NetBSD/sys/msdosfs/msdosfs_denode.c

698 lines
18 KiB
C
Raw Normal View History

1994-07-17 01:32:06 +04:00
/* $NetBSD: msdosfs_denode.c,v 1.6 1994/07/16 21:33:20 cgd Exp $ */
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
/*
* Written by Paul Popelka (paulp@uts.amdahl.com)
*
* You can do anything you want with this software, just don't say you wrote
* it, and don't reoove this notice.
*
* This software is provided "as is".
*
* The authop supplies this software to be publicly redistributed on the
* understanding that the author is not responsible for the correct
* functioning of this software in any circumstances and is not liable for
* any damages caused by this software.
*
* October 1992
*/
1993-12-18 03:50:51 +03:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/mount.h>
1994-07-17 01:32:06 +04:00
#include <sys/malloc.h>
1993-12-18 03:50:51 +03:00
#include <sys/proc.h>
#include <sys/buf.h>
#include <sys/vnode.h>
#include <sys/kernel.h> /* defines "time" */
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
1993-12-18 03:50:51 +03:00
#include <msdosfs/bpb.h>
#include <msdosfs/msdosfsmount.h>
#include <msdosfs/direntry.h>
#include <msdosfs/denode.h>
#include <msdosfs/fat.h>
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
1994-07-17 01:32:06 +04:00
struct denode **dehashtbl;
u_long dehash; /* size of hash table - 1 */
#define DEHASH(dev, deno) (((dev) + (deno)) & dehash)
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
msdosfs_init()
{
1994-07-17 01:32:06 +04:00
dehashtbl = hashinit(desiredvnodes/2, M_MSDOSFSMNT, &dehash);
}
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
1994-07-17 01:32:06 +04:00
static struct denode *
msdosfs_hashget(dev, dirclust, diroff)
dev_t dev;
u_long dirclust;
u_long diroff;
{
struct denode *dep;
for (;;)
for (dep = dehashtbl[DEHASH(dev, dirclust + diroff)];;
dep = dep->de_next) {
if (dep == NULL)
return NULL;
if (dirclust != dep->de_dirclust
|| diroff != dep->de_diroffset
|| dev != dep->de_dev
|| dep->de_refcnt == 0)
continue;
if (dep->de_flag & DELOCKED) {
dep->de_flag |= DEWANT;
sleep((caddr_t)dep, PINOD);
break;
}
if (!vget(DETOV(dep), 1))
return dep;
break;
}
/* NOTREACHED */
}
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
1994-07-17 01:32:06 +04:00
static void
msdosfs_hashins(dep)
struct denode *dep;
{
struct denode **depp, *deq;
depp = &dehashtbl[DEHASH(dep->de_dev, dep->de_dirclust + dep->de_diroffset)];
if (deq = *depp)
deq->de_prev = &dep->de_next;
dep->de_next = deq;
dep->de_prev = depp;
*depp = dep;
if (dep->de_flag & DELOCKED)
panic("msdosfs_hashins: already locked");
if (curproc)
dep->de_lockholder = curproc->p_pid;
else
dep->de_lockholder = -1;
dep->de_flag |= DELOCKED;
}
static void
msdosfs_hashrem(dep)
struct denode *dep;
{
struct denode *deq;
if (deq = dep->de_next)
deq->de_prev = dep->de_prev;
*dep->de_prev = deq;
#ifdef DIAGNOSTIC
dep->de_next = NULL;
dep->de_prev = NULL;
#endif
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
}
/*
* If deget() succeeds it returns with the gotten denode locked().
1994-07-17 01:32:06 +04:00
*
* pmp - address of msdosfsmount structure of the filesystem containing
* the denode of interest. The pm_dev field and the address of
* the msdosfsmount structure are used.
* dirclust - which cluster bp contains, if dirclust is 0 (root directory)
* diroffset is relative to the beginning of the root directory,
* otherwise it is cluster relative.
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
* diroffset - offset past begin of cluster of denode we want
1994-07-17 01:32:06 +04:00
* direntptr - address of the direntry structure of interest. If direntptr is
* NULL, the block is read if necessary.
* depp - returns the address of the gotten denode.
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
*/
int
deget(pmp, dirclust, diroffset, direntptr, depp)
struct msdosfsmount *pmp; /* so we know the maj/min number */
u_long dirclust; /* cluster this dir entry came from */
u_long diroffset; /* index of entry within the cluster */
struct direntry *direntptr;
struct denode **depp; /* returns the addr of the gotten denode */
{
int error;
dev_t dev = pmp->pm_dev;
struct mount *mntp = pmp->pm_mountp;
1994-07-17 01:32:06 +04:00
extern int (**msdosfs_vnodeop_p)();
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
struct denode *ldep;
struct vnode *nvp;
struct buf *bp;
#if defined(MSDOSFSDEBUG)
printf("deget(pmp %08x, dirclust %d, diroffset %x, direntptr %x, depp %08x)\n",
1994-07-17 01:32:06 +04:00
pmp, dirclust, diroffset, direntptr, depp);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
#endif /* defined(MSDOSFSDEBUG) */
/*
* If dir entry is given and refers to a directory, convert to
* canonical form
*/
if (direntptr && (direntptr->deAttributes & ATTR_DIRECTORY)) {
dirclust = getushort(direntptr->deStartCluster);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
if (dirclust == MSDOSFSROOT)
diroffset = MSDOSFSROOT_OFS;
else
diroffset = 0;
}
/*
* See if the denode is in the denode cache. Use the location of
* the directory entry to compute the hash value. For subdir use
* address of "." entry. for root dir use cluster MSDOSFSROOT,
* offset MSDOSFSROOT_OFS
*
* NOTE: The check for de_refcnt > 0 below insures the denode being
* examined does not represent an unlinked but still open file.
* These files are not to be accessible even when the directory
* entry that represented the file happens to be reused while the
* deleted file is still open.
*/
1994-07-17 01:32:06 +04:00
if (ldep = msdosfs_hashget(dev, dirclust, diroffset)) {
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
*depp = ldep;
return 0;
}
/*
* Directory entry was not in cache, have to create a vnode and
* copy it from the passed disk buffer.
*/
/* getnewvnode() does a VREF() on the vnode */
1994-07-17 01:32:06 +04:00
if (error = getnewvnode(VT_MSDOSFS, mntp, msdosfs_vnodeop_p, &nvp)) {
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
*depp = 0;
return error;
}
1994-07-17 01:32:06 +04:00
MALLOC(ldep, struct denode *, sizeof(struct denode), M_MSDOSFSNODE, M_WAITOK);
bzero((caddr_t)ldep, sizeof *ldep);
nvp->v_data = ldep;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
ldep->de_vnode = nvp;
ldep->de_flag = 0;
ldep->de_devvp = 0;
ldep->de_lockf = 0;
ldep->de_dev = dev;
1994-07-17 01:32:06 +04:00
ldep->de_dirclust = dirclust;
ldep->de_diroffset = diroffset;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
fc_purge(ldep, 0); /* init the fat cache for this denode */
/*
* Insert the denode into the hash queue and lock the denode so it
* can't be accessed until we've read it in and have done what we
* need to it.
*/
1994-07-17 01:32:06 +04:00
msdosfs_hashins(ldep);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
/*
* Copy the directory entry into the denode area of the vnode.
*/
if (dirclust == MSDOSFSROOT && diroffset == MSDOSFSROOT_OFS) {
/*
* Directory entry for the root directory. There isn't one,
* so we manufacture one. We should probably rummage
* through the root directory and find a label entry (if it
* exists), and then use the time and date from that entry
* as the time and date for the root denode.
*/
ldep->de_Attributes = ATTR_DIRECTORY;
ldep->de_StartCluster = MSDOSFSROOT;
ldep->de_FileSize = pmp->pm_rootdirsize * pmp->pm_BytesPerSec;
/*
* fill in time and date so that dos2unixtime() doesn't
* spit up when called from msdosfs_getattr() with root
* denode
*/
ldep->de_Time = 0x0000; /* 00:00:00 */
ldep->de_Date = (0 << DD_YEAR_SHIFT) | (1 << DD_MONTH_SHIFT)
| (1 << DD_DAY_SHIFT);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
/* Jan 1, 1980 */
/* leave the other fields as garbage */
1994-07-17 01:32:06 +04:00
} else {
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
bp = NULL;
if (!direntptr) {
error = readep(pmp, dirclust, diroffset, &bp,
&direntptr);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
if (error)
return error;
}
DE_INTERNALIZE(ldep, direntptr);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
if (bp)
brelse(bp);
}
/*
* Fill in a few fields of the vnode and finish filling in the
* denode. Then return the address of the found denode.
*/
ldep->de_pmp = pmp;
ldep->de_devvp = pmp->pm_devvp;
ldep->de_refcnt = 1;
if (ldep->de_Attributes & ATTR_DIRECTORY) {
/*
* Since DOS directory entries that describe directories
* have 0 in the filesize field, we take this opportunity
* to find out the length of the directory and plug it into
* the denode structure.
*/
u_long size;
nvp->v_type = VDIR;
if (ldep->de_StartCluster == MSDOSFSROOT)
nvp->v_flag |= VROOT;
else {
error = pcbmap(ldep, 0xffff, 0, &size);
if (error == E2BIG) {
ldep->de_FileSize = size << pmp->pm_cnshift;
error = 0;
1994-07-17 01:32:06 +04:00
} else
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
printf("deget(): pcbmap returned %d\n", error);
}
1994-07-17 01:32:06 +04:00
} else
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
nvp->v_type = VREG;
VREF(ldep->de_devvp);
*depp = ldep;
return 0;
}
int
deupdat(dep, tp, waitfor)
struct denode *dep;
struct timespec *tp;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
int waitfor;
{
int error;
daddr_t bn;
int diro;
struct buf *bp;
struct direntry *dirp;
struct msdosfsmount *pmp = dep->de_pmp;
struct timespec ts;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
struct vnode *vp = DETOV(dep);
#if defined(MSDOSFSDEBUG)
printf("deupdat(): dep %08x\n", dep);
#endif /* defined(MSDOSFSDEBUG) */
/*
* If the update bit is off, or this denode is from a readonly
* filesystem, or this denode is for a directory, or the denode
* represents an open but unlinked file then don't do anything. DOS
* directory entries that describe a directory do not ever get
* updated. This is the way dos treats them.
*/
if ((dep->de_flag & DEUPD) == 0 ||
vp->v_mount->mnt_flag & MNT_RDONLY ||
dep->de_Attributes & ATTR_DIRECTORY ||
dep->de_refcnt <= 0)
return 0;
/*
* Read in the cluster containing the directory entry we want to
* update.
*/
if (error = readde(dep, &bp, &dirp))
return error;
/*
* Put the passed in time into the directory entry.
*/
TIMEVAL_TO_TIMESPEC(&time, &ts);
unix2dostime(&ts, &dep->de_Date, &dep->de_Time);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
dep->de_flag &= ~DEUPD;
/*
* Copy the directory entry out of the denode into the cluster it
* came from.
*/
DE_EXTERNALIZE(dirp, dep);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
/*
* Write the cluster back to disk. If they asked for us to wait
* for the write to complete, then use bwrite() otherwise use
* bdwrite().
*/
error = 0; /* note that error is 0 from above, but ... */
if (waitfor)
error = bwrite(bp);
else
bdwrite(bp);
return error;
}
/*
* Truncate the file described by dep to the length specified by length.
*/
int
1994-07-17 01:32:06 +04:00
detrunc(dep, length, flags, cred, p)
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
struct denode *dep;
u_long length;
int flags;
1994-07-17 01:32:06 +04:00
struct ucred *cred;
struct proc *p;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
{
int error;
int allerror;
1994-07-17 01:32:06 +04:00
int vflags;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
u_long eofentry;
u_long chaintofree;
daddr_t bn;
int boff;
int isadir = dep->de_Attributes & ATTR_DIRECTORY;
struct buf *bp;
struct msdosfsmount *pmp = dep->de_pmp;
#if defined(MSDOSFSDEBUG)
printf("detrunc(): file %s, length %d, flags %d\n", dep->de_Name, length, flags);
#endif /* defined(MSDOSFSDEBUG) */
/*
* Disallow attempts to truncate the root directory since it is of
* fixed size. That's just the way dos filesystems are. We use
* the VROOT bit in the vnode because checking for the directory
* bit and a startcluster of 0 in the denode is not adequate to
* recognize the root directory at this point in a file or
* directory's life.
*/
if (DETOV(dep)->v_flag & VROOT) {
printf("detrunc(): can't truncate root directory, clust %d, offset %d\n",
dep->de_dirclust, dep->de_diroffset);
return EINVAL;
}
vnode_pager_setsize(DETOV(dep), length);
if (dep->de_FileSize <= length) {
dep->de_flag |= DEUPD;
error = deupdat(dep, &time, 1);
#if defined(MSDOSFSDEBUG)
printf("detrunc(): file is shorter return point, errno %d\n", error);
#endif /* defined(MSDOSFSDEBUG) */
return error;
}
/*
* If the desired length is 0 then remember the starting cluster of
* the file and set the StartCluster field in the directory entry
* to 0. If the desired length is not zero, then get the number of
* the last cluster in the shortened file. Then get the number of
* the first cluster in the part of the file that is to be freed.
* Then set the next cluster pointer in the last cluster of the
* file to CLUST_EOFE.
*/
if (length == 0) {
chaintofree = dep->de_StartCluster;
dep->de_StartCluster = 0;
eofentry = ~0;
1994-07-17 01:32:06 +04:00
} else {
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
error = pcbmap(dep, (length - 1) >> pmp->pm_cnshift,
0, &eofentry);
if (error) {
#if defined(MSDOSFSDEBUG)
printf("detrunc(): pcbmap fails %d\n", error);
#endif /* defined(MSDOSFSDEBUG) */
return error;
}
}
fc_purge(dep, (length + pmp->pm_crbomask) >> pmp->pm_cnshift);
/*
* If the new length is not a multiple of the cluster size then we
* must zero the tail end of the new last cluster in case it
* becomes part of the file again because of a seek.
*/
if ((boff = length & pmp->pm_crbomask) != 0) {
/*
* should read from file vnode or filesystem vnode
* depending on if file or dir
*/
if (isadir) {
bn = cntobn(pmp, eofentry);
error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster,
NOCRED, &bp);
1994-07-17 01:32:06 +04:00
} else {
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
bn = (length - 1) >> pmp->pm_cnshift;
error = bread(DETOV(dep), bn, pmp->pm_bpcluster,
NOCRED, &bp);
}
if (error) {
#if defined(MSDOSFSDEBUG)
printf("detrunc(): bread fails %d\n", error);
#endif /* defined(MSDOSFSDEBUG) */
return error;
}
vnode_pager_uncache(DETOV(dep)); /* what's this for? */
/*
* is this the right place for it?
*/
1994-07-17 01:32:06 +04:00
bzero(bp->b_data + boff, pmp->pm_bpcluster - boff);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
if (flags & IO_SYNC)
bwrite(bp);
else
bdwrite(bp);
}
/*
* Write out the updated directory entry. Even if the update fails
* we free the trailing clusters.
*/
dep->de_FileSize = length;
dep->de_flag |= DEUPD;
1994-07-17 01:32:06 +04:00
vflags = (length > 0 ? V_SAVE : 0) | V_SAVEMETA;
vinvalbuf(DETOV(dep), vflags, cred, p, 0, 0);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
allerror = deupdat(dep, &time, MNT_WAIT);
#if defined(MSDOSFSDEBUG)
printf("detrunc(): allerror %d, eofentry %d\n",
1994-07-17 01:32:06 +04:00
allerror, eofentry);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
#endif /* defined(MSDOSFSDEBUG) */
/*
* If we need to break the cluster chain for the file then do it
* now.
*/
if (eofentry != ~0) {
error = fatentry(FAT_GET_AND_SET, pmp, eofentry,
1994-07-17 01:32:06 +04:00
&chaintofree, CLUST_EOFE);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
if (error) {
#if defined(MSDOSFSDEBUG)
printf("detrunc(): fatentry errors %d\n", error);
#endif /* defined(MSDOSFSDEBUG) */
return error;
}
fc_setcache(dep, FC_LASTFC, (length - 1) >> pmp->pm_cnshift,
1994-07-17 01:32:06 +04:00
eofentry);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
}
/*
* Now free the clusters removed from the file because of the
* truncation.
*/
if (chaintofree != 0 && !MSDOSFSEOF(chaintofree))
freeclusterchain(pmp, chaintofree);
return allerror;
}
1994-07-17 01:32:06 +04:00
/*
* Extend the file described by dep to length specified by length.
*/
int
deextend(dep, length, cred)
struct denode *dep;
off_t length;
struct ucred *cred;
{
struct msdosfsmount *pmp = dep->de_pmp;
off_t foff;
int error;
struct buf *bp;
/*
* The root of a DOS filesystem cannot be extended.
*/
if (DETOV(dep)->v_flag & VROOT)
return EINVAL;
/*
* Directories can only be extended by the superuser.
* Is this really important?
*/
if (dep->de_Attributes & ATTR_DIRECTORY) {
if (error = suser(cred, NULL))
return error;
}
if (length <= dep->de_FileSize)
panic("deextend: file too large");
/*
* Compute the offset of the first byte after the last block in the
* file.
*/
foff = (dep->de_FileSize + (pmp->pm_bpcluster - 1)) & ~pmp->pm_crbomask;
while (foff < length) {
if (error = extendfile(dep, &bp, 0))
return error;
dep->de_flag |= DEUPD;
bdwrite(bp);
foff += pmp->pm_bpcluster;
dep->de_FileSize += pmp->pm_bpcluster;
}
dep->de_FileSize = length;
return deupdat(dep, &time);
}
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
/*
* Move a denode to its correct hash queue after the file it represents has
* been moved to a new directory.
*/
reinsert(dep)
struct denode *dep;
{
struct msdosfsmount *pmp = dep->de_pmp;
union dehead *deh;
/*
* Fix up the denode cache. If the denode is for a directory,
* there is nothing to do since the hash is based on the starting
* cluster of the directory file and that hasn't changed. If for a
* file the hash is based on the location of the directory entry,
* so we must remove it from the cache and re-enter it with the
* hash based on the new location of the directory entry.
*/
if ((dep->de_Attributes & ATTR_DIRECTORY) == 0) {
1994-07-17 01:32:06 +04:00
msdosfs_hashrem(dep);
msdosfs_hashins(dep);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
}
}
int
1994-07-17 01:32:06 +04:00
msdosfs_reclaim(ap)
struct vop_reclaim_args /* {
struct vnode *a_vp;
} */ *ap;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
{
1994-07-17 01:32:06 +04:00
struct vnode *vp = ap->a_vp;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
struct denode *dep = VTODE(vp);
int i;
1994-07-17 01:32:06 +04:00
extern int prtactive;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
#if defined(MSDOSFSDEBUG)
printf("msdosfs_reclaim(): dep %08x, file %s, refcnt %d\n",
dep, dep->de_Name, dep->de_refcnt);
#endif /* defined(MSDOSFSDEBUG) */
1994-07-17 01:32:06 +04:00
if (prtactive && vp->v_usecount != 0)
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
vprint("msdosfs_reclaim(): pushing active", vp);
/*
* Remove the denode from the denode hash chain we are in.
*/
1994-07-17 01:32:06 +04:00
msdosfs_hashrem(dep);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
cache_purge(vp);
/*
* Indicate that one less file on the filesystem is open.
*/
if (dep->de_devvp) {
vrele(dep->de_devvp);
dep->de_devvp = 0;
}
dep->de_flag = 0;
1994-07-17 01:32:06 +04:00
FREE(dep, M_MSDOSFSNODE);
vp->v_data = NULL;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
return 0;
}
int
1994-07-17 01:32:06 +04:00
msdosfs_inactive(ap)
struct vop_inactive_args /* {
struct vnode *a_vp;
} */ *ap;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
{
1994-07-17 01:32:06 +04:00
struct vnode *vp = ap->a_vp;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
struct denode *dep = VTODE(vp);
int error = 0;
1994-07-17 01:32:06 +04:00
extern int prtactive;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
#if defined(MSDOSFSDEBUG)
printf("msdosfs_inactive(): dep %08x, de_Name[0] %x\n", dep, dep->de_Name[0]);
#endif /* defined(MSDOSFSDEBUG) */
1994-07-17 01:32:06 +04:00
if (prtactive && vp->v_usecount != 0)
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
vprint("msdosfs_inactive(): pushing active", vp);
/*
* Get rid of denodes related to stale file handles. Hmmm, what
* does this really do?
*/
if (dep->de_Name[0] == SLOT_DELETED) {
if ((vp->v_flag & VXLOCK) == 0)
vgone(vp);
return 0;
}
/*
* If the file has been deleted and it is on a read/write
* filesystem, then truncate the file, and mark the directory slot
1994-07-17 01:32:06 +04:00
* as empty. (This may not be necessary for the dos filesystem.)
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
*/
#if defined(MSDOSFSDEBUG)
printf("msdosfs_inactive(): dep %08x, refcnt %d, mntflag %x, MNT_RDONLY %x\n",
1994-07-17 01:32:06 +04:00
dep, dep->de_refcnt, vp->v_mount->mnt_flag, MNT_RDONLY);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
#endif /* defined(MSDOSFSDEBUG) */
DELOCK(dep);
if (dep->de_refcnt <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1994-07-17 01:32:06 +04:00
error = detrunc(dep, (u_long) 0, 0, NOCRED, NULL);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
dep->de_flag |= DEUPD;
dep->de_Name[0] = SLOT_DELETED;
}
DEUPDAT(dep, &time, 0);
DEUNLOCK(dep);
dep->de_flag = 0;
/*
* If we are done with the denode, then reclaim it so that it can
* be reused now.
*/
#if defined(MSDOSFSDEBUG)
printf("msdosfs_inactive(): v_usecount %d, de_Name[0] %x\n", vp->v_usecount,
1994-07-17 01:32:06 +04:00
dep->de_Name[0]);
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
#endif /* defined(MSDOSFSDEBUG) */
if (vp->v_usecount == 0 && dep->de_Name[0] == SLOT_DELETED)
vgone(vp);
return error;
}
int
delock(dep)
struct denode *dep;
{
while (dep->de_flag & DELOCKED) {
dep->de_flag |= DEWANT;
1994-07-17 01:32:06 +04:00
if (dep->de_lockholder == curproc->p_pid)
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
panic("delock: locking against myself");
1994-07-17 01:32:06 +04:00
dep->de_lockwaiter = curproc->p_pid;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
(void) sleep((caddr_t) dep, PINOD);
}
1994-07-17 01:32:06 +04:00
dep->de_lockwaiter = 0;
dep->de_lockholder = curproc->p_pid;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
dep->de_flag |= DELOCKED;
return 0;
}
int
deunlock(dep)
struct denode *dep;
{
if ((dep->de_flag & DELOCKED) == 0)
vprint("deunlock: found unlocked denode", DETOV(dep));
1994-07-17 01:32:06 +04:00
dep->de_lockholder = 0;
brought in fixed/renamed/matching MS-DOS FS code, from Jeff Polk <polk@bsdi.com>. His notes are as follows: ------------------------------------------------------------------------------ July 22, 1993 - Changed name of entire package from PCFS to MSDOSFS - Fixed bugs: root directory size in clusters instead of bytes growing directory didn't update in-core size link, symlink, mknod didn't free locked parent (deadlock) lookup returned real error on create and rename instead of EJUSTRETURN rename changed `.' entry in child instead of name entry in parent rename removed `.' entry in child instead of removing entry in parent when moving a directory from one dir to another createde() left new node locked when write of parent failed (deadlock) removede() decremented refcount even on error (rmdir's which failed due to write errors left in-core cache entries inconsistent) changed validation for filesystem to not check for the boot signature since some disks (e.g., mtools) aren't bootable directories are always show current time as modify time (needed for NFS export since DOS never updates dir mod times -- ctime is true create time). - Added support for cookies changes to the readdir() vnode interface (#ifdef __bsdi__) - Punted on the whole problem of inode generation numbers. This means that there's a chance of using a stale file handle to access a new file, but it doesn't appear to be the common case, and I don't see how to generate reasonable generation numbers without changing something on the disk (which is the way the SVR4 filesystem survival kit guys did it). I don't think it would be very safe to change the on-disk format. Jeff Polk (polk@BSDI.COM) ------------------------------------------------------------------------------
1993-08-13 15:35:13 +04:00
dep->de_flag &= ~DELOCKED;
if (dep->de_flag & DEWANT) {
dep->de_flag &= ~DEWANT;
wakeup((caddr_t) dep);
}
return 0;
}