freebsd-skq/sys/coda/coda_subr.c

574 lines
14 KiB
C
Raw Normal View History

/*-
1998-09-02 19:09:53 +00:00
* Coda: an Experimental Distributed File System
* Release 3.1
*
* Copyright (c) 1987-1998 Carnegie Mellon University
* All Rights Reserved
*
* Permission to use, copy, modify and distribute this software and its
* documentation is hereby granted, provided that both the copyright
* notice and this permission notice appear in all copies of the
* software, derivative works or modified versions, and any portions
* thereof, and that both notices appear in supporting documentation, and
* that credit is given to Carnegie Mellon University in all documents
* and publicity pertaining to direct or indirect use of this code or its
* derivatives.
*
* CODA IS AN EXPERIMENTAL SOFTWARE SYSTEM AND IS KNOWN TO HAVE BUGS,
* SOME OF WHICH MAY HAVE SERIOUS CONSEQUENCES. CARNEGIE MELLON ALLOWS
* FREE USE OF THIS SOFTWARE IN ITS "AS IS" CONDITION. CARNEGIE MELLON
* DISCLAIMS ANY LIABILITY OF ANY KIND FOR ANY DAMAGES WHATSOEVER
* RESULTING DIRECTLY OR INDIRECTLY FROM THE USE OF THIS SOFTWARE OR OF
* ANY DERIVATIVE WORK.
*
* Carnegie Mellon encourages users of this software to return any
* improvements or extensions that they make, and to grant Carnegie
* Mellon the rights to redistribute these changes without encumbrance.
*
1998-09-13 13:57:59 +00:00
* @(#) src/sys/coda/coda_subr.c,v 1.1.1.1 1998/08/29 21:14:52 rvb Exp $
2003-06-10 21:29:12 +00:00
*/
/*-
1998-08-29 21:14:52 +00:00
* Mach Operating System
* Copyright (c) 1989 Carnegie-Mellon University
* All rights reserved. The CMU software License Agreement specifies
* the terms and conditions for use and redistribution.
*/
/*
2002-05-16 21:28:32 +00:00
* This code was written for the Coda filesystem at Carnegie Mellon
1998-08-29 21:14:52 +00:00
* University. Contributers include David Steere, James Kistler, and
* M. Satyanarayanan.
*/
1998-08-29 21:14:52 +00:00
/* NOTES: rvb
* 1. Added coda_unmounting to mark all cnodes as being UNMOUNTING. This has to
1998-08-29 21:14:52 +00:00
* be done before dounmount is called. Because some of the routines that
* dounmount calls before coda_unmounted might try to force flushes to venus.
1998-08-29 21:14:52 +00:00
* The vnode pager does this.
* 2. coda_unmounting marks all cnodes scanning coda_cache.
1998-08-29 21:14:52 +00:00
* 3. cfs_checkunmounting (under DEBUG) checks all cnodes by chasing the vnodes
* under the /coda mount point.
* 4. coda_cacheprint (under DEBUG) prints names with vnode/cnode address
1998-08-29 21:14:52 +00:00
*/
2003-06-10 21:29:12 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
1998-08-29 21:14:52 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/lock.h>
1998-09-02 19:09:53 +00:00
#include <sys/malloc.h>
#include <sys/mutex.h>
1998-08-29 21:14:52 +00:00
#include <sys/mount.h>
1998-09-13 13:57:59 +00:00
#include <coda/coda.h>
#include <coda/cnode.h>
#include <coda/coda_subr.h>
#include <coda/coda_namecache.h>
1998-08-29 21:14:52 +00:00
int coda_active = 0;
int coda_reuse = 0;
int coda_new = 0;
1998-08-29 21:14:52 +00:00
struct cnode *coda_freelist = NULL;
struct cnode *coda_cache[CODA_CACHESIZE];
1998-08-29 21:14:52 +00:00
#define CNODE_NEXT(cp) ((cp)->c_next)
#ifdef CODA_COMPAT_5
#define coda_hash(fid) \
(((fid)->Volume + (fid)->Vnode) & (CODA_CACHESIZE-1))
#define IS_DIR(cnode) (cnode.Vnode & 0x1)
#else
#define coda_hash(fid) (coda_f2i(fid) & (CODA_CACHESIZE-1))
#define IS_DIR(cnode) (cnode.opaque[2] & 0x1)
#endif
1998-08-29 21:14:52 +00:00
/*
* Allocate a cnode.
*/
struct cnode *
coda_alloc(void)
1998-08-29 21:14:52 +00:00
{
struct cnode *cp;
if (coda_freelist) {
cp = coda_freelist;
coda_freelist = CNODE_NEXT(cp);
coda_reuse++;
1998-08-29 21:14:52 +00:00
}
else {
CODA_ALLOC(cp, struct cnode *, sizeof(struct cnode));
1998-08-29 21:14:52 +00:00
/* NetBSD vnodes don't have any Pager info in them ('cause there are
no external pagers, duh!) */
#define VNODE_VM_INFO_INIT(vp) /* MT */
VNODE_VM_INFO_INIT(CTOV(cp));
coda_new++;
1998-08-29 21:14:52 +00:00
}
bzero(cp, sizeof (struct cnode));
return(cp);
}
/*
* Deallocate a cnode.
*/
void
coda_free(cp)
1998-08-29 21:14:52 +00:00
register struct cnode *cp;
{
CNODE_NEXT(cp) = coda_freelist;
coda_freelist = cp;
1998-08-29 21:14:52 +00:00
}
/*
* Put a cnode in the hash table
*/
void
coda_save(cp)
1998-08-29 21:14:52 +00:00
struct cnode *cp;
{
CNODE_NEXT(cp) = coda_cache[coda_hash(&cp->c_fid)];
coda_cache[coda_hash(&cp->c_fid)] = cp;
1998-08-29 21:14:52 +00:00
}
/*
* Remove a cnode from the hash table
*/
void
coda_unsave(cp)
1998-08-29 21:14:52 +00:00
struct cnode *cp;
{
struct cnode *ptr;
struct cnode *ptrprev = NULL;
ptr = coda_cache[coda_hash(&cp->c_fid)];
1998-08-29 21:14:52 +00:00
while (ptr != NULL) {
if (ptr == cp) {
if (ptrprev == NULL) {
coda_cache[coda_hash(&cp->c_fid)]
1998-08-29 21:14:52 +00:00
= CNODE_NEXT(ptr);
} else {
CNODE_NEXT(ptrprev) = CNODE_NEXT(ptr);
}
CNODE_NEXT(cp) = (struct cnode *)NULL;
return;
}
ptrprev = ptr;
ptr = CNODE_NEXT(ptr);
}
}
/*
* Lookup a cnode by fid. If the cnode is dying, it is bogus so skip it.
* NOTE: this allows multiple cnodes with same fid -- dcs 1/25/95
*/
struct cnode *
coda_find(fid)
CodaFid *fid;
1998-08-29 21:14:52 +00:00
{
struct cnode *cp;
cp = coda_cache[coda_hash(fid)];
1998-08-29 21:14:52 +00:00
while (cp) {
if (coda_fid_eq(&(cp->c_fid), fid) &&
1998-08-29 21:14:52 +00:00
(!IS_UNMOUNTING(cp)))
{
coda_active++;
1998-08-29 21:14:52 +00:00
return(cp);
}
cp = CNODE_NEXT(cp);
}
return(NULL);
}
/*
* coda_kill is called as a side effect to vcopen. To prevent any
1998-08-29 21:14:52 +00:00
* cnodes left around from an earlier run of a venus or warden from
* causing problems with the new instance, mark any outstanding cnodes
* as dying. Future operations on these cnodes should fail (excepting
* coda_inactive of course!). Since multiple venii/wardens can be
1998-08-29 21:14:52 +00:00
* running, only kill the cnodes for a particular entry in the
* coda_mnttbl. -- DCS 12/1/94 */
1998-08-29 21:14:52 +00:00
int
coda_kill(whoIam, dcstat)
1998-08-29 21:14:52 +00:00
struct mount *whoIam;
enum dc_status dcstat;
{
int hash, count = 0;
struct cnode *cp;
/*
* Algorithm is as follows:
* Second, flush whatever vnodes we can from the name cache.
*
* Finally, step through whatever is left and mark them dying.
* This prevents any operation at all.
*/
/* This is slightly overkill, but should work. Eventually it'd be
* nice to only flush those entries from the namecache that
* reference a vnode in this vfs. */
coda_nc_flush(dcstat);
1998-08-29 21:14:52 +00:00
for (hash = 0; hash < CODA_CACHESIZE; hash++) {
for (cp = coda_cache[hash]; cp != NULL; cp = CNODE_NEXT(cp)) {
1998-08-29 21:14:52 +00:00
if (CTOV(cp)->v_mount == whoIam) {
#ifdef DEBUG
printf("coda_kill: vp %p, cp %p\n", CTOV(cp), cp);
1998-08-29 21:14:52 +00:00
#endif
count++;
CODADEBUG(CODA_FLUSH,
myprintf(("Live cnode fid %s flags %d count %d\n",
coda_f2s(&cp->c_fid),
cp->c_flags,
vrefcnt(CTOV(cp)))); );
1998-08-29 21:14:52 +00:00
}
}
}
return count;
}
/*
* There are two reasons why a cnode may be in use, it may be in the
* name cache or it may be executing.
*/
void
coda_flush(dcstat)
1998-08-29 21:14:52 +00:00
enum dc_status dcstat;
{
int hash;
struct cnode *cp;
coda_clstat.ncalls++;
coda_clstat.reqs[CODA_FLUSH]++;
1998-08-29 21:14:52 +00:00
coda_nc_flush(dcstat); /* flush files from the name cache */
1998-08-29 21:14:52 +00:00
for (hash = 0; hash < CODA_CACHESIZE; hash++) {
for (cp = coda_cache[hash]; cp != NULL; cp = CNODE_NEXT(cp)) {
if (!IS_DIR(cp->c_fid)) /* only files can be executed */
coda_vmflush(cp);
1998-08-29 21:14:52 +00:00
}
}
}
/*
* As a debugging measure, print out any cnodes that lived through a
* name cache flush.
*/
void
coda_testflush(void)
1998-08-29 21:14:52 +00:00
{
int hash;
struct cnode *cp;
for (hash = 0; hash < CODA_CACHESIZE; hash++) {
for (cp = coda_cache[hash];
1998-08-29 21:14:52 +00:00
cp != NULL;
cp = CNODE_NEXT(cp)) {
myprintf(("Live cnode fid %s count %d\n",
coda_f2s(&cp->c_fid), CTOV(cp)->v_usecount));
1998-08-29 21:14:52 +00:00
}
}
}
/*
* First, step through all cnodes and mark them unmounting.
* NetBSD kernels may try to fsync them now that venus
* is dead, which would be a bad thing.
*
*/
void
coda_unmounting(whoIam)
1998-08-29 21:14:52 +00:00
struct mount *whoIam;
{
int hash;
struct cnode *cp;
for (hash = 0; hash < CODA_CACHESIZE; hash++) {
for (cp = coda_cache[hash]; cp != NULL; cp = CNODE_NEXT(cp)) {
1998-08-29 21:14:52 +00:00
if (CTOV(cp)->v_mount == whoIam) {
if (cp->c_flags & (C_LOCKED|C_WANTED)) {
printf("coda_unmounting: Unlocking %p\n", cp);
1998-08-29 21:14:52 +00:00
cp->c_flags &= ~(C_LOCKED|C_WANTED);
wakeup((caddr_t) cp);
}
cp->c_flags |= C_UNMOUNTING;
}
}
}
}
#ifdef DEBUG
void
coda_checkunmounting(mp)
1998-08-29 21:14:52 +00:00
struct mount *mp;
{
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
struct vnode *vp, *nvp;
1998-08-29 21:14:52 +00:00
struct cnode *cp;
int count = 0, bad = 0;
MNT_ILOCK(mp);
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
MNT_VNODE_FOREACH(vp, mp, nvp) {
VI_LOCK(vp);
if (vp->v_iflag & VI_DOOMED) {
VI_UNLOCK(vp);
continue;
}
1998-08-29 21:14:52 +00:00
cp = VTOC(vp);
count++;
if (!(cp->c_flags & C_UNMOUNTING)) {
bad++;
printf("vp %p, cp %p missed\n", vp, cp);
cp->c_flags |= C_UNMOUNTING;
}
VI_UNLOCK(vp);
1998-08-29 21:14:52 +00:00
}
MNT_IUNLOCK(mp);
1998-08-29 21:14:52 +00:00
}
void
coda_cacheprint(whoIam)
1998-08-29 21:14:52 +00:00
struct mount *whoIam;
{
int hash;
struct cnode *cp;
int count = 0;
printf("coda_cacheprint: coda_ctlvp %p, cp %p", coda_ctlvp, VTOC(coda_ctlvp));
coda_nc_name(VTOC(coda_ctlvp));
1998-08-29 21:14:52 +00:00
printf("\n");
for (hash = 0; hash < CODA_CACHESIZE; hash++) {
for (cp = coda_cache[hash]; cp != NULL; cp = CNODE_NEXT(cp)) {
1998-08-29 21:14:52 +00:00
if (CTOV(cp)->v_mount == whoIam) {
printf("coda_cacheprint: vp %p, cp %p", CTOV(cp), cp);
coda_nc_name(cp);
1998-08-29 21:14:52 +00:00
printf("\n");
count++;
}
}
}
printf("coda_cacheprint: count %d\n", count);
1998-08-29 21:14:52 +00:00
}
#endif
/*
* There are 6 cases where invalidations occur. The semantics of each
* is listed here.
*
* CODA_FLUSH -- flush all entries from the name cache and the cnode cache.
* CODA_PURGEUSER -- flush all entries from the name cache for a specific user
1998-08-29 21:14:52 +00:00
* This call is a result of token expiration.
*
* The next two are the result of callbacks on a file or directory.
* CODA_ZAPDIR -- flush the attributes for the dir from its cnode.
1998-08-29 21:14:52 +00:00
* Zap all children of this directory from the namecache.
* CODA_ZAPFILE -- flush the attributes for a file.
1998-08-29 21:14:52 +00:00
*
* The fifth is a result of Venus detecting an inconsistent file.
* CODA_PURGEFID -- flush the attribute for the file
1998-08-29 21:14:52 +00:00
* If it is a dir (odd vnode), purge its
* children from the namecache
* remove the file from the namecache.
*
* The sixth allows Venus to replace local fids with global ones
* during reintegration.
*
* CODA_REPLACE -- replace one CodaFid with another throughout the name cache
1998-08-29 21:14:52 +00:00
*/
int handleDownCall(opcode, out)
int opcode; union outputArgs *out;
{
int error;
/* Handle invalidate requests. */
switch (opcode) {
case CODA_FLUSH : {
1998-08-29 21:14:52 +00:00
coda_flush(IS_DOWNCALL);
1998-08-29 21:14:52 +00:00
CODADEBUG(CODA_FLUSH,coda_testflush();) /* print remaining cnodes */
1998-08-29 21:14:52 +00:00
return(0);
}
case CODA_PURGEUSER : {
coda_clstat.ncalls++;
coda_clstat.reqs[CODA_PURGEUSER]++;
1998-08-29 21:14:52 +00:00
/* XXX - need to prevent fsync's */
#ifdef CODA_COMPAT_5
coda_nc_purge_user(out->coda_purgeuser.cred.cr_uid, IS_DOWNCALL);
#else
coda_nc_purge_user(out->coda_purgeuser.uid, IS_DOWNCALL);
#endif
1998-08-29 21:14:52 +00:00
return(0);
}
case CODA_ZAPFILE : {
1998-08-29 21:14:52 +00:00
struct cnode *cp;
error = 0;
coda_clstat.ncalls++;
coda_clstat.reqs[CODA_ZAPFILE]++;
1998-08-29 21:14:52 +00:00
cp = coda_find(&out->coda_zapfile.Fid);
1998-08-29 21:14:52 +00:00
if (cp != NULL) {
vref(CTOV(cp));
cp->c_flags &= ~C_VATTR;
ASSERT_VOP_LOCKED(CTOV(cp), "coda HandleDownCall");
if (CTOV(cp)->v_vflag & VV_TEXT)
error = coda_vmflush(cp);
CODADEBUG(CODA_ZAPFILE,
myprintf(("zapfile: fid = %s, refcnt = %d, error = %d\n",
coda_f2s(&cp->c_fid), CTOV(cp)->v_usecount - 1, error)););
if (vrefcnt(CTOV(cp)) == 1) {
1998-08-29 21:14:52 +00:00
cp->c_flags |= C_PURGING;
}
vrele(CTOV(cp));
}
return(error);
}
case CODA_ZAPDIR : {
1998-08-29 21:14:52 +00:00
struct cnode *cp;
coda_clstat.ncalls++;
coda_clstat.reqs[CODA_ZAPDIR]++;
1998-08-29 21:14:52 +00:00
cp = coda_find(&out->coda_zapdir.Fid);
1998-08-29 21:14:52 +00:00
if (cp != NULL) {
vref(CTOV(cp));
cp->c_flags &= ~C_VATTR;
coda_nc_zapParentfid(&out->coda_zapdir.Fid, IS_DOWNCALL);
CODADEBUG(CODA_ZAPDIR, myprintf((
"zapdir: fid = %s, refcnt = %d\n",
coda_f2s(&cp->c_fid), CTOV(cp)->v_usecount - 1)););
if (vrefcnt(CTOV(cp)) == 1) {
1998-08-29 21:14:52 +00:00
cp->c_flags |= C_PURGING;
}
vrele(CTOV(cp));
}
return(0);
}
case CODA_PURGEFID : {
1998-08-29 21:14:52 +00:00
struct cnode *cp;
error = 0;
coda_clstat.ncalls++;
coda_clstat.reqs[CODA_PURGEFID]++;
1998-08-29 21:14:52 +00:00
cp = coda_find(&out->coda_purgefid.Fid);
1998-08-29 21:14:52 +00:00
if (cp != NULL) {
vref(CTOV(cp));
if (IS_DIR(out->coda_purgefid.Fid)) { /* Vnode is a directory */
coda_nc_zapParentfid(&out->coda_purgefid.Fid,IS_DOWNCALL);
1998-08-29 21:14:52 +00:00
}
cp->c_flags &= ~C_VATTR;
coda_nc_zapfid(&out->coda_purgefid.Fid, IS_DOWNCALL);
ASSERT_VOP_LOCKED(CTOV(cp), "coda HandleDownCall");
if (!(IS_DIR(out->coda_purgefid.Fid))
&& (CTOV(cp)->v_vflag & VV_TEXT)) {
1998-08-29 21:14:52 +00:00
error = coda_vmflush(cp);
1998-08-29 21:14:52 +00:00
}
CODADEBUG(CODA_PURGEFID, myprintf((
"purgefid: fid = %s, refcnt = %d, error = %d\n",
coda_f2s(&cp->c_fid), CTOV(cp)->v_usecount - 1, error)););
if (vrefcnt(CTOV(cp)) == 1) {
1998-08-29 21:14:52 +00:00
cp->c_flags |= C_PURGING;
}
vrele(CTOV(cp));
}
return(error);
}
case CODA_REPLACE : {
1998-08-29 21:14:52 +00:00
struct cnode *cp = NULL;
coda_clstat.ncalls++;
coda_clstat.reqs[CODA_REPLACE]++;
1998-08-29 21:14:52 +00:00
cp = coda_find(&out->coda_replace.OldFid);
1998-08-29 21:14:52 +00:00
if (cp != NULL) {
/* remove the cnode from the hash table, replace the fid, and reinsert */
vref(CTOV(cp));
coda_unsave(cp);
cp->c_fid = out->coda_replace.NewFid;
coda_save(cp);
CODADEBUG(CODA_REPLACE, myprintf((
"replace: oldfid = %s, newfid = %s, cp = %p\n",
coda_f2s(&out->coda_replace.OldFid),
coda_f2s(&cp->c_fid), cp));) vrele(CTOV(cp));
1998-08-29 21:14:52 +00:00
}
return (0);
}
default:
myprintf(("handleDownCall: unknown opcode %d\n", opcode));
return (EINVAL);
}
}
/* coda_grab_vnode: lives in either cfs_mach.c or cfs_nbsd.c */
1998-08-29 21:14:52 +00:00
int
coda_vmflush(cp)
1998-08-29 21:14:52 +00:00
struct cnode *cp;
{
return 0;
}
/*
* kernel-internal debugging switches
*/
void coda_debugon(void)
1998-08-29 21:14:52 +00:00
{
codadebug = -1;
coda_nc_debug = -1;
coda_vnop_print_entry = 1;
coda_psdev_print_entry = 1;
coda_vfsop_print_entry = 1;
1998-08-29 21:14:52 +00:00
}
void coda_debugoff(void)
1998-08-29 21:14:52 +00:00
{
codadebug = 0;
coda_nc_debug = 0;
coda_vnop_print_entry = 0;
coda_psdev_print_entry = 0;
coda_vfsop_print_entry = 0;
1998-08-29 21:14:52 +00:00
}
/*
* Utilities used by both client and server
* Standard levels:
* 0) no debugging
* 1) hard failures
* 2) soft failures
* 3) current test software
* 4) main procedure entry points
* 5) main procedure exit points
* 6) utility procedure entry points
* 7) utility procedure exit points
* 8) obscure procedure entry points
* 9) obscure procedure exit points
* 10) random stuff
* 11) all <= 1
* 12) all <= 2
* 13) all <= 3
* ...
*/