC_ACCESS]++;
mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(wmode);
nfsm_request(vp, NFSPROC_ACCESS, td, cred);
nfsm_postop_attr(vp, attrflag);
if (!error) {
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
rmode = fxdr_unsigned(u_int32_t, *tl);
np->n_mode = rmode;
np->n_modeuid = cred->cr_uid;
np->n_modestamp = time_second;
}
m_freem(mrep);
nfsmout:
return error;
}
1994-05-24 10:09:53 +00:00
/*
* nfs access vnode op.
* For nfs version 2, just return ok. File accesses may fail later.
* For nfs version 3, use the access rpc to check accessibility. If file modes
* are changed on the server, accesses might still fail later.
1994-05-24 10:09:53 +00:00
*/
static int
nfs_access(struct vop_access_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
int error = 0;
u_int32_t mode, wmode;
int v3 = NFS_ISV3(vp);
struct nfsnode *np = VTONFS(vp);
1994-05-24 10:09:53 +00:00
/*
* Disallow write attempts on filesystems mounted read-only;
* unless the file is a socket, fifo, or a block or character
* device resident on the filesystem.
*/
if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
switch (vp->v_type) {
case VREG:
case VDIR:
case VLNK:
return (EROFS);
default:
break;
}
}
1994-05-24 10:09:53 +00:00
/*
* For nfs v3, check to see if we have done this recently, and if
* so return our cached result instead of making an ACCESS call.
* If not, do an access rpc, otherwise you are stuck emulating
1994-05-24 10:09:53 +00:00
* ufs_access() locally using the vattr. This may not be correct,
* since the server may apply other access criteria such as
* client uid-->server uid mapping that we do not know about.
1994-05-24 10:09:53 +00:00
*/
if (v3) {
1994-05-24 10:09:53 +00:00
if (ap->a_mode & VREAD)
mode = NFSV3ACCESS_READ;
1994-05-24 10:09:53 +00:00
else
mode = 0;
if (vp->v_type != VDIR) {
if (ap->a_mode & VWRITE)
mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
if (ap->a_mode & VEXEC)
mode |= NFSV3ACCESS_EXECUTE;
} else {
if (ap->a_mode & VWRITE)
mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
NFSV3ACCESS_DELETE);
if (ap->a_mode & VEXEC)
mode |= NFSV3ACCESS_LOOKUP;
}
/* XXX safety belt, only make blanket request if caching */
if (nfsaccess_cache_timeout > 0) {
wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
} else {
wmode = mode;
}
/*
* Does our cached result allow us to give a definite yes to
* this request?
*/
if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
(ap->a_cred->cr_uid == np->n_modeuid) &&
((np->n_mode & mode) == mode)) {
nfsstats.accesscache_hits++;
} else {
/*
* Either a no, or a don't know. Go to the wire.
*/
nfsstats.accesscache_misses++;
error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
if (!error) {
if ((np->n_mode & mode) != mode) {
error = EACCES;
}
}
}
1994-05-24 10:09:53 +00:00
return (error);
} else {
if ((error = nfsspec_access(ap)) != 0)
return (error);
/*
* Attempt to prevent a mapped root from accessing a file
* which it shouldn't. We try to read a byte from the file
* if the user is root and the file is not zero length.
* After calling nfsspec_access, we should have the correct
* file size cached.
*/
if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
&& VTONFS(vp)->n_size > 0) {
struct iovec aiov;
struct uio auio;
char buf[1];
aiov.iov_base = buf;
aiov.iov_len = 1;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_offset = 0;
auio.uio_resid = 1;
auio.uio_segflg = UIO_SYSSPACE;
auio.uio_rw = UIO_READ;
auio.uio_td = ap->a_td;
if (vp->v_type == VREG)
error = nfs_readrpc(vp, &auio, ap->a_cred);
else if (vp->v_type == VDIR) {
char* bp;
bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
aiov.iov_base = bp;
aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
error = nfs_readdirrpc(vp, &auio, ap->a_cred);
free(bp, M_TEMP);
} else if (vp->v_type == VLNK)
error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
else
error = EACCES;
}
return (error);
}
1994-05-24 10:09:53 +00:00
}
/*
* nfs open vnode op
* Check to see if the type is ok
* and that deletion is not in progress.
* For paged in text files, you will need to flush the page cache
* if consistency is lost.
*/
/* ARGSUSED */
static int
nfs_open(struct vop_open_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
1994-05-24 10:09:53 +00:00
struct nfsnode *np = VTONFS(vp);
struct vattr vattr;
int error;
if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
#ifdef DIAGNOSTIC
printf("open eacces vtyp=%d\n", vp->v_type);
#endif
1994-05-24 10:09:53 +00:00
return (EACCES);
}
/*
* Get a valid lease. If cached data is stale, flush it.
*/
if (np->n_flag & NMODIFIED) {
if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
ap->a_td, 1)) == EINTR)
1994-05-24 10:09:53 +00:00
return (error);
np->n_attrstamp = 0;
if (vp->v_type == VDIR)
np->n_direofoffset = 0;
error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
if (error)
return (error);
np->n_mtime = vattr.va_mtime.tv_sec;
} else {
error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
if (error)
return (error);
if (np->n_mtime != vattr.va_mtime.tv_sec) {
if (vp->v_type == VDIR)
np->n_direofoffset = 0;
if ((error = nfs_vinvalbuf(vp, V_SAVE,
ap->a_cred, ap->a_td, 1)) == EINTR)
1994-05-24 10:09:53 +00:00
return (error);
np->n_mtime = vattr.va_mtime.tv_sec;
1994-05-24 10:09:53 +00:00
}
}
np->n_attrstamp = 0; /* For Open/Close consistency */
1994-05-24 10:09:53 +00:00
return (0);
}
/*
* nfs close vnode op
* What an NFS client should do upon close after writing is a debatable issue.
* Most NFS clients push delayed writes to the server upon close, basically for
* two reasons:
* 1 - So that any write errors may be reported back to the client process
* doing the close system call. By far the two most likely errors are
* NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
* 2 - To put a worst case upper bound on cache inconsistency between
* multiple clients for the file.
* There is also a consistency problem for Version 2 of the protocol w.r.t.
* not being able to tell if other clients are writing a file concurrently,
* since there is no way of knowing if the changed modify time in the reply
* is only due to the write for this client.
* (NFS Version 3 provides weak cache consistency data in the reply that
* should be sufficient to detect and handle this case.)
*
* The current code does the following:
* for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
* for NFS Version 3 - flush dirty buffers to the server but don't invalidate
* or commit them (this satisfies 1 and 2 except for the
* case where the server crashes after this close but
* before the commit RPC, which is felt to be "good
* enough". Changing the last argument to nfs_flush() to
* a 1 would force a commit operation, if it is felt a
* commit is necessary now.
1994-05-24 10:09:53 +00:00
*/
/* ARGSUSED */
static int
nfs_close(struct vop_close_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
1994-05-24 10:09:53 +00:00
int error = 0;
if (vp->v_type == VREG) {
if (np->n_flag & NMODIFIED) {
if (NFS_ISV3(vp)) {
/*
* Under NFSv3 we have dirty buffers to dispose of. We
* must flush them to the NFS server. We have the option
* of waiting all the way through the commit rpc or just
* waiting for the initial write. The default is to only
* wait through the initial write so the data is in the
* server's cache, which is roughly similar to the state
* a standard disk subsystem leaves the file in on close().
*
* We cannot clear the NMODIFIED bit in np->n_flag due to
* potential races with other processes, and certainly
* cannot clear it if we don't commit.
*/
int cm = nfsv3_commit_on_close ? 1 : 0;
error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_td, cm);
/* np->n_flag &= ~NMODIFIED; */
} else {
error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_td, 1);
}
1994-05-24 10:09:53 +00:00
np->n_attrstamp = 0;
}
if (np->n_flag & NWRITEERR) {
np->n_flag &= ~NWRITEERR;
error = np->n_error;
}
}
return (error);
}
/*
* nfs getattr call from vfs.
*/
static int
nfs_getattr(struct vop_getattr_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
1994-05-24 10:09:53 +00:00
caddr_t bpos, dpos;
int error = 0;
struct mbuf *mreq, *mrep, *md, *mb;
int v3 = NFS_ISV3(vp);
1994-05-24 10:09:53 +00:00
/*
* Update local times for special files.
*/
if (np->n_flag & (NACC | NUPD))
np->n_flag |= NCHG;
/*
* First look in the cache.
*/
if (nfs_getattrcache(vp, ap->a_vap) == 0)
return (0);
if (v3 && nfsaccess_cache_timeout > 0) {
nfsstats.accesscache_misses++;
nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
if (nfs_getattrcache(vp, ap->a_vap) == 0)
return (0);
}
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_GETATTR]++;
mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred);
if (!error) {
nfsm_loadattr(vp, ap->a_vap);
}
m_freem(mrep);
nfsmout:
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs setattr call.
*/
static int
nfs_setattr(struct vop_setattr_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
struct vattr *vap = ap->a_vap;
int error = 0;
u_quad_t tsize;
1994-05-24 10:09:53 +00:00
#ifndef nolint
tsize = (u_quad_t)0;
#endif
/*
* Setting of flags is not supported.
*/
if (vap->va_flags != VNOVAL)
return (EOPNOTSUPP);
/*
* Disallow write attempts if the filesystem is mounted read-only.
*/
if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
(vp->v_mount->mnt_flag & MNT_RDONLY))
return (EROFS);
if (vap->va_size != VNOVAL) {
switch (vp->v_type) {
case VDIR:
return (EISDIR);
case VCHR:
case VBLK:
case VSOCK:
case VFIFO:
if (vap->va_mtime.tv_sec == VNOVAL &&
vap->va_atime.tv_sec == VNOVAL &&
vap->va_mode == (mode_t)VNOVAL &&
vap->va_uid == (uid_t)VNOVAL &&
vap->va_gid == (gid_t)VNOVAL)
return (0);
vap->va_size = VNOVAL;
break;
default:
/*
* Disallow write attempts if the filesystem is
* mounted read-only.
*/
if (vp->v_mount->mnt_flag & MNT_RDONLY)
return (EROFS);
This fixes a large number of bugs in our NFS client side code. A recent commit by Kirk also fixed a softupdates bug that could easily be triggered by server side NFS. * An edge case with shared R+W mmap()'s and truncate whereby the system would inappropriately clear the dirty bits on still-dirty data. (applicable to all filesystems) THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING. see vm/vm_page.c line 1641 * The straddle case for VM pages and buffer cache buffers when truncating. (applicable to NFS client side) * Possible SMP database corruption due to vm_pager_unmap_page() not clearing the TLB for the other cpu's. (applicable to NFS client side but could effect all filesystems). Note: not considered serious since the corruption occurs beyond the file EOF. * When flusing a dirty buffer due to B_CACHE getting cleared, we were accidently setting B_CACHE again (that is, bwrite() sets B_CACHE), when we really want it to stay clear after the write is complete. This resulted in a corrupt buffer. (applicable to all filesystems but probably only triggered by NFS) * We have to call vtruncbuf() when ftruncate()ing to remove any buffer cache buffers. This is still tentitive, I may be able to remove it due to the second bug fix. (applicable to NFS client side) * vnode_pager_setsize() race against nfs_vinvalbuf()... we have to set n_size before calling nfs_vinvalbuf or the NFS code may recursively vnode_pager_setsize() to the original value before the truncate. This is what was causing the user mmap bus faults in the nfs tester program. (applicable to NFS client side) * Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made by Kirk). Testing program written by: Avadis Tevanian, Jr. Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step') MFC after: 1 week
2001-12-14 01:16:57 +00:00
/*
* We run vnode_pager_setsize() early (why?),
* we must set np->n_size now to avoid vinvalbuf
* V_SAVE races that might setsize a lower
* value.
*/
tsize = np->n_size;
error = nfs_meta_setsize(vp, ap->a_cred,
ap->a_td, vap->va_size);
if (np->n_flag & NMODIFIED) {
if (vap->va_size == 0)
error = nfs_vinvalbuf(vp, 0,
ap->a_cred, ap->a_td, 1);
else
error = nfs_vinvalbuf(vp, V_SAVE,
ap->a_cred, ap->a_td, 1);
if (error) {
vnode_pager_setsize(vp, np->n_size);
return (error);
}
}
This fixes a large number of bugs in our NFS client side code. A recent commit by Kirk also fixed a softupdates bug that could easily be triggered by server side NFS. * An edge case with shared R+W mmap()'s and truncate whereby the system would inappropriately clear the dirty bits on still-dirty data. (applicable to all filesystems) THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING. see vm/vm_page.c line 1641 * The straddle case for VM pages and buffer cache buffers when truncating. (applicable to NFS client side) * Possible SMP database corruption due to vm_pager_unmap_page() not clearing the TLB for the other cpu's. (applicable to NFS client side but could effect all filesystems). Note: not considered serious since the corruption occurs beyond the file EOF. * When flusing a dirty buffer due to B_CACHE getting cleared, we were accidently setting B_CACHE again (that is, bwrite() sets B_CACHE), when we really want it to stay clear after the write is complete. This resulted in a corrupt buffer. (applicable to all filesystems but probably only triggered by NFS) * We have to call vtruncbuf() when ftruncate()ing to remove any buffer cache buffers. This is still tentitive, I may be able to remove it due to the second bug fix. (applicable to NFS client side) * vnode_pager_setsize() race against nfs_vinvalbuf()... we have to set n_size before calling nfs_vinvalbuf or the NFS code may recursively vnode_pager_setsize() to the original value before the truncate. This is what was causing the user mmap bus faults in the nfs tester program. (applicable to NFS client side) * Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made by Kirk). Testing program written by: Avadis Tevanian, Jr. Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step') MFC after: 1 week
2001-12-14 01:16:57 +00:00
np->n_vattr.va_size = vap->va_size;
};
} else if ((vap->va_mtime.tv_sec != VNOVAL ||
vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
vp->v_type == VREG &&
(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
ap->a_td, 1)) == EINTR)
return (error);
error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
if (error && vap->va_size != VNOVAL) {
np->n_size = np->n_vattr.va_size = tsize;
vnode_pager_setsize(vp, np->n_size);
1994-05-24 10:09:53 +00:00
}
return (error);
}
/*
* Do an nfs setattr rpc.
*/
static int
nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
struct thread *td)
{
struct nfsv2_sattr *sp;
caddr_t bpos, dpos;
u_int32_t *tl;
int error = 0, wccflag = NFSV3_WCCRATTR;
struct mbuf *mreq, *mrep, *md, *mb;
int v3 = NFS_ISV3(vp);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_SETATTR]++;
mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
if (v3) {
nfsm_v3attrbuild(vap, TRUE);
tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
*tl = nfs_false;
1994-05-24 10:09:53 +00:00
} else {
sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
if (vap->va_mode == (mode_t)VNOVAL)
sp->sa_mode = nfs_xdrneg1;
else
sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
if (vap->va_uid == (uid_t)VNOVAL)
sp->sa_uid = nfs_xdrneg1;
else
sp->sa_uid = txdr_unsigned(vap->va_uid);
if (vap->va_gid == (gid_t)VNOVAL)
sp->sa_gid = nfs_xdrneg1;
else
sp->sa_gid = txdr_unsigned(vap->va_gid);
sp->sa_size = txdr_unsigned(vap->va_size);
txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1994-05-24 10:09:53 +00:00
}
nfsm_request(vp, NFSPROC_SETATTR, td, cred);
if (v3) {
nfsm_wcc_data(vp, wccflag);
} else
nfsm_loadattr(vp, (struct vattr *)0);
m_freem(mrep);
nfsmout:
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs lookup call, one step at a time...
* First look in cache
* If not found, unlock the directory nfsnode and do the rpc
1994-05-24 10:09:53 +00:00
*/
static int
nfs_lookup(struct vop_lookup_args *ap)
1994-05-24 10:09:53 +00:00
{
struct componentname *cnp = ap->a_cnp;
struct vnode *dvp = ap->a_dvp;
struct vnode **vpp = ap->a_vpp;
int flags = cnp->cn_flags;
struct vnode *newvp;
1994-05-24 10:09:53 +00:00
struct nfsmount *nmp;
caddr_t bpos, dpos;
struct mbuf *mreq, *mrep, *md, *mb;
1994-05-24 10:09:53 +00:00
long len;
nfsfh_t *fhp;
1994-05-24 10:09:53 +00:00
struct nfsnode *np;
int lockparent, wantparent, error = 0, attrflag, fhsize;
int v3 = NFS_ISV3(dvp);
struct thread *td = cnp->cn_thread;
1994-05-24 10:09:53 +00:00
*vpp = NULLVP;
cnp->cn_flags &= ~PDIRUNLOCK;
if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
(cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
return (EROFS);
if (dvp->v_type != VDIR)
return (ENOTDIR);
1994-05-24 10:09:53 +00:00
lockparent = flags & LOCKPARENT;
wantparent = flags & (LOCKPARENT|WANTPARENT);
nmp = VFSTONFS(dvp->v_mount);
np = VTONFS(dvp);
if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
struct vattr vattr;
int vpid;
if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
*vpp = NULLVP;
return (error);
}
vhold(*vpp);
newvp = *vpp;
vpid = newvp->v_id;
/*
* See the comment starting `Step through' in ufs/ufs_lookup.c
* for an explanation of the locking protocol
*/
if (dvp == newvp) {
VREF(newvp);
error = 0;
} else if (flags & ISDOTDOT) {
VOP_UNLOCK(dvp, 0, td);
cnp->cn_flags |= PDIRUNLOCK;
error = vget(newvp, LK_EXCLUSIVE, td);
if (!error && lockparent && (flags & ISLASTCN)) {
error = vn_lock(dvp, LK_EXCLUSIVE, td);
if (error == 0)
cnp->cn_flags &= ~PDIRUNLOCK;
}
} else {
error = vget(newvp, LK_EXCLUSIVE, td);
if (!lockparent || error || !(flags & ISLASTCN)) {
VOP_UNLOCK(dvp, 0, td);
cnp->cn_flags |= PDIRUNLOCK;
}
}
if (!error) {
if (vpid == newvp->v_id) {
if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td)
&& vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
nfsstats.lookupcache_hits++;
if (cnp->cn_nameiop != LOOKUP &&
(flags & ISLASTCN))
cnp->cn_flags |= SAVENAME;
vdrop(newvp);
return (0);
}
cache_purge(newvp);
}
vput(newvp);
if (lockparent && dvp != newvp && (flags & ISLASTCN))
VOP_UNLOCK(dvp, 0, td);
}
vdrop(newvp);
error = vn_lock(dvp, LK_EXCLUSIVE, td);
*vpp = NULLVP;
if (error) {
cnp->cn_flags |= PDIRUNLOCK;
return (error);
}
cnp->cn_flags &= ~PDIRUNLOCK;
}
1994-05-24 10:09:53 +00:00
error = 0;
newvp = NULLVP;
1994-05-24 10:09:53 +00:00
nfsstats.lookupcache_misses++;
nfsstats.rpccnt[NFSPROC_LOOKUP]++;
len = cnp->cn_namelen;
mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
1994-05-24 10:09:53 +00:00
nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
1994-05-24 10:09:53 +00:00
if (error) {
nfsm_postop_attr(dvp, attrflag);
m_freem(mrep);
goto nfsmout;
1994-05-24 10:09:53 +00:00
}
nfsm_getfh(fhp, fhsize, v3);
1994-05-24 10:09:53 +00:00
/*
* Handle RENAME case...
*/
if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
if (NFS_CMPFH(np, fhp, fhsize)) {
1994-05-24 10:09:53 +00:00
m_freem(mrep);
return (EISDIR);
}
error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
if (error) {
1994-05-24 10:09:53 +00:00
m_freem(mrep);
return (error);
}
newvp = NFSTOV(np);
if (v3) {
nfsm_postop_attr(newvp, attrflag);
nfsm_postop_attr(dvp, attrflag);
} else
nfsm_loadattr(newvp, (struct vattr *)0);
1994-05-24 10:09:53 +00:00
*vpp = newvp;
m_freem(mrep);
cnp->cn_flags |= SAVENAME;
if (!lockparent) {
VOP_UNLOCK(dvp, 0, td);
cnp->cn_flags |= PDIRUNLOCK;
}
1994-05-24 10:09:53 +00:00
return (0);
}
if (flags & ISDOTDOT) {
VOP_UNLOCK(dvp, 0, td);
error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
if (error) {
vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
return (error);
}
newvp = NFSTOV(np);
if (lockparent && (flags & ISLASTCN)) {
error = vn_lock(dvp, LK_EXCLUSIVE, td);
if (error) {
cnp->cn_flags |= PDIRUNLOCK;
vput(newvp);
return (error);
}
} else
cnp->cn_flags |= PDIRUNLOCK;
} else if (NFS_CMPFH(np, fhp, fhsize)) {
1994-05-24 10:09:53 +00:00
VREF(dvp);
newvp = dvp;
} else {
error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
if (error) {
1994-05-24 10:09:53 +00:00
m_freem(mrep);
return (error);
}
if (!lockparent || !(flags & ISLASTCN)) {
cnp->cn_flags |= PDIRUNLOCK;
VOP_UNLOCK(dvp, 0, td);
}
1994-05-24 10:09:53 +00:00
newvp = NFSTOV(np);
}
if (v3) {
nfsm_postop_attr(newvp, attrflag);
nfsm_postop_attr(dvp, attrflag);
} else
nfsm_loadattr(newvp, (struct vattr *)0);
1994-05-24 10:09:53 +00:00
if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
cnp->cn_flags |= SAVENAME;
if ((cnp->cn_flags & MAKEENTRY) &&
(cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
np->n_ctime = np->n_vattr.va_ctime.tv_sec;
cache_enter(dvp, newvp, cnp);
1994-05-24 10:09:53 +00:00
}
*vpp = newvp;
m_freem(mrep);
nfsmout:
if (error) {
if (newvp != NULLVP) {
vrele(newvp);
*vpp = NULLVP;
}
if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
(flags & ISLASTCN) && error == ENOENT) {
if (!lockparent) {
VOP_UNLOCK(dvp, 0, td);
cnp->cn_flags |= PDIRUNLOCK;
}
if (dvp->v_mount->mnt_flag & MNT_RDONLY)
error = EROFS;
else
error = EJUSTRETURN;
}
if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
cnp->cn_flags |= SAVENAME;
}
return (error);
1994-05-24 10:09:53 +00:00
}
/*
* nfs read call.
* Just call nfs_bioread() to do the work.
*/
static int
nfs_read(struct vop_read_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
1994-05-24 10:09:53 +00:00
if (vp->v_type != VREG)
return (EPERM);
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1994-05-24 10:09:53 +00:00
}
/*
* nfs readlink call
*/
static int
nfs_readlink(struct vop_readlink_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
1994-05-24 10:09:53 +00:00
if (vp->v_type != VLNK)
return (EINVAL);
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1994-05-24 10:09:53 +00:00
}
/*
* Do a readlink rpc.
* Called by nfs_doio() from below the buffer cache.
*/
int
nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1994-05-24 10:09:53 +00:00
{
caddr_t bpos, dpos;
int error = 0, len, attrflag;
struct mbuf *mreq, *mrep, *md, *mb;
int v3 = NFS_ISV3(vp);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_READLINK]++;
mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
if (v3)
nfsm_postop_attr(vp, attrflag);
if (!error) {
nfsm_strsiz(len, NFS_MAXPATHLEN);
if (len == NFS_MAXPATHLEN) {
struct nfsnode *np = VTONFS(vp);
if (np->n_size && np->n_size < NFS_MAXPATHLEN)
len = np->n_size;
}
nfsm_mtouio(uiop, len);
}
m_freem(mrep);
nfsmout:
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs read rpc call
* Ditto above
*/
int
nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1994-05-24 10:09:53 +00:00
{
u_int32_t *tl;
caddr_t bpos, dpos;
struct mbuf *mreq, *mrep, *md, *mb;
1994-05-24 10:09:53 +00:00
struct nfsmount *nmp;
int error = 0, len, retlen, tsiz, eof, attrflag;
int v3 = NFS_ISV3(vp);
1994-05-24 10:09:53 +00:00
#ifndef nolint
eof = 0;
#endif
1994-05-24 10:09:53 +00:00
nmp = VFSTONFS(vp->v_mount);
tsiz = uiop->uio_resid;
if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1994-05-24 10:09:53 +00:00
return (EFBIG);
while (tsiz > 0) {
nfsstats.rpccnt[NFSPROC_READ]++;
len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
if (v3) {
txdr_hyper(uiop->uio_offset, tl);
1994-05-24 10:09:53 +00:00
*(tl + 2) = txdr_unsigned(len);
} else {
*tl++ = txdr_unsigned(uiop->uio_offset);
*tl++ = txdr_unsigned(len);
*tl = 0;
}
nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
if (v3) {
nfsm_postop_attr(vp, attrflag);
if (error) {
m_freem(mrep);
goto nfsmout;
}
tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
eof = fxdr_unsigned(int, *(tl + 1));
} else
nfsm_loadattr(vp, (struct vattr *)0);
1994-05-24 10:09:53 +00:00
nfsm_strsiz(retlen, nmp->nm_rsize);
nfsm_mtouio(uiop, retlen);
m_freem(mrep);
tsiz -= retlen;
if (v3) {
This fixes a large number of bugs in our NFS client side code. A recent commit by Kirk also fixed a softupdates bug that could easily be triggered by server side NFS. * An edge case with shared R+W mmap()'s and truncate whereby the system would inappropriately clear the dirty bits on still-dirty data. (applicable to all filesystems) THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING. see vm/vm_page.c line 1641 * The straddle case for VM pages and buffer cache buffers when truncating. (applicable to NFS client side) * Possible SMP database corruption due to vm_pager_unmap_page() not clearing the TLB for the other cpu's. (applicable to NFS client side but could effect all filesystems). Note: not considered serious since the corruption occurs beyond the file EOF. * When flusing a dirty buffer due to B_CACHE getting cleared, we were accidently setting B_CACHE again (that is, bwrite() sets B_CACHE), when we really want it to stay clear after the write is complete. This resulted in a corrupt buffer. (applicable to all filesystems but probably only triggered by NFS) * We have to call vtruncbuf() when ftruncate()ing to remove any buffer cache buffers. This is still tentitive, I may be able to remove it due to the second bug fix. (applicable to NFS client side) * vnode_pager_setsize() race against nfs_vinvalbuf()... we have to set n_size before calling nfs_vinvalbuf or the NFS code may recursively vnode_pager_setsize() to the original value before the truncate. This is what was causing the user mmap bus faults in the nfs tester program. (applicable to NFS client side) * Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made by Kirk). Testing program written by: Avadis Tevanian, Jr. Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step') MFC after: 1 week
2001-12-14 01:16:57 +00:00
if (eof || retlen == 0) {
tsiz = 0;
This fixes a large number of bugs in our NFS client side code. A recent commit by Kirk also fixed a softupdates bug that could easily be triggered by server side NFS. * An edge case with shared R+W mmap()'s and truncate whereby the system would inappropriately clear the dirty bits on still-dirty data. (applicable to all filesystems) THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING. see vm/vm_page.c line 1641 * The straddle case for VM pages and buffer cache buffers when truncating. (applicable to NFS client side) * Possible SMP database corruption due to vm_pager_unmap_page() not clearing the TLB for the other cpu's. (applicable to NFS client side but could effect all filesystems). Note: not considered serious since the corruption occurs beyond the file EOF. * When flusing a dirty buffer due to B_CACHE getting cleared, we were accidently setting B_CACHE again (that is, bwrite() sets B_CACHE), when we really want it to stay clear after the write is complete. This resulted in a corrupt buffer. (applicable to all filesystems but probably only triggered by NFS) * We have to call vtruncbuf() when ftruncate()ing to remove any buffer cache buffers. This is still tentitive, I may be able to remove it due to the second bug fix. (applicable to NFS client side) * vnode_pager_setsize() race against nfs_vinvalbuf()... we have to set n_size before calling nfs_vinvalbuf or the NFS code may recursively vnode_pager_setsize() to the original value before the truncate. This is what was causing the user mmap bus faults in the nfs tester program. (applicable to NFS client side) * Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made by Kirk). Testing program written by: Avadis Tevanian, Jr. Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step') MFC after: 1 week
2001-12-14 01:16:57 +00:00
}
} else if (retlen < len) {
1994-05-24 10:09:53 +00:00
tsiz = 0;
This fixes a large number of bugs in our NFS client side code. A recent commit by Kirk also fixed a softupdates bug that could easily be triggered by server side NFS. * An edge case with shared R+W mmap()'s and truncate whereby the system would inappropriately clear the dirty bits on still-dirty data. (applicable to all filesystems) THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING. see vm/vm_page.c line 1641 * The straddle case for VM pages and buffer cache buffers when truncating. (applicable to NFS client side) * Possible SMP database corruption due to vm_pager_unmap_page() not clearing the TLB for the other cpu's. (applicable to NFS client side but could effect all filesystems). Note: not considered serious since the corruption occurs beyond the file EOF. * When flusing a dirty buffer due to B_CACHE getting cleared, we were accidently setting B_CACHE again (that is, bwrite() sets B_CACHE), when we really want it to stay clear after the write is complete. This resulted in a corrupt buffer. (applicable to all filesystems but probably only triggered by NFS) * We have to call vtruncbuf() when ftruncate()ing to remove any buffer cache buffers. This is still tentitive, I may be able to remove it due to the second bug fix. (applicable to NFS client side) * vnode_pager_setsize() race against nfs_vinvalbuf()... we have to set n_size before calling nfs_vinvalbuf or the NFS code may recursively vnode_pager_setsize() to the original value before the truncate. This is what was causing the user mmap bus faults in the nfs tester program. (applicable to NFS client side) * Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made by Kirk). Testing program written by: Avadis Tevanian, Jr. Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step') MFC after: 1 week
2001-12-14 01:16:57 +00:00
}
1994-05-24 10:09:53 +00:00
}
nfsmout:
return (error);
}
/*
* nfs write call
*/
int
nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
int *iomode, int *must_commit)
1994-05-24 10:09:53 +00:00
{
u_int32_t *tl;
int32_t backup;
caddr_t bpos, dpos;
struct mbuf *mreq, *mrep, *md, *mb;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1994-05-24 10:09:53 +00:00
#ifndef DIAGNOSTIC
if (uiop->uio_iovcnt != 1)
panic("nfs: writerpc iovcnt > 1");
#endif
*must_commit = 0;
1994-05-24 10:09:53 +00:00
tsiz = uiop->uio_resid;
if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1994-05-24 10:09:53 +00:00
return (EFBIG);
while (tsiz > 0) {
nfsstats.rpccnt[NFSPROC_WRITE]++;
len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
if (v3) {
tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
txdr_hyper(uiop->uio_offset, tl);
1994-05-24 10:09:53 +00:00
tl += 2;
*tl++ = txdr_unsigned(len);
*tl++ = txdr_unsigned(*iomode);
*tl = txdr_unsigned(len);
1994-05-24 10:09:53 +00:00
} else {
u_int32_t x;
tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
/* Set both "begin" and "current" to non-garbage. */
x = txdr_unsigned((u_int32_t)uiop->uio_offset);
*tl++ = x; /* "begin offset" */
*tl++ = x; /* "current offset" */
x = txdr_unsigned(len);
*tl++ = x; /* total to this offset */
*tl = x; /* size of this write */
1994-05-24 10:09:53 +00:00
}
nfsm_uiotom(uiop, len);
nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
if (v3) {
wccflag = NFSV3_WCCCHK;
nfsm_wcc_data(vp, wccflag);
if (!error) {
tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
+ NFSX_V3WRITEVERF);
rlen = fxdr_unsigned(int, *tl++);
if (rlen == 0) {
error = NFSERR_IO;
m_freem(mrep);
break;
} else if (rlen < len) {
backup = len - rlen;
uiop->uio_iov->iov_base -= backup;
uiop->uio_iov->iov_len += backup;
uiop->uio_offset -= backup;
uiop->uio_resid += backup;
len = rlen;
}
commit = fxdr_unsigned(int, *tl++);
/*
* Return the lowest committment level
* obtained by any of the RPCs.
*/
if (committed == NFSV3WRITE_FILESYNC)
committed = commit;
else if (committed == NFSV3WRITE_DATASYNC &&
commit == NFSV3WRITE_UNSTABLE)
committed = commit;
if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
NFSX_V3WRITEVERF);
nmp->nm_state |= NFSSTA_HASWRITEVERF;
} else if (bcmp((caddr_t)tl,
(caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
*must_commit = 1;
bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
NFSX_V3WRITEVERF);
}
}
} else
nfsm_loadattr(vp, (struct vattr *)0);
if (wccflag)
VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec;
1994-05-24 10:09:53 +00:00
m_freem(mrep);
if (error)
break;
1994-05-24 10:09:53 +00:00
tsiz -= len;
}
nfsmout:
if (vp->v_mount->mnt_flag & MNT_ASYNC)
committed = NFSV3WRITE_FILESYNC;
*iomode = committed;
1994-05-24 10:09:53 +00:00
if (error)
uiop->uio_resid = tsiz;
return (error);
}
/*
* nfs mknod rpc
* For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
* mode set to specify the file type and the size field for rdev.
1994-05-24 10:09:53 +00:00
*/
static int
nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
struct vattr *vap)
1994-05-24 10:09:53 +00:00
{
struct nfsv2_sattr *sp;
u_int32_t *tl;
struct vnode *newvp = (struct vnode *)0;
struct nfsnode *np = (struct nfsnode *)0;
1994-05-24 10:09:53 +00:00
struct vattr vattr;
caddr_t bpos, dpos;
int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
struct mbuf *mreq, *mrep, *md, *mb;
u_int32_t rdev;
int v3 = NFS_ISV3(dvp);
1994-05-24 10:09:53 +00:00
if (vap->va_type == VCHR || vap->va_type == VBLK)
rdev = txdr_unsigned(vap->va_rdev);
else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
rdev = nfs_xdrneg1;
1994-05-24 10:09:53 +00:00
else {
return (EOPNOTSUPP);
}
if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1994-05-24 10:09:53 +00:00
return (error);
}
nfsstats.rpccnt[NFSPROC_MKNOD]++;
mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
1994-05-24 10:09:53 +00:00
nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
if (v3) {
tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
*tl++ = vtonfsv3_type(vap->va_type);
nfsm_v3attrbuild(vap, FALSE);
if (vap->va_type == VCHR || vap->va_type == VBLK) {
tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
Divorce "dev_t" from the "major|minor" bitmap, which is now called udev_t in the kernel but still called dev_t in userland. Provide functions to manipulate both types: major() umajor() minor() uminor() makedev() umakedev() dev2udev() udev2dev() For now they're functions, they will become in-line functions after one of the next two steps in this process. Return major/minor/makedev to macro-hood for userland. Register a name in cdevsw[] for the "filedescriptor" driver. In the kernel the udev_t appears in places where we have the major/minor number combination, (ie: a potential device: we may not have the driver nor the device), like in inodes, vattr, cdevsw registration and so on, whereas the dev_t appears where we carry around a reference to a actual device. In the future the cdevsw and the aliased-from vnode will be hung directly from the dev_t, along with up to two softc pointers for the device driver and a few houskeeping bits. This will essentially replace the current "alias" check code (same buck, bigger bang). A little stunt has been provided to try to catch places where the wrong type is being used (dev_t vs udev_t), if you see something not working, #undef DEVT_FASCIST in kern/kern_conf.c and see if it makes a difference. If it does, please try to track it down (many hands make light work) or at least try to reproduce it as simply as possible, and describe how to do that. Without DEVT_FASCIST I belive this patch is a no-op. Stylistic/posixoid comments about the userland view of the <sys/*.h> files welcome now, from userland they now contain the end result. Next planned step: make all dev_t's refer to the same devsw[] which means convert BLK's to CHR's at the perimeter of the vnodes and other places where they enter the game (bootdev, mknod, sysctl).
1999-05-11 19:55:07 +00:00
*tl++ = txdr_unsigned(umajor(vap->va_rdev));
*tl = txdr_unsigned(uminor(vap->va_rdev));
}
1994-05-24 10:09:53 +00:00
} else {
sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
sp->sa_uid = nfs_xdrneg1;
sp->sa_gid = nfs_xdrneg1;
sp->sa_size = rdev;
txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1994-05-24 10:09:53 +00:00
}
nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
if (!error) {
nfsm_mtofh(dvp, newvp, v3, gotvp);
if (!gotvp) {
if (newvp) {
vput(newvp);
newvp = (struct vnode *)0;
}
error = nfs_lookitup(dvp, cnp->cn_nameptr,
cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
if (!error)
newvp = NFSTOV(np);
}
}
if (v3)
nfsm_wcc_data(dvp, wccflag);
m_freem(mrep);
nfsmout:
if (error) {
if (newvp)
vput(newvp);
} else {
if (cnp->cn_flags & MAKEENTRY)
cache_enter(dvp, newvp, cnp);
*vpp = newvp;
}
1994-05-24 10:09:53 +00:00
VTONFS(dvp)->n_flag |= NMODIFIED;
if (!wccflag)
VTONFS(dvp)->n_attrstamp = 0;
return (error);
}
/*
* nfs mknod vop
* just call nfs_mknodrpc() to do the work.
*/
/* ARGSUSED */
static int
nfs_mknod(struct vop_mknod_args *ap)
{
return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1994-05-24 10:09:53 +00:00
}
static u_long create_verf;
1994-05-24 10:09:53 +00:00
/*
* nfs file create call
*/
static int
nfs_create(struct vop_create_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *dvp = ap->a_dvp;
struct vattr *vap = ap->a_vap;
struct componentname *cnp = ap->a_cnp;
struct nfsv2_sattr *sp;
u_int32_t *tl;
struct nfsnode *np = (struct nfsnode *)0;
struct vnode *newvp = (struct vnode *)0;
caddr_t bpos, dpos;
int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
struct mbuf *mreq, *mrep, *md, *mb;
1994-05-24 10:09:53 +00:00
struct vattr vattr;
int v3 = NFS_ISV3(dvp);
1994-05-24 10:09:53 +00:00
/*
* Oops, not for me..
*/
if (vap->va_type == VSOCK)
return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1994-05-24 10:09:53 +00:00
return (error);
}
if (vap->va_vaflags & VA_EXCLUSIVE)
fmode |= O_EXCL;
again:
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_CREATE]++;
mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
1994-05-24 10:09:53 +00:00
nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
if (v3) {
tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
if (fmode & O_EXCL) {
*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
#ifdef INET
if (!TAILQ_EMPTY(&in_ifaddrhead))
*tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
else
#endif
*tl++ = create_verf;
*tl = ++create_verf;
} else {
*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
nfsm_v3attrbuild(vap, FALSE);
}
1994-05-24 10:09:53 +00:00
} else {
sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
sp->sa_uid = nfs_xdrneg1;
sp->sa_gid = nfs_xdrneg1;
sp->sa_size = 0;
txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1994-05-24 10:09:53 +00:00
}
nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
if (!error) {
nfsm_mtofh(dvp, newvp, v3, gotvp);
if (!gotvp) {
if (newvp) {
vput(newvp);
newvp = (struct vnode *)0;
}
error = nfs_lookitup(dvp, cnp->cn_nameptr,
cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
if (!error)
newvp = NFSTOV(np);
}
}
if (v3)
nfsm_wcc_data(dvp, wccflag);
m_freem(mrep);
nfsmout:
if (error) {
if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
fmode &= ~O_EXCL;
goto again;
}
if (newvp)
vput(newvp);
} else if (v3 && (fmode & O_EXCL)) {
/*
* 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;
error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread);
}
if (!error) {
if (cnp->cn_flags & MAKEENTRY)
cache_enter(dvp, newvp, cnp);
*ap->a_vpp = newvp;
}
1994-05-24 10:09:53 +00:00
VTONFS(dvp)->n_flag |= NMODIFIED;
if (!wccflag)
VTONFS(dvp)->n_attrstamp = 0;
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs file remove call
* To try and make nfs semantics closer to ufs semantics, a file that has
* other processes using the vnode is renamed instead of removed and then
* removed later on the last close.
* - If v_usecount > 1
* If a rename is not already in the works
* call nfs_sillyrename() to set it up
* else
* do the remove rpc
*/
static int
nfs_remove(struct vop_remove_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct vnode *dvp = ap->a_dvp;
struct componentname *cnp = ap->a_cnp;
struct nfsnode *np = VTONFS(vp);
1994-05-24 10:09:53 +00:00
int error = 0;
struct vattr vattr;
1994-05-24 10:09:53 +00:00
#ifndef DIAGNOSTIC
if ((cnp->cn_flags & HASBUF) == 0)
panic("nfs_remove: no name");
if (vp->v_usecount < 1)
panic("nfs_remove: bad v_usecount");
#endif
if (vp->v_type == VDIR)
error = EPERM;
else if (vp->v_usecount == 1 || (np->n_sillyrename &&
VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
vattr.va_nlink > 1)) {
1994-05-24 10:09:53 +00:00
/*
* Purge the name cache so that the chance of a lookup for
* the name succeeding while the remove is in progress is
* minimized. Without node locking it can still happen, such
* that an I/O op returns ESTALE, but since you get this if
* another host removes the file..
*/
cache_purge(vp);
/*
* throw away biocache buffers, mainly to avoid
* unnecessary delayed writes later.
1994-05-24 10:09:53 +00:00
*/
error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_thread, 1);
1994-05-24 10:09:53 +00:00
/* Do the rpc */
if (error != EINTR)
error = nfs_removerpc(dvp, cnp->cn_nameptr,
cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1994-05-24 10:09:53 +00:00
/*
* Kludge City: If the first reply to the remove rpc is lost..
* the reply to the retransmitted request will be ENOENT
* since the file was in fact removed
* Therefore, we cheat and return success.
*/
if (error == ENOENT)
error = 0;
} else if (!np->n_sillyrename)
error = nfs_sillyrename(dvp, vp, cnp);
1994-05-24 10:09:53 +00:00
np->n_attrstamp = 0;
return (error);
}
/*
* nfs file remove rpc called from nfs_inactive
*/
int
nfs_removeit(struct sillyrename *sp)
1994-05-24 10:09:53 +00:00
{
return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
(struct thread *)0));
}
/*
* Nfs remove rpc, called from nfs_remove() and nfs_removeit().
*/
static int
nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
struct ucred *cred, struct thread *td)
{
caddr_t bpos, dpos;
int error = 0, wccflag = NFSV3_WCCRATTR;
struct mbuf *mreq, *mrep, *md, *mb;
int v3 = NFS_ISV3(dvp);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_REMOVE]++;
mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
if (v3)
nfsm_wcc_data(dvp, wccflag);
m_freem(mrep);
nfsmout:
VTONFS(dvp)->n_flag |= NMODIFIED;
if (!wccflag)
VTONFS(dvp)->n_attrstamp = 0;
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs file rename call
*/
static int
nfs_rename(struct vop_rename_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *fvp = ap->a_fvp;
struct vnode *tvp = ap->a_tvp;
struct vnode *fdvp = ap->a_fdvp;
struct vnode *tdvp = ap->a_tdvp;
struct componentname *tcnp = ap->a_tcnp;
struct componentname *fcnp = ap->a_fcnp;
int error;
1994-05-24 10:09:53 +00:00
#ifndef DIAGNOSTIC
if ((tcnp->cn_flags & HASBUF) == 0 ||
(fcnp->cn_flags & HASBUF) == 0)
panic("nfs_rename: no name");
#endif
1994-05-24 10:09:53 +00:00
/* Check for cross-device rename */
if ((fvp->v_mount != tdvp->v_mount) ||
(tvp && (fvp->v_mount != tvp->v_mount))) {
error = EXDEV;
goto out;
}
/*
* We have to flush B_DELWRI data prior to renaming
* the file. If we don't, the delayed-write buffers
* can be flushed out later after the file has gone stale
* under NFSV3. NFSV2 does not have this problem because
* ( as far as I can tell ) it flushes dirty buffers more
* often.
*/
VOP_FSYNC(fvp, fcnp->cn_cred, MNT_WAIT, fcnp->cn_thread);
if (tvp)
VOP_FSYNC(tvp, tcnp->cn_cred, MNT_WAIT, tcnp->cn_thread);
/*
* If the tvp exists and is in use, sillyrename it before doing the
* rename of the new file over it.
* XXX Can't sillyrename a directory.
*/
if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
vput(tvp);
tvp = NULL;
}
error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
tcnp->cn_thread);
1994-05-24 10:09:53 +00:00
if (fvp->v_type == VDIR) {
if (tvp != NULL && tvp->v_type == VDIR)
cache_purge(tdvp);
cache_purge(fdvp);
}
1994-05-24 10:09:53 +00:00
out:
if (tdvp == tvp)
vrele(tdvp);
else
vput(tdvp);
if (tvp)
vput(tvp);
vrele(fdvp);
vrele(fvp);
/*
* Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
*/
if (error == ENOENT)
error = 0;
return (error);
}
/*
* nfs file rename rpc called from nfs_remove() above
*/
static int
nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
struct sillyrename *sp)
{
return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
}
/*
* Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
*/
static int
nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
struct thread *td)
1994-05-24 10:09:53 +00:00
{
caddr_t bpos, dpos;
int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
struct mbuf *mreq, *mrep, *md, *mb;
int v3 = NFS_ISV3(fdvp);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_RENAME]++;
mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
(NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
nfsm_rndup(tnamelen));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(fdvp, v3);
nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
nfsm_fhtom(tdvp, v3);
nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
if (v3) {
nfsm_wcc_data(fdvp, fwccflag);
nfsm_wcc_data(tdvp, twccflag);
}
m_freem(mrep);
nfsmout:
VTONFS(fdvp)->n_flag |= NMODIFIED;
VTONFS(tdvp)->n_flag |= NMODIFIED;
if (!fwccflag)
VTONFS(fdvp)->n_attrstamp = 0;
if (!twccflag)
VTONFS(tdvp)->n_attrstamp = 0;
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs hard link create call
*/
static int
nfs_link(struct vop_link_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct vnode *tdvp = ap->a_tdvp;
struct componentname *cnp = ap->a_cnp;
caddr_t bpos, dpos;
int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
struct mbuf *mreq, *mrep, *md, *mb;
int v3;
1994-05-24 10:09:53 +00:00
if (vp->v_mount != tdvp->v_mount) {
return (EXDEV);
}
/*
* Push all writes to the server, so that the attribute cache
* doesn't get "out of sync" with the server.
* XXX There should be a better way!
*/
VOP_FSYNC(vp, cnp->cn_cred, MNT_WAIT, cnp->cn_thread);
v3 = NFS_ISV3(vp);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_LINK]++;
mreq = nfsm_reqhead(vp, NFSPROC_LINK,
NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
nfsm_fhtom(tdvp, v3);
1994-05-24 10:09:53 +00:00
nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
if (v3) {
nfsm_postop_attr(vp, attrflag);
nfsm_wcc_data(tdvp, wccflag);
}
m_freem(mrep);
nfsmout:
VTONFS(tdvp)->n_flag |= NMODIFIED;
if (!attrflag)
VTONFS(vp)->n_attrstamp = 0;
if (!wccflag)
VTONFS(tdvp)->n_attrstamp = 0;
1994-05-24 10:09:53 +00:00
/*
* Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
*/
if (error == EEXIST)
error = 0;
return (error);
}
/*
* nfs symbolic link create call
*/
static int
nfs_symlink(struct vop_symlink_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *dvp = ap->a_dvp;
struct vattr *vap = ap->a_vap;
struct componentname *cnp = ap->a_cnp;
struct nfsv2_sattr *sp;
caddr_t bpos, dpos;
int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
struct mbuf *mreq, *mrep, *md, *mb;
struct vnode *newvp = (struct vnode *)0;
int v3 = NFS_ISV3(dvp);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_SYMLINK]++;
slen = strlen(ap->a_target);
mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
1994-05-24 10:09:53 +00:00
nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
if (v3) {
nfsm_v3attrbuild(vap, FALSE);
}
1994-05-24 10:09:53 +00:00
nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
if (!v3) {
sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
sp->sa_uid = nfs_xdrneg1;
sp->sa_gid = nfs_xdrneg1;
sp->sa_size = nfs_xdrneg1;
txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1994-05-24 10:09:53 +00:00
}
/*
* Issue the NFS request and get the rpc response.
*
* Only NFSv3 responses returning an error of 0 actually return
* a file handle that can be converted into newvp without having
* to do an extra lookup rpc.
*/
nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
if (v3) {
if (error == 0)
nfsm_mtofh(dvp, newvp, v3, gotvp);
nfsm_wcc_data(dvp, wccflag);
}
/*
* out code jumps -> here, mrep is also freed.
*/
m_freem(mrep);
nfsmout:
1994-05-24 10:09:53 +00:00
/*
* If we get an EEXIST error, silently convert it to no-error
* in case of an NFS retry.
1994-05-24 10:09:53 +00:00
*/
if (error == EEXIST)
error = 0;
/*
* If we do not have (or no longer have) an error, and we could
* not extract the newvp from the response due to the request being
* NFSv2 or the error being EEXIST. We have to do a lookup in order
* to obtain a newvp to return.
*/
if (error == 0 && newvp == NULL) {
struct nfsnode *np = NULL;
error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
cnp->cn_cred, cnp->cn_thread, &np);
if (!error)
newvp = NFSTOV(np);
}
if (error) {
if (newvp)
vput(newvp);
} else {
*ap->a_vpp = newvp;
}
VTONFS(dvp)->n_flag |= NMODIFIED;
if (!wccflag)
VTONFS(dvp)->n_attrstamp = 0;
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs make dir call
*/
static int
nfs_mkdir(struct vop_mkdir_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *dvp = ap->a_dvp;
struct vattr *vap = ap->a_vap;
struct componentname *cnp = ap->a_cnp;
struct nfsv2_sattr *sp;
int len;
struct nfsnode *np = (struct nfsnode *)0;
struct vnode *newvp = (struct vnode *)0;
caddr_t bpos, dpos;
int error = 0, wccflag = NFSV3_WCCRATTR;
int gotvp = 0;
struct mbuf *mreq, *mrep, *md, *mb;
1994-05-24 10:09:53 +00:00
struct vattr vattr;
int v3 = NFS_ISV3(dvp);
1994-05-24 10:09:53 +00:00
if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1994-05-24 10:09:53 +00:00
return (error);
}
len = cnp->cn_namelen;
nfsstats.rpccnt[NFSPROC_MKDIR]++;
mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
1994-05-24 10:09:53 +00:00
nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
if (v3) {
nfsm_v3attrbuild(vap, FALSE);
1994-05-24 10:09:53 +00:00
} else {
sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
sp->sa_uid = nfs_xdrneg1;
sp->sa_gid = nfs_xdrneg1;
sp->sa_size = nfs_xdrneg1;
txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1994-05-24 10:09:53 +00:00
}
nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
if (!error)
nfsm_mtofh(dvp, newvp, v3, gotvp);
if (v3)
nfsm_wcc_data(dvp, wccflag);
m_freem(mrep);
nfsmout:
1994-05-24 10:09:53 +00:00
VTONFS(dvp)->n_flag |= NMODIFIED;
if (!wccflag)
VTONFS(dvp)->n_attrstamp = 0;
1994-05-24 10:09:53 +00:00
/*
* Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
* if we can succeed in looking up the directory.
*/
if (error == EEXIST || (!error && !gotvp)) {
if (newvp) {
vrele(newvp);
newvp = (struct vnode *)0;
}
error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
cnp->cn_thread, &np);
if (!error) {
newvp = NFSTOV(np);
if (newvp->v_type != VDIR)
error = EEXIST;
1994-05-24 10:09:53 +00:00
}
}
if (error) {
if (newvp)
vrele(newvp);
} else
*ap->a_vpp = newvp;
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs remove directory call
*/
static int
nfs_rmdir(struct vop_rmdir_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct vnode *dvp = ap->a_dvp;
struct componentname *cnp = ap->a_cnp;
caddr_t bpos, dpos;
int error = 0, wccflag = NFSV3_WCCRATTR;
struct mbuf *mreq, *mrep, *md, *mb;
int v3 = NFS_ISV3(dvp);
1994-05-24 10:09:53 +00:00
if (dvp == vp)
return (EINVAL);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_RMDIR]++;
mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
1994-05-24 10:09:53 +00:00
nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
if (v3)
nfsm_wcc_data(dvp, wccflag);
m_freem(mrep);
nfsmout:
1994-05-24 10:09:53 +00:00
VTONFS(dvp)->n_flag |= NMODIFIED;
if (!wccflag)
VTONFS(dvp)->n_attrstamp = 0;
1994-05-24 10:09:53 +00:00
cache_purge(dvp);
cache_purge(vp);
/*
* Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
*/
if (error == ENOENT)
error = 0;
return (error);
}
/*
* nfs readdir call
*/
static int
nfs_readdir(struct vop_readdir_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
struct uio *uio = ap->a_uio;
1994-05-24 10:09:53 +00:00
int tresid, error;
struct vattr vattr;
if (vp->v_type != VDIR)
return (EPERM);
/*
* First, check for hit on the EOF offset cache
*/
if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
1994-05-24 10:09:53 +00:00
(np->n_flag & NMODIFIED) == 0) {
if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0 &&
np->n_mtime == vattr.va_mtime.tv_sec) {
1994-05-24 10:09:53 +00:00
nfsstats.direofcache_hits++;
return (0);
}
}
/*
* Call nfs_bioread() to do the real work.
*/
tresid = uio->uio_resid;
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
error = nfs_bioread(vp, uio, 0, ap->a_cred);
1994-05-24 10:09:53 +00:00
if (!error && uio->uio_resid == tresid)
nfsstats.direofcache_misses++;
return (error);
}
/*
* Readdir rpc call.
* Called from below the buffer cache by nfs_doio().
*/
int
nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1994-05-24 10:09:53 +00:00
{
int len, left;
struct dirent *dp = NULL;
u_int32_t *tl;
caddr_t cp;
nfsuint64 *cookiep;
caddr_t bpos, dpos;
struct mbuf *mreq, *mrep, *md, *mb;
nfsuint64 cookie;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
struct nfsnode *dnp = VTONFS(vp);
u_quad_t fileno;
int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
int attrflag;
int v3 = NFS_ISV3(vp);
#ifndef DIAGNOSTIC
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
(uiop->uio_resid & (DIRBLKSIZ - 1)))
panic("nfs readdirrpc bad uio");
#endif
1994-05-24 10:09:53 +00:00
/*
* If there is no cookie, assume directory was stale.
*/
cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
if (cookiep)
cookie = *cookiep;
else
return (NFSERR_BAD_COOKIE);
/*
* Loop around doing readdir rpc's of size nm_readdirsize
* truncated to a multiple of DIRBLKSIZ.
1994-05-24 10:09:53 +00:00
* The stopping criteria is EOF or buffer full.
*/
while (more_dirs && bigenough) {
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_READDIR]++;
mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
NFSX_READDIR(v3));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
if (v3) {
tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
*tl++ = cookie.nfsuquad[0];
*tl++ = cookie.nfsuquad[1];
*tl++ = dnp->n_cookieverf.nfsuquad[0];
*tl++ = dnp->n_cookieverf.nfsuquad[1];
} else {
tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
*tl++ = cookie.nfsuquad[0];
}
*tl = txdr_unsigned(nmp->nm_readdirsize);
nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
if (v3) {
nfsm_postop_attr(vp, attrflag);
if (!error) {
tl = nfsm_dissect(u_int32_t *,
2 * NFSX_UNSIGNED);
dnp->n_cookieverf.nfsuquad[0] = *tl++;
dnp->n_cookieverf.nfsuquad[1] = *tl;
} else {
m_freem(mrep);
goto nfsmout;
}
}
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1994-05-24 10:09:53 +00:00
more_dirs = fxdr_unsigned(int, *tl);
1994-05-24 10:09:53 +00:00
/* loop thru the dir entries, doctoring them to 4bsd form */
while (more_dirs && bigenough) {
if (v3) {
tl = nfsm_dissect(u_int32_t *,
3 * NFSX_UNSIGNED);
fileno = fxdr_hyper(tl);
len = fxdr_unsigned(int, *(tl + 2));
} else {
tl = nfsm_dissect(u_int32_t *,
2 * NFSX_UNSIGNED);
fileno = fxdr_unsigned(u_quad_t, *tl++);
len = fxdr_unsigned(int, *tl);
}
1994-05-24 10:09:53 +00:00
if (len <= 0 || len > NFS_MAXNAMLEN) {
error = EBADRPC;
m_freem(mrep);
goto nfsmout;
}
tlen = nfsm_rndup(len);
if (tlen == len)
tlen += 4; /* To ensure null termination */
left = DIRBLKSIZ - blksiz;
if ((tlen + DIRHDSIZ) > left) {
dp->d_reclen += left;
uiop->uio_iov->iov_base += left;
uiop->uio_iov->iov_len -= left;
uiop->uio_offset += left;
uiop->uio_resid -= left;
blksiz = 0;
1994-05-24 10:09:53 +00:00
}
if ((tlen + DIRHDSIZ) > uiop->uio_resid)
bigenough = 0;
if (bigenough) {
dp = (struct dirent *)uiop->uio_iov->iov_base;
dp->d_fileno = (int)fileno;
dp->d_namlen = len;
dp->d_reclen = tlen + DIRHDSIZ;
dp->d_type = DT_UNKNOWN;
blksiz += dp->d_reclen;
if (blksiz == DIRBLKSIZ)
blksiz = 0;
uiop->uio_offset += DIRHDSIZ;
uiop->uio_resid -= DIRHDSIZ;
uiop->uio_iov->iov_base += DIRHDSIZ;
uiop->uio_iov->iov_len -= DIRHDSIZ;
nfsm_mtouio(uiop, len);
cp = uiop->uio_iov->iov_base;
tlen -= len;
*cp = '\0'; /* null terminate */
uiop->uio_iov->iov_base += tlen;
uiop->uio_iov->iov_len -= tlen;
uiop->uio_offset += tlen;
uiop->uio_resid -= tlen;
} else
nfsm_adv(nfsm_rndup(len));
if (v3) {
tl = nfsm_dissect(u_int32_t *,
3 * NFSX_UNSIGNED);
} else {
tl = nfsm_dissect(u_int32_t *,
2 * NFSX_UNSIGNED);
}
if (bigenough) {
cookie.nfsuquad[0] = *tl++;
if (v3)
cookie.nfsuquad[1] = *tl++;
} else if (v3)
tl += 2;
else
tl++;
1994-05-24 10:09:53 +00:00
more_dirs = fxdr_unsigned(int, *tl);
}
/*
* If at end of rpc data, get the eof boolean
*/
if (!more_dirs) {
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1994-05-24 10:09:53 +00:00
more_dirs = (fxdr_unsigned(int, *tl) == 0);
}
m_freem(mrep);
}
/*
* Fill last record, iff any, out to a multiple of DIRBLKSIZ
1994-05-24 10:09:53 +00:00
* by increasing d_reclen for the last record.
*/
if (blksiz > 0) {
left = DIRBLKSIZ - blksiz;
dp->d_reclen += left;
uiop->uio_iov->iov_base += left;
uiop->uio_iov->iov_len -= left;
uiop->uio_offset += left;
uiop->uio_resid -= left;
}
/*
* We are now either at the end of the directory or have filled the
* block.
*/
if (bigenough)
dnp->n_direofoffset = uiop->uio_offset;
else {
if (uiop->uio_resid > 0)
printf("EEK! readdirrpc resid > 0\n");
cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
*cookiep = cookie;
1994-05-24 10:09:53 +00:00
}
nfsmout:
return (error);
}
/*
* NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
1994-05-24 10:09:53 +00:00
*/
int
nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1994-05-24 10:09:53 +00:00
{
int len, left;
struct dirent *dp;
u_int32_t *tl;
caddr_t cp;
struct vnode *newvp;
nfsuint64 *cookiep;
caddr_t bpos, dpos, dpossav1, dpossav2;
struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
1994-05-24 10:09:53 +00:00
struct nameidata nami, *ndp = &nami;
struct componentname *cnp = &ndp->ni_cnd;
nfsuint64 cookie;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
struct nfsnode *dnp = VTONFS(vp), *np;
nfsfh_t *fhp;
u_quad_t fileno;
int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
int attrflag, fhsize;
#ifndef nolint
dp = (struct dirent *)0;
#endif
#ifndef DIAGNOSTIC
if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
(uiop->uio_resid & (DIRBLKSIZ - 1)))
panic("nfs readdirplusrpc bad uio");
#endif
1994-05-24 10:09:53 +00:00
ndp->ni_dvp = vp;
newvp = NULLVP;
/*
* If there is no cookie, assume directory was stale.
*/
cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
if (cookiep)
cookie = *cookiep;
else
return (NFSERR_BAD_COOKIE);
1994-05-24 10:09:53 +00:00
/*
* Loop around doing readdir rpc's of size nm_readdirsize
* truncated to a multiple of DIRBLKSIZ.
1994-05-24 10:09:53 +00:00
* The stopping criteria is EOF or buffer full.
*/
while (more_dirs && bigenough) {
nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
NFSX_FH(1) + 6 * NFSX_UNSIGNED);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, 1);
tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
*tl++ = cookie.nfsuquad[0];
*tl++ = cookie.nfsuquad[1];
*tl++ = dnp->n_cookieverf.nfsuquad[0];
*tl++ = dnp->n_cookieverf.nfsuquad[1];
*tl++ = txdr_unsigned(nmp->nm_readdirsize);
*tl = txdr_unsigned(nmp->nm_rsize);
nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
nfsm_postop_attr(vp, attrflag);
if (error) {
m_freem(mrep);
goto nfsmout;
}
tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
dnp->n_cookieverf.nfsuquad[0] = *tl++;
dnp->n_cookieverf.nfsuquad[1] = *tl++;
1994-05-24 10:09:53 +00:00
more_dirs = fxdr_unsigned(int, *tl);
1995-05-30 08:16:23 +00:00
1994-05-24 10:09:53 +00:00
/* loop thru the dir entries, doctoring them to 4bsd form */
while (more_dirs && bigenough) {
tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
fileno = fxdr_hyper(tl);
len = fxdr_unsigned(int, *(tl + 2));
1994-05-24 10:09:53 +00:00
if (len <= 0 || len > NFS_MAXNAMLEN) {
error = EBADRPC;
m_freem(mrep);
goto nfsmout;
}
tlen = nfsm_rndup(len);
if (tlen == len)
tlen += 4; /* To ensure null termination*/
left = DIRBLKSIZ - blksiz;
if ((tlen + DIRHDSIZ) > left) {
dp->d_reclen += left;
uiop->uio_iov->iov_base += left;
uiop->uio_iov->iov_len -= left;
uiop->uio_offset += left;
uiop->uio_resid -= left;
blksiz = 0;
}
1994-05-24 10:09:53 +00:00
if ((tlen + DIRHDSIZ) > uiop->uio_resid)
bigenough = 0;
if (bigenough) {
1994-05-24 10:09:53 +00:00
dp = (struct dirent *)uiop->uio_iov->iov_base;
dp->d_fileno = (int)fileno;
1994-05-24 10:09:53 +00:00
dp->d_namlen = len;
dp->d_reclen = tlen + DIRHDSIZ;
dp->d_type = DT_UNKNOWN;
blksiz += dp->d_reclen;
if (blksiz == DIRBLKSIZ)
blksiz = 0;
uiop->uio_offset += DIRHDSIZ;
1994-05-24 10:09:53 +00:00
uiop->uio_resid -= DIRHDSIZ;
uiop->uio_iov->iov_base += DIRHDSIZ;
uiop->uio_iov->iov_len -= DIRHDSIZ;
cnp->cn_nameptr = uiop->uio_iov->iov_base;
cnp->cn_namelen = len;
nfsm_mtouio(uiop, len);
cp = uiop->uio_iov->iov_base;
tlen -= len;
*cp = '\0';
1994-05-24 10:09:53 +00:00
uiop->uio_iov->iov_base += tlen;
uiop->uio_iov->iov_len -= tlen;
uiop->uio_offset += tlen;
1994-05-24 10:09:53 +00:00
uiop->uio_resid -= tlen;
} else
nfsm_adv(nfsm_rndup(len));
tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
if (bigenough) {
cookie.nfsuquad[0] = *tl++;
cookie.nfsuquad[1] = *tl++;
} else
tl += 2;
/*
* Since the attributes are before the file handle
* (sigh), we must skip over the attributes and then
* come back and get them.
*/
attrflag = fxdr_unsigned(int, *tl);
if (attrflag) {
dpossav1 = dpos;
mdsav1 = md;
nfsm_adv(NFSX_V3FATTR);
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
doit = fxdr_unsigned(int, *tl);
if (doit) {
nfsm_getfh(fhp, fhsize, 1);
if (NFS_CMPFH(dnp, fhp, fhsize)) {
VREF(vp);
newvp = vp;
np = dnp;
} else {
error = nfs_nget(vp->v_mount, fhp,
fhsize, &np);
if (error)
doit = 0;
else
newvp = NFSTOV(np);
}
}
if (doit && bigenough) {
dpossav2 = dpos;
dpos = dpossav1;
mdsav2 = md;
md = mdsav1;
nfsm_loadattr(newvp, (struct vattr *)0);
dpos = dpossav2;
md = mdsav2;
dp->d_type =
IFTODT(VTTOIF(np->n_vattr.va_type));
ndp->ni_vp = newvp;
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
}
1994-05-24 10:09:53 +00:00
} else {
/* Just skip over the file handle */
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
i = fxdr_unsigned(int, *tl);
nfsm_adv(nfsm_rndup(i));
1994-05-24 10:09:53 +00:00
}
if (newvp != NULLVP) {
if (newvp == vp)
vrele(newvp);
else
vput(newvp);
newvp = NULLVP;
1994-05-24 10:09:53 +00:00
}
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1994-05-24 10:09:53 +00:00
more_dirs = fxdr_unsigned(int, *tl);
}
/*
* If at end of rpc data, get the eof boolean
*/
if (!more_dirs) {
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1994-05-24 10:09:53 +00:00
more_dirs = (fxdr_unsigned(int, *tl) == 0);
}
m_freem(mrep);
}
/*
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* Fill last record, iff any, out to a multiple of DIRBLKSIZ
1994-05-24 10:09:53 +00:00
* by increasing d_reclen for the last record.
*/
if (blksiz > 0) {
left = DIRBLKSIZ - blksiz;
dp->d_reclen += left;
uiop->uio_iov->iov_base += left;
uiop->uio_iov->iov_len -= left;
uiop->uio_offset += left;
uiop->uio_resid -= left;
}
/*
* We are now either at the end of the directory or have filled the
* block.
*/
if (bigenough)
dnp->n_direofoffset = uiop->uio_offset;
else {
if (uiop->uio_resid > 0)
printf("EEK! readdirplusrpc resid > 0\n");
cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
*cookiep = cookie;
1994-05-24 10:09:53 +00:00
}
nfsmout:
if (newvp != NULLVP) {
if (newvp == vp)
vrele(newvp);
else
vput(newvp);
newvp = NULLVP;
}
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* Silly rename. To make the NFS filesystem that is stateless look a little
* more like the "ufs" a remove of an active vnode is translated to a rename
* to a funny looking filename that is removed by nfs_inactive on the
* nfsnode. There is the potential for another process on a different client
* to create the same funny name between the nfs_lookitup() fails and the
* nfs_rename() completes, but...
*/
static int
nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
1994-05-24 10:09:53 +00:00
{
struct sillyrename *sp;
struct nfsnode *np;
1994-05-24 10:09:53 +00:00
int error;
short pid;
cache_purge(dvp);
np = VTONFS(vp);
#ifndef DIAGNOSTIC
if (vp->v_type == VDIR)
panic("nfs: sillyrename dir");
#endif
1994-05-24 10:09:53 +00:00
MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
M_NFSREQ, M_WAITOK);
sp->s_cred = crhold(cnp->cn_cred);
1994-05-24 10:09:53 +00:00
sp->s_dvp = dvp;
VREF(dvp);
/* Fudge together a funny name */
pid = cnp->cn_thread->td_proc->p_pid;
sp->s_namlen = sprintf(sp->s_name, ".nfsA%04x4.4", pid);
1994-05-24 10:09:53 +00:00
/* Try lookitups until we get one that isn't there */
while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
cnp->cn_thread, (struct nfsnode **)0) == 0) {
1994-05-24 10:09:53 +00:00
sp->s_name[4]++;
if (sp->s_name[4] > 'z') {
error = EINVAL;
goto bad;
}
}
error = nfs_renameit(dvp, cnp, sp);
if (error)
1994-05-24 10:09:53 +00:00
goto bad;
error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
cnp->cn_thread, &np);
1994-05-24 10:09:53 +00:00
np->n_sillyrename = sp;
return (0);
bad:
vrele(sp->s_dvp);
crfree(sp->s_cred);
free((caddr_t)sp, M_NFSREQ);
return (error);
}
/*
* Look up a file name and optionally either update the file handle or
* allocate an nfsnode, depending on the value of npp.
* npp == NULL --> just do the lookup
* *npp == NULL --> allocate a new nfsnode and make sure attributes are
* handled too
* *npp != NULL --> update the file handle in the vnode
1994-05-24 10:09:53 +00:00
*/
static int
nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
struct thread *td, struct nfsnode **npp)
1994-05-24 10:09:53 +00:00
{
struct vnode *newvp = (struct vnode *)0;
struct nfsnode *np, *dnp = VTONFS(dvp);
caddr_t bpos, dpos;
int error = 0, fhlen, attrflag;
struct mbuf *mreq, *mrep, *md, *mb;
nfsfh_t *nfhp;
int v3 = NFS_ISV3(dvp);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_LOOKUP]++;
mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
nfsm_strtom(name, len, NFS_MAXNAMLEN);
nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
if (npp && !error) {
nfsm_getfh(nfhp, fhlen, v3);
if (*npp) {
np = *npp;
if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
free((caddr_t)np->n_fhp, M_NFSBIGFH);
np->n_fhp = &np->n_fh;
} else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
np->n_fhsize = fhlen;
newvp = NFSTOV(np);
} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
VREF(dvp);
newvp = dvp;
} else {
error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
if (error) {
m_freem(mrep);
return (error);
}
newvp = NFSTOV(np);
}
if (v3) {
nfsm_postop_attr(newvp, attrflag);
if (!attrflag && *npp == NULL) {
m_freem(mrep);
if (newvp == dvp)
vrele(newvp);
else
vput(newvp);
return (ENOENT);
}
} else
nfsm_loadattr(newvp, (struct vattr *)0);
1994-05-24 10:09:53 +00:00
}
m_freem(mrep);
nfsmout:
if (npp && *npp == NULL) {
if (error) {
if (newvp) {
if (newvp == dvp)
vrele(newvp);
else
vput(newvp);
}
} else
*npp = np;
}
return (error);
}
/*
* Nfs Version 3 commit rpc
*/
int
nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
struct thread *td)
{
u_int32_t *tl;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
caddr_t bpos, dpos;
int error = 0, wccflag = NFSV3_WCCRATTR;
struct mbuf *mreq, *mrep, *md, *mb;
if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
return (0);
nfsstats.rpccnt[NFSPROC_COMMIT]++;
mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, 1);
tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
txdr_hyper(offset, tl);
tl += 2;
*tl = txdr_unsigned(cnt);
nfsm_request(vp, NFSPROC_COMMIT, td, cred);
nfsm_wcc_data(vp, wccflag);
if (!error) {
tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
NFSX_V3WRITEVERF)) {
bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
NFSX_V3WRITEVERF);
error = NFSERR_STALEWRITEVERF;
}
1994-05-24 10:09:53 +00:00
}
m_freem(mrep);
nfsmout:
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* Strategy routine.
* For async requests when nfsiod(s) are running, queue the request by
* calling nfs_asyncio(), otherwise just all nfs_doio() to do the
* request.
*/
static int
nfs_strategy(struct vop_strategy_args *ap)
1994-05-24 10:09:53 +00:00
{
struct buf *bp = ap->a_bp;
1994-05-24 10:09:53 +00:00
struct ucred *cr;
struct thread *td;
1994-05-24 10:09:53 +00:00
int error = 0;
KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
if (bp->b_flags & B_PHYS)
panic("nfs physio");
1994-05-24 10:09:53 +00:00
if (bp->b_flags & B_ASYNC)
td = (struct thread *)0;
1994-05-24 10:09:53 +00:00
else
td = curthread; /* XXX */
if (bp->b_iocmd == BIO_READ)
1994-05-24 10:09:53 +00:00
cr = bp->b_rcred;
else
cr = bp->b_wcred;
1994-05-24 10:09:53 +00:00
/*
* If the op is asynchronous and an i/o daemon is waiting
* queue the request, wake it up and wait for completion
* otherwise just do it ourselves.
*/
if ((bp->b_flags & B_ASYNC) == 0 ||
nfs_asyncio(bp, NOCRED, td))
error = nfs_doio(bp, cr, td);
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* fsync vnode op. Just call nfs_flush() with commit == 1.
1994-05-24 10:09:53 +00:00
*/
/* ARGSUSED */
static int
nfs_fsync(struct vop_fsync_args *ap)
1994-05-24 10:09:53 +00:00
{
return (nfs_flush(ap->a_vp, ap->a_cred, ap->a_waitfor, ap->a_td, 1));
}
/*
* Flush all the blocks associated with a vnode.
* Walk through the buffer pool and push any dirty pages
* associated with the vnode.
*/
static int
nfs_flush(struct vnode *vp, struct ucred *cred, int waitfor, struct thread *td,
int commit)
{
struct nfsnode *np = VTONFS(vp);
struct buf *bp;
int i;
1994-05-24 10:09:53 +00:00
struct buf *nbp;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
int passone = 1;
u_quad_t off, endoff, toff;
struct ucred* wcred = NULL;
struct buf **bvec = NULL;
#ifndef NFS_COMMITBVECSIZ
#define NFS_COMMITBVECSIZ 20
#endif
struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
int bvecsize = 0, bveccount;
1994-05-24 10:09:53 +00:00
if (nmp->nm_flag & NFSMNT_INT)
slpflag = PCATCH;
if (!commit)
passone = 0;
/*
* A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
* server, but nas not been committed to stable storage on the server
* yet. On the first pass, the byte range is worked out and the commit
* rpc is done. On the second pass, nfs_writebp() is called to do the
* job.
*/
again:
off = (u_quad_t)-1;
endoff = 0;
bvecpos = 0;
if (NFS_ISV3(vp) && commit) {
s = splbio();
/*
* Count up how many buffers waiting for a commit.
*/
bveccount = 0;
for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
nbp = TAILQ_NEXT(bp, b_vnbufs);
if (BUF_REFCNT(bp) == 0 &&
(bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
== (B_DELWRI | B_NEEDCOMMIT))
bveccount++;
}
/*
* Allocate space to remember the list of bufs to commit. It is
* important to use M_NOWAIT here to avoid a race with nfs_write.
* If we can't get memory (for whatever reason), we will end up
* committing the buffers one-by-one in the loop below.
*/
if (bvec != NULL && bvec != bvec_on_stack)
free(bvec, M_TEMP);
if (bveccount > NFS_COMMITBVECSIZ) {
bvec = (struct buf **)
malloc(bveccount * sizeof(struct buf *),
M_TEMP, M_NOWAIT);
if (bvec == NULL) {
bvec = bvec_on_stack;
bvecsize = NFS_COMMITBVECSIZ;
} else
bvecsize = bveccount;
} else {
bvec = bvec_on_stack;
bvecsize = NFS_COMMITBVECSIZ;
}
for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
nbp = TAILQ_NEXT(bp, b_vnbufs);
if (bvecpos >= bvecsize)
break;
if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
(B_DELWRI | B_NEEDCOMMIT) ||
BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
continue;
bremfree(bp);
/*
* Work out if all buffers are using the same cred
* so we can deal with them all with one commit.
*
* NOTE: we are not clearing B_DONE here, so we have
* to do it later on in this routine if we intend to
* initiate I/O on the bp.
*
* Note: to avoid loopback deadlocks, we do not
* assign b_runningbufspace.
*/
if (wcred == NULL)
wcred = bp->b_wcred;
else if (wcred != bp->b_wcred)
wcred = NOCRED;
bp->b_flags |= B_WRITEINPROG;
vfs_busy_pages(bp, 1);
/*
* bp is protected by being locked, but nbp is not
* and vfs_busy_pages() may sleep. We have to
* recalculate nbp.
*/
nbp = TAILQ_NEXT(bp, b_vnbufs);
/*
* A list of these buffers is kept so that the
* second loop knows which buffers have actually
* been committed. This is necessary, since there
* may be a race between the commit rpc and new
* uncommitted writes on the file.
*/
bvec[bvecpos++] = bp;
toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
bp->b_dirtyoff;
if (toff < off)
off = toff;
toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
if (toff > endoff)
endoff = toff;
}
splx(s);
}
if (bvecpos > 0) {
/*
* Commit data on the server, as required.
* If all bufs are using the same wcred, then use that with
* one call for all of them, otherwise commit each one
* separately.
*/
if (wcred != NOCRED)
retv = nfs_commit(vp, off, (int)(endoff - off),
wcred, td);
else {
retv = 0;
for (i = 0; i < bvecpos; i++) {
off_t off, size;
bp = bvec[i];
off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
bp->b_dirtyoff;
size = (u_quad_t)(bp->b_dirtyend
- bp->b_dirtyoff);
retv = nfs_commit(vp, off, (int)size,
bp->b_wcred, td);
if (retv) break;
}
}
if (retv == NFSERR_STALEWRITEVERF)
nfs_clearcommit(vp->v_mount);
/*
* Now, either mark the blocks I/O done or mark the
* blocks dirty, depending on whether the commit
* succeeded.
*/
for (i = 0; i < bvecpos; i++) {
bp = bvec[i];
Synopsis of problem being fixed: Dan Nelson originally reported that blocks of zeros could wind up in a file written to over NFS by a client. The problem only occurs a few times per several gigabytes of data. This problem turned out to be bug #3 below. bug #1: B_CLUSTEROK must be cleared when an NFS buffer is reverted from stage 2 (ready for commit rpc) to stage 1 (ready for write). Reversions can occur when a dirty NFS buffer is redirtied with new data. Otherwise the VFS/BIO system may end up thinking that a stage 1 NFS buffer is clusterable. Stage 1 NFS buffers are not clusterable. bug #2: B_CLUSTEROK was inappropriately set for a 'short' NFS buffer (short buffers only occur near the EOF of the file). Change to only set when the buffer is a full biosize (usually 8K). This bug has no effect but should be fixed in -current anyway. It need not be backported. bug #3: B_NEEDCOMMIT was inappropriately set in nfs_flush() (which is typically only called by the update daemon). nfs_flush() does a multi-pass loop but due to the lack of vnode locking it is possible for new buffers to be added to the dirtyblkhd list while a flush operation is going on. This may result in nfs_flush() setting B_NEEDCOMMIT on a buffer which has *NOT* yet gone through its stage 1 write, causing only the commit rpc to be made and thus causing the contents of the buffer to be thrown away (never sent to the server). The patch also contains some cleanup, which only applies to the commit into -current. Reviewed by: dg, julian Originally Reported by: Dan Nelson <dnelson@emsphone.com>
1999-12-12 06:09:57 +00:00
bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG | B_CLUSTEROK);
if (retv) {
/*
* Error, leave B_DELWRI intact
*/
vfs_unbusy_pages(bp);
brelse(bp);
} else {
/*
* Success, remove B_DELWRI ( bundirty() ).
*
* b_dirtyoff/b_dirtyend seem to be NFS
* specific. We should probably move that
* into bundirty(). XXX
*/
s = splbio();
vp->v_numoutput++;
bp->b_flags |= B_ASYNC;
bundirty(bp);
bp->b_flags &= ~B_DONE;
bp->b_ioflags &= ~BIO_ERROR;
bp->b_dirtyoff = bp->b_dirtyend = 0;
splx(s);
bufdone(bp);
}
}
}
/*
* Start/do any write(s) that are required.
*/
1994-05-24 10:09:53 +00:00
loop:
s = splbio();
for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
nbp = TAILQ_NEXT(bp, b_vnbufs);
if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
if (waitfor != MNT_WAIT || passone)
1994-05-24 10:09:53 +00:00
continue;
error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL,
"nfsfsync", slpflag, slptimeo);
1994-05-24 10:09:53 +00:00
splx(s);
if (error == 0)
panic("nfs_fsync: inconsistent lock");
if (error == ENOLCK)
goto loop;
if (nfs_sigintr(nmp, (struct nfsreq *)0, td->td_proc)) {
error = EINTR;
goto done;
}
if (slpflag == PCATCH) {
1994-05-24 10:09:53 +00:00
slpflag = 0;
slptimeo = 2 * hz;
}
goto loop;
}
if ((bp->b_flags & B_DELWRI) == 0)
panic("nfs_fsync: not dirty");
if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
BUF_UNLOCK(bp);
continue;
}
1994-05-24 10:09:53 +00:00
bremfree(bp);
if (passone || !commit)
bp->b_flags |= B_ASYNC;
else
Synopsis of problem being fixed: Dan Nelson originally reported that blocks of zeros could wind up in a file written to over NFS by a client. The problem only occurs a few times per several gigabytes of data. This problem turned out to be bug #3 below. bug #1: B_CLUSTEROK must be cleared when an NFS buffer is reverted from stage 2 (ready for commit rpc) to stage 1 (ready for write). Reversions can occur when a dirty NFS buffer is redirtied with new data. Otherwise the VFS/BIO system may end up thinking that a stage 1 NFS buffer is clusterable. Stage 1 NFS buffers are not clusterable. bug #2: B_CLUSTEROK was inappropriately set for a 'short' NFS buffer (short buffers only occur near the EOF of the file). Change to only set when the buffer is a full biosize (usually 8K). This bug has no effect but should be fixed in -current anyway. It need not be backported. bug #3: B_NEEDCOMMIT was inappropriately set in nfs_flush() (which is typically only called by the update daemon). nfs_flush() does a multi-pass loop but due to the lack of vnode locking it is possible for new buffers to be added to the dirtyblkhd list while a flush operation is going on. This may result in nfs_flush() setting B_NEEDCOMMIT on a buffer which has *NOT* yet gone through its stage 1 write, causing only the commit rpc to be made and thus causing the contents of the buffer to be thrown away (never sent to the server). The patch also contains some cleanup, which only applies to the commit into -current. Reviewed by: dg, julian Originally Reported by: Dan Nelson <dnelson@emsphone.com>
1999-12-12 06:09:57 +00:00
bp->b_flags |= B_ASYNC | B_WRITEINPROG;
1994-05-24 10:09:53 +00:00
splx(s);
BUF_WRITE(bp);
1994-05-24 10:09:53 +00:00
goto loop;
}
splx(s);
if (passone) {
passone = 0;
goto again;
}
if (waitfor == MNT_WAIT) {
1994-05-24 10:09:53 +00:00
while (vp->v_numoutput) {
vp->v_flag |= VBWAIT;
error = tsleep((caddr_t)&vp->v_numoutput,
slpflag | (PRIBIO + 1), "nfsfsync", slptimeo);
if (error) {
if (nfs_sigintr(nmp, (struct nfsreq *)0, td->td_proc)) {
error = EINTR;
goto done;
}
1994-05-24 10:09:53 +00:00
if (slpflag == PCATCH) {
slpflag = 0;
slptimeo = 2 * hz;
}
}
}
if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) && commit) {
1994-05-24 10:09:53 +00:00
goto loop;
}
}
if (np->n_flag & NWRITEERR) {
error = np->n_error;
np->n_flag &= ~NWRITEERR;
}
done:
if (bvec != NULL && bvec != bvec_on_stack)
free(bvec, M_TEMP);
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* NFS advisory byte-level locks.
*/
static int
nfs_advlock(struct vop_advlock_args *ap)
1994-05-24 10:09:53 +00:00
{
if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
struct nfsnode *np = VTONFS(ap->a_vp);
return (lf_advlock(ap, &(np->n_lockf), np->n_size));
}
return (nfs_dolock(ap));
1994-05-24 10:09:53 +00:00
}
/*
* Print out the contents of an nfsnode.
*/
static int
nfs_print(struct vop_print_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
1994-05-24 10:09:53 +00:00
printf("tag VT_NFS, fileid %ld fsid 0x%x",
1994-05-24 10:09:53 +00:00
np->n_vattr.va_fileid, np->n_vattr.va_fsid);
if (vp->v_type == VFIFO)
fifo_printinfo(vp);
printf("\n");
return (0);
1994-05-24 10:09:53 +00:00
}
/*
* This is the "real" nfs::bwrite(struct buf*).
* B_WRITEINPROG isn't set unless the force flag is one and it
* handles the B_NEEDCOMMIT flag.
* We set B_CACHE if this is a VMIO buffer.
*/
int
nfs_writebp(struct buf *bp, int force, struct thread *td)
{
int s;
int oldflags = bp->b_flags;
#if 0
int retv = 1;
off_t off;
#endif
if (BUF_REFCNT(bp) == 0)
panic("bwrite: buffer is not locked???");
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
if (bp->b_flags & B_INVAL) {
brelse(bp);
return(0);
}
bp->b_flags |= B_CACHE;
This mega-commit is meant to fix numerous interrelated problems. There has been some bitrot and incorrect assumptions in the vfs_bio code. These problems have manifest themselves worse on NFS type filesystems, but can still affect local filesystems under certain circumstances. Most of the problems have involved mmap consistancy, and as a side-effect broke the vfs.ioopt code. This code might have been committed seperately, but almost everything is interrelated. 1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that are fully valid. 2) Rather than deactivating erroneously read initial (header) pages in kern_exec, we now free them. 3) Fix the rundown of non-VMIO buffers that are in an inconsistent (missing vp) state. 4) Fix the disassociation of pages from buffers in brelse. The previous code had rotted and was faulty in a couple of important circumstances. 5) Remove a gratuitious buffer wakeup in vfs_vmio_release. 6) Remove a crufty and currently unused cluster mechanism for VBLK files in vfs_bio_awrite. When the code is functional, I'll add back a cleaner version. 7) The page busy count wakeups assocated with the buffer cache usage were incorrectly cleaned up in a previous commit by me. Revert to the original, correct version, but with a cleaner implementation. 8) The cluster read code now tries to keep data associated with buffers more aggressively (without breaking the heuristics) when it is presumed that the read data (buffers) will be soon needed. 9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The delay loop waiting is not useful for filesystem locks, due to the length of the time intervals. 10) Correct and clean-up spec_getpages. 11) Implement a fully functional nfs_getpages, nfs_putpages. 12) Fix nfs_write so that modifications are coherent with the NFS data on the server disk (at least as well as NFS seems to allow.) 13) Properly support MS_INVALIDATE on NFS. 14) Properly pass down MS_INVALIDATE to lower levels of the VM code from vm_map_clean. 15) Better support the notion of pages being busy but valid, so that fewer in-transit waits occur. (use p->busy more for pageouts instead of PG_BUSY.) Since the page is fully valid, it is still usable for reads. 16) It is possible (in error) for cached pages to be busy. Make the page allocation code handle that case correctly. (It should probably be a printf or panic, but I want the system to handle coding errors robustly. I'll probably add a printf.) 17) Correct the design and usage of vm_page_sleep. It didn't handle consistancy problems very well, so make the design a little less lofty. After vm_page_sleep, if it ever blocked, it is still important to relookup the page (if the object generation count changed), and verify it's status (always.) 18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up. 19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush. 20) Fix vm_pager_put_pages and it's descendents to support an int flag instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
/*
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* Undirty the bp. We will redirty it later if the I/O fails.
*/
s = splbio();
bundirty(bp);
bp->b_flags &= ~B_DONE;
bp->b_ioflags &= ~BIO_ERROR;
bp->b_iocmd = BIO_WRITE;
bp->b_vp->v_numoutput++;
curthread->td_proc->p_stats->p_ru.ru_oublock++;
splx(s);
/*
* Note: to avoid loopback deadlocks, we do not
* assign b_runningbufspace.
*/
vfs_busy_pages(bp, 1);
if (force)
bp->b_flags |= B_WRITEINPROG;
BUF_KERNPROC(bp);
BUF_STRATEGY(bp);
if( (oldflags & B_ASYNC) == 0) {
int rtval = bufwait(bp);
if (oldflags & B_DELWRI) {
s = splbio();
reassignbuf(bp, bp->b_vp);
splx(s);
}
This mega-commit is meant to fix numerous interrelated problems. There has been some bitrot and incorrect assumptions in the vfs_bio code. These problems have manifest themselves worse on NFS type filesystems, but can still affect local filesystems under certain circumstances. Most of the problems have involved mmap consistancy, and as a side-effect broke the vfs.ioopt code. This code might have been committed seperately, but almost everything is interrelated. 1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that are fully valid. 2) Rather than deactivating erroneously read initial (header) pages in kern_exec, we now free them. 3) Fix the rundown of non-VMIO buffers that are in an inconsistent (missing vp) state. 4) Fix the disassociation of pages from buffers in brelse. The previous code had rotted and was faulty in a couple of important circumstances. 5) Remove a gratuitious buffer wakeup in vfs_vmio_release. 6) Remove a crufty and currently unused cluster mechanism for VBLK files in vfs_bio_awrite. When the code is functional, I'll add back a cleaner version. 7) The page busy count wakeups assocated with the buffer cache usage were incorrectly cleaned up in a previous commit by me. Revert to the original, correct version, but with a cleaner implementation. 8) The cluster read code now tries to keep data associated with buffers more aggressively (without breaking the heuristics) when it is presumed that the read data (buffers) will be soon needed. 9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The delay loop waiting is not useful for filesystem locks, due to the length of the time intervals. 10) Correct and clean-up spec_getpages. 11) Implement a fully functional nfs_getpages, nfs_putpages. 12) Fix nfs_write so that modifications are coherent with the NFS data on the server disk (at least as well as NFS seems to allow.) 13) Properly support MS_INVALIDATE on NFS. 14) Properly pass down MS_INVALIDATE to lower levels of the VM code from vm_map_clean. 15) Better support the notion of pages being busy but valid, so that fewer in-transit waits occur. (use p->busy more for pageouts instead of PG_BUSY.) Since the page is fully valid, it is still usable for reads. 16) It is possible (in error) for cached pages to be busy. Make the page allocation code handle that case correctly. (It should probably be a printf or panic, but I want the system to handle coding errors robustly. I'll probably add a printf.) 17) Correct the design and usage of vm_page_sleep. It didn't handle consistancy problems very well, so make the design a little less lofty. After vm_page_sleep, if it ever blocked, it is still important to relookup the page (if the object generation count changed), and verify it's status (always.) 18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up. 19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush. 20) Fix vm_pager_put_pages and it's descendents to support an int flag instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
brelse(bp);
return (rtval);
}
return (0);
}
1994-05-24 10:09:53 +00:00
/*
* nfs special file access vnode op.
* Essentially just get vattr and then imitate iaccess() since the device is
* local to the client.
*/
static int
nfsspec_access(struct vop_access_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vattr *vap;
gid_t *gp;
struct ucred *cred = ap->a_cred;
struct vnode *vp = ap->a_vp;
1994-05-24 10:09:53 +00:00
mode_t mode = ap->a_mode;
struct vattr vattr;
int i;
1994-05-24 10:09:53 +00:00
int error;
/*
* Disallow write attempts on filesystems mounted read-only;
* unless the file is a socket, fifo, or a block or character
* device resident on the filesystem.
*/
if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
switch (vp->v_type) {
case VREG:
case VDIR:
case VLNK:
return (EROFS);
default:
break;
}
}
1994-05-24 10:09:53 +00:00
/*
* If you're the super-user,
* you always get access.
*/
if (cred->cr_uid == 0)
return (0);
vap = &vattr;
error = VOP_GETATTR(vp, vap, cred, ap->a_td);
if (error)
1994-05-24 10:09:53 +00:00
return (error);
/*
* Access check is based on only one of owner, group, public.
* If not owner, then check group. If not a member of the
* group, then check public access.
*/
if (cred->cr_uid != vap->va_uid) {
mode >>= 3;
gp = cred->cr_groups;
for (i = 0; i < cred->cr_ngroups; i++, gp++)
if (vap->va_gid == *gp)
goto found;
mode >>= 3;
found:
;
}
error = (vap->va_mode & mode) == mode ? 0 : EACCES;
return (error);
1994-05-24 10:09:53 +00:00
}
/*
* Read wrapper for special devices.
*/
static int
nfsspec_read(struct vop_read_args *ap)
1994-05-24 10:09:53 +00:00
{
struct nfsnode *np = VTONFS(ap->a_vp);
1994-05-24 10:09:53 +00:00
/*
* Set access flag.
*/
np->n_flag |= NACC;
getnanotime(&np->n_atim);
1994-05-24 10:09:53 +00:00
return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
}
/*
* Write wrapper for special devices.
*/
static int
nfsspec_write(struct vop_write_args *ap)
1994-05-24 10:09:53 +00:00
{
struct nfsnode *np = VTONFS(ap->a_vp);
1994-05-24 10:09:53 +00:00
/*
* Set update flag.
*/
np->n_flag |= NUPD;
getnanotime(&np->n_mtim);
1994-05-24 10:09:53 +00:00
return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
}
/*
* Close wrapper for special devices.
*
* Update the times on the nfsnode then do device close.
*/
static int
nfsspec_close(struct vop_close_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
1994-05-24 10:09:53 +00:00
struct vattr vattr;
if (np->n_flag & (NACC | NUPD)) {
np->n_flag |= NCHG;
if (vp->v_usecount == 1 &&
(vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
VATTR_NULL(&vattr);
if (np->n_flag & NACC)
vattr.va_atime = np->n_atim;
if (np->n_flag & NUPD)
vattr.va_mtime = np->n_mtim;
(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
1994-05-24 10:09:53 +00:00
}
}
return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
}
/*
* Read wrapper for fifos.
*/
static int
nfsfifo_read(struct vop_read_args *ap)
1994-05-24 10:09:53 +00:00
{
struct nfsnode *np = VTONFS(ap->a_vp);
1994-05-24 10:09:53 +00:00
/*
* Set access flag.
*/
np->n_flag |= NACC;
getnanotime(&np->n_atim);
1994-05-24 10:09:53 +00:00
return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
}
/*
* Write wrapper for fifos.
*/
static int
nfsfifo_write(struct vop_write_args *ap)
1994-05-24 10:09:53 +00:00
{
struct nfsnode *np = VTONFS(ap->a_vp);
1994-05-24 10:09:53 +00:00
/*
* Set update flag.
*/
np->n_flag |= NUPD;
getnanotime(&np->n_mtim);
1994-05-24 10:09:53 +00:00
return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
}
/*
* Close wrapper for fifos.
*
* Update the times on the nfsnode then do fifo close.
*/
static int
nfsfifo_close(struct vop_close_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
1994-05-24 10:09:53 +00:00
struct vattr vattr;
struct timespec ts;
1994-05-24 10:09:53 +00:00
if (np->n_flag & (NACC | NUPD)) {
getnanotime(&ts);
if (np->n_flag & NACC)
np->n_atim = ts;
if (np->n_flag & NUPD)
np->n_mtim = ts;
1994-05-24 10:09:53 +00:00
np->n_flag |= NCHG;
if (vp->v_usecount == 1 &&
(vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
VATTR_NULL(&vattr);
if (np->n_flag & NACC)
vattr.va_atime = np->n_atim;
if (np->n_flag & NUPD)
vattr.va_mtime = np->n_mtim;
(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
1994-05-24 10:09:53 +00:00
}
}
return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
}
This fixes a large number of bugs in our NFS client side code. A recent commit by Kirk also fixed a softupdates bug that could easily be triggered by server side NFS. * An edge case with shared R+W mmap()'s and truncate whereby the system would inappropriately clear the dirty bits on still-dirty data. (applicable to all filesystems) THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING. see vm/vm_page.c line 1641 * The straddle case for VM pages and buffer cache buffers when truncating. (applicable to NFS client side) * Possible SMP database corruption due to vm_pager_unmap_page() not clearing the TLB for the other cpu's. (applicable to NFS client side but could effect all filesystems). Note: not considered serious since the corruption occurs beyond the file EOF. * When flusing a dirty buffer due to B_CACHE getting cleared, we were accidently setting B_CACHE again (that is, bwrite() sets B_CACHE), when we really want it to stay clear after the write is complete. This resulted in a corrupt buffer. (applicable to all filesystems but probably only triggered by NFS) * We have to call vtruncbuf() when ftruncate()ing to remove any buffer cache buffers. This is still tentitive, I may be able to remove it due to the second bug fix. (applicable to NFS client side) * vnode_pager_setsize() race against nfs_vinvalbuf()... we have to set n_size before calling nfs_vinvalbuf or the NFS code may recursively vnode_pager_setsize() to the original value before the truncate. This is what was causing the user mmap bus faults in the nfs tester program. (applicable to NFS client side) * Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made by Kirk). Testing program written by: Avadis Tevanian, Jr. Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step') MFC after: 1 week
2001-12-14 01:16:57 +00:00