NetBSD/sys/nfs/nfs_node.c

317 lines
7.9 KiB
C
Raw Normal View History

2014-05-30 12:47:45 +04:00
/* $NetBSD: nfs_node.c,v 1.118 2014/05/30 08:47:45 hannken Exp $ */
1993-03-21 12:45:37 +03:00
/*
* Copyright (c) 1989, 1993
* The Regents of the University of California. All rights reserved.
1993-03-21 12:45:37 +03:00
*
* This code is derived from software contributed to Berkeley by
* Rick Macklem at The University of Guelph.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the University nor the names of its contributors
1993-03-21 12:45:37 +03:00
* 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_node.c 8.6 (Berkeley) 5/22/95
1993-03-21 12:45:37 +03:00
*/
2001-11-10 13:59:08 +03:00
#include <sys/cdefs.h>
2014-05-30 12:47:45 +04:00
__KERNEL_RCSID(0, "$NetBSD: nfs_node.c,v 1.118 2014/05/30 08:47:45 hannken Exp $");
2001-11-10 13:59:08 +03:00
#ifdef _KERNEL_OPT
#include "opt_nfs.h"
#endif
1993-12-18 03:40:47 +03:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/mount.h>
#include <sys/namei.h>
#include <sys/vnode.h>
#include <sys/kernel.h>
#include <sys/pool.h>
#include <sys/lock.h>
#include <sys/hash.h>
2006-05-15 01:31:52 +04:00
#include <sys/kauth.h>
1993-03-21 12:45:37 +03:00
#include <nfs/rpcv2.h>
#include <nfs/nfsproto.h>
1993-12-18 03:40:47 +03:00
#include <nfs/nfs.h>
#include <nfs/nfsnode.h>
#include <nfs/nfsmount.h>
1996-02-10 00:48:19 +03:00
#include <nfs/nfs_var.h>
1993-03-21 12:45:37 +03:00
struct pool nfs_node_pool;
struct pool nfs_vattr_pool;
static struct workqueue *nfs_sillyworkq;
extern int prtactive;
static void nfs_gop_size(struct vnode *, off_t, off_t *, int);
static int nfs_gop_alloc(struct vnode *, off_t, off_t, int, kauth_cred_t);
static int nfs_gop_write(struct vnode *, struct vm_page **, int, int);
static void nfs_sillyworker(struct work *, void *);
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
static const struct genfs_ops nfs_genfsops = {
.gop_size = nfs_gop_size,
.gop_alloc = nfs_gop_alloc,
.gop_write = nfs_gop_write,
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
};
1993-03-21 12:45:37 +03:00
/*
* Reinitialize inode hash table.
1993-03-21 12:45:37 +03:00
*/
1996-02-10 00:48:19 +03:00
void
2009-03-15 20:14:40 +03:00
nfs_node_init(void)
1993-03-21 12:45:37 +03:00
{
pool_init(&nfs_node_pool, sizeof(struct nfsnode), 0, 0, 0, "nfsnodepl",
&pool_allocator_nointr, IPL_NONE);
pool_init(&nfs_vattr_pool, sizeof(struct vattr), 0, 0, 0, "nfsvapl",
&pool_allocator_nointr, IPL_NONE);
if (workqueue_create(&nfs_sillyworkq, "nfssilly", nfs_sillyworker,
NULL, PRI_NONE, IPL_NONE, 0) != 0) {
panic("nfs_node_init");
}
1993-03-21 12:45:37 +03:00
}
/*
* Free resources previously allocated in nfs_node_reinit().
*/
void
2009-03-15 20:14:40 +03:00
nfs_node_done(void)
{
pool_destroy(&nfs_node_pool);
pool_destroy(&nfs_vattr_pool);
workqueue_destroy(nfs_sillyworkq);
}
1993-03-21 12:45:37 +03:00
/*
2014-05-30 12:47:45 +04:00
* Initialize this vnode / nfs node pair.
* Caller assures no other thread will try to load this node.
1993-03-21 12:45:37 +03:00
*/
1996-02-10 00:48:19 +03:00
int
2014-05-30 12:47:45 +04:00
nfs_loadvnode(struct mount *mp, struct vnode *vp,
const void *key, size_t key_len, const void **new_key)
1993-03-21 12:45:37 +03:00
{
2014-05-30 12:47:45 +04:00
int fhsize = key_len;
const nfsfh_t *fhp = key;
struct nfsnode *np;
2014-05-30 12:47:45 +04:00
/* Aloocate and initialize the nfsnode. */
np = pool_get(&nfs_node_pool, PR_WAITOK);
memset(np, 0, sizeof *np);
if (fhsize > NFS_SMALLFH) {
2008-01-02 22:26:45 +03:00
np->n_fhp = kmem_alloc(fhsize, KM_SLEEP);
} else
np->n_fhp = &np->n_fh;
2014-05-30 12:47:45 +04:00
vp->v_tag = VT_NFS;
vp->v_type = VNON;
vp->v_op = nfsv2_vnodeop_p;
vp->v_data = np;
memcpy(np->n_fhp, fhp, fhsize);
np->n_fhsize = fhsize;
np->n_accstamp = -1;
np->n_vattr = pool_get(&nfs_vattr_pool, PR_WAITOK);
2014-05-30 12:47:45 +04:00
np->n_vnode = vp;
2014-05-30 12:47:45 +04:00
/* Initialize genfs node. */
genfs_node_init(vp, &nfs_genfsops);
/*
* Initalize read/write creds to useful values. VOP_OPEN will
* overwrite these.
*/
np->n_rcred = curlwp->l_cred;
2006-05-15 01:31:52 +04:00
kauth_cred_hold(np->n_rcred);
np->n_wcred = curlwp->l_cred;
2006-05-15 01:31:52 +04:00
kauth_cred_hold(np->n_wcred);
NFS_INVALIDATE_ATTRCACHE(np);
uvm_vnp_setsize(vp, 0);
2014-05-30 12:47:45 +04:00
*new_key = np->n_fhp;
return 0;
}
2014-05-30 12:47:45 +04:00
/*
* Look up a vnode/nfsnode by file handle.
* Callers must check for mount points!!
* In all cases, a pointer to a
* nfsnode structure is returned.
*/
int
nfs_nget1(struct mount *mntp, nfsfh_t *fhp, int fhsize, struct nfsnode **npp,
int lkflags)
{
int error;
struct vnode *vp;
error = vcache_get(mntp, fhp, fhsize, &vp);
if (error)
return error;
error = vn_lock(vp, LK_EXCLUSIVE | lkflags);
if (error) {
vrele(vp);
return error;
}
*npp = VTONFS(vp);
return 0;
1993-03-21 12:45:37 +03:00
}
1996-02-10 00:48:19 +03:00
int
nfs_inactive(void *v)
1996-02-10 00:48:19 +03:00
{
struct vop_inactive_args /* {
struct vnode *a_vp;
2008-01-02 14:48:20 +03:00
bool *a_recycle;
1996-02-10 00:48:19 +03:00
} */ *ap = v;
2000-03-30 16:51:13 +04:00
struct nfsnode *np;
struct sillyrename *sp;
struct vnode *vp = ap->a_vp;
1993-03-21 12:45:37 +03:00
np = VTONFS(vp);
if (vp->v_type != VDIR) {
sp = np->n_sillyrename;
np->n_sillyrename = (struct sillyrename *)0;
} else
sp = NULL;
if (sp != NULL)
2008-01-02 14:48:20 +03:00
nfs_vinvalbuf(vp, 0, sp->s_cred, curlwp, 1);
*ap->a_recycle = (np->n_flag & NREMOVED) != 0;
np->n_flag &=
(NMODIFIED | NFLUSHINPROG | NFLUSHWANT | NEOFVALID | NTRUNCDELAYED);
if (vp->v_type == VDIR && np->n_dircache)
nfs_invaldircache(vp,
NFS_INVALDIRCACHE_FORCE | NFS_INVALDIRCACHE_KEEPEOF);
VOP_UNLOCK(vp);
if (sp != NULL) {
workqueue_enqueue(nfs_sillyworkq, &sp->s_work, NULL);
}
1993-03-21 12:45:37 +03:00
return (0);
}
/*
* Reclaim an nfsnode so that it can be used for other purposes.
*/
1996-02-10 00:48:19 +03:00
int
nfs_reclaim(void *v)
1996-02-10 00:48:19 +03:00
{
struct vop_reclaim_args /* {
struct vnode *a_vp;
1996-02-10 00:48:19 +03:00
} */ *ap = v;
2000-03-30 16:51:13 +04:00
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
1993-03-21 12:45:37 +03:00
2008-01-17 13:27:43 +03:00
if (prtactive && vp->v_usecount > 1)
1993-03-21 12:45:37 +03:00
vprint("nfs_reclaim: pushing active", vp);
2014-05-30 12:47:45 +04:00
vcache_remove(vp->v_mount, np->n_fhp, np->n_fhsize);
/*
* Free up any directory cookie structures and
* large file handle structures that might be associated with
* this nfs node.
*/
if (vp->v_type == VDIR && np->n_dircache != NULL) {
nfs_invaldircache(vp, NFS_INVALDIRCACHE_FORCE);
hashdone(np->n_dircache, HASH_LIST, nfsdirhashmask);
}
2003-05-22 18:14:02 +04:00
KASSERT(np->n_dirgens == NULL);
if (np->n_fhsize > NFS_SMALLFH)
2008-01-02 22:26:45 +03:00
kmem_free(np->n_fhp, np->n_fhsize);
pool_put(&nfs_vattr_pool, np->n_vattr);
if (np->n_rcred)
2006-05-15 01:31:52 +04:00
kauth_cred_free(np->n_rcred);
if (np->n_wcred)
2006-05-15 01:31:52 +04:00
kauth_cred_free(np->n_wcred);
if (vp->v_type == VREG) {
mutex_destroy(&np->n_commitlock);
}
genfs_node_destroy(vp);
pool_put(&nfs_node_pool, np);
vp->v_data = NULL;
1993-03-21 12:45:37 +03:00
return (0);
}
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
void
nfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags)
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
{
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
*eobp = MAX(size, vp->v_size);
}
int
nfs_gop_alloc(struct vnode *vp, off_t off, off_t len, int flags,
kauth_cred_t cred)
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
{
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
return 0;
}
int
nfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, int flags)
{
int i;
mutex_enter(vp->v_interlock);
for (i = 0; i < npages; i++) {
pmap_page_protect(pgs[i], VM_PROT_READ);
}
mutex_exit(vp->v_interlock);
return genfs_gop_write(vp, pgs, npages, flags);
}
/*
* Remove a silly file that was rename'd earlier
*/
static void
nfs_sillyworker(struct work *work, void *arg)
{
struct sillyrename *sp;
int error;
sp = (struct sillyrename *)work;
error = vn_lock(sp->s_dvp, LK_EXCLUSIVE);
if (error || sp->s_dvp->v_data == NULL) {
/* XXX should recover */
printf("%s: vp=%p error=%d\n", __func__, sp->s_dvp, error);
if (error == 0) {
vput(sp->s_dvp);
} else {
vrele(sp->s_dvp);
}
} else {
nfs_removeit(sp);
vput(sp->s_dvp);
}
kauth_cred_free(sp->s_cred);
kmem_free(sp, sizeof(*sp));
}