afs_ops.c

早期freebsd实现

/*
 * Copyright (c) 1990 Jan-Simon Pendry
 * Copyright (c) 1990 Imperial College of Science, Technology & Medicine
 * Copyright (c) 1990, 1993
 *	The Regents of the University of California.  All rights reserved.
 *
 * This code is derived from software contributed to Berkeley by
 * Jan-Simon Pendry at Imperial College, London.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *	This product includes software developed by the University of
 *	California, Berkeley and its contributors.
 * 4. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 *	@(#)afs_ops.c	8.1 (Berkeley) 6/6/93
 *
 * $Id: afs_ops.c,v 5.2.2.4 1992/05/31 16:36:36 jsp Exp $
 *
 */

#include "am.h"

#define NFS
#define NFSCLIENT

#include <sys/stat.h>
#ifdef NFS_3
typedef nfs_fh fhandle_t;
#endif /* NFS_3 */
#ifdef NFS_HDR
#include NFS_HDR
#endif /* NFS_HDR */
#include <sys/mount.h>
#include "mount.h"

/*
 * Automount file system
 * Direct file system
 * Root file system
 * Top-level file system
 */

/*
 * Interval between forced retries of a mount.
 */
#define RETRY_INTERVAL	2

/*
 * AFS needs nothing in particular.
 */
static char *afs_match P((am_opts *fo));
static char *afs_match(fo)
am_opts *fo;
{
	char *p = fo->opt_rfs;
	if (!fo->opt_rfs) {
		plog(XLOG_USER, "auto: no mount point named (rfs:=)");
		return 0;
	}
	if (!fo->opt_fs) {
		plog(XLOG_USER, "auto: no map named (fs:=)");
		return 0;
	}
	/*
	 * Swap round fs:= and rfs:= options
	 * ... historical (jsp)
	 */
	fo->opt_rfs = fo->opt_fs;
	fo->opt_fs = p;
	/*
	 * mtab entry turns out to be the name of the mount map
	 */
	return strdup(fo->opt_rfs ? fo->opt_rfs : ".");
}

/*
 * Mount an automounter directory.
 * The automounter is connected into the system
 * as a user-level NFS server.  mount_toplvl constructs
 * the necessary NFS parameters to be given to the
 * kernel so that it will talk back to us.
 */
static int mount_toplvl P((char *dir, char *opts));
static int mount_toplvl(dir, opts)
char *dir;
char *opts;
{
	struct nfs_args nfs_args;
	struct mntent mnt;
	int retry;
	struct sockaddr_in sin;
	unsigned short port;
	int flags;
	extern nfs_fh *root_fh();
	nfs_fh *fhp;
	char fs_hostname[MAXHOSTNAMELEN+MAXPATHLEN+1];

	MTYPE_TYPE type = MOUNT_TYPE_NFS;

	bzero((voidp) &nfs_args, sizeof(nfs_args));	/* Paranoid */

	mnt.mnt_dir = dir;
	mnt.mnt_fsname = pid_fsname;
	mnt.mnt_type = MNTTYPE_AUTO;
	mnt.mnt_opts = opts;
	mnt.mnt_freq = 0;
	mnt.mnt_passno = 0;

	retry = hasmntval(&mnt, "retry");
	if (retry <= 0)
		retry = 2;	/* XXX */

	/*
	 * get fhandle of remote path for automount point
	 */
	
	fhp = root_fh(dir);
	if (!fhp) {
		plog(XLOG_FATAL, "Can't find root file handle for %s", dir);
		return EINVAL;
	}

	NFS_FH_DREF(nfs_args.fh, (NFS_FH_TYPE) fhp);

	/*
	 * Create sockaddr to point to the local machine.  127.0.0.1
	 * is not used since that will not work in HP-UX clusters and
	 * this is no more expensive.
	 */
	bzero((voidp) &sin, sizeof(sin));
	sin.sin_family = AF_INET;
	sin.sin_addr = myipaddr;
	if (port = hasmntval(&mnt, "port")) {
		sin.sin_port = htons(port);
	} else {
		plog(XLOG_ERROR, "no port number specified for %s", dir);
		return EINVAL;
	}

	/*
	 * set mount args
	 */
	NFS_SA_DREF(nfs_args, &sin);

	/*
	 * Make a ``hostname'' string for the kernel
	 */
#ifndef HOSTNAMESZ
#define	SHORT_MOUNT_NAME
#endif /* HOSTNAMESZ */
#ifdef SHORT_MOUNT_NAME
	sprintf(fs_hostname, "amd:%d", foreground ? mypid : getppid());
#else
	sprintf(fs_hostname, "pid%d@%s:%s", foreground ? mypid : getppid(), hostname, dir);
#endif /* SHORT_MOUNT_NAME */
	nfs_args.hostname = fs_hostname;
	nfs_args.flags |= NFSMNT_HOSTNAME;
#ifdef HOSTNAMESZ
	/*
	 * Most kernels have a name length restriction.
	 */
	if (strlen(fs_hostname) >= HOSTNAMESZ)
		strcpy(fs_hostname + HOSTNAMESZ - 3, "..");
#endif /* HOSTNAMESZ */

#ifdef NFSMNT_DUMBTIMR
	nfs_args.flags |= NFSMNT_DUMBTIMR;
	plog(XLOG_INFO, "defeating nfs window computation");
#endif

	/*
	 * Parse a subset of the standard nfs options.  The
	 * others are probably irrelevant for this application
	 */
	if (nfs_args.timeo = hasmntval(&mnt, "timeo"))
		nfs_args.flags |= NFSMNT_TIMEO;

	if (nfs_args.retrans = hasmntval(&mnt, "retrans"))
		nfs_args.flags |= NFSMNT_RETRANS;

#ifdef NFSMNT_BIODS
	if (nfs_args.biods = hasmntval(&mnt, "biods"))
		nfs_args.flags |= NFSMNT_BIODS;

#endif /* NFSMNT_BIODS */

#if defined(NFSMNT_ACREGMIN) && defined(NFSMNT_ACREGMAX)
	/*
	 * Don't cache attributes - they are changing under
	 * the kernel's feet...
	 */
	nfs_args.acregmin = nfs_args.acregmax = 1;
	nfs_args.flags |= NFSMNT_ACREGMIN|NFSMNT_ACREGMAX;
#endif /* defined(NFSMNT_ACREGMIN) && defined(NFSMNT_ACREGMAX) */
	/*
	 * These two are constructed internally by the calling routine
	 */
	if (hasmntopt(&mnt, MNTOPT_SOFT) != NULL)
		nfs_args.flags |= NFSMNT_SOFT;

#ifdef MNTOPT_INTR
	if (hasmntopt(&mnt, MNTOPT_INTR) != NULL)
		nfs_args.flags |= NFSMNT_INT;
#endif /* MNTOPT_INTR */

	flags = compute_mount_flags(&mnt);
#ifdef ULTRIX_HACK
	nfs_args.gfs_flags = flags;
	flags &= M_RDONLY;
	if (flags & M_RDONLY)
		nfs_args.flags |= NFSMNT_RONLY;
#endif /* ULTRIX_HACK */
	return mount_fs(&mnt, flags, (caddr_t) &nfs_args, retry, type);
}

static void afs_mkcacheref P((mntfs *mf));
static void afs_mkcacheref(mf)
mntfs *mf;
{
	/*
	 * Build a new map cache for this node, or re-use
	 * an existing cache for the same map.
	 */
	char *cache;
	if (mf->mf_fo && mf->mf_fo->opt_cache)
	  	cache = mf->mf_fo->opt_cache;
	else
	  	cache = "none";
	mf->mf_private = (voidp) mapc_find(mf->mf_info, cache);
	mf->mf_prfree = mapc_free;
}

/*
 * Mount the root...
 */
static int root_mount P((am_node *mp));
static int root_mount(mp)
am_node *mp;
{
	mntfs *mf = mp->am_mnt;

	mf->mf_mount = strealloc(mf->mf_mount, pid_fsname);
	mf->mf_private = (voidp) mapc_find(mf->mf_info, "");
	mf->mf_prfree = mapc_free;

	return 0;
}

/*
 * Mount a sub-mount
 */
static int afs_mount P((am_node *mp));
static int afs_mount(mp)
am_node *mp;
{
	mntfs *mf = mp->am_mnt;

	/*
	 * Pseudo-directories are used to provide some structure
	 * to the automounted directories instead
	 * of putting them all in the top-level automount directory.
	 *
	 * Here, just increment the parent's link count.
	 */
	mp->am_parent->am_fattr.nlink++;
	/*
	 * Info field of . means use parent's info field.
	 * Historical - not documented.
	 */
	if (mf->mf_info[0] == '.' && mf->mf_info[1] == '\0')
		mf->mf_info = strealloc(mf->mf_info, mp->am_parent->am_mnt->mf_info);
	/*
	 * Compute prefix:
	 *
	 * If there is an option prefix then use that else
	 * If the parent had a prefix then use that with name
	 *	of this node appended else
	 * Use the name of this node.
	 *
	 * That means if you want no prefix you must say so
	 * in the map.
	 */
	if (mf->mf_fo->opt_pref) {
		/*
		 * the prefix specified as an option
		 */
		mp->am_pref = strdup(mf->mf_fo->opt_pref);
	} else {
		/*
		 * else the parent's prefix
		 * followed by the name
		 * followed by /
		 */
		char *ppref = mp->am_parent->am_pref;
		if (ppref == 0)
			ppref = "";
		mp->am_pref = str3cat((char *) 0, ppref, mp->am_name, "/");
	}

	/*
	 * Attach a map cache
	 */
	afs_mkcacheref(mf);

	return 0;
}

/*
 * Mount the top-level
 */
static int toplvl_mount P((am_node *mp));
static int toplvl_mount(mp)
am_node *mp;
{
	mntfs *mf = mp->am_mnt;
	struct stat stb;
	char opts[256];
	int error;
	char *mnttype;

	/*
	 * Mounting the automounter.
	 * Make sure the mount directory exists, construct
	 * the mount options and call the mount_toplvl routine.
	 */

	if (stat(mp->am_path, &stb) < 0) {
		return errno;
	} else if ((stb.st_mode & S_IFMT) != S_IFDIR) {
		plog(XLOG_WARNING, "%s is not a directory", mp->am_path);
		return ENOTDIR;
	}

	if (mf->mf_ops == &toplvl_ops) mnttype = "indirect";
	else if (mf->mf_ops == &dfs_ops) mnttype = "direct";
#ifdef HAS_UNION_FS
	else if (mf->mf_ops == &union_ops) mnttype = "union";
#endif
	else mnttype = "auto";

	/*
	 * Construct some mount options
	 */
	sprintf(opts,
#ifdef MNTOPT_INTR
		"%s,%s,%s=%d,%s=%d,%s=%d,%s",
		MNTOPT_INTR,
#else
		"%s,%s=%d,%s=%d,%s=%d,%s",
#endif /* MNTOPT_INTR */
		"rw",
		"port", nfs_port,
		"timeo", afs_timeo,
		"retrans", afs_retrans,
		mnttype);

	error = mount_toplvl(mf->mf_mount, opts);
	if (error) {
		errno = error;
		plog(XLOG_FATAL, "mount_toplvl: %m");
		return error;
	}

	return 0;
}

static void toplvl_mounted P((mntfs *mf));
static void toplvl_mounted(mf)
mntfs *mf;
{
	afs_mkcacheref(mf);
}

#ifdef HAS_UNION_FS
/*
 * Create a reference to a union'ed entry
 */
static int create_union_node P((char *dir, voidp arg));
static int create_union_node(dir, arg)
char *dir;
voidp arg;
{
	if (strcmp(dir, "/defaults") != 0) {
		int error = 0;
		(void) toplvl_ops.lookuppn(arg, dir, &error, VLOOK_CREATE);
		if (error > 0) {
			errno = error; /* XXX */
			plog(XLOG_ERROR, "Could not mount %s: %m", dir);
		}
		return error;
	}
	return 0;
}

static void union_mounted P((mntfs *mf));
static void union_mounted(mf)
mntfs *mf;
{
	int i;
	
	afs_mkcacheref(mf);

	/*
	 * Having made the union mount point,
	 * populate all the entries...
	 */
	for (i = 0; i <= last_used_map; i++) {
		am_node *mp = exported_ap[i];
		if (mp && mp->am_mnt == mf) {
			/* return value from create_union_node is ignored by mapc_keyiter */
			(void) mapc_keyiter((mnt_map *) mp->am_mnt->mf_private,
				(void (*)P((char*,void*))) create_union_node, mp);
			break;
		}
	}

#ifdef notdef
	/*
	 * would be nice to flush most of the cache, but we need to
	 * keep the wildcard and /defaults entries...
	 */
	mapc_free(mf->mf_private);
	mf->mf_private = (voidp) mapc_find(mf->mf_info, "inc");
/*	mapc_add_kv(mf->mf_private, strdup("/defaults"),
		strdup("type:=link;opts:=nounmount;sublink:=${key}")); */
#endif
}
#endif /* HAS_UNION_FS */

/*
 * Unmount an automount sub-node
 */
static int afs_umount P((am_node *mp));
static int afs_umount(mp)
am_node *mp;
{
	return 0;
}

/*
 * Unmount a top-level automount node
 */
static int toplvl_umount P((am_node *mp));
static int toplvl_umount(mp)
am_node *mp;
{
	int error;

	struct stat stb;
again:
	/*
	 * The lstat is needed if this mount is type=direct.
	 * When that happens, the kernel cache gets confused
	 * between the underlying type (dir) and the mounted
	 * type (link) and so needs to be re-synced before
	 * the unmount.  This is all because the unmount system
	 * call follows links and so can't actually unmount
	 * a link (stupid!).  It was noted that doing an ls -ld
	 * of the mount point to see why things were not working
	 * actually fixed the problem - so simulate an ls -ld here.
	 */
	if (lstat(mp->am_path, &stb) < 0) {
#ifdef DEBUG
		dlog("lstat(%s): %m", mp->am_path);
#endif /* DEBUG */
	}
	error = UMOUNT_FS(mp->am_path);
	if (error == EBUSY) {
		plog(XLOG_WARNING, "afs_unmount retrying %s in 1s", mp->am_path);
		sleep(1);	/* XXX */
		goto again;
	}

	return error;
}

/*
 * Unmount an automount node
 */
static void afs_umounted P((am_node *mp));
static void afs_umounted(mp)
am_node *mp;
{
	/*
	 * If this is a pseudo-directory then just adjust the link count
	 * in the parent, otherwise call the generic unmount routine
	 */
	if (mp->am_parent && mp->am_parent->am_parent)
		--mp->am_parent->am_fattr.nlink;
}

/*
 * Mounting a file system may take a significant period of time.  The
 * problem is that if this is done in the main process thread then
 * the entire automounter could be blocked, possibly hanging lots of
 * processes on the system.  Instead we use a continuation scheme to
 * allow mounts to be attempted in a sub-process.  When the sub-process
 * exits we pick up the exit status (by convention a UN*X error number)
 * and continue in a notifier.  The notifier gets handed a data structure
 * and can then determine whether the mount was successful or not.  If
 * not, it updates the data structure and tries again until there are no
 * more ways to try the mount, or some other permanent error occurs.
 * In the mean time no RPC reply is sent, even after the mount is succesful.
 * We rely on the RPC retry mechanism to resend the lookup request which
 * can then be handled.
 */


struct continuation {
	char **ivec;		/* Current mount info */
	am_node *mp;		/* Node we are trying to mount */
	char *key;		/* Map key */
	char *info;		/* Info string */
	char **xivec;		/* Saved strsplit vector */
	char *auto_opts;	/* Automount options */
	am_opts fs_opts;	/* Filesystem options */
	char *def_opts;		/* Default automount options */
	int retry;		/* Try again? */
	int tried;		/* Have we tried any yet? */
	time_t start;		/* Time we started this mount */
	int callout;		/* Callout identifier */
};

#define	IN_PROGRESS(cp) ((cp)->mp->am_mnt->mf_flags & MFF_MOUNTING)

/*
 * Discard an old continuation
 */
static void free_continuation P((struct continuation *cp));
static void free_continuation(cp)
struct continuation *cp;
{
	if (cp->callout)
		untimeout(cp->callout);
	free((voidp) cp->key);
	free((voidp) cp->xivec);
	free((voidp) cp->info);
	free((voidp) cp->auto_opts);
	free((voidp) cp->def_opts);
	free_opts(&cp->fs_opts);
	free((voidp) cp);
}

static int afs_bgmount P((struct continuation*, int));

/*
 * Discard the underlying mount point and replace
 * with a reference to an error filesystem.
 */
static void assign_error_mntfs P((am_node *mp));
static void assign_error_mntfs(mp)
am_node *mp;
{
	if (mp->am_error > 0) {
		/*
		 * Save the old error code
		 */
		int error = mp->am_error;
		if (error <= 0)
			error = mp->am_mnt->mf_error;
		/*
		 * Discard the old filesystem
		 */
		free_mntfs(mp->am_mnt);
		/*
		 * Allocate a new error reference
		 */
		mp->am_mnt = new_mntfs();
		/*
		 * Put back the error code
		 */
		mp->am_mnt->mf_error = error;
		mp->am_mnt->mf_flags |= MFF_ERROR;
		/*
		 * Zero the error in the mount point
		 */
		mp->am_error = 0;
	}
}

/*
 * The continuation function.  This is called by
 * the task notifier when a background mount attempt