vfs_subr.c

来自「基于组件方式开发操作系统的OSKIT源代码」· C语言 代码 · 共 2,733 行 · 第 1/5 页

	if (argp->ex_flags & MNT_DELEXPORT) {
		if (mp->mnt_flag & MNT_EXPUBLIC) {
			vfs_setpublicfs(NULL, NULL, NULL);
			mp->mnt_flag &= ~MNT_EXPUBLIC;
		}
		vfs_free_addrlist(nep);
		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
	}
	if (argp->ex_flags & MNT_EXPORTED) {
		if (argp->ex_flags & MNT_EXPUBLIC) {
			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
				return (error);
			mp->mnt_flag |= MNT_EXPUBLIC;
		}
		if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0)
			return (error);
		mp->mnt_flag |= MNT_EXPORTED;
	}
	return (0);
}

/*
 * Set the publicly exported filesystem (WebNFS). Currently, only
 * one public filesystem is possible in the spec (RFC 2054 and 2055)
 */
int
vfs_setpublicfs(mp, nep, argp)
	struct mount *mp;
	struct netexport *nep;
	struct export_args *argp;
{
	int error;
	struct vnode *rvp;
	char *cp;

	/*
	 * mp == NULL -> invalidate the current info, the FS is
	 * no longer exported. May be called from either vfs_export
	 * or unmount, so check if it hasn't already been done.
	 */
	if (mp == NULL) {
		if (nfs_pub.np_valid) {
			nfs_pub.np_valid = 0;
			if (nfs_pub.np_index != NULL) {
				FREE(nfs_pub.np_index, M_TEMP);
				nfs_pub.np_index = NULL;
			}
		}
		return (0);
	}

	/*
	 * Only one allowed at a time.
	 */
	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
		return (EBUSY);

	/*
	 * Get real filehandle for root of exported FS.
	 */
	memset((caddr_t)&nfs_pub.np_handle, 0, sizeof(nfs_pub.np_handle));
	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;

	if ((error = VFS_ROOT(mp, &rvp)))
		return (error);

	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
		return (error);

	vput(rvp);

	/*
	 * If an indexfile was specified, pull it in.
	 */
	if (argp->ex_indexfile != NULL) {
		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
		    M_WAITOK);
		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
		    MAXNAMLEN, (size_t *)0);
		if (!error) {
			/*
			 * Check for illegal filenames.
			 */
			for (cp = nfs_pub.np_index; *cp; cp++) {
				if (*cp == '/') {
					error = EINVAL;
					break;
				}
			}
		}
		if (error) {
			FREE(nfs_pub.np_index, M_TEMP);
			return (error);
		}
	}

	nfs_pub.np_mount = mp;
	nfs_pub.np_valid = 1;
	return (0);
}

struct netcred *
vfs_export_lookup(mp, nep, nam)
	struct mount *mp;
	struct netexport *nep;
	struct mbuf *nam;
{
	struct netcred *np;
	struct radix_node_head *rnh;
	struct sockaddr *saddr;

	np = NULL;
	if (mp->mnt_flag & MNT_EXPORTED) {
		/*
		 * Lookup in the export list first.
		 */
		if (nam != NULL) {
			saddr = mtod(nam, struct sockaddr *);
			rnh = nep->ne_rtable[saddr->sa_family];
			if (rnh != NULL) {
				np = (struct netcred *)
					(*rnh->rnh_matchaddr)((caddr_t)saddr,
							      rnh);
				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
					np = NULL;
			}
		}
		/*
		 * If no address match, use the default if it exists.
		 */
		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
			np = &nep->ne_defexported;
	}
	return (np);
}

/*
 * Do the usual access checking.
 * file_mode, uid and gid are from the vnode in question,
 * while acc_mode and cred are from the VOP_ACCESS parameter list
 */
int
vaccess(type, file_mode, uid, gid, acc_mode, cred)
	enum vtype type;
	mode_t file_mode;
	uid_t uid;
	gid_t gid;
	mode_t acc_mode;
	struct ucred *cred;
{
	mode_t mask;
	
	/*
	 * Super-user always gets read/write access, but execute access depends
	 * on at least one execute bit being set.
	 */
	if (cred->cr_uid == 0) {
		if ((acc_mode & VEXEC) && type != VDIR &&
		    (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
			return (EACCES);
		return (0);
	}
	
	mask = 0;
	
	/* Otherwise, check the owner. */
	if (cred->cr_uid == uid) {
		if (acc_mode & VEXEC)
			mask |= S_IXUSR;
		if (acc_mode & VREAD)
			mask |= S_IRUSR;
		if (acc_mode & VWRITE)
			mask |= S_IWUSR;
		return ((file_mode & mask) == mask ? 0 : EACCES);
	}
	
	/* Otherwise, check the groups. */
	if (cred->cr_gid == gid || groupmember(gid, cred)) {
		if (acc_mode & VEXEC)
			mask |= S_IXGRP;
		if (acc_mode & VREAD)
			mask |= S_IRGRP;
		if (acc_mode & VWRITE)
			mask |= S_IWGRP;
		return ((file_mode & mask) == mask ? 0 : EACCES);
	}
	
	/* Otherwise, check everyone else. */
	if (acc_mode & VEXEC)
		mask |= S_IXOTH;
	if (acc_mode & VREAD)
		mask |= S_IROTH;
	if (acc_mode & VWRITE)
		mask |= S_IWOTH;
	return ((file_mode & mask) == mask ? 0 : EACCES);
}

/*
 * Unmount all file systems.
 * We traverse the list in reverse order under the assumption that doing so
 * will avoid needing to worry about dependencies.
 */
void
vfs_unmountall(p)
	struct proc *p;
{
	struct mount *mp, *nmp;
	int allerror, error;

	for (allerror = 0,
	     mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
		nmp = mp->mnt_list.cqe_prev;
#ifdef DEBUG
		printf("unmounting %s (%s)...\n",
		    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname);
#endif
		if (vfs_busy(mp, 0, 0))
			continue;
		if ((error = dounmount(mp, MNT_FORCE, p)) != 0) {
			printf("unmount of %s failed with error %d\n",
			    mp->mnt_stat.f_mntonname, error);
			allerror = 1;
		}
	}
	if (allerror)
		printf("WARNING: some file systems would not unmount\n");
}

/*
 * Sync and unmount file systems before shutting down.
 */
void
vfs_shutdown()
{
	struct buf *bp;
	int iter, nbusy, nbusy_prev = 0, dcount, s;
	struct proc *p = curproc;

	/* XXX we're certainly not running in proc0's context! */
	if (p == NULL)
		p = &proc0;
	
	printf("syncing disks... ");

	/* remove user process from run queue */
	suspendsched();
	(void) spl0();

	/* avoid coming back this way again if we panic. */
	doing_shutdown = 1;

	sys_sync(p, NULL, NULL);

	/* Wait for sync to finish. */
	dcount = 10000;
	for (iter = 0; iter < 20;) {
		nbusy = 0;
		for (bp = &buf[nbuf]; --bp >= buf; ) {
			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
				nbusy++;
			/*
			 * With soft updates, some buffers that are
			 * written will be remarked as dirty until other
			 * buffers are written.
			 */
			if (bp->b_vp && bp->b_vp->v_mount
			    && (bp->b_vp->v_mount->mnt_flag & MNT_SOFTDEP)
			    && (bp->b_flags & B_DELWRI)) {
				s = splbio();
				bremfree(bp);
				bp->b_flags |= B_BUSY;
				splx(s);
				nbusy++;
				bawrite(bp);
				if (dcount-- <= 0) {
					printf("softdep ");
					goto fail;
				}
			}
		}
		if (nbusy == 0)
			break;
		if (nbusy_prev == 0)
			nbusy_prev = nbusy;
		printf("%d ", nbusy);
		tsleep(&nbusy, PRIBIO, "bflush",
		    (iter == 0) ? 1 : hz / 25 * iter);
		if (nbusy >= nbusy_prev) /* we didn't flush anything */
			iter++;
		else
			nbusy_prev = nbusy;
	}
	if (nbusy) {
fail:
#if defined(DEBUG) || defined(DEBUG_HALT_BUSY)
		printf("giving up\nPrinting vnodes for busy buffers\n");
		for (bp = &buf[nbuf]; --bp >= buf; )
			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
				vprint(NULL, bp->b_vp);

#if defined(DDB) && defined(DEBUG_HALT_BUSY)
		Debugger();
#endif

#else  /* defined(DEBUG) || defined(DEBUG_HALT_BUSY) */
		printf("giving up\n");
#endif /* defined(DEBUG) || defined(DEBUG_HALT_BUSY) */
		return;
	} else
		printf("done\n");

	/*
	 * If we've panic'd, don't make the situation potentially
	 * worse by unmounting the file systems.
	 */
	if (panicstr != NULL)
		return;

	/* Release inodes held by texts before update. */
#ifdef notdef
	vnshutdown();
#endif
	/* Unmount file systems. */
	vfs_unmountall(p);
}

/*
 * Mount the root file system.  If the operator didn't specify a
 * file system to use, try all possible file systems until one
 * succeeds.
 */
int
vfs_mountroot()
{
	extern int (*mountroot) __P((void));
	struct vfsops *v;

	if (root_device == NULL)
		panic("vfs_mountroot: root device unknown");

	switch (root_device->dv_class) {
	case DV_IFNET:
		if (rootdev != NODEV)
			panic("vfs_mountroot: rootdev set for DV_IFNET");
		break;

	case DV_DISK:
		if (rootdev == NODEV)
			panic("vfs_mountroot: rootdev not set for DV_DISK");
		break;

	default:
		printf("%s: inappropriate for root file system\n",
		    root_device->dv_xname);
		return (ENODEV);
	}

	/*
	 * If user specified a file system, use it.
	 */
	if (mountroot != NULL)
		return ((*mountroot)());

	/*
	 * Try each file system currently configured into the kernel.
	 */
	for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) {
		if (v->vfs_mountroot == NULL)
			continue;
#ifdef DEBUG
		printf("mountroot: trying %s...\n", v->vfs_name);
#endif
		if ((*v->vfs_mountroot)() == 0) {
			printf("root file system type: %s\n", v->vfs_name);
			break;
		}
	}

	if (v == NULL) {
		printf("no file system for %s", root_device->dv_xname);
		if (root_device->dv_class == DV_DISK)
			printf(" (dev 0x%x)", rootdev);
		printf("\n");
		return (EFTYPE);
	}
	return (0);
}

/*
 * Given a file system name, look up the vfsops for that
 * file system, or return NULL if file system isn't present
 * in the kernel.
 */
struct vfsops *
vfs_getopsbyname(name)
	const char *name;
{
	struct vfsops *v;

	for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) {
		if (strcmp(v->vfs_name, name) == 0)
			break;
	}

	return (v);
}

/*
 * Establish a file system and initialize it.
 */
int
vfs_attach(vfs)
	struct vfsops *vfs;
{
	struct vfsops *v;
	int error = 0;


	/*
	 * Make sure this file system doesn't already exist.
	 */
	for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) {
		if (strcmp(vfs->vfs_name, v->vfs_name) == 0) {
			error = EEXIST;
			goto out;
		}
	}

	/*
	 * Initialize the vnode operations for this file system.
	 */
	vfs_opv_init(vfs->vfs_opv_descs);

	/*
	 * Now initialize the file system itself.
	 */
	(*vfs->vfs_init)();

	/*
	 * ...and link it into the kernel's list.
	 */
	LIST_INSERT_HEAD(&vfs_list, vfs, vfs_list);

	/*
	 * Sanity: make sure the reference count is 0.
	 */
	vfs->vfs_refcount = 0;

 out:
	return (error);
}

/*
 * Remove a file system from the kernel.
 */
int
vfs_detach(vfs)
	struct vfsops *vfs;
{
	struct vfsops *v;

	/*
	 * Make sure no one is using the filesystem.
	 */
	if (vfs->vfs_refcount != 0)
		return (EBUSY);

	/*
	 * ...and remove it from the kernel's list.
	 */
	for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) {
		if (v == vfs) {
			LIST_REMOVE(v, vfs_list);
			break;
		}
	}

	if (v == NULL)
		return (ESRCH);

	/*
	 * Now run the file system-specific cleanups.
	 */
	(*vfs->vfs_done)();

	/*
	 * Free the vnode operations vector.
	 */
	vfs_opv_free(vfs->vfs_opv_descs);
	return (0);
}

#ifdef DDB
const char buf_flagbits[] =
	"\20\1AGE\2NEEDCOMMIT\3ASYNC\4BAD\5BUSY\6SCANNED\7CALL\10DELWRI"
	"\11DIRTY\12DONE\13EINTR\14ERROR\15GATHERED\16INVAL\17LOCKED\20NOCACHE"
	"\21ORDERED\22CACHE\23PHYS\24RAW\25READ\26TAPE\30WANTED"
	"\32XXX\33VFLUSH";

void
vfs_buf_print(bp, full, pr)
	struct buf *bp;
	int full;
	void (*pr) __P((const char *, ...));
{
	char buf[1024];

	(*pr)("  vp %p lblkno 0x%x blkno 0x%x dev 0x%x\n",
		  bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_dev);

	bitmask_snprintf(bp->b_flags, buf_flagbits, buf, sizeof(buf));
	(*pr)("  error %d flags 0x%s\n", bp->b_error, buf);

	(*pr)("  bufsize 0x%x bcount 0x%x resid 0x%x\n",
		  bp->b_bufsize, bp->b_bcount, bp->b_resid);
	(*pr)("  data %p saveaddr %p dep %p\n",
		  bp->b_data, bp->b_saveaddr, LIST_FIRST(&bp->b_dep));
	(*pr)("  iodone %p\n", bp->b_iodone);
}


const char vnode_flagbits[] =
	"\20\1ROOT\2TEXT\3SYSTEM\4ISTTY\11XLOCK\12XWANT\13BWAIT\14ALIASED"
	"\15DIROP\17DIRTY";

const char *vnode_types[] = {
	"VNON",
	"VREG",
	"VDIR",
	"VBLK",
	"VCHR",
	"VLNK",
	"VSOCK",
	"VFIFO",
	"VBAD",
};

const char *vnode_tags[] = {
	"VT_NON",
	"VT_UFS",
	"VT_NFS",
	"VT_MFS",
	"VT_MSDOSFS",
	"VT_LFS",
	"V