minix/servers/vfs/mount.c
Thomas Veerman 1efb51b1de VFS: improve crashed FS resource cleanup
When VFS detects that an FS has crashed and tries to clean up
resources, it marks fairly late in the process that a vmnt is not
to be used again (to send requests to). This allows a thread to
become blocked on a vmnt after all blocked threads were stopped, but
before it finds out it shouldn't try to send to that vmnt.
2012-02-22 13:54:35 +00:00

611 lines
17 KiB
C

/* This file performs the MOUNT and UMOUNT system calls.
*
* The entry points into this file are
* do_fsready: perform the FS_READY system call
* do_mount: perform the MOUNT system call
* do_umount: perform the UMOUNT system call
* unmount: unmount a file system
*/
#include "fs.h"
#include <fcntl.h>
#include <string.h>
#include <minix/callnr.h>
#include <minix/com.h>
#include <minix/keymap.h>
#include <minix/const.h>
#include <minix/endpoint.h>
#include <minix/syslib.h>
#include <minix/bitmap.h>
#include <minix/ds.h>
#include <unistd.h>
#include <sys/stat.h>
#include <sys/mount.h>
#include <dirent.h>
#include <assert.h>
#include "file.h"
#include "fproc.h"
#include "dmap.h"
#include <minix/vfsif.h>
#include "vnode.h"
#include "vmnt.h"
#include "path.h"
#include "param.h"
/* Allow the root to be replaced before the first 'real' mount. */
PRIVATE int have_root = 0;
/* Bitmap of in-use "none" pseudo devices. */
PRIVATE bitchunk_t nonedev[BITMAP_CHUNKS(NR_NONEDEVS)] = { 0 };
#define alloc_nonedev(dev) SET_BIT(nonedev, minor(dev) - 1)
#define free_nonedev(dev) UNSET_BIT(nonedev, minor(dev) - 1)
FORWARD _PROTOTYPE( dev_t name_to_dev, (int allow_mountpt,
char path[PATH_MAX]) );
FORWARD _PROTOTYPE( dev_t find_free_nonedev, (void) );
FORWARD _PROTOTYPE( void update_bspec, (dev_t dev, endpoint_t fs_e,
int send_drv_e) );
/*===========================================================================*
* update_bspec *
*===========================================================================*/
PRIVATE void update_bspec(dev_t dev, endpoint_t fs_e, int send_drv_e)
{
/* Update all block special files for a certain device, to use a new FS endpt
* to route raw block I/O requests through.
*/
struct vnode *vp;
struct dmap *dp;
int r, major;
for (vp = &vnode[0]; vp < &vnode[NR_VNODES]; ++vp)
if (vp->v_ref_count > 0 && S_ISBLK(vp->v_mode) && vp->v_sdev == dev) {
vp->v_bfs_e = fs_e;
if (send_drv_e) {
major = major(dev);
if (major < 0 || major >= NR_DEVICES) {
/* Can't update for out-of-range major */
continue;
}
dp = &dmap[major(dev)];
if (dp->dmap_driver == NONE) {
/* Can't update for vanished driver */
printf("VFS: can't send new driver label\n");
continue;
}
if ((r = req_newdriver(fs_e, vp->v_sdev,
dp->dmap_label)) != OK) {
printf("VFS: Failed to send new driver label"
" for moved block special file to %d\n",
fs_e);
}
}
}
}
/*===========================================================================*
* do_fsready *
*===========================================================================*/
PUBLIC int do_fsready()
{
/* deprecated */
return(SUSPEND);
}
/*===========================================================================*
* do_mount *
*===========================================================================*/
PUBLIC int do_mount()
{
/* Perform the mount(name, mfile, mount_flags) system call. */
endpoint_t fs_e;
int r, slot, rdonly, nodev;
char fullpath[PATH_MAX];
char mount_label[LABEL_MAX];
dev_t dev;
/* Only the super-user may do MOUNT. */
if (!super_user) return(EPERM);
/* FS process' endpoint number */
if (m_in.mount_flags & MS_LABEL16) {
/* Get the label from the caller, and ask DS for the endpoint. */
r = sys_datacopy(who_e, (vir_bytes) m_in.fs_label, SELF,
(vir_bytes) mount_label, (phys_bytes) sizeof(mount_label));
if (r != OK) return(r);
mount_label[sizeof(mount_label)-1] = 0;
r = ds_retrieve_label_endpt(mount_label, &fs_e);
if (r != OK) return(r);
} else {
/* Legacy support: get the endpoint from the request itself. */
fs_e = (endpoint_t) m_in.fs_label;
mount_label[0] = 0;
}
/* Sanity check on process number. */
if (isokendpt(fs_e, &slot) != OK) return(EINVAL);
/* Should the file system be mounted read-only? */
rdonly = (m_in.mount_flags & MS_RDONLY);
/* A null string for block special device means don't use a device at all. */
nodev = (m_in.name1_length == 0);
if (!nodev) {
/* If 'name' is not for a block special file, return error. */
if (fetch_name(m_in.name1, m_in.name1_length, M1, fullpath) != OK)
return(err_code);
if ((dev = name_to_dev(FALSE /*allow_mountpt*/, fullpath)) == NO_DEV)
return(err_code);
} else {
/* Find a free pseudo-device as substitute for an actual device. */
if ((dev = find_free_nonedev()) == NO_DEV)
return(err_code);
}
/* Fetch the name of the mountpoint */
if (fetch_name(m_in.name2, m_in.name2_length, M1, fullpath) != OK)
return(err_code);
/* Do the actual job */
return mount_fs(dev, fullpath, fs_e, rdonly, mount_label);
}
/*===========================================================================*
* mount_fs *
*===========================================================================*/
PUBLIC int mount_fs(
dev_t dev,
char mountpoint[PATH_MAX],
endpoint_t fs_e,
int rdonly,
char mount_label[LABEL_MAX] )
{
int rdir, mdir; /* TRUE iff {root|mount} file is dir */
int i, r = OK, found, isroot, mount_root, con_reqs, slot;
struct fproc *tfp, *rfp;
struct dmap *dp;
struct vnode *root_node, *vp = NULL;
struct vmnt *new_vmp, *parent_vmp;
char *label;
struct node_details res;
struct lookup resolve;
/* Look up block device driver label when dev is not a pseudo-device */
label = "";
if (!is_nonedev(dev)) {
/* Get driver process' endpoint */
dp = &dmap[major(dev)];
if (dp->dmap_driver == NONE) {
printf("VFS: no driver for dev %d\n", dev);
return(EINVAL);
}
label = dp->dmap_label;
assert(strlen(label) > 0);
}
/* Scan vmnt table to see if dev already mounted. If not, find a free slot.*/
found = FALSE;
for (i = 0; i < NR_MNTS; ++i) {
if (vmnt[i].m_dev == dev) found = TRUE;
}
if (found) {
return(EBUSY);
} else if ((new_vmp = get_free_vmnt()) == NULL) {
return(ENOMEM);
}
if ((r = lock_vmnt(new_vmp, VMNT_EXCL)) != OK) return(r);
isroot = (strcmp(mountpoint, "/") == 0);
mount_root = (isroot && have_root < 2); /* Root can be mounted twice:
* 1: ramdisk
* 2: boot disk (e.g., harddisk)
*/
if (!mount_root) {
/* Get vnode of mountpoint */
lookup_init(&resolve, mountpoint, PATH_NOFLAGS, &parent_vmp, &vp);
resolve.l_vmnt_lock = VMNT_EXCL;
resolve.l_vnode_lock = VNODE_WRITE;
if ((vp = eat_path(&resolve, fp)) == NULL)
r = err_code;
else if (vp->v_ref_count == 1) {
/*Tell FS on which vnode it is mounted (glue into mount tree)*/
r = req_mountpoint(vp->v_fs_e, vp->v_inode_nr);
} else
r = EBUSY;
if (vp != NULL) {
/* Quickly unlock to allow back calls (from e.g. FUSE) to
* relock */
unlock_vmnt(parent_vmp);
}
if (r != OK) {
if (vp != NULL) {
unlock_vnode(vp);
put_vnode(vp);
}
unlock_vmnt(new_vmp);
return(r);
}
}
/* We'll need a vnode for the root inode */
if ((root_node = get_free_vnode()) == NULL) {
if (vp != NULL) {
unlock_vnode(vp);
put_vnode(vp);
}
unlock_vmnt(new_vmp);
return(err_code);
}
lock_vnode(root_node, VNODE_OPCL);
/* Record process as a system process */
if (isokendpt(fs_e, &slot) != OK) {
if (vp != NULL) {
unlock_vnode(vp);
put_vnode(vp);
}
unlock_vnode(root_node);
unlock_vmnt(new_vmp);
return(EINVAL);
}
rfp = &fproc[slot];
rfp->fp_flags |= FP_SYS_PROC; /* Process is an FS */
/* Store some essential vmnt data first */
new_vmp->m_fs_e = fs_e;
new_vmp->m_dev = dev;
if (rdonly) new_vmp->m_flags |= VMNT_READONLY;
else new_vmp->m_flags &= ~VMNT_READONLY;
/* Tell FS which device to mount */
new_vmp->m_flags |= VMNT_MOUNTING;
r = req_readsuper(fs_e, label, dev, rdonly, isroot, &res, &con_reqs);
new_vmp->m_flags &= ~VMNT_MOUNTING;
if (r != OK) {
mark_vmnt_free(new_vmp);
unlock_vnode(root_node);
if (vp != NULL) {
unlock_vnode(vp);
put_vnode(vp);
}
unlock_vmnt(new_vmp);
return(r);
}
lock_bsf();
/* Fill in root node's fields */
root_node->v_fs_e = res.fs_e;
root_node->v_inode_nr = res.inode_nr;
root_node->v_mode = res.fmode;
root_node->v_uid = res.uid;
root_node->v_gid = res.gid;
root_node->v_size = res.fsize;
root_node->v_sdev = NO_DEV;
root_node->v_fs_count = 1;
root_node->v_ref_count = 1;
/* Root node is indeed on the partition */
root_node->v_vmnt = new_vmp;
root_node->v_dev = new_vmp->m_dev;
if (con_reqs == 0)
new_vmp->m_comm.c_max_reqs = 1; /* Default if FS doesn't tell us */
else
new_vmp->m_comm.c_max_reqs = con_reqs;
new_vmp->m_comm.c_cur_reqs = 0;
if (mount_root) {
/* Superblock and root node already read.
* Nothing else can go wrong. Perform the mount. */
new_vmp->m_root_node = root_node;
new_vmp->m_mounted_on = NULL;
strcpy(new_vmp->m_label, mount_label);
if (is_nonedev(dev)) alloc_nonedev(dev);
update_bspec(dev, fs_e, 0 /* Don't send new driver endpoint */);
ROOT_DEV = dev;
ROOT_FS_E = fs_e;
/* Replace all root and working directories */
for (i = 0, tfp = fproc; i < NR_PROCS; i++, tfp++) {
if (tfp->fp_pid == PID_FREE)
continue;
#define MAKEROOT(what) { \
if (what) put_vnode(what); \
dup_vnode(root_node); \
what = root_node; \
}
MAKEROOT(tfp->fp_rd);
MAKEROOT(tfp->fp_wd);
}
unlock_vnode(root_node);
unlock_vmnt(new_vmp);
have_root++; /* We have a (new) root */
unlock_bsf();
return(OK);
}
/* File types may not conflict. */
mdir = ((vp->v_mode & I_TYPE) == I_DIRECTORY); /*TRUE iff dir*/
rdir = ((root_node->v_mode & I_TYPE) == I_DIRECTORY);
if (!mdir && rdir) r = EISDIR;
/* If error, return the super block and both inodes; release the vmnt. */
if (r != OK) {
unlock_vnode(vp);
unlock_vnode(root_node);
mark_vmnt_free(new_vmp);
unlock_vmnt(new_vmp);
put_vnode(vp);
put_vnode(root_node);
unlock_bsf();
return(r);
}
/* Nothing else can go wrong. Perform the mount. */
new_vmp->m_mounted_on = vp;
new_vmp->m_root_node = root_node;
strcpy(new_vmp->m_label, mount_label);
/* Allocate the pseudo device that was found, if not using a real device. */
if (is_nonedev(dev)) alloc_nonedev(dev);
/* The new FS will handle block I/O requests for its device now. */
if (!(new_vmp->m_flags & VMNT_FORCEROOTBSF))
update_bspec(dev, fs_e, 0 /* Don't send new driver endpoint */);
unlock_vnode(vp);
unlock_vnode(root_node);
unlock_vmnt(new_vmp);
unlock_bsf();
return(OK);
}
/*===========================================================================*
* mount_pfs *
*===========================================================================*/
PUBLIC void mount_pfs(void)
{
/* Mount the Pipe File Server. It's not really mounted onto the file system,
but it's necessary it has a vmnt entry to make locking easier */
dev_t dev;
struct vmnt *vmp;
struct fproc *rfp;
if ((dev = find_free_nonedev()) == NO_DEV)
panic("VFS: no nonedev to initialize PFS");
if ((vmp = get_free_vmnt()) == NULL)
panic("VFS: no vmnt to initialize PFS");
alloc_nonedev(dev);
vmp->m_dev = dev;
vmp->m_fs_e = PFS_PROC_NR;
strcpy(vmp->m_label, "pfs");
rfp = &fproc[_ENDPOINT_P(PFS_PROC_NR)];
rfp->fp_flags |= FP_SYS_PROC; /* PFS is a driver and an FS */
}
/*===========================================================================*
* do_umount *
*===========================================================================*/
PUBLIC int do_umount(void)
{
/* Perform the umount(name) system call. */
char label[LABEL_MAX];
dev_t dev;
int r;
char fullpath[PATH_MAX];
/* Only the super-user may do umount. */
if (!super_user) return(EPERM);
/* If 'name' is not for a block special file or mountpoint, return error. */
if (fetch_name(m_in.name, m_in.name_length, M3, fullpath) != OK)
return(err_code);
if ((dev = name_to_dev(TRUE /*allow_mountpt*/, fullpath)) == NO_DEV)
return(err_code);
if ((r = unmount(dev, label)) != OK) return(r);
/* Return the label of the mounted file system, so that the caller
* can shut down the corresponding server process.
*/
if (strlen(label) >= M3_LONG_STRING) /* should never evaluate to true */
label[M3_LONG_STRING-1] = 0;
strcpy(m_out.umount_label, label);
return(OK);
}
/*===========================================================================*
* unmount *
*===========================================================================*/
PUBLIC int unmount(
dev_t dev, /* block-special device */
char *label /* buffer to retrieve label, or NULL */
)
{
struct vnode *vp;
struct vmnt *vmp_i = NULL, *vmp = NULL;
int count, locks, r;
/* Find vmnt that is to be unmounted */
for (vmp_i = &vmnt[0]; vmp_i < &vmnt[NR_MNTS]; ++vmp_i) {
if (vmp_i->m_dev == dev) {
if(vmp) panic("device mounted more than once: %d", dev);
vmp = vmp_i;
}
}
/* Did we find the vmnt (i.e., was dev a mounted device)? */
if(!vmp) return(EINVAL);
if ((r = lock_vmnt(vmp, VMNT_EXCL)) != OK) return(r);
/* See if the mounted device is busy. Only 1 vnode using it should be
* open -- the root vnode -- and that inode only 1 time. */
locks = count = 0;
for (vp = &vnode[0]; vp < &vnode[NR_VNODES]; vp++)
if (vp->v_ref_count > 0 && vp->v_dev == dev) {
count += vp->v_ref_count;
if (is_vnode_locked(vp)) locks++;
}
if (count > 1 || locks > 1 || tll_haspendinglock(&vmp->m_lock)) {
unlock_vmnt(vmp);
return(EBUSY); /* can't umount a busy file system */
}
/* Tell FS to drop all inode references for root inode except 1. */
vnode_clean_refs(vmp->m_root_node);
if (vmp->m_mounted_on) {
put_vnode(vmp->m_mounted_on);
vmp->m_mounted_on = NULL;
}
vmp->m_comm.c_max_reqs = 1; /* Force max concurrent reqs to just one, so
* we won't send any messages after the
* unmount request */
/* Tell FS to unmount */
if ((r = req_unmount(vmp->m_fs_e)) != OK) /* Not recoverable. */
printf("VFS: ignoring failed umount attempt FS endpoint: %d (%d)\n",
vmp->m_fs_e, r);
if (is_nonedev(vmp->m_dev)) free_nonedev(vmp->m_dev);
if (label != NULL) strcpy(label, vmp->m_label);
if (vmp->m_root_node) { /* PFS lacks a root node */
vmp->m_root_node->v_ref_count = 0;
vmp->m_root_node->v_fs_count = 0;
vmp->m_root_node->v_sdev = NO_DEV;
vmp->m_root_node = NULL;
}
mark_vmnt_free(vmp);
unlock_vmnt(vmp);
/* The root FS will handle block I/O requests for this device now. */
lock_bsf();
update_bspec(dev, ROOT_FS_E, 1 /* send new driver endpoint */);
unlock_bsf();
return(OK);
}
/*===========================================================================*
* unmount_all *
*===========================================================================*/
PUBLIC void unmount_all(void)
{
/* Unmount all filesystems. File systems are mounted on other file systems,
* so you have to pull off the loose bits repeatedly to get it all undone.
*/
int i;
struct vmnt *vmp;
/* Now unmount the rest */
for (i = 0; i < NR_MNTS; i++) {
/* Unmount at least one. */
for (vmp = &vmnt[0]; vmp < &vmnt[NR_MNTS]; vmp++) {
if (vmp->m_dev != NO_DEV)
unmount(vmp->m_dev, NULL);
}
}
check_vnode_locks();
check_vmnt_locks();
check_filp_locks();
check_bsf_lock();
}
/*===========================================================================*
* name_to_dev *
*===========================================================================*/
PRIVATE dev_t name_to_dev(int allow_mountpt, char path[PATH_MAX])
{
/* Convert the block special file in 'user_fullpath' to a device number.
* If the given path is not a block special file, but 'allow_mountpt' is set
* and the path is the root node of a mounted file system, return that device
* number. In all other cases, return NO_DEV and an error code in 'err_code'.
*/
dev_t dev;
struct vnode *vp;
struct vmnt *vmp;
struct lookup resolve;
lookup_init(&resolve, path, PATH_NOFLAGS, &vmp, &vp);
resolve.l_vmnt_lock = VMNT_READ;
resolve.l_vnode_lock = VNODE_READ;
/* Request lookup */
if ((vp = eat_path(&resolve, fp)) == NULL) return(NO_DEV);
if ((vp->v_mode & I_TYPE) == I_BLOCK_SPECIAL) {
dev = vp->v_sdev;
} else if (allow_mountpt && vp->v_vmnt->m_root_node == vp) {
dev = vp->v_dev;
} else {
err_code = ENOTBLK;
dev = NO_DEV;
}
unlock_vnode(vp);
unlock_vmnt(vmp);
put_vnode(vp);
return(dev);
}
/*===========================================================================*
* is_nonedev *
*===========================================================================*/
PUBLIC int is_nonedev(dev_t dev)
{
/* Return whether the given device is a "none" pseudo device.
*/
return (major(dev) == NONE_MAJOR &&
minor(dev) > 0 && minor(dev) <= NR_NONEDEVS);
}
/*===========================================================================*
* find_free_nonedev *
*===========================================================================*/
PRIVATE dev_t find_free_nonedev(void)
{
/* Find a free "none" pseudo device. Do not allocate it yet.
*/
int i;
for (i = 0; i < NR_NONEDEVS; i++)
if (!GET_BIT(nonedev, i))
return makedev(NONE_MAJOR, i + 1);
err_code = EMFILE;
return NO_DEV;
}