Aren/NetBSD
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Initial integration of the Unified Buffer Cache project.

Browse Source
This commit is contained in:
chs 2000-11-27 08:39:39 +00:00
parent c29a1b4461
commit aeda8d3b77
80 changed files with 4961 additions and 3435 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
sys/adosfs/advnops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: advnops.c,v 1.54 2000/08/03 00:54:23 thorpej Exp $ */
/* $NetBSD: advnops.c,v 1.55 2000/11/27 08:39:39 chs Exp $ */
/*
* Copyright (c) 1994 Christian E. Hopps
@ -143,7 +143,9 @@ struct vnodeopv_entry_desc adosfs_vnodeop_entries[] = {
{ &vop_truncate_desc, adosfs_truncate }, /* truncate */
{ &vop_update_desc, adosfs_update }, /* update */
{ &vop_bwrite_desc, adosfs_bwrite }, /* bwrite */
{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
{ &vop_getpages_desc, genfs_getpages }, /* getpages */
{ &vop_size_desc, genfs_size }, /* size */
{ NULL, NULL }
};
struct vnodeopv_desc adosfs_vnodeop_opv_desc =
@ -226,6 +228,7 @@ adosfs_read(v)
int a_ioflag;
struct ucred *a_cred;
} */ *sp = v;
struct vnode *vp = sp->a_vp;
struct adosfsmount *amp;
struct anode *ap;
struct uio *uio;
@ -265,6 +268,28 @@ adosfs_read(v)
/*
* taken from ufs_read()
*/
if (vp->v_type == VREG) {
error = 0;
while (uio->uio_resid > 0) {
void *win;
vsize_t bytelen = min(ap->fsize - uio->uio_offset,
uio->uio_resid);
if (bytelen == 0) {
break;
}
win = ubc_alloc(&vp->v_uvm.u_obj, uio->uio_offset,
&bytelen, UBC_READ);
error = uiomove(win, bytelen, uio);
ubc_release(win, 0);
if (error) {
break;
}
}
goto out;
}
do {
/*
* we are only supporting ADosFFS currently
@ -326,6 +351,8 @@ adosfs_read(v)
amp->bsize - amp->dbsize, (int)n, uio);
brelse(bp);
} while (error == 0 && uio->uio_resid > 0 && n != 0);
out:
reterr:
#ifdef ADOSFS_DIAGNOSTIC
printf(" %d)", error);

3
sys/arch/atari/dev/md_root.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: md_root.c,v 1.14 2000/01/21 23:29:02 thorpej Exp $ */
/* $NetBSD: md_root.c,v 1.15 2000/11/27 08:39:40 chs Exp $ */
/*
* Copyright (c) 1996 Leo Weppelman.
@ -159,7 +159,6 @@ struct proc *proc;
* Initialize our buffer header:
*/
memset(&buf, 0, sizeof(buf));
buf.b_rcred = buf.b_wcred = proc->p_ucred;
buf.b_vnbufs.le_next = NOLIST;
buf.b_flags = B_BUSY;
buf.b_dev = ld_dev;

12
sys/coda/coda_subr.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: coda_subr.c,v 1.9 2000/03/30 11:24:16 augustss Exp $ */
/* $NetBSD: coda_subr.c,v 1.10 2000/11/27 08:39:40 chs Exp $ */
/*
*
@ -227,7 +227,7 @@ coda_kill(whoIam, dcstat)
#endif
count++;
CODADEBUG(CODA_FLUSH,
myprintf(("Live cnode fid %lx.%lx.%lx flags %d count %ld\n",
myprintf(("Live cnode fid %lx.%lx.%lx flags %d count %d\n",
(cp->c_fid).Volume,
(cp->c_fid).Vnode,
(cp->c_fid).Unique,
@ -277,7 +277,7 @@ coda_testflush(void)
for (cp = coda_cache[hash];
cp != NULL;
cp = CNODE_NEXT(cp)) {
myprintf(("Live cnode fid %lx.%lx.%lx count %ld\n",
myprintf(("Live cnode fid %lx.%lx.%lx count %d\n",
(cp->c_fid).Volume,(cp->c_fid).Vnode,
(cp->c_fid).Unique, CTOV(cp)->v_usecount));
}
@ -424,7 +424,7 @@ int handleDownCall(opcode, out)
if (CTOV(cp)->v_flag & VTEXT)
error = coda_vmflush(cp);
CODADEBUG(CODA_ZAPFILE, myprintf(("zapfile: fid = (%lx.%lx.%lx),
refcnt = %ld, error = %d\n",
refcnt = %d, error = %d\n",
cp->c_fid.Volume,
cp->c_fid.Vnode,
cp->c_fid.Unique,
@ -452,7 +452,7 @@ int handleDownCall(opcode, out)
coda_nc_zapParentfid(&out->coda_zapdir.CodaFid, IS_DOWNCALL);
CODADEBUG(CODA_ZAPDIR, myprintf(("zapdir: fid = (%lx.%lx.%lx),
refcnt = %ld\n",cp->c_fid.Volume,
refcnt = %d\n",cp->c_fid.Volume,
cp->c_fid.Vnode,
cp->c_fid.Unique,
CTOV(cp)->v_usecount - 1)););
@ -486,7 +486,7 @@ int handleDownCall(opcode, out)
error = coda_vmflush(cp);
}
CODADEBUG(CODA_PURGEFID, myprintf(("purgefid: fid = (%lx.%lx.%lx), refcnt = %ld, error = %d\n",
CODADEBUG(CODA_PURGEFID, myprintf(("purgefid: fid = (%lx.%lx.%lx), refcnt = %d, error = %d\n",
cp->c_fid.Volume, cp->c_fid.Vnode,
cp->c_fid.Unique,
CTOV(cp)->v_usecount - 1, error)););

10
sys/coda/coda_vnops.c
View File

@ -6,7 +6,7 @@ mkdir
rmdir
symlink
*/
/* $NetBSD: coda_vnops.c,v 1.21 2000/09/19 22:00:01 fvdl Exp $ */
/* $NetBSD: coda_vnops.c,v 1.22 2000/11/27 08:39:40 chs Exp $ */
/*
*
@ -453,7 +453,7 @@ printf("coda_rdwr: Internally Opening %p\n", vp);
}
/* Have UFS handle the call. */
CODADEBUG(CODA_RDWR, myprintf(("indirect rdwr: fid = (%lx.%lx.%lx), refcnt = %ld\n",
CODADEBUG(CODA_RDWR, myprintf(("indirect rdwr: fid = (%lx.%lx.%lx), refcnt = %d\n",
cp->c_fid.Volume, cp->c_fid.Vnode,
cp->c_fid.Unique, CTOV(cp)->v_usecount)); )
@ -873,9 +873,9 @@ coda_inactive(v)
if (IS_UNMOUNTING(cp)) {
#ifdef DEBUG
printf("coda_inactive: IS_UNMOUNTING use %ld: vp %p, cp %p\n", vp->v_usecount, vp, cp);
printf("coda_inactive: IS_UNMOUNTING use %d: vp %p, cp %p\n", vp->v_usecount, vp, cp);
if (cp->c_ovp != NULL)
printf("coda_inactive: cp->ovp != NULL use %ld: vp %p, cp %p\n",
printf("coda_inactive: cp->ovp != NULL use %d: vp %p, cp %p\n",
vp->v_usecount, vp, cp);
#endif
lockmgr(&vp->v_lock, LK_RELEASE, &vp->v_interlock);
@ -1727,7 +1727,7 @@ printf("coda_readdir: Internally Opening %p\n", vp);
}
/* Have UFS handle the call. */
CODADEBUG(CODA_READDIR, myprintf(("indirect readdir: fid = (%lx.%lx.%lx), refcnt = %ld\n",cp->c_fid.Volume, cp->c_fid.Vnode, cp->c_fid.Unique, vp->v_usecount)); )
CODADEBUG(CODA_READDIR, myprintf(("indirect readdir: fid = (%lx.%lx.%lx), refcnt = %d\n",cp->c_fid.Volume, cp->c_fid.Vnode, cp->c_fid.Unique, vp->v_usecount)); )
error = VOP_READDIR(cp->c_ovp, uiop, cred, eofflag, cookies,
ncookies);
if (error)

3
sys/conf/files
View File

@ -1,4 +1,4 @@
# $NetBSD: files,v 1.404 2000/11/26 17:44:02 ad Exp $
# $NetBSD: files,v 1.405 2000/11/27 08:39:41 chs Exp $
# @(#)files.newconf 7.5 (Berkeley) 5/10/93
@ -1060,6 +1060,7 @@ file ufs/ufs/ufs_vnops.c ffs | lfs | mfs | ext2fs
file uvm/uvm_amap.c
file uvm/uvm_anon.c
file uvm/uvm_aobj.c
file uvm/uvm_bio.c
file uvm/uvm_device.c
file uvm/uvm_fault.c
file uvm/uvm_glue.c

24
sys/dev/vnd.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: vnd.c,v 1.68 2000/09/12 08:03:24 enami Exp $ */
/* $NetBSD: vnd.c,v 1.69 2000/11/27 08:39:41 chs Exp $ */
/*-
* Copyright (c) 1996, 1997, 1998 The NetBSD Foundation, Inc.
@ -445,29 +445,7 @@ vndstrategy(bp)
nbp->vb_buf.b_proc = bp->b_proc;
nbp->vb_buf.b_iodone = vndiodone;
nbp->vb_buf.b_vp = NULLVP;
nbp->vb_buf.b_rcred = vnd->sc_cred; /* XXX crdup? */
nbp->vb_buf.b_wcred = vnd->sc_cred; /* XXX crdup? */
LIST_INIT(&nbp->vb_buf.b_dep);
if (bp->b_dirtyend == 0) {
nbp->vb_buf.b_dirtyoff = 0;
nbp->vb_buf.b_dirtyend = sz;
} else {
nbp->vb_buf.b_dirtyoff =
max(0, bp->b_dirtyoff - (bp->b_bcount - resid));
nbp->vb_buf.b_dirtyend =
min(sz,
max(0, bp->b_dirtyend - (bp->b_bcount-resid)));
}
if (bp->b_validend == 0) {
nbp->vb_buf.b_validoff = 0;
nbp->vb_buf.b_validend = sz;
} else {
nbp->vb_buf.b_validoff =
max(0, bp->b_validoff - (bp->b_bcount - resid));
nbp->vb_buf.b_validend =
min(sz,
max(0, bp->b_validend - (bp->b_bcount-resid)));
}
nbp->vb_xfer = vnx;

5
sys/filecorefs/filecore_vfsops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: filecore_vfsops.c,v 1.11 2000/03/16 18:08:22 jdolecek Exp $ */
/* $NetBSD: filecore_vfsops.c,v 1.12 2000/11/27 08:39:41 chs Exp $ */
/*-
* Copyright (c) 1998 Andrew McMurry
@ -324,6 +324,9 @@ filecore_mountfs(devvp, mp, p, argp)
mp->mnt_stat.f_fsid.val[1] = makefstype(MOUNT_FILECORE);
mp->mnt_maxsymlinklen = 0;
mp->mnt_flag |= MNT_LOCAL;
mp->mnt_dev_bshift = fcdr->log2secsize;
mp->mnt_fs_bshift = fcmp->log2bsize;
fcmp->fc_mountp = mp;
fcmp->fc_dev = dev;
fcmp->fc_devvp = devvp;

30
sys/filecorefs/filecore_vnops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: filecore_vnops.c,v 1.9 2000/08/03 03:38:39 thorpej Exp $ */
/* $NetBSD: filecore_vnops.c,v 1.10 2000/11/27 08:39:42 chs Exp $ */
/*-
* Copyright (c) 1998 Andrew McMurry
@ -162,6 +162,28 @@ filecore_read(v)
return (EINVAL);
ip->i_flag |= IN_ACCESS;
fcmp = ip->i_mnt;
if (vp->v_type == VREG) {
error = 0;
while (uio->uio_resid > 0) {
void *win;
vsize_t bytelen = min(ip->i_size - uio->uio_offset,
uio->uio_resid);
if (bytelen == 0) {
break;
}
win = ubc_alloc(&vp->v_uvm.u_obj, uio->uio_offset,
&bytelen, UBC_READ);
error = uiomove(win, bytelen, uio);
ubc_release(win, 0);
if (error) {
break;
}
}
goto out;
}
do {
lbn = lblkno(fcmp, uio->uio_offset);
on = blkoff(fcmp, uio->uio_offset);
@ -213,6 +235,8 @@ filecore_read(v)
#endif
brelse(bp);
} while (error == 0 && uio->uio_resid > 0 && n != 0);
out:
return (error);
}
@ -571,7 +595,9 @@ struct vnodeopv_entry_desc filecore_vnodeop_entries[] = {
{ &vop_truncate_desc, filecore_truncate }, /* truncate */
{ &vop_update_desc, filecore_update }, /* update */
{ &vop_bwrite_desc, vn_bwrite }, /* bwrite */
{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
{ &vop_getpages_desc, genfs_getpages }, /* getpages */
{ &vop_size_desc, genfs_size }, /* size */
{ NULL, NULL }
};
struct vnodeopv_desc filecore_vnodeop_opv_desc =
{ &filecore_vnodeop_p, filecore_vnodeop_entries };

4
sys/isofs/cd9660/cd9660_vfsops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: cd9660_vfsops.c,v 1.49 2000/07/15 21:40:44 jdolecek Exp $ */
/* $NetBSD: cd9660_vfsops.c,v 1.50 2000/11/27 08:39:42 chs Exp $ */
/*-
* Copyright (c) 1994
@ -399,6 +399,8 @@ iso_mountfs(devvp, mp, p, argp)
mp->mnt_stat.f_fsid.val[1] = makefstype(MOUNT_CD9660);
mp->mnt_maxsymlinklen = 0;
mp->mnt_flag |= MNT_LOCAL;
mp->mnt_dev_bshift = iso_bsize;
mp->mnt_fs_bshift = isomp->im_bshift;
isomp->im_mountp = mp;
isomp->im_dev = dev;
isomp->im_devvp = devvp;

32
sys/isofs/cd9660/cd9660_vnops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: cd9660_vnops.c,v 1.60 2000/11/14 22:26:32 thorpej Exp $ */
/* $NetBSD: cd9660_vnops.c,v 1.61 2000/11/27 08:39:42 chs Exp $ */
/*-
* Copyright (c) 1994
@ -278,6 +278,26 @@ cd9660_read(v)
return (EINVAL);
ip->i_flag |= IN_ACCESS;
imp = ip->i_mnt;
if (vp->v_type == VREG) {
error = 0;
while (uio->uio_resid > 0) {
void *win;
vsize_t bytelen = min(ip->i_size - uio->uio_offset,
uio->uio_resid);
if (bytelen == 0)
break;
win = ubc_alloc(&vp->v_uvm.u_obj, uio->uio_offset,
&bytelen, UBC_READ);
error = uiomove(win, bytelen, uio);
ubc_release(win, 0);
if (error)
break;
}
goto out;
}
do {
lbn = lblkno(imp, uio->uio_offset);
on = blkoff(imp, uio->uio_offset);
@ -315,6 +335,8 @@ cd9660_read(v)
error = uiomove(bp->b_data + on, (int)n, uio);
brelse(bp);
} while (error == 0 && uio->uio_resid > 0 && n != 0);
out:
return (error);
}
@ -955,7 +977,9 @@ struct vnodeopv_entry_desc cd9660_vnodeop_entries[] = {
{ &vop_truncate_desc, cd9660_truncate }, /* truncate */
{ &vop_update_desc, cd9660_update }, /* update */
{ &vop_bwrite_desc, vn_bwrite }, /* bwrite */
{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
{ &vop_getpages_desc, genfs_getpages }, /* getpages */
{ &vop_size_desc, genfs_size }, /* size */
{ NULL, NULL }
};
struct vnodeopv_desc cd9660_vnodeop_opv_desc =
{ &cd9660_vnodeop_p, cd9660_vnodeop_entries };
@ -1009,7 +1033,7 @@ struct vnodeopv_entry_desc cd9660_specop_entries[] = {
{ &vop_truncate_desc, spec_truncate }, /* truncate */
{ &vop_update_desc, cd9660_update }, /* update */
{ &vop_bwrite_desc, vn_bwrite }, /* bwrite */
{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
{ NULL, NULL }
};
struct vnodeopv_desc cd9660_specop_opv_desc =
{ &cd9660_specop_p, cd9660_specop_entries };
@ -1060,7 +1084,7 @@ struct vnodeopv_entry_desc cd9660_fifoop_entries[] = {
{ &vop_truncate_desc, fifo_truncate }, /* truncate */
{ &vop_update_desc, cd9660_update }, /* update */
{ &vop_bwrite_desc, vn_bwrite }, /* bwrite */
{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
{ NULL, NULL }
};
struct vnodeopv_desc cd9660_fifoop_opv_desc =
{ &cd9660_fifoop_p, cd9660_fifoop_entries };

3
sys/kern/exec_subr.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: exec_subr.c,v 1.25 2000/11/05 22:41:35 tv Exp $ */
/* $NetBSD: exec_subr.c,v 1.26 2000/11/27 08:39:42 chs Exp $ */
/*
* Copyright (c) 1993, 1994, 1996 Christopher G. Demetriou
@ -163,6 +163,7 @@ vmcmd_map_pagedvn(struct proc *p, struct exec_vmcmd *cmd)
uobj = uvn_attach((void *) cmd->ev_vp, VM_PROT_READ|VM_PROT_EXECUTE);
if (uobj == NULL)
return(ENOMEM);
VREF(cmd->ev_vp);
/*
* do the map

8
sys/kern/init_main.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: init_main.c,v 1.184 2000/11/21 00:37:56 jdolecek Exp $ */
/* $NetBSD: init_main.c,v 1.185 2000/11/27 08:39:43 chs Exp $ */
/*
* Copyright (c) 1995 Christopher G. Demetriou. All rights reserved.
@ -323,6 +323,8 @@ main(void)
/* Configure the system hardware. This will enable interrupts. */
configure();
ubc_init(); /* must be after autoconfig */
/* Lock the kernel on behalf of proc0. */
KERNEL_PROC_LOCK(p);
@ -472,6 +474,10 @@ main(void)
if (kthread_create1(sched_sync, NULL, NULL, "ioflush"))
panic("fork syncer");
/* Create the aiodone daemon kernel thread. */
if (kthread_create1(uvm_aiodone_daemon, NULL, NULL, "aiodoned"))
panic("fork aiodoned");
#if defined(MULTIPROCESSOR)
/* Boot the secondary processors. */
cpu_boot_secondary_processors();

3
sys/kern/kern_exec.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: kern_exec.c,v 1.124 2000/11/21 00:37:56 jdolecek Exp $ */
/* $NetBSD: kern_exec.c,v 1.125 2000/11/27 08:39:43 chs Exp $ */
/*-
* Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou
@ -152,6 +152,7 @@ check_exec(struct proc *p, struct exec_package *epp)
VOP_UNLOCK(vp, 0);
/* now we have the file, get the exec header */
uvn_attach(vp, VM_PROT_READ);
error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
UIO_SYSSPACE, 0, p->p_ucred, &resid, p);
if (error)

5
sys/kern/kern_physio.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: kern_physio.c,v 1.44 2000/09/29 13:27:12 ad Exp $ */
/* $NetBSD: kern_physio.c,v 1.45 2000/11/27 08:39:43 chs Exp $ */
/*-
* Copyright (c) 1994 Christopher G. Demetriou
@ -290,8 +290,7 @@ getphysbuf()
splx(s);
memset(bp, 0, sizeof(*bp));
/* XXXCDC: are the following two lines necessary? */
bp->b_rcred = bp->b_wcred = NOCRED;
/* XXXCDC: is the following line necessary? */
bp->b_vnbufs.le_next = NOLIST;
return(bp);

89
sys/kern/vfs_bio.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: vfs_bio.c,v 1.72 2000/11/18 15:58:13 simonb Exp $ */
/* $NetBSD: vfs_bio.c,v 1.73 2000/11/27 08:39:43 chs Exp $ */
/*-
* Copyright (c) 1994 Christopher G. Demetriou
@ -59,7 +59,7 @@
#include <sys/resourcevar.h>
#include <sys/conf.h>
#include <uvm/uvm_extern.h>
#include <uvm/uvm.h>
#include <miscfs/specfs/specdev.h>
@ -72,7 +72,7 @@
* Definitions for the buffer hash lists.
*/
#define BUFHASH(dvp, lbn) \
(&bufhashtbl[((long)(dvp) / sizeof(*(dvp)) + (int)(lbn)) & bufhash])
(&bufhashtbl[(((long)(dvp) >> 8) + (int)(lbn)) & bufhash])
LIST_HEAD(bufhashhdr, buf) *bufhashtbl, invalhash;
u_long bufhash;
struct bio_ops bioops; /* I/O operation notification */
@ -134,7 +134,6 @@ bremfree(bp)
panic("bremfree: lost tail");
}
TAILQ_REMOVE(dp, bp, b_freelist);
splx(s);
}
@ -166,8 +165,6 @@ bufinit()
bp = &buf[i];
memset((char *)bp, 0, sizeof(*bp));
bp->b_dev = NODEV;
bp->b_rcred = NOCRED;
bp->b_wcred = NOCRED;
bp->b_vnbufs.le_next = NOLIST;
LIST_INIT(&bp->b_dep);
bp->b_data = buffers + i * MAXBSIZE;
@ -201,12 +198,8 @@ bio_doread(vp, blkno, size, cred, async)
* Therefore, it's valid if it's I/O has completed or been delayed.
*/
if (!ISSET(bp->b_flags, (B_DONE | B_DELWRI))) {
/* Start I/O for the buffer (keeping credentials). */
/* Start I/O for the buffer. */
SET(bp->b_flags, B_READ | async);
if (cred != NOCRED && bp->b_rcred == NOCRED) {
crhold(cred);
bp->b_rcred = cred;
}
VOP_STRATEGY(bp);
/* Pay for the read. */
@ -375,7 +368,6 @@ bwrite(bp)
bp->b_vp->v_numoutput++;
splx(s);
SET(bp->b_flags, B_WRITEINPROG);
VOP_STRATEGY(bp);
if (sync) {
@ -509,6 +501,8 @@ brelse(bp)
struct bqueues *bufq;
int s;
KASSERT(ISSET(bp->b_flags, B_BUSY));
/* Wake up any processes waiting for any buffer to become free. */
if (needbuffer) {
needbuffer = 0;
@ -602,6 +596,7 @@ brelse(bp)
already_queued:
/* Unlock the buffer. */
CLR(bp->b_flags, B_AGE|B_ASYNC|B_BUSY|B_NOCACHE|B_ORDERED);
SET(bp->b_flags, B_CACHE);
/* Allow disk interrupts. */
splx(s);
@ -630,7 +625,7 @@ incore(vp, blkno)
return (bp);
}
return (0);
return (NULL);
}
/*
@ -647,56 +642,38 @@ getblk(vp, blkno, size, slpflag, slptimeo)
daddr_t blkno;
int size, slpflag, slptimeo;
{
struct bufhashhdr *bh;
struct buf *bp;
int s, err;
/*
* XXX
* The following is an inlined version of 'incore()', but with
* the 'invalid' test moved to after the 'busy' test. It's
* necessary because there are some cases in which the NFS
* code sets B_INVAL prior to writing data to the server, but
* in which the buffers actually contain valid data. In this
* case, we can't allow the system to allocate a new buffer for
* the block until the write is finished.
*/
bh = BUFHASH(vp, blkno);
start:
bp = bh->lh_first;
for (; bp != NULL; bp = bp->b_hash.le_next) {
if (bp->b_lblkno != blkno || bp->b_vp != vp)
continue;
bp = incore(vp, blkno);
if (bp != NULL) {
s = splbio();
if (ISSET(bp->b_flags, B_BUSY)) {
if (curproc == uvm.pagedaemon_proc) {
splx(s);
return NULL;
}
SET(bp->b_flags, B_WANTED);
err = tsleep(bp, slpflag | (PRIBIO + 1), "getblk",
slptimeo);
slptimeo);
splx(s);
if (err)
return (NULL);
goto start;
}
if (!ISSET(bp->b_flags, B_INVAL)) {
#ifdef DIAGNOSTIC
if (ISSET(bp->b_flags, B_DONE|B_DELWRI) &&
bp->b_bcount < size)
panic("getblk: block size invariant failed");
if (ISSET(bp->b_flags, B_DONE|B_DELWRI) && bp->b_bcount < size)
panic("getblk: block size invariant failed");
#endif
SET(bp->b_flags, B_BUSY);
bremfree(bp);
splx(s);
break;
}
SET(bp->b_flags, B_BUSY);
bremfree(bp);
splx(s);
}
if (bp == NULL) {
} else {
if ((bp = getnewbuf(slpflag, slptimeo)) == NULL)
goto start;
binshash(bp, bh);
binshash(bp, BUFHASH(vp, blkno));
bp->b_blkno = bp->b_lblkno = bp->b_rawblkno = blkno;
s = splbio();
bgetvp(vp, bp);
@ -720,7 +697,6 @@ geteblk(size)
SET(bp->b_flags, B_INVAL);
binshash(bp, &invalhash);
allocbuf(bp, size);
return (bp);
}
@ -737,9 +713,9 @@ allocbuf(bp, size)
struct buf *bp;
int size;
{
struct buf *nbp;
vsize_t desired_size;
int s;
struct buf *nbp;
vsize_t desired_size;
int s;
desired_size = round_page((vsize_t)size);
if (desired_size > MAXBSIZE)
@ -759,6 +735,7 @@ allocbuf(bp, size)
/* find a buffer */
while ((nbp = getnewbuf(0, 0)) == NULL)
;
SET(nbp->b_flags, B_INVAL);
binshash(nbp, &invalhash);
@ -836,7 +813,7 @@ start:
needbuffer = 1;
tsleep(&needbuffer, slpflag|(PRIBIO+1), "getnewbuf", slptimeo);
splx(s);
return (0);
return (NULL);
}
if (ISSET(bp->b_flags, B_VFLUSH)) {
@ -882,18 +859,6 @@ start:
bp->b_error = 0;
bp->b_resid = 0;
bp->b_bcount = 0;
bp->b_dirtyoff = bp->b_dirtyend = 0;
bp->b_validoff = bp->b_validend = 0;
/* nuke any credentials we were holding */
if (bp->b_rcred != NOCRED) {
crfree(bp->b_rcred);
bp->b_rcred = NOCRED;
}
if (bp->b_wcred != NOCRED) {
crfree(bp->b_wcred);
bp->b_wcred = NOCRED;
}
bremhash(bp);
return (bp);

134
sys/kern/vfs_subr.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: vfs_subr.c,v 1.141 2000/11/24 03:59:09 chs Exp $ */
/* $NetBSD: vfs_subr.c,v 1.142 2000/11/27 08:39:44 chs Exp $ */
/*-
* Copyright (c) 1997, 1998 The NetBSD Foundation, Inc.
@ -413,6 +413,8 @@ getnewvnode(tag, mp, vops, vpp)
int (**vops) __P((void *));
struct vnode **vpp;
{
extern struct uvm_pagerops uvm_vnodeops;
struct uvm_object *uobj;
struct proc *p = curproc; /* XXX */
struct freelst *listhd;
static int toggle;
@ -451,6 +453,7 @@ getnewvnode(tag, mp, vops, vpp)
* vnode_hold_list because we will lose the identity of all its
* referencing buffers.
*/
toggle ^= 1;
if (numvnodes > 2 * desiredvnodes)
toggle = 0;
@ -461,7 +464,7 @@ getnewvnode(tag, mp, vops, vpp)
(TAILQ_FIRST(listhd = &vnode_hold_list) == NULL || toggle))) {
simple_unlock(&vnode_free_list_slock);
vp = pool_get(&vnode_pool, PR_WAITOK);
memset((char *)vp, 0, sizeof(*vp));
memset(vp, 0, sizeof(*vp));
simple_lock_init(&vp->v_interlock);
numvnodes++;
} else {
@ -522,6 +525,7 @@ getnewvnode(tag, mp, vops, vpp)
vp->v_type = VNON;
vp->v_vnlock = &vp->v_lock;
lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0);
lockinit(&vp->v_glock, PVFS, "glock", 0, 0);
cache_purge(vp);
vp->v_tag = tag;
vp->v_op = vops;
@ -530,6 +534,16 @@ getnewvnode(tag, mp, vops, vpp)
vp->v_usecount = 1;
vp->v_data = 0;
simple_lock_init(&vp->v_uvm.u_obj.vmobjlock);
/*
* initialize uvm_object within vnode.
*/
uobj = &vp->v_uvm.u_obj;
uobj->pgops = &uvm_vnodeops;
TAILQ_INIT(&uobj->memq);
vp->v_uvm.u_size = VSIZENOTSET;
if (mp && error != EDEADLK)
vfs_unbusy(mp);
return (0);
@ -606,7 +620,6 @@ vwakeup(bp)
{
struct vnode *vp;
bp->b_flags &= ~B_WRITEINPROG;
if ((vp = bp->b_vp) != NULL) {
if (--vp->v_numoutput < 0)
panic("vwakeup: neg numoutput, vp %p", vp);
@ -630,9 +643,21 @@ vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
struct proc *p;
int slpflag, slptimeo;
{
struct uvm_object *uobj = &vp->v_uvm.u_obj;
struct buf *bp, *nbp;
int s, error;
int s, error, rv;
int flushflags = PGO_ALLPAGES|PGO_FREE|PGO_SYNCIO|
(flags & V_SAVE ? PGO_CLEANIT : 0);
/* XXXUBC this doesn't look at flags or slp* */
if (vp->v_type == VREG) {
simple_lock(&uobj->vmobjlock);
rv = (uobj->pgops->pgo_flush)(uobj, 0, 0, flushflags);
simple_unlock(&uobj->vmobjlock);
if (!rv) {
return EIO;
}
}
if (flags & V_SAVE) {
error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0, p);
if (error)
@ -714,10 +739,22 @@ vtruncbuf(vp, lbn, slpflag, slptimeo)
daddr_t lbn;
int slpflag, slptimeo;
{
struct uvm_object *uobj = &vp->v_uvm.u_obj;
struct buf *bp, *nbp;
int s, error;
int s, error, rv;
s = splbio();
if (vp->v_type == VREG) {
simple_lock(&uobj->vmobjlock);
rv = (uobj->pgops->pgo_flush)(uobj,
round_page(lbn << vp->v_mount->mnt_fs_bshift),
vp->v_uvm.u_size, PGO_FREE);
simple_unlock(&uobj->vmobjlock);
if (!rv) {
splx(s);
return EIO;
}
}
restart:
for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
@ -726,7 +763,7 @@ restart:
continue;
if (bp->b_flags & B_BUSY) {
bp->b_flags |= B_WANTED;
error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
error = tsleep(bp, slpflag | (PRIBIO + 1),
"vtruncbuf", slptimeo);
if (error) {
splx(s);
@ -744,7 +781,7 @@ restart:
continue;
if (bp->b_flags & B_BUSY) {
bp->b_flags |= B_WANTED;
error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
error = tsleep(bp, slpflag | (PRIBIO + 1),
"vtruncbuf", slptimeo);
if (error) {
splx(s);
@ -766,9 +803,18 @@ vflushbuf(vp, sync)
struct vnode *vp;
int sync;
{
struct uvm_object *uobj = &vp->v_uvm.u_obj;
struct buf *bp, *nbp;
int s;
if (vp->v_type == VREG) {
int flags = PGO_CLEANIT|PGO_ALLPAGES| (sync ? PGO_SYNCIO : 0);
simple_lock(&uobj->vmobjlock);
(uobj->pgops->pgo_flush)(uobj, 0, 0, flags);
simple_unlock(&uobj->vmobjlock);
}
loop:
s = splbio();
for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
@ -850,11 +896,14 @@ brelvp(bp)
*/
if (bp->b_vnbufs.le_next != NOLIST)
bufremvn(bp);
if ((vp->v_flag & VONWORKLST) && LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
if (vp->v_type != VREG && (vp->v_flag & VONWORKLST) &&
LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
vp->v_flag &= ~VONWORKLST;
LIST_REMOVE(vp, v_synclist);
}
bp->b_vp = (struct vnode *) 0;
bp->b_vp = NULL;
HOLDRELE(vp);
splx(s);
}
@ -874,11 +923,6 @@ reassignbuf(bp, newvp)
struct buflists *listheadp;
int delay;
if (newvp == NULL) {
printf("reassignbuf: NULL");
return;
}
/*
* Delete from old vnode list, if on one.
*/
@ -890,7 +934,8 @@ reassignbuf(bp, newvp)
*/
if ((bp->b_flags & B_DELWRI) == 0) {
listheadp = &newvp->v_cleanblkhd;
if ((newvp->v_flag & VONWORKLST) &&
if (newvp->v_type != VREG &&
(newvp->v_flag & VONWORKLST) &&
LIST_FIRST(&newvp->v_dirtyblkhd) == NULL) {
newvp->v_flag &= ~VONWORKLST;
LIST_REMOVE(newvp, v_synclist);
@ -1074,9 +1119,13 @@ vget(vp, flags)
* return failure. Cleaning is determined by checking that
* the VXLOCK flag is set.
*/
if ((flags & LK_INTERLOCK) == 0)
simple_lock(&vp->v_interlock);
if (vp->v_flag & VXLOCK) {
if (flags & LK_NOWAIT) {
return EBUSY;
}
vp->v_flag |= VXWANT;
ltsleep((caddr_t)vp, PINOD|PNORELOCK,
"vget", 0, &vp->v_interlock);
@ -1167,6 +1216,7 @@ vput(vp)
else
TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
simple_unlock(&vnode_free_list_slock);
vp->v_flag &= ~VTEXT;
simple_unlock(&vp->v_interlock);
VOP_INACTIVE(vp, p);
}
@ -1194,7 +1244,7 @@ vrele(vp)
#ifdef DIAGNOSTIC
if (vp->v_usecount < 0 || vp->v_writecount != 0) {
vprint("vrele: bad ref count", vp);
panic("vrele: ref cnt");
panic("vrele: ref cnt vp %p", vp);
}
#endif
/*
@ -1206,6 +1256,7 @@ vrele(vp)
else
TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
simple_unlock(&vnode_free_list_slock);
vp->v_flag &= ~VTEXT;
if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0)
VOP_INACTIVE(vp, p);
}
@ -1256,6 +1307,7 @@ holdrele(vp)
if (vp->v_holdcnt <= 0)
panic("holdrele: holdcnt vp %p", vp);
vp->v_holdcnt--;
/*
* If it is on the holdlist and the hold count drops to
* zero, move it to the free list. The test of the back
@ -1269,6 +1321,7 @@ holdrele(vp)
* getnewvnode after removing it from a freelist to ensure
* that we do not try to move it here.
*/
if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) &&
vp->v_holdcnt == 0 && vp->v_usecount == 0) {
simple_lock(&vnode_free_list_slock);
@ -1427,6 +1480,8 @@ vclean(vp, flags, p)
if (vp->v_flag & VXLOCK)
panic("vclean: deadlock, vp %p", vp);
vp->v_flag |= VXLOCK;
vp->v_flag &= ~VTEXT;
/*
* Even if the count is zero, the VOP_INACTIVE routine may still
* have the object locked while it cleans it out. The VOP_LOCK
@ -1437,11 +1492,7 @@ vclean(vp, flags, p)
VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK);
/*
* clean out any VM data associated with the vnode.
*/
uvm_vnp_terminate(vp);
/*
* Clean out any buffers associated with the vnode.
* Clean out any cached data associated with the vnode.
*/
if (flags & DOCLOSE)
vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
@ -1467,7 +1518,6 @@ vclean(vp, flags, p)
*/
if (VOP_RECLAIM(vp, p))
panic("vclean: cannot reclaim, vp %p", vp);
if (active) {
/*
* Inline copy of vrele() since VOP_INACTIVE
@ -1484,6 +1534,7 @@ vclean(vp, flags, p)
/*
* Insert at tail of LRU list.
*/
simple_unlock(&vp->v_interlock);
simple_lock(&vnode_free_list_slock);
#ifdef DIAGNOSTIC
@ -1740,7 +1791,7 @@ vprint(label, vp)
if (label != NULL)
printf("%s: ", label);
printf("tag %d type %s, usecount %ld, writecount %ld, refcount %ld,",
printf("tag %d type %s, usecount %d, writecount %ld, refcount %ld,",
vp->v_tag, typename[vp->v_type], vp->v_usecount, vp->v_writecount,
vp->v_holdcnt);
buf[0] = '\0';
@ -2365,7 +2416,7 @@ vfs_shutdown()
/* avoid coming back this way again if we panic. */
doing_shutdown = 1;
sys_sync(p, (void *)0, (register_t *)0);
sys_sync(p, NULL, NULL);
/* Wait for sync to finish. */
dcount = 10000;
@ -2608,10 +2659,10 @@ vfs_detach(vfs)
#ifdef DDB
const char buf_flagbits[] =
"\20\1AGE\2NEEDCOMMIT\3ASYNC\4BAD\5BUSY\6CACHE\7CALL\10DELWRI"
"\20\1AGE\2NEEDCOMMIT\3ASYNC\4BAD\5BUSY\6SCANNED\7CALL\10DELWRI"
"\11DIRTY\12DONE\13EINTR\14ERROR\15GATHERED\16INVAL\17LOCKED\20NOCACHE"
"\21PAGET\22PGIN\23PHYS\24RAW\25READ\26TAPE\27UAREA\30WANTED"
"\31WRITEINPROG\32XXX\33VFLUSH";
"\21ORDERED\22CACHE\23PHYS\24RAW\25READ\26TAPE\30WANTED"
"\32XXX\33VFLUSH";
void
vfs_buf_print(bp, full, pr)
@ -2629,15 +2680,9 @@ vfs_buf_print(bp, full, pr)
(*pr)(" bufsize 0x%x bcount 0x%x resid 0x%x\n",
bp->b_bufsize, bp->b_bcount, bp->b_resid);
(*pr)(" data %p saveaddr %p\n",
bp->b_data, bp->b_saveaddr);
(*pr)(" data %p saveaddr %p dep %p\n",
bp->b_data, bp->b_saveaddr, LIST_FIRST(&bp->b_dep));
(*pr)(" iodone %p\n", bp->b_iodone);
(*pr)(" dirtyoff 0x%x dirtyend 0x%x validoff 0x%x validend 0x%x\n",
bp->b_dirtyoff, bp->b_dirtyend,
bp->b_validoff, bp->b_validend);
(*pr)(" rcred %p wcred %p\n", bp->b_rcred, bp->b_wcred);
}
@ -2689,16 +2734,17 @@ vfs_vnode_print(vp, full, pr)
int full;
void (*pr) __P((const char *, ...));
{
char buf[1024];
char buf[256];
const char *vtype, *vtag;
uvm_object_printit(&vp->v_uvm.u_obj, full, pr);
bitmask_snprintf(vp->v_flag, vnode_flagbits, buf, sizeof(buf));
(*pr)("\nVNODE flags %s\n", buf);
(*pr)("nio %d size 0x%x wlist %s\n",
vp->v_uvm.u_nio, vp->v_uvm.u_size,
vp->v_uvm.u_wlist.le_next ? "YES" : "NO");
(*pr)("mp %p nio %d size 0x%x rwlock 0x%x glock 0x%x\n",
vp->v_mount, vp->v_uvm.u_nio, (int)vp->v_uvm.u_size,
vp->v_vnlock ? lockstatus(vp->v_vnlock) : 0x999,
lockstatus(&vp->v_glock));
(*pr)("data %p usecount %d writecount %d holdcnt %d numoutput %d\n",
vp->v_data, vp->v_usecount, vp->v_writecount,
@ -2723,16 +2769,14 @@ vfs_vnode_print(vp, full, pr)
struct buf *bp;
(*pr)("clean bufs:\n");
for (bp = LIST_FIRST(&vp->v_cleanblkhd);
bp != NULL;
bp = LIST_NEXT(bp, b_vnbufs)) {
LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
(*pr)(" bp %p\n", bp);
vfs_buf_print(bp, full, pr);
}
(*pr)("dirty bufs:\n");
for (bp = LIST_FIRST(&vp->v_dirtyblkhd);
bp != NULL;
bp = LIST_NEXT(bp, b_vnbufs)) {
LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
(*pr)(" bp %p\n", bp);
vfs_buf_print(bp, full, pr);
}
}

11
sys/kern/vfs_syscalls.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: vfs_syscalls.c,v 1.163 2000/09/28 06:43:20 enami Exp $ */
/* $NetBSD: vfs_syscalls.c,v 1.164 2000/11/27 08:39:44 chs Exp $ */
/*
* Copyright (c) 1989, 1993
@ -571,7 +571,6 @@ sys_sync(p, v, retval)
if ((mp->mnt_flag & MNT_RDONLY) == 0) {
asyncflag = mp->mnt_flag & MNT_ASYNC;
mp->mnt_flag &= ~MNT_ASYNC;
uvm_vnp_sync(mp);
VFS_SYNC(mp, MNT_NOWAIT, p->p_ucred, p);
if (asyncflag)
mp->mnt_flag |= MNT_ASYNC;
@ -1181,6 +1180,11 @@ sys_fhopen(p, v, retval)
}
if ((error = VOP_OPEN(vp, flags, cred, p)) != 0)
goto bad;
if (vp->v_type == VREG &&
uvn_attach(vp, flags & FWRITE ? VM_PROT_WRITE : 0) == NULL) {
error = EIO;
goto bad;
}
if (flags & FWRITE)
vp->v_writecount++;
@ -1583,8 +1587,6 @@ sys_unlink(p, v, retval)
goto out;
}
(void)uvm_vnp_uncache(vp);
VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE);
VOP_LEASE(vp, p, p->p_ucred, LEASE_WRITE);
error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
@ -2852,7 +2854,6 @@ out:
if (fromnd.ni_dvp != tdvp)
VOP_LEASE(fromnd.ni_dvp, p, p->p_ucred, LEASE_WRITE);
if (tvp) {
(void)uvm_vnp_uncache(tvp);
VOP_LEASE(tvp, p, p->p_ucred, LEASE_WRITE);
}
error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,

15
sys/kern/vfs_vnops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: vfs_vnops.c,v 1.44 2000/08/12 16:43:00 sommerfeld Exp $ */
/* $NetBSD: vfs_vnops.c,v 1.45 2000/11/27 08:39:44 chs Exp $ */
/*
* Copyright (c) 1982, 1986, 1989, 1993
@ -156,8 +156,14 @@ vn_open(ndp, fmode, cmode)
}
if ((error = VOP_OPEN(vp, fmode, cred, p)) != 0)
goto bad;
if (vp->v_type == VREG &&
uvn_attach(vp, fmode & FWRITE ? VM_PROT_WRITE : 0) == NULL) {
error = EIO;
goto bad;
}
if (fmode & FWRITE)
vp->v_writecount++;
return (0);
bad:
vput(vp);
@ -174,11 +180,10 @@ vn_writechk(vp)
{
/*
* If there's shared text associated with
* the vnode, try to free it up once. If
* we fail, we can't allow writing.
* If the vnode is in use as a process's text,
* we can't allow writing.
*/
if ((vp->v_flag & VTEXT) && !uvm_vnp_uncache(vp))
if (vp->v_flag & VTEXT)
return (ETXTBSY);
return (0);
}

47
sys/kern/vnode_if.src
View File

@ -1,4 +1,4 @@
# $NetBSD: vnode_if.src,v 1.25 2000/09/19 21:57:14 fvdl Exp $
# $NetBSD: vnode_if.src,v 1.26 2000/11/27 08:39:45 chs Exp $
#
# Copyright (c) 1992, 1993
# The Regents of the University of California. All rights reserved.
@ -502,6 +502,17 @@ vop_balloc {
OUT struct buf **bpp;
};
#
#% ballocn vp L L L
#
vop_ballocn {
IN struct vnode *vp;
IN off_t offset;
IN off_t length;
IN struct ucred *cred;
IN int flags;
};
#
#% reallocblks vp L L L
#
@ -569,3 +580,37 @@ vop_whiteout {
#vop_bwrite {
# IN struct buf *bp;
#};
#
#% getpages vp L L L
#
vop_getpages {
IN struct vnode *vp;
IN voff_t offset;
IN vm_page_t *m;
IN int *count;
IN int centeridx;
IN vm_prot_t access_type;
IN int advice;
IN int flags;
};
#
#% putpages vp L L L
#
vop_putpages {
IN struct vnode *vp;
IN vm_page_t *m;
IN int count;
IN int flags;
IN int *rtvals;
};
#
#% size vp = = =
#
vop_size {
IN struct vnode *vp;
IN off_t size;
OUT off_t *eobp;
};

5
sys/miscfs/genfs/genfs.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: genfs.h,v 1.10 1999/08/03 20:19:19 wrstuden Exp $ */
/* $NetBSD: genfs.h,v 1.11 2000/11/27 08:39:45 chs Exp $ */
int genfs_badop __P((void *));
int genfs_nullop __P((void *));
@ -22,3 +22,6 @@ int genfs_lease_check __P((void *));
int genfs_lock __P((void *));
int genfs_islocked __P((void *));
int genfs_unlock __P((void *));
int genfs_getpages __P((void *));
int genfs_putpages __P((void *));
int genfs_size __P((void *));

661
sys/miscfs/genfs/genfs_vnops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: genfs_vnops.c,v 1.20 2000/09/19 22:01:59 fvdl Exp $ */
/* $NetBSD: genfs_vnops.c,v 1.21 2000/11/27 08:39:45 chs Exp $ */
/*
* Copyright (c) 1982, 1986, 1989, 1993
@ -50,6 +50,9 @@
#include <miscfs/genfs/genfs.h>
#include <miscfs/specfs/specdev.h>
#include <uvm/uvm.h>
#include <uvm/uvm_pager.h>
#ifdef NFSSERVER
#include <nfs/rpcv2.h>
#include <nfs/nfsproto.h>
@ -414,3 +417,659 @@ genfs_lease_check(v)
return (0);
#endif /* NFSSERVER */
}
/*
* generic VM getpages routine.
* Return PG_BUSY pages for the given range,
* reading from backing store if necessary.
*/
int
genfs_getpages(v)
void *v;
{
struct vop_getpages_args /* {
struct vnode *a_vp;
voff_t a_offset;
vm_page_t *a_m;
int *a_count;
int a_centeridx;
vm_prot_t a_access_type;
int a_advice;
int a_flags;
} */ *ap = v;
off_t eof, offset, origoffset, startoffset, endoffset, raoffset;
daddr_t lbn, blkno;
int s, i, error, npages, orignpages, npgs, run, ridx, pidx, pcount;
int fs_bshift, fs_bsize, dev_bshift, dev_bsize;
int flags = ap->a_flags;
size_t bytes, iobytes, tailbytes, totalbytes, skipbytes;
vaddr_t kva;
struct buf *bp, *mbp;
struct vnode *vp = ap->a_vp;
struct uvm_object *uobj = &vp->v_uvm.u_obj;
struct vm_page *pgs[16]; /* XXXUBC 16 */
struct ucred *cred = curproc->p_ucred; /* XXXUBC curproc */
boolean_t async = (flags & PGO_SYNCIO) == 0;
boolean_t write = (ap->a_access_type & VM_PROT_WRITE) != 0;
boolean_t sawhole = FALSE;
UVMHIST_FUNC("genfs_getpages"); UVMHIST_CALLED(ubchist);
/* XXXUBC temp limit */
if (*ap->a_count > 16) {
return EINVAL;
}
error = VOP_SIZE(vp, vp->v_uvm.u_size, &eof);
if (error) {
return error;
}
#ifdef DIAGNOSTIC
if (ap->a_centeridx < 0 || ap->a_centeridx > *ap->a_count) {
panic("genfs_getpages: centeridx %d out of range",
ap->a_centeridx);
}
if (ap->a_offset & (PAGE_SIZE - 1) || ap->a_offset < 0) {
panic("genfs_getpages: offset 0x%x", (int)ap->a_offset);
}
if (*ap->a_count < 0) {
panic("genfs_getpages: count %d < 0", *ap->a_count);
}
#endif
/*
* Bounds-check the request.
*/
error = 0;
origoffset = ap->a_offset;
if (origoffset + (ap->a_centeridx << PAGE_SHIFT) >= eof &&
(flags & PGO_PASTEOF) == 0) {
if ((flags & PGO_LOCKED) == 0) {
simple_unlock(&uobj->vmobjlock);
}
UVMHIST_LOG(ubchist, "off 0x%x count %d goes past EOF 0x%x",
origoffset, *ap->a_count, eof,0);
return EINVAL;
}
/*
* For PGO_LOCKED requests, just return whatever's in memory.
*/
if (flags & PGO_LOCKED) {
uvn_findpages(uobj, origoffset, ap->a_count, ap->a_m,
UFP_NOWAIT|UFP_NOALLOC|UFP_NORDONLY);
return ap->a_m[ap->a_centeridx] == NULL ? EBUSY : 0;
}
/* vnode is VOP_LOCKed, uobj is locked */
if (write && (vp->v_flag & VONWORKLST) == 0) {
vn_syncer_add_to_worklist(vp, filedelay);
}
/*
* find the requested pages and make some simple checks.
* leave space in the page array for a whole block.
*/
fs_bshift = vp->v_mount->mnt_fs_bshift;
fs_bsize = 1 << fs_bshift;
dev_bshift = vp->v_mount->mnt_dev_bshift;
dev_bsize = 1 << dev_bshift;
KASSERT((eof & (dev_bsize - 1)) == 0);
orignpages = min(*ap->a_count,
round_page(eof - origoffset) >> PAGE_SHIFT);
if (flags & PGO_PASTEOF) {
orignpages = *ap->a_count;
}
npages = orignpages;
startoffset = origoffset & ~(fs_bsize - 1);
endoffset = round_page((origoffset + (npages << PAGE_SHIFT)
+ fs_bsize - 1) & ~(fs_bsize - 1));
endoffset = min(endoffset, round_page(eof));
ridx = (origoffset - startoffset) >> PAGE_SHIFT;
memset(pgs, 0, sizeof(pgs));
uvn_findpages(uobj, origoffset, &npages, &pgs[ridx], UFP_ALL);
/*
* if PGO_OVERWRITE is set, don't bother reading the pages.
* PGO_OVERWRITE also means that the caller guarantees
* that the pages already have backing store allocated.
*/
if (flags & PGO_OVERWRITE) {
UVMHIST_LOG(ubchist, "PGO_OVERWRITE",0,0,0,0);
for (i = 0; i < npages; i++) {
struct vm_page *pg = pgs[ridx + i];
if (pg->flags & PG_FAKE) {
uvm_pagezero(pg);
pg->flags &= ~(PG_FAKE);
}
pg->flags &= ~(PG_RDONLY);
}
goto out;
}
/*
* if the pages are already resident, just return them.
*/
for (i = 0; i < npages; i++) {
struct vm_page *pg = pgs[ridx + i];
if ((pg->flags & PG_FAKE) ||
(write && (pg->flags & PG_RDONLY))) {
break;
}
}
if (i == npages) {
UVMHIST_LOG(ubchist, "returning cached pages", 0,0,0,0);
raoffset = origoffset + (orignpages << PAGE_SHIFT);
goto raout;
}
/*
* the page wasn't resident and we're not overwriting,
* so we're going to have to do some i/o.
* find any additional pages needed to cover the expanded range.
*/
if (startoffset != origoffset) {
/*
* XXXUBC we need to avoid deadlocks caused by locking
* additional pages at lower offsets than pages we
* already have locked. for now, unlock them all and
* start over.
*/
for (i = 0; i < npages; i++) {
struct vm_page *pg = pgs[ridx + i];
if (pg->flags & PG_FAKE) {
pg->flags |= PG_RELEASED;
}
}
uvm_page_unbusy(&pgs[ridx], npages);
memset(pgs, 0, sizeof(pgs));
UVMHIST_LOG(ubchist, "reset npages start 0x%x end 0x%x",
startoffset, endoffset, 0,0);
npages = (endoffset - startoffset) >> PAGE_SHIFT;
npgs = npages;
uvn_findpages(uobj, startoffset, &npgs, pgs, UFP_ALL);
}
simple_unlock(&uobj->vmobjlock);
/*
* read the desired page(s).
*/
totalbytes = npages << PAGE_SHIFT;
bytes = min(totalbytes, eof - startoffset);
tailbytes = totalbytes - bytes;
skipbytes = 0;
kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WAITOK |
UVMPAGER_MAPIN_READ);
s = splbio();
mbp = pool_get(&bufpool, PR_WAITOK);
splx(s);
mbp->b_bufsize = totalbytes;
mbp->b_data = (void *)kva;
mbp->b_resid = mbp->b_bcount = bytes;
mbp->b_flags = B_BUSY|B_READ| (async ? B_CALL : 0);
mbp->b_iodone = uvm_aio_biodone;
mbp->b_vp = vp;
LIST_INIT(&mbp->b_dep);
/*
* if EOF is in the middle of the last page, zero the part past EOF.
*/
if (tailbytes > 0) {
memset((void *)(kva + bytes), 0, tailbytes);
}
/*
* now loop over the pages, reading as needed.
*/
if (write) {
lockmgr(&vp->v_glock, LK_EXCLUSIVE, NULL);
} else {
lockmgr(&vp->v_glock, LK_SHARED, NULL);
}
bp = NULL;
for (offset = startoffset;
bytes > 0;
offset += iobytes, bytes -= iobytes) {
/*
* skip pages which don't need to be read.
*/
pidx = (offset - startoffset) >> PAGE_SHIFT;
while ((pgs[pidx]->flags & PG_FAKE) == 0) {
size_t b;
#ifdef DEBUG
if (offset & (PAGE_SIZE - 1)) {
panic("genfs_getpages: skipping from middle "
"of page");
}
#endif
b = min(PAGE_SIZE, bytes);
offset += b;
bytes -= b;
skipbytes += b;
pidx++;
UVMHIST_LOG(ubchist, "skipping, new offset 0x%x",
offset, 0,0,0);
if (bytes == 0) {
goto loopdone;
}
}
/*
* bmap the file to find out the blkno to read from and
* how much we can read in one i/o. if bmap returns an error,
* skip the rest of the top-level i/o.
*/
lbn = offset >> fs_bshift;
error = VOP_BMAP(vp, lbn, NULL, &blkno, &run);
if (error) {
UVMHIST_LOG(ubchist, "VOP_BMAP lbn 0x%x -> %d\n",
lbn, error,0,0);
skipbytes += bytes;
goto loopdone;
}
/*
* see how many pages can be read with this i/o.
* reduce the i/o size if necessary to avoid
* overwriting pages with valid data.
*/
iobytes = min(((lbn + 1 + run) << fs_bshift) - offset, bytes);
if (offset + iobytes > round_page(offset)) {
pcount = 1;
while (pidx + pcount < npages &&
pgs[pidx + pcount]->flags & PG_FAKE) {
pcount++;
}
iobytes = min(iobytes, (pcount << PAGE_SHIFT) -
(offset - trunc_page(offset)));
}
/*
* if this block isn't allocated, zero it instead of reading it.
* if this is a read access, mark the pages we zeroed PG_RDONLY.
*/
if (blkno < 0) {
UVMHIST_LOG(ubchist, "lbn 0x%x -> HOLE", lbn,0,0,0);
sawhole = TRUE;
memset((char *)kva + (offset - startoffset), 0,
iobytes);
skipbytes += iobytes;
if (!write) {
int holepages =
(round_page(offset + iobytes) -
trunc_page(offset)) >> PAGE_SHIFT;
for (i = 0; i < holepages; i++) {
pgs[pidx + i]->flags |= PG_RDONLY;
}
}
continue;
}
/*
* allocate a sub-buf for this piece of the i/o
* (or just use mbp if there's only 1 piece),
* and start it going.
*/
if (offset == startoffset && iobytes == bytes) {
bp = mbp;
} else {
s = splbio();
bp = pool_get(&bufpool, PR_WAITOK);
splx(s);
bp->b_data = (char *)kva + offset - startoffset;
bp->b_resid = bp->b_bcount = iobytes;
bp->b_flags = B_BUSY|B_READ|B_CALL;
bp->b_iodone = uvm_aio_biodone1;
bp->b_vp = vp;
LIST_INIT(&bp->b_dep);
}
bp->b_lblkno = 0;
bp->b_private = mbp;
/* adjust physical blkno for partial blocks */
bp->b_blkno = blkno + ((offset - (lbn << fs_bshift)) >>
dev_bshift);
UVMHIST_LOG(ubchist, "bp %p offset 0x%x bcount 0x%x blkno 0x%x",
bp, offset, iobytes, bp->b_blkno);
VOP_STRATEGY(bp);
}
loopdone:
if (skipbytes) {
s = splbio();
if (error) {
mbp->b_flags |= B_ERROR;
mbp->b_error = error;
}
mbp->b_resid -= skipbytes;
if (mbp->b_resid == 0) {
biodone(mbp);
}
splx(s);
}
if (async) {
UVMHIST_LOG(ubchist, "returning PEND",0,0,0,0);
lockmgr(&vp->v_glock, LK_RELEASE, NULL);
return EINPROGRESS;
}
if (bp != NULL) {
error = biowait(mbp);
}
s = splbio();
pool_put(&bufpool, mbp);
splx(s);
uvm_pagermapout(kva, npages);
raoffset = offset;
/*
* if this we encountered a hole then we have to do a little more work.
* for read faults, we marked the page PG_RDONLY so that future
* write accesses to the page will fault again.
* for write faults, we must make sure that the backing store for
* the page is completely allocated while the pages are locked.
*/
if (error == 0 && sawhole && write) {
error = VOP_BALLOCN(vp, startoffset, npages << PAGE_SHIFT,
cred, 0);
if (error) {
UVMHIST_LOG(ubchist, "balloc lbn 0x%x -> %d",
lbn, error,0,0);
lockmgr(&vp->v_glock, LK_RELEASE, NULL);
simple_lock(&uobj->vmobjlock);
goto out;
}
}
lockmgr(&vp->v_glock, LK_RELEASE, NULL);
simple_lock(&uobj->vmobjlock);
/*
* see if we want to start any readahead.
* XXXUBC for now, just read the next 128k on 64k boundaries.
* this is pretty nonsensical, but it is 50% faster than reading
* just the next 64k.
*/
raout:
if (!async && !write && ((int)raoffset & 0xffff) == 0 &&
PAGE_SHIFT <= 16) {
int racount;
racount = 1 << (16 - PAGE_SHIFT);
(void) VOP_GETPAGES(vp, raoffset, NULL, &racount, 0,
VM_PROT_READ, 0, 0);
simple_lock(&uobj->vmobjlock);
racount = 1 << (16 - PAGE_SHIFT);
(void) VOP_GETPAGES(vp, raoffset + 0x10000, NULL, &racount, 0,
VM_PROT_READ, 0, 0);
simple_lock(&uobj->vmobjlock);
}
/*
* we're almost done! release the pages...
* for errors, we free the pages.
* otherwise we activate them and mark them as valid and clean.
* also, unbusy pages that were not actually requested.
*/
out:
if (error) {
uvm_lock_pageq();
for (i = 0; i < npages; i++) {
if (pgs[i] == NULL) {
continue;
}
UVMHIST_LOG(ubchist, "examining pg %p flags 0x%x",
pgs[i], pgs[i]->flags, 0,0);
if ((pgs[i]->flags & PG_FAKE) == 0) {
continue;
}
if (pgs[i]->flags & PG_WANTED) {
wakeup(pgs[i]);
}
uvm_pagefree(pgs[i]);
}
uvm_unlock_pageq();
simple_unlock(&uobj->vmobjlock);
UVMHIST_LOG(ubchist, "returning error %d", error,0,0,0);
return error;
}
UVMHIST_LOG(ubchist, "succeeding, npages %d", npages,0,0,0);
for (i = 0; i < npages; i++) {
if (pgs[i] == NULL) {
continue;
}
UVMHIST_LOG(ubchist, "examining pg %p flags 0x%x",
pgs[i], pgs[i]->flags, 0,0);
if (pgs[i]->flags & PG_FAKE) {
UVMHIST_LOG(ubchist, "unfaking pg %p offset 0x%x",
pgs[i], pgs[i]->offset,0,0);
pgs[i]->flags &= ~(PG_FAKE);
pmap_clear_modify(pgs[i]);
pmap_clear_reference(pgs[i]);
}
if (write) {
pgs[i]->flags &= ~(PG_RDONLY);
}
if (i < ridx || i >= ridx + orignpages || async) {
UVMHIST_LOG(ubchist, "unbusy pg %p offset 0x%x",
pgs[i], pgs[i]->offset,0,0);
if (pgs[i]->flags & PG_WANTED) {
wakeup(pgs[i]);
}
if (pgs[i]->wire_count == 0) {
uvm_pageactivate(pgs[i]);
}
pgs[i]->flags &= ~(PG_WANTED|PG_BUSY);
UVM_PAGE_OWN(pgs[i], NULL);
}
}
simple_unlock(&uobj->vmobjlock);
if (ap->a_m != NULL) {
memcpy(ap->a_m, &pgs[ridx],
orignpages * sizeof(struct vm_page *));
}
return 0;
}
/*
* generic VM putpages routine.
* Write the given range of pages to backing store.
*/
int
genfs_putpages(v)
void *v;
{
struct vop_putpages_args /* {
struct vnode *a_vp;
struct vm_page **a_m;
int a_count;
int a_flags;
int *a_rtvals;
} */ *ap = v;
int s, error, error2, npages, run;
int fs_bshift, dev_bshift, dev_bsize;
vaddr_t kva;
off_t eof, offset, startoffset;
size_t bytes, iobytes, skipbytes;
daddr_t lbn, blkno;
struct vm_page *pg;
struct buf *mbp, *bp;
struct vnode *vp = ap->a_vp;
boolean_t async = (ap->a_flags & PGO_SYNCIO) == 0;
UVMHIST_FUNC("genfs_putpages"); UVMHIST_CALLED(ubchist);
simple_unlock(&vp->v_uvm.u_obj.vmobjlock);
error = VOP_SIZE(vp, vp->v_uvm.u_size, &eof);
if (error) {
return error;
}
error = error2 = 0;
npages = ap->a_count;
fs_bshift = vp->v_mount->mnt_fs_bshift;
dev_bshift = vp->v_mount->mnt_dev_bshift;
dev_bsize = 1 << dev_bshift;
KASSERT((eof & (dev_bsize - 1)) == 0);
pg = ap->a_m[0];
startoffset = pg->offset;
bytes = min(npages << PAGE_SHIFT, eof - startoffset);
skipbytes = 0;
KASSERT(bytes != 0);
kva = uvm_pagermapin(ap->a_m, npages, UVMPAGER_MAPIN_WAITOK);
s = splbio();
vp->v_numoutput += 2;
mbp = pool_get(&bufpool, PR_WAITOK);
UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x",
vp, mbp, vp->v_numoutput, bytes);
splx(s);
mbp->b_bufsize = npages << PAGE_SHIFT;
mbp->b_data = (void *)kva;
mbp->b_resid = mbp->b_bcount = bytes;
mbp->b_flags = B_BUSY|B_WRITE|B_AGE |
(async ? B_CALL : 0) |
(curproc == uvm.pagedaemon_proc ? B_PDAEMON : 0);
mbp->b_iodone = uvm_aio_biodone;
mbp->b_vp = vp;
LIST_INIT(&mbp->b_dep);
bp = NULL;
for (offset = startoffset;
bytes > 0;
offset += iobytes, bytes -= iobytes) {
lbn = offset >> fs_bshift;
error = VOP_BMAP(vp, lbn, NULL, &blkno, &run);
if (error) {
UVMHIST_LOG(ubchist, "VOP_BMAP() -> %d", error,0,0,0);
skipbytes += bytes;
bytes = 0;
break;
}
iobytes = min(((lbn + 1 + run) << fs_bshift) - offset, bytes);
if (blkno == (daddr_t)-1) {
skipbytes += iobytes;
continue;
}
/* if it's really one i/o, don't make a second buf */
if (offset == startoffset && iobytes == bytes) {
bp = mbp;
} else {
s = splbio();
vp->v_numoutput++;
bp = pool_get(&bufpool, PR_WAITOK);
UVMHIST_LOG(ubchist, "vp %p bp %p num now %d",
vp, bp, vp->v_numoutput, 0);
splx(s);
bp->b_data = (char *)kva +
(vaddr_t)(offset - pg->offset);
bp->b_resid = bp->b_bcount = iobytes;
bp->b_flags = B_BUSY|B_WRITE|B_CALL|B_ASYNC;
bp->b_iodone = uvm_aio_biodone1;
bp->b_vp = vp;
LIST_INIT(&bp->b_dep);
}
bp->b_lblkno = 0;
bp->b_private = mbp;
/* adjust physical blkno for partial blocks */
bp->b_blkno = blkno + ((offset - (lbn << fs_bshift)) >>
dev_bshift);
UVMHIST_LOG(ubchist, "vp %p offset 0x%x bcount 0x%x blkno 0x%x",
vp, offset, bp->b_bcount, bp->b_blkno);
VOP_STRATEGY(bp);
}
if (skipbytes) {
UVMHIST_LOG(ubchist, "skipbytes %d", bytes, 0,0,0);
s = splbio();
mbp->b_resid -= skipbytes;
if (mbp->b_resid == 0) {
biodone(mbp);
}
splx(s);
}
if (async) {
UVMHIST_LOG(ubchist, "returning PEND", 0,0,0,0);
return EINPROGRESS;
}
if (bp != NULL) {
UVMHIST_LOG(ubchist, "waiting for mbp %p", mbp,0,0,0);
error2 = biowait(mbp);
}
{
/* XXXUBC */
void softdep_pageiodone(struct buf *);
softdep_pageiodone(mbp);
}
s = splbio();
vwakeup(mbp);
pool_put(&bufpool, mbp);
splx(s);
uvm_pagermapout(kva, npages);
UVMHIST_LOG(ubchist, "returning, error %d", error,0,0,0);
return error ? error : error2;
}
int
genfs_size(v)
void *v;
{
struct vop_size_args /* {
struct vnode *a_vp;
off_t a_size;
off_t *a_eobp;
} */ *ap = v;
int bsize;
bsize = 1 << ap->a_vp->v_mount->mnt_fs_bshift;
*ap->a_eobp = (ap->a_size + bsize) & ~(bsize - 1);
return 0;
}

3
sys/miscfs/genfs/layer_subr.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: layer_subr.c,v 1.6 2000/03/16 18:08:24 jdolecek Exp $ */
/* $NetBSD: layer_subr.c,v 1.7 2000/11/27 08:39:45 chs Exp $ */
/*
* Copyright (c) 1999 National Aeronautics & Space Administration
@ -272,6 +272,7 @@ layer_node_alloc(mp, lowervp, vpp)
VREF(lowervp); /* Take into account reference held in layer_node */
hd = LAYER_NHASH(lmp, lowervp);
LIST_INSERT_HEAD(hd, xp, layer_hash);
uvm_vnp_setsize(vp, 0);
simple_unlock(&lmp->layerm_hashlock);
return (0);
}

3
sys/miscfs/procfs/procfs_subr.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: procfs_subr.c,v 1.33 2000/11/24 18:58:37 chs Exp $ */
/* $NetBSD: procfs_subr.c,v 1.34 2000/11/27 08:39:46 chs Exp $ */
/*
* Copyright (c) 1994 Christopher G. Demetriou. All rights reserved.
@ -167,6 +167,7 @@ procfs_allocvp(mp, vpp, pid, pfs_type)
}
procfs_hashins(pfs);
uvm_vnp_setsize(vp, 0);
lockmgr(&pfs_hashlock, LK_RELEASE, NULL);
return (error);

33
sys/miscfs/syncfs/sync_subr.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: sync_subr.c,v 1.7 2000/10/06 19:08:00 jdolecek Exp $ */
/* $NetBSD: sync_subr.c,v 1.8 2000/11/27 08:39:46 chs Exp $ */
/*
* Copyright 1997 Marshall Kirk McKusick. All Rights Reserved.
@ -188,16 +188,14 @@ sched_sync(v)
}
s = splbio();
if (LIST_FIRST(slp) == vp) {
if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL &&
vp->v_type != VBLK)
panic("sched_sync: fsync failed vp %p tag %d",
vp, vp->v_tag);
/*
* Put us back on the worklist. The worklist
* routine will remove us from our current
* position and then add us back in at a later
* position.
*/
vn_syncer_add_to_worklist(vp, syncdelay);
}
splx(s);
@ -235,7 +233,7 @@ sched_sync(v)
* filesystem activity.
*/
if (time.tv_sec == starttime)
tsleep(&lbolt, PPAUSE, "syncer", 0);
tsleep(&rushjob, PPAUSE, "syncer", hz);
}
}
@ -247,21 +245,12 @@ sched_sync(v)
int
speedup_syncer()
{
int s;
/*
* XXX Should not be doing this, should be using ltsleep()
* XXX with a timeout, rather than sleeping on lbolt.
*/
SCHED_LOCK(s);
if (updateproc && updateproc->p_wchan == &lbolt)
setrunnable(updateproc);
SCHED_UNLOCK(s);
if (rushjob < syncdelay / 2) {
rushjob += 1;
stat_rush_requests += 1;
return (1);
if (rushjob >= syncdelay / 2) {
return (0);
}
return(0);
rushjob++;
wakeup(&rushjob);
stat_rush_requests += 1;
return (1);
}

5
sys/msdosfs/msdosfs_denode.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: msdosfs_denode.c,v 1.46 2000/11/08 14:28:14 ad Exp $ */
/* $NetBSD: msdosfs_denode.c,v 1.47 2000/11/27 08:39:46 chs Exp $ */
/*-
* Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
@ -323,6 +323,7 @@ deget(pmp, dirclust, diroffset, depp)
nvp->v_type = VREG;
VREF(ldep->de_devvp);
*depp = ldep;
nvp->v_uvm.u_size = ldep->de_FileSize;
return (0);
}
@ -427,7 +428,7 @@ detrunc(dep, length, flags, cred, p)
#endif
return (error);
}
uvm_vnp_uncache(DETOV(dep)); /* what's this for? */
/*
* is this the right place for it?
*/

51
sys/msdosfs/msdosfs_fat.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: msdosfs_fat.c,v 1.33 2000/05/13 06:04:42 cgd Exp $ */
/* $NetBSD: msdosfs_fat.c,v 1.34 2000/11/27 08:39:46 chs Exp $ */
/*-
* Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
@ -965,6 +965,7 @@ fillinusemap(pmp)
* the de_flag field of the denode and it does not change the de_FileSize
* field. This is left for the caller to do.
*/
int
extendfile(dep, count, bpp, ncp, flags)
struct denode *dep;
@ -974,8 +975,7 @@ extendfile(dep, count, bpp, ncp, flags)
int flags;
{
int error;
u_long frcn;
u_long cn, got;
u_long frcn, cn, got, origcount;
struct msdosfsmount *pmp = dep->de_pmp;
struct buf *bp;
@ -1002,16 +1002,19 @@ extendfile(dep, count, bpp, ncp, flags)
return (error);
}
origcount = count;
while (count > 0) {
/*
* Allocate a new cluster chain and cat onto the end of the
* file. * If the file is empty we make de_StartCluster point
* file. If the file is empty we make de_StartCluster point
* to the new block. Note that de_StartCluster being 0 is
* sufficient to be sure the file is empty since we exclude
* attempts to extend the root directory above, and the root
* dir is the only file with a startcluster of 0 that has
* blocks allocated (sort of).
*/
if (dep->de_StartCluster == 0)
cn = 0;
else
@ -1046,41 +1049,33 @@ extendfile(dep, count, bpp, ncp, flags)
}
/*
* Update the "last cluster of the file" entry in the denode's fat
* cache.
* Update the "last cluster of the file" entry in the
* denode's fat cache.
*/
fc_setcache(dep, FC_LASTFC, frcn + got - 1, cn + got - 1);
if (flags & DE_CLEAR) {
fc_setcache(dep, FC_LASTFC, frcn + got - 1, cn + got - 1);
if ((flags & DE_CLEAR) &&
(dep->de_Attributes & ATTR_DIRECTORY)) {
while (got-- > 0) {
/*
* Get the buf header for the new block of the file.
*/
if (dep->de_Attributes & ATTR_DIRECTORY)
bp = getblk(pmp->pm_devvp, cntobn(pmp, cn++),
pmp->pm_bpcluster, 0, 0);
else {
bp = getblk(DETOV(dep), de_cn2bn(pmp, frcn++),
bp = getblk(pmp->pm_devvp, cntobn(pmp, cn++),
pmp->pm_bpcluster, 0, 0);
/*
* Do the bmap now, as in msdosfs_write
*/
if (pcbmap(dep,
de_bn2cn(pmp, bp->b_lblkno),
&bp->b_blkno, 0, 0))
bp->b_blkno = -1;
if (bp->b_blkno == -1)
panic("extendfile: pcbmap");
}
clrbuf(bp);
if (bpp) {
*bpp = bp;
bpp = NULL;
} else
bpp = NULL;
} else {
bdwrite(bp);
}
}
}
}
if ((flags & DE_CLEAR) && !(dep->de_Attributes & ATTR_DIRECTORY)) {
int cnshift = pmp->pm_cnshift;
uvm_vnp_zerorange(DETOV(dep), frcn << cnshift,
origcount << cnshift);
}
return (0);
}

5
sys/msdosfs/msdosfs_vfsops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: msdosfs_vfsops.c,v 1.70 2000/09/19 22:02:10 fvdl Exp $ */
/* $NetBSD: msdosfs_vfsops.c,v 1.71 2000/11/27 08:39:47 chs Exp $ */
/*-
* Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
@ -699,6 +699,9 @@ msdosfs_mountfs(devvp, mp, p, argp)
mp->mnt_stat.f_fsid.val[0] = (long)dev;
mp->mnt_stat.f_fsid.val[1] = makefstype(MOUNT_MSDOS);
mp->mnt_flag |= MNT_LOCAL;
mp->mnt_dev_bshift = pmp->pm_bnshift;
mp->mnt_fs_bshift = pmp->pm_cnshift;
#ifdef QUOTA
/*
* If we ever do quotas for DOS filesystems this would be a place

188
sys/msdosfs/msdosfs_vnops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: msdosfs_vnops.c,v 1.104 2000/08/03 20:41:29 thorpej Exp $ */
/* $NetBSD: msdosfs_vnops.c,v 1.105 2000/11/27 08:39:47 chs Exp $ */
/*-
* Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
@ -464,11 +464,11 @@ msdosfs_read(v)
int error = 0;
int64_t diff;
int blsize;
int isadir;
long n;
long on;
daddr_t lbn;
daddr_t rablock;
void *win;
vsize_t bytelen;
struct buf *bp;
struct vnode *vp = ap->a_vp;
struct denode *dep = VTODE(vp);
@ -478,12 +478,31 @@ msdosfs_read(v)
/*
* If they didn't ask for any data, then we are done.
*/
if (uio->uio_resid == 0)
return (0);
if (uio->uio_offset < 0)
return (EINVAL);
isadir = dep->de_Attributes & ATTR_DIRECTORY;
if (vp->v_type == VREG) {
while (uio->uio_resid > 0) {
bytelen = min(dep->de_FileSize - uio->uio_offset,
uio->uio_resid);
if (bytelen == 0)
break;
win = ubc_alloc(&vp->v_uvm.u_obj, uio->uio_offset,
&bytelen, UBC_READ);
error = uiomove(win, bytelen, uio);
ubc_release(win, 0);
if (error)
break;
}
dep->de_flag |= DE_ACCESS;
goto out;
}
/* this loop is only for directories now */
do {
lbn = de_cluster(pmp, uio->uio_offset);
on = uio->uio_offset & pmp->pm_crbomask;
@ -494,41 +513,28 @@ msdosfs_read(v)
diff = dep->de_FileSize - uio->uio_offset;
if (diff < n)
n = (long) diff;
/* convert cluster # to block # if a directory */
if (isadir) {
error = pcbmap(dep, lbn, &lbn, 0, &blsize);
if (error)
return (error);
}
/* convert cluster # to block # */
error = pcbmap(dep, lbn, &lbn, 0, &blsize);
if (error)
return (error);
/*
* If we are operating on a directory file then be sure to
* do i/o with the vnode for the filesystem instead of the
* vnode for the directory.
*/
if (isadir) {
error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
} else {
rablock = lbn + 1;
if (vp->v_lastr + 1 == lbn &&
de_cn2off(pmp, rablock) < dep->de_FileSize)
error = breada(vp, de_cn2bn(pmp, lbn),
pmp->pm_bpcluster, de_cn2bn(pmp, rablock),
pmp->pm_bpcluster, NOCRED, &bp);
else
error = bread(vp, de_cn2bn(pmp, lbn),
pmp->pm_bpcluster, NOCRED, &bp);
vp->v_lastr = lbn;
}
error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
n = min(n, pmp->pm_bpcluster - bp->b_resid);
if (error) {
brelse(bp);
return (error);
}
error = uiomove(bp->b_data + on, (int) n, uio);
if (!isadir)
dep->de_flag |= DE_ACCESS;
brelse(bp);
} while (error == 0 && uio->uio_resid > 0 && n != 0);
out:
if ((ap->a_ioflag & IO_SYNC) == IO_SYNC)
error = deupdat(dep, 1);
return (error);
@ -547,19 +553,19 @@ msdosfs_write(v)
int a_ioflag;
struct ucred *a_cred;
} */ *ap = v;
int n;
int croffset;
int resid;
u_long osize;
int error = 0;
u_long count;
daddr_t bn, lastcn;
struct buf *bp;
daddr_t lastcn;
int ioflag = ap->a_ioflag;
void *win;
vsize_t bytelen;
off_t oldoff;
boolean_t rv;
struct uio *uio = ap->a_uio;
struct proc *p = uio->uio_procp;
struct vnode *vp = ap->a_vp;
struct vnode *thisvp;
struct denode *dep = VTODE(vp);
struct msdosfsmount *pmp = dep->de_pmp;
struct ucred *cred = ap->a_cred;
@ -575,7 +581,6 @@ msdosfs_write(v)
case VREG:
if (ioflag & IO_APPEND)
uio->uio_offset = dep->de_FileSize;
thisvp = vp;
break;
case VDIR:
return EISDIR;
@ -630,84 +635,53 @@ msdosfs_write(v)
} else
lastcn = de_clcount(pmp, osize) - 1;
if (dep->de_FileSize < uio->uio_offset + resid) {
dep->de_FileSize = uio->uio_offset + resid;
uvm_vnp_setsize(vp, dep->de_FileSize);
}
do {
if (de_cluster(pmp, uio->uio_offset) > lastcn) {
oldoff = uio->uio_offset;
if (de_cluster(pmp, oldoff) > lastcn) {
error = ENOSPC;
break;
}
bytelen = min(dep->de_FileSize - oldoff, uio->uio_resid);
bn = de_blk(pmp, uio->uio_offset);
if ((uio->uio_offset & pmp->pm_crbomask) == 0
&& (de_blk(pmp, uio->uio_offset + uio->uio_resid) > de_blk(pmp, uio->uio_offset)
|| uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
/*
* If either the whole cluster gets written,
* or we write the cluster from its start beyond EOF,
* then no need to read data from disk.
*/
bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0);
clrbuf(bp);
/*
* Do the bmap now, since pcbmap needs buffers
* for the fat table. (see msdosfs_strategy)
*/
if (bp->b_blkno == bp->b_lblkno) {
error = pcbmap(dep,
de_bn2cn(pmp, bp->b_lblkno),
&bp->b_blkno, 0, 0);
if (error)
bp->b_blkno = -1;
}
if (bp->b_blkno == -1) {
brelse(bp);
if (!error)
error = EIO; /* XXX */
break;
}
} else {
/*
* The block we need to write into exists, so read it in.
*/
error = bread(thisvp, bn, pmp->pm_bpcluster,
NOCRED, &bp);
if (error) {
brelse(bp);
break;
}
/*
* XXXUBC if file is mapped and this is the last block,
* process one page at a time.
*/
if (bytelen == 0)
break;
win = ubc_alloc(&vp->v_uvm.u_obj, oldoff, &bytelen, UBC_READ);
error = uiomove(win, bytelen, uio);
ubc_release(win, 0);
if (error) {
break;
}
croffset = uio->uio_offset & pmp->pm_crbomask;
n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
if (uio->uio_offset + n > dep->de_FileSize) {
dep->de_FileSize = uio->uio_offset + n;
uvm_vnp_setsize(vp, dep->de_FileSize);/* why? */
/*
* flush what we just wrote if necessary.
* XXXUBC simplistic async flushing.
*/
if (ioflag & IO_SYNC) {
simple_lock(&vp->v_uvm.u_obj.vmobjlock);
rv = vp->v_uvm.u_obj.pgops->pgo_flush(
&vp->v_uvm.u_obj, oldoff,
oldoff + bytelen, PGO_CLEANIT|PGO_SYNCIO);
simple_unlock(&vp->v_uvm.u_obj.vmobjlock);
} else if (oldoff >> 16 != uio->uio_offset >> 16) {
simple_lock(&vp->v_uvm.u_obj.vmobjlock);
rv = vp->v_uvm.u_obj.pgops->pgo_flush(
&vp->v_uvm.u_obj, (oldoff >> 16) << 16,
(uio->uio_offset >> 16) << 16, PGO_CLEANIT);
simple_unlock(&vp->v_uvm.u_obj.vmobjlock);
}
(void) uvm_vnp_uncache(vp); /* why not? */
/*
* Should these vnode_pager_* functions be done on dir
* files?
*/
/*
* Copy the data from user space into the buf header.
*/
error = uiomove(bp->b_data + croffset, n, uio);
/*
* If they want this synchronous then write it and wait for
* it. Otherwise, if on a cluster boundary write it
* asynchronously so we can move on to the next block
* without delay. Otherwise do a delayed write because we
* may want to write somemore into the block later.
*/
if (ioflag & IO_SYNC)
(void) bwrite(bp);
else if (n + croffset == pmp->pm_bpcluster)
bawrite(bp);
else
bdwrite(bp);
dep->de_flag |= DE_UPDATE;
} while (error == 0 && uio->uio_resid > 0);
dep->de_flag |= DE_UPDATE;
/*
* If the write failed and they want us to, truncate the file back
@ -720,7 +694,8 @@ errexit:
uio->uio_offset -= resid - uio->uio_resid;
uio->uio_resid = resid;
} else {
detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED, NULL);
detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED,
NULL);
if (uio->uio_resid != resid)
error = 0;
}
@ -1805,12 +1780,12 @@ msdosfs_strategy(v)
biodone(bp);
return (error);
}
#ifdef DIAGNOSTIC
#endif
/*
* Read/write the block from/to the disk that contains the desired
* file block.
*/
vp = dep->de_devvp;
bp->b_dev = vp->v_rdev;
VOCALL(vp->v_op, VOFFSET(vop_strategy), ap);
@ -1934,7 +1909,10 @@ struct vnodeopv_entry_desc msdosfs_vnodeop_entries[] = {
{ &vop_reallocblks_desc, msdosfs_reallocblks }, /* reallocblks */
{ &vop_update_desc, msdosfs_update }, /* update */
{ &vop_bwrite_desc, vn_bwrite }, /* bwrite */
{ (struct vnodeop_desc *)NULL, (int (*) __P((void *)))NULL }
{ &vop_getpages_desc, genfs_getpages }, /* getpages */
{ &vop_putpages_desc, genfs_putpages }, /* putpages */
{ &vop_size_desc, genfs_size }, /* size */
{ NULL, NULL }
};
struct vnodeopv_desc msdosfs_vnodeop_opv_desc =
{ &msdosfs_vnodeop_p, msdosfs_vnodeop_entries };

18
sys/nfs/nfs.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: nfs.h,v 1.24 2000/09/19 22:05:55 fvdl Exp $ */
/* $NetBSD: nfs.h,v 1.25 2000/11/27 08:39:48 chs Exp $ */
/*
* Copyright (c) 1989, 1993, 1995
* The Regents of the University of California. All rights reserved.
@ -84,8 +84,18 @@ extern int nfs_niothreads; /* Number of async_daemons desired */
* DIRBLKSIZ.
*/
#if 1
/*
* XXXUBC temp hack because of the removal of b_validend.
* eventually we'll store NFS VDIR data in the page cache as well,
* we'll fix this at that point.
*/
#define NFS_DIRBLKSIZ PAGE_SIZE
#define NFS_DIRFRAGSIZ PAGE_SIZE
#else
#define NFS_DIRBLKSIZ 8192 /* Must be a multiple of DIRBLKSIZ */
#define NFS_DIRFRAGSIZ 512 /* Same as DIRBLKSIZ, generally */
#endif
/*
* Maximum number of directory entries cached per NFS node, to avoid
@ -120,10 +130,10 @@ extern int nfs_niothreads; /* Number of async_daemons desired */
#endif
/*
* The B_INVAFTERWRITE flag should be set to whatever is required by the
* buffer cache code to say "Invalidate the block after it is written back".
* Use the vm_page flag reserved for pager use to indicate pages
* which have been written to the server but not yet committed.
*/
#define B_INVAFTERWRITE B_INVAL
#define PG_NEEDCOMMIT PG_PAGER1
/*
* The IO_METASYNC flag should be implemented for local file systems.

1058
sys/nfs/nfs_bio.c
View File

File diff suppressed because it is too large Load Diff

33
sys/nfs/nfs_node.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: nfs_node.c,v 1.37 2000/11/08 14:28:15 ad Exp $ */
/* $NetBSD: nfs_node.c,v 1.38 2000/11/27 08:39:48 chs Exp $ */
/*
* Copyright (c) 1989, 1993
@ -140,7 +140,7 @@ nfs_nget(mntp, fhp, fhsize, npp)
loop:
for (np = nhpp->lh_first; np != 0; np = np->n_hash.le_next) {
if (mntp != NFSTOV(np)->v_mount || np->n_fhsize != fhsize ||
memcmp((caddr_t)fhp, (caddr_t)np->n_fhp, fhsize))
memcmp(fhp, np->n_fhp, fhsize))
continue;
vp = NFSTOV(np);
if (vget(vp, LK_EXCLUSIVE))
@ -159,10 +159,11 @@ loop:
nvp->v_vnlock = 0; /* XXX At least untill we do locking */
vp = nvp;
np = pool_get(&nfs_node_pool, PR_WAITOK);
memset((caddr_t)np, 0, sizeof *np);
memset(np, 0, sizeof *np);
lockinit(&np->n_commitlock, PINOD, "nfsclock", 0, 0);
vp->v_data = np;
np->n_vnode = vp;
/*
* Insert the nfsnode in the hash queue for its new file handle
*/
@ -171,11 +172,21 @@ loop:
np->n_fhp = malloc(fhsize, M_NFSBIGFH, M_WAITOK);
} else
np->n_fhp = &np->n_fh;
memcpy((caddr_t)np->n_fhp, (caddr_t)fhp, fhsize);
memcpy(np->n_fhp, fhp, fhsize);
np->n_fhsize = fhsize;
np->n_accstamp = -1;
np->n_vattr = pool_get(&nfs_vattr_pool, PR_WAITOK);
memset(np->n_vattr, 0, sizeof (struct vattr));
/*
* XXXUBC doing this while holding the nfs_hashlock is bad,
* but there's no alternative at the moment.
*/
error = VOP_GETATTR(vp, np->n_vattr, curproc->p_ucred, curproc);
if (error) {
return error;
}
uvm_vnp_setsize(vp, np->n_vattr->va_size);
lockmgr(&nfs_hashlock, LK_RELEASE, 0);
*npp = np;
return (0);
@ -227,7 +238,7 @@ nfs_inactive(v)
nfs_removeit(sp);
crfree(sp->s_cred);
vrele(sp->s_dvp);
FREE((caddr_t)sp, M_NFSREQ);
FREE(sp, M_NFSREQ);
}
np->n_flag &= (NMODIFIED | NFLUSHINPROG | NFLUSHWANT | NQNFSEVICTED |
NQNFSNONCACHE | NQNFSWRITE);
@ -272,12 +283,18 @@ nfs_reclaim(v)
FREE(np->n_dircache, M_NFSDIROFF);
}
if (np->n_fhsize > NFS_SMALLFH) {
free((caddr_t)np->n_fhp, M_NFSBIGFH);
free(np->n_fhp, M_NFSBIGFH);
}
pool_put(&nfs_vattr_pool, np->n_vattr);
if (np->n_rcred) {
crfree(np->n_rcred);
}
if (np->n_wcred) {
crfree(np->n_wcred);
}
cache_purge(vp);
pool_put(&nfs_node_pool, vp->v_data);
vp->v_data = (void *)0;
vp->v_data = NULL;
return (0);
}

12
sys/nfs/nfs_serv.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: nfs_serv.c,v 1.58 2000/09/19 22:05:29 fvdl Exp $ */
/* $NetBSD: nfs_serv.c,v 1.59 2000/11/27 08:39:49 chs Exp $ */
/*
* Copyright (c) 1989, 1993
@ -1729,7 +1729,6 @@ nfsrv_remove(nfsd, slp, procp, mrq)
}
out:
if (!error) {
(void)uvm_vnp_uncache(vp);
nqsrv_getl(nd.ni_dvp, ND_WRITE);
nqsrv_getl(vp, ND_WRITE);
error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
@ -1904,7 +1903,6 @@ out:
nqsrv_getl(fromnd.ni_dvp, ND_WRITE);
nqsrv_getl(tdvp, ND_WRITE);
if (tvp) {
(void)uvm_vnp_uncache(tvp);
nqsrv_getl(tvp, ND_WRITE);
}
error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
@ -3389,12 +3387,12 @@ nfsrv_access(vp, flags, cred, rdonly, p, override)
break;
}
}
/*
* If there's shared text associated with
* the inode, try to free it up once. If
* we fail, we can't allow writing.
* If the vnode is in use as a process's text,
* we can't allow writing.
*/
if ((vp->v_flag & VTEXT) && !uvm_vnp_uncache(vp))
if (vp->v_flag & VTEXT)
return (ETXTBSY);
}
error = VOP_GETATTR(vp, &vattr, cred, p);

88
sys/nfs/nfs_subs.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: nfs_subs.c,v 1.88 2000/11/08 14:28:15 ad Exp $ */
/* $NetBSD: nfs_subs.c,v 1.89 2000/11/27 08:39:49 chs Exp $ */
/*
* Copyright (c) 1989, 1993
@ -1665,17 +1665,14 @@ nfs_loadattrcache(vpp, fp, vaper)
vap->va_filerev = 0;
}
if (vap->va_size != np->n_size) {
if (vap->va_type == VREG) {
if (np->n_flag & NMODIFIED) {
if (vap->va_size < np->n_size)
vap->va_size = np->n_size;
else
np->n_size = vap->va_size;
} else
np->n_size = vap->va_size;
uvm_vnp_setsize(vp, np->n_size);
} else
if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) {
vap->va_size = np->n_size;
} else {
np->n_size = vap->va_size;
if (vap->va_type == VREG) {
uvm_vnp_setsize(vp, np->n_size);
}
}
}
np->n_attrstamp = time.tv_sec;
if (vaper != NULL) {
@ -2366,7 +2363,6 @@ netaddr_match(family, haddr, nam)
return (0);
}
/*
* The write verifier has changed (probably due to a server reboot), so all
* B_NEEDCOMMIT blocks will have to be written again. Since they are on the
@ -2377,17 +2373,14 @@ void
nfs_clearcommit(mp)
struct mount *mp;
{
struct vnode *vp, *nvp;
struct buf *bp, *nbp;
struct vnode *vp;
struct nfsnode *np;
struct vm_page *pg;
int s;
s = splbio();
loop:
for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) {
if (vp->v_mount != mp) /* Paranoia */
goto loop;
nvp = vp->v_mntvnodes.le_next;
LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
KASSERT(vp->v_mount == mp);
if (vp->v_type == VNON)
continue;
np = VTONFS(vp);
@ -2395,12 +2388,11 @@ loop:
np->n_pushedhi = 0;
np->n_commitflags &=
~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = nbp) {
nbp = bp->b_vnbufs.le_next;
if ((bp->b_flags & (B_BUSY | B_DELWRI | B_NEEDCOMMIT))
== (B_DELWRI | B_NEEDCOMMIT))
bp->b_flags &= ~B_NEEDCOMMIT;
simple_lock(&vp->v_uvm.u_obj.vmobjlock);
TAILQ_FOREACH(pg, &vp->v_uvm.u_obj.memq, listq) {
pg->flags &= ~PG_NEEDCOMMIT;
}
simple_unlock(&vp->v_uvm.u_obj.vmobjlock);
}
splx(s);
}
@ -2432,47 +2424,47 @@ nfs_merge_commit_ranges(vp)
}
int
nfs_in_committed_range(vp, bp)
nfs_in_committed_range(vp, off, len)
struct vnode *vp;
struct buf *bp;
off_t off, len;
{
struct nfsnode *np = VTONFS(vp);
off_t lo, hi;
if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
return 0;
lo = (off_t)bp->b_blkno * DEV_BSIZE;
hi = lo + bp->b_dirtyend;
lo = off;
hi = lo + len;
return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
}
int
nfs_in_tobecommitted_range(vp, bp)
nfs_in_tobecommitted_range(vp, off, len)
struct vnode *vp;
struct buf *bp;
off_t off, len;
{
struct nfsnode *np = VTONFS(vp);
off_t lo, hi;
if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
return 0;
lo = (off_t)bp->b_blkno * DEV_BSIZE;
hi = lo + bp->b_dirtyend;
lo = off;
hi = lo + len;
return (lo >= np->n_pushlo && hi <= np->n_pushhi);
}
void
nfs_add_committed_range(vp, bp)
nfs_add_committed_range(vp, off, len)
struct vnode *vp;
struct buf *bp;
off_t off, len;
{
struct nfsnode *np = VTONFS(vp);
off_t lo, hi;
lo = (off_t)bp->b_blkno * DEV_BSIZE;
hi = lo + bp->b_dirtyend;
lo = off;
hi = lo + len;
if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
np->n_pushedlo = lo;
@ -2491,9 +2483,9 @@ nfs_add_committed_range(vp, bp)
}
void
nfs_del_committed_range(vp, bp)
nfs_del_committed_range(vp, off, len)
struct vnode *vp;
struct buf *bp;
off_t off, len;
{
struct nfsnode *np = VTONFS(vp);
off_t lo, hi;
@ -2501,8 +2493,8 @@ nfs_del_committed_range(vp, bp)
if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
return;
lo = (off_t)bp->b_blkno * DEV_BSIZE;
hi = lo + bp->b_dirtyend;
lo = off;
hi = lo + len;
if (lo > np->n_pushedhi || hi < np->n_pushedlo)
return;
@ -2528,15 +2520,15 @@ nfs_del_committed_range(vp, bp)
}
void
nfs_add_tobecommitted_range(vp, bp)
nfs_add_tobecommitted_range(vp, off, len)
struct vnode *vp;
struct buf *bp;
off_t off, len;
{
struct nfsnode *np = VTONFS(vp);
off_t lo, hi;
lo = (off_t)bp->b_blkno * DEV_BSIZE;
hi = lo + bp->b_dirtyend;
lo = off;
hi = lo + len;
if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
np->n_pushlo = lo;
@ -2555,9 +2547,9 @@ nfs_add_tobecommitted_range(vp, bp)
}
void
nfs_del_tobecommitted_range(vp, bp)
nfs_del_tobecommitted_range(vp, off, len)
struct vnode *vp;
struct buf *bp;
off_t off, len;
{
struct nfsnode *np = VTONFS(vp);
off_t lo, hi;
@ -2565,8 +2557,8 @@ nfs_del_tobecommitted_range(vp, bp)
if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
return;
lo = (off_t)bp->b_blkno * DEV_BSIZE;
hi = lo + bp->b_dirtyend;
lo = off;
hi = lo + len;
if (lo > np->n_pushhi || hi < np->n_pushlo)
return;

7
sys/nfs/nfs_syscalls.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: nfs_syscalls.c,v 1.47 2000/11/24 23:30:03 chs Exp $ */
/* $NetBSD: nfs_syscalls.c,v 1.48 2000/11/27 08:39:50 chs Exp $ */
/*
* Copyright (c) 1989, 1993
@ -970,10 +970,7 @@ nfssvc_iod(p)
nmp->nm_bufqwant = FALSE;
wakeup(&nmp->nm_bufq);
}
if (bp->b_flags & B_READ)
(void) nfs_doio(bp, bp->b_rcred, (struct proc *)0);
else
(void) nfs_doio(bp, bp->b_wcred, (struct proc *)0);
(void) nfs_doio(bp, NULL);
/*
* If there are more than one iod on this mount, then defect
* so that the iods can be shared out fairly between the mounts

26
sys/nfs/nfs_var.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: nfs_var.h,v 1.18 2000/09/19 22:14:59 fvdl Exp $ */
/* $NetBSD: nfs_var.h,v 1.19 2000/11/27 08:39:50 chs Exp $ */
/*-
* Copyright (c) 1996 The NetBSD Foundation, Inc.
@ -74,8 +74,8 @@ int nfs_bioread __P((struct vnode *, struct uio *, int, struct ucred *, int));
struct buf *nfs_getcacheblk __P((struct vnode *, daddr_t, int, struct proc *));
int nfs_vinvalbuf __P((struct vnode *, int, struct ucred *, struct proc *,
int));
int nfs_asyncio __P((struct buf *, struct ucred *));
int nfs_doio __P((struct buf *, struct ucred *, struct proc *));
int nfs_asyncio __P((struct buf *));
int nfs_doio __P((struct buf *, struct proc *));
/* nfs_boot.c */
/* see nfsdiskless.h */
@ -91,9 +91,8 @@ int nfs_null __P((struct vnode *, struct ucred *, struct proc *));
int nfs_setattrrpc __P((struct vnode *, struct vattr *, struct ucred *,
struct proc *));
int nfs_readlinkrpc __P((struct vnode *, struct uio *, struct ucred *));
int nfs_readrpc __P((struct vnode *, struct uio *, struct ucred *));
int nfs_writerpc __P((struct vnode *, struct uio *, struct ucred *, int *,
int *));
int nfs_readrpc __P((struct vnode *, struct uio *));
int nfs_writerpc __P((struct vnode *, struct uio *, int *, int *));
int nfs_mknodrpc __P((struct vnode *, struct vnode **, struct componentname *,
struct vattr *));
int nfs_removeit __P((struct sillyrename *));
@ -109,8 +108,7 @@ int nfs_sillyrename __P((struct vnode *, struct vnode *,
struct componentname *));
int nfs_lookitup __P((struct vnode *, const char *, int, struct ucred *,
struct proc *, struct nfsnode **));
int nfs_commit __P((struct vnode *, u_quad_t, unsigned, struct ucred *,
struct proc *));
int nfs_commit __P((struct vnode *, off_t, uint32_t, struct proc *));
int nfs_flush __P((struct vnode *, struct ucred *, int, struct proc *, int));
/* nfs_nqlease.c */
@ -267,12 +265,12 @@ int netaddr_match __P((int, union nethostaddr *, struct mbuf *));
void nfs_clearcommit __P((struct mount *));
void nfs_merge_commit_ranges __P((struct vnode *));
int nfs_in_committed_range __P((struct vnode *, struct buf *));
int nfs_in_tobecommitted_range __P((struct vnode *, struct buf *));
void nfs_add_committed_range __P((struct vnode *, struct buf *));
void nfs_del_committed_range __P((struct vnode *, struct buf *));
void nfs_add_tobecommitted_range __P((struct vnode *, struct buf *));
void nfs_del_tobecommitted_range __P((struct vnode *, struct buf *));
int nfs_in_committed_range __P((struct vnode *, off_t, off_t));
int nfs_in_tobecommitted_range __P((struct vnode *, off_t, off_t));
void nfs_add_committed_range __P((struct vnode *, off_t, off_t));
void nfs_del_committed_range __P((struct vnode *, off_t, off_t));
void nfs_add_tobecommitted_range __P((struct vnode *, off_t, off_t));
void nfs_del_tobecommitted_range __P((struct vnode *, off_t, off_t));
int nfsrv_errmap __P((struct nfsrv_descript *, int));
void nfsrvw_sort __P((gid_t *, int));

4
sys/nfs/nfs_vfsops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: nfs_vfsops.c,v 1.96 2000/09/19 22:15:41 fvdl Exp $ */
/* $NetBSD: nfs_vfsops.c,v 1.97 2000/11/27 08:39:50 chs Exp $ */
/*
* Copyright (c) 1989, 1993, 1995
@ -689,6 +689,8 @@ mountnfs(argp, mp, nam, pth, hst, vpp, p)
#else
mp->mnt_stat.f_type = 0;
#endif
mp->mnt_fs_bshift = DEV_BSHIFT;
mp->mnt_dev_bshift = -1;
strncpy(&mp->mnt_stat.f_fstypename[0], mp->mnt_op->vfs_name,
MFSNAMELEN);
memcpy(mp->mnt_stat.f_mntfromname, hst, MNAMELEN);

171
sys/nfs/nfs_vnops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: nfs_vnops.c,v 1.123 2000/11/08 05:20:32 chs Exp $ */
/* $NetBSD: nfs_vnops.c,v 1.124 2000/11/27 08:39:51 chs Exp $ */
/*
* Copyright (c) 1989, 1993
@ -43,6 +43,7 @@
*/
#include "opt_nfs.h"
#include "opt_uvmhist.h"
#include <sys/param.h>
#include <sys/proc.h>
@ -63,6 +64,7 @@
#include <sys/unistd.h>
#include <uvm/uvm_extern.h>
#include <uvm/uvm.h>
#include <miscfs/fifofs/fifo.h>
#include <miscfs/genfs/genfs.h>
@ -136,7 +138,9 @@ struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
{ &vop_truncate_desc, nfs_truncate }, /* truncate */
{ &vop_update_desc, nfs_update }, /* update */
{ &vop_bwrite_desc, nfs_bwrite }, /* bwrite */
{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
{ &vop_getpages_desc, nfs_getpages }, /* getpages */
{ &vop_putpages_desc, nfs_putpages }, /* putpages */
{ NULL, NULL }
};
struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
@ -163,7 +167,7 @@ struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
{ &vop_poll_desc, spec_poll }, /* poll */
{ &vop_revoke_desc, spec_revoke }, /* revoke */
{ &vop_mmap_desc, spec_mmap }, /* mmap */
{ &vop_fsync_desc, nfs_fsync }, /* fsync */
{ &vop_fsync_desc, spec_fsync }, /* fsync */
{ &vop_seek_desc, spec_seek }, /* seek */
{ &vop_remove_desc, spec_remove }, /* remove */
{ &vop_link_desc, spec_link }, /* link */
@ -191,7 +195,7 @@ struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
{ &vop_truncate_desc, spec_truncate }, /* truncate */
{ &vop_update_desc, nfs_update }, /* update */
{ &vop_bwrite_desc, vn_bwrite }, /* bwrite */
{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
{ NULL, NULL }
};
struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
{ &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
@ -243,7 +247,7 @@ struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
{ &vop_truncate_desc, fifo_truncate }, /* truncate */
{ &vop_update_desc, nfs_update }, /* update */
{ &vop_bwrite_desc, vn_bwrite }, /* bwrite */
{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
{ NULL, NULL }
};
struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
{ &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
@ -432,11 +436,9 @@ nfs_open(v)
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
return (EACCES);
}
#ifndef NFS_V2_ONLY
/*
* Get a valid lease. If cached data is stale, flush it.
@ -454,7 +456,6 @@ nfs_open(v)
if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
ap->a_p, 1)) == EINTR)
return (error);
(void) uvm_vnp_uncache(vp);
np->n_brev = np->n_lrev;
}
}
@ -465,7 +466,6 @@ nfs_open(v)
if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
ap->a_p, 1)) == EINTR)
return (error);
(void) uvm_vnp_uncache(vp);
np->n_attrstamp = 0;
if (vp->v_type == VDIR) {
nfs_invaldircache(vp, 0);
@ -487,7 +487,6 @@ nfs_open(v)
if ((error = nfs_vinvalbuf(vp, V_SAVE,
ap->a_cred, ap->a_p, 1)) == EINTR)
return (error);
(void) uvm_vnp_uncache(vp);
np->n_mtime = vattr.va_mtime.tv_sec;
}
}
@ -542,6 +541,7 @@ nfs_close(v)
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
int error = 0;
UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
if (vp->v_type == VREG) {
if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
@ -558,6 +558,7 @@ nfs_close(v)
error = np->n_error;
}
}
UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
return (error);
}
@ -1020,10 +1021,9 @@ nfs_readlinkrpc(vp, uiop, cred)
* Ditto above
*/
int
nfs_readrpc(vp, uiop, cred)
nfs_readrpc(vp, uiop)
struct vnode *vp;
struct uio *uiop;
struct ucred *cred;
{
u_int32_t *tl;
caddr_t cp;
@ -1055,7 +1055,8 @@ nfs_readrpc(vp, uiop, cred)
*tl++ = txdr_unsigned(len);
*tl = 0;
}
nfsm_request(vp, NFSPROC_READ, uiop->uio_procp, cred);
nfsm_request(vp, NFSPROC_READ, uiop->uio_procp,
VTONFS(vp)->n_rcred);
if (v3) {
nfsm_postop_attr(vp, attrflag);
if (error) {
@ -1084,10 +1085,9 @@ nfsmout:
* nfs write call
*/
int
nfs_writerpc(vp, uiop, cred, iomode, must_commit)
nfs_writerpc(vp, uiop, iomode, must_commit)
struct vnode *vp;
struct uio *uiop;
struct ucred *cred;
int *iomode, *must_commit;
{
u_int32_t *tl;
@ -1110,7 +1110,7 @@ nfs_writerpc(vp, uiop, cred, iomode, must_commit)
return (EFBIG);
while (tsiz > 0) {
nfsstats.rpccnt[NFSPROC_WRITE]++;
len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
len = min(tsiz, nmp->nm_wsize);
nfsm_reqhead(vp, NFSPROC_WRITE,
NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
nfsm_fhtom(vp, v3);
@ -1135,7 +1135,8 @@ nfs_writerpc(vp, uiop, cred, iomode, must_commit)
}
nfsm_uiotom(uiop, len);
nfsm_request(vp, NFSPROC_WRITE, uiop->uio_procp, cred);
nfsm_request(vp, NFSPROC_WRITE, uiop->uio_procp,
VTONFS(vp)->n_wcred);
if (v3) {
wccflag = NFSV3_WCCCHK;
nfsm_wcc_data(vp, wccflag);
@ -2595,11 +2596,10 @@ nfs_lookitup(dvp, name, len, cred, procp, npp)
* Nfs Version 3 commit rpc
*/
int
nfs_commit(vp, offset, cnt, cred, procp)
nfs_commit(vp, offset, cnt, procp)
struct vnode *vp;
u_quad_t offset;
unsigned cnt;
struct ucred *cred;
off_t offset;
uint32_t cnt;
struct proc *procp;
{
caddr_t cp;
@ -2624,7 +2624,7 @@ nfs_commit(vp, offset, cnt, cred, procp)
txdr_hyper(offset, tl);
tl += 2;
*tl = txdr_unsigned(cnt);
nfsm_request(vp, NFSPROC_COMMIT, procp, cred);
nfsm_request(vp, NFSPROC_COMMIT, procp, VTONFS(vp)->n_wcred);
nfsm_wcc_data(vp, wccflag);
if (!error) {
nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
@ -2680,28 +2680,25 @@ nfs_strategy(v)
{
struct vop_strategy_args *ap = v;
struct buf *bp = ap->a_bp;
struct ucred *cr;
struct proc *p;
int error = 0;
if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
panic("nfs physio/async");
if (bp->b_flags & B_ASYNC)
p = (struct proc *)0;
p = NULL;
else
p = curproc; /* XXX */
if (bp->b_flags & B_READ)
cr = bp->b_rcred;
else
cr = bp->b_wcred;
/*
* 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))
error = nfs_doio(bp, cr, p);
nfs_asyncio(bp))
error = nfs_doio(bp, p);
return (error);
}
@ -2750,16 +2747,7 @@ nfs_fsync(v)
}
/*
* Flush all the blocks associated with a vnode.
* Walk through the buffer pool and push any dirty pages
* associated with the vnode.
*
* Don't bother to cluster commits; the commitrange code will
* do that. In the first pass, push all dirty buffers to the
* server, using stable writes if commit is set to 1.
* In the 2nd pass, push anything that might be left,
* i.e. the buffer was busy in the first pass, or it wasn't
* committed in the first pass.
* Flush all the data associated with a vnode.
*/
int
nfs_flush(vp, cred, waitfor, p, commit)
@ -2769,104 +2757,25 @@ nfs_flush(vp, cred, waitfor, p, commit)
struct proc *p;
int commit;
{
struct uvm_object *uobj = &vp->v_uvm.u_obj;
struct nfsnode *np = VTONFS(vp);
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
struct buf *bp;
struct buf *nbp;
int pass, s, error, slpflag, slptimeo;
int error;
int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
int rv;
UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
pass = 1;
error = 0;
slptimeo = 0;
slpflag = nmp->nm_flag & NFSMNT_INT ? PCATCH : 0;
loop:
s = splbio();
for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = nbp) {
nbp = bp->b_vnbufs.le_next;
if (bp->b_flags & B_BUSY) {
if (pass == 2 && waitfor == MNT_WAIT) {
bp->b_flags |= B_WANTED;
error = tsleep((caddr_t)bp,
slpflag | (PRIBIO + 1),
"nfsfsync", slptimeo);
splx(s);
if (error) {
if (nfs_sigintr(nmp, (struct nfsreq *)0, p))
return (EINTR);
if (slpflag == PCATCH) {
slpflag = 0;
slptimeo = 2 * hz;
}
}
goto loop;
} else
continue;
}
#ifdef DIAGNOSTIC
if ((bp->b_flags & B_DELWRI) == 0)
panic("nfs_fsync: not dirty");
#endif
if (!commit && (bp->b_flags & B_NEEDCOMMIT))
continue;
/*
* Note: can't use B_VFLUSH here, since there is no
* real vnode lock, so we can't leave the buffer on
* the freelist.
*/
bremfree(bp);
if (commit && vp->v_type == VREG)
/*
* Setting B_NOCACHE has the effect
* effect of nfs_doio using a stable write
* RPC. XXX this abuses the B_NOCACHE flag,
* but it is needed to tell nfs_strategy
* that this buffer is async, but needs to
* be written with a stable RPC. nfs_doio
* will remove B_NOCACHE again.
*/
bp->b_flags |= B_NOCACHE;
bp->b_flags |= B_BUSY | B_ASYNC;
splx(s);
VOP_BWRITE(bp);
goto loop;
}
splx(s);
if (commit && pass == 1) {
pass = 2;
goto loop;
}
if (waitfor == MNT_WAIT) {
s = splbio();
while (vp->v_numoutput) {
vp->v_flag |= VBWAIT;
error = tsleep((caddr_t)&vp->v_numoutput,
slpflag | (PRIBIO + 1), "nfsfsync", slptimeo);
if (error) {
splx(s);
if (nfs_sigintr(nmp, (struct nfsreq *)0, p))
return (EINTR);
if (slpflag == PCATCH) {
slpflag = 0;
slptimeo = 2 * hz;
}
s = splbio();
}
}
splx(s);
if (vp->v_dirtyblkhd.lh_first && commit) {
#if 0
vprint("nfs_fsync: dirty", vp);
#endif
goto loop;
}
simple_lock(&uobj->vmobjlock);
rv = (uobj->pgops->pgo_flush)(uobj, 0, 0, flushflags);
simple_unlock(&uobj->vmobjlock);
if (!rv) {
error = EIO;
}
if (np->n_flag & NWRITEERR) {
error = np->n_error;
np->n_flag &= ~NWRITEERR;
}
UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
return (error);
}

8
sys/nfs/nfsnode.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: nfsnode.h,v 1.30 2000/09/19 22:18:40 fvdl Exp $ */
/* $NetBSD: nfsnode.h,v 1.31 2000/11/27 08:39:51 chs Exp $ */
/*
* Copyright (c) 1989, 1993
@ -140,6 +140,8 @@ struct nfsnode {
off_t n_pushhi; /* Last block in range */
struct lock n_commitlock; /* Serialize commits XXX */
int n_commitflags;
struct ucred *n_rcred;
struct ucred *n_wcred;
};
/*
@ -173,7 +175,7 @@ struct nfsnode {
* Convert between nfsnode pointers and vnode pointers
*/
#define VTONFS(vp) ((struct nfsnode *)(vp)->v_data)
#define NFSTOV(np) ((struct vnode *)(np)->n_vnode)
#define NFSTOV(np) ((np)->n_vnode)
/*
* Queue head for nfsiod's
@ -235,6 +237,8 @@ int nfs_bwrite __P((void *));
#define nfs_vfree genfs_nullop
int nfs_truncate __P((void *));
int nfs_update __P((void *));
int nfs_getpages __P((void *));
int nfs_putpages __P((void *));
extern int (**nfsv2_vnodeop_p) __P((void *));

15
sys/sys/buf.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: buf.h,v 1.43 2000/04/10 02:22:15 chs Exp $ */
/* $NetBSD: buf.h,v 1.44 2000/11/27 08:39:51 chs Exp $ */
/*-
* Copyright (c) 1999, 2000 The NetBSD Foundation, Inc.
@ -190,13 +190,8 @@ struct buf {
number (not partition relative) */
/* Function to call upon completion. */
void (*b_iodone) __P((struct buf *));
struct vnode *b_vp; /* Device vnode. */
int b_dirtyoff; /* Offset in buffer of dirty region. */
int b_dirtyend; /* Offset of end of dirty region. */
struct ucred *b_rcred; /* Read credentials reference. */
struct ucred *b_wcred; /* Write credentials reference. */
int b_validoff; /* Offset in buffer of valid region. */
int b_validend; /* Offset of end of valid region. */
struct vnode *b_vp; /* File vnode. */
void *b_private; /* Private data for owner */
off_t b_dcookie; /* Offset cookie if dir block */
struct workhead b_dep; /* List of filesystem dependencies. */
};
@ -230,15 +225,16 @@ struct buf {
#define B_LOCKED 0x00004000 /* Locked in core (not reusable). */
#define B_NOCACHE 0x00008000 /* Do not cache block after use. */
#define B_ORDERED 0x00010000 /* ordered I/O request */
#define B_CACHE 0x00020000 /* Bread found us in the cache. */
#define B_PHYS 0x00040000 /* I/O to user memory. */
#define B_RAW 0x00080000 /* Set by physio for raw transfers. */
#define B_READ 0x00100000 /* Read buffer. */
#define B_TAPE 0x00200000 /* Magnetic tape I/O. */
#define B_WANTED 0x00800000 /* Process wants this buffer. */
#define B_WRITE 0x00000000 /* Write buffer (pseudo flag). */
#define B_WRITEINPROG 0x01000000 /* Write in progress. */
#define B_XXX 0x02000000 /* Debugging flag. */
#define B_VFLUSH 0x04000000 /* Buffer is being synced. */
#define B_PDAEMON 0x10000000 /* I/O initiated by pagedaemon. */
/*
* This structure describes a clustered I/O. It is stored in the b_saveaddr
@ -268,6 +264,7 @@ do { \
#define B_SYNC 0x02 /* Do all allocations synchronously. */
#ifdef _KERNEL
extern int nbuf; /* The number of buffer headers */
extern struct buf *buf; /* The buffer headers. */
extern char *buffers; /* The buffer contents. */

4
sys/sys/mount.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: mount.h,v 1.86 2000/06/26 21:10:34 christos Exp $ */
/* $NetBSD: mount.h,v 1.87 2000/11/27 08:39:52 chs Exp $ */
/*
* Copyright (c) 1989, 1991, 1993
@ -131,6 +131,8 @@ struct mount {
struct lock mnt_lock; /* mount structure lock */
int mnt_flag; /* flags */
int mnt_maxsymlinklen; /* max size of short symlink */
int mnt_fs_bshift; /* offset shift for lblkno */
int mnt_dev_bshift; /* shift for device sectors */
struct statfs mnt_stat; /* cache of filesystem stats */
qaddr_t mnt_data; /* private data */
int mnt_wcnt; /* count of vfs_busy waiters */

13
sys/sys/param.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: param.h,v 1.111 2000/11/11 00:53:24 thorpej Exp $ */
/* $NetBSD: param.h,v 1.112 2000/11/27 08:39:52 chs Exp $ */
/*-
* Copyright (c) 1982, 1986, 1989, 1993
@ -252,4 +252,15 @@
#define FSHIFT 11 /* bits to right of fixed binary point */
#define FSCALE (1<<FSHIFT)
/*
* Defaults for Unified Buffer Cache parameters.
*/
#ifndef UBC_WINSIZE
#define UBC_WINSIZE 8192
#endif
#ifndef UBC_NWINS
#define UBC_NWINS 1024
#endif
#endif /* !_SYS_PARAM_H_ */

22
sys/sys/vnode.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: vnode.h,v 1.83 2000/07/09 00:59:05 mycroft Exp $ */
/* $NetBSD: vnode.h,v 1.84 2000/11/27 08:39:52 chs Exp $ */
/*
* Copyright (c) 1989, 1993
@ -85,8 +85,10 @@ LIST_HEAD(buflists, buf);
*/
struct vnode {
struct uvm_vnode v_uvm; /* uvm data */
u_long v_flag; /* vnode flags (see below) */
long v_usecount; /* reference count of users */
#define v_flag v_uvm.u_flags
#define v_usecount v_uvm.u_obj.uo_refs
#define v_interlock v_uvm.u_obj.vmobjlock
#define v_numoutput v_uvm.u_nio
long v_writecount; /* reference count of writers */
long v_holdcnt; /* page & buffer references */
daddr_t v_lastr; /* last read (read-ahead) */
@ -97,13 +99,11 @@ struct vnode {
LIST_ENTRY(vnode) v_mntvnodes; /* vnodes for mount point */
struct buflists v_cleanblkhd; /* clean blocklist head */
struct buflists v_dirtyblkhd; /* dirty blocklist head */
long v_numoutput; /* num of writes in progress */
LIST_ENTRY(vnode) v_synclist; /* vnodes with dirty buffers */
enum vtype v_type; /* vnode type */
union {
struct mount *vu_mountedhere;/* ptr to mounted vfs (VDIR) */
struct socket *vu_socket; /* unix ipc (VSOCK) */
caddr_t vu_vmdata; /* private data for vm (VREG) */
struct specinfo *vu_specinfo; /* device (VCHR, VBLK) */
struct fifoinfo *vu_fifoinfo; /* fifo (VFIFO) */
} v_un;
@ -114,15 +114,14 @@ struct vnode {
int v_clen; /* length of current cluster */
int v_ralen; /* Read-ahead length */
daddr_t v_maxra; /* last readahead block */
struct simplelock v_interlock; /* lock on usecount and flag */
struct lock v_lock; /* lock for this vnode */
struct lock v_glock; /* getpage lock */
struct lock *v_vnlock; /* pointer to vnode lock */
enum vtagtype v_tag; /* type of underlying data */
void *v_data; /* private data for fs */
};
#define v_mountedhere v_un.vu_mountedhere
#define v_socket v_un.vu_socket
#define v_vmdata v_un.vu_vmdata
#define v_specinfo v_un.vu_specinfo
#define v_fifoinfo v_un.vu_fifoinfo
/*
@ -143,7 +142,9 @@ struct vnode {
*/
#define VROOT 0x0001 /* root of its file system */
#define VTEXT 0x0002 /* vnode is a pure text prototype */
/* VSYSTEM only used to skip vflush()ing quota files */
#define VSYSTEM 0x0004 /* vnode being used by kernel */
/* VISTTY used when reading dead vnodes */
#define VISTTY 0x0008 /* vnode represents a tty */
#define VXLOCK 0x0100 /* vnode is locked to change underlying type */
#define VXWANT 0x0200 /* process is waiting for vnode */
@ -152,6 +153,9 @@ struct vnode {
#define VDIROP 0x1000 /* LFS: vnode is involved in a directory op */
#define VLAYER 0x2000 /* vnode is on a layer filesystem */
#define VONWORKLST 0x4000 /* On syncer work-list */
#define VDIRTY 0x8000 /* vnode possibly has dirty pages */
#define VSIZENOTSET ((voff_t)-1)
/*
* Vnode attributes. A field value of VNOVAL represents a field whose value
@ -453,6 +457,10 @@ struct vop_generic_args {
* vclean changes the ops vector and then wants to call ops with the old
* vector.
*/
/*
* actually, vclean doesn't use it anymore, but nfs does,
* for device specials and fifos.
*/
#define VOCALL(OPSV,OFF,AP) (( *((OPSV)[(OFF)])) (AP))
/*

268
sys/ufs/ext2fs/ext2fs_balloc.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ext2fs_balloc.c,v 1.6 2000/06/28 14:16:37 mrg Exp $ */
/* $NetBSD: ext2fs_balloc.c,v 1.7 2000/11/27 08:39:53 chs Exp $ */
/*
* Copyright (c) 1997 Manuel Bouyer.
@ -37,12 +37,19 @@
* Modified for ext2fs by Manuel Bouyer.
*/
#if defined(_KERNEL) && !defined(_LKM)
#include "opt_uvmhist.h"
#endif
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/buf.h>
#include <sys/proc.h>
#include <sys/file.h>
#include <sys/vnode.h>
#include <sys/mount.h>
#include <uvm/uvm.h>
#include <ufs/ufs/quota.h>
#include <ufs/ufs/inode.h>
@ -75,8 +82,13 @@ ext2fs_balloc(ip, bn, size, cred, bpp, flags)
u_int deallocated;
ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
int unwindidx = -1;
UVMHIST_FUNC("ext2fs_balloc"); UVMHIST_CALLED(ubchist);
*bpp = NULL;
UVMHIST_LOG(ubchist, "bn 0x%x", bn,0,0,0);
if (bpp != NULL) {
*bpp = NULL;
}
if (bn < 0)
return (EFBIG);
fs = ip->i_e2fs;
@ -88,29 +100,43 @@ ext2fs_balloc(ip, bn, size, cred, bpp, flags)
if (bn < NDADDR) {
nb = fs2h32(ip->i_e2fs_blocks[bn]);
if (nb != 0) {
error = bread(vp, bn, fs->e2fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
/*
* the block is already allocated, just read it.
*/
if (bpp != NULL) {
error = bread(vp, bn, fs->e2fs_bsize, NOCRED,
&bp);
if (error) {
brelse(bp);
return (error);
}
*bpp = bp;
}
*bpp = bp;
return (0);
} else {
error = ext2fs_alloc(ip, bn,
ext2fs_blkpref(ip, bn, (int)bn, &ip->i_e2fs_blocks[0]),
cred, &newb);
if (error)
return (error);
ip->i_e2fs_last_lblk = lbn;
ip->i_e2fs_last_blk = newb;
}
/*
* allocate a new direct block.
*/
error = ext2fs_alloc(ip, bn,
ext2fs_blkpref(ip, bn, bn, &ip->i_e2fs_blocks[0]),
cred, &newb);
if (error)
return (error);
ip->i_e2fs_last_lblk = lbn;
ip->i_e2fs_last_blk = newb;
ip->i_e2fs_blocks[bn] = h2fs32(newb);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (bpp != NULL) {
bp = getblk(vp, bn, fs->e2fs_bsize, 0, 0);
bp->b_blkno = fsbtodb(fs, newb);
if (flags & B_CLRBUF)
clrbuf(bp);
*bpp = bp;
}
ip->i_e2fs_blocks[bn] = h2fs32(dbtofsb(fs, bp->b_blkno));
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*bpp = bp;
return (0);
}
/*
@ -218,10 +244,6 @@ ext2fs_balloc(ip, bn, size, cred, bpp, flags)
*allocblk++ = nb;
ip->i_e2fs_last_lblk = lbn;
ip->i_e2fs_last_blk = newb;
nbp = getblk(vp, lbn, fs->e2fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & B_CLRBUF)
clrbuf(nbp);
bap[indirs[num].in_off] = h2fs32(nb);
/*
* If required, write synchronously, otherwise use
@ -232,21 +254,30 @@ ext2fs_balloc(ip, bn, size, cred, bpp, flags)
} else {
bdwrite(bp);
}
*bpp = nbp;
if (bpp != NULL) {
nbp = getblk(vp, lbn, fs->e2fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & B_CLRBUF)
clrbuf(nbp);
*bpp = nbp;
}
return (0);
}
brelse(bp);
if (flags & B_CLRBUF) {
error = bread(vp, lbn, (int)fs->e2fs_bsize, NOCRED, &nbp);
if (error) {
brelse(nbp);
goto fail;
if (bpp != NULL) {
if (flags & B_CLRBUF) {
error = bread(vp, lbn, (int)fs->e2fs_bsize, NOCRED,
&nbp);
if (error) {
brelse(nbp);
goto fail;
}
} else {
nbp = getblk(vp, lbn, fs->e2fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
}
} else {
nbp = getblk(vp, lbn, fs->e2fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
*bpp = nbp;
}
*bpp = nbp;
return (0);
fail:
/*
@ -288,5 +319,178 @@ fail:
ip->i_e2fs_nblock -= btodb(deallocated);
ip->i_e2fs_flags |= IN_CHANGE | IN_UPDATE;
}
return error;
}
int
ext2fs_ballocn(v)
void *v;
{
struct vop_ballocn_args /* {
struct vnode *a_vp;
off_t a_offset;
off_t a_length;
struct ucred *a_cred;
int a_flags;
} */ *ap = v;
off_t off, len;
struct vnode *vp = ap->a_vp;
struct inode *ip = VTOI(vp);
struct m_ext2fs *fs = ip->i_e2fs;
int error, delta, bshift, bsize;
UVMHIST_FUNC("ext2fs_ballocn"); UVMHIST_CALLED(ubchist);
bshift = fs->e2fs_bshift;
bsize = 1 << bshift;
off = ap->a_offset;
len = ap->a_length;
delta = off & (bsize - 1);
off -= delta;
len += delta;
while (len > 0) {
bsize = min(bsize, len);
UVMHIST_LOG(ubchist, "off 0x%x len 0x%x bsize 0x%x",
off, len, bsize, 0);
error = ext2fs_balloc(ip, lblkno(fs, off), bsize, ap->a_cred,
NULL, ap->a_flags);
if (error) {
UVMHIST_LOG(ubchist, "error %d", error, 0,0,0);
return error;
}
/*
* increase file size now, VOP_BALLOC() requires that
* EOF be up-to-date before each call.
*/
if (ip->i_e2fs_size < off + bsize) {
UVMHIST_LOG(ubchist, "old 0x%x new 0x%x",
ip->i_e2fs_size, off + bsize,0,0);
ip->i_e2fs_size = off + bsize;
if (vp->v_uvm.u_size < ip->i_e2fs_size) {
uvm_vnp_setsize(vp, ip->i_e2fs_size);
}
}
off += bsize;
len -= bsize;
}
return 0;
}
/*
* allocate a range of blocks in a file.
* after this function returns, any page entirely contained within the range
* will map to invalid data and thus must be overwritten before it is made
* accessible to others.
*/
int
ext2fs_balloc_range(vp, off, len, cred, flags)
struct vnode *vp;
off_t off, len;
struct ucred *cred;
int flags;
{
off_t eof, pagestart, pageend;
struct uvm_object *uobj;
struct inode *ip = VTOI(vp);
int i, delta, error, npages1, npages2;
int bshift = vp->v_mount->mnt_fs_bshift;
int bsize = 1 << bshift;
int ppb = max(bsize >> PAGE_SHIFT, 1);
struct vm_page *pgs1[ppb], *pgs2[ppb];
UVMHIST_FUNC("ext2fs_balloc_range"); UVMHIST_CALLED(ubchist);
UVMHIST_LOG(ubchist, "vp %p off 0x%x len 0x%x u_size 0x%x",
vp, off, len, vp->v_uvm.u_size);
error = 0;
uobj = &vp->v_uvm.u_obj;
eof = max(vp->v_uvm.u_size, off + len);
vp->v_uvm.u_size = eof;
UVMHIST_LOG(ubchist, "new eof 0x%x", eof,0,0,0);
pgs1[0] = pgs2[0] = NULL;
/*
* if the range does not start on a page and block boundary,
* cache the first block if the file so the page(s) will contain
* the correct data. hold the page(s) busy while we allocate
* the backing store for the range.
*/
pagestart = trunc_page(off) & ~(bsize - 1);
if (off != pagestart) {
npages1 = min(ppb, (round_page(eof) - pagestart) >>
PAGE_SHIFT);
memset(pgs1, 0, npages1);
simple_lock(&uobj->vmobjlock);
error = VOP_GETPAGES(vp, pagestart, pgs1, &npages1, 0,
VM_PROT_READ, 0, PGO_SYNCIO);
if (error) {
UVMHIST_LOG(ubchist, "gp1 %d", error,0,0,0);
goto errout;
}
for (i = 0; i < npages1; i++) {
UVMHIST_LOG(ubchist, "got pgs1[%d] %p", i, pgs1[i],0,0);
}
}
/*
* similarly if the range does not end on a page and block boundary.
*/
pageend = trunc_page(off + len) & ~(bsize - 1);
if (off + len < ip->i_e2fs_size &&
off + len != pageend &&
pagestart != pageend) {
npages2 = min(ppb, (round_page(eof) - pageend) >>
PAGE_SHIFT);
memset(pgs2, 0, npages2);
simple_lock(&uobj->vmobjlock);
error = VOP_GETPAGES(vp, pageend, pgs2, &npages2, 0,
VM_PROT_READ, 0, PGO_SYNCIO);
if (error) {
UVMHIST_LOG(ubchist, "gp2 %d", error,0,0,0);
goto errout;
}
for (i = 0; i < npages2; i++) {
UVMHIST_LOG(ubchist, "got pgs2[%d] %p", i, pgs2[i],0,0);
}
}
/*
* adjust off to be block-aligned.
*/
delta = off & (bsize - 1);
off -= delta;
len += delta;
/*
* now allocate the range.
*/
lockmgr(&vp->v_glock, LK_EXCLUSIVE, NULL);
error = VOP_BALLOCN(vp, off, len, cred, flags);
UVMHIST_LOG(ubchist, "ballocn %d", error,0,0,0);
lockmgr(&vp->v_glock, LK_RELEASE, NULL);
/*
* unbusy any pages we are holding.
*/
errout:
simple_lock(&uobj->vmobjlock);
if (pgs1[0] != NULL) {
uvm_page_unbusy(pgs1, npages1);
}
if (pgs2[0] != NULL) {
uvm_page_unbusy(pgs2, npages2);
}
simple_unlock(&uobj->vmobjlock);
return (error);
}

5
sys/ufs/ext2fs/ext2fs_extern.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: ext2fs_extern.h,v 1.8 2000/03/16 18:08:32 jdolecek Exp $ */
/* $NetBSD: ext2fs_extern.h,v 1.9 2000/11/27 08:39:53 chs Exp $ */
/*-
* Copyright (c) 1997 Manuel Bouyer.
@ -71,6 +71,9 @@ int ext2fs_vfree __P((void *));
/* ext2fs_balloc.c */
int ext2fs_balloc __P((struct inode *, ufs_daddr_t, int, struct ucred *,
struct buf **, int));
int ext2fs_ballocn __P((void *));
int ext2fs_balloc_range __P((struct vnode *, off_t, off_t, struct ucred *,
int));
/* ext2fs_bmap.c */
int ext2fs_bmap __P((void *));

67
sys/ufs/ext2fs/ext2fs_inode.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ext2fs_inode.c,v 1.20 2000/06/28 14:16:37 mrg Exp $ */
/* $NetBSD: ext2fs_inode.c,v 1.21 2000/11/27 08:39:53 chs Exp $ */
/*
* Copyright (c) 1997 Manuel Bouyer.
@ -101,7 +101,7 @@ out:
* so that it can be reused immediately.
*/
if (ip->i_e2fs_dtime != 0)
vrecycle(vp, (struct simplelock *)0, p);
vrecycle(vp, NULL, p);
return (error);
}
@ -187,15 +187,14 @@ ext2fs_truncate(v)
struct vnode *ovp = ap->a_vp;
ufs_daddr_t lastblock;
struct inode *oip;
ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
ufs_daddr_t bn, lastiblock[NIADDR], indir_lbn[NIADDR];
ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
off_t length = ap->a_length;
struct m_ext2fs *fs;
struct buf *bp;
int offset, size, level;
long count, nblocks, blocksreleased = 0;
int i;
int aflags, error, allerror = 0;
int error, allerror = 0;
off_t osize;
if (length < 0)
@ -232,24 +231,10 @@ ext2fs_truncate(v)
if (length > fs->fs_maxfilesize)
return (EFBIG);
#endif
offset = blkoff(fs, length - 1);
lbn = lblkno(fs, length - 1);
aflags = B_CLRBUF;
if (ap->a_flags & IO_SYNC)
aflags |= B_SYNC;
error = ext2fs_balloc(oip, lbn, offset + 1, ap->a_cred, &bp,
aflags);
if (error)
return (error);
oip->i_e2fs_size = length;
uvm_vnp_setsize(ovp, length);
(void) uvm_vnp_uncache(ovp);
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
ext2fs_balloc_range(ovp, length - 1, 1, ap->a_cred,
ap->a_flags & IO_SYNC ? B_SYNC : 0);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (VOP_UPDATE(ovp, NULL, NULL, UPDATE_WAIT));
return (VOP_UPDATE(ovp, NULL, NULL, 1));
}
/*
* Shorten the size of the file. If the file is not being
@ -259,26 +244,13 @@ ext2fs_truncate(v)
* of subsequent file growth.
*/
offset = blkoff(fs, length);
if (offset == 0) {
oip->i_e2fs_size = length;
} else {
lbn = lblkno(fs, length);
aflags = B_CLRBUF;
if (ap->a_flags & IO_SYNC)
aflags |= B_SYNC;
error = ext2fs_balloc(oip, lbn, offset, ap->a_cred, &bp, aflags);
if (error)
return (error);
oip->i_e2fs_size = length;
if (offset != 0) {
size = fs->e2fs_bsize;
(void) uvm_vnp_uncache(ovp);
memset((char *)bp->b_data + offset, 0, (u_int)(size - offset));
allocbuf(bp, size);
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
/* XXXUBC we should handle more than just VREG */
uvm_vnp_zerorange(ovp, length, size - offset);
}
oip->i_e2fs_size = length;
uvm_vnp_setsize(ovp, length);
/*
@ -317,6 +289,7 @@ ext2fs_truncate(v)
* Note that we save the new block configuration so we can check it
* when we are done.
*/
memcpy((caddr_t)newblks, (caddr_t)&oip->i_e2fs_blocks[0], sizeof newblks);
memcpy((caddr_t)&oip->i_e2fs_blocks[0], (caddr_t)oldblks, sizeof oldblks);
oip->i_e2fs_size = osize;
@ -359,20 +332,20 @@ ext2fs_truncate(v)
ext2fs_blkfree(oip, bn);
blocksreleased += btodb(fs->e2fs_bsize);
}
if (lastblock < 0)
goto done;
done:
#ifdef DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[NDADDR + level] != oip->i_e2fs_blocks[NDADDR + level])
panic("itrunc1");
if (newblks[NDADDR + level] !=
oip->i_e2fs_blocks[NDADDR + level])
panic("ext2fs_truncate1");
for (i = 0; i < NDADDR; i++)
if (newblks[i] != oip->i_e2fs_blocks[i])
panic("itrunc2");
panic("ext2fs_truncate2");
if (length == 0 &&
(!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
panic("itrunc3");
(!LIST_EMPTY(&ovp->v_cleanblkhd) ||
!LIST_EMPTY(&ovp->v_dirtyblkhd)))
panic("ext2fs_truncate3");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.

107
sys/ufs/ext2fs/ext2fs_readwrite.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ext2fs_readwrite.c,v 1.13 2000/06/28 14:16:38 mrg Exp $ */
/* $NetBSD: ext2fs_readwrite.c,v 1.14 2000/11/27 08:39:53 chs Exp $ */
/*-
* Copyright (c) 1997 Manuel Bouyer.
@ -79,6 +79,8 @@ ext2fs_read(v)
struct uio *uio;
struct m_ext2fs *fs;
struct buf *bp;
void *win;
vsize_t bytelen;
ufs_daddr_t lbn, nextlbn;
off_t bytesinfile;
long size, xfersize, blkoffset;
@ -107,6 +109,27 @@ ext2fs_read(v)
if (uio->uio_resid == 0)
return (0);
if (vp->v_type == VREG) {
error = 0;
while (uio->uio_resid > 0) {
bytelen = min(ip->i_e2fs_size - uio->uio_offset,
uio->uio_resid);
if (bytelen == 0) {
break;
}
win = ubc_alloc(&vp->v_uvm.u_obj, uio->uio_offset,
&bytelen, UBC_READ);
error = uiomove(win, bytelen, uio);
ubc_release(win, 0);
if (error) {
break;
}
}
goto out;
}
for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
if ((bytesinfile = ip->i_e2fs_size - uio->uio_offset) <= 0)
break;
@ -148,14 +171,15 @@ ext2fs_read(v)
break;
xfersize = size;
}
error = uiomove((char *)bp->b_data + blkoffset, (int)xfersize,
uio);
error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio);
if (error)
break;
brelse(bp);
}
if (bp != NULL)
brelse(bp);
out:
if (!(vp->v_mount->mnt_flag & MNT_NOATIME)) {
ip->i_flag |= IN_ACCESS;
if ((ap->a_ioflag & IO_SYNC) == IO_SYNC)
@ -185,12 +209,17 @@ ext2fs_write(v)
struct proc *p;
ufs_daddr_t lbn;
off_t osize;
int blkoffset, error, flags, ioflag, resid, size, xfersize;
int blkoffset, error, flags, ioflag, resid, xfersize;
vsize_t bytelen;
void *win;
off_t oldoff;
boolean_t rv;
ioflag = ap->a_ioflag;
uio = ap->a_uio;
vp = ap->a_vp;
ip = VTOI(vp);
error = 0;
#ifdef DIAGNOSTIC
if (uio->uio_rw != UIO_WRITE)
@ -234,35 +263,66 @@ ext2fs_write(v)
resid = uio->uio_resid;
osize = ip->i_e2fs_size;
flags = ioflag & IO_SYNC ? B_SYNC : 0;
if (vp->v_type == VREG) {
while (uio->uio_resid > 0) {
oldoff = uio->uio_offset;
blkoffset = blkoff(fs, uio->uio_offset);
bytelen = min(fs->e2fs_bsize - blkoffset,
uio->uio_resid);
/*
* XXXUBC if file is mapped and this is the last block,
* process one page at a time.
*/
error = ext2fs_balloc_range(vp, uio->uio_offset,
bytelen, ap->a_cred, 0);
if (error) {
break;
}
win = ubc_alloc(&vp->v_uvm.u_obj, uio->uio_offset,
&bytelen, UBC_WRITE);
error = uiomove(win, bytelen, uio);
ubc_release(win, 0);
if (error) {
break;
}
/*
* flush what we just wrote if necessary.
* XXXUBC simplistic async flushing.
*/
if (oldoff >> 16 != uio->uio_offset >> 16) {
simple_lock(&vp->v_uvm.u_obj.vmobjlock);
rv = vp->v_uvm.u_obj.pgops->pgo_flush(
&vp->v_uvm.u_obj, (oldoff >> 16) << 16,
(uio->uio_offset >> 16) << 16, PGO_CLEANIT);
simple_unlock(&vp->v_uvm.u_obj.vmobjlock);
}
}
goto out;
}
flags = ioflag & IO_SYNC ? B_SYNC : 0;
for (error = 0; uio->uio_resid > 0;) {
lbn = lblkno(fs, uio->uio_offset);
blkoffset = blkoff(fs, uio->uio_offset);
xfersize = fs->e2fs_bsize - blkoffset;
if (uio->uio_resid < xfersize)
xfersize = uio->uio_resid;
if (fs->e2fs_bsize > xfersize)
xfersize = min(fs->e2fs_bsize - blkoffset, uio->uio_resid);
if (xfersize < fs->e2fs_bsize)
flags |= B_CLRBUF;
else
flags &= ~B_CLRBUF;
error = ext2fs_balloc(ip,
lbn, blkoffset + xfersize, ap->a_cred, &bp, flags);
error = VOP_BALLOC(vp, lblktosize(fs, lbn),
blkoffset + xfersize, ap->a_cred, flags,
&bp);
if (error)
break;
if (uio->uio_offset + xfersize > ip->i_e2fs_size) {
if (ip->i_e2fs_size < uio->uio_offset + xfersize) {
ip->i_e2fs_size = uio->uio_offset + xfersize;
uvm_vnp_setsize(vp, ip->i_e2fs_size);
}
(void)uvm_vnp_uncache(vp);
size = fs->e2fs_bsize - bp->b_resid;
if (size < xfersize)
xfersize = size;
error =
uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio);
if (ioflag & IO_SYNC)
(void)bwrite(bp);
else if (xfersize + blkoffset == fs->e2fs_bsize)
@ -274,13 +334,14 @@ ext2fs_write(v)
bdwrite(bp);
if (error || xfersize == 0)
break;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* If we successfully wrote any data, and we are not the superuser
* we clear the setuid and setgid bits as a precaution against
* tampering.
*/
out:
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (resid > uio->uio_resid && ap->a_cred && ap->a_cred->cr_uid != 0)
ip->i_e2fs_mode &= ~(ISUID | ISGID);
if (error) {

7
sys/ufs/ext2fs/ext2fs_vfsops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ext2fs_vfsops.c,v 1.39 2000/09/19 22:03:05 fvdl Exp $ */
/* $NetBSD: ext2fs_vfsops.c,v 1.40 2000/11/27 08:39:53 chs Exp $ */
/*
* Copyright (c) 1997 Manuel Bouyer.
@ -592,15 +592,19 @@ ext2fs_mountfs(devvp, mp, p)
mp->mnt_stat.f_fsid.val[1] = makefstype(MOUNT_EXT2FS);
mp->mnt_maxsymlinklen = EXT2_MAXSYMLINKLEN;
mp->mnt_flag |= MNT_LOCAL;
mp->mnt_dev_bshift = DEV_BSHIFT; /* XXX */
mp->mnt_fs_bshift = m_fs->e2fs_bshift;
ump->um_flags = 0;
ump->um_mountp = mp;
ump->um_dev = dev;
ump->um_devvp = devvp;
ump->um_nindir = NINDIR(m_fs);
ump->um_lognindir = ffs(NINDIR(m_fs)) - 1;
ump->um_bptrtodb = m_fs->e2fs_fsbtodb;
ump->um_seqinc = 1; /* no frags */
devvp->v_specmountpoint = mp;
return (0);
out:
if (bp)
brelse(bp);
@ -931,6 +935,7 @@ ext2fs_vget(mp, ino, vpp)
ip->i_flag |= IN_MODIFIED;
}
vp->v_uvm.u_size = ip->i_e2fs_size;
*vpp = vp;
return (0);
}

15
sys/ufs/ext2fs/ext2fs_vnops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ext2fs_vnops.c,v 1.29 2000/08/03 20:41:36 thorpej Exp $ */
/* $NetBSD: ext2fs_vnops.c,v 1.30 2000/11/27 08:39:53 chs Exp $ */
/*
* Copyright (c) 1997 Manuel Bouyer.
@ -196,7 +196,6 @@ ext2fs_access(v)
struct inode *ip = VTOI(vp);
mode_t mode = ap->a_mode;
/*
* Disallow write attempts on read-only file systems;
* unless the file is a socket, fifo, or a block or
@ -421,8 +420,6 @@ ext2fs_chmod(vp, mode, cred, p)
ip->i_e2fs_mode &= ~ALLPERMS;
ip->i_e2fs_mode |= (mode & ALLPERMS);
ip->i_flag |= IN_CHANGE;
if ((vp->v_flag & VTEXT) && (ip->i_e2fs_mode & S_ISTXT) == 0)
(void) uvm_vnp_uncache(vp);
return (0);
}
@ -1465,7 +1462,11 @@ struct vnodeopv_entry_desc ext2fs_vnodeop_entries[] = {
{ &vop_truncate_desc, ext2fs_truncate }, /* truncate */
{ &vop_update_desc, ext2fs_update }, /* update */
{ &vop_bwrite_desc, vn_bwrite }, /* bwrite */
{ (struct vnodeop_desc*)NULL, (int(*) __P((void*)))NULL }
{ &vop_ballocn_desc, ext2fs_ballocn }, /* ballocn */
{ &vop_getpages_desc, genfs_getpages }, /* getpages */
{ &vop_putpages_desc, genfs_putpages }, /* putpages */
{ &vop_size_desc, genfs_size }, /* size */
{ NULL, NULL }
};
struct vnodeopv_desc ext2fs_vnodeop_opv_desc =
{ &ext2fs_vnodeop_p, ext2fs_vnodeop_entries };
@ -1516,7 +1517,7 @@ struct vnodeopv_entry_desc ext2fs_specop_entries[] = {
{ &vop_truncate_desc, spec_truncate }, /* truncate */
{ &vop_update_desc, ext2fs_update }, /* update */
{ &vop_bwrite_desc, vn_bwrite }, /* bwrite */
{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
{ NULL, NULL }
};
struct vnodeopv_desc ext2fs_specop_opv_desc =
{ &ext2fs_specop_p, ext2fs_specop_entries };
@ -1567,7 +1568,7 @@ struct vnodeopv_entry_desc ext2fs_fifoop_entries[] = {
{ &vop_truncate_desc, fifo_truncate }, /* truncate */
{ &vop_update_desc, ext2fs_update }, /* update */
{ &vop_bwrite_desc, vn_bwrite }, /* bwrite */
{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
{ NULL, NULL }
};
struct vnodeopv_desc ext2fs_fifoop_opv_desc =
{ &ext2fs_fifoop_p, ext2fs_fifoop_entries };

100
sys/ufs/ffs/ffs_alloc.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ffs_alloc.c,v 1.36 2000/06/28 14:16:39 mrg Exp $ */
/* $NetBSD: ffs_alloc.c,v 1.37 2000/11/27 08:39:54 chs Exp $ */
/*
* Copyright (c) 1982, 1986, 1989, 1993
@ -109,15 +109,33 @@ ffs_alloc(ip, lbn, bpref, size, cred, bnp)
struct ucred *cred;
ufs_daddr_t *bnp;
{
struct fs *fs;
struct fs *fs = ip->i_fs;
ufs_daddr_t bno;
int cg;
#ifdef QUOTA
int error;
#endif
#ifdef UVM_PAGE_TRKOWN
if (ITOV(ip)->v_type == VREG && lbn > 0) {
struct vm_page *pg;
struct uvm_object *uobj = &ITOV(ip)->v_uvm.u_obj;
voff_t off = trunc_page(lblktosize(fs, lbn));
voff_t endoff = round_page(lblktosize(fs, lbn) + size);
simple_lock(&uobj->vmobjlock);
while (off < endoff) {
pg = uvm_pagelookup(uobj, off);
KASSERT(pg != NULL);
KASSERT(pg->owner == curproc->p_pid);
KASSERT((pg->flags & PG_CLEAN) == 0);
off += PAGE_SIZE;
}
simple_unlock(&uobj->vmobjlock);
}
#endif
*bnp = 0;
fs = ip->i_fs;
#ifdef DIAGNOSTIC
if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
printf("dev = 0x%x, bsize = %d, size = %d, fs = %s\n",
@ -170,21 +188,39 @@ nospace:
* invoked to get an appropriate block.
*/
int
ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp, blknop)
struct inode *ip;
ufs_daddr_t lbprev;
ufs_daddr_t bpref;
int osize, nsize;
struct ucred *cred;
struct buf **bpp;
ufs_daddr_t *blknop;
{
struct fs *fs;
struct fs *fs = ip->i_fs;
struct buf *bp;
int cg, request, error;
ufs_daddr_t bprev, bno;
*bpp = 0;
fs = ip->i_fs;
#ifdef UVM_PAGE_TRKOWN
if (ITOV(ip)->v_type == VREG) {
struct vm_page *pg;
struct uvm_object *uobj = &ITOV(ip)->v_uvm.u_obj;
voff_t off = trunc_page(lblktosize(fs, lbprev));
voff_t endoff = round_page(lblktosize(fs, lbprev) + osize);
simple_lock(&uobj->vmobjlock);
while (off < endoff) {
pg = uvm_pagelookup(uobj, off);
KASSERT(pg != NULL);
KASSERT(pg->owner == curproc->p_pid);
KASSERT((pg->flags & PG_CLEAN) == 0);
off += PAGE_SIZE;
}
simple_unlock(&uobj->vmobjlock);
}
#endif
#ifdef DIAGNOSTIC
if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
(u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
@ -206,7 +242,8 @@ ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
/*
* Allocate the extra space in the buffer.
*/
if ((error = bread(ITOV(ip), lbprev, osize, NOCRED, &bp)) != 0) {
if (bpp != NULL &&
(error = bread(ITOV(ip), lbprev, osize, NOCRED, &bp)) != 0) {
brelse(bp);
return (error);
}
@ -221,14 +258,20 @@ ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
*/
cg = dtog(fs, bprev);
if ((bno = ffs_fragextend(ip, cg, (long)bprev, osize, nsize)) != 0) {
if (bp->b_blkno != fsbtodb(fs, bno))
panic("bad blockno");
ip->i_ffs_blocks += btodb(nsize - osize);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
allocbuf(bp, nsize);
bp->b_flags |= B_DONE;
memset((char *)bp->b_data + osize, 0, (u_int)nsize - osize);
*bpp = bp;
if (bpp != NULL) {
if (bp->b_blkno != fsbtodb(fs, bno))
panic("bad blockno");
allocbuf(bp, nsize);
bp->b_flags |= B_DONE;
memset(bp->b_data + osize, 0, nsize - osize);
*bpp = bp;
}
if (blknop != NULL) {
*blknop = bno;
}
return (0);
}
/*
@ -292,8 +335,6 @@ ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request,
ffs_alloccg);
if (bno > 0) {
bp->b_blkno = fsbtodb(fs, bno);
(void) uvm_vnp_uncache(ITOV(ip));
if (!DOINGSOFTDEP(ITOV(ip)))
ffs_blkfree(ip, bprev, (long)osize);
if (nsize < request)
@ -301,10 +342,16 @@ ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
(long)(request - nsize));
ip->i_ffs_blocks += btodb(nsize - osize);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
allocbuf(bp, nsize);
bp->b_flags |= B_DONE;
memset((char *)bp->b_data + osize, 0, (u_int)nsize - osize);
*bpp = bp;
if (bpp != NULL) {
bp->b_blkno = fsbtodb(fs, bno);
allocbuf(bp, nsize);
bp->b_flags |= B_DONE;
memset(bp->b_data + osize, 0, (u_int)nsize - osize);
*bpp = bp;
}
if (blknop != NULL) {
*blknop = bno;
}
return (0);
}
#ifdef QUOTA
@ -313,7 +360,10 @@ ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
*/
(void) chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE);
#endif
brelse(bp);
if (bpp != NULL) {
brelse(bp);
}
nospace:
/*
* no space available
@ -344,7 +394,7 @@ struct ctldebug debug15 = { "prtrealloc", &prtrealloc };
#endif
int doasyncfree = 1;
extern int doreallocblks;
int doreallocblks;
int
ffs_reallocblks(v)
@ -364,6 +414,9 @@ ffs_reallocblks(v)
struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
int i, len, start_lvl, end_lvl, pref, ssize;
/* XXXUBC don't reallocblks for now */
return ENOSPC;
vp = ap->a_vp;
ip = VTOI(vp);
fs = ip->i_fs;
@ -1725,5 +1778,6 @@ ffs_fserr(fs, uid, cp)
char *cp;
{
log(LOG_ERR, "uid %d on %s: %s\n", uid, fs->fs_fsmnt, cp);
log(LOG_ERR, "uid %d comm %s on %s: %s\n",
uid, curproc->p_comm, fs->fs_fsmnt, cp);
}

221
sys/ufs/ffs/ffs_balloc.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ffs_balloc.c,v 1.22 2000/09/19 22:04:08 fvdl Exp $ */
/* $NetBSD: ffs_balloc.c,v 1.23 2000/11/27 08:39:54 chs Exp $ */
/*
* Copyright (c) 1982, 1986, 1989, 1993
@ -57,6 +57,8 @@
#include <ufs/ffs/fs.h>
#include <ufs/ffs/ffs_extern.h>
#include <uvm/uvm.h>
/*
* Balloc defines the structure of file system storage
* by allocating the physical blocks on a device given
@ -72,7 +74,7 @@ ffs_balloc(v)
int a_size;
struct ucred *a_cred;
int a_flags;
struct buf *a_bpp;
struct buf **a_bpp;
} */ *ap = v;
ufs_daddr_t lbn;
int size;
@ -88,15 +90,22 @@ ffs_balloc(v)
int deallocated, osize, nsize, num, i, error;
ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
int unwindidx = -1;
struct buf **bpp = ap->a_bpp;
#ifdef FFS_EI
const int needswap = UFS_FSNEEDSWAP(fs);
#endif
UVMHIST_FUNC("ffs_balloc"); UVMHIST_CALLED(ubchist);
lbn = lblkno(fs, ap->a_startoffset);
size = blkoff(fs, ap->a_startoffset) + ap->a_size;
if (size > fs->fs_bsize)
panic("ffs_balloc: blk too big");
*ap->a_bpp = NULL;
if (bpp != NULL) {
*bpp = NULL;
}
UVMHIST_LOG(ubchist, "vp %p lbn 0x%x size 0x%x", vp, lbn, size,0);
KASSERT(size <= fs->fs_bsize);
if (lbn < 0)
return (EFBIG);
cred = ap->a_cred;
@ -107,71 +116,109 @@ ffs_balloc(v)
* and the file is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
nb = lblkno(fs, ip->i_ffs_size);
if (nb < NDADDR && nb < lbn) {
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
error = ffs_realloccg(ip, nb,
ffs_blkpref(ip, nb, (int)nb, &ip->i_ffs_db[0]),
osize, (int)fs->fs_bsize, cred, &bp);
osize, (int)fs->fs_bsize, cred, bpp, &newb);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb,
dbtofsb(fs, bp->b_blkno),
softdep_setup_allocdirect(ip, nb, newb,
ufs_rw32(ip->i_ffs_db[nb], needswap),
fs->fs_bsize, osize, bp);
ip->i_ffs_size = (nb + 1) * fs->fs_bsize;
fs->fs_bsize, osize, bpp ? *bpp : NULL);
ip->i_ffs_size = lblktosize(fs, nb + 1);
uvm_vnp_setsize(vp, ip->i_ffs_size);
ip->i_ffs_db[nb] = ufs_rw32(dbtofsb(fs, bp->b_blkno),
needswap);
ip->i_ffs_db[nb] = ufs_rw32(newb, needswap);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (flags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
if (bpp) {
if (flags & B_SYNC)
bwrite(*bpp);
else
bawrite(*bpp);
}
}
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
nb = ufs_rw32(ip->i_ffs_db[lbn], needswap);
if (nb != 0 && ip->i_ffs_size >= (lbn + 1) * fs->fs_bsize) {
error = bread(vp, lbn, fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
if (nb != 0 && ip->i_ffs_size >= lblktosize(fs, lbn + 1)) {
/*
* The block is an already-allocated direct block
* and the file already extends past this block,
* thus this must be a whole block.
* Just read the block (if requested).
*/
if (bpp != NULL) {
error = bread(vp, lbn, fs->fs_bsize, NOCRED,
bpp);
if (error) {
brelse(*bpp);
return (error);
}
}
*ap->a_bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_ffs_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread(vp, lbn, osize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
/*
* The existing block is already
* at least as big as we want.
* Just read the block (if requested).
*/
if (bpp != NULL) {
error = bread(vp, lbn, osize, NOCRED,
bpp);
if (error) {
brelse(*bpp);
return (error);
}
}
return 0;
} else {
/*
* The existing block is smaller than we want,
* grow it.
*/
error = ffs_realloccg(ip, lbn,
ffs_blkpref(ip, lbn, (int)lbn,
&ip->i_ffs_db[0]), osize, nsize, cred,
&bp);
bpp, &newb);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
dbtofsb(fs, bp->b_blkno), nb,
nsize, osize, bp);
newb, nb, nsize, osize,
bpp ? *bpp : NULL);
}
} else {
if (ip->i_ffs_size < (lbn + 1) * fs->fs_bsize)
/*
* the block was not previously allocated,
* allocate a new block or fragment.
*/
if (ip->i_ffs_size < lblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
@ -180,18 +227,20 @@ ffs_balloc(v)
nsize, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lbn, nsize, 0, 0);
bp->b_blkno = fsbtodb(fs, newb);
if (flags & B_CLRBUF)
clrbuf(bp);
if (DOINGSOFTDEP(vp))
if (bpp != NULL) {
bp = getblk(vp, lbn, nsize, 0, 0);
bp->b_blkno = fsbtodb(fs, newb);
if (flags & B_CLRBUF)
clrbuf(bp);
*bpp = bp;
}
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bp);
nsize, 0, bpp ? *bpp : NULL);
}
}
ip->i_ffs_db[lbn] = ufs_rw32(dbtofsb(fs, bp->b_blkno),
needswap);
ip->i_ffs_db[lbn] = ufs_rw32(newb, needswap);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*ap->a_bpp = bp;
return (0);
}
/*
@ -200,6 +249,7 @@ ffs_balloc(v)
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef DIAGNOSTIC
if (num < 1)
panic ("ffs_balloc: ufs_bmaparray returned indirect block\n");
@ -311,14 +361,20 @@ ffs_balloc(v)
}
nb = newb;
*allocblk++ = nb;
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & B_CLRBUF)
clrbuf(nbp);
if (bpp != NULL) {
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & B_CLRBUF)
clrbuf(nbp);
*bpp = nbp;
}
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[num].in_off, nb, 0, nbp);
indirs[num].in_off, nb, 0, bpp ? *bpp : NULL);
bap[indirs[num].in_off] = ufs_rw32(nb, needswap);
if (allocib == NULL && unwindidx < 0) {
unwindidx = i - 1;
}
/*
* If required, write synchronously, otherwise use
* delayed write.
@ -328,21 +384,23 @@ ffs_balloc(v)
} else {
bdwrite(bp);
}
*ap->a_bpp = nbp;
return (0);
}
brelse(bp);
if (flags & B_CLRBUF) {
error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp);
if (error) {
brelse(nbp);
goto fail;
if (bpp != NULL) {
if (flags & B_CLRBUF) {
error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp);
if (error) {
brelse(nbp);
goto fail;
}
} else {
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
clrbuf(nbp);
}
} else {
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
*bpp = nbp;
}
*ap->a_bpp = nbp;
return (0);
fail:
/*
@ -401,3 +459,62 @@ fail:
(void) VOP_FSYNC(vp, cred, FSYNC_WAIT, 0, 0, curproc);
return (error);
}
int
ffs_ballocn(v)
void *v;
{
struct vop_ballocn_args /* {
struct vnode *a_vp;
off_t a_offset;
off_t a_length;
struct ucred *a_cred;
int a_flags;
} */ *ap = v;
off_t off, len;
struct vnode *vp = ap->a_vp;
struct inode *ip = VTOI(vp);
struct fs *fs = ip->i_fs;
int error, delta, bshift, bsize;
error = 0;
bshift = fs->fs_bshift;
bsize = 1 << bshift;
off = ap->a_offset;
len = ap->a_length;
delta = off & (bsize - 1);
off -= delta;
len += delta;
while (len > 0) {
bsize = min(bsize, len);
error = VOP_BALLOC(vp, off, bsize, ap->a_cred, ap->a_flags,
NULL);
if (error) {
goto out;
}
/*
* increase file size now, VOP_BALLOC() requires that
* EOF be up-to-date before each call.
*/
if (ip->i_ffs_size < off + bsize) {
ip->i_ffs_size = off + bsize;
if (vp->v_uvm.u_size < ip->i_ffs_size) {
uvm_vnp_setsize(vp, ip->i_ffs_size);
}
}
off += bsize;
len -= bsize;
}
out:
return error;
}

6
sys/ufs/ffs/ffs_extern.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: ffs_extern.h,v 1.16 2000/04/04 09:23:20 jdolecek Exp $ */
/* $NetBSD: ffs_extern.h,v 1.17 2000/11/27 08:39:54 chs Exp $ */
/*-
* Copyright (c) 1991, 1993, 1994
@ -79,7 +79,7 @@ __BEGIN_DECLS
int ffs_alloc __P((struct inode *, ufs_daddr_t, ufs_daddr_t , int, struct ucred *,
ufs_daddr_t *));
int ffs_realloccg __P((struct inode *, ufs_daddr_t, ufs_daddr_t, int, int ,
struct ucred *, struct buf **));
struct ucred *, struct buf **, ufs_daddr_t *));
int ffs_reallocblks __P((void *));
int ffs_valloc __P((void *));
ufs_daddr_t ffs_blkpref __P((struct inode *, ufs_daddr_t, int, ufs_daddr_t *));
@ -89,6 +89,7 @@ void ffs_clusteracct __P((struct fs *, struct cg *, ufs_daddr_t, int));
/* ffs_balloc.c */
int ffs_balloc __P((void *));
int ffs_ballocn __P((void *));
/* ffs_bswap.c */
void ffs_sb_swap __P((struct fs*, struct fs *, int));
@ -137,6 +138,7 @@ int ffs_read __P((void *));
int ffs_write __P((void *));
int ffs_fsync __P((void *));
int ffs_reclaim __P((void *));
int ffs_size __P((void *));
__END_DECLS

130
sys/ufs/ffs/ffs_inode.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ffs_inode.c,v 1.37 2000/09/19 22:04:09 fvdl Exp $ */
/* $NetBSD: ffs_inode.c,v 1.38 2000/11/27 08:39:54 chs Exp $ */
/*
* Copyright (c) 1982, 1986, 1989, 1993
@ -170,37 +170,25 @@ ffs_truncate(v)
struct vnode *ovp = ap->a_vp;
ufs_daddr_t lastblock;
struct inode *oip;
ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
ufs_daddr_t bn, lastiblock[NIADDR], indir_lbn[NIADDR];
ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
off_t length = ap->a_length;
struct fs *fs;
struct buf *bp;
int offset, size, level;
long count, nblocks, blocksreleased = 0;
int i;
int aflags, error, allerror = 0;
int error, allerror = 0;
off_t osize;
if (length < 0)
return (EINVAL);
oip = VTOI(ovp);
#if 1
/*
* XXX. Was in Kirk's patches. Is it good behavior to just
* return and not update modification times?
*/
if (oip->i_ffs_size == length)
return (0);
#endif
if (ovp->v_type == VLNK &&
(oip->i_ffs_size < ovp->v_mount->mnt_maxsymlinklen ||
(ovp->v_mount->mnt_maxsymlinklen == 0 &&
oip->i_din.ffs_din.di_blocks == 0))) {
#ifdef DIAGNOSTIC
if (length != 0)
panic("ffs_truncate: partial truncate of symlink");
#endif
memset((char *)&oip->i_ffs_shortlink, 0, (u_int)oip->i_ffs_size);
KDASSERT(length == 0);
memset(&oip->i_ffs_shortlink, 0, (size_t)oip->i_ffs_size);
oip->i_ffs_size = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (VOP_UPDATE(ovp, NULL, NULL, UPDATE_WAIT));
@ -214,12 +202,56 @@ ffs_truncate(v)
return (error);
#endif
fs = oip->i_fs;
if (length > fs->fs_maxfilesize)
return (EFBIG);
osize = oip->i_ffs_size;
ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0;
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
*/
if (osize < length) {
ufs_balloc_range(ovp, length - 1, 1, ap->a_cred,
ap->a_flags & IO_SYNC ? B_SYNC : 0);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (VOP_UPDATE(ovp, NULL, NULL, 1));
}
/*
* When truncating a regular file down to a non-block-aligned size,
* we must zero the part of last block which is past the new EOF.
* We must synchronously flush the zeroed pages to disk
* since the new pages will be invalidated as soon as we
* inform the VM system of the new, smaller size.
* We must to this before acquiring the GLOCK, since fetching
* the pages will acquire the GLOCK internally.
* So there is a window where another thread could see a whole
* zeroed page past EOF, but that's life.
*/
offset = blkoff(fs, length);
if (ovp->v_type == VREG && length < osize && offset != 0) {
struct uvm_object *uobj;
voff_t eoz;
size = blksize(fs, oip, lblkno(fs, length));
eoz = min(lblktosize(fs, lblkno(fs, length)) + size, osize);
uvm_vnp_zerorange(ovp, length, eoz - length);
uobj = &ovp->v_uvm.u_obj;
simple_lock(&uobj->vmobjlock);
uobj->pgops->pgo_flush(uobj, length, eoz,
PGO_CLEANIT|PGO_DEACTIVATE|PGO_SYNCIO);
simple_unlock(&ovp->v_uvm.u_obj.vmobjlock);
}
lockmgr(&ovp->v_glock, LK_EXCLUSIVE, NULL);
if (DOINGSOFTDEP(ovp)) {
uvm_vnp_setsize(ovp, length);
(void) uvm_vnp_uncache(ovp);
if (length > 0) {
/*
* If a file is only partially truncated, then
@ -231,73 +263,26 @@ ffs_truncate(v)
* so that it will have no data structures left.
*/
if ((error = VOP_FSYNC(ovp, ap->a_cred, FSYNC_WAIT,
0, 0, ap->a_p)) != 0)
0, 0, ap->a_p)) != 0) {
lockmgr(&ovp->v_glock, LK_RELEASE, NULL);
return (error);
}
} else {
#ifdef QUOTA
(void) chkdq(oip, -oip->i_ffs_blocks, NOCRED, 0);
#endif
softdep_setup_freeblocks(oip, length);
(void) vinvalbuf(ovp, 0, ap->a_cred, ap->a_p, 0, 0);
lockmgr(&ovp->v_glock, LK_RELEASE, NULL);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (VOP_UPDATE(ovp, NULL, NULL, 0));
}
}
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
* Reduce the size of the file.
*/
if (osize < length) {
if (length > fs->fs_maxfilesize)
return (EFBIG);
aflags = B_CLRBUF;
if (ap->a_flags & IO_SYNC)
aflags |= B_SYNC;
error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp);
if (error)
return (error);
oip->i_ffs_size = length;
uvm_vnp_setsize(ovp, length);
(void) uvm_vnp_uncache(ovp);
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (VOP_UPDATE(ovp, NULL, NULL, UPDATE_WAIT));
}
/*
* Shorten the size of the file. If the file is not being
* truncated to a block boundary, the contents of the
* partial block following the end of the file must be
* zero'ed in case it ever becomes accessible again because
* of subsequent file growth. Directories however are not
* zero'ed as they should grow back initialized to empty.
*/
offset = blkoff(fs, length);
if (offset == 0) {
oip->i_ffs_size = length;
} else {
lbn = lblkno(fs, length);
aflags = B_CLRBUF;
if (ap->a_flags & IO_SYNC)
aflags |= B_SYNC;
error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp);
if (error)
return (error);
oip->i_ffs_size = length;
size = blksize(fs, oip, lbn);
(void) uvm_vnp_uncache(ovp);
if (ovp->v_type != VDIR)
memset((char *)bp->b_data + offset, 0,
(u_int)(size - offset));
allocbuf(bp, size);
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
}
oip->i_ffs_size = length;
uvm_vnp_setsize(ovp, length);
/*
* Calculate index into inode's block list of
@ -431,6 +416,7 @@ done:
oip->i_ffs_blocks -= blocksreleased;
if (oip->i_ffs_blocks < 0) /* sanity */
oip->i_ffs_blocks = 0;
lockmgr(&ovp->v_glock, LK_RELEASE, NULL);
oip->i_flag |= IN_CHANGE;
#ifdef QUOTA
(void) chkdq(oip, -blocksreleased, NOCRED, 0);

347
sys/ufs/ffs/ffs_softdep.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ffs_softdep.c,v 1.7 2000/11/08 14:28:16 ad Exp $ */
/* $NetBSD: ffs_softdep.c,v 1.8 2000/11/27 08:39:54 chs Exp $ */
/*
* Copyright 1998 Marshall Kirk McKusick. All Rights Reserved.
@ -53,6 +53,10 @@
#include <ufs/ufs/ufs_extern.h>
#include <ufs/ufs/ufs_bswap.h>
#include <uvm/uvm.h>
struct pool sdpcpool;
int softdep_lockedbufs;
/*
* For now we want the safety net that the DIAGNOSTIC and DEBUG flags provide.
*/
@ -98,6 +102,13 @@ extern char *memname[];
* End system adaptaion definitions.
*/
/*
* Definitions for page cache info hashtable.
*/
#define PCBPHASHSIZE 1024
LIST_HEAD(, buf) pcbphashhead[PCBPHASHSIZE];
#define PCBPHASH(vp, lbn) ((((vaddr_t)(vp) >> 8) ^ (lbn)) & (PCBPHASHSIZE - 1))
/*
* Internal function prototypes.
*/
@ -149,6 +160,16 @@ static int pagedep_lookup __P((struct inode *, ufs_lbn_t, int,
static void pause_timer __P((void *));
static int request_cleanup __P((int, int));
static void add_to_worklist __P((struct worklist *));
static struct buf *softdep_setup_pagecache __P((struct inode *, ufs_lbn_t,
long));
static void softdep_collect_pagecache __P((struct vnode *,
struct bufq_head *));
static void softdep_free_pagecache __P((struct bufq_head *));
static struct vnode *softdep_lookupvp(struct fs *, ino_t);
static struct buf *softdep_lookup_pcbp __P((struct vnode *, ufs_lbn_t));
void softdep_pageiodone __P((struct buf *));
void softdep_flush_vnode __P((struct vnode *, ufs_lbn_t));
static void softdep_flush_indir __P((struct vnode *));
/*
* Exported softdep operations.
@ -889,6 +910,7 @@ top:
void
softdep_initialize()
{
int i;
LIST_INIT(&mkdirlisthd);
LIST_INIT(&softdep_workitem_pending);
@ -902,6 +924,11 @@ softdep_initialize()
newblk_hashtbl = hashinit(64, HASH_LIST, M_NEWBLK, M_WAITOK,
&newblk_hash);
sema_init(&newblk_in_progress, "newblk", PRIBIO, 0);
pool_init(&sdpcpool, sizeof(struct buf), 0, 0, 0, "sdpcpool",
0, pool_page_alloc_nointr, pool_page_free_nointr, M_TEMP);
for (i = 0; i < PCBPHASHSIZE; i++) {
LIST_INIT(&pcbphashhead[i]);
}
}
/*
@ -1161,6 +1188,18 @@ softdep_setup_allocdirect(ip, lbn, newblkno, oldblkno, newsize, oldsize, bp)
LIST_REMOVE(newblk, nb_hash);
FREE(newblk, M_NEWBLK);
/*
* If we were not passed a bp to attach the dep to,
* then this must be for a regular file.
* Allocate a buffer to represent the page cache pages
* that are the real dependency. The pages themselves
* cannot refer to the dependency since we don't want to
* add a field to struct vm_page for this.
*/
if (bp == NULL) {
bp = softdep_setup_pagecache(ip, lbn, newsize);
}
WORKLIST_INSERT(&bp->b_dep, &adp->ad_list);
if (lbn >= NDADDR) {
/* allocating an indirect block */
@ -1310,7 +1349,10 @@ handle_workitem_freefrag(freefrag)
vp.v_data = &tip;
vp.v_mount = freefrag->ff_devvp->v_specmountpoint;
tip.i_vnode = &vp;
lockinit(&vp.v_glock, PVFS, "fglock", 0, 0);
lockmgr(&vp.v_glock, LK_EXCLUSIVE, NULL);
ffs_blkfree(&tip, freefrag->ff_blkno, freefrag->ff_fragsize);
lockmgr(&vp.v_glock, LK_RELEASE, NULL);
FREE(freefrag, M_FREEFRAG);
}
@ -1380,6 +1422,18 @@ softdep_setup_allocindir_page(ip, lbn, bp, ptrno, newblkno, oldblkno, nbp)
struct allocindir *aip;
struct pagedep *pagedep;
/*
* If we are already holding "many" buffers busy (as the safe copies
* of indirect blocks) flush the dependency for one of those before
* potentially tying up more. otherwise we could fill the
* buffer cache with busy buffers and deadlock.
* XXXUBC I'm sure there's a better way to deal with this.
*/
while (softdep_lockedbufs > nbuf >> 2) {
softdep_flush_indir(ITOV(ip));
}
aip = newallocindir(ip, ptrno, newblkno, oldblkno);
ACQUIRE_LOCK(&lk);
/*
@ -1390,6 +1444,9 @@ softdep_setup_allocindir_page(ip, lbn, bp, ptrno, newblkno, oldblkno, nbp)
if ((ip->i_ffs_mode & IFMT) == IFDIR &&
pagedep_lookup(ip, lbn, DEPALLOC, &pagedep) == 0)
WORKLIST_INSERT(&nbp->b_dep, &pagedep->pd_list);
if (nbp == NULL) {
nbp = softdep_setup_pagecache(ip, lbn, ip->i_fs->fs_bsize);
}
WORKLIST_INSERT(&nbp->b_dep, &aip->ai_list);
FREE_LOCK(&lk);
setup_allocindir_phase2(bp, ip, aip);
@ -1495,8 +1552,10 @@ setup_allocindir_phase2(bp, ip, aip)
FREE_LOCK(&lk);
}
if (newindirdep) {
if (indirdep->ir_savebp != NULL)
if (indirdep->ir_savebp != NULL) {
brelse(newindirdep->ir_savebp);
softdep_lockedbufs--;
}
WORKITEM_FREE((caddr_t)newindirdep, D_INDIRDEP);
}
if (indirdep)
@ -1513,6 +1572,7 @@ setup_allocindir_phase2(bp, ip, aip)
}
newindirdep->ir_savebp =
getblk(ip->i_devvp, bp->b_blkno, bp->b_bcount, 0, 0);
softdep_lockedbufs++;
newindirdep->ir_savebp->b_flags |= B_ASYNC;
bcopy(bp->b_data, newindirdep->ir_savebp->b_data, bp->b_bcount);
}
@ -1555,8 +1615,9 @@ softdep_setup_freeblocks(ip, length)
struct freeblks *freeblks;
struct inodedep *inodedep;
struct allocdirect *adp;
struct vnode *vp;
struct vnode *vp = ITOV(ip);
struct buf *bp;
struct bufq_head fbqh;
struct fs *fs = ip->i_fs;
int i, error;
#ifdef FFS_EI
@ -1616,7 +1677,13 @@ softdep_setup_freeblocks(ip, length)
* with this inode are obsolete and can simply be de-allocated.
* We must first merge the two dependency lists to get rid of
* any duplicate freefrag structures, then purge the merged list.
* We must remove any pagecache markers from the pagecache
* hashtable first because any I/Os in flight will want to see
* dependencies attached to their pagecache markers. We cannot
* free the pagecache markers until after we've freed all the
* dependencies that reference them later.
*/
softdep_collect_pagecache(vp, &fbqh);
merge_inode_lists(inodedep);
while ((adp = TAILQ_FIRST(&inodedep->id_inoupdt)) != 0)
free_allocdirect(&inodedep->id_inoupdt, adp, 1);
@ -1628,7 +1695,6 @@ softdep_setup_freeblocks(ip, length)
* Once they are all there, walk the list and get rid of
* any dependencies.
*/
vp = ITOV(ip);
ACQUIRE_LOCK(&lk);
drain_output(vp, 1);
while (getdirtybuf(&vp->v_dirtyblkhd.lh_first, MNT_WAIT)) {
@ -1640,6 +1706,7 @@ softdep_setup_freeblocks(ip, length)
brelse(bp);
ACQUIRE_LOCK(&lk);
}
softdep_free_pagecache(&fbqh);
/*
* Add the freeblks structure to the list of operations that
* must await the zero'ed inode being written to disk. If we
@ -1730,8 +1797,8 @@ deallocate_dependencies(bp, inodedep)
* If the inode has already been written, then they
* can be dumped directly onto the work list.
*/
for (dirrem = LIST_FIRST(&pagedep->pd_dirremhd); dirrem;
dirrem = LIST_NEXT(dirrem, dm_next)) {
while ((dirrem = LIST_FIRST(&pagedep->pd_dirremhd))
!= NULL) {
LIST_REMOVE(dirrem, dm_next);
dirrem->dm_dirinum = pagedep->pd_ino;
if (inodedep == NULL ||
@ -1944,6 +2011,10 @@ handle_workitem_freeblocks(freeblks)
}
nblocks = btodb(fs->fs_bsize);
blocksreleased = 0;
lockinit(&vp.v_glock, PVFS, "fglock", 0, 0);
lockmgr(&vp.v_glock, LK_EXCLUSIVE, NULL);
/*
* Indirect blocks first.
*/
@ -1966,6 +2037,7 @@ handle_workitem_freeblocks(freeblks)
ffs_blkfree(&tip, bn, bsize);
blocksreleased += btodb(bsize);
}
lockmgr(&vp.v_glock, LK_RELEASE, NULL);
#ifdef DIAGNOSTIC
if (freeblks->fb_chkcnt != blocksreleased)
@ -2034,6 +2106,7 @@ indir_trunc(ip, dbn, level, lbn, countp)
error = bread(ip->i_devvp, dbn, (int)fs->fs_bsize, NOCRED, &bp);
if (error)
return (error);
softdep_lockedbufs++;
}
/*
* Recursively free indirect blocks.
@ -2053,6 +2126,7 @@ indir_trunc(ip, dbn, level, lbn, countp)
}
bp->b_flags |= B_INVAL | B_NOCACHE;
brelse(bp);
softdep_lockedbufs--;
return (allerror);
}
@ -2793,6 +2867,8 @@ softdep_disk_io_initiation(bp)
if (LIST_FIRST(&indirdep->ir_deplisthd) == NULL) {
indirdep->ir_savebp->b_flags |= B_INVAL | B_NOCACHE;
brelse(indirdep->ir_savebp);
softdep_lockedbufs--;
/* inline expand WORKLIST_REMOVE(wk); */
wk->wk_state &= ~ONWORKLIST;
LIST_REMOVE(wk, wk_list);
@ -3681,8 +3757,9 @@ merge_inode_lists(inodedep)
{
struct allocdirect *listadp, *newadp;
listadp = TAILQ_FIRST(&inodedep->id_inoupdt);
newadp = TAILQ_FIRST(&inodedep->id_newinoupdt);
for (listadp = TAILQ_FIRST(&inodedep->id_inoupdt); listadp && newadp;) {
while (listadp && newadp) {
if (listadp->ad_lbn < newadp->ad_lbn) {
listadp = TAILQ_NEXT(listadp, ad_next);
continue;
@ -3935,6 +4012,7 @@ loop:
switch (wk->wk_type) {
case D_ALLOCDIRECT:
KASSERT(vp->v_type != VREG);
adp = WK_ALLOCDIRECT(wk);
if (adp->ad_state & DEPCOMPLETE)
break;
@ -4141,6 +4219,7 @@ flush_inodedep_deps(fs, ino)
struct allocdirect *adp;
int error, waitfor;
struct buf *bp;
struct vnode *vp;
/*
* This work is done in two passes. The first pass grabs most
@ -4160,6 +4239,27 @@ flush_inodedep_deps(fs, ino)
ACQUIRE_LOCK(&lk);
if (inodedep_lookup(fs, ino, 0, &inodedep) == 0)
return (0);
/*
* When file data was in the buffer cache,
* softdep_sync_metadata() would start i/o on
* file data buffers itself. But now that
* we're using the page cache to hold file data,
* we need something else to trigger those flushes.
* let's just do it here.
*/
vp = softdep_lookupvp(fs, ino);
if (vp) {
struct uvm_object *uobj = &vp->v_uvm.u_obj;
simple_lock(&uobj->vmobjlock);
(uobj->pgops->pgo_flush)(uobj, 0, 0,
PGO_ALLPAGES|PGO_CLEANIT|
(waitfor == MNT_NOWAIT ? 0: PGO_SYNCIO));
simple_unlock(&uobj->vmobjlock);
}
for (adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp;
adp = TAILQ_NEXT(adp, ad_next)) {
if (adp->ad_state & DEPCOMPLETE)
@ -4727,3 +4827,236 @@ softdep_error(func, error)
/* XXX should do something better! */
printf("%s: got error %d while accessing filesystem\n", func, error);
}
/*
* Allocate a buffer on which to attach a dependency.
*/
static struct buf *
softdep_setup_pagecache(ip, lbn, size)
struct inode *ip;
ufs_lbn_t lbn;
long size;
{
struct vnode *vp = ITOV(ip);
struct buf *bp;
int s;
/*
* Enter pagecache dependency buf in hash.
*/
bp = softdep_lookup_pcbp(vp, lbn);
if (bp == NULL) {
s = splbio();
bp = pool_get(&sdpcpool, PR_WAITOK);
splx(s);
memset(bp, 0, sizeof(*bp));
bp->b_vp = vp;
bp->b_lblkno = lbn;
bp->b_bcount = bp->b_resid = size;
LIST_INIT(&bp->b_dep);
LIST_INSERT_HEAD(&pcbphashhead[PCBPHASH(vp, lbn)], bp, b_hash);
} else {
KASSERT(size >= bp->b_bcount);
bp->b_resid += size - bp->b_bcount;
bp->b_bcount = size;
}
return bp;
}
/*
* softdep_collect_pagecache() and softdep_free_pagecache()
* are used to remove page cache dependency buffers when
* a file is being truncated to 0.
*/
static void
softdep_collect_pagecache(vp, bqhp)
struct vnode *vp;
struct bufq_head *bqhp;
{
struct buf *bp, *nextbp;
int i;
TAILQ_INIT(bqhp);
for (i = 0; i < PCBPHASHSIZE; i++) {
for (bp = LIST_FIRST(&pcbphashhead[i]);
bp != NULL;
bp = nextbp) {
nextbp = LIST_NEXT(bp, b_hash);
if (bp->b_vp == vp) {
LIST_REMOVE(bp, b_hash);
TAILQ_INSERT_HEAD(bqhp, bp, b_freelist);
}
}
}
}
static void
softdep_free_pagecache(bqhp)
struct bufq_head *bqhp;
{
struct buf *bp, *nextbp;
for (bp = TAILQ_FIRST(bqhp); bp != NULL; bp = nextbp) {
nextbp = TAILQ_NEXT(bp, b_freelist);
TAILQ_REMOVE(bqhp, bp, b_freelist);
KASSERT(LIST_FIRST(&bp->b_dep) == NULL);
pool_put(&sdpcpool, bp);
}
}
static struct vnode *
softdep_lookupvp(fs, ino)
struct fs *fs;
ino_t ino;
{
struct mount *mp;
extern struct vfsops ffs_vfsops;
CIRCLEQ_FOREACH(mp, &mountlist, mnt_list) {
if (mp->mnt_op == &ffs_vfsops &&
VFSTOUFS(mp)->um_fs == fs) {
break;
}
}
if (mp == NULL) {
return NULL;
}
return ufs_ihashlookup(VFSTOUFS(mp)->um_dev, ino);
}
/*
* Flush some dependent page cache data for any vnode *except*
* the one specified.
* XXXUBC this is a horrible hack and it's probably not too hard to deadlock
* even with this, but it's better than nothing.
*/
static void
softdep_flush_indir(vp)
struct vnode *vp;
{
struct buf *bp;
int i;
for (i = 0; i < PCBPHASHSIZE; i++) {
LIST_FOREACH(bp, &pcbphashhead[i], b_hash) {
if (bp->b_vp == vp ||
LIST_FIRST(&bp->b_dep)->wk_type != D_ALLOCINDIR) {
continue;
}
VOP_FSYNC(bp->b_vp, curproc->p_ucred, FSYNC_WAIT, 0, 0,
curproc);
return;
}
}
printf("softdep_flush_indir: nothing to flush?\n");
}
static struct buf *
softdep_lookup_pcbp(vp, lbn)
struct vnode *vp;
ufs_lbn_t lbn;
{
struct buf *bp;
LIST_FOREACH(bp, &pcbphashhead[PCBPHASH(vp, lbn)], b_hash) {
if (bp->b_vp == vp && bp->b_lblkno == lbn) {
break;
}
}
return bp;
}
/*
* Do softdep i/o completion processing for page cache writes.
*/
void
softdep_pageiodone(bp)
struct buf *bp;
{
int npages = bp->b_bufsize >> PAGE_SHIFT;
struct vnode *vp = bp->b_vp;
struct vm_page *pg;
struct buf *pcbp = NULL;
struct allocdirect *adp;
struct allocindir *aip;
struct worklist *wk;
ufs_lbn_t lbn;
voff_t off;
long iosize = bp->b_bcount;
int size, asize, bshift, bsize;
int i;
KASSERT(!(bp->b_flags & B_READ));
bshift = vp->v_mount->mnt_fs_bshift;
bsize = 1 << bshift;
asize = min(PAGE_SIZE, bsize);
ACQUIRE_LOCK(&lk);
for (i = 0; i < npages; i++) {
pg = uvm_pageratop((vaddr_t)bp->b_data + (i << PAGE_SHIFT));
if (pg == NULL) {
continue;
}
for (off = pg->offset;
off < pg->offset + PAGE_SIZE;
off += bsize) {
size = min(asize, iosize);
iosize -= size;
lbn = off >> bshift;
if (pcbp == NULL || pcbp->b_lblkno != lbn) {
pcbp = softdep_lookup_pcbp(vp, lbn);
}
if (pcbp == NULL) {
continue;
}
pcbp->b_resid -= size;
if (pcbp->b_resid < 0) {
panic("softdep_pageiodone: "
"resid < 0, vp %p lbn 0x%lx pcbp %p",
vp, lbn, pcbp);
}
if (pcbp->b_resid > 0) {
continue;
}
/*
* We've completed all the i/o for this block.
* mark the dep complete.
*/
KASSERT(LIST_FIRST(&pcbp->b_dep) != NULL);
while ((wk = LIST_FIRST(&pcbp->b_dep))) {
WORKLIST_REMOVE(wk);
switch (wk->wk_type) {
case D_ALLOCDIRECT:
adp = WK_ALLOCDIRECT(wk);
adp->ad_state |= COMPLETE;
handle_allocdirect_partdone(adp);
break;
case D_ALLOCINDIR:
aip = WK_ALLOCINDIR(wk);
aip->ai_state |= COMPLETE;
handle_allocindir_partdone(aip);
break;
default:
panic("softdep_pageiodone: "
"bad type %d, pcbp %p wk %p",
wk->wk_type, pcbp, wk);
}
}
LIST_REMOVE(pcbp, b_hash);
pool_put(&sdpcpool, pcbp);
pcbp = NULL;
}
}
FREE_LOCK(&lk);
}

9
sys/ufs/ffs/ffs_vfsops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ffs_vfsops.c,v 1.72 2000/10/13 16:40:26 simonb Exp $ */
/* $NetBSD: ffs_vfsops.c,v 1.73 2000/11/27 08:39:55 chs Exp $ */
/*
* Copyright (c) 1989, 1991, 1993, 1994
@ -690,6 +690,8 @@ ffs_mountfs(devvp, mp, p)
mp->mnt_stat.f_fsid.val[0] = (long)dev;
mp->mnt_stat.f_fsid.val[1] = makefstype(MOUNT_FFS);
mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
mp->mnt_fs_bshift = fs->fs_bshift;
mp->mnt_dev_bshift = DEV_BSHIFT; /* XXX */
mp->mnt_flag |= MNT_LOCAL;
#ifdef FFS_EI
if (needswap)
@ -699,6 +701,7 @@ ffs_mountfs(devvp, mp, p)
ump->um_dev = dev;
ump->um_devvp = devvp;
ump->um_nindir = fs->fs_nindir;
ump->um_lognindir = ffs(fs->fs_nindir) - 1;
ump->um_bptrtodb = fs->fs_fsbtodb;
ump->um_seqinc = fs->fs_frag;
for (i = 0; i < MAXQUOTAS; i++)
@ -797,6 +800,9 @@ ffs_unmount(mp, mntflags, p)
if (ump->um_devvp->v_type != VBAD)
ump->um_devvp->v_specmountpoint = NULL;
vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
if (LIST_FIRST(&ump->um_devvp->v_dirtyblkhd)) {
panic("ffs_unmount: flush left dirty bufs %p", ump->um_devvp);
}
error = VOP_CLOSE(ump->um_devvp, fs->fs_ronly ? FREAD : FREAD|FWRITE,
NOCRED, p);
vput(ump->um_devvp);
@ -1107,6 +1113,7 @@ ffs_vget(mp, ino, vpp)
ip->i_ffs_uid = ip->i_din.ffs_din.di_ouid; /* XXX */
ip->i_ffs_gid = ip->i_din.ffs_din.di_ogid; /* XXX */
} /* XXX */
uvm_vnp_setsize(vp, ip->i_ffs_size);
*vpp = vp;
return (0);

104
sys/ufs/ffs/ffs_vnops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ffs_vnops.c,v 1.34 2000/10/24 14:43:32 fvdl Exp $ */
/* $NetBSD: ffs_vnops.c,v 1.35 2000/11/27 08:39:55 chs Exp $ */
/*
* Copyright (c) 1982, 1986, 1989, 1993
@ -108,12 +108,16 @@ struct vnodeopv_entry_desc ffs_vnodeop_entries[] = {
{ &vop_blkatoff_desc, ffs_blkatoff }, /* blkatoff */
{ &vop_valloc_desc, ffs_valloc }, /* valloc */
{ &vop_balloc_desc, ffs_balloc }, /* balloc */
{ &vop_ballocn_desc, ffs_ballocn }, /* balloc */
{ &vop_reallocblks_desc, ffs_reallocblks }, /* reallocblks */
{ &vop_vfree_desc, ffs_vfree }, /* vfree */
{ &vop_truncate_desc, ffs_truncate }, /* truncate */
{ &vop_update_desc, ffs_update }, /* update */
{ &vop_bwrite_desc, vn_bwrite }, /* bwrite */
{ (struct vnodeop_desc*)NULL, (int(*) __P((void*)))NULL }
{ &vop_getpages_desc, genfs_getpages }, /* getpages */
{ &vop_putpages_desc, genfs_putpages }, /* putpages */
{ &vop_size_desc, ffs_size }, /* size */
{ NULL, NULL }
};
struct vnodeopv_desc ffs_vnodeop_opv_desc =
{ &ffs_vnodeop_p, ffs_vnodeop_entries };
@ -165,7 +169,7 @@ struct vnodeopv_entry_desc ffs_specop_entries[] = {
{ &vop_truncate_desc, spec_truncate }, /* truncate */
{ &vop_update_desc, ffs_update }, /* update */
{ &vop_bwrite_desc, vn_bwrite }, /* bwrite */
{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
{ NULL, NULL }
};
struct vnodeopv_desc ffs_specop_opv_desc =
{ &ffs_specop_p, ffs_specop_entries };
@ -217,7 +221,7 @@ struct vnodeopv_entry_desc ffs_fifoop_entries[] = {
{ &vop_truncate_desc, fifo_truncate }, /* truncate */
{ &vop_update_desc, ffs_update }, /* update */
{ &vop_bwrite_desc, vn_bwrite }, /* bwrite */
{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
{ NULL, NULL }
};
struct vnodeopv_desc ffs_fifoop_opv_desc =
{ &ffs_fifoop_p, ffs_fifoop_entries };
@ -239,7 +243,7 @@ ffs_fsync(v)
off_t offhi;
struct proc *a_p;
} */ *ap = v;
struct buf *bp, *nbp, *ibp;
struct buf *bp;
int s, num, error, i;
struct indir ia[NIADDR + 1];
int bsize;
@ -260,38 +264,32 @@ ffs_fsync(v)
if (ap->a_offhi % bsize != 0)
blk_high++;
/*
* First, flush all data blocks in range.
*/
loop:
s = splbio();
for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
nbp = LIST_NEXT(bp, b_vnbufs);
if ((bp->b_flags & B_BUSY))
continue;
if (bp->b_lblkno < blk_low || bp->b_lblkno > blk_high)
continue;
bp->b_flags |= B_BUSY | B_VFLUSH;
splx(s);
bawrite(bp);
goto loop;
}
/*
* Then, flush possibly unwritten indirect blocks. Without softdeps,
* these should be the only ones left.
* First, flush all pages in range.
*/
simple_lock(&vp->v_uvm.u_obj.vmobjlock);
(vp->v_uvm.u_obj.pgops->pgo_flush)(&vp->v_uvm.u_obj,
ap->a_offlo, ap->a_offhi - ap->a_offlo, PGO_CLEANIT|PGO_SYNCIO);
simple_unlock(&vp->v_uvm.u_obj.vmobjlock);
/*
* Then, flush indirect blocks.
*/
if (!(ap->a_flags & FSYNC_DATAONLY) && blk_high >= NDADDR) {
error = ufs_getlbns(vp, blk_high, ia, &num);
if (error != 0)
if (error)
return error;
for (i = 0; i < num; i++) {
ibp = incore(vp, ia[i].in_lbn);
if (ibp != NULL && !(ibp->b_flags & B_BUSY) &&
(ibp->b_flags & B_DELWRI)) {
ibp->b_flags |= B_BUSY | B_VFLUSH;
bp = incore(vp, ia[i].in_lbn);
if (bp != NULL && !(bp->b_flags & B_BUSY) &&
(bp->b_flags & B_DELWRI)) {
bp->b_flags |= B_BUSY | B_VFLUSH;
splx(s);
bawrite(ibp);
bawrite(bp);
s = splbio();
}
}
@ -300,11 +298,9 @@ loop:
if (ap->a_flags & FSYNC_WAIT) {
while (vp->v_numoutput > 0) {
vp->v_flag |= VBWAIT;
tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1,
"fsync_range", 0);
tsleep(&vp->v_numoutput, PRIBIO + 1, "fsync_range", 0);
}
}
splx(s);
return (VOP_UPDATE(vp, NULL, NULL,
@ -330,23 +326,33 @@ ffs_full_fsync(v)
struct vnode *vp = ap->a_vp;
struct buf *bp, *nbp;
int s, error, passes, skipmeta;
struct uvm_object *uobj;
if (vp->v_type == VBLK &&
vp->v_specmountpoint != NULL &&
(vp->v_specmountpoint->mnt_flag & MNT_SOFTDEP))
softdep_fsync_mountdev(vp);
/*
* Flush all dirty buffers associated with a vnode
/*
* Flush all dirty data associated with a vnode.
*/
if (vp->v_type == VREG) {
uobj = &vp->v_uvm.u_obj;
simple_lock(&uobj->vmobjlock);
(uobj->pgops->pgo_flush)(uobj, 0, 0, PGO_ALLPAGES|PGO_CLEANIT|
((ap->a_flags & FSYNC_WAIT) ? PGO_SYNCIO : 0));
simple_unlock(&uobj->vmobjlock);
}
passes = NIADDR + 1;
skipmeta = 0;
if (ap->a_flags & (FSYNC_DATAONLY|FSYNC_WAIT))
skipmeta = 1;
s = splbio();
loop:
for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp;
bp = LIST_NEXT(bp, b_vnbufs))
LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs)
bp->b_flags &= ~B_SCANNED;
for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
nbp = LIST_NEXT(bp, b_vnbufs);
@ -445,3 +451,31 @@ ffs_reclaim(v)
vp->v_data = NULL;
return (0);
}
/*
* Return the last logical file offset that should be written for this file
* if we're doing a write that ends at "size".
*/
int
ffs_size(v)
void *v;
{
struct vop_size_args /* {
struct vnode *a_vp;
off_t a_size;
off_t *a_eobp;
} */ *ap = v;
struct inode *ip = VTOI(ap->a_vp);
struct fs *fs = ip->i_fs;
ufs_lbn_t olbn, nlbn;
olbn = lblkno(fs, ip->i_ffs_size);
nlbn = lblkno(fs, ap->a_size);
if (nlbn < NDADDR && olbn <= nlbn) {
*ap->a_eobp = fragroundup(fs, ap->a_size);
} else {
*ap->a_eobp = blkroundup(fs, ap->a_size);
}
return 0;
}

3
sys/ufs/lfs/lfs_alloc.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: lfs_alloc.c,v 1.44 2000/11/27 03:33:57 perseant Exp $ */
/* $NetBSD: lfs_alloc.c,v 1.45 2000/11/27 08:39:55 chs Exp $ */
/*-
* Copyright (c) 1999, 2000 The NetBSD Foundation, Inc.
@ -229,7 +229,6 @@ extend_ifile(struct lfs *fs, struct ucred *cred)
}
ip->i_ffs_size += fs->lfs_bsize;
uvm_vnp_setsize(vp, ip->i_ffs_size);
(void)uvm_vnp_uncache(vp);
VOP_UNLOCK(vp, 0);
i = (blkno - fs->lfs_segtabsz - fs->lfs_cleansz) *

4
sys/ufs/lfs/lfs_inode.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: lfs_inode.c,v 1.48 2000/11/27 03:33:57 perseant Exp $ */
/* $NetBSD: lfs_inode.c,v 1.49 2000/11/27 08:39:56 chs Exp $ */
/*-
* Copyright (c) 1999, 2000 The NetBSD Foundation, Inc.
@ -288,7 +288,6 @@ lfs_truncate(v)
return (error);
oip->i_ffs_size = length;
uvm_vnp_setsize(ovp, length);
(void) uvm_vnp_uncache(ovp);
(void) VOP_BWRITE(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (VOP_UPDATE(ovp, NULL, NULL, 0));
@ -338,7 +337,6 @@ lfs_truncate(v)
odb = btodb(bp->b_bcount);
oip->i_ffs_size = length;
size = blksize(fs, oip, lbn);
(void) uvm_vnp_uncache(ovp);
if (ovp->v_type != VDIR)
memset((char *)bp->b_data + offset, 0,
(u_int)(size - offset));

8
sys/ufs/lfs/lfs_segment.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: lfs_segment.c,v 1.63 2000/11/27 03:33:57 perseant Exp $ */
/* $NetBSD: lfs_segment.c,v 1.64 2000/11/27 08:39:56 chs Exp $ */
/*-
* Copyright (c) 1999, 2000 The NetBSD Foundation, Inc.
@ -309,7 +309,7 @@ lfs_vflush(vp)
/* panic("VDIROP being flushed...this can\'t happen"); */
}
if(vp->v_usecount<0) {
printf("usecount=%ld\n",vp->v_usecount);
printf("usecount=%d\n",vp->v_usecount);
panic("lfs_vflush: usecount<0");
}
#endif
@ -1864,8 +1864,8 @@ lfs_vunref(vp)
#ifdef DIAGNOSTIC
if(vp->v_usecount<=0) {
printf("lfs_vunref: inum is %d\n", VTOI(vp)->i_number);
printf("lfs_vunref: flags are 0x%lx\n", vp->v_flag);
printf("lfs_vunref: usecount = %ld\n", vp->v_usecount);
printf("lfs_vunref: flags are 0x%x\n", vp->v_flag);
printf("lfs_vunref: usecount = %d\n", vp->v_usecount);
panic("lfs_vunref: v_usecount<0");
}
#endif

4
sys/ufs/mfs/mfs_vnops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: mfs_vnops.c,v 1.25 2000/10/09 18:07:06 thorpej Exp $ */
/* $NetBSD: mfs_vnops.c,v 1.26 2000/11/27 08:39:57 chs Exp $ */
/*
* Copyright (c) 1989, 1993
@ -263,7 +263,7 @@ mfs_close(v)
* vnode, so if we find any other uses, it is a panic.
*/
if (vp->v_usecount > 1)
printf("mfs_close: ref count %ld > 1\n", vp->v_usecount);
printf("mfs_close: ref count %d > 1\n", vp->v_usecount);
if (vp->v_usecount > 1 || BUFQ_FIRST(&mfsp->mfs_buflist) != NULL)
panic("mfs_close");
/*

20
sys/ufs/ufs/ufs_bmap.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ufs_bmap.c,v 1.9 2000/03/30 12:41:14 augustss Exp $ */
/* $NetBSD: ufs_bmap.c,v 1.10 2000/11/27 08:39:57 chs Exp $ */
/*
* Copyright (c) 1989, 1991, 1993
@ -186,6 +186,9 @@ ufs_bmaparray(vp, bn, bnp, ap, nump, runp)
xap->in_exists = 1;
bp = getblk(vp, metalbn, mp->mnt_stat.f_iosize, 0, 0);
if (bp == NULL) {
return ENOMEM;
}
if (bp->b_flags & (B_DONE | B_DELWRI)) {
trace(TR_BREADHIT, pack(vp, size), metalbn);
}
@ -243,6 +246,7 @@ ufs_getlbns(vp, bn, ap, nump)
long metalbn, realbn;
struct ufsmount *ump;
int64_t blockcnt;
int lbc;
int i, numlevels, off;
ump = VFSTOUFS(vp->v_mount);
@ -263,10 +267,15 @@ ufs_getlbns(vp, bn, ap, nump)
* at the given level of indirection, and NIADDR - i is the number
* of levels of indirection needed to locate the requested block.
*/
for (blockcnt = 1, i = NIADDR, bn -= NDADDR;; i--, bn -= blockcnt) {
bn -= NDADDR;
for (lbc = 0, i = NIADDR;; i--, bn -= blockcnt) {
if (i == 0)
return (EFBIG);
blockcnt *= MNINDIR(ump);
lbc += ump->um_lognindir;
blockcnt = (int64_t)1 << lbc;
if (bn < blockcnt)
break;
}
@ -292,8 +301,9 @@ ufs_getlbns(vp, bn, ap, nump)
if (metalbn == realbn)
break;
blockcnt /= MNINDIR(ump);
off = (bn / blockcnt) % MNINDIR(ump);
lbc -= ump->um_lognindir;
blockcnt = (int64_t)1 << lbc;
off = (bn >> lbc) & (MNINDIR(ump) - 1);
++numlevels;
ap->in_lbn = metalbn;

3
sys/ufs/ufs/ufs_extern.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: ufs_extern.h,v 1.23 2000/03/16 18:26:49 jdolecek Exp $ */
/* $NetBSD: ufs_extern.h,v 1.24 2000/11/27 08:39:57 chs Exp $ */
/*-
* Copyright (c) 1991, 1993, 1994
@ -113,6 +113,7 @@ void ufs_ihashrem __P((struct inode *));
/* ufs_inode.c */
int ufs_reclaim __P((struct vnode *, struct proc *));
int ufs_balloc_range __P((struct vnode *, off_t, off_t, struct ucred *, int));
/* ufs_lookup.c */
void ufs_dirbad __P((struct inode *, doff_t, char *));

146
sys/ufs/ufs/ufs_inode.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ufs_inode.c,v 1.15 2000/05/29 18:04:31 mycroft Exp $ */
/* $NetBSD: ufs_inode.c,v 1.16 2000/11/27 08:39:57 chs Exp $ */
/*
* Copyright (c) 1991, 1993
@ -55,6 +55,8 @@
#include <ufs/ufs/ufsmount.h>
#include <ufs/ufs/ufs_extern.h>
#include <uvm/uvm.h>
/*
* Last reference to an inode. If necessary, write or delete it.
*/
@ -73,7 +75,7 @@ ufs_inactive(v)
extern int prtactive;
if (prtactive && vp->v_usecount != 0)
vprint("ffs_inactive: pushing active", vp);
vprint("ufs_inactive: pushing active", vp);
/*
* Ignore inodes related to stale file handles.
@ -102,8 +104,9 @@ out:
* If we are done with the inode, reclaim it
* so that it can be reused immediately.
*/
if (ip->i_ffs_mode == 0)
vrecycle(vp, (struct simplelock *)0, p);
vrecycle(vp, NULL, p);
return (error);
}
@ -146,3 +149,140 @@ ufs_reclaim(vp, p)
#endif
return (0);
}
/*
* allocate a range of blocks in a file.
* after this function returns, any page entirely contained within the range
* will map to invalid data and thus must be overwritten before it is made
* accessible to others.
*/
int
ufs_balloc_range(vp, off, len, cred, flags)
struct vnode *vp;
off_t off, len;
struct ucred *cred;
int flags;
{
off_t oldeof, neweof, oldeob, neweob, oldpagestart, pagestart;
struct uvm_object *uobj;
int i, delta, error, npages1, npages2;
int bshift = vp->v_mount->mnt_fs_bshift;
int bsize = 1 << bshift;
int ppb = max(bsize >> PAGE_SHIFT, 1);
struct vm_page *pgs1[ppb], *pgs2[ppb];
UVMHIST_FUNC("ufs_balloc_range"); UVMHIST_CALLED(ubchist);
UVMHIST_LOG(ubchist, "vp %p off 0x%x len 0x%x u_size 0x%x",
vp, off, len, vp->v_uvm.u_size);
oldeof = vp->v_uvm.u_size;
error = VOP_SIZE(vp, oldeof, &oldeob);
if (error) {
return error;
}
neweof = max(vp->v_uvm.u_size, off + len);
error = VOP_SIZE(vp, neweof, &neweob);
if (error) {
return error;
}
error = 0;
uobj = &vp->v_uvm.u_obj;
pgs1[0] = pgs2[0] = NULL;
/*
* if the last block in the file is not a full block (ie. it is a
* fragment), and this allocation is causing the fragment to change
* size (either to expand the fragment or promote it to a full block),
* cache the old last block (at its new size).
*/
oldpagestart = trunc_page(oldeof) & ~(bsize - 1);
if ((oldeob & (bsize - 1)) != 0 && oldeob != neweob) {
npages1 = min(ppb, (round_page(neweob) - oldpagestart) >>
PAGE_SHIFT);
memset(pgs1, 0, npages1 * sizeof(struct vm_page *));
simple_lock(&uobj->vmobjlock);
error = VOP_GETPAGES(vp, oldpagestart, pgs1, &npages1,
0, VM_PROT_READ, 0, PGO_SYNCIO|PGO_PASTEOF);
if (error) {
goto out;
}
for (i = 0; i < npages1; i++) {
UVMHIST_LOG(ubchist, "got pgs1[%d] %p", i, pgs1[i],0,0);
KASSERT((pgs1[i]->flags & PG_RELEASED) == 0);
pgs1[i]->flags &= ~PG_CLEAN;
}
}
/*
* cache the new range as well. this will create zeroed pages
* where the new block will be and keep them locked until the
* new block is allocated, so there will be no window where
* the old contents of the new block is visible to racing threads.
*/
pagestart = trunc_page(off) & ~(bsize - 1);
if (pagestart != oldpagestart || pgs1[0] == NULL) {
npages2 = min(ppb, (round_page(neweob) - pagestart) >>
PAGE_SHIFT);
memset(pgs2, 0, npages2 * sizeof(struct vm_page *));
simple_lock(&uobj->vmobjlock);
error = VOP_GETPAGES(vp, pagestart, pgs2, &npages2, 0,
VM_PROT_READ, 0, PGO_SYNCIO|PGO_PASTEOF);
if (error) {
goto out;
}
for (i = 0; i < npages2; i++) {
UVMHIST_LOG(ubchist, "got pgs2[%d] %p", i, pgs2[i],0,0);
KASSERT((pgs2[i]->flags & PG_RELEASED) == 0);
pgs2[i]->flags &= ~PG_CLEAN;
}
}
/*
* adjust off to be block-aligned.
*/
delta = off & (bsize - 1);
off -= delta;
len += delta;
/*
* now allocate the range.
*/
lockmgr(&vp->v_glock, LK_EXCLUSIVE, NULL);
error = VOP_BALLOCN(vp, off, len, cred, flags);
lockmgr(&vp->v_glock, LK_RELEASE, NULL);
/*
* unbusy any pages we are holding.
* if we got an error, set the vnode size back to what it was before.
* this will free any pages we created past the old eof.
*/
out:
if (error) {
uvm_vnp_setsize(vp, oldeof);
}
simple_lock(&uobj->vmobjlock);
if (pgs1[0] != NULL) {
uvm_page_unbusy(pgs1, npages1);
/*
* The data in the frag might be moving to a new disk location.
* We need to flush pages to the new disk locations.
*/
(uobj->pgops->pgo_flush)(uobj, oldeof & ~(bsize - 1),
min((oldeof + bsize) & ~(bsize - 1), neweof),
PGO_CLEANIT | ((flags & B_SYNC) ? PGO_SYNCIO : 0));
}
if (pgs2[0] != NULL) {
uvm_page_unbusy(pgs2, npages2);
}
simple_unlock(&uobj->vmobjlock);
return error;
}

139
sys/ufs/ufs/ufs_readwrite.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ufs_readwrite.c,v 1.27 2000/09/09 04:49:55 perseant Exp $ */
/* $NetBSD: ufs_readwrite.c,v 1.28 2000/11/27 08:39:57 chs Exp $ */
/*-
* Copyright (c) 1993
@ -73,17 +73,20 @@ READ(v)
struct inode *ip;
struct uio *uio;
FS *fs;
#ifndef LFS_READWRITE
void *win;
vsize_t bytelen;
#endif
struct buf *bp;
ufs_daddr_t lbn, nextlbn;
off_t bytesinfile;
long size, xfersize, blkoffset;
int error;
u_short mode;
vp = ap->a_vp;
ip = VTOI(vp);
mode = ip->i_ffs_mode;
uio = ap->a_uio;
error = 0;
#ifdef DIAGNOSTIC
if (uio->uio_rw != UIO_READ)
@ -102,19 +105,39 @@ READ(v)
return (EFBIG);
if (uio->uio_resid == 0)
return (0);
if (uio->uio_offset >= ip->i_ffs_size) {
goto out;
}
#ifndef LFS_READWRITE
if (vp->v_type == VREG) {
while (uio->uio_resid > 0) {
bytelen = min(ip->i_ffs_size - uio->uio_offset,
uio->uio_resid);
if (bytelen == 0)
break;
win = ubc_alloc(&vp->v_uvm.u_obj, uio->uio_offset,
&bytelen, UBC_READ);
error = uiomove(win, bytelen, uio);
ubc_release(win, 0);
if (error)
break;
}
goto out;
}
#endif
for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
if ((bytesinfile = ip->i_ffs_size - uio->uio_offset) <= 0)
bytesinfile = ip->i_ffs_size - uio->uio_offset;
if (bytesinfile <= 0)
break;
lbn = lblkno(fs, uio->uio_offset);
nextlbn = lbn + 1;
size = BLKSIZE(fs, ip, lbn);
blkoffset = blkoff(fs, uio->uio_offset);
xfersize = fs->fs_bsize - blkoffset;
if (uio->uio_resid < xfersize)
xfersize = uio->uio_resid;
if (bytesinfile < xfersize)
xfersize = bytesinfile;
xfersize = min(min(fs->fs_bsize - blkoffset, uio->uio_resid),
bytesinfile);
#ifdef LFS_READWRITE
(void)lfs_check(vp, lbn, 0);
@ -122,9 +145,6 @@ READ(v)
#else
if (lblktosize(fs, nextlbn) >= ip->i_ffs_size)
error = bread(vp, lbn, size, NOCRED, &bp);
else if (doclusterread)
error = cluster_read(vp,
ip->i_ffs_size, lbn, size, NOCRED, &bp);
else if (lbn - 1 == vp->v_lastr) {
int nextsize = BLKSIZE(fs, ip, nextlbn);
error = breadn(vp, lbn,
@ -149,14 +169,15 @@ READ(v)
break;
xfersize = size;
}
error = uiomove((char *)bp->b_data + blkoffset, (int)xfersize,
uio);
error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio);
if (error)
break;
brelse(bp);
}
if (bp != NULL)
brelse(bp);
out:
if (!(vp->v_mount->mnt_flag & MNT_NOATIME)) {
ip->i_flag |= IN_ACCESS;
if ((ap->a_ioflag & IO_SYNC) == IO_SYNC)
@ -187,6 +208,12 @@ WRITE(v)
ufs_daddr_t lbn;
off_t osize;
int blkoffset, error, flags, ioflag, resid, size, xfersize;
#ifndef LFS_READWRITE
void *win;
vsize_t bytelen;
off_t oldoff;
boolean_t rv;
#endif
ioflag = ap->a_ioflag;
uio = ap->a_uio;
@ -240,14 +267,65 @@ WRITE(v)
resid = uio->uio_resid;
osize = ip->i_ffs_size;
flags = ioflag & IO_SYNC ? B_SYNC : 0;
error = 0;
for (error = 0; uio->uio_resid > 0;) {
#ifndef LFS_READWRITE
if (vp->v_type != VREG) {
goto bcache;
}
while (uio->uio_resid > 0) {
oldoff = uio->uio_offset;
blkoffset = blkoff(fs, uio->uio_offset);
bytelen = min(fs->fs_bsize - blkoffset, uio->uio_resid);
/*
* XXXUBC if file is mapped and this is the last block,
* process one page at a time.
*/
error = ufs_balloc_range(vp, uio->uio_offset, bytelen,
ap->a_cred, ioflag & IO_SYNC ? B_SYNC : 0);
if (error) {
return error;
}
win = ubc_alloc(&vp->v_uvm.u_obj, uio->uio_offset, &bytelen,
UBC_WRITE);
error = uiomove(win, bytelen, uio);
ubc_release(win, 0);
/*
* flush what we just wrote if necessary.
* XXXUBC simplistic async flushing.
*/
if (ioflag & IO_SYNC) {
simple_lock(&vp->v_uvm.u_obj.vmobjlock);
rv = vp->v_uvm.u_obj.pgops->pgo_flush(
&vp->v_uvm.u_obj, oldoff, oldoff + bytelen,
PGO_CLEANIT|PGO_SYNCIO);
simple_unlock(&vp->v_uvm.u_obj.vmobjlock);
} else if (oldoff >> 16 != uio->uio_offset >> 16) {
simple_lock(&vp->v_uvm.u_obj.vmobjlock);
rv = vp->v_uvm.u_obj.pgops->pgo_flush(
&vp->v_uvm.u_obj, (oldoff >> 16) << 16,
(uio->uio_offset >> 16) << 16, PGO_CLEANIT);
simple_unlock(&vp->v_uvm.u_obj.vmobjlock);
}
if (error) {
break;
}
}
goto out;
bcache:
#endif
flags = ioflag & IO_SYNC ? B_SYNC : 0;
while (uio->uio_resid > 0) {
lbn = lblkno(fs, uio->uio_offset);
blkoffset = blkoff(fs, uio->uio_offset);
xfersize = fs->fs_bsize - blkoffset;
if (uio->uio_resid < xfersize)
xfersize = uio->uio_resid;
xfersize = min(fs->fs_bsize - blkoffset, uio->uio_resid);
if (fs->fs_bsize > xfersize)
flags |= B_CLRBUF;
else
@ -262,14 +340,22 @@ WRITE(v)
ip->i_ffs_size = uio->uio_offset + xfersize;
uvm_vnp_setsize(vp, ip->i_ffs_size);
}
(void)uvm_vnp_uncache(vp);
size = BLKSIZE(fs, ip, lbn) - bp->b_resid;
if (size < xfersize)
if (xfersize > size)
xfersize = size;
error =
uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio);
/*
* if we didn't clear the block and the uiomove failed,
* the buf will now contain part of some other file,
* so we need to invalidate it.
*/
if (error && (flags & B_CLRBUF) == 0) {
bp->b_flags |= B_INVAL;
brelse(bp);
break;
}
#ifdef LFS_READWRITE
if (!error)
error = lfs_reserve(fs, vp, fsbtodb(fs, NIADDR + 1));
@ -289,13 +375,16 @@ WRITE(v)
#endif
if (error || xfersize == 0)
break;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* If we successfully wrote any data, and we are not the superuser
* we clear the setuid and setgid bits as a precaution against
* tampering.
*/
#ifndef LFS_READWRITE
out:
#endif
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (resid > uio->uio_resid && ap->a_cred && ap->a_cred->cr_uid != 0)
ip->i_ffs_mode &= ~(ISUID | ISGID);
if (error) {

5
sys/ufs/ufs/ufs_vnops.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: ufs_vnops.c,v 1.74 2000/10/19 10:55:35 pk Exp $ */
/* $NetBSD: ufs_vnops.c,v 1.75 2000/11/27 08:40:02 chs Exp $ */
/*
* Copyright (c) 1982, 1986, 1989, 1993, 1995
@ -461,8 +461,6 @@ ufs_chmod(vp, mode, cred, p)
ip->i_ffs_mode &= ~ALLPERMS;
ip->i_ffs_mode |= (mode & ALLPERMS);
ip->i_flag |= IN_CHANGE;
if ((vp->v_flag & VTEXT) && (ip->i_ffs_mode & S_ISTXT) == 0)
(void) uvm_vnp_uncache(vp);
return (0);
}
@ -1632,6 +1630,7 @@ ufs_strategy(v)
ip = VTOI(vp);
if (vp->v_type == VBLK || vp->v_type == VCHR)
panic("ufs_strategy: spec");
KASSERT(bp->b_bcount != 0);
if (bp->b_blkno == bp->b_lblkno) {
error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno,
NULL);

3
sys/ufs/ufs/ufsmount.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: ufsmount.h,v 1.7 1998/03/18 15:57:29 bouyer Exp $ */
/* $NetBSD: ufsmount.h,v 1.8 2000/11/27 08:40:02 chs Exp $ */
/*
* Copyright (c) 1982, 1986, 1989, 1993
@ -82,6 +82,7 @@ struct ufsmount {
struct vnode *um_quotas[MAXQUOTAS]; /* pointer to quota files */
struct ucred *um_cred[MAXQUOTAS]; /* quota file access cred */
u_long um_nindir; /* indirect ptrs per block */
u_long um_lognindir; /* log2 of um_nindir */
u_long um_bptrtodb; /* indir ptr to disk block */
u_long um_seqinc; /* inc between seq blocks */
time_t um_btime[MAXQUOTAS]; /* block quota time limit */

13
sys/uvm/uvm.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: uvm.h,v 1.23 2000/06/26 14:21:16 mrg Exp $ */
/* $NetBSD: uvm.h,v 1.24 2000/11/27 08:40:02 chs Exp $ */
/*
*
@ -76,6 +76,7 @@
struct uvm {
/* vm_page related parameters */
/* vm_page queues */
struct pgfreelist page_free[VM_NFREELIST]; /* unallocated pages */
struct pglist page_active; /* allocated pages, in use */
@ -86,10 +87,17 @@ struct uvm {
boolean_t page_init_done; /* TRUE if uvm_page_init() finished */
boolean_t page_idle_zero; /* TRUE if we should try to zero
pages in the idle loop */
/* page daemon trigger */
int pagedaemon; /* daemon sleeps on this */
struct proc *pagedaemon_proc; /* daemon's pid */
simple_lock_data_t pagedaemon_lock;
/* aiodone daemon trigger */
int aiodoned; /* daemon sleeps on this */
struct proc *aiodoned_proc; /* daemon's pid */
simple_lock_data_t aiodoned_lock;
/* page hash */
struct pglist *page_hash; /* page hash table (vp/off->page) */
int page_nhash; /* number of buckets */
@ -105,7 +113,7 @@ struct uvm {
simple_lock_data_t kentry_lock;
/* aio_done is locked by uvm.pagedaemon_lock and splbio! */
struct uvm_aiohead aio_done; /* done async i/o reqs */
TAILQ_HEAD(, buf) aio_done; /* done async i/o reqs */
/* pager VM area bounds */
vaddr_t pager_sva; /* start of pager VA area */
@ -145,6 +153,7 @@ extern struct uvm uvm;
UVMHIST_DECL(maphist);
UVMHIST_DECL(pdhist);
UVMHIST_DECL(ubchist);
/*
* UVM_UNLOCK_AND_WAIT: atomic unlock+wait... wrapper around the

549
sys/uvm/uvm_bio.c Normal file
View File

@ -0,0 +1,549 @@
/* $NetBSD: uvm_bio.c,v 1.2 2000/11/27 08:43:40 chs Exp $ */
/*
* Copyright (c) 1998 Chuck Silvers.
* All rights reserved.
*
* 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. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
* AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
*/
#include "opt_uvmhist.h"
/*
* uvm_bio.c: buffered i/o vnode mapping cache
*/
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/malloc.h>
#include <sys/kernel.h>
#include <sys/vnode.h>
#include <uvm/uvm.h>
#include <uvm/uvm_page.h>
/*
* global data structures
*/
/*
* local functions
*/
static int ubc_fault __P((struct uvm_faultinfo *, vaddr_t,
vm_page_t *, int, int, vm_fault_t, vm_prot_t,
int));
static struct ubc_map *ubc_find_mapping __P((struct uvm_object *, voff_t));
/*
* local data structues
*/
#define UBC_HASH(uobj, offset) (((((u_long)(uobj)) >> 8) + \
(((u_long)(offset)) >> PAGE_SHIFT)) & \
ubc_object.hashmask)
#define UBC_QUEUE(offset) (&ubc_object.inactive[((offset) / UBC_WINSIZE) & \
(UBC_NQUEUES - 1)])
struct ubc_map
{
struct uvm_object * uobj; /* mapped object */
voff_t offset; /* offset into uobj */
int refcount; /* refcount on mapping */
voff_t writeoff; /* overwrite offset */
vsize_t writelen; /* overwrite len */
LIST_ENTRY(ubc_map) hash; /* hash table */
TAILQ_ENTRY(ubc_map) inactive; /* inactive queue */
};
static struct ubc_object
{
struct uvm_object uobj; /* glue for uvm_map() */
char *kva; /* where ubc_object is mapped */
struct ubc_map *umap; /* array of ubc_map's */
LIST_HEAD(, ubc_map) *hash; /* hashtable for cached ubc_map's */
u_long hashmask; /* mask for hashtable */
TAILQ_HEAD(ubc_inactive_head, ubc_map) *inactive;
/* inactive queues for ubc_map's */
} ubc_object;
struct uvm_pagerops ubc_pager =
{
NULL, /* init */
NULL, /* reference */
NULL, /* detach */
ubc_fault, /* fault */
/* ... rest are NULL */
};
int ubc_nwins = UBC_NWINS;
int ubc_winsize = UBC_WINSIZE;
#ifdef PMAP_PREFER
int ubc_nqueues;
boolean_t ubc_release_unmap = FALSE;
#define UBC_NQUEUES ubc_nqueues
#define UBC_RELEASE_UNMAP ubc_release_unmap
#else
#define UBC_NQUEUES 1
#define UBC_RELEASE_UNMAP FALSE
#endif
/*
* ubc_init
*
* init pager private data structures.
*/
void
ubc_init(void)
{
struct ubc_map *umap;
vaddr_t va;
int i;
/*
* init ubc_object.
* alloc and init ubc_map's.
* init inactive queues.
* alloc and init hashtable.
* map in ubc_object.
*/
simple_lock_init(&ubc_object.uobj.vmobjlock);
ubc_object.uobj.pgops = &ubc_pager;
TAILQ_INIT(&ubc_object.uobj.memq);
ubc_object.uobj.uo_npages = 0;
ubc_object.uobj.uo_refs = UVM_OBJ_KERN;
ubc_object.umap = malloc(ubc_nwins * sizeof(struct ubc_map),
M_TEMP, M_NOWAIT);
bzero(ubc_object.umap, ubc_nwins * sizeof(struct ubc_map));
va = (vaddr_t)1L;
#ifdef PMAP_PREFER
PMAP_PREFER(0, &va);
if (va < UBC_WINSIZE) {
va = UBC_WINSIZE;
}
ubc_nqueues = va / UBC_WINSIZE;
if (ubc_nqueues != 1) {
ubc_release_unmap = TRUE;
}
#endif
ubc_object.inactive = malloc(UBC_NQUEUES *
sizeof(struct ubc_inactive_head),
M_TEMP, M_NOWAIT);
for (i = 0; i < UBC_NQUEUES; i++) {
TAILQ_INIT(&ubc_object.inactive[i]);
}
for (i = 0; i < ubc_nwins; i++) {
umap = &ubc_object.umap[i];
TAILQ_INSERT_TAIL(&ubc_object.inactive[i & (UBC_NQUEUES - 1)],
umap, inactive);
}
ubc_object.hash = hashinit(ubc_nwins, HASH_LIST, M_TEMP, M_NOWAIT,
&ubc_object.hashmask);
for (i = 0; i <= ubc_object.hashmask; i++) {
LIST_INIT(&ubc_object.hash[i]);
}
if (uvm_map(kernel_map, (vaddr_t *)&ubc_object.kva,
ubc_nwins * UBC_WINSIZE, &ubc_object.uobj, 0, (vsize_t)va,
UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
UVM_ADV_RANDOM, UVM_FLAG_NOMERGE))
!= KERN_SUCCESS) {
panic("ubc_init: failed to map ubc_object\n");
}
UVMHIST_INIT(ubchist, 300);
}
/*
* ubc_fault: fault routine for ubc mapping
*/
static int
ubc_fault(ufi, ign1, ign2, ign3, ign4, fault_type, access_type, flags)
struct uvm_faultinfo *ufi;
vaddr_t ign1;
vm_page_t *ign2;
int ign3, ign4;
vm_fault_t fault_type;
vm_prot_t access_type;
int flags;
{
struct uvm_object *uobj;
struct vnode *vp;
struct ubc_map *umap;
vaddr_t va, eva, ubc_offset, slot_offset;
int i, rv, npages;
struct vm_page *pgs[UBC_WINSIZE >> PAGE_SHIFT], *pg;
UVMHIST_FUNC("ubc_fault"); UVMHIST_CALLED(ubchist);
/*
* no need to try with PGO_LOCKED...
* we don't need to have the map locked since we know that
* no one will mess with it until our reference is released.
*/
if (flags & PGO_LOCKED) {
#if 0
return VM_PAGER_UNLOCK;
#else
uvmfault_unlockall(ufi, NULL, &ubc_object.uobj, NULL);
flags &= ~PGO_LOCKED;
#endif
}
va = ufi->orig_rvaddr;
ubc_offset = va - (vaddr_t)ubc_object.kva;
UVMHIST_LOG(ubchist, "va 0x%lx ubc_offset 0x%lx at %d",
va, ubc_offset, access_type,0);
umap = &ubc_object.umap[ubc_offset / UBC_WINSIZE];
KASSERT(umap->refcount != 0);
slot_offset = trunc_page(ubc_offset & (UBC_WINSIZE - 1));
/* no umap locking needed since we have a ref on the umap */
uobj = umap->uobj;
vp = (struct vnode *)uobj;
KASSERT(uobj != NULL);
npages = (UBC_WINSIZE - slot_offset) >> PAGE_SHIFT;
/*
* XXXUBC
* if npages is more than 1 we have to be sure that
* we set PGO_OVERWRITE correctly.
*/
if (access_type == VM_PROT_WRITE) {
npages = 1;
}
again:
memset(pgs, 0, sizeof (pgs));
simple_lock(&uobj->vmobjlock);
UVMHIST_LOG(ubchist, "slot_offset 0x%x writeoff 0x%x writelen 0x%x "
"u_size 0x%x", slot_offset, umap->writeoff, umap->writelen,
vp->v_uvm.u_size);
if (access_type & VM_PROT_WRITE &&
slot_offset >= umap->writeoff &&
(slot_offset + PAGE_SIZE <= umap->writeoff + umap->writelen ||
slot_offset + PAGE_SIZE >= vp->v_uvm.u_size - umap->offset)) {
UVMHIST_LOG(ubchist, "setting PGO_OVERWRITE", 0,0,0,0);
flags |= PGO_OVERWRITE;
}
else { UVMHIST_LOG(ubchist, "NOT setting PGO_OVERWRITE", 0,0,0,0); }
/* XXX be sure to zero any part of the page past EOF */
/*
* XXX
* ideally we'd like to pre-fault all of the pages we're overwriting.
* so for PGO_OVERWRITE, we should call VOP_GETPAGES() with all of the
* pages in [writeoff, writeoff+writesize] instead of just the one.
*/
UVMHIST_LOG(ubchist, "getpages vp %p offset 0x%x npages %d",
uobj, umap->offset + slot_offset, npages, 0);
rv = VOP_GETPAGES(vp, umap->offset + slot_offset, pgs, &npages, 0,
access_type, 0, flags);
UVMHIST_LOG(ubchist, "getpages rv %d npages %d", rv, npages,0,0);
switch (rv) {
case VM_PAGER_OK:
break;
case VM_PAGER_AGAIN:
tsleep(&lbolt, PVM, "ubc_fault", 0);
goto again;
default:
return rv;
}
if (npages == 0) {
return VM_PAGER_OK;
}
va = ufi->orig_rvaddr;
eva = ufi->orig_rvaddr + (npages << PAGE_SHIFT);
UVMHIST_LOG(ubchist, "va 0x%lx eva 0x%lx", va, eva, 0,0);
simple_lock(&uobj->vmobjlock);
for (i = 0; va < eva; i++, va += PAGE_SIZE) {
UVMHIST_LOG(ubchist, "pgs[%d] = %p", i, pgs[i],0,0);
pg = pgs[i];
if (pg == NULL || pg == PGO_DONTCARE) {
continue;
}
if (pg->flags & PG_WANTED) {
wakeup(pg);
}
KASSERT((pg->flags & PG_FAKE) == 0);
if (pg->flags & PG_RELEASED) {
rv = uobj->pgops->pgo_releasepg(pg, NULL);
KASSERT(rv);
continue;
}
KASSERT(access_type == VM_PROT_READ ||
(pg->flags & PG_RDONLY) == 0);
uvm_lock_pageq();
uvm_pageactivate(pg);
uvm_unlock_pageq();
pmap_enter(ufi->orig_map->pmap, va, VM_PAGE_TO_PHYS(pg),
VM_PROT_ALL, access_type);
pg->flags &= ~(PG_BUSY);
UVM_PAGE_OWN(pg, NULL);
}
simple_unlock(&uobj->vmobjlock);
return VM_PAGER_OK;
}
/*
* local functions
*/
static struct ubc_map *
ubc_find_mapping(uobj, offset)
struct uvm_object *uobj;
voff_t offset;
{
struct ubc_map *umap;
LIST_FOREACH(umap, &ubc_object.hash[UBC_HASH(uobj, offset)], hash) {
if (umap->uobj == uobj && umap->offset == offset) {
return umap;
}
}
return NULL;
}
/*
* ubc interface functions
*/
/*
* ubc_alloc: allocate a buffer mapping
*/
void *
ubc_alloc(uobj, offset, lenp, flags)
struct uvm_object *uobj;
voff_t offset;
vsize_t *lenp;
int flags;
{
int s;
vaddr_t umap_offset, slot_offset, va;
struct ubc_map *umap;
UVMHIST_FUNC("ubc_alloc"); UVMHIST_CALLED(ubchist);
UVMHIST_LOG(ubchist, "uobj %p offset 0x%lx len 0x%lx filesize 0x%x",
uobj, offset, *lenp, ((struct uvm_vnode *)uobj)->u_size);
umap_offset = (vaddr_t)(offset & ~((voff_t)UBC_WINSIZE - 1));
slot_offset = (vaddr_t)(offset & ((voff_t)UBC_WINSIZE - 1));
*lenp = min(*lenp, UBC_WINSIZE - slot_offset);
/*
* the vnode is always locked here, so we don't need to add a ref.
*/
s = splbio();
again:
simple_lock(&ubc_object.uobj.vmobjlock);
umap = ubc_find_mapping(uobj, umap_offset);
if (umap == NULL) {
umap = TAILQ_FIRST(UBC_QUEUE(offset));
if (umap == NULL) {
simple_unlock(&ubc_object.uobj.vmobjlock);
tsleep(&lbolt, PVM, "ubc_alloc", 0);
goto again;
}
/*
* remove from old hash (if any),
* add to new hash.
*/
if (umap->uobj != NULL) {
LIST_REMOVE(umap, hash);
}
umap->uobj = uobj;
umap->offset = umap_offset;
LIST_INSERT_HEAD(&ubc_object.hash[UBC_HASH(uobj, umap_offset)],
umap, hash);
va = (vaddr_t)(ubc_object.kva +
(umap - ubc_object.umap) * UBC_WINSIZE);
pmap_remove(pmap_kernel(), va, va + UBC_WINSIZE);
}
if (umap->refcount == 0) {
TAILQ_REMOVE(UBC_QUEUE(offset), umap, inactive);
}
#ifdef DIAGNOSTIC
if ((flags & UBC_WRITE) &&
(umap->writeoff || umap->writelen)) {
panic("ubc_fault: concurrent writes vp %p", uobj);
}
#endif
if (flags & UBC_WRITE) {
umap->writeoff = slot_offset;
umap->writelen = *lenp;
}
umap->refcount++;
simple_unlock(&ubc_object.uobj.vmobjlock);
splx(s);
UVMHIST_LOG(ubchist, "umap %p refs %d va %p",
umap, umap->refcount,
ubc_object.kva + (umap - ubc_object.umap) * UBC_WINSIZE,0);
return ubc_object.kva +
(umap - ubc_object.umap) * UBC_WINSIZE + slot_offset;
}
void
ubc_release(va, wlen)
void *va;
vsize_t wlen;
{
struct ubc_map *umap;
struct uvm_object *uobj;
int s;
UVMHIST_FUNC("ubc_release"); UVMHIST_CALLED(ubchist);
UVMHIST_LOG(ubchist, "va %p", va,0,0,0);
s = splbio();
simple_lock(&ubc_object.uobj.vmobjlock);
umap = &ubc_object.umap[((char *)va - ubc_object.kva) / UBC_WINSIZE];
uobj = umap->uobj;
KASSERT(uobj != NULL);
umap->writeoff = 0;
umap->writelen = 0;
umap->refcount--;
if (umap->refcount == 0) {
if (UBC_RELEASE_UNMAP &&
(((struct vnode *)uobj)->v_flag & VTEXT)) {
vaddr_t va;
/*
* if this file is the executable image of
* some process, that process will likely have
* the file mapped at an alignment other than
* what PMAP_PREFER() would like. we'd like
* to have process text be able to use the
* cache even if someone is also reading the
* file, so invalidate mappings of such files
* as soon as possible.
*/
va = (vaddr_t)(ubc_object.kva +
(umap - ubc_object.umap) * UBC_WINSIZE);
pmap_remove(pmap_kernel(), va, va + UBC_WINSIZE);
LIST_REMOVE(umap, hash);
umap->uobj = NULL;
TAILQ_INSERT_HEAD(UBC_QUEUE(umap->offset), umap,
inactive);
} else {
TAILQ_INSERT_TAIL(UBC_QUEUE(umap->offset), umap,
inactive);
}
}
UVMHIST_LOG(ubchist, "umap %p refs %d", umap, umap->refcount,0,0);
simple_unlock(&ubc_object.uobj.vmobjlock);
splx(s);
}
/*
* removing a range of mappings from the ubc mapping cache.
*/
void
ubc_flush(uobj, start, end)
struct uvm_object *uobj;
voff_t start, end;
{
struct ubc_map *umap;
vaddr_t va;
int s;
UVMHIST_FUNC("ubc_flush"); UVMHIST_CALLED(ubchist);
UVMHIST_LOG(ubchist, "uobj %p start 0x%lx end 0x%lx",
uobj, start, end,0);
s = splbio();
simple_lock(&ubc_object.uobj.vmobjlock);
for (umap = ubc_object.umap;
umap < &ubc_object.umap[ubc_nwins];
umap++) {
if (umap->uobj != uobj ||
umap->offset < start ||
(umap->offset >= end && end != 0) ||
umap->refcount > 0) {
continue;
}
/*
* remove from hash,
* move to head of inactive queue.
*/
va = (vaddr_t)(ubc_object.kva +
(umap - ubc_object.umap) * UBC_WINSIZE);
pmap_remove(pmap_kernel(), va, va + UBC_WINSIZE);
LIST_REMOVE(umap, hash);
umap->uobj = NULL;
TAILQ_REMOVE(UBC_QUEUE(umap->offset), umap, inactive);
TAILQ_INSERT_HEAD(UBC_QUEUE(umap->offset), umap, inactive);
}
simple_unlock(&ubc_object.uobj.vmobjlock);
splx(s);
}

44
sys/uvm/uvm_extern.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: uvm_extern.h,v 1.52 2000/11/27 04:36:40 nisimura Exp $ */
/* $NetBSD: uvm_extern.h,v 1.53 2000/11/27 08:40:03 chs Exp $ */
/*
*
@ -192,6 +192,21 @@ typedef struct vm_page *vm_page_t;
#define UVM_PGA_USERESERVE 0x0001 /* ok to use reserve pages */
#define UVM_PGA_ZERO 0x0002 /* returned page must be zero'd */
/*
* the following defines are for ubc_alloc's flags
*/
#define UBC_READ 0
#define UBC_WRITE 1
/*
* flags for uvn_findpages().
*/
#define UFP_ALL 0x0
#define UFP_NOWAIT 0x1
#define UFP_NOALLOC 0x2
#define UFP_NOCACHE 0x4
#define UFP_NORDONLY 0x8
/*
* lockflags that control the locking behavior of various functions.
*/
@ -213,8 +228,11 @@ struct vm_anon;
struct vmspace;
struct pmap;
struct vnode;
struct pool;
struct simplelock;
extern struct pool *uvm_aiobuf_pool;
/*
* uvmexp: global data structures that are exported to parts of the kernel
* other than the vm system.
@ -414,9 +432,16 @@ void uao_detach_locked __P((struct uvm_object *));
void uao_reference __P((struct uvm_object *));
void uao_reference_locked __P((struct uvm_object *));
/* uvm_bio.c */
void ubc_init __P((void));
void * ubc_alloc __P((struct uvm_object *, voff_t, vsize_t *,
int));
void ubc_release __P((void *, vsize_t));
void ubc_flush __P((struct uvm_object *, voff_t, voff_t));
/* uvm_fault.c */
int uvm_fault __P((vm_map_t, vaddr_t,
vm_fault_t, vm_prot_t));
int uvm_fault __P((vm_map_t, vaddr_t, vm_fault_t,
vm_prot_t));
/* handle a page fault */
/* uvm_glue.c */
@ -511,8 +536,14 @@ void uvm_page_physload __P((paddr_t, paddr_t,
paddr_t, paddr_t, int));
void uvm_setpagesize __P((void));
/* uvm_pager.c */
void uvm_aio_biodone1 __P((struct buf *));
void uvm_aio_biodone __P((struct buf *));
void uvm_aio_aiodone __P((struct buf *));
/* uvm_pdaemon.c */
void uvm_pageout __P((void *));
void uvm_aiodone_daemon __P((void *));
/* uvm_pglist.c */
int uvm_pglistalloc __P((psize_t, paddr_t,
@ -538,10 +569,11 @@ int uvm_deallocate __P((vm_map_t, vaddr_t, vsize_t));
/* uvm_vnode.c */
void uvm_vnp_setsize __P((struct vnode *, voff_t));
void uvm_vnp_sync __P((struct mount *));
void uvm_vnp_terminate __P((struct vnode *));
/* terminate a uvm/uvn object */
boolean_t uvm_vnp_uncache __P((struct vnode *));
struct uvm_object *uvn_attach __P((void *, vm_prot_t));
void uvn_findpages __P((struct uvm_object *, voff_t,
int *, struct vm_page **, int));
void uvm_vnp_zerorange __P((struct vnode *, off_t, size_t));
void uvm_vnp_asyncget __P((struct vnode *, off_t, size_t));
/* kern_malloc.c */
void kmeminit_nkmempages __P((void));

181
sys/uvm/uvm_fault.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: uvm_fault.c,v 1.51 2000/08/06 00:22:53 thorpej Exp $ */
/* $NetBSD: uvm_fault.c,v 1.52 2000/11/27 08:40:03 chs Exp $ */
/*
*
@ -458,12 +458,8 @@ uvmfault_anonget(ufi, amap, anon)
}
if (result != VM_PAGER_OK) {
#ifdef DIAGNOSTIC
if (result == VM_PAGER_PEND) {
panic("uvmfault_anonget: "
"got PENDING for non-async I/O");
}
#endif
KASSERT(result != VM_PAGER_PEND);
/* remove page from anon */
anon->u.an_page = NULL;
@ -569,7 +565,7 @@ uvm_fault(orig_map, vaddr, fault_type, access_type)
vm_prot_t enter_prot;
boolean_t wired, narrow, promote, locked, shadowed;
int npages, nback, nforw, centeridx, result, lcv, gotpages;
vaddr_t startva, objaddr, currva, offset;
vaddr_t startva, objaddr, currva, offset, uoff;
paddr_t pa;
struct vm_amap *amap;
struct uvm_object *uobj;
@ -580,7 +576,8 @@ uvm_fault(orig_map, vaddr, fault_type, access_type)
UVMHIST_LOG(maphist, "(map=0x%x, vaddr=0x%x, ft=%d, at=%d)",
orig_map, vaddr, fault_type, access_type);
anon = NULL; /* XXX: shut up gcc */
anon = NULL;
pg = NULL;
uvmexp.faults++; /* XXX: locking? */
@ -717,10 +714,8 @@ ReFault:
if (narrow == FALSE) {
/* wide fault (!narrow) */
#ifdef DIAGNOSTIC
if (uvmadvice[ufi.entry->advice].advice != ufi.entry->advice)
panic("fault: advice mismatch!");
#endif
KASSERT(uvmadvice[ufi.entry->advice].advice ==
ufi.entry->advice);
nback = min(uvmadvice[ufi.entry->advice].nback,
(ufi.orig_rvaddr - ufi.entry->start) >> PAGE_SHIFT);
startva = ufi.orig_rvaddr - (nback << PAGE_SHIFT);
@ -793,7 +788,7 @@ ReFault:
/* now forget about the backpages */
if (amap)
anons += nback;
startva = startva + (nback << PAGE_SHIFT);
startva += (nback << PAGE_SHIFT);
npages -= nback;
nback = centeridx = 0;
}
@ -814,12 +809,10 @@ ReFault:
* dont play with VAs that are already mapped
* except for center)
*/
if (lcv != centeridx) {
if (pmap_extract(ufi.orig_map->pmap, currva, &pa) ==
TRUE) {
pages[lcv] = PGO_DONTCARE;
continue;
}
if (lcv != centeridx &&
pmap_extract(ufi.orig_map->pmap, currva, &pa)) {
pages[lcv] = PGO_DONTCARE;
continue;
}
/*
@ -851,11 +844,13 @@ ReFault:
" MAPPING: n anon: pm=0x%x, va=0x%x, pg=0x%x",
ufi.orig_map->pmap, currva, anon->u.an_page, 0);
uvmexp.fltnamap++;
/*
* Since this isn't the page that's actually faulting,
* ignore pmap_enter() failures; it's not critical
* that we enter these right now.
*/
(void) pmap_enter(ufi.orig_map->pmap, currva,
VM_PAGE_TO_PHYS(anon->u.an_page),
(anon->an_ref > 1) ? (enter_prot & ~VM_PROT_WRITE) :
@ -888,13 +883,13 @@ ReFault:
*/
if (uobj && shadowed == FALSE && uobj->pgops->pgo_fault != NULL) {
simple_lock(&uobj->vmobjlock);
/* locked: maps(read), amap (if there), uobj */
result = uobj->pgops->pgo_fault(&ufi, startva, pages, npages,
centeridx, fault_type, access_type,
PGO_LOCKED);
PGO_LOCKED|PGO_SYNCIO);
/* locked: nothing, pgo_fault has unlocked everything */
if (result == VM_PAGER_OK)
@ -925,7 +920,7 @@ ReFault:
uvmexp.fltlget++;
gotpages = npages;
result = uobj->pgops->pgo_get(uobj, ufi.entry->offset +
(void) uobj->pgops->pgo_get(uobj, ufi.entry->offset +
(startva - ufi.entry->start),
pages, &gotpages, centeridx,
access_type & MASK(ufi.entry),
@ -946,29 +941,22 @@ ReFault:
pages[lcv] == PGO_DONTCARE)
continue;
#ifdef DIAGNOSTIC
/*
* pager sanity check: pgo_get with
* PGO_LOCKED should never return a
* released page to us.
*/
if (pages[lcv]->flags & PG_RELEASED)
panic("uvm_fault: pgo_get PGO_LOCKED gave us a RELEASED page");
#endif
KASSERT((pages[lcv]->flags & PG_RELEASED) == 0);
/*
* if center page is resident and not
* PG_BUSY|PG_RELEASED then pgo_get
* made it PG_BUSY for us and gave
* us a handle to it. remember this
* page as "uobjpage." (for later use).
*/
if (lcv == centeridx) {
uobjpage = pages[lcv];
UVMHIST_LOG(maphist, " got uobjpage (0x%x) with locked get",
/*
* if center page is resident and not
* PG_BUSY|PG_RELEASED then pgo_get
* made it PG_BUSY for us and gave
* us a handle to it. remember this
* page as "uobjpage." (for later use).
*/
if (lcv == centeridx) {
uobjpage = pages[lcv];
UVMHIST_LOG(maphist, " got uobjpage "
"(0x%x) with locked get",
uobjpage, 0,0,0);
continue;
continue;
}
/*
@ -987,15 +975,18 @@ ReFault:
" MAPPING: n obj: pm=0x%x, va=0x%x, pg=0x%x",
ufi.orig_map->pmap, currva, pages[lcv], 0);
uvmexp.fltnomap++;
/*
* Since this page isn't the page that's
* actually fauling, ignore pmap_enter()
* failures; it's not critical that we
* enter these right now.
*/
(void) pmap_enter(ufi.orig_map->pmap, currva,
VM_PAGE_TO_PHYS(pages[lcv]),
enter_prot & MASK(ufi.entry),
pages[lcv]->flags & PG_RDONLY ?
VM_PROT_READ : enter_prot & MASK(ufi.entry),
PMAP_CANFAIL |
(wired ? PMAP_WIRED : 0));
@ -1004,18 +995,14 @@ ReFault:
* because we've held the lock the whole time
* we've had the handle.
*/
pages[lcv]->flags &= ~(PG_BUSY); /* un-busy! */
UVM_PAGE_OWN(pages[lcv], NULL);
/* done! */
} /* for "lcv" loop */
} /* "gotpages" != 0 */
/* note: object still _locked_ */
} else {
uobjpage = NULL;
}
/* locked (shadowed): maps(read), amap */
@ -1078,13 +1065,9 @@ ReFault:
case VM_PAGER_REFAULT:
goto ReFault;
case VM_PAGER_ERROR:
/*
* An error occurred while trying to bring in the
* page -- this is the only error we return right
* now.
*/
return (KERN_PROTECTION_FAILURE); /* XXX */
case VM_PAGER_AGAIN:
tsleep(&lbolt, PVM, "fltagain1", 0);
goto ReFault;
default:
#ifdef DIAGNOSTIC
@ -1105,6 +1088,7 @@ ReFault:
/*
* special handling for loaned pages
*/
if (anon->u.an_page->loan_count) {
if ((access_type & VM_PROT_WRITE) == 0) {
@ -1198,21 +1182,13 @@ ReFault:
anon = uvm_analloc();
if (anon)
pg = uvm_pagealloc(NULL, 0, anon, 0);
#ifdef __GNUC__
else
pg = NULL; /* XXX: gcc */
#endif
/* check for out of RAM */
if (anon == NULL || pg == NULL) {
if (anon)
uvm_anfree(anon);
uvmfault_unlockall(&ufi, amap, uobj, oanon);
#ifdef DIAGNOSTIC
if (uvmexp.swpgonly > uvmexp.swpages) {
panic("uvmexp.swpgonly botch");
}
#endif
KASSERT(uvmexp.swpgonly <= uvmexp.swpages);
if (anon == NULL || uvmexp.swpgonly == uvmexp.swpages) {
UVMHIST_LOG(maphist,
"<- failed. out of VM",0,0,0,0);
@ -1243,7 +1219,7 @@ ReFault:
*/
} else {
uvmexp.flt_anon++;
oanon = anon; /* old, locked anon is same as anon */
pg = anon->u.an_page;
@ -1252,7 +1228,7 @@ ReFault:
}
/* locked: maps(read), amap, anon */
/* locked: maps(read), amap, oanon */
/*
* now map the page in ...
@ -1274,10 +1250,7 @@ ReFault:
* as the map may change while we're asleep.
*/
uvmfault_unlockall(&ufi, amap, uobj, oanon);
#ifdef DIAGNOSTIC
if (uvmexp.swpgonly > uvmexp.swpages)
panic("uvmexp.swpgonly botch");
#endif
KASSERT(uvmexp.swpgonly <= uvmexp.swpages);
if (uvmexp.swpgonly == uvmexp.swpages) {
UVMHIST_LOG(maphist,
"<- failed. out of VM",0,0,0,0);
@ -1343,7 +1316,7 @@ Case2:
uobjpage = PGO_DONTCARE;
promote = TRUE; /* always need anon here */
} else {
/* assert(uobjpage != PGO_DONTCARE) */
KASSERT(uobjpage != PGO_DONTCARE);
promote = (access_type & VM_PROT_WRITE) &&
UVM_ET_ISCOPYONWRITE(ufi.entry);
}
@ -1372,24 +1345,19 @@ Case2:
uvmexp.fltget++;
gotpages = 1;
result = uobj->pgops->pgo_get(uobj,
(ufi.orig_rvaddr - ufi.entry->start) + ufi.entry->offset,
&uobjpage, &gotpages, 0,
access_type & MASK(ufi.entry),
ufi.entry->advice, 0);
uoff = (ufi.orig_rvaddr - ufi.entry->start) + ufi.entry->offset;
result = uobj->pgops->pgo_get(uobj, uoff, &uobjpage, &gotpages,
0, access_type & MASK(ufi.entry), ufi.entry->advice,
PGO_SYNCIO);
/* locked: uobjpage(if result OK) */
/*
* recover from I/O
*/
if (result != VM_PAGER_OK) {
#ifdef DIAGNOSTIC
if (result == VM_PAGER_PEND)
panic("uvm_fault: pgo_get got PENDing "
"on non-async I/O");
#endif
KASSERT(result != VM_PAGER_PEND);
if (result == VM_PAGER_AGAIN) {
UVMHIST_LOG(maphist,
@ -1448,11 +1416,8 @@ Case2:
if (uobjpage->flags & PG_RELEASED) {
uvmexp.fltpgrele++;
#ifdef DIAGNOSTIC
if (uobj->pgops->pgo_releasepg == NULL)
panic("uvm_fault: object has no "
"releasepg function");
#endif
KASSERT(uobj->pgops->pgo_releasepg != NULL);
/* frees page */
if (uobj->pgops->pgo_releasepg(uobjpage,NULL))
/* unlock if still alive */
@ -1479,7 +1444,6 @@ Case2:
*/
/* locked: maps(read), amap(if !null), uobj, uobjpage */
}
/*
@ -1616,10 +1580,6 @@ Case2:
pg = uvm_pagealloc(NULL, 0, anon,
(uobjpage == PGO_DONTCARE) ? UVM_PGA_ZERO : 0);
}
#ifdef __GNUC__
else
pg = NULL; /* XXX: gcc */
#endif
/*
* out of memory resources?
@ -1635,21 +1595,15 @@ Case2:
wakeup(uobjpage);
uvm_lock_pageq();
/* make sure it is in queues */
uvm_pageactivate(uobjpage);
uvm_unlock_pageq();
/* un-busy! (still locked) */
uobjpage->flags &= ~(PG_BUSY|PG_WANTED);
UVM_PAGE_OWN(uobjpage, NULL);
}
/* unlock and fail ... */
uvmfault_unlockall(&ufi, amap, uobj, NULL);
#ifdef DIAGNOSTIC
if (uvmexp.swpgonly > uvmexp.swpages) {
panic("uvmexp.swpgonly botch");
}
#endif
KASSERT(uvmexp.swpgonly <= uvmexp.swpages);
if (anon == NULL || uvmexp.swpgonly == uvmexp.swpages) {
UVMHIST_LOG(maphist, " promote: out of VM",
0,0,0,0);
@ -1659,6 +1613,7 @@ Case2:
UVMHIST_LOG(maphist, " out of RAM, waiting for more",
0,0,0,0);
anon->an_ref--;
uvm_anfree(anon);
uvmexp.fltnoram++;
uvm_wait("flt_noram5");
@ -1684,8 +1639,8 @@ Case2:
/*
* dispose of uobjpage. it can't be PG_RELEASED
* since we still hold the object lock. drop
* handle to uobj as well.
* since we still hold the object lock.
* drop handle to uobj as well.
*/
if (uobjpage->flags & PG_WANTED)
@ -1694,10 +1649,11 @@ Case2:
uobjpage->flags &= ~(PG_BUSY|PG_WANTED);
UVM_PAGE_OWN(uobjpage, NULL);
uvm_lock_pageq();
uvm_pageactivate(uobjpage); /* put it back */
uvm_pageactivate(uobjpage);
uvm_unlock_pageq();
simple_unlock(&uobj->vmobjlock);
uobj = NULL;
UVMHIST_LOG(maphist,
" promote uobjpage 0x%x to anon/page 0x%x/0x%x",
uobjpage, anon, pg, 0);
@ -1732,9 +1688,12 @@ Case2:
UVMHIST_LOG(maphist,
" MAPPING: case2: pm=0x%x, va=0x%x, pg=0x%x, promote=%d",
ufi.orig_map->pmap, ufi.orig_rvaddr, pg, promote);
KASSERT(access_type == VM_PROT_READ || (pg->flags & PG_RDONLY) == 0);
if (pmap_enter(ufi.orig_map->pmap, ufi.orig_rvaddr, VM_PAGE_TO_PHYS(pg),
enter_prot, access_type | PMAP_CANFAIL | (wired ? PMAP_WIRED : 0))
pg->flags & PG_RDONLY ? VM_PROT_READ : enter_prot,
access_type | PMAP_CANFAIL | (wired ? PMAP_WIRED : 0))
!= KERN_SUCCESS) {
/*
* No need to undo what we did; we can simply think of
* this as the pmap throwing away the mapping information.
@ -1742,6 +1701,7 @@ Case2:
* We do, however, have to go through the ReFault path,
* as the map may change while we're asleep.
*/
if (pg->flags & PG_WANTED)
wakeup(pg); /* lock still held */
@ -1753,10 +1713,7 @@ Case2:
pg->flags &= ~(PG_BUSY|PG_FAKE|PG_WANTED);
UVM_PAGE_OWN(pg, NULL);
uvmfault_unlockall(&ufi, amap, uobj, NULL);
#ifdef DIAGNOSTIC
if (uvmexp.swpgonly > uvmexp.swpages)
panic("uvmexp.swpgonly botch");
#endif
KASSERT(uvmexp.swpgonly <= uvmexp.swpages);
if (uvmexp.swpgonly == uvmexp.swpages) {
UVMHIST_LOG(maphist,
"<- failed. out of VM",0,0,0,0);
@ -1788,7 +1745,6 @@ Case2:
/* activate it */
uvm_pageactivate(pg);
}
uvm_unlock_pageq();
if (pg->flags & PG_WANTED)
@ -1880,10 +1836,7 @@ uvm_fault_unwire_locked(map, start, end)
paddr_t pa;
struct vm_page *pg;
#ifdef DIAGNOSTIC
if (map->flags & VM_MAP_INTRSAFE)
panic("uvm_fault_unwire_locked: intrsafe map");
#endif
KASSERT((map->flags & VM_MAP_INTRSAFE) == 0);
/*
* we assume that the area we are unwiring has actually been wired

18
sys/uvm/uvm_map.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: uvm_map.c,v 1.85 2000/11/25 06:27:59 chs Exp $ */
/* $NetBSD: uvm_map.c,v 1.86 2000/11/27 08:40:03 chs Exp $ */
/*
* Copyright (c) 1997 Charles D. Cranor and Washington University.
@ -3267,16 +3267,16 @@ uvm_object_printit(uobj, full, pr)
}
}
const char page_flagbits[] =
"\20\4CLEAN\5BUSY\6WANTED\7TABLED\12FAKE\13FILLED\14DIRTY\15RELEASED"
"\16FAULTING\17CLEANCHK";
const char page_pqflagbits[] =
"\20\1FREE\2INACTIVE\3ACTIVE\4LAUNDRY\5ANON\6AOBJ";
/*
* uvm_page_printit: actually print the page
*/
static const char page_flagbits[] =
"\20\1BUSY\2WANTED\3TABLED\4CLEAN\5CLEANCHK\6RELEASED\7FAKE\10RDONLY"
"\11ZERO\15PAGER1";
static const char page_pqflagbits[] =
"\20\1FREE\2INACTIVE\3ACTIVE\4LAUNDRY\5ANON\6AOBJ";
void
uvm_page_printit(pg, full, pr)
struct vm_page *pg;
@ -3294,8 +3294,8 @@ uvm_page_printit(pg, full, pr)
bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf));
(*pr)(" flags=%s, pqflags=%s, vers=%d, wire_count=%d, pa=0x%lx\n",
pgbuf, pqbuf, pg->version, pg->wire_count, (long)pg->phys_addr);
(*pr)(" uobject=%p, uanon=%p, offset=0x%lx loan_count=%d\n",
pg->uobject, pg->uanon, pg->offset, pg->loan_count);
(*pr)(" uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n",
pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count);
#if defined(UVM_PAGE_TRKOWN)
if (pg->flags & PG_BUSY)
(*pr)(" owning process = %d, tag=%s\n",

26
sys/uvm/uvm_map_i.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: uvm_map_i.h,v 1.17 2000/05/08 22:59:35 thorpej Exp $ */
/* $NetBSD: uvm_map_i.h,v 1.18 2000/11/27 08:40:04 chs Exp $ */
/*
* Copyright (c) 1997 Charles D. Cranor and Washington University.
@ -69,8 +69,6 @@
#ifndef _UVM_UVM_MAP_I_H_
#define _UVM_UVM_MAP_I_H_
#include "opt_uvmhist.h"
/*
* uvm_map_i.h
*/
@ -197,16 +195,6 @@ MAP_INLINE void
uvm_map_reference(map)
vm_map_t map;
{
if (__predict_false(map == NULL)) {
#ifdef DIAGNOSTIC
printf("uvm_map_reference: reference to NULL map\n");
#ifdef DDB
Debugger();
#endif
#endif
return;
}
simple_lock(&map->ref_lock);
map->ref_count++;
simple_unlock(&map->ref_lock);
@ -225,20 +213,9 @@ uvm_map_deallocate(map)
{
int c;
if (__predict_false(map == NULL)) {
#ifdef DIAGNOSTIC
printf("uvm_map_deallocate: reference to NULL map\n");
#ifdef DDB
Debugger();
#endif
#endif
return;
}
simple_lock(&map->ref_lock);
c = --map->ref_count;
simple_unlock(&map->ref_lock);
if (c > 0) {
return;
}
@ -249,7 +226,6 @@ uvm_map_deallocate(map)
uvm_unmap(map, map->min_offset, map->max_offset);
pmap_destroy(map->pmap);
FREE(map, M_VMMAP);
}

58
sys/uvm/uvm_mmap.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: uvm_mmap.c,v 1.45 2000/11/24 23:30:01 soren Exp $ */
/* $NetBSD: uvm_mmap.c,v 1.46 2000/11/27 08:40:04 chs Exp $ */
/*
* Copyright (c) 1997 Charles D. Cranor and Washington University.
@ -262,26 +262,6 @@ sys_mincore(p, v, retval)
return (error);
}
#if 0
/*
* munmapfd: unmap file descriptor
*
* XXX: is this acutally a useful function? could it be useful?
*/
void
munmapfd(p, fd)
struct proc *p;
int fd;
{
/*
* XXX should vm_deallocate any regions mapped to this file
*/
p->p_fd->fd_ofileflags[fd] &= ~UF_MAPPED;
}
#endif
/*
* sys_mmap: mmap system call.
*
@ -375,7 +355,9 @@ sys_mmap(p, v, retval)
* not fixed: make sure we skip over the largest possible heap.
* we will refine our guess later (e.g. to account for VAC, etc)
*/
if (addr < round_page((vaddr_t)p->p_vmspace->vm_daddr+MAXDSIZ))
if (addr < round_page((vaddr_t)p->p_vmspace->vm_daddr +
MAXDSIZ))
addr = round_page((vaddr_t)p->p_vmspace->vm_daddr +
MAXDSIZ);
}
@ -1157,36 +1139,8 @@ uvm_mmap(map, addr, size, prot, maxprot, flags, handle, foff, locklimit)
uobj = uvn_attach((void *) vp, (flags & MAP_SHARED) ?
maxprot : (maxprot & ~VM_PROT_WRITE));
/*
* XXXCDC: hack from old code
* don't allow vnodes which have been mapped
* shared-writeable to persist [forces them to be
* flushed out when last reference goes].
* XXXCDC: interesting side effect: avoids a bug.
* note that in WRITE [ufs_readwrite.c] that we
* allocate buffer, uncache, and then do the write.
* the problem with this is that if the uncache causes
* VM data to be flushed to the same area of the file
* we are writing to... in that case we've got the
* buffer locked and our process goes to sleep forever.
*
* XXXCDC: checking maxprot protects us from the
* "persistbug" program but this is not a long term
* solution.
*
* XXXCDC: we don't bother calling uncache with the vp
* VOP_LOCKed since we know that we are already
* holding a valid reference to the uvn (from the
* uvn_attach above), and thus it is impossible for
* the uncache to kill the uvn and trigger I/O.
*/
if (flags & MAP_SHARED) {
if ((prot & VM_PROT_WRITE) ||
(maxprot & VM_PROT_WRITE)) {
uvm_vnp_uncache(vp);
}
}
/* XXX for now, attach doesn't gain a ref */
VREF(vp);
} else {
uobj = udv_attach((void *) &vp->v_rdev,
(flags & MAP_SHARED) ?

154
sys/uvm/uvm_page.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: uvm_page.c,v 1.43 2000/11/09 19:15:28 christos Exp $ */
/* $NetBSD: uvm_page.c,v 1.44 2000/11/27 08:40:04 chs Exp $ */
/*
* Copyright (c) 1997 Charles D. Cranor and Washington University.
@ -70,10 +70,13 @@
* uvm_page.c: page ops.
*/
#include "opt_uvmhist.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/malloc.h>
#include <sys/sched.h>
#include <sys/kernel.h>
#define UVM_PAGE /* pull in uvm_page.h functions */
#include <uvm/uvm.h>
@ -94,8 +97,15 @@ int vm_nphysseg = 0; /* XXXCDC: uvm.nphysseg */
* of the things necessary to do idle page zero'ing efficiently.
* We therefore provide a way to disable it from machdep code here.
*/
/*
* XXX disabled until we can find a way to do this without causing
* problems for either cpu caches or DMA latency.
*/
boolean_t vm_page_zero_enable = FALSE;
boolean_t vm_page_zero_enable = TRUE;
u_long uvm_pgcnt_anon;
u_long uvm_pgcnt_vnode;
extern struct uvm_pagerops uvm_vnodeops;
/*
* local variables
@ -123,7 +133,7 @@ static struct pglist uvm_bootbucket;
*/
static void uvm_pageinsert __P((struct vm_page *));
static void uvm_pageremove __P((struct vm_page *));
/*
* inline functions
@ -160,7 +170,6 @@ uvm_pageinsert(pg)
TAILQ_INSERT_TAIL(&pg->uobject->memq, pg, listq); /* put in object */
pg->flags |= PG_TABLED;
pg->uobject->uo_npages++;
}
/*
@ -170,21 +179,14 @@ uvm_pageinsert(pg)
* => caller must lock page queues
*/
void __inline
static __inline void
uvm_pageremove(pg)
struct vm_page *pg;
{
struct pglist *buck;
int s;
#ifdef DIAGNOSTIC
if ((pg->flags & (PG_FAULTING)) != 0)
panic("uvm_pageremove: page is faulting");
#endif
if ((pg->flags & PG_TABLED) == 0)
return; /* XXX: log */
KASSERT(pg->flags & PG_TABLED);
buck = &uvm.page_hash[uvm_pagehash(pg->uobject,pg->offset)];
s = splimp();
simple_lock(&uvm.hashlock);
@ -192,6 +194,10 @@ uvm_pageremove(pg)
simple_unlock(&uvm.hashlock);
splx(s);
if (pg->uobject->pgops == &uvm_vnodeops) {
uvm_pgcnt_vnode--;
}
/* object should be locked */
TAILQ_REMOVE(&pg->uobject->memq, pg, listq);
@ -199,7 +205,6 @@ uvm_pageremove(pg)
pg->uobject->uo_npages--;
pg->uobject = NULL;
pg->version++;
}
/*
@ -217,7 +222,6 @@ uvm_page_init(kvm_startp, kvm_endp)
int lcv, i;
paddr_t paddr;
/*
* step 1: init the page queues and page queue locks
*/
@ -238,7 +242,7 @@ uvm_page_init(kvm_startp, kvm_endp)
*/
uvm.page_nhash = 1; /* 1 bucket */
uvm.page_hashmask = 0; /* mask for hash function */
uvm.page_hashmask = 0; /* mask for hash function */
uvm.page_hash = &uvm_bootbucket; /* install bootstrap bucket */
TAILQ_INIT(uvm.page_hash); /* init hash table */
simple_lock_init(&uvm.hashlock); /* init hash table lock */
@ -291,7 +295,6 @@ uvm_page_init(kvm_startp, kvm_endp)
*/
for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) {
n = vm_physmem[lcv].end - vm_physmem[lcv].start;
if (n > pagecount) {
printf("uvm_page_init: lost %ld page(s) in init\n",
@ -317,6 +320,7 @@ uvm_page_init(kvm_startp, kvm_endp)
}
}
}
/*
* step 5: pass up the values of virtual_space_start and
* virtual_space_end (obtained by uvm_pageboot_alloc) to the upper
@ -327,10 +331,11 @@ uvm_page_init(kvm_startp, kvm_endp)
*kvm_endp = trunc_page(virtual_space_end);
/*
* step 6: init pagedaemon lock
* step 6: init locks for kernel threads
*/
simple_lock_init(&uvm.pagedaemon_lock);
simple_lock_init(&uvm.aiodoned_lock);
/*
* step 7: init reserve thresholds
@ -342,10 +347,6 @@ uvm_page_init(kvm_startp, kvm_endp)
/*
* step 8: determine if we should zero pages in the idle
* loop.
*
* XXXJRT - might consider zero'ing up to the target *now*,
* but that could take an awfully long time if you
* have a lot of memory.
*/
uvm.page_idle_zero = vm_page_zero_enable;
@ -360,7 +361,6 @@ uvm_page_init(kvm_startp, kvm_endp)
* uvm_setpagesize: set the page size
*
* => sets page_shift and page_mask from uvmexp.pagesize.
* => XXXCDC: move global vars.
*/
void
@ -889,22 +889,20 @@ uvm_pagealloc_strat(obj, off, anon, flags, strat, free_list)
struct pgfreelist *pgfl;
boolean_t use_reserve;
#ifdef DIAGNOSTIC
/* sanity check */
if (obj && anon)
panic("uvm_pagealloc: obj and anon != NULL");
#endif
s = uvm_lock_fpageq(); /* lock free page queue */
KASSERT(obj == NULL || anon == NULL);
KASSERT(off == trunc_page(off));
s = uvm_lock_fpageq();
/*
* check to see if we need to generate some free pages waking
* the pagedaemon.
*/
if (uvmexp.free < uvmexp.freemin || (uvmexp.free < uvmexp.freetarg &&
uvmexp.inactive < uvmexp.inactarg))
if (uvmexp.free + uvmexp.paging < uvmexp.freemin ||
(uvmexp.free + uvmexp.paging < uvmexp.freetarg &&
uvmexp.inactive < uvmexp.inactarg)) {
wakeup(&uvm.pagedaemon);
}
/*
* fail if any of these conditions is true:
@ -957,11 +955,7 @@ uvm_pagealloc_strat(obj, off, anon, flags, strat, free_list)
case UVM_PGA_STRAT_ONLY:
case UVM_PGA_STRAT_FALLBACK:
/* Attempt to allocate from the specified free list. */
#ifdef DIAGNOSTIC
if (free_list >= VM_NFREELIST || free_list < 0)
panic("uvm_pagealloc_strat: bad free list %d",
free_list);
#endif
KASSERT(free_list >= 0 && free_list < VM_NFREELIST);
pgfl = &uvm.page_free[free_list];
if ((pg = TAILQ_FIRST((freeq =
&pgfl->pgfl_queues[try1]))) != NULL ||
@ -1012,11 +1006,10 @@ uvm_pagealloc_strat(obj, off, anon, flags, strat, free_list)
pg->uanon = anon;
pg->flags = PG_BUSY|PG_CLEAN|PG_FAKE;
pg->version++;
pg->wire_count = 0;
pg->loan_count = 0;
if (anon) {
anon->u.an_page = pg;
pg->pqflags = PQ_ANON;
uvm_pgcnt_anon++;
} else {
if (obj)
uvm_pageinsert(pg);
@ -1074,8 +1067,6 @@ uvm_pagerealloc(pg, newobj, newoff)
pg->version++;
uvm_pageinsert(pg);
}
return;
}
@ -1089,14 +1080,20 @@ uvm_pagerealloc(pg, newobj, newoff)
* => assumes all valid mappings of pg are gone
*/
void uvm_pagefree(pg)
struct vm_page *pg;
void
uvm_pagefree(pg)
struct vm_page *pg;
{
int s;
int saved_loan_count = pg->loan_count;
#ifdef DEBUG
if (pg->uobject == (void *)0xdeadbeef &&
pg->uanon == (void *)0xdeadbeef) {
panic("uvm_pagefree: freeing free page %p\n", pg);
}
#endif
/*
* if the page was an object page (and thus "TABLED"), remove it
* from the object.
@ -1105,7 +1102,7 @@ struct vm_page *pg;
if (pg->flags & PG_TABLED) {
/*
* if the object page is on loan we are going to drop ownership.
* if the object page is on loan we are going to drop ownership.
* it is possible that an anon will take over as owner for this
* page later on. the anon will want a !PG_CLEAN page so that
* it knows it needs to allocate swap if it wants to page the
@ -1114,7 +1111,6 @@ struct vm_page *pg;
if (saved_loan_count)
pg->flags &= ~PG_CLEAN; /* in case an anon takes over */
uvm_pageremove(pg);
/*
@ -1125,9 +1121,9 @@ struct vm_page *pg;
* return (when the last loan is dropped, then the page can be
* freed by whatever was holding the last loan).
*/
if (saved_loan_count)
return;
} else if (saved_loan_count && (pg->pqflags & PQ_ANON)) {
/*
@ -1137,19 +1133,12 @@ struct vm_page *pg;
* note that the kernel can't change the loan status of our
* page as long as we are holding PQ lock.
*/
pg->pqflags &= ~PQ_ANON;
pg->uanon = NULL;
return;
}
#ifdef DIAGNOSTIC
if (saved_loan_count) {
printf("uvm_pagefree: warning: freeing page with a loan "
"count of %d\n", saved_loan_count);
panic("uvm_pagefree: loan count");
}
#endif
KASSERT(saved_loan_count == 0);
/*
* now remove the page from the queues
@ -1172,13 +1161,17 @@ struct vm_page *pg;
/*
* if the page was wired, unwire it now.
*/
if (pg->wire_count) {
pg->wire_count = 0;
uvmexp.wired--;
}
if (pg->uanon) {
uvm_pgcnt_anon--;
}
/*
* and put on free queue
* and put on free queue
*/
pg->flags &= ~PG_ZERO;
@ -1200,6 +1193,51 @@ struct vm_page *pg;
uvm_unlock_fpageq(s);
}
/*
* uvm_page_unbusy: unbusy an array of pages.
*
* => pages must either all belong to the same object, or all belong to anons.
* => if pages are object-owned, object must be locked.
* => if pages are anon-owned, anons must be unlockd and have 0 refcount.
*/
void
uvm_page_unbusy(pgs, npgs)
struct vm_page **pgs;
int npgs;
{
struct vm_page *pg;
struct uvm_object *uobj;
int i;
UVMHIST_FUNC("uvm_page_unbusy"); UVMHIST_CALLED(ubchist);
for (i = 0; i < npgs; i++) {
pg = pgs[i];
if (pg == NULL) {
continue;
}
if (pg->flags & PG_WANTED) {
wakeup(pg);
}
if (pg->flags & PG_RELEASED) {
UVMHIST_LOG(ubchist, "releasing pg %p", pg,0,0,0);
uobj = pg->uobject;
if (uobj != NULL) {
uobj->pgops->pgo_releasepg(pg, NULL);
} else {
pg->flags &= ~(PG_BUSY);
UVM_PAGE_OWN(pg, NULL);
uvm_anfree(pg->uanon);
}
} else {
UVMHIST_LOG(ubchist, "unbusying pg %p", pg,0,0,0);
pg->flags &= ~(PG_WANTED|PG_BUSY);
UVM_PAGE_OWN(pg, NULL);
}
}
}
#if defined(UVM_PAGE_TRKOWN)
/*
* uvm_page_own: set or release page ownership

74
sys/uvm/uvm_page.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: uvm_page.h,v 1.17 2000/10/03 20:50:49 mrg Exp $ */
/* $NetBSD: uvm_page.h,v 1.18 2000/11/27 08:40:05 chs Exp $ */
/*
* Copyright (c) 1997 Charles D. Cranor and Washington University.
@ -118,27 +118,27 @@
#include <uvm/uvm_pglist.h>
struct vm_page {
TAILQ_ENTRY(vm_page) pageq; /* queue info for FIFO
* queue or free list (P) */
TAILQ_ENTRY(vm_page) hashq; /* hash table links (O)*/
TAILQ_ENTRY(vm_page) listq; /* pages in same object (O)*/
TAILQ_ENTRY(vm_page) pageq; /* queue info for FIFO
* queue or free list (P) */
TAILQ_ENTRY(vm_page) hashq; /* hash table links (O)*/
TAILQ_ENTRY(vm_page) listq; /* pages in same object (O)*/
struct vm_anon *uanon; /* anon (O,P) */
struct uvm_object *uobject; /* object (O,P) */
voff_t offset; /* offset into object (O,P) */
struct vm_anon *uanon; /* anon (O,P) */
struct uvm_object *uobject; /* object (O,P) */
voff_t offset; /* offset into object (O,P) */
u_short flags; /* object flags [O] */
u_short version; /* version count [O] */
u_short wire_count; /* wired down map refs [P] */
u_short pqflags; /* page queue flags [P] */
u_int loan_count; /* number of active loans
* to read: [O or P]
* to modify: [O _and_ P] */
paddr_t phys_addr; /* physical address of page */
u_short flags; /* object flags [O] */
u_short version; /* version count [O] */
u_short wire_count; /* wired down map refs [P] */
u_short pqflags; /* page queue flags [P] */
u_int loan_count; /* number of active loans
* to read: [O or P]
* to modify: [O _and_ P] */
paddr_t phys_addr; /* physical address of page */
#if defined(UVM_PAGE_TRKOWN)
/* debugging fields to track page ownership */
pid_t owner; /* proc that set PG_BUSY */
char *owner_tag; /* why it was set busy */
/* debugging fields to track page ownership */
pid_t owner; /* proc that set PG_BUSY */
char *owner_tag; /* why it was set busy */
#endif
};
@ -157,25 +157,23 @@ struct vm_page {
* PG_ZERO is used to indicate that a page has been pre-zero'd. This flag
* is only set when the page is on no queues, and is cleared when the page
* is placed on the free list.
*
* possible deadwood: PG_FAULTING, PQ_LAUNDRY
*/
#define PG_BUSY 0x0001 /* page is locked */
#define PG_WANTED 0x0002 /* someone is waiting for page */
#define PG_TABLED 0x0004 /* page is in VP table */
#define PG_CLEAN 0x0008 /* page has not been modified */
#define PG_BUSY 0x0010 /* page is in transit */
#define PG_WANTED 0x0020 /* someone is waiting for page */
#define PG_TABLED 0x0040 /* page is in VP table */
#define PG_ZERO 0x0100 /* page is pre-zero'd */
#define PG_FAKE 0x0200 /* page is placeholder for pagein */
#define PG_FILLED 0x0400 /* client flag to set when filled */
#define PG_DIRTY 0x0800 /* client flag to set when dirty */
#define PG_RELEASED 0x1000 /* page released while paging */
#define PG_FAULTING 0x2000 /* page is being faulted in */
#define PG_CLEANCHK 0x4000 /* clean bit has been checked */
#define PG_CLEANCHK 0x0010 /* clean bit has been checked */
#define PG_RELEASED 0x0020 /* page released while paging */
#define PG_FAKE 0x0040 /* page is not yet initialized */
#define PG_RDONLY 0x0080 /* page must be mapped read-only */
#define PG_ZERO 0x0100 /* page is pre-zero'd */
#define PG_PAGER1 0x1000 /* pager-specific flag */
#define PQ_FREE 0x0001 /* page is on free list */
#define PQ_INACTIVE 0x0002 /* page is in inactive list */
#define PQ_ACTIVE 0x0004 /* page is in active list */
#define PQ_LAUNDRY 0x0008 /* page is being cleaned now */
#define PQ_ANON 0x0010 /* page is part of an anon, rather
than an uvm_object */
#define PQ_AOBJ 0x0020 /* page is part of an anonymous
@ -237,12 +235,9 @@ extern boolean_t vm_page_zero_enable;
* ordered, in LRU-like fashion.
*/
extern
struct pglist vm_page_queue_free; /* memory free queue */
extern
struct pglist vm_page_queue_active; /* active memory queue */
extern
struct pglist vm_page_queue_inactive; /* inactive memory queue */
extern struct pglist vm_page_queue_free; /* memory free queue */
extern struct pglist vm_page_queue_active; /* active memory queue */
extern struct pglist vm_page_queue_inactive; /* inactive memory queue */
/*
* physical memory config is stored in vm_physmem.
@ -283,9 +278,8 @@ vaddr_t uvm_pageboot_alloc __P((vsize_t));
PAGE_INLINE void uvm_pagecopy __P((struct vm_page *, struct vm_page *));
PAGE_INLINE void uvm_pagedeactivate __P((struct vm_page *));
void uvm_pagefree __P((struct vm_page *));
void uvm_page_unbusy __P((struct vm_page **, int));
PAGE_INLINE struct vm_page *uvm_pagelookup __P((struct uvm_object *, voff_t));
void uvm_pageremove __P((struct vm_page *));
/* uvm_pagerename: not needed */
PAGE_INLINE void uvm_pageunwire __P((struct vm_page *));
PAGE_INLINE void uvm_pagewait __P((struct vm_page *, int));
PAGE_INLINE void uvm_pagewake __P((struct vm_page *));

282
sys/uvm/uvm_pager.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: uvm_pager.c,v 1.34 2000/11/24 22:41:39 chs Exp $ */
/* $NetBSD: uvm_pager.c,v 1.35 2000/11/27 08:40:05 chs Exp $ */
/*
*
@ -44,21 +44,27 @@
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/malloc.h>
#include <sys/pool.h>
#include <sys/vnode.h>
#define UVM_PAGER
#include <uvm/uvm.h>
struct pool *uvm_aiobuf_pool;
/*
* list of uvm pagers in the system
*/
extern struct uvm_pagerops uvm_deviceops;
extern struct uvm_pagerops uvm_vnodeops;
extern struct uvm_pagerops ubc_pager;
struct uvm_pagerops *uvmpagerops[] = {
&aobj_pager,
&uvm_deviceops,
&uvm_vnodeops,
&ubc_pager,
};
/*
@ -68,7 +74,8 @@ struct uvm_pagerops *uvmpagerops[] = {
vm_map_t pager_map; /* XXX */
simple_lock_data_t pager_map_wanted_lock;
boolean_t pager_map_wanted; /* locked by pager map */
static vaddr_t emergva;
static boolean_t emerginuse;
/*
* uvm_pager_init: init pagers (at boot time)
@ -83,10 +90,12 @@ uvm_pager_init()
* init pager map
*/
pager_map = uvm_km_suballoc(kernel_map, &uvm.pager_sva, &uvm.pager_eva,
PAGER_MAP_SIZE, 0, FALSE, NULL);
simple_lock_init(&pager_map_wanted_lock);
pager_map_wanted = FALSE;
pager_map = uvm_km_suballoc(kernel_map, &uvm.pager_sva, &uvm.pager_eva,
PAGER_MAP_SIZE, 0, FALSE, NULL);
simple_lock_init(&pager_map_wanted_lock);
pager_map_wanted = FALSE;
emergva = uvm_km_valloc(kernel_map, MAXBSIZE);
emerginuse = FALSE;
/*
* init ASYNC I/O queue
@ -112,22 +121,19 @@ uvm_pager_init()
*/
vaddr_t
uvm_pagermapin(pps, npages, aiop, flags)
uvm_pagermapin(pps, npages, flags)
struct vm_page **pps;
int npages;
struct uvm_aiodesc **aiop; /* OUT */
int flags;
{
vsize_t size;
vaddr_t kva;
struct uvm_aiodesc *aio;
vaddr_t cva;
struct vm_page *pp;
vm_prot_t prot;
UVMHIST_FUNC("uvm_pagermapin"); UVMHIST_CALLED(maphist);
UVMHIST_LOG(maphist,"(pps=0x%x, npages=%d, aiop=0x%x, flags=0x%x)",
pps, npages, aiop, flags);
UVMHIST_LOG(maphist,"(pps=0x%x, npages=%d)", pps, npages,0,0);
/*
* compute protection. outgoing I/O only needs read
@ -139,24 +145,26 @@ uvm_pagermapin(pps, npages, aiop, flags)
prot |= VM_PROT_WRITE;
ReStart:
if (aiop) {
MALLOC(aio, struct uvm_aiodesc *, sizeof(*aio), M_TEMP,
(flags & UVMPAGER_MAPIN_WAITOK));
if (aio == NULL)
return(0);
*aiop = aio;
} else {
aio = NULL;
}
size = npages << PAGE_SHIFT;
kva = 0; /* let system choose VA */
if (uvm_map(pager_map, &kva, size, NULL,
UVM_UNKNOWN_OFFSET, 0, UVM_FLAG_NOMERGE) != KERN_SUCCESS) {
if (curproc == uvm.pagedaemon_proc) {
simple_lock(&pager_map_wanted_lock);
if (emerginuse) {
UVM_UNLOCK_AND_WAIT(&emergva,
&pager_map_wanted_lock, FALSE,
"emergva", 0);
goto ReStart;
}
emerginuse = TRUE;
simple_unlock(&pager_map_wanted_lock);
kva = emergva;
KASSERT(npages <= MAXBSIZE >> PAGE_SHIFT);
goto enter;
}
if ((flags & UVMPAGER_MAPIN_WAITOK) == 0) {
if (aio)
FREE(aio, M_TEMP);
UVMHIST_LOG(maphist,"<- NOWAIT failed", 0,0,0,0);
return(0);
}
@ -164,16 +172,17 @@ ReStart:
pager_map_wanted = TRUE;
UVMHIST_LOG(maphist, " SLEEPING on pager_map",0,0,0,0);
UVM_UNLOCK_AND_WAIT(pager_map, &pager_map_wanted_lock, FALSE,
"pager_map",0);
"pager_map", 0);
goto ReStart;
}
enter:
/* got it */
for (cva = kva ; size != 0 ; size -= PAGE_SIZE, cva += PAGE_SIZE) {
pp = *pps++;
#ifdef DEBUG
if ((pp->flags & PG_BUSY) == 0)
panic("uvm_pagermapin: page not busy");
panic("uvm_pagermapin: pg %p not busy", pp);
#endif
pmap_enter(vm_map_pmap(pager_map), cva, VM_PAGE_TO_PHYS(pp),
prot, PMAP_WIRED | prot);
@ -198,13 +207,22 @@ uvm_pagermapout(kva, npages)
vsize_t size = npages << PAGE_SHIFT;
vm_map_entry_t entries;
UVMHIST_FUNC("uvm_pagermapout"); UVMHIST_CALLED(maphist);
UVMHIST_LOG(maphist, " (kva=0x%x, npages=%d)", kva, npages,0,0);
/*
* duplicate uvm_unmap, but add in pager_map_wanted handling.
*/
if (kva == emergva) {
simple_lock(&pager_map_wanted_lock);
emerginuse = FALSE;
wakeup(&emergva);
simple_unlock(&pager_map_wanted_lock);
entries = NULL;
goto remove;
}
vm_map_lock(pager_map);
(void) uvm_unmap_remove(pager_map, kva, kva + size, &entries);
simple_lock(&pager_map_wanted_lock);
@ -214,6 +232,8 @@ uvm_pagermapout(kva, npages)
}
simple_unlock(&pager_map_wanted_lock);
vm_map_unlock(pager_map);
remove:
pmap_remove(pmap_kernel(), kva, kva + (npages << PAGE_SHIFT));
if (entries)
uvm_unmap_detach(entries, 0);
@ -251,7 +271,7 @@ uvm_mk_pcluster(uobj, pps, npages, center, flags, mlo, mhi)
{
struct vm_page **ppsp, *pclust;
voff_t lo, hi, curoff;
int center_idx, forward;
int center_idx, forward, incr;
UVMHIST_FUNC("uvm_mk_pcluster"); UVMHIST_CALLED(maphist);
/*
@ -273,9 +293,11 @@ uvm_mk_pcluster(uobj, pps, npages, center, flags, mlo, mhi)
if (hi > mhi)
hi = mhi;
}
if ((hi - lo) >> PAGE_SHIFT > *npages) { /* pps too small, bail out! */
if ((hi - lo) >> PAGE_SHIFT > *npages) { /* pps too small, bail out! */
#ifdef DIAGNOSTIC
printf("uvm_mk_pcluster: provided page array too small (fixed)\n");
printf("uvm_mk_pcluster uobj %p npages %d lo 0x%llx hi 0x%llx "
"flags 0x%x\n", uobj, *npages, (long long)lo,
(long long)hi, flags);
#endif
pps[0] = center;
*npages = 1;
@ -291,7 +313,7 @@ uvm_mk_pcluster(uobj, pps, npages, center, flags, mlo, mhi)
pps[center_idx] = center; /* plug in the center page */
ppsp = &pps[center_idx];
*npages = 1;
/*
* attempt to cluster around the left [backward], and then
* the right side [forward].
@ -303,21 +325,23 @@ uvm_mk_pcluster(uobj, pps, npages, center, flags, mlo, mhi)
*/
for (forward = 0 ; forward <= 1 ; forward++) {
curoff = center->offset + (forward ? PAGE_SIZE : -PAGE_SIZE);
incr = forward ? PAGE_SIZE : -PAGE_SIZE;
curoff = center->offset + incr;
for ( ;(forward == 0 && curoff >= lo) ||
(forward && curoff < hi);
curoff += (forward ? 1 : -1) << PAGE_SHIFT) {
curoff += incr) {
pclust = uvm_pagelookup(uobj, curoff); /* lookup page */
if (pclust == NULL)
if (pclust == NULL) {
break; /* no page */
}
/* handle active pages */
/* NOTE: inactive pages don't have pmap mappings */
if ((pclust->pqflags & PQ_INACTIVE) == 0) {
if ((flags & PGO_DOACTCLUST) == 0)
if ((flags & PGO_DOACTCLUST) == 0) {
/* dont want mapped pages at all */
break;
}
/* make sure "clean" bit is sync'd */
if ((pclust->flags & PG_CLEANCHK) == 0) {
@ -330,13 +354,16 @@ uvm_mk_pcluster(uobj, pps, npages, center, flags, mlo, mhi)
pclust->flags |= PG_CLEANCHK;
}
}
/* is page available for cleaning and does it need it */
if ((pclust->flags & (PG_CLEAN|PG_BUSY)) != 0)
if ((pclust->flags & (PG_CLEAN|PG_BUSY)) != 0) {
break; /* page is already clean or is busy */
}
/* yes! enroll the page in our array */
pclust->flags |= PG_BUSY; /* busy! */
UVM_PAGE_OWN(pclust, "uvm_mk_pcluster");
/* XXX: protect wired page? see above comment. */
pmap_page_protect(pclust, VM_PROT_READ);
if (!forward) {
@ -346,7 +373,7 @@ uvm_mk_pcluster(uobj, pps, npages, center, flags, mlo, mhi)
/* move forward one page */
ppsp[*npages] = pclust;
}
*npages = *npages + 1;
(*npages)++;
}
}
@ -409,6 +436,7 @@ uvm_pager_put(uobj, pg, ppsp_ptr, npages, flags, start, stop)
int result;
daddr_t swblk;
struct vm_page **ppsp = *ppsp_ptr;
UVMHIST_FUNC("uvm_pager_put"); UVMHIST_CALLED(ubchist);
/*
* note that uobj is null if we are doing a swap-backed pageout.
@ -459,12 +487,12 @@ uvm_pager_put(uobj, pg, ppsp_ptr, npages, flags, start, stop)
ReTry:
if (uobj) {
/* object is locked */
result = uobj->pgops->pgo_put(uobj, ppsp, *npages,
flags & PGO_SYNCIO);
result = uobj->pgops->pgo_put(uobj, ppsp, *npages, flags);
UVMHIST_LOG(ubchist, "put -> %d", result, 0,0,0);
/* object is now unlocked */
} else {
/* nothing locked */
result = uvm_swap_put(swblk, ppsp, *npages, flags & PGO_SYNCIO);
result = uvm_swap_put(swblk, ppsp, *npages, flags);
/* nothing locked */
}
@ -565,7 +593,7 @@ ReTry:
/*
* a pager error occured (even after dropping the cluster, if there
* was one). give up! the caller only has one page ("pg")
* was one). give up! the caller only has one page ("pg")
* to worry about.
*/
@ -610,7 +638,8 @@ uvm_pager_dropcluster(uobj, pg, ppsp, npages, flags)
for (lcv = 0 ; lcv < *npages ; lcv++) {
if (ppsp[lcv] == pg) /* skip "pg" */
/* skip "pg" or empty slot */
if (ppsp[lcv] == pg || ppsp[lcv] == NULL)
continue;
/*
@ -637,9 +666,10 @@ uvm_pager_dropcluster(uobj, pg, ppsp, npages, flags)
}
/* did someone want the page while we had it busy-locked? */
if (ppsp[lcv]->flags & PG_WANTED)
if (ppsp[lcv]->flags & PG_WANTED) {
/* still holding obj lock */
wakeup(ppsp[lcv]);
}
/* if page was released, release it. otherwise un-busy it */
if (ppsp[lcv]->flags & PG_RELEASED) {
@ -690,7 +720,7 @@ uvm_pager_dropcluster(uobj, pg, ppsp, npages, flags)
continue; /* next page */
} else {
ppsp[lcv]->flags &= ~(PG_BUSY|PG_WANTED);
ppsp[lcv]->flags &= ~(PG_BUSY|PG_WANTED|PG_FAKE);
UVM_PAGE_OWN(ppsp[lcv], NULL);
}
@ -713,3 +743,167 @@ uvm_pager_dropcluster(uobj, pg, ppsp, npages, flags)
}
}
}
/*
* interrupt-context iodone handler for nested i/o bufs.
*
* => must be at splbio().
*/
void
uvm_aio_biodone1(bp)
struct buf *bp;
{
struct buf *mbp = bp->b_private;
KASSERT(mbp != bp);
if (bp->b_flags & B_ERROR) {
mbp->b_flags |= B_ERROR;
mbp->b_error = bp->b_error;
}
mbp->b_resid -= bp->b_bcount;
pool_put(&bufpool, bp);
if (mbp->b_resid == 0) {
biodone(mbp);
}
}
/*
* interrupt-context iodone handler for single-buf i/os
* or the top-level buf of a nested-buf i/o.
*
* => must be at splbio().
*/
void
uvm_aio_biodone(bp)
struct buf *bp;
{
/* reset b_iodone for when this is a single-buf i/o. */
bp->b_iodone = uvm_aio_aiodone;
simple_lock(&uvm.aiodoned_lock); /* locks uvm.aio_done */
TAILQ_INSERT_TAIL(&uvm.aio_done, bp, b_freelist);
wakeup(&uvm.aiodoned);
simple_unlock(&uvm.aiodoned_lock);
}
/*
* uvm_aio_aiodone: do iodone processing for async i/os.
* this should be called in thread context, not interrupt context.
*/
void
uvm_aio_aiodone(bp)
struct buf *bp;
{
int npages = bp->b_bufsize >> PAGE_SHIFT;
struct vm_page *pg, *pgs[npages];
struct uvm_object *uobj;
int s, i;
boolean_t release, write, swap;
UVMHIST_FUNC("uvm_aio_aiodone"); UVMHIST_CALLED(ubchist);
UVMHIST_LOG(ubchist, "bp %p", bp, 0,0,0);
release = (bp->b_flags & (B_ERROR|B_READ)) == (B_ERROR|B_READ);
write = (bp->b_flags & B_READ) == 0;
/* XXXUBC B_NOCACHE is for swap pager, should be done differently */
if (write && !(bp->b_flags & B_NOCACHE)) {
/* XXXUBC */
void softdep_pageiodone(struct buf *);
softdep_pageiodone(bp);
}
uobj = NULL;
for (i = 0; i < npages; i++) {
pgs[i] = uvm_pageratop((vaddr_t)bp->b_data + (i << PAGE_SHIFT));
UVMHIST_LOG(ubchist, "pgs[%d] = %p", i, pgs[i],0,0);
}
uvm_pagermapout((vaddr_t)bp->b_data, npages);
for (i = 0; i < npages; i++) {
pg = pgs[i];
if (i == 0) {
swap = (pg->pqflags & PQ_SWAPBACKED) != 0;
if (!swap) {
uobj = pg->uobject;
simple_lock(&uobj->vmobjlock);
}
}
KASSERT(swap || pg->uobject == uobj);
if (swap) {
if (pg->pqflags & PQ_ANON) {
simple_lock(&pg->uanon->an_lock);
} else {
simple_lock(&pg->uobject->vmobjlock);
}
}
/*
* if this is a read and we got an error, mark the pages
* PG_RELEASED so that uvm_page_unbusy() will free them.
*/
if (release) {
pg->flags |= PG_RELEASED;
continue;
}
KASSERT(!write || (pgs[i]->flags & PG_FAKE) == 0);
/*
* if this is a read and the page is PG_FAKE
* or this was a write, mark the page PG_CLEAN and not PG_FAKE.
*/
if (pgs[i]->flags & PG_FAKE || write) {
pmap_clear_reference(pgs[i]);
pmap_clear_modify(pgs[i]);
pgs[i]->flags |= PG_CLEAN;
pgs[i]->flags &= ~PG_FAKE;
}
if (swap) {
if (pg->pqflags & PQ_ANON) {
simple_unlock(&pg->uanon->an_lock);
} else {
simple_unlock(&pg->uobject->vmobjlock);
}
}
}
uvm_page_unbusy(pgs, npages);
if (!swap) {
simple_unlock(&uobj->vmobjlock);
}
s = splbio();
if (write && (bp->b_flags & B_AGE) != 0) {
vwakeup(bp);
}
pool_put(&bufpool, bp);
splx(s);
}
/*
* translate unix errno values to VM_PAGER_*.
*/
int
uvm_errno2vmerror(errno)
int errno;
{
switch (errno) {
case 0:
return VM_PAGER_OK;
case EINVAL:
return VM_PAGER_BAD;
case EINPROGRESS:
return VM_PAGER_PEND;
case EIO:
return VM_PAGER_ERROR;
case EAGAIN:
return VM_PAGER_AGAIN;
case EBUSY:
return VM_PAGER_UNLOCK;
default:
return VM_PAGER_ERROR;
}
}

33
sys/uvm/uvm_pager.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: uvm_pager.h,v 1.19 2000/11/27 08:19:51 chs Exp $ */
/* $NetBSD: uvm_pager.h,v 1.20 2000/11/27 08:40:05 chs Exp $ */
/*
*
@ -81,21 +81,6 @@
* uvm_pager.h
*/
/*
* async pager i/o descriptor structure
*/
TAILQ_HEAD(uvm_aiohead, uvm_aiodesc);
struct uvm_aiodesc {
void (*aiodone) __P((struct uvm_aiodesc *));
/* aio done function */
vaddr_t kva; /* KVA of mapped page(s) */
int npages; /* # of pages in I/O req */
void *pd_ptr; /* pager-dependent pointer */
TAILQ_ENTRY(uvm_aiodesc) aioq; /* linked list of aio's */
};
/*
* pager ops
*/
@ -132,22 +117,22 @@ struct uvm_pagerops {
/* pager flags [mostly for flush] */
#define PGO_CLEANIT 0x001 /* write dirty pages to backing store */
#define PGO_SYNCIO 0x002 /* if PGO_CLEAN: use sync I/O? */
/*
* obviously if neither PGO_INVALIDATE or PGO_FREE are set then the pages
* stay where they are.
*/
#define PGO_SYNCIO 0x002 /* if PGO_CLEANIT: use sync I/O? */
#define PGO_DEACTIVATE 0x004 /* deactivate flushed pages */
#define PGO_FREE 0x008 /* free flushed pages */
/* if PGO_FREE is not set then the pages stay where they are. */
#define PGO_ALLPAGES 0x010 /* flush whole object/get all pages */
#define PGO_DOACTCLUST 0x020 /* flag to mk_pcluster to include active */
#define PGO_LOCKED 0x040 /* fault data structures are locked [get] */
#define PGO_PDFREECLUST 0x080 /* daemon's free cluster flag [uvm_pager_put] */
#define PGO_REALLOCSWAP 0x100 /* reallocate swap area [pager_dropcluster] */
#define PGO_OVERWRITE 0x200 /* pages will be overwritten before unlocked */
#define PGO_WEAK 0x400 /* "weak" put, for nfs */
#define PGO_PASTEOF 0x800 /* allow allocation of pages past EOF */
/* page we are not interested in getting */
#define PGO_DONTCARE ((struct vm_page *) -1) /* [get only] */
#define PGO_DONTCARE ((struct vm_page *) -1L) /* [get only] */
#ifdef _KERNEL
@ -175,12 +160,12 @@ int uvm_pager_put __P((struct uvm_object *, struct vm_page *,
PAGER_INLINE struct vm_page *uvm_pageratop __P((vaddr_t));
vaddr_t uvm_pagermapin __P((struct vm_page **, int,
struct uvm_aiodesc **, int));
vaddr_t uvm_pagermapin __P((struct vm_page **, int, int));
void uvm_pagermapout __P((vaddr_t, int));
struct vm_page **uvm_mk_pcluster __P((struct uvm_object *, struct vm_page **,
int *, struct vm_page *, int,
voff_t, voff_t));
int uvm_errno2vmerror __P((int));
/* Flags to uvm_pagermapin() */
#define UVMPAGER_MAPIN_WAITOK 0x01 /* it's okay to wait */

476
sys/uvm/uvm_pdaemon.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: uvm_pdaemon.c,v 1.23 2000/08/20 10:24:14 bjh21 Exp $ */
/* $NetBSD: uvm_pdaemon.c,v 1.24 2000/11/27 08:40:05 chs Exp $ */
/*
* Copyright (c) 1997 Charles D. Cranor and Washington University.
@ -77,9 +77,13 @@
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/pool.h>
#include <sys/buf.h>
#include <uvm/uvm.h>
extern u_long uvm_pgcnt_vnode;
extern struct uvm_pagerops uvm_vnodeops;
/*
* UVMPD_NUMDIRTYREACTS is how many dirty pages the pagedeamon will reactivate
* in a pass thru the inactive list when swap is full. the value should be
@ -194,10 +198,8 @@ void
uvm_pageout(void *arg)
{
int npages = 0;
int s;
struct uvm_aiodesc *aio, *nextaio;
UVMHIST_FUNC("uvm_pageout"); UVMHIST_CALLED(pdhist);
UVMHIST_LOG(pdhist,"<starting uvm pagedaemon>", 0, 0, 0, 0);
/*
@ -214,7 +216,82 @@ uvm_pageout(void *arg)
/*
* main loop
*/
while (TRUE) {
for (;;) {
simple_lock(&uvm.pagedaemon_lock);
UVMHIST_LOG(pdhist," <<SLEEPING>>",0,0,0,0);
UVM_UNLOCK_AND_WAIT(&uvm.pagedaemon,
&uvm.pagedaemon_lock, FALSE, "pgdaemon", 0);
uvmexp.pdwoke++;
UVMHIST_LOG(pdhist," <<WOKE UP>>",0,0,0,0);
/* drain pool resources */
pool_drain(0);
/*
* now lock page queues and recompute inactive count
*/
uvm_lock_pageq();
if (npages != uvmexp.npages) { /* check for new pages? */
npages = uvmexp.npages;
uvmpd_tune();
}
uvmexp.inactarg = (uvmexp.active + uvmexp.inactive) / 3;
if (uvmexp.inactarg <= uvmexp.freetarg) {
uvmexp.inactarg = uvmexp.freetarg + 1;
}
UVMHIST_LOG(pdhist," free/ftarg=%d/%d, inact/itarg=%d/%d",
uvmexp.free, uvmexp.freetarg, uvmexp.inactive,
uvmexp.inactarg);
/*
* scan if needed
*/
if (uvmexp.free + uvmexp.paging < uvmexp.freetarg ||
uvmexp.inactive < uvmexp.inactarg ||
uvm_pgcnt_vnode >
(uvmexp.active + uvmexp.inactive + uvmexp.wired +
uvmexp.free) * 13 / 16) {
uvmpd_scan();
}
/*
* if there's any free memory to be had,
* wake up any waiters.
*/
if (uvmexp.free > uvmexp.reserve_kernel ||
uvmexp.paging == 0) {
wakeup(&uvmexp.free);
}
/*
* scan done. unlock page queues (the only lock we are holding)
*/
uvm_unlock_pageq();
}
/*NOTREACHED*/
}
/*
* uvm_aiodone_daemon: main loop for the aiodone daemon.
*/
void
uvm_aiodone_daemon(void *arg)
{
int s, free;
struct buf *bp, *nbp;
UVMHIST_FUNC("uvm_aiodoned"); UVMHIST_CALLED(pdhist);
for (;;) {
/*
* carefully attempt to go to sleep (without losing "wakeups"!).
@ -223,95 +300,58 @@ uvm_pageout(void *arg)
*/
s = splbio();
simple_lock(&uvm.pagedaemon_lock);
/*
* if we've got done aio's, then bypass the sleep
*/
if (uvm.aio_done.tqh_first == NULL) {
UVMHIST_LOG(maphist," <<SLEEPING>>",0,0,0,0);
UVM_UNLOCK_AND_WAIT(&uvm.pagedaemon,
&uvm.pagedaemon_lock, FALSE, "daemon_slp", 0);
uvmexp.pdwoke++;
simple_lock(&uvm.aiodoned_lock);
if (TAILQ_FIRST(&uvm.aio_done) == NULL) {
UVMHIST_LOG(pdhist," <<SLEEPING>>",0,0,0,0);
UVM_UNLOCK_AND_WAIT(&uvm.aiodoned,
&uvm.aiodoned_lock, FALSE, "aiodoned", 0);
UVMHIST_LOG(pdhist," <<WOKE UP>>",0,0,0,0);
/* relock pagedaemon_lock, still at splbio */
simple_lock(&uvm.pagedaemon_lock);
/* relock aiodoned_lock, still at splbio */
simple_lock(&uvm.aiodoned_lock);
}
/*
* check for done aio structures
*/
aio = uvm.aio_done.tqh_first; /* save current list (if any)*/
if (aio) {
TAILQ_INIT(&uvm.aio_done); /* zero global list */
bp = TAILQ_FIRST(&uvm.aio_done);
if (bp) {
TAILQ_INIT(&uvm.aio_done);
}
simple_unlock(&uvm.pagedaemon_lock); /* unlock */
splx(s); /* drop splbio */
simple_unlock(&uvm.aiodoned_lock);
splx(s);
/*
* first clear out any pending aios (to free space in case we
* want to pageout more stuff).
* process each i/o that's done.
*/
for (/*null*/; aio != NULL ; aio = nextaio) {
uvmexp.paging -= aio->npages;
nextaio = aio->aioq.tqe_next;
aio->aiodone(aio);
free = uvmexp.free;
while (bp != NULL) {
if (bp->b_flags & B_PDAEMON) {
uvmexp.paging -= bp->b_bufsize >> PAGE_SHIFT;
}
nbp = TAILQ_NEXT(bp, b_freelist);
(*bp->b_iodone)(bp);
bp = nbp;
}
/* Next, drain pool resources */
pool_drain(0);
/*
* now lock page queues and recompute inactive count
*/
uvm_lock_pageq();
if (npages != uvmexp.npages) { /* check for new pages? */
npages = uvmexp.npages;
uvmpd_tune();
}
uvmexp.inactarg = (uvmexp.active + uvmexp.inactive) / 3;
if (uvmexp.inactarg <= uvmexp.freetarg)
uvmexp.inactarg = uvmexp.freetarg + 1;
UVMHIST_LOG(pdhist," free/ftarg=%d/%d, inact/itarg=%d/%d",
uvmexp.free, uvmexp.freetarg, uvmexp.inactive,
uvmexp.inactarg);
/*
* scan if needed
* [XXX: note we are reading uvm.free without locking]
*/
if (uvmexp.free < uvmexp.freetarg ||
uvmexp.inactive < uvmexp.inactarg)
uvmpd_scan();
/*
* done scan. unlock page queues (the only lock we are holding)
*/
uvm_unlock_pageq();
/*
* done! restart loop.
*/
if (uvmexp.free > uvmexp.reserve_kernel ||
uvmexp.paging == 0)
if (free <= uvmexp.reserve_kernel) {
s = uvm_lock_fpageq();
wakeup(&uvm.pagedaemon);
uvm_unlock_fpageq(s);
} else {
simple_lock(&uvm.pagedaemon_lock);
wakeup(&uvmexp.free);
simple_unlock(&uvm.pagedaemon_lock);
}
}
/*NOTREACHED*/
}
/*
* uvmpd_scan_inactive: the first loop of uvmpd_scan broken out into
* its own function for ease of reading.
* uvmpd_scan_inactive: scan an inactive list for pages to clean or free.
*
* => called with page queues locked
* => we work on meeting our free target by converting inactive pages
@ -334,9 +374,9 @@ uvmpd_scan_inactive(pglst)
int swnpages, swcpages; /* XXX: see below */
int swslot;
struct vm_anon *anon;
boolean_t swap_backed;
boolean_t swap_backed, vnode_only;
vaddr_t start;
int dirtyreacts;
int dirtyreacts, vpgs;
UVMHIST_FUNC("uvmpd_scan_inactive"); UVMHIST_CALLED(pdhist);
/*
@ -349,75 +389,81 @@ uvmpd_scan_inactive(pglst)
/*
* swslot is non-zero if we are building a swap cluster. we want
* to stay in the loop while we have a page to scan or we have
* to stay in the loop while we have a page to scan or we have
* a swap-cluster to build.
*/
swslot = 0;
swnpages = swcpages = 0;
free = 0;
dirtyreacts = 0;
vnode_only = FALSE;
for (p = pglst->tqh_first ; p != NULL || swslot != 0 ; p = nextpg) {
for (p = TAILQ_FIRST(pglst); p != NULL || swslot != 0; p = nextpg) {
/*
* note that p can be NULL iff we have traversed the whole
* list and need to do one final swap-backed clustered pageout.
*/
uobj = NULL;
anon = NULL;
if (p) {
/*
* update our copy of "free" and see if we've met
* our target
*/
s = uvm_lock_fpageq();
free = uvmexp.free;
uvm_unlock_fpageq(s);
/* XXXUBC */
vpgs = uvm_pgcnt_vnode -
(uvmexp.active + uvmexp.inactive +
uvmexp.wired + uvmexp.free) * 13 / 16;
if (free + uvmexp.paging >= uvmexp.freetarg << 2 ||
dirtyreacts == UVMPD_NUMDIRTYREACTS) {
UVMHIST_LOG(pdhist," met free target: "
"exit loop", 0, 0, 0, 0);
retval = TRUE; /* hit the target! */
if (vpgs <= 0) {
UVMHIST_LOG(pdhist," met free target: "
"exit loop", 0, 0, 0, 0);
retval = TRUE;
if (swslot == 0)
/* exit now if no swap-i/o pending */
break;
if (swslot == 0)
/* exit now if no
swap-i/o pending */
break;
/* set p to null to signal final swap i/o */
p = NULL;
/* set p to null to signal final
swap i/o */
p = NULL;
} else {
vnode_only = TRUE;
}
}
}
uobj = NULL; /* be safe and shut gcc up */
anon = NULL; /* be safe and shut gcc up */
if (p) { /* if (we have a new page to consider) */
/*
* we are below target and have a new page to consider.
*/
uvmexp.pdscans++;
nextpg = p->pageq.tqe_next;
/*
* move referenced pages back to active queue and
* skip to next page (unlikely to happen since
* inactive pages shouldn't have any valid mappings
* and we cleared reference before deactivating).
*/
if (pmap_is_referenced(p)) {
uvm_pageactivate(p);
uvmexp.pdreact++;
continue;
}
uvmexp.pdscans++;
nextpg = TAILQ_NEXT(p, pageq);
/*
* first we attempt to lock the object that this page
* belongs to. if our attempt fails we skip on to
* the next page (no harm done). it is important to
* "try" locking the object as we are locking in the
* wrong order (pageq -> object) and we don't want to
* get deadlocked.
* deadlock.
*
* the only time we exepct to see an ownerless page
* the only time we expect to see an ownerless page
* (i.e. a page with no uobject and !PQ_ANON) is if an
* anon has loaned a page from a uvm_object and the
* uvm_object has dropped the ownership. in that
@ -427,17 +473,12 @@ uvmpd_scan_inactive(pglst)
/* is page part of an anon or ownerless ? */
if ((p->pqflags & PQ_ANON) || p->uobject == NULL) {
if (vnode_only) {
uvm_pageactivate(p);
continue;
}
anon = p->uanon;
#ifdef DIAGNOSTIC
/* to be on inactive q, page must be part
* of _something_ */
if (anon == NULL)
panic("pagedaemon: page with no anon "
"or object detected - loop 1");
#endif
KASSERT(anon != NULL);
if (!simple_lock_try(&anon->an_lock))
/* lock failed, skip this page */
continue;
@ -446,41 +487,38 @@ uvmpd_scan_inactive(pglst)
* if the page is ownerless, claim it in the
* name of "anon"!
*/
if ((p->pqflags & PQ_ANON) == 0) {
#ifdef DIAGNOSTIC
if (p->loan_count < 1)
panic("pagedaemon: non-loaned "
"ownerless page detected -"
" loop 1");
#endif
p->loan_count--;
p->pqflags |= PQ_ANON; /* anon now owns it */
}
if ((p->pqflags & PQ_ANON) == 0) {
KASSERT(p->loan_count > 0);
p->loan_count--;
p->pqflags |= PQ_ANON;
/* anon now owns it */
}
if (p->flags & PG_BUSY) {
simple_unlock(&anon->an_lock);
uvmexp.pdbusy++;
/* someone else owns page, skip it */
continue;
}
uvmexp.pdanscan++;
} else {
uobj = p->uobject;
KASSERT(uobj != NULL);
if (vnode_only &&
uobj->pgops != &uvm_vnodeops) {
uvm_pageactivate(p);
continue;
}
if (!simple_lock_try(&uobj->vmobjlock))
/* lock failed, skip this page */
continue;
continue;
if (p->flags & PG_BUSY) {
simple_unlock(&uobj->vmobjlock);
uvmexp.pdbusy++;
/* someone else owns page, skip it */
continue;
continue;
}
uvmexp.pdobscan++;
}
@ -498,21 +536,18 @@ uvmpd_scan_inactive(pglst)
simple_unlock(&uvm.swap_data_lock);
}
/* zap all mappings with pmap_page_protect... */
pmap_page_protect(p, VM_PROT_NONE);
uvm_pagefree(p);
uvmexp.pdfreed++;
if (anon) {
#ifdef DIAGNOSTIC
/*
* an anonymous page can only be clean
* if it has valid backing store.
* if it has backing store assigned.
*/
if (anon->an_swslot == 0)
panic("pagedaemon: clean anon "
"page without backing store?");
#endif
KASSERT(anon->an_swslot != 0);
/* remove from object */
anon->u.an_page = NULL;
simple_unlock(&anon->an_lock);
@ -528,6 +563,7 @@ uvmpd_scan_inactive(pglst)
* this page is dirty, skip it if we'll have met our
* free target when all the current pageouts complete.
*/
if (free + uvmexp.paging > uvmexp.freetarg << 2) {
if (anon) {
simple_unlock(&anon->an_lock);
@ -543,11 +579,8 @@ uvmpd_scan_inactive(pglst)
* reactivate it so that we eventually cycle
* all pages thru the inactive queue.
*/
#ifdef DIAGNOSTIC
if (uvmexp.swpgonly > uvmexp.swpages) {
panic("uvmexp.swpgonly botch");
}
#endif
KASSERT(uvmexp.swpgonly <= uvmexp.swpages);
if ((p->pqflags & PQ_SWAPBACKED) &&
uvmexp.swpgonly == uvmexp.swpages) {
dirtyreacts++;
@ -565,11 +598,8 @@ uvmpd_scan_inactive(pglst)
* is full, free any swap allocated to the page
* so that other pages can be paged out.
*/
#ifdef DIAGNOSTIC
if (uvmexp.swpginuse > uvmexp.swpages) {
panic("uvmexp.swpginuse botch");
}
#endif
KASSERT(uvmexp.swpginuse <= uvmexp.swpages);
if ((p->pqflags & PQ_SWAPBACKED) &&
uvmexp.swpginuse == uvmexp.swpages) {
@ -588,26 +618,25 @@ uvmpd_scan_inactive(pglst)
* the page we are looking at is dirty. we must
* clean it before it can be freed. to do this we
* first mark the page busy so that no one else will
* touch the page. we write protect all the mappings
* of the page so that no one touches it while it is
* in I/O.
* touch the page.
*/
swap_backed = ((p->pqflags & PQ_SWAPBACKED) != 0);
p->flags |= PG_BUSY; /* now we own it */
UVM_PAGE_OWN(p, "scan_inactive");
pmap_page_protect(p, VM_PROT_READ);
uvmexp.pgswapout++;
/*
* for swap-backed pages we need to (re)allocate
* swap space.
*/
if (swap_backed) {
/*
* free old swap slot (if any)
*/
if (anon) {
if (anon->an_swslot) {
uvm_swap_free(anon->an_swslot,
@ -622,13 +651,11 @@ uvmpd_scan_inactive(pglst)
/*
* start new cluster (if necessary)
*/
if (swslot == 0) {
/* want this much */
swnpages = MAXBSIZE >> PAGE_SHIFT;
if (swslot == 0) {
swnpages = MAXBSIZE >> PAGE_SHIFT;
swslot = uvm_swap_alloc(&swnpages,
TRUE);
if (swslot == 0) {
/* no swap? give up! */
p->flags &= ~PG_BUSY;
@ -647,6 +674,7 @@ uvmpd_scan_inactive(pglst)
/*
* add block to cluster
*/
swpps[swcpages] = p;
if (anon)
anon->an_swslot = swslot + swcpages;
@ -655,11 +683,7 @@ uvmpd_scan_inactive(pglst)
p->offset >> PAGE_SHIFT,
swslot + swcpages);
swcpages++;
/* done (swap-backed) */
}
/* end: if (p) ["if we have new page to consider"] */
} else {
/* if p == NULL we must be doing a last swap i/o */
@ -667,16 +691,16 @@ uvmpd_scan_inactive(pglst)
}
/*
* now consider doing the pageout.
* now consider doing the pageout.
*
* for swap-backed pages, we do the pageout if we have either
* filled the cluster (in which case (swnpages == swcpages) or
* for swap-backed pages, we do the pageout if we have either
* filled the cluster (in which case (swnpages == swcpages) or
* run out of pages (p == NULL).
*
* for object pages, we always do the pageout.
*/
if (swap_backed) {
if (swap_backed) {
if (p) { /* if we just added a page to cluster */
if (anon)
simple_unlock(&anon->an_lock);
@ -699,21 +723,18 @@ uvmpd_scan_inactive(pglst)
if (swcpages < swnpages) {
uvm_swap_free(swslot + swcpages,
(swnpages - swcpages));
}
}
} else {
/* normal object pageout */
ppsp = pps;
npages = sizeof(pps) / sizeof(struct vm_page *);
/* not looked at because PGO_ALLPAGES is set */
start = 0;
}
/*
* now do the pageout.
*
*
* for swap_backed pages we have already built the cluster.
* for !swap_backed pages, uvm_pager_put will call the object's
* "make put cluster" function to build a cluster on our behalf.
@ -734,7 +755,7 @@ uvmpd_scan_inactive(pglst)
/* locked: uobj (if !swap_backed), page queues */
uvmexp.pdpageouts++;
result = uvm_pager_put((swap_backed) ? NULL : uobj, p,
result = uvm_pager_put(swap_backed ? NULL : uobj, p,
&ppsp, &npages, PGO_ALLPAGES|PGO_PDFREECLUST, start, 0);
/* locked: uobj (if !swap_backed && result != PEND) */
/* unlocked: pageqs, object (if swap_backed ||result == PEND) */
@ -762,21 +783,27 @@ uvmpd_scan_inactive(pglst)
if (result == VM_PAGER_PEND) {
uvmexp.paging += npages;
uvm_lock_pageq(); /* relock page queues */
uvm_lock_pageq();
uvmexp.pdpending++;
if (p) {
if (p->pqflags & PQ_INACTIVE)
/* reload! */
nextpg = p->pageq.tqe_next;
nextpg = TAILQ_NEXT(p, pageq);
else
/* reload! */
nextpg = pglst->tqh_first;
} else {
nextpg = NULL; /* done list */
nextpg = TAILQ_FIRST(pglst);
} else {
nextpg = NULL;
}
continue;
}
if (result == VM_PAGER_ERROR &&
curproc == uvm.pagedaemon_proc) {
uvm_lock_pageq();
nextpg = TAILQ_NEXT(p, pageq);
uvm_pageactivate(p);
continue;
}
/*
* clean up "p" if we have one
*/
@ -812,12 +839,6 @@ uvmpd_scan_inactive(pglst)
simple_lock(&uobj->vmobjlock);
}
#ifdef DIAGNOSTIC
if (result == VM_PAGER_UNLOCK)
panic("pagedaemon: pageout returned "
"invalid 'unlock' code");
#endif
/* handle PG_WANTED now */
if (p->flags & PG_WANTED)
/* still holding object lock */
@ -837,24 +858,19 @@ uvmpd_scan_inactive(pglst)
pmap_page_protect(p, VM_PROT_NONE);
anon = NULL;
uvm_lock_pageq();
nextpg = p->pageq.tqe_next;
nextpg = TAILQ_NEXT(p, pageq);
/* free released page */
uvm_pagefree(p);
} else {
#ifdef DIAGNOSTIC
if (uobj->pgops->pgo_releasepg == NULL)
panic("pagedaemon: no "
"pgo_releasepg function");
#endif
/*
/*
* pgo_releasepg nukes the page and
* gets "nextpg" for us. it returns
* with the page queues locked (when
* given nextpg ptr).
*/
if (!uobj->pgops->pgo_releasepg(p,
&nextpg))
/* uobj died after release */
@ -864,35 +880,27 @@ uvmpd_scan_inactive(pglst)
* lock page queues here so that they're
* always locked at the end of the loop.
*/
uvm_lock_pageq();
}
} else { /* page was not released during I/O */
uvm_lock_pageq();
nextpg = p->pageq.tqe_next;
nextpg = TAILQ_NEXT(p, pageq);
if (result != VM_PAGER_OK) {
/* pageout was a failure... */
if (result != VM_PAGER_AGAIN)
uvm_pageactivate(p);
pmap_clear_reference(p);
/* XXXCDC: if (swap_backed) FREE p's
* swap block? */
} else {
/* pageout was a success... */
pmap_clear_reference(p);
pmap_clear_modify(p);
p->flags |= PG_CLEAN;
/* XXX: could free page here, but old
* pagedaemon does not */
}
}
/*
* drop object lock (if there is an object left). do
* a safety check of nextpg to make sure it is on the
@ -906,26 +914,27 @@ uvmpd_scan_inactive(pglst)
else if (uobj)
simple_unlock(&uobj->vmobjlock);
} /* if (p) */ else {
} else {
/*
* if p is null in this loop, make sure it stays null
* in the next loop.
*/
/* if p is null in this loop, make sure it stays null
* in next loop */
nextpg = NULL;
/*
* lock page queues here just so they're always locked
* at the end of the loop.
*/
uvm_lock_pageq();
}
if (nextpg && (nextpg->pqflags & PQ_INACTIVE) == 0) {
printf("pagedaemon: invalid nextpg! reverting to "
"queue head\n");
nextpg = pglst->tqh_first; /* reload! */
nextpg = TAILQ_FIRST(pglst); /* reload! */
}
} /* end of "inactive" 'for' loop */
}
return (retval);
}
@ -945,10 +954,8 @@ uvmpd_scan()
UVMHIST_FUNC("uvmpd_scan"); UVMHIST_CALLED(pdhist);
uvmexp.pdrevs++; /* counter */
uobj = NULL;
#ifdef __GNUC__
uobj = NULL; /* XXX gcc */
#endif
/*
* get current "free" page count
*/
@ -962,13 +969,11 @@ uvmpd_scan()
* we need to unlock the page queues for this.
*/
if (free < uvmexp.freetarg) {
uvmexp.pdswout++;
UVMHIST_LOG(pdhist," free %d < target %d: swapout", free,
uvmexp.freetarg, 0, 0);
uvm_unlock_pageq();
uvm_swapout_threads();
pmap_update(); /* update so we can scan inactive q */
uvm_lock_pageq();
}
@ -984,8 +989,8 @@ uvmpd_scan()
UVMHIST_LOG(pdhist, " starting 'free' loop",0,0,0,0);
/*
* do loop #1! alternate starting queue between swap and object based
* on the low bit of uvmexp.pdrevs (which we bump by one each call).
* alternate starting queue between swap and object based on the
* low bit of uvmexp.pdrevs (which we bump by one each call).
*/
got_it = FALSE;
@ -1009,6 +1014,7 @@ uvmpd_scan()
* detect if we're not going to be able to page anything out
* until we free some swap resources from active pages.
*/
swap_shortage = 0;
if (uvmexp.free < uvmexp.freetarg &&
uvmexp.swpginuse == uvmexp.swpages &&
@ -1016,13 +1022,13 @@ uvmpd_scan()
pages_freed == 0) {
swap_shortage = uvmexp.freetarg - uvmexp.free;
}
UVMHIST_LOG(pdhist, " loop 2: inactive_shortage=%d swap_shortage=%d",
inactive_shortage, swap_shortage,0,0);
for (p = TAILQ_FIRST(&uvm.page_active);
for (p = TAILQ_FIRST(&uvm.page_active);
p != NULL && (inactive_shortage > 0 || swap_shortage > 0);
p = nextpg) {
nextpg = p->pageq.tqe_next;
nextpg = TAILQ_NEXT(p, pageq);
if (p->flags & PG_BUSY)
continue; /* quick check before trying to lock */
@ -1031,22 +1037,13 @@ uvmpd_scan()
*/
/* is page anon owned or ownerless? */
if ((p->pqflags & PQ_ANON) || p->uobject == NULL) {
#ifdef DIAGNOSTIC
if (p->uanon == NULL)
panic("pagedaemon: page with no anon or "
"object detected - loop 2");
#endif
KASSERT(p->uanon != NULL);
if (!simple_lock_try(&p->uanon->an_lock))
continue;
/* take over the page? */
if ((p->pqflags & PQ_ANON) == 0) {
#ifdef DIAGNOSTIC
if (p->loan_count < 1)
panic("pagedaemon: non-loaned "
"ownerless page detected - loop 2");
#endif
KASSERT(p->loan_count > 0);
p->loan_count--;
p->pqflags |= PQ_ANON;
}
@ -1054,9 +1051,11 @@ uvmpd_scan()
if (!simple_lock_try(&p->uobject->vmobjlock))
continue;
}
/*
* skip this page if it's busy.
*/
if ((p->flags & PG_BUSY) != 0) {
if (p->pqflags & PQ_ANON)
simple_unlock(&p->uanon->an_lock);
@ -1064,11 +1063,12 @@ uvmpd_scan()
simple_unlock(&p->uobject->vmobjlock);
continue;
}
/*
* if there's a shortage of swap, free any swap allocated
* to this page so that other pages can be paged out.
*/
if (swap_shortage > 0) {
if ((p->pqflags & PQ_ANON) && p->uanon->an_swslot) {
uvm_swap_free(p->uanon->an_swslot, 1);
@ -1086,11 +1086,12 @@ uvmpd_scan()
}
}
}
/*
* deactivate this page if there's a shortage of
* inactive pages.
*/
if (inactive_shortage > 0) {
pmap_page_protect(p, VM_PROT_NONE);
/* no need to check wire_count as pg is "active" */
@ -1098,7 +1099,6 @@ uvmpd_scan()
uvmexp.pddeact++;
inactive_shortage--;
}
if (p->pqflags & PQ_ANON)
simple_unlock(&p->uanon->an_lock);
else

267
sys/uvm/uvm_swap.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: uvm_swap.c,v 1.40 2000/11/17 11:39:39 mrg Exp $ */
/* $NetBSD: uvm_swap.c,v 1.41 2000/11/27 08:40:05 chs Exp $ */
/*
* Copyright (c) 1995, 1996, 1997 Matthew R. Green
@ -34,6 +34,7 @@
#include "fs_nfs.h"
#include "opt_uvmhist.h"
#include "opt_compat_netbsd.h"
#include "opt_ddb.h"
#include <sys/param.h>
#include <sys/systm.h>
@ -77,11 +78,6 @@
* by the "swap_priority" global var. each "swappri" contains a
* CIRCLEQ of "swapdev" structures at that priority.
*
* the system maintains a fixed pool of "swapbuf" structures for use
* at swap i/o time. a swapbuf includes a "buf" structure and an
* "aiodone" [we want to avoid malloc()'ing anything at swapout time
* since memory may be low].
*
* locking:
* - swap_syscall_lock (sleep lock): this lock serializes the swapctl
* system call and prevents the swap priority list from changing
@ -89,8 +85,6 @@
* - uvm.swap_data_lock (simple_lock): this lock protects all swap data
* structures including the priority list, the swapdev structures,
* and the swapmap extent.
* - swap_buf_lock (simple_lock): this lock protects the free swapbuf
* pool.
*
* each swap device has the following info:
* - swap device in use (could be disabled, preventing future use)
@ -157,15 +151,6 @@ struct swappri {
LIST_ENTRY(swappri) spi_swappri; /* global list of pri's */
};
/*
* swapbuf, swapbuffer plus async i/o info
*/
struct swapbuf {
struct buf sw_buf; /* a buffer structure */
struct uvm_aiodesc sw_aio; /* aiodesc structure, used if ASYNC */
SIMPLEQ_ENTRY(swapbuf) sw_sq; /* free list pointer */
};
/*
* The following two structures are used to keep track of data transfers
* on swap devices associated with regular files.
@ -222,8 +207,6 @@ cdev_decl(sw);
* local variables
*/
static struct extent *swapmap; /* controls the mapping of /dev/drum */
SIMPLEQ_HEAD(swapbufhead, swapbuf);
struct pool *swapbuf_pool;
/* list of all active swap devices [by priority] */
LIST_HEAD(swap_priority, swappri);
@ -250,8 +233,6 @@ static void sw_reg_strategy __P((struct swapdev *, struct buf *, int));
static void sw_reg_iodone __P((struct buf *));
static void sw_reg_start __P((struct swapdev *));
static void uvm_swap_aiodone __P((struct uvm_aiodesc *));
static void uvm_swap_bufdone __P((struct buf *));
static int uvm_swap_io __P((struct vm_page **, int, int, int));
/*
@ -292,18 +273,9 @@ uvm_swap_init()
panic("uvm_swap_init: extent_create failed");
/*
* allocate our private pool of "swapbuf" structures (includes
* a "buf" structure). ["nswbuf" comes from param.c and can
* be adjusted by MD code before we get here].
* allocate pools for structures used for swapping to files.
*/
swapbuf_pool =
pool_create(sizeof(struct swapbuf), 0, 0, 0, "swp buf", 0,
NULL, NULL, 0);
if (swapbuf_pool == NULL)
panic("swapinit: pool_create failed");
/* XXX - set a maximum on swapbuf_pool? */
vndxfer_pool =
pool_create(sizeof(struct vndxfer), 0, 0, 0, "swp vnx", 0,
NULL, NULL, 0);
@ -1120,7 +1092,7 @@ swstrategy(bp)
* be yanked out from under us because we are holding resources
* in it (i.e. the blocks we are doing I/O on).
*/
pageno = dbtob(bp->b_blkno) >> PAGE_SHIFT;
pageno = dbtob((int64_t)bp->b_blkno) >> PAGE_SHIFT;
simple_lock(&uvm.swap_data_lock);
sdp = swapdrum_getsdp(pageno);
simple_unlock(&uvm.swap_data_lock);
@ -1139,7 +1111,7 @@ swstrategy(bp)
pageno -= sdp->swd_drumoffset; /* page # on swapdev */
bn = btodb(pageno << PAGE_SHIFT); /* convert to diskblock */
UVMHIST_LOG(pdhist, " %s: mapoff=%x bn=%x bcount=%ld\n",
UVMHIST_LOG(pdhist, " %s: mapoff=%x bn=%x bcount=%ld",
((bp->b_flags & B_READ) == 0) ? "write" : "read",
sdp->swd_drumoffset, bn, bp->b_bcount);
@ -1174,14 +1146,14 @@ swstrategy(bp)
vp->v_numoutput++; /* put it on swapdev */
}
/*
/*
* dissassocate buffer with /dev/drum vnode
* [could be null if buf was from physio]
*/
if (bp->b_vp != NULLVP)
if (bp->b_vp != NULL)
brelvp(bp);
/*
/*
* finally plug in swapdev vnode and start I/O
*/
bp->b_vp = vp;
@ -1279,18 +1251,15 @@ sw_reg_strategy(sdp, bp, bn)
/*
* compute the size ("sz") of this transfer (in bytes).
* XXXCDC: ignores read-ahead for non-zero offset
*/
if ((off = (byteoff % sdp->swd_bsize)) != 0)
sz = sdp->swd_bsize - off;
else
sz = (1 + nra) * sdp->swd_bsize;
if (resid < sz)
off = byteoff % sdp->swd_bsize;
sz = (1 + nra) * sdp->swd_bsize - off;
if (sz > resid)
sz = resid;
UVMHIST_LOG(pdhist, "sw_reg_strategy: vp %p/%p offset 0x%x/0x%x",
sdp->swd_vp, vp, byteoff, nbn);
UVMHIST_LOG(pdhist, "sw_reg_strategy: "
"vp %p/%p offset 0x%x/0x%x",
sdp->swd_vp, vp, byteoff, nbn);
/*
* now get a buf structure. note that the vb_buf is
@ -1303,42 +1272,13 @@ sw_reg_strategy(sdp, bp, bn)
nbp->vb_buf.b_bufsize = sz;
nbp->vb_buf.b_error = 0;
nbp->vb_buf.b_data = addr;
nbp->vb_buf.b_lblkno = 0;
nbp->vb_buf.b_blkno = nbn + btodb(off);
nbp->vb_buf.b_rawblkno = nbp->vb_buf.b_blkno;
nbp->vb_buf.b_proc = bp->b_proc;
nbp->vb_buf.b_iodone = sw_reg_iodone;
nbp->vb_buf.b_vp = NULLVP;
nbp->vb_buf.b_vnbufs.le_next = NOLIST;
nbp->vb_buf.b_rcred = sdp->swd_cred;
nbp->vb_buf.b_wcred = sdp->swd_cred;
nbp->vb_buf.b_vp = NULL;
LIST_INIT(&nbp->vb_buf.b_dep);
/*
* set b_dirtyoff/end and b_validoff/end. this is
* required by the NFS client code (otherwise it will
* just discard our I/O request).
*/
if (bp->b_dirtyend == 0) {
nbp->vb_buf.b_dirtyoff = 0;
nbp->vb_buf.b_dirtyend = sz;
} else {
nbp->vb_buf.b_dirtyoff =
max(0, bp->b_dirtyoff - (bp->b_bcount-resid));
nbp->vb_buf.b_dirtyend =
min(sz,
max(0, bp->b_dirtyend - (bp->b_bcount-resid)));
}
if (bp->b_validend == 0) {
nbp->vb_buf.b_validoff = 0;
nbp->vb_buf.b_validend = sz;
} else {
nbp->vb_buf.b_validoff =
max(0, bp->b_validoff - (bp->b_bcount-resid));
nbp->vb_buf.b_validend =
min(sz,
max(0, bp->b_validend - (bp->b_bcount-resid)));
}
nbp->vb_xfer = vnx; /* patch it back in to vnx */
/*
@ -1352,7 +1292,7 @@ sw_reg_strategy(sdp, bp, bn)
vnx->vx_pending++;
/* assoc new buffer with underlying vnode */
bgetvp(vp, &nbp->vb_buf);
bgetvp(vp, &nbp->vb_buf);
/* sort it in and start I/O if we are not over our limit */
disksort_blkno(&sdp->swd_tab, &nbp->vb_buf);
@ -1411,6 +1351,7 @@ sw_reg_start(sdp)
bp, bp->b_vp, bp->b_blkno, bp->b_bcount);
if ((bp->b_flags & B_READ) == 0)
bp->b_vp->v_numoutput++;
VOP_STRATEGY(bp);
}
sdp->swd_flags &= ~SWF_BUSY;
@ -1455,11 +1396,9 @@ sw_reg_iodone(bp)
}
/*
* disassociate this buffer from the vnode (if any).
* disassociate this buffer from the vnode.
*/
if (vbp->vb_buf.b_vp != NULLVP) {
brelvp(&vbp->vb_buf);
}
brelvp(&vbp->vb_buf);
/*
* kill vbp structure
@ -1598,8 +1537,9 @@ uvm_swap_markbad(startslot, nslots)
* we assume here that the range of slots will all be within
* one swap device.
*/
sdp->swd_npgbad += nslots;
sdp->swd_npgbad += nslots;
UVMHIST_LOG(pdhist, "now %d bad", sdp->swd_npgbad, 0,0,0);
simple_unlock(&uvm.swap_data_lock);
}
@ -1735,15 +1675,18 @@ uvm_swap_io(pps, startslot, npages, flags)
int startslot, npages, flags;
{
daddr_t startblk;
struct swapbuf *sbp;
struct buf *bp;
vaddr_t kva;
int result, s, mapinflags, pflag;
boolean_t write, async;
UVMHIST_FUNC("uvm_swap_io"); UVMHIST_CALLED(pdhist);
UVMHIST_LOG(pdhist, "<- called, startslot=%d, npages=%d, flags=%d",
startslot, npages, flags, 0);
write = (flags & B_READ) == 0;
async = (flags & B_ASYNC) != 0;
/*
* convert starting drum slot to block number
*/
@ -1751,43 +1694,37 @@ uvm_swap_io(pps, startslot, npages, flags)
/*
* first, map the pages into the kernel (XXX: currently required
* by buffer system). note that we don't let pagermapin alloc
* an aiodesc structure because we don't want to chance a malloc.
* we've got our own pool of aiodesc structures (in swapbuf).
* by buffer system).
*/
mapinflags = (flags & B_READ) ? UVMPAGER_MAPIN_READ :
UVMPAGER_MAPIN_WRITE;
if ((flags & B_ASYNC) == 0)
mapinflags = !write ? UVMPAGER_MAPIN_READ : UVMPAGER_MAPIN_WRITE;
if (!async)
mapinflags |= UVMPAGER_MAPIN_WAITOK;
kva = uvm_pagermapin(pps, npages, NULL, mapinflags);
kva = uvm_pagermapin(pps, npages, mapinflags);
if (kva == 0)
return (VM_PAGER_AGAIN);
/*
* now allocate a swap buffer off of freesbufs
* now allocate a buf for the i/o.
* [make sure we don't put the pagedaemon to sleep...]
*/
s = splbio();
pflag = ((flags & B_ASYNC) != 0 || curproc == uvm.pagedaemon_proc)
? 0
: PR_WAITOK;
sbp = pool_get(swapbuf_pool, pflag);
splx(s); /* drop splbio */
pflag = (async || curproc == uvm.pagedaemon_proc) ? 0 : PR_WAITOK;
bp = pool_get(&bufpool, pflag);
splx(s);
/*
* if we failed to get a swapbuf, return "try again"
* if we failed to get a buf, return "try again"
*/
if (sbp == NULL)
if (bp == NULL)
return (VM_PAGER_AGAIN);
/*
* fill in the bp/sbp. we currently route our i/o through
* /dev/drum's vnode [swapdev_vp].
*/
bp = &sbp->sw_buf;
bp->b_flags = B_BUSY | B_NOCACHE | (flags & (B_READ|B_ASYNC));
bp->b_proc = &proc0; /* XXX */
bp->b_rcred = bp->b_wcred = proc0.p_ucred;
bp->b_vnbufs.le_next = NOLIST;
bp->b_data = (caddr_t)kva;
bp->b_blkno = startblk;
@ -1799,49 +1736,43 @@ uvm_swap_io(pps, startslot, npages, flags)
/* XXXMRG: probably -- this is obviously something inherited... */
if (swapdev_vp->v_type == VBLK)
bp->b_dev = swapdev_vp->v_rdev;
bp->b_bcount = npages << PAGE_SHIFT;
bp->b_bufsize = bp->b_bcount = npages << PAGE_SHIFT;
LIST_INIT(&bp->b_dep);
/*
* for pageouts we must set "dirtyoff" [NFS client code needs it].
* and we bump v_numoutput (counter of number of active outputs).
* bump v_numoutput (counter of number of active outputs).
*/
if ((bp->b_flags & B_READ) == 0) {
bp->b_dirtyoff = 0;
bp->b_dirtyend = npages << PAGE_SHIFT;
if (write) {
s = splbio();
swapdev_vp->v_numoutput++;
splx(s);
}
/*
* for async ops we must set up the aiodesc and setup the callback
* XXX: we expect no async-reads, but we don't prevent it here.
* for async ops we must set up the iodone handler.
*/
if (flags & B_ASYNC) {
sbp->sw_aio.aiodone = uvm_swap_aiodone;
sbp->sw_aio.kva = kva;
sbp->sw_aio.npages = npages;
sbp->sw_aio.pd_ptr = sbp; /* backpointer */
bp->b_flags |= B_CALL; /* set callback */
bp->b_iodone = uvm_swap_bufdone;/* "buf" iodone function */
if (async) {
/* XXXUBC pagedaemon */
bp->b_flags |= B_CALL | (curproc == uvm.pagedaemon_proc ?
B_PDAEMON : 0);
bp->b_iodone = uvm_aio_biodone;
UVMHIST_LOG(pdhist, "doing async!", 0, 0, 0, 0);
}
UVMHIST_LOG(pdhist,
"about to start io: data = 0x%p blkno = 0x%x, bcount = %ld",
"about to start io: data = %p blkno = 0x%x, bcount = %ld",
bp->b_data, bp->b_blkno, bp->b_bcount, 0);
/*
* now we start the I/O, and if async, return.
*/
VOP_STRATEGY(bp);
if (flags & B_ASYNC)
if (async)
return (VM_PAGER_PEND);
/*
* must be sync i/o. wait for it to finish
*/
bp->b_error = biowait(bp);
(void) biowait(bp);
result = (bp->b_flags & B_ERROR) ? VM_PAGER_ERROR : VM_PAGER_OK;
/*
@ -1850,13 +1781,14 @@ uvm_swap_io(pps, startslot, npages, flags)
uvm_pagermapout(kva, npages);
/*
* now dispose of the swap buffer
* now dispose of the buf
*/
s = splbio();
if (bp->b_vp)
brelvp(bp);
pool_put(swapbuf_pool, sbp);
if (write)
vwakeup(bp);
pool_put(&bufpool, bp);
splx(s);
/*
@ -1865,96 +1797,3 @@ uvm_swap_io(pps, startslot, npages, flags)
UVMHIST_LOG(pdhist, "<- done (sync) result=%d", result, 0, 0, 0);
return (result);
}
/*
* uvm_swap_bufdone: called from the buffer system when the i/o is done
*/
static void
uvm_swap_bufdone(bp)
struct buf *bp;
{
struct swapbuf *sbp = (struct swapbuf *) bp;
int s = splbio();
UVMHIST_FUNC("uvm_swap_bufdone"); UVMHIST_CALLED(pdhist);
UVMHIST_LOG(pdhist, "cleaning buf %p", buf, 0, 0, 0);
#ifdef DIAGNOSTIC
/*
* sanity check: swapbufs are private, so they shouldn't be wanted
*/
if (bp->b_flags & B_WANTED)
panic("uvm_swap_bufdone: private buf wanted");
#endif
/*
* drop the buffer's reference to the vnode.
*/
if (bp->b_vp)
brelvp(bp);
/*
* now put the aio on the uvm.aio_done list and wake the
* pagedaemon (which will finish up our job in its context).
*/
simple_lock(&uvm.pagedaemon_lock); /* locks uvm.aio_done */
TAILQ_INSERT_TAIL(&uvm.aio_done, &sbp->sw_aio, aioq);
simple_unlock(&uvm.pagedaemon_lock);
wakeup(&uvm.pagedaemon);
splx(s);
}
/*
* uvm_swap_aiodone: aiodone function for anonymous memory
*
* => this is called in the context of the pagedaemon (but with the
* page queues unlocked!)
* => our "aio" structure must be part of a "swapbuf"
*/
static void
uvm_swap_aiodone(aio)
struct uvm_aiodesc *aio;
{
struct swapbuf *sbp = aio->pd_ptr;
struct vm_page *pps[MAXBSIZE >> PAGE_SHIFT];
int lcv, s;
vaddr_t addr;
UVMHIST_FUNC("uvm_swap_aiodone"); UVMHIST_CALLED(pdhist);
UVMHIST_LOG(pdhist, "done with aio %p", aio, 0, 0, 0);
#ifdef DIAGNOSTIC
/*
* sanity check
*/
if (aio->npages > (MAXBSIZE >> PAGE_SHIFT))
panic("uvm_swap_aiodone: aio too big!");
#endif
/*
* first, we have to recover the page pointers (pps) by poking in the
* kernel pmap (XXX: should be saved in the buf structure).
*/
for (addr = aio->kva, lcv = 0 ; lcv < aio->npages ;
addr += PAGE_SIZE, lcv++) {
pps[lcv] = uvm_pageratop(addr);
}
/*
* now we can dispose of the kernel mappings of the buffer
*/
uvm_pagermapout(aio->kva, aio->npages);
/*
* now we can dispose of the pages by using the dropcluster function
* [note that we have no "page of interest" so we pass in null]
*/
uvm_pager_dropcluster(NULL, NULL, pps, &aio->npages,
PGO_PDFREECLUST);
/*
* finally, we can dispose of the swapbuf
*/
s = splbio();
pool_put(swapbuf_pool, sbp);
splx(s);
}

1546
sys/uvm/uvm_vnode.c
View File

File diff suppressed because it is too large Load Diff

52
sys/uvm/uvm_vnode.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: uvm_vnode.h,v 1.9 2000/03/26 20:54:48 kleink Exp $ */
/* $NetBSD: uvm_vnode.h,v 1.10 2000/11/27 08:40:06 chs Exp $ */
/*
*
@ -54,56 +54,6 @@ struct uvm_vnode {
int u_flags; /* flags */
int u_nio; /* number of running I/O requests */
voff_t u_size; /* size of object */
/* the following entry is locked by uvn_wl_lock */
LIST_ENTRY(uvm_vnode) u_wlist; /* list of writeable vnode objects */
/* the following entry is locked by uvn_sync_lock */
SIMPLEQ_ENTRY(uvm_vnode) u_syncq; /* vnode objects due for a "sync" */
};
/*
* u_flags values
*/
#define UVM_VNODE_VALID 0x001 /* we are attached to the vnode */
#define UVM_VNODE_CANPERSIST 0x002 /* we can persist after ref == 0 */
#define UVM_VNODE_ALOCK 0x004 /* uvn_attach is locked out */
#define UVM_VNODE_DYING 0x008 /* final detach/terminate in
progress */
#define UVM_VNODE_RELKILL 0x010 /* uvn should be killed by releasepg
when final i/o is done */
#define UVM_VNODE_WANTED 0x020 /* someone is waiting for alock,
dying, or relkill to clear */
#define UVM_VNODE_VNISLOCKED 0x040 /* underlying vnode struct is locked
(valid when DYING is true) */
#define UVM_VNODE_IOSYNC 0x080 /* I/O sync in progress ... setter
sleeps on &uvn->u_nio */
#define UVM_VNODE_IOSYNCWANTED 0x100 /* a process is waiting for the
i/o sync to clear so it can do
i/o */
#define UVM_VNODE_WRITEABLE 0x200 /* uvn has pages that are writeable */
/*
* UVM_VNODE_BLOCKED: any condition that should new processes from
* touching the vnode [set WANTED and sleep to wait for it to clear]
*/
#define UVM_VNODE_BLOCKED (UVM_VNODE_ALOCK|UVM_VNODE_DYING|UVM_VNODE_RELKILL)
#ifdef _KERNEL
/*
* prototypes
*/
#if 0
/*
* moved uvn_attach to uvm_extern.h because uvm_vnode.h is needed to
* include sys/vnode.h, and files that include sys/vnode.h don't know
* what a vm_prot_t is.
*/
struct uvm_object *uvn_attach __P((void *, vm_prot_t));
#endif
#endif /* _KERNEL */
#endif /* _UVM_UVM_VNODE_H_ */