59ff5cbd87
. also implement now-possible fsck -p option . allows unconditional fsck -p invocation at startup, only checking each filesystem if not marked clean . mounting unclean is allowed but is forced readonly . updating the superblock while mounted is now not allowed by mfs - must be done (e.g. by fsck.mfs) on an unmounted fs . clean flag is unset by mfs on mounting, and set by mfs on clean unmounting (if clean flag was set at mount time) Signed-off-by: Ben Gras <ben@minix3.org>
196 lines
5.5 KiB
C
196 lines
5.5 KiB
C
#include "fs.h"
|
|
#include "inode.h"
|
|
#include "super.h"
|
|
#include <minix/vfsif.h>
|
|
#include <minix/bdev.h>
|
|
|
|
PRIVATE int cleanmount = 1;
|
|
|
|
/*===========================================================================*
|
|
* fs_readsuper *
|
|
*===========================================================================*/
|
|
PUBLIC int fs_readsuper()
|
|
{
|
|
/* This function reads the superblock of the partition, gets the root inode
|
|
* and sends back the details of them. Note, that the FS process does not
|
|
* know the index of the vmnt object which refers to it, whenever the pathname
|
|
* lookup leaves a partition an ELEAVEMOUNT error is transferred back
|
|
* so that the VFS knows that it has to find the vnode on which this FS
|
|
* process' partition is mounted on.
|
|
*/
|
|
struct inode *root_ip;
|
|
cp_grant_id_t label_gid;
|
|
size_t label_len;
|
|
int r;
|
|
int readonly, isroot;
|
|
|
|
fs_dev = (dev_t) fs_m_in.REQ_DEV;
|
|
label_gid = (cp_grant_id_t) fs_m_in.REQ_GRANT;
|
|
label_len = (size_t) fs_m_in.REQ_PATH_LEN;
|
|
readonly = (fs_m_in.REQ_FLAGS & REQ_RDONLY) ? 1 : 0;
|
|
isroot = (fs_m_in.REQ_FLAGS & REQ_ISROOT) ? 1 : 0;
|
|
|
|
if (label_len > sizeof(fs_dev_label))
|
|
return(EINVAL);
|
|
|
|
r = sys_safecopyfrom(fs_m_in.m_source, label_gid, (vir_bytes) 0,
|
|
(vir_bytes) fs_dev_label, label_len, D);
|
|
if (r != OK) {
|
|
printf("MFS %s:%d safecopyfrom failed: %d\n", __FILE__, __LINE__, r);
|
|
return(EINVAL);
|
|
}
|
|
|
|
/* Map the driver label for this major. */
|
|
bdev_driver(fs_dev, fs_dev_label);
|
|
|
|
/* Open the device the file system lives on. */
|
|
if (bdev_open(fs_dev, readonly ? R_BIT : (R_BIT|W_BIT) ) != OK) {
|
|
return(EINVAL);
|
|
}
|
|
|
|
/* Fill in the super block. */
|
|
superblock.s_dev = fs_dev; /* read_super() needs to know which dev */
|
|
r = read_super(&superblock);
|
|
|
|
/* Is it recognized as a Minix filesystem? */
|
|
if (r != OK) {
|
|
superblock.s_dev = NO_DEV;
|
|
bdev_close(fs_dev);
|
|
return(r);
|
|
}
|
|
|
|
/* Remember whether we were mounted cleanly so we know what to
|
|
* do at unmount time
|
|
*/
|
|
if(superblock.s_flags & MFSFLAG_CLEAN)
|
|
cleanmount = 1;
|
|
|
|
/* clean check: if rw and not clean, switch to readonly */
|
|
if(!(superblock.s_flags & MFSFLAG_CLEAN) && !readonly) {
|
|
if(bdev_close(fs_dev) != OK)
|
|
panic("couldn't bdev_close after found unclean FS");
|
|
readonly = 1;
|
|
|
|
if (bdev_open(fs_dev, R_BIT) != OK) {
|
|
panic("couldn't bdev_open after found unclean FS");
|
|
return(EINVAL);
|
|
}
|
|
printf("MFS: WARNING: FS 0x%x unclean, mounting readonly\n", fs_dev);
|
|
}
|
|
|
|
set_blocksize(&superblock);
|
|
|
|
/* Get the root inode of the mounted file system. */
|
|
if( (root_ip = get_inode(fs_dev, ROOT_INODE)) == NULL) {
|
|
printf("MFS: couldn't get root inode\n");
|
|
superblock.s_dev = NO_DEV;
|
|
bdev_close(fs_dev);
|
|
return(EINVAL);
|
|
}
|
|
|
|
if(root_ip->i_mode == 0) {
|
|
printf("%s:%d zero mode for root inode?\n", __FILE__, __LINE__);
|
|
put_inode(root_ip);
|
|
superblock.s_dev = NO_DEV;
|
|
bdev_close(fs_dev);
|
|
return(EINVAL);
|
|
}
|
|
|
|
superblock.s_rd_only = readonly;
|
|
superblock.s_is_root = isroot;
|
|
|
|
/* Root inode properties */
|
|
fs_m_out.RES_INODE_NR = root_ip->i_num;
|
|
fs_m_out.RES_MODE = root_ip->i_mode;
|
|
fs_m_out.RES_FILE_SIZE_LO = root_ip->i_size;
|
|
fs_m_out.RES_UID = root_ip->i_uid;
|
|
fs_m_out.RES_GID = root_ip->i_gid;
|
|
|
|
fs_m_out.RES_CONREQS = 1; /* We can handle only 1 request at a time */
|
|
|
|
/* Mark it dirty */
|
|
if(!superblock.s_rd_only) {
|
|
superblock.s_flags &= ~MFSFLAG_CLEAN;
|
|
if(write_super(&superblock) != OK)
|
|
panic("mounting: couldn't write dirty superblock");
|
|
}
|
|
|
|
return(r);
|
|
}
|
|
|
|
|
|
/*===========================================================================*
|
|
* fs_mountpoint *
|
|
*===========================================================================*/
|
|
PUBLIC int fs_mountpoint()
|
|
{
|
|
/* This function looks up the mount point, it checks the condition whether
|
|
* the partition can be mounted on the inode or not.
|
|
*/
|
|
register struct inode *rip;
|
|
int r = OK;
|
|
mode_t bits;
|
|
|
|
/* Temporarily open the file. */
|
|
if( (rip = get_inode(fs_dev, (ino_t) fs_m_in.REQ_INODE_NR)) == NULL)
|
|
return(EINVAL);
|
|
|
|
|
|
if(rip->i_mountpoint) r = EBUSY;
|
|
|
|
/* It may not be special. */
|
|
bits = rip->i_mode & I_TYPE;
|
|
if (bits == I_BLOCK_SPECIAL || bits == I_CHAR_SPECIAL) r = ENOTDIR;
|
|
|
|
put_inode(rip);
|
|
|
|
if(r == OK) rip->i_mountpoint = TRUE;
|
|
|
|
return(r);
|
|
}
|
|
|
|
|
|
/*===========================================================================*
|
|
* fs_unmount *
|
|
*===========================================================================*/
|
|
PUBLIC int fs_unmount()
|
|
{
|
|
/* Unmount a file system by device number. */
|
|
int count;
|
|
struct inode *rip, *root_ip;
|
|
|
|
if(superblock.s_dev != fs_dev) return(EINVAL);
|
|
|
|
/* See if the mounted device is busy. Only 1 inode using it should be
|
|
* open --the root inode-- and that inode only 1 time. */
|
|
count = 0;
|
|
for (rip = &inode[0]; rip < &inode[NR_INODES]; rip++)
|
|
if (rip->i_count > 0 && rip->i_dev == fs_dev) count += rip->i_count;
|
|
|
|
if ((root_ip = find_inode(fs_dev, ROOT_INODE)) == NULL) {
|
|
panic("MFS: couldn't find root inode\n");
|
|
return(EINVAL);
|
|
}
|
|
|
|
if (count > 1) return(EBUSY); /* can't umount a busy file system */
|
|
put_inode(root_ip);
|
|
|
|
/* force any cached blocks out of memory */
|
|
(void) fs_sync();
|
|
|
|
/* Mark it clean if we're allowed to write _and_ it was clean originally. */
|
|
if(cleanmount && !superblock.s_rd_only) {
|
|
superblock.s_flags |= MFSFLAG_CLEAN;
|
|
write_super(&superblock);
|
|
}
|
|
|
|
/* Close the device the file system lives on. */
|
|
bdev_close(fs_dev);
|
|
|
|
/* Finish off the unmount. */
|
|
superblock.s_dev = NO_DEV;
|
|
unmountdone = TRUE;
|
|
|
|
return(OK);
|
|
}
|
|
|