2004-01-24 01:20:20 +03:00
|
|
|
/* $NetBSD: nfs_node.c,v 1.72 2004/01/23 22:20:20 wrstuden Exp $ */
|
1994-06-29 10:39:25 +04:00
|
|
|
|
1993-03-21 12:45:37 +03:00
|
|
|
/*
|
1994-06-08 15:33:09 +04:00
|
|
|
* 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.
|
2003-08-07 20:26:28 +04:00
|
|
|
* 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.
|
|
|
|
|
*
|
1996-02-18 14:53:36 +03:00
|
|
|
* @(#)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>
|
2004-01-24 01:20:20 +03:00
|
|
|
__KERNEL_RCSID(0, "$NetBSD: nfs_node.c,v 1.72 2004/01/23 22:20:20 wrstuden Exp $");
|
2001-11-10 13:59:08 +03:00
|
|
|
|
2000-09-19 21:04:50 +04:00
|
|
|
#include "opt_nfs.h"
|
1996-02-18 14:53:36 +03:00
|
|
|
|
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/malloc.h>
|
1998-09-01 07:11:36 +04:00
|
|
|
#include <sys/pool.h>
|
1997-07-08 03:34:55 +04:00
|
|
|
#include <sys/lock.h>
|
2001-12-06 04:26:36 +03:00
|
|
|
#include <sys/hash.h>
|
1993-03-21 12:45:37 +03:00
|
|
|
|
1994-06-08 15:33:09 +04:00
|
|
|
#include <nfs/rpcv2.h>
|
1996-02-18 14:53:36 +03:00
|
|
|
#include <nfs/nfsproto.h>
|
1993-12-18 03:40:47 +03:00
|
|
|
#include <nfs/nfs.h>
|
|
|
|
|
#include <nfs/nfsnode.h>
|
|
|
|
|
#include <nfs/nfsmount.h>
|
1994-06-08 15:33:09 +04:00
|
|
|
#include <nfs/nqnfs.h>
|
1996-02-10 00:48:19 +03:00
|
|
|
#include <nfs/nfs_var.h>
|
1993-03-21 12:45:37 +03:00
|
|
|
|
2002-12-02 02:02:10 +03:00
|
|
|
struct nfsnodehashhead *nfsnodehashtbl;
|
1994-08-19 02:47:43 +04:00
|
|
|
u_long nfsnodehash;
|
1997-07-08 03:34:55 +04:00
|
|
|
struct lock nfs_hashlock;
|
1993-03-21 12:45:37 +03:00
|
|
|
|
1998-09-01 07:11:36 +04:00
|
|
|
struct pool nfs_node_pool; /* memory pool for nfs nodes */
|
|
|
|
|
struct pool nfs_vattr_pool; /* memory pool for nfs vattrs */
|
|
|
|
|
|
2003-02-01 09:23:35 +03:00
|
|
|
MALLOC_DEFINE(M_NFSBIGFH, "NFS bigfh", "NFS big filehandle");
|
|
|
|
|
MALLOC_DEFINE(M_NFSNODE, "NFS node", "NFS vnode private part");
|
|
|
|
|
|
2001-02-07 15:40:43 +03:00
|
|
|
extern int prtactive;
|
|
|
|
|
|
2001-12-06 04:26:36 +03:00
|
|
|
#define nfs_hash(x,y) hash32_buf((x), (y), HASH32_BUF_INIT)
|
|
|
|
|
|
Add code to UBCify LFS. This is still behind "#ifdef LFS_UBC" for now
(there are still some details to work out) but expect that to go
away soon. To support these basic changes (creation of lfs_putpages,
lfs_gop_write, mods to lfs_balloc) several other changes were made, to
wit:
* Create a writer daemon kernel thread whose purpose is to handle page
writes for the pagedaemon, but which also takes over some of the
functions of lfs_check(). This thread is started the first time an
LFS is mounted.
* Add a "flags" parameter to GOP_SIZE. Current values are
GOP_SIZE_READ, meaning that the call should return the size of the
in-core version of the file, and GOP_SIZE_WRITE, meaning that it
should return the on-disk size. One of GOP_SIZE_READ or
GOP_SIZE_WRITE must be specified.
* Instead of using malloc(...M_WAITOK) for everything, reserve enough
resources to get by and use malloc(...M_NOWAIT), using the reserves if
necessary. Use the pool subsystem for structures small enough that
this is feasible. This also obsoletes LFS_THROTTLE.
And a few that are not strictly necessary:
* Moves the LFS inode extensions off onto a separately allocated
structure; getting closer to LFS as an LKM. "Welcome to 1.6O."
* Unified GOP_ALLOC between FFS and LFS.
* Update LFS copyright headers to correct values.
* Actually cast to unsigned in lfs_shellsort, like the comment says.
* Keep track of which segments were empty before the previous
checkpoint; any segments that pass two checkpoints both dirty and
empty can be summarily cleaned. Do this. Right now lfs_segclean
still works, but this should be turned into an effectless
compatibility syscall.
2003-02-18 02:48:08 +03:00
|
|
|
void nfs_gop_size(struct vnode *, off_t, off_t *, int);
|
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
|
|
|
int nfs_gop_alloc(struct vnode *, off_t, off_t, int, struct ucred *);
|
2002-03-17 02:05:25 +03:00
|
|
|
int nfs_gop_write(struct vnode *, struct vm_page **, int, int);
|
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
|
|
|
|
|
|
|
|
struct genfs_ops nfs_genfsops = {
|
|
|
|
|
nfs_gop_size,
|
|
|
|
|
nfs_gop_alloc,
|
2002-03-17 02:05:25 +03:00
|
|
|
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
|
|
|
/*
|
|
|
|
|
* Initialize hash links for nfsnodes
|
|
|
|
|
* and build nfsnode free list.
|
|
|
|
|
*/
|
1996-02-10 00:48:19 +03:00
|
|
|
void
|
1993-03-21 12:45:37 +03:00
|
|
|
nfs_nhinit()
|
|
|
|
|
{
|
|
|
|
|
|
2000-11-08 17:28:12 +03:00
|
|
|
nfsnodehashtbl = hashinit(desiredvnodes, HASH_LIST, M_NFSNODE,
|
|
|
|
|
M_WAITOK, &nfsnodehash);
|
1997-07-08 03:34:55 +04:00
|
|
|
lockinit(&nfs_hashlock, PINOD, "nfs_hashlock", 0, 0);
|
1998-09-01 07:11:36 +04:00
|
|
|
|
|
|
|
|
pool_init(&nfs_node_pool, sizeof(struct nfsnode), 0, 0, 0, "nfsnodepl",
|
2002-03-08 23:48:27 +03:00
|
|
|
&pool_allocator_nointr);
|
1998-09-01 07:11:36 +04:00
|
|
|
pool_init(&nfs_vattr_pool, sizeof(struct vattr), 0, 0, 0, "nfsvapl",
|
2002-03-08 23:48:27 +03:00
|
|
|
&pool_allocator_nointr);
|
1993-03-21 12:45:37 +03:00
|
|
|
}
|
|
|
|
|
|
2001-09-15 20:12:54 +04:00
|
|
|
/*
|
|
|
|
|
* Reinitialize inode hash table.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
void
|
|
|
|
|
nfs_nhreinit()
|
|
|
|
|
{
|
|
|
|
|
struct nfsnode *np;
|
|
|
|
|
struct nfsnodehashhead *oldhash, *hash;
|
|
|
|
|
u_long oldmask, mask, val;
|
|
|
|
|
int i;
|
|
|
|
|
|
|
|
|
|
hash = hashinit(desiredvnodes, HASH_LIST, M_NFSNODE, M_WAITOK,
|
|
|
|
|
&mask);
|
|
|
|
|
|
|
|
|
|
lockmgr(&nfs_hashlock, LK_EXCLUSIVE, NULL);
|
|
|
|
|
oldhash = nfsnodehashtbl;
|
|
|
|
|
oldmask = nfsnodehash;
|
|
|
|
|
nfsnodehashtbl = hash;
|
|
|
|
|
nfsnodehash = mask;
|
|
|
|
|
for (i = 0; i <= oldmask; i++) {
|
|
|
|
|
while ((np = LIST_FIRST(&oldhash[i])) != NULL) {
|
|
|
|
|
LIST_REMOVE(np, n_hash);
|
|
|
|
|
val = NFSNOHASH(nfs_hash(np->n_fhp, np->n_fhsize));
|
|
|
|
|
LIST_INSERT_HEAD(&hash[val], np, n_hash);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
lockmgr(&nfs_hashlock, LK_RELEASE, NULL);
|
|
|
|
|
hashdone(oldhash, M_NFSNODE);
|
|
|
|
|
}
|
|
|
|
|
|
2000-03-16 21:08:17 +03:00
|
|
|
/*
|
|
|
|
|
* Free resources previoslu allocated in nfs_nhinit().
|
|
|
|
|
*/
|
|
|
|
|
void
|
|
|
|
|
nfs_nhdone()
|
|
|
|
|
{
|
|
|
|
|
hashdone(nfsnodehashtbl, M_NFSNODE);
|
|
|
|
|
pool_destroy(&nfs_node_pool);
|
|
|
|
|
pool_destroy(&nfs_vattr_pool);
|
|
|
|
|
}
|
|
|
|
|
|
1993-03-21 12:45:37 +03:00
|
|
|
/*
|
|
|
|
|
* 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.
|
|
|
|
|
*/
|
1996-02-10 00:48:19 +03:00
|
|
|
int
|
2003-06-30 02:28:00 +04:00
|
|
|
nfs_nget(mntp, fhp, fhsize, npp)
|
1993-03-21 12:45:37 +03:00
|
|
|
struct mount *mntp;
|
2000-03-30 16:51:13 +04:00
|
|
|
nfsfh_t *fhp;
|
1996-02-18 14:53:36 +03:00
|
|
|
int fhsize;
|
1993-03-21 12:45:37 +03:00
|
|
|
struct nfsnode **npp;
|
|
|
|
|
{
|
2000-03-30 16:51:13 +04:00
|
|
|
struct nfsnode *np;
|
1994-08-19 02:47:43 +04:00
|
|
|
struct nfsnodehashhead *nhpp;
|
2000-03-30 16:51:13 +04:00
|
|
|
struct vnode *vp;
|
1993-03-21 12:45:37 +03:00
|
|
|
struct vnode *nvp;
|
|
|
|
|
int error;
|
|
|
|
|
|
2001-09-15 20:12:54 +04:00
|
|
|
nhpp = &nfsnodehashtbl[NFSNOHASH(nfs_hash(fhp, fhsize))];
|
1993-03-21 12:45:37 +03:00
|
|
|
loop:
|
2001-09-15 20:12:54 +04:00
|
|
|
LIST_FOREACH(np, nhpp, n_hash) {
|
1996-02-18 14:53:36 +03:00
|
|
|
if (mntp != NFSTOV(np)->v_mount || np->n_fhsize != fhsize ||
|
2000-11-27 11:39:39 +03:00
|
|
|
memcmp(fhp, np->n_fhp, fhsize))
|
1993-03-21 12:45:37 +03:00
|
|
|
continue;
|
|
|
|
|
vp = NFSTOV(np);
|
2003-06-29 22:43:21 +04:00
|
|
|
if (vget(vp, LK_EXCLUSIVE))
|
1993-03-21 12:45:37 +03:00
|
|
|
goto loop;
|
|
|
|
|
*npp = np;
|
|
|
|
|
return(0);
|
|
|
|
|
}
|
1998-03-01 05:20:01 +03:00
|
|
|
if (lockmgr(&nfs_hashlock, LK_EXCLUSIVE|LK_SLEEPFAIL, 0))
|
1997-07-08 03:34:55 +04:00
|
|
|
goto loop;
|
1996-02-10 00:48:19 +03:00
|
|
|
error = getnewvnode(VT_NFS, mntp, nfsv2_vnodeop_p, &nvp);
|
|
|
|
|
if (error) {
|
1993-03-21 12:45:37 +03:00
|
|
|
*npp = 0;
|
1998-03-01 05:20:01 +03:00
|
|
|
lockmgr(&nfs_hashlock, LK_RELEASE, 0);
|
1993-03-21 12:45:37 +03:00
|
|
|
return (error);
|
|
|
|
|
}
|
|
|
|
|
vp = nvp;
|
1998-09-01 07:11:36 +04:00
|
|
|
np = pool_get(&nfs_node_pool, PR_WAITOK);
|
2000-11-27 11:39:39 +03:00
|
|
|
memset(np, 0, sizeof *np);
|
2000-09-20 02:13:01 +04:00
|
|
|
lockinit(&np->n_commitlock, PINOD, "nfsclock", 0, 0);
|
1994-06-08 15:33:09 +04:00
|
|
|
vp->v_data = np;
|
1993-03-21 12:45:37 +03:00
|
|
|
np->n_vnode = vp;
|
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
|
|
|
genfs_node_init(vp, &nfs_genfsops);
|
2000-11-27 11:39:39 +03:00
|
|
|
|
1993-03-21 12:45:37 +03:00
|
|
|
/*
|
|
|
|
|
* Insert the nfsnode in the hash queue for its new file handle
|
|
|
|
|
*/
|
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
|
|
|
|
1994-08-19 02:47:43 +04:00
|
|
|
LIST_INSERT_HEAD(nhpp, np, n_hash);
|
1996-02-18 14:53:36 +03:00
|
|
|
if (fhsize > NFS_SMALLFH) {
|
2000-08-03 10:15:02 +04:00
|
|
|
np->n_fhp = malloc(fhsize, M_NFSBIGFH, M_WAITOK);
|
1996-02-18 14:53:36 +03:00
|
|
|
} else
|
|
|
|
|
np->n_fhp = &np->n_fh;
|
2000-11-27 11:39:39 +03:00
|
|
|
memcpy(np->n_fhp, fhp, fhsize);
|
1996-02-18 14:53:36 +03:00
|
|
|
np->n_fhsize = fhsize;
|
1999-11-30 02:34:00 +03:00
|
|
|
np->n_accstamp = -1;
|
1998-09-01 07:11:36 +04:00
|
|
|
np->n_vattr = pool_get(&nfs_vattr_pool, PR_WAITOK);
|
2003-12-08 00:15:46 +03:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Initalize read/write creds to useful values. VOP_OPEN will
|
|
|
|
|
* overwrite these.
|
|
|
|
|
*/
|
|
|
|
|
np->n_rcred = curproc->p_ucred;
|
|
|
|
|
crhold(np->n_rcred);
|
|
|
|
|
np->n_wcred = curproc->p_ucred;
|
|
|
|
|
crhold(np->n_wcred);
|
2001-09-15 20:12:54 +04:00
|
|
|
lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
|
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
|
|
|
lockmgr(&nfs_hashlock, LK_RELEASE, NULL);
|
2003-06-30 02:28:00 +04:00
|
|
|
error = VOP_GETATTR(vp, np->n_vattr, curproc->p_ucred, curproc);
|
2000-11-27 11:39:39 +03:00
|
|
|
if (error) {
|
2002-01-21 16:48:51 +03:00
|
|
|
vput(vp);
|
2000-11-27 11:39:39 +03:00
|
|
|
return error;
|
|
|
|
|
}
|
|
|
|
|
uvm_vnp_setsize(vp, np->n_vattr->va_size);
|
1993-03-21 12:45:37 +03:00
|
|
|
*npp = np;
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
1996-02-10 00:48:19 +03:00
|
|
|
int
|
|
|
|
|
nfs_inactive(v)
|
|
|
|
|
void *v;
|
|
|
|
|
{
|
1994-06-08 15:33:09 +04:00
|
|
|
struct vop_inactive_args /* {
|
|
|
|
|
struct vnode *a_vp;
|
2003-06-30 02:28:00 +04:00
|
|
|
struct proc *a_p;
|
1996-02-10 00:48:19 +03:00
|
|
|
} */ *ap = v;
|
2000-03-30 16:51:13 +04:00
|
|
|
struct nfsnode *np;
|
|
|
|
|
struct sillyrename *sp;
|
2003-06-30 02:28:00 +04:00
|
|
|
struct proc *p = ap->a_p;
|
2001-02-06 19:25:48 +03:00
|
|
|
struct vnode *vp = ap->a_vp;
|
2003-02-12 17:50:52 +03:00
|
|
|
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
|
2003-07-30 16:25:39 +04:00
|
|
|
boolean_t removed;
|
2004-01-24 01:20:20 +03:00
|
|
|
int err;
|
1993-03-21 12:45:37 +03:00
|
|
|
|
2001-02-06 19:25:48 +03:00
|
|
|
np = VTONFS(vp);
|
|
|
|
|
if (prtactive && vp->v_usecount != 0)
|
|
|
|
|
vprint("nfs_inactive: pushing active", vp);
|
|
|
|
|
if (vp->v_type != VDIR) {
|
1996-02-18 14:53:36 +03:00
|
|
|
sp = np->n_sillyrename;
|
1997-02-12 20:26:56 +03:00
|
|
|
np->n_sillyrename = (struct sillyrename *)0;
|
|
|
|
|
} else
|
2001-05-03 19:53:04 +04:00
|
|
|
sp = NULL;
|
|
|
|
|
if (sp != NULL)
|
2003-06-30 02:28:00 +04:00
|
|
|
nfs_vinvalbuf(vp, 0, sp->s_cred, p, 1);
|
2003-07-30 16:25:39 +04:00
|
|
|
removed = (np->n_flag & NREMOVED) != 0;
|
2001-05-03 19:53:04 +04:00
|
|
|
np->n_flag &= (NMODIFIED | NFLUSHINPROG | NFLUSHWANT | NQNFSEVICTED |
|
|
|
|
|
NQNFSNONCACHE | NQNFSWRITE);
|
|
|
|
|
VOP_UNLOCK(vp, 0);
|
|
|
|
|
if (sp != NULL) {
|
1997-02-22 05:45:48 +03:00
|
|
|
|
1993-03-21 12:45:37 +03:00
|
|
|
/*
|
|
|
|
|
* Remove the silly file that was rename'd earlier
|
2004-01-24 01:20:20 +03:00
|
|
|
*
|
|
|
|
|
* Just in case our thread also has the parent node locked,
|
|
|
|
|
* we let vn_lock() fail.
|
1993-03-21 12:45:37 +03:00
|
|
|
*/
|
2001-02-06 19:25:48 +03:00
|
|
|
|
2004-01-24 01:20:20 +03:00
|
|
|
err = vn_lock(sp->s_dvp, LK_EXCLUSIVE | LK_RETRY
|
|
|
|
|
| LK_RECURSEFAIL);
|
1994-06-08 15:33:09 +04:00
|
|
|
nfs_removeit(sp);
|
1993-03-21 12:45:37 +03:00
|
|
|
crfree(sp->s_cred);
|
2004-01-24 01:20:20 +03:00
|
|
|
if (err != EDEADLK)
|
|
|
|
|
vput(sp->s_dvp);
|
|
|
|
|
else
|
|
|
|
|
vrele(sp->s_dvp);
|
2000-11-27 11:39:39 +03:00
|
|
|
FREE(sp, M_NFSREQ);
|
1994-06-08 15:33:09 +04:00
|
|
|
}
|
2003-02-12 17:50:52 +03:00
|
|
|
|
2003-04-02 19:14:19 +04:00
|
|
|
if ((nmp->nm_flag & NFSMNT_NQNFS) && CIRCLEQ_NEXT(np, n_timer) != 0) {
|
2003-02-12 17:50:52 +03:00
|
|
|
CIRCLEQ_REMOVE(&nmp->nm_timerhead, np, n_timer);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (vp->v_type == VDIR && np->n_dircache)
|
|
|
|
|
nfs_invaldircache(vp, 1);
|
|
|
|
|
|
2003-07-30 16:25:39 +04:00
|
|
|
if (removed)
|
|
|
|
|
vrecycle(vp, NULL, p);
|
|
|
|
|
|
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(v)
|
|
|
|
|
void *v;
|
|
|
|
|
{
|
1994-06-08 15:33:09 +04:00
|
|
|
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
|
|
|
|
|
|
|
|
if (prtactive && vp->v_usecount != 0)
|
|
|
|
|
vprint("nfs_reclaim: pushing active", vp);
|
1996-02-18 14:53:36 +03:00
|
|
|
|
2003-02-15 21:00:25 +03:00
|
|
|
LIST_REMOVE(np, n_hash);
|
|
|
|
|
|
1996-02-18 14:53:36 +03:00
|
|
|
/*
|
|
|
|
|
* Free up any directory cookie structures and
|
|
|
|
|
* large file handle structures that might be associated with
|
|
|
|
|
* this nfs node.
|
|
|
|
|
*/
|
2003-02-12 17:50:52 +03:00
|
|
|
if (vp->v_type == VDIR && np->n_dircache)
|
2003-05-07 17:16:35 +04:00
|
|
|
hashdone(np->n_dircache, M_NFSDIROFF);
|
2003-05-22 18:14:02 +04:00
|
|
|
KASSERT(np->n_dirgens == NULL);
|
2003-02-12 17:50:52 +03:00
|
|
|
|
|
|
|
|
if (np->n_fhsize > NFS_SMALLFH)
|
2000-11-27 11:39:39 +03:00
|
|
|
free(np->n_fhp, M_NFSBIGFH);
|
1996-02-18 14:53:36 +03:00
|
|
|
|
1998-09-01 07:11:36 +04:00
|
|
|
pool_put(&nfs_vattr_pool, np->n_vattr);
|
2003-02-12 17:50:52 +03:00
|
|
|
if (np->n_rcred)
|
2000-11-27 11:39:39 +03:00
|
|
|
crfree(np->n_rcred);
|
2003-02-12 17:50:52 +03:00
|
|
|
|
|
|
|
|
if (np->n_wcred)
|
2000-11-27 11:39:39 +03:00
|
|
|
crfree(np->n_wcred);
|
2003-02-12 17:50:52 +03:00
|
|
|
|
1993-03-21 12:45:37 +03:00
|
|
|
cache_purge(vp);
|
1998-09-01 07:11:36 +04:00
|
|
|
pool_put(&nfs_node_pool, vp->v_data);
|
2000-11-27 11:39:39 +03:00
|
|
|
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
|
Add code to UBCify LFS. This is still behind "#ifdef LFS_UBC" for now
(there are still some details to work out) but expect that to go
away soon. To support these basic changes (creation of lfs_putpages,
lfs_gop_write, mods to lfs_balloc) several other changes were made, to
wit:
* Create a writer daemon kernel thread whose purpose is to handle page
writes for the pagedaemon, but which also takes over some of the
functions of lfs_check(). This thread is started the first time an
LFS is mounted.
* Add a "flags" parameter to GOP_SIZE. Current values are
GOP_SIZE_READ, meaning that the call should return the size of the
in-core version of the file, and GOP_SIZE_WRITE, meaning that it
should return the on-disk size. One of GOP_SIZE_READ or
GOP_SIZE_WRITE must be specified.
* Instead of using malloc(...M_WAITOK) for everything, reserve enough
resources to get by and use malloc(...M_NOWAIT), using the reserves if
necessary. Use the pool subsystem for structures small enough that
this is feasible. This also obsoletes LFS_THROTTLE.
And a few that are not strictly necessary:
* Moves the LFS inode extensions off onto a separately allocated
structure; getting closer to LFS as an LKM. "Welcome to 1.6O."
* Unified GOP_ALLOC between FFS and LFS.
* Update LFS copyright headers to correct values.
* Actually cast to unsigned in lfs_shellsort, like the comment says.
* Keep track of which segments were empty before the previous
checkpoint; any segments that pass two checkpoints both dirty and
empty can be summarily cleaned. Do this. Right now lfs_segclean
still works, but this should be turned into an effectless
compatibility syscall.
2003-02-18 02:48:08 +03:00
|
|
|
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
|
|
|
{
|
Add code to UBCify LFS. This is still behind "#ifdef LFS_UBC" for now
(there are still some details to work out) but expect that to go
away soon. To support these basic changes (creation of lfs_putpages,
lfs_gop_write, mods to lfs_balloc) several other changes were made, to
wit:
* Create a writer daemon kernel thread whose purpose is to handle page
writes for the pagedaemon, but which also takes over some of the
functions of lfs_check(). This thread is started the first time an
LFS is mounted.
* Add a "flags" parameter to GOP_SIZE. Current values are
GOP_SIZE_READ, meaning that the call should return the size of the
in-core version of the file, and GOP_SIZE_WRITE, meaning that it
should return the on-disk size. One of GOP_SIZE_READ or
GOP_SIZE_WRITE must be specified.
* Instead of using malloc(...M_WAITOK) for everything, reserve enough
resources to get by and use malloc(...M_NOWAIT), using the reserves if
necessary. Use the pool subsystem for structures small enough that
this is feasible. This also obsoletes LFS_THROTTLE.
And a few that are not strictly necessary:
* Moves the LFS inode extensions off onto a separately allocated
structure; getting closer to LFS as an LKM. "Welcome to 1.6O."
* Unified GOP_ALLOC between FFS and LFS.
* Update LFS copyright headers to correct values.
* Actually cast to unsigned in lfs_shellsort, like the comment says.
* Keep track of which segments were empty before the previous
checkpoint; any segments that pass two checkpoints both dirty and
empty can be summarily cleaned. Do this. Right now lfs_segclean
still works, but this should be turned into an effectless
compatibility syscall.
2003-02-18 02:48:08 +03:00
|
|
|
KASSERT(flags & (GOP_SIZE_READ | GOP_SIZE_WRITE));
|
|
|
|
|
KASSERT((flags & (GOP_SIZE_READ | GOP_SIZE_WRITE))
|
|
|
|
|
!= (GOP_SIZE_READ | GOP_SIZE_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
|
|
|
*eobp = MAX(size, vp->v_size);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
int
|
|
|
|
|
nfs_gop_alloc(struct vnode *vp, off_t off, off_t len, int flags,
|
|
|
|
|
struct ucred *cred)
|
|
|
|
|
{
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
2002-03-17 02:05:25 +03:00
|
|
|
|
|
|
|
|
int
|
|
|
|
|
nfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, int flags)
|
|
|
|
|
{
|
|
|
|
|
int i;
|
|
|
|
|
|
|
|
|
|
for (i = 0; i < npages; i++) {
|
|
|
|
|
pmap_page_protect(pgs[i], VM_PROT_READ);
|
|
|
|
|
}
|
|
|
|
|
return genfs_gop_write(vp, pgs, npages, flags);
|
|
|
|
|
}
|