1004 lines
33 KiB
C
1004 lines
33 KiB
C
/*
|
|
* Copyright (c) 1997-2006 Erez Zadok
|
|
* Copyright (c) 1990 Jan-Simon Pendry
|
|
* Copyright (c) 1990 Imperial College of Science, Technology & Medicine
|
|
* Copyright (c) 1990 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 acknowledgment:
|
|
* 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.
|
|
*
|
|
*
|
|
* File: am-utils/libamu/mount_fs.c
|
|
*
|
|
*/
|
|
|
|
#ifdef HAVE_CONFIG_H
|
|
# include <config.h>
|
|
#endif /* HAVE_CONFIG_H */
|
|
#include <am_defs.h>
|
|
#include <amu.h>
|
|
|
|
|
|
/* ensure that mount table options are delimited by a comma */
|
|
#define append_opts(old, l, new) { \
|
|
if (*(old) != '\0') \
|
|
xstrlcat(old, ",", l); \
|
|
xstrlcat(old, new, l); }
|
|
|
|
/*
|
|
* Standard mount flags
|
|
*/
|
|
struct opt_tab mnt_flags[] =
|
|
{
|
|
#if defined(MNT2_GEN_OPT_RDONLY) && defined(MNTTAB_OPT_RO)
|
|
{MNTTAB_OPT_RO, MNT2_GEN_OPT_RDONLY},
|
|
#endif /* defined(MNT2_GEN_OPT_RDONLY) && defined(MNTTAB_OPT_RO) */
|
|
|
|
#if defined(MNT2_GEN_OPT_NOCACHE) && defined(MNTTAB_OPT_NOCACHE)
|
|
{MNTTAB_OPT_NOCACHE, MNT2_GEN_OPT_NOCACHE},
|
|
#endif /* defined(MNT2_GEN_OPT_NOCACHE) && defined(MNTTAB_OPT_NOCACHE) */
|
|
|
|
/* the "grpid" mount option can be offered as generic of NFS */
|
|
#ifdef MNTTAB_OPT_GRPID
|
|
# ifdef MNT2_GEN_OPT_GRPID
|
|
{MNTTAB_OPT_GRPID, MNT2_GEN_OPT_GRPID},
|
|
# endif /* MNT2_GEN_OPT_GRPID */
|
|
# ifdef MNT2_NFS_OPT_GRPID
|
|
{MNTTAB_OPT_GRPID, MNT2_NFS_OPT_GRPID},
|
|
# endif /* MNT2_NFS_OPT_GRPID */
|
|
#endif /* MNTTAB_OPT_GRPID */
|
|
|
|
#if defined(MNT2_GEN_OPT_MULTI) && defined(MNTTAB_OPT_MULTI)
|
|
{MNTTAB_OPT_MULTI, MNT2_GEN_OPT_MULTI},
|
|
#endif /* defined(MNT2_GEN_OPT_MULTI) && defined(MNTTAB_OPT_MULTI) */
|
|
|
|
#if defined(MNT2_GEN_OPT_NODEV) && defined(MNTTAB_OPT_NODEV)
|
|
{MNTTAB_OPT_NODEV, MNT2_GEN_OPT_NODEV},
|
|
#endif /* defined(MNT2_GEN_OPT_NODEV) && defined(MNTTAB_OPT_NODEV) */
|
|
|
|
#if defined(MNT2_GEN_OPT_NOEXEC) && defined(MNTTAB_OPT_NOEXEC)
|
|
{MNTTAB_OPT_NOEXEC, MNT2_GEN_OPT_NOEXEC},
|
|
#endif /* defined(MNT2_GEN_OPT_NOEXEC) && defined(MNTTAB_OPT_NOEXEC) */
|
|
|
|
#if defined(MNT2_GEN_OPT_NOSUB) && defined(MNTTAB_OPT_NOSUB)
|
|
{MNTTAB_OPT_NOSUB, MNT2_GEN_OPT_NOSUB},
|
|
#endif /* defined(MNT2_GEN_OPT_NOSUB) && defined(MNTTAB_OPT_NOSUB) */
|
|
|
|
#if defined(MNT2_GEN_OPT_NOSUID) && defined(MNTTAB_OPT_NOSUID)
|
|
{MNTTAB_OPT_NOSUID, MNT2_GEN_OPT_NOSUID},
|
|
#endif /* defined(MNT2_GEN_OPT_NOSUID) && defined(MNTTAB_OPT_NOSUID) */
|
|
|
|
#if defined(MNT2_GEN_OPT_SYNC) && defined(MNTTAB_OPT_SYNC)
|
|
{MNTTAB_OPT_SYNC, MNT2_GEN_OPT_SYNC},
|
|
#endif /* defined(MNT2_GEN_OPT_SYNC) && defined(MNTTAB_OPT_SYNC) */
|
|
|
|
#if defined(MNT2_GEN_OPT_OVERLAY) && defined(MNTTAB_OPT_OVERLAY)
|
|
{MNTTAB_OPT_OVERLAY, MNT2_GEN_OPT_OVERLAY},
|
|
#endif /* defined(MNT2_GEN_OPT_OVERLAY) && defined(MNTTAB_OPT_OVERLAY) */
|
|
|
|
/*
|
|
* Do not define MNT2_NFS_OPT_* entries here! This is for generic
|
|
* mount(2) options only, not for NFS mount options. If you need to put
|
|
* something here, it's probably not the right place: see
|
|
* include/am_compat.h.
|
|
*/
|
|
|
|
{0, 0}
|
|
};
|
|
|
|
|
|
/* compute generic mount flags */
|
|
int
|
|
compute_mount_flags(mntent_t *mntp)
|
|
{
|
|
struct opt_tab *opt;
|
|
int flags = 0;
|
|
|
|
#ifdef MNT2_GEN_OPT_NEWTYPE
|
|
flags |= MNT2_GEN_OPT_NEWTYPE;
|
|
#endif /* MNT2_GEN_OPT_NEWTYPE */
|
|
#ifdef MNT2_GEN_OPT_AUTOMOUNTED
|
|
flags |= MNT2_GEN_OPT_AUTOMOUNTED;
|
|
#endif /* not MNT2_GEN_OPT_AUTOMOUNTED */
|
|
|
|
/*
|
|
* Crack basic mount options
|
|
*/
|
|
for (opt = mnt_flags; opt->opt; opt++) {
|
|
flags |= amu_hasmntopt(mntp, opt->opt) ? opt->flag : 0;
|
|
}
|
|
|
|
return flags;
|
|
}
|
|
|
|
|
|
/* compute generic mount flags for automounter mounts */
|
|
int
|
|
compute_automounter_mount_flags(mntent_t *mntp)
|
|
{
|
|
int flags = 0;
|
|
|
|
#ifdef MNT2_GEN_OPT_IGNORE
|
|
flags |= MNT2_GEN_OPT_IGNORE;
|
|
#endif /* not MNT2_GEN_OPT_IGNORE */
|
|
#ifdef MNT2_GEN_OPT_AUTOMNTFS
|
|
flags |= MNT2_GEN_OPT_AUTOMNTFS;
|
|
#endif /* not MNT2_GEN_OPT_AUTOMNTFS */
|
|
|
|
return flags;
|
|
}
|
|
|
|
|
|
int
|
|
mount_fs(mntent_t *mnt, int flags, caddr_t mnt_data, int retry, MTYPE_TYPE type, u_long nfs_version, const char *nfs_proto, const char *mnttabname, int on_autofs)
|
|
{
|
|
int error = 0;
|
|
#ifdef MOUNT_TABLE_ON_FILE
|
|
char *zopts = NULL, *xopts = NULL;
|
|
size_t l;
|
|
#endif /* MOUNT_TABLE_ON_FILE */
|
|
char *mnt_dir = 0;
|
|
|
|
#ifdef NEED_AUTOFS_SPACE_HACK
|
|
char *old_mnt_dir = 0;
|
|
/* perform space hack */
|
|
if (on_autofs) {
|
|
old_mnt_dir = mnt->mnt_dir;
|
|
mnt->mnt_dir = mnt_dir = autofs_strdup_space_hack(old_mnt_dir);
|
|
} else
|
|
#endif /* NEED_AUTOFS_SPACE_HACK */
|
|
mnt_dir = strdup(mnt->mnt_dir);
|
|
|
|
dlog("'%s' fstype " MTYPE_PRINTF_TYPE " (%s) flags %#x (%s)",
|
|
mnt_dir, type, mnt->mnt_type, flags, mnt->mnt_opts);
|
|
|
|
again:
|
|
error = MOUNT_TRAP(type, mnt, flags, mnt_data);
|
|
|
|
if (error < 0) {
|
|
plog(XLOG_ERROR, "'%s': mount: %m", mnt_dir);
|
|
/*
|
|
* The following code handles conditions which shouldn't
|
|
* occur. They are possible either because amd screws up
|
|
* in preparing for the mount, or because some human
|
|
* messed with the mount point. Both have been known to
|
|
* happen. -- stolcke 2/22/95
|
|
*/
|
|
if (errno == EBUSY) {
|
|
/*
|
|
* Also, sometimes unmount isn't called, e.g., because
|
|
* our mountlist is garbled. This leaves old mount
|
|
* points around which need to be removed before we
|
|
* can mount something new in their place.
|
|
*/
|
|
errno = umount_fs(mnt_dir, mnttabname, on_autofs);
|
|
if (errno != 0)
|
|
plog(XLOG_ERROR, "'%s': umount: %m", mnt_dir);
|
|
else {
|
|
plog(XLOG_WARNING, "extra umount required for '%s'", mnt_dir);
|
|
error = MOUNT_TRAP(type, mnt, flags, mnt_data);
|
|
}
|
|
}
|
|
}
|
|
|
|
if (error < 0 && --retry > 0) {
|
|
sleep(1);
|
|
goto again;
|
|
}
|
|
|
|
#ifdef NEED_AUTOFS_SPACE_HACK
|
|
/* Undo space hack */
|
|
if (on_autofs)
|
|
mnt->mnt_dir = old_mnt_dir;
|
|
#endif /* NEED_AUTOFS_SPACE_HACK */
|
|
|
|
if (error < 0) {
|
|
error = errno;
|
|
goto out;
|
|
}
|
|
|
|
#ifdef MOUNT_TABLE_ON_FILE
|
|
/*
|
|
* Allocate memory for options:
|
|
* dev=..., vers={2,3}, proto={tcp,udp}
|
|
*/
|
|
l = strlen(mnt->mnt_opts) + 48;
|
|
zopts = (char *) xmalloc(l);
|
|
|
|
/* copy standard options */
|
|
xopts = mnt->mnt_opts;
|
|
|
|
xstrlcpy(zopts, xopts, l);
|
|
|
|
# ifdef MNTTAB_OPT_DEV
|
|
{
|
|
/* add the extra dev= field to the mount table */
|
|
struct stat stb;
|
|
if (lstat(mnt_dir, &stb) == 0) {
|
|
char optsbuf[48];
|
|
if (sizeof(stb.st_dev) == 2) /* e.g. SunOS 4.1 */
|
|
xsnprintf(optsbuf, sizeof(optsbuf), "%s=%04lx",
|
|
MNTTAB_OPT_DEV, (u_long) stb.st_dev & 0xffff);
|
|
else /* e.g. System Vr4 */
|
|
xsnprintf(optsbuf, sizeof(optsbuf), "%s=%08lx",
|
|
MNTTAB_OPT_DEV, (u_long) stb.st_dev);
|
|
append_opts(zopts, l, optsbuf);
|
|
}
|
|
}
|
|
# endif /* MNTTAB_OPT_DEV */
|
|
|
|
# if defined(HAVE_FS_NFS3) && defined(MNTTAB_OPT_VERS)
|
|
/*
|
|
* add the extra vers={2,3} field to the mount table,
|
|
* unless already specified by user
|
|
*/
|
|
if (nfs_version == NFS_VERSION3 &&
|
|
hasmntval(mnt, MNTTAB_OPT_VERS) != NFS_VERSION3) {
|
|
char optsbuf[48];
|
|
xsnprintf(optsbuf, sizeof(optsbuf),
|
|
"%s=%d", MNTTAB_OPT_VERS, NFS_VERSION3);
|
|
append_opts(zopts, l, optsbuf);
|
|
}
|
|
# endif /* defined(HAVE_FS_NFS3) && defined(MNTTAB_OPT_VERS) */
|
|
|
|
# ifdef MNTTAB_OPT_PROTO
|
|
/*
|
|
* add the extra proto={tcp,udp} field to the mount table,
|
|
* unless already specified by user.
|
|
*/
|
|
if (nfs_proto && !amu_hasmntopt(mnt, MNTTAB_OPT_PROTO)) {
|
|
char optsbuf[48];
|
|
xsnprintf(optsbuf, sizeof(optsbuf), "%s=%s", MNTTAB_OPT_PROTO, nfs_proto);
|
|
append_opts(zopts, l, optsbuf);
|
|
}
|
|
# endif /* MNTTAB_OPT_PROTO */
|
|
|
|
/* finally, store the options into the mount table structure */
|
|
mnt->mnt_opts = zopts;
|
|
|
|
/*
|
|
* Additional fields in mntent_t
|
|
* are fixed up here
|
|
*/
|
|
# ifdef HAVE_MNTENT_T_MNT_CNODE
|
|
mnt->mnt_cnode = 0;
|
|
# endif /* HAVE_MNTENT_T_MNT_CNODE */
|
|
|
|
# ifdef HAVE_MNTENT_T_MNT_RO
|
|
mnt->mnt_ro = (amu_hasmntopt(mnt, MNTTAB_OPT_RO) != NULL);
|
|
# endif /* HAVE_MNTENT_T_MNT_RO */
|
|
|
|
# ifdef HAVE_MNTENT_T_MNT_TIME
|
|
# ifdef HAVE_MNTENT_T_MNT_TIME_STRING
|
|
{ /* allocate enough space for a long */
|
|
size_t l = 13 * sizeof(char);
|
|
char *str = (char *) xmalloc(l);
|
|
xsnprintf(str, l, "%ld", time((time_t *) NULL));
|
|
mnt->mnt_time = str;
|
|
}
|
|
# else /* not HAVE_MNTENT_T_MNT_TIME_STRING */
|
|
mnt->mnt_time = time((time_t *) NULL);
|
|
# endif /* not HAVE_MNTENT_T_MNT_TIME_STRING */
|
|
# endif /* HAVE_MNTENT_T_MNT_TIME */
|
|
|
|
write_mntent(mnt, mnttabname);
|
|
|
|
# ifdef MNTTAB_OPT_DEV
|
|
if (xopts) {
|
|
XFREE(mnt->mnt_opts);
|
|
mnt->mnt_opts = xopts;
|
|
}
|
|
# endif /* MNTTAB_OPT_DEV */
|
|
#endif /* MOUNT_TABLE_ON_FILE */
|
|
|
|
out:
|
|
XFREE(mnt_dir);
|
|
return error;
|
|
}
|
|
|
|
|
|
/*
|
|
* Compute all NFS attribute cache related flags separately. Note that this
|
|
* function now computes attribute-cache flags for both Amd's automount
|
|
* points (NFS) as well as any normal NFS mount that Amd performs. Edit
|
|
* with caution.
|
|
*/
|
|
static void
|
|
compute_nfs_attrcache_flags(nfs_args_t *nap, mntent_t *mntp)
|
|
{
|
|
int acval = 0;
|
|
int err_acval = 1; /* 1 means we found no 'actimeo' value */
|
|
#if defined(HAVE_NFS_ARGS_T_ACREGMIN) || defined(HAVE_NFS_ARGS_T_ACREGMAX) || defined(HAVE_NFS_ARGS_T_ACDIRMIN) || defined(HAVE_NFS_ARGS_T_ACDIRMAX)
|
|
int err_acrdmm; /* for ac{reg,dir}{min,max} */
|
|
#endif /* HAVE_NFS_ARGS_T_AC{REG,DIR}{MIN,MAX} */
|
|
|
|
/************************************************************************/
|
|
/*** ATTRIBUTE CACHES ***/
|
|
/************************************************************************/
|
|
/*
|
|
* acval is set to 0 at the top of the function. If actimeo mount option
|
|
* exists and defined in mntopts, then its acval is set to it.
|
|
* If the value is non-zero, then we set all attribute cache fields to it.
|
|
* If acval is zero, it means it was never defined in mntopts or the
|
|
* actimeo mount option does not exist, in which case we check for
|
|
* individual mount options per attribute cache.
|
|
* Regardless of the value of acval, mount flags are set based directly
|
|
* on the values of the attribute caches.
|
|
*/
|
|
#ifdef MNTTAB_OPT_ACTIMEO
|
|
err_acval = hasmntvalerr(mntp, MNTTAB_OPT_ACTIMEO, &acval); /* attr cache timeout (sec) */
|
|
#endif /* MNTTAB_OPT_ACTIMEO */
|
|
|
|
/*** acregmin ***/
|
|
#ifdef HAVE_NFS_ARGS_T_ACREGMIN
|
|
err_acrdmm = 1; /* 1 means we found no acregmin value */
|
|
if (!err_acval) {
|
|
nap->acregmin = acval; /* min ac timeout for reg files (sec) */
|
|
} else {
|
|
# ifdef MNTTAB_OPT_ACREGMIN
|
|
err_acrdmm = hasmntvalerr(mntp, MNTTAB_OPT_ACREGMIN, (int *) &nap->acregmin);
|
|
# else /* not MNTTAB_OPT_ACREGMIN */
|
|
nap->acregmin = 0;
|
|
# endif /* not MNTTAB_OPT_ACREGMIN */
|
|
}
|
|
/* set this flag iff we changed acregmin (possibly to zero) */
|
|
# ifdef MNT2_NFS_OPT_ACREGMIN
|
|
if (!err_acval || !err_acrdmm)
|
|
nap->flags |= MNT2_NFS_OPT_ACREGMIN;
|
|
# endif /* MNT2_NFS_OPT_ACREGMIN */
|
|
#endif /* HAVE_NFS_ARGS_T_ACREGMIN */
|
|
|
|
/*** acregmax ***/
|
|
#ifdef HAVE_NFS_ARGS_T_ACREGMAX
|
|
err_acrdmm = 1; /* 1 means we found no acregmax value */
|
|
if (!err_acval) {
|
|
nap->acregmax = acval; /* max ac timeout for reg files (sec) */
|
|
} else {
|
|
# ifdef MNTTAB_OPT_ACREGMAX
|
|
err_acrdmm = hasmntvalerr(mntp, MNTTAB_OPT_ACREGMAX, (int *) &nap->acregmax);
|
|
# else /* not MNTTAB_OPT_ACREGMAX */
|
|
nap->acregmax = 0;
|
|
# endif /* not MNTTAB_OPT_ACREGMAX */
|
|
}
|
|
/* set this flag iff we changed acregmax (possibly to zero) */
|
|
# ifdef MNT2_NFS_OPT_ACREGMAX
|
|
if (!err_acval || !err_acrdmm)
|
|
nap->flags |= MNT2_NFS_OPT_ACREGMAX;
|
|
# endif /* MNT2_NFS_OPT_ACREGMAX */
|
|
#endif /* HAVE_NFS_ARGS_T_ACREGMAX */
|
|
|
|
/*** acdirmin ***/
|
|
#ifdef HAVE_NFS_ARGS_T_ACDIRMIN
|
|
err_acrdmm = 1; /* 1 means we found no acdirmin value */
|
|
if (!err_acval) {
|
|
nap->acdirmin = acval; /* min ac timeout for dirs (sec) */
|
|
} else {
|
|
# ifdef MNTTAB_OPT_ACDIRMIN
|
|
err_acrdmm = hasmntvalerr(mntp, MNTTAB_OPT_ACDIRMIN, (int *) &nap->acdirmin);
|
|
# else /* not MNTTAB_OPT_ACDIRMIN */
|
|
nap->acdirmin = 0;
|
|
# endif /* not MNTTAB_OPT_ACDIRMIN */
|
|
}
|
|
/* set this flag iff we changed acdirmin (possibly to zero) */
|
|
# ifdef MNT2_NFS_OPT_ACDIRMIN
|
|
if (!err_acval || !err_acrdmm)
|
|
nap->flags |= MNT2_NFS_OPT_ACDIRMIN;
|
|
# endif /* MNT2_NFS_OPT_ACDIRMIN */
|
|
#endif /* HAVE_NFS_ARGS_T_ACDIRMIN */
|
|
|
|
/*** acdirmax ***/
|
|
#ifdef HAVE_NFS_ARGS_T_ACDIRMAX
|
|
err_acrdmm = 1; /* 1 means we found no acdirmax value */
|
|
if (!err_acval) {
|
|
nap->acdirmax = acval; /* max ac timeout for dirs (sec) */
|
|
} else {
|
|
# ifdef MNTTAB_OPT_ACDIRMAX
|
|
err_acrdmm = hasmntvalerr(mntp, MNTTAB_OPT_ACDIRMAX, (int *) &nap->acdirmax);
|
|
# else /* not MNTTAB_OPT_ACDIRMAX */
|
|
nap->acdirmax = 0;
|
|
# endif /* not MNTTAB_OPT_ACDIRMAX */
|
|
}
|
|
/* set this flag iff we changed acdirmax (possibly to zero) */
|
|
# ifdef MNT2_NFS_OPT_ACDIRMAX
|
|
if (!err_acval || !err_acrdmm)
|
|
nap->flags |= MNT2_NFS_OPT_ACDIRMAX;
|
|
# endif /* MNT2_NFS_OPT_ACDIRMAX */
|
|
#endif /* HAVE_NFS_ARGS_T_ACDIRMAX */
|
|
|
|
|
|
/* don't cache attributes */
|
|
#if defined(MNTTAB_OPT_NOAC) && defined(MNT2_NFS_OPT_NOAC)
|
|
if (amu_hasmntopt(mntp, MNTTAB_OPT_NOAC) != NULL)
|
|
nap->flags |= MNT2_NFS_OPT_NOAC;
|
|
#endif /* defined(MNTTAB_OPT_NOAC) && defined(MNT2_NFS_OPT_NOAC) */
|
|
}
|
|
|
|
|
|
/*
|
|
* Fill in the many possible fields and flags of struct nfs_args.
|
|
*
|
|
* nap: pre-allocated structure to fill in.
|
|
* mntp: mount entry structure (includes options)
|
|
* genflags: generic mount flags already determined
|
|
* nfsncp: (TLI only) netconfig entry for this NFS mount
|
|
* ip_addr: IP address of file server
|
|
* nfs_version: 2, 3, (4 in the future), or 0 if unknown
|
|
* nfs_proto: "udp", "tcp", or NULL.
|
|
* fhp: file handle structure pointer
|
|
* host_name: name of remote NFS host
|
|
* fs_name: remote file system name to mount
|
|
*/
|
|
void
|
|
compute_nfs_args(nfs_args_t *nap,
|
|
mntent_t *mntp,
|
|
int genflags,
|
|
struct netconfig *nfsncp,
|
|
struct sockaddr_in *ip_addr,
|
|
u_long nfs_version,
|
|
char *nfs_proto,
|
|
am_nfs_handle_t *fhp,
|
|
char *host_name,
|
|
char *fs_name)
|
|
{
|
|
/* initialize just in case */
|
|
memset((voidp) nap, 0, sizeof(nfs_args_t));
|
|
|
|
/* compute all of the NFS attribute-cache flags */
|
|
compute_nfs_attrcache_flags(nap, mntp);
|
|
|
|
/************************************************************************/
|
|
/*** FILEHANDLE DATA AND LENGTH ***/
|
|
/************************************************************************/
|
|
#ifdef HAVE_FS_NFS3
|
|
if (nfs_version == NFS_VERSION3) {
|
|
# if defined(HAVE_NFS_ARGS_T_FHSIZE) || defined(HAVE_NFS_ARGS_T_FH_LEN)
|
|
/*
|
|
* Some systems (Irix/bsdi3) have a separate field in nfs_args for
|
|
* the length of the file handle for NFS V3. They insist that
|
|
* the file handle set in nfs_args be plain bytes, and not
|
|
* include the length field.
|
|
*/
|
|
NFS_FH_DREF(nap->NFS_FH_FIELD, &fhp->v3.am_fh3_data);
|
|
# else /* not defined(HAVE_NFS_ARGS_T_FHSIZE) || defined(HAVE_NFS_ARGS_T_FH_LEN) */
|
|
NFS_FH_DREF(nap->NFS_FH_FIELD, &fhp->v3);
|
|
# endif /* not defined(HAVE_NFS_ARGS_T_FHSIZE) || defined(HAVE_NFS_ARGS_T_FH_LEN) */
|
|
# ifdef MNT2_NFS_OPT_NFSV3
|
|
nap->flags |= MNT2_NFS_OPT_NFSV3;
|
|
# endif /* MNT2_NFS_OPT_NFSV3 */
|
|
# ifdef MNT2_NFS_OPT_VER3
|
|
nap->flags |= MNT2_NFS_OPT_VER3;
|
|
# endif /* MNT2_NFS_OPT_VER3 */
|
|
} else
|
|
#endif /* HAVE_FS_NFS3 */
|
|
NFS_FH_DREF(nap->NFS_FH_FIELD, &fhp->v2);
|
|
|
|
#ifdef HAVE_NFS_ARGS_T_FHSIZE
|
|
# ifdef HAVE_FS_NFS3
|
|
if (nfs_version == NFS_VERSION3)
|
|
nap->fhsize = fhp->v3.am_fh3_length;
|
|
else
|
|
# endif /* HAVE_FS_NFS3 */
|
|
nap->fhsize = FHSIZE;
|
|
#endif /* HAVE_NFS_ARGS_T_FHSIZE */
|
|
|
|
/* this is the version of the nfs_args structure, not of NFS! */
|
|
#ifdef HAVE_NFS_ARGS_T_FH_LEN
|
|
# ifdef HAVE_FS_NFS3
|
|
if (nfs_version == NFS_VERSION3)
|
|
nap->fh_len = fhp->v3.am_fh3_length;
|
|
else
|
|
# endif /* HAVE_FS_NFS3 */
|
|
nap->fh_len = FHSIZE;
|
|
#endif /* HAVE_NFS_ARGS_T_FH_LEN */
|
|
|
|
/************************************************************************/
|
|
/*** HOST NAME ***/
|
|
/************************************************************************/
|
|
/*
|
|
* XXX: warning, using xstrlcpy in NFS_HN_DREF, which may corrupt a
|
|
* struct nfs_args, or truncate our concocted "hostname:/path"
|
|
* string prematurely.
|
|
*/
|
|
NFS_HN_DREF(nap->hostname, host_name);
|
|
#ifdef MNT2_NFS_OPT_HOSTNAME
|
|
nap->flags |= MNT2_NFS_OPT_HOSTNAME;
|
|
#endif /* MNT2_NFS_OPT_HOSTNAME */
|
|
|
|
/************************************************************************/
|
|
/*** IP ADDRESS OF REMOTE HOST ***/
|
|
/************************************************************************/
|
|
if (ip_addr) {
|
|
#ifdef HAVE_TRANSPORT_TYPE_TLI
|
|
nap->addr = ALLOC(struct netbuf); /* free()'ed at end of mount_nfs_fh() */
|
|
#endif /* HAVE_TRANSPORT_TYPE_TLI */
|
|
NFS_SA_DREF(nap, ip_addr);
|
|
}
|
|
|
|
/************************************************************************/
|
|
/*** NFS PROTOCOL (UDP, TCP) AND VERSION ***/
|
|
/************************************************************************/
|
|
#ifdef MNT2_NFS_OPT_TCP
|
|
if (nfs_proto && STREQ(nfs_proto, "tcp"))
|
|
nap->flags |= MNT2_NFS_OPT_TCP;
|
|
#endif /* MNT2_NFS_OPT_TCP */
|
|
|
|
#ifdef HAVE_NFS_ARGS_T_SOTYPE
|
|
/* bsdi3 uses this */
|
|
if (nfs_proto) {
|
|
if (STREQ(nfs_proto, "tcp"))
|
|
nap->sotype = SOCK_STREAM;
|
|
else if (STREQ(nfs_proto, "udp"))
|
|
nap->sotype = SOCK_DGRAM;
|
|
}
|
|
#endif /* HAVE_NFS_ARGS_T_SOTYPE */
|
|
|
|
#ifdef HAVE_NFS_ARGS_T_PROTO
|
|
nap->proto = 0; /* bsdi3 sets this field to zero */
|
|
# ifdef IPPROTO_TCP
|
|
if (nfs_proto) {
|
|
if (STREQ(nfs_proto, "tcp")) /* AIX 4.2.x needs this */
|
|
nap->proto = IPPROTO_TCP;
|
|
else if (STREQ(nfs_proto, "udp"))
|
|
nap->proto = IPPROTO_UDP;
|
|
}
|
|
# endif /* IPPROTO_TCP */
|
|
#endif /* HAVE_NFS_ARGS_T_SOTYPE */
|
|
|
|
#ifdef HAVE_NFS_ARGS_T_VERSION
|
|
# ifdef NFS_ARGSVERSION
|
|
nap->version = NFS_ARGSVERSION; /* BSDI 3.0 and OpenBSD 2.2 */
|
|
# endif /* NFS_ARGSVERSION */
|
|
# ifdef DG_MOUNT_NFS_VERSION
|
|
nap->version = DG_MOUNT_NFS_VERSION; /* dg-ux */
|
|
# endif /* DG_MOUNT_NFS_VERSION */
|
|
#endif /* HAVE_NFS_ARGS_VERSION */
|
|
|
|
/************************************************************************/
|
|
/*** OTHER NFS SOCKET RELATED OPTIONS AND FLAGS ***/
|
|
/************************************************************************/
|
|
#ifdef MNT2_NFS_OPT_NOCONN
|
|
/* check if user specified to use unconnected or connected sockets */
|
|
if (amu_hasmntopt(mntp, MNTTAB_OPT_NOCONN) != NULL)
|
|
nap->flags |= MNT2_NFS_OPT_NOCONN;
|
|
else if (amu_hasmntopt(mntp, MNTTAB_OPT_CONN) != NULL)
|
|
nap->flags &= ~MNT2_NFS_OPT_NOCONN;
|
|
else {
|
|
/*
|
|
* Some OSs want you to set noconn always. Some want you to always turn
|
|
* it off. Others want you to turn it on/off only if NFS V.3 is used.
|
|
* And all of that changes from revision to another. This is
|
|
* particularly true of OpenBSD, NetBSD, and FreeBSD. So, rather than
|
|
* attempt to auto-detect this, I'm forced to "fix" it in the individual
|
|
* conf/nfs_prot/nfs_prot_*.h files.
|
|
*/
|
|
# ifdef USE_UNCONNECTED_NFS_SOCKETS
|
|
if (!(nap->flags & MNT2_NFS_OPT_NOCONN)) {
|
|
nap->flags |= MNT2_NFS_OPT_NOCONN;
|
|
plog(XLOG_WARNING, "noconn option not specified, and was just turned ON (OS override)! (May cause NFS hangs on some systems...)");
|
|
}
|
|
# endif /* USE_UNCONNECTED_NFS_SOCKETS */
|
|
# ifdef USE_CONNECTED_NFS_SOCKETS
|
|
if (nap->flags & MNT2_NFS_OPT_NOCONN) {
|
|
nap->flags &= ~MNT2_NFS_OPT_NOCONN;
|
|
plog(XLOG_WARNING, "noconn option specified, and was just turned OFF (OS override)! (May cause NFS hangs on some systems...)");
|
|
}
|
|
# endif /* USE_CONNECTED_NFS_SOCKETS */
|
|
}
|
|
#endif /* MNT2_NFS_OPT_NOCONN */
|
|
|
|
#ifdef MNT2_NFS_OPT_RESVPORT
|
|
# ifdef MNTTAB_OPT_RESVPORT
|
|
if (amu_hasmntopt(mntp, MNTTAB_OPT_RESVPORT) != NULL)
|
|
nap->flags |= MNT2_NFS_OPT_RESVPORT;
|
|
# else /* not MNTTAB_OPT_RESVPORT */
|
|
nap->flags |= MNT2_NFS_OPT_RESVPORT;
|
|
# endif /* not MNTTAB_OPT_RESVPORT */
|
|
#endif /* MNT2_NFS_OPT_RESVPORT */
|
|
|
|
/************************************************************************/
|
|
/*** OTHER FLAGS AND OPTIONS ***/
|
|
/************************************************************************/
|
|
|
|
#ifdef HAVE_TRANSPORT_TYPE_TLI
|
|
/* set up syncaddr field */
|
|
nap->syncaddr = (struct netbuf *) NULL;
|
|
|
|
/* set up knconf field */
|
|
if (get_knetconfig(&nap->knconf, nfsncp, nfs_proto) < 0) {
|
|
plog(XLOG_FATAL, "cannot fill knetconfig structure for nfs_args");
|
|
going_down(1);
|
|
}
|
|
/* update the flags field for knconf */
|
|
nap->flags |= MNT2_NFS_OPT_KNCONF;
|
|
#endif /* HAVE_TRANSPORT_TYPE_TLI */
|
|
|
|
#ifdef MNT2_NFS_OPT_FSNAME
|
|
nap->fsname = fs_name;
|
|
nap->flags |= MNT2_NFS_OPT_FSNAME;
|
|
#endif /* MNT2_NFS_OPT_FSNAME */
|
|
|
|
nap->rsize = hasmntval(mntp, MNTTAB_OPT_RSIZE);
|
|
#ifdef MNT2_NFS_OPT_RSIZE
|
|
if (nap->rsize)
|
|
nap->flags |= MNT2_NFS_OPT_RSIZE;
|
|
#endif /* MNT2_NFS_OPT_RSIZE */
|
|
if (nfs_version == NFS_VERSION && nap->rsize > 8192)
|
|
nap->rsize = 8192;
|
|
|
|
nap->wsize = hasmntval(mntp, MNTTAB_OPT_WSIZE);
|
|
#ifdef MNT2_NFS_OPT_WSIZE
|
|
if (nap->wsize)
|
|
nap->flags |= MNT2_NFS_OPT_WSIZE;
|
|
#endif /* MNT2_NFS_OPT_WSIZE */
|
|
if (nfs_version == NFS_VERSION && nap->wsize > 8192)
|
|
nap->wsize = 8192;
|
|
|
|
nap->timeo = hasmntval(mntp, MNTTAB_OPT_TIMEO);
|
|
#ifdef MNT2_NFS_OPT_TIMEO
|
|
if (nap->timeo)
|
|
nap->flags |= MNT2_NFS_OPT_TIMEO;
|
|
#endif /* MNT2_NFS_OPT_TIMEO */
|
|
|
|
nap->retrans = hasmntval(mntp, MNTTAB_OPT_RETRANS);
|
|
#ifdef MNT2_NFS_OPT_RETRANS
|
|
if (nap->retrans)
|
|
nap->flags |= MNT2_NFS_OPT_RETRANS;
|
|
#endif /* MNT2_NFS_OPT_RETRANS */
|
|
|
|
#ifdef MNT2_NFS_OPT_BIODS
|
|
if ((nap->biods = hasmntval(mntp, MNTTAB_OPT_BIODS)))
|
|
nap->flags |= MNT2_NFS_OPT_BIODS;
|
|
#endif /* MNT2_NFS_OPT_BIODS */
|
|
|
|
#ifdef MNT2_NFS_OPT_SOFT
|
|
if (amu_hasmntopt(mntp, MNTTAB_OPT_SOFT) != NULL)
|
|
nap->flags |= MNT2_NFS_OPT_SOFT;
|
|
#endif /* MNT2_NFS_OPT_SOFT */
|
|
|
|
#ifdef MNT2_NFS_OPT_SPONGY
|
|
if (amu_hasmntopt(mntp, MNTTAB_OPT_SPONGY) != NULL) {
|
|
nap->flags |= MNT2_NFS_OPT_SPONGY;
|
|
if (nap->flags & MNT2_NFS_OPT_SOFT) {
|
|
plog(XLOG_USER, "Mount opts soft and spongy are incompatible - soft ignored");
|
|
nap->flags &= ~MNT2_NFS_OPT_SOFT;
|
|
}
|
|
}
|
|
#endif /* MNT2_NFS_OPT_SPONGY */
|
|
|
|
#if defined(MNT2_GEN_OPT_RONLY) && defined(MNT2_NFS_OPT_RONLY)
|
|
/* Ultrix has separate generic and NFS ro flags */
|
|
if (genflags & MNT2_GEN_OPT_RONLY)
|
|
nap->flags |= MNT2_NFS_OPT_RONLY;
|
|
#endif /* defined(MNT2_GEN_OPT_RONLY) && defined(MNT2_NFS_OPT_RONLY) */
|
|
|
|
#ifdef MNTTAB_OPT_INTR
|
|
if (amu_hasmntopt(mntp, MNTTAB_OPT_INTR) != NULL)
|
|
/*
|
|
* Either turn on the "allow interrupts" option, or
|
|
* turn off the "disallow interrupts" option"
|
|
*/
|
|
# ifdef MNT2_NFS_OPT_INTR
|
|
nap->flags |= MNT2_NFS_OPT_INTR;
|
|
# endif /* MNT2_NFS_OPT_INTR */
|
|
# ifdef MNT2_NFS_OPT_NOINTR
|
|
nap->flags &= ~MNT2_NFS_OPT_NOINTR;
|
|
# endif /* MNT2_NFS_OPT_NOINTR */
|
|
# ifdef MNT2_NFS_OPT_INT
|
|
nap->flags |= MNT2_NFS_OPT_INT;
|
|
# endif /* MNT2_NFS_OPT_INT */
|
|
# ifdef MNT2_NFS_OPT_NOINT
|
|
nap->flags &= ~MNT2_NFS_OPT_NOINT;
|
|
# endif /* MNT2_NFS_OPT_NOINT */
|
|
#endif /* MNTTAB_OPT_INTR */
|
|
|
|
#ifdef MNTTAB_OPT_NODEVS
|
|
if (amu_hasmntopt(mntp, MNTTAB_OPT_NODEVS) != NULL)
|
|
nap->flags |= MNT2_NFS_OPT_NODEVS;
|
|
#endif /* MNTTAB_OPT_NODEVS */
|
|
|
|
#ifdef MNTTAB_OPT_COMPRESS
|
|
if (amu_hasmntopt(mntp, MNTTAB_OPT_COMPRESS) != NULL)
|
|
nap->flags |= MNT2_NFS_OPT_COMPRESS;
|
|
#endif /* MNTTAB_OPT_COMPRESS */
|
|
|
|
#ifdef MNTTAB_OPT_PRIVATE /* mount private, single-client tree */
|
|
if (amu_hasmntopt(mntp, MNTTAB_OPT_PRIVATE) != NULL)
|
|
nap->flags |= MNT2_NFS_OPT_PRIVATE;
|
|
#endif /* MNTTAB_OPT_PRIVATE */
|
|
|
|
#ifdef MNTTAB_OPT_SYMTTL /* symlink cache time-to-live */
|
|
if ((nap->symttl = hasmntval(mntp, MNTTAB_OPT_SYMTTL)))
|
|
nap->flags |= MNT2_NFS_OPT_SYMTTL;
|
|
#endif /* MNTTAB_OPT_SYMTTL */
|
|
|
|
#ifdef MNT2_NFS_OPT_PGTHRESH /* paging threshold */
|
|
if ((nap->pg_thresh = hasmntval(mntp, MNTTAB_OPT_PGTHRESH)))
|
|
nap->flags |= MNT2_NFS_OPT_PGTHRESH;
|
|
#endif /* MNT2_NFS_OPT_PGTHRESH */
|
|
|
|
#if defined(MNT2_NFS_OPT_NOCTO) && defined(MNTTAB_OPT_NOCTO)
|
|
if (amu_hasmntopt(mntp, MNTTAB_OPT_NOCTO) != NULL)
|
|
nap->flags |= MNT2_NFS_OPT_NOCTO;
|
|
#endif /* defined(MNT2_NFS_OPT_NOCTO) && defined(MNTTAB_OPT_NOCTO) */
|
|
|
|
#if defined(MNT2_NFS_OPT_POSIX) && defined(MNTTAB_OPT_POSIX)
|
|
if (amu_hasmntopt(mntp, MNTTAB_OPT_POSIX) != NULL) {
|
|
nap->flags |= MNT2_NFS_OPT_POSIX;
|
|
nap->pathconf = NULL;
|
|
}
|
|
#endif /* MNT2_NFS_OPT_POSIX && MNTTAB_OPT_POSIX */
|
|
|
|
#if defined(MNT2_NFS_OPT_PROPLIST) && defined(MNTTAB_OPT_PROPLIST)
|
|
if (amu_hasmntopt(mntp, MNTTAB_OPT_PROPLIST) != NULL)
|
|
nap->flags |= MNT2_NFS_OPT_PROPLIST;
|
|
#endif /* defined(MNT2_NFS_OPT_PROPLIST) && defined(MNTTAB_OPT_PROPLIST) */
|
|
|
|
#if defined(MNT2_NFS_OPT_MAXGRPS) && defined(MNTTAB_OPT_MAXGROUPS)
|
|
nap->maxgrouplist = hasmntval(mntp, MNTTAB_OPT_MAXGROUPS);
|
|
if (nap->maxgrouplist != 0)
|
|
nap->flags |= MNT2_NFS_OPT_MAXGRPS;
|
|
#endif /* defined(MNT2_NFS_OPT_MAXGRPS) && defined(MNTTAB_OPT_MAXGROUPS) */
|
|
|
|
#if defined(MNT2_NFS_OPT_NONLM) && defined(MNTTAB_OPT_NOLOCK)
|
|
if (amu_hasmntopt(mntp, MNTTAB_OPT_NOLOCK) != NULL)
|
|
nap->flags |= MNT2_NFS_OPT_NONLM;
|
|
#endif /* defined(MNT2_NFS_OPT_NONLM) && defined(MNTTAB_OPT_NOLOCK) */
|
|
|
|
#if defined(MNT2_NFS_OPT_XLATECOOKIE) && defined(MNTTAB_OPT_XLATECOOKIE)
|
|
if (amu_hasmntopt(mntp, MNTTAB_OPT_XLATECOOKIE) != NULL)
|
|
nap->flags |= MNT2_NFS_OPT_XLATECOOKIE;
|
|
#endif /* defined(MNT2_NFS_OPT_XLATECOOKIE) && defined(MNTTAB_OPT_XLATECOOKIE) */
|
|
|
|
#ifdef HAVE_NFS_ARGS_T_OPTSTR
|
|
nap->optstr = mntp->mnt_opts;
|
|
#endif /* HAVE_NFS_ARGS_T_OPTSTR */
|
|
|
|
/************************************************************************/
|
|
/*** FINAL ACTIONS ***/
|
|
/************************************************************************/
|
|
|
|
#ifdef HAVE_NFS_ARGS_T_GFS_FLAGS
|
|
/* Ultrix stores generic flags in nfs_args.gfs_flags. */
|
|
nap->gfs_flags = genflags;
|
|
#endif /* HAVE_NFS_ARGS_T_FLAGS */
|
|
|
|
return; /* end of compute_nfs_args() function */
|
|
}
|
|
|
|
|
|
/*
|
|
* Fill in special values for flags and fields of nfs_args, for an
|
|
* automounter NFS mount.
|
|
*/
|
|
void
|
|
compute_automounter_nfs_args(nfs_args_t *nap, mntent_t *mntp)
|
|
{
|
|
#ifdef MNT2_NFS_OPT_SYMTTL
|
|
/*
|
|
* Don't let the kernel cache symbolic links we generate, or else lookups
|
|
* will bypass amd and fail to remount stuff as needed.
|
|
*/
|
|
plog(XLOG_INFO, "turning on NFS option symttl and setting value to 0");
|
|
nap->flags |= MNT2_NFS_OPT_SYMTTL;
|
|
nap->symttl = 0;
|
|
#endif /* MNT2_NFS_OPT_SYMTTL */
|
|
|
|
/*
|
|
* This completes the flags for the HIDE_MOUNT_TYPE code in the
|
|
* mount_amfs_toplvl() function in amd/amfs_toplvl.c.
|
|
* Some systems don't have a mount type, but a mount flag.
|
|
*/
|
|
#ifdef MNT2_NFS_OPT_AUTO
|
|
nap->flags |= MNT2_NFS_OPT_AUTO;
|
|
#endif /* MNT2_NFS_OPT_AUTO */
|
|
#ifdef MNT2_NFS_OPT_IGNORE
|
|
nap->flags |= MNT2_NFS_OPT_IGNORE;
|
|
#endif /* MNT2_NFS_OPT_IGNORE */
|
|
#ifdef MNT2_GEN_OPT_AUTOMNTFS
|
|
nap->flags |= MNT2_GEN_OPT_AUTOMNTFS;
|
|
#endif /* not MNT2_GEN_OPT_AUTOMNTFS */
|
|
|
|
#ifdef MNT2_NFS_OPT_DUMBTIMR
|
|
/*
|
|
* Don't let the kernel start computing throughput of Amd. The numbers
|
|
* will be meaningless because of the way Amd does mount retries.
|
|
*/
|
|
plog(XLOG_INFO, "%s: disabling nfs congestion window", mntp->mnt_dir);
|
|
nap->flags |= MNT2_NFS_OPT_DUMBTIMR;
|
|
#endif /* MNT2_NFS_OPT_DUMBTIMR */
|
|
|
|
/* compute all of the NFS attribute-cache flags */
|
|
compute_nfs_attrcache_flags(nap, mntp);
|
|
|
|
/*
|
|
* Provide a slight bit more security by requiring the kernel to use
|
|
* reserved ports.
|
|
*/
|
|
#ifdef MNT2_NFS_OPT_RESVPORT
|
|
nap->flags |= MNT2_NFS_OPT_RESVPORT;
|
|
#endif /* MNT2_NFS_OPT_RESVPORT */
|
|
}
|
|
|
|
|
|
#ifdef DEBUG
|
|
/* get string version (in hex) of identifier */
|
|
static char *
|
|
get_hex_string(u_int len, const char *fhdata)
|
|
{
|
|
u_int i;
|
|
static char buf[128]; /* better not go over it! */
|
|
char str[16];
|
|
short int arr[64];
|
|
|
|
if (!fhdata)
|
|
return NULL;
|
|
buf[0] = '\0';
|
|
memset(&arr[0], 0, (64 * sizeof(short int)));
|
|
memcpy(&arr[0], &fhdata[0], len);
|
|
for (i=0; i<len/sizeof(unsigned short int); i++) {
|
|
xsnprintf(str, sizeof(str), "%04x", ntohs(arr[i]));
|
|
xstrlcat(buf, str, sizeof(buf));
|
|
}
|
|
return buf;
|
|
}
|
|
|
|
|
|
/*
|
|
* print a subset of fields from "struct nfs_args" that are otherwise
|
|
* not being provided anywhere else.
|
|
*/
|
|
void
|
|
print_nfs_args(const nfs_args_t *nap, u_long nfs_version)
|
|
{
|
|
int fhlen = 32; /* default: NFS V.2 file handle length is 32 */
|
|
#ifdef HAVE_TRANSPORT_TYPE_TLI
|
|
struct netbuf *nbp;
|
|
struct knetconfig *kncp;
|
|
#else /* not HAVE_TRANSPORT_TYPE_TLI */
|
|
struct sockaddr_in *sap;
|
|
#endif /* not HAVE_TRANSPORT_TYPE_TLI */
|
|
|
|
if (!nap) {
|
|
plog(XLOG_DEBUG, "NULL nfs_args!");
|
|
return;
|
|
}
|
|
|
|
/* override default file handle size */
|
|
#ifdef FHSIZE
|
|
fhlen = FHSIZE;
|
|
#endif /* FHSIZE */
|
|
#ifdef NFS_FHSIZE
|
|
fhlen = NFS_FHSIZE;
|
|
#endif /* NFS_FHSIZE */
|
|
|
|
#ifdef HAVE_TRANSPORT_TYPE_TLI
|
|
nbp = nap->addr;
|
|
plog(XLOG_DEBUG, "NA->addr {netbuf} (maxlen=%d, len=%d) = \"%s\"",
|
|
nbp->maxlen, nbp->len,
|
|
get_hex_string(nbp->len, nbp->buf));
|
|
nbp = nap->syncaddr;
|
|
plog(XLOG_DEBUG, "NA->syncaddr {netbuf} %p", nbp);
|
|
kncp = nap->knconf;
|
|
plog(XLOG_DEBUG, "NA->knconf->semantics %lu", (u_long) kncp->knc_semantics);
|
|
plog(XLOG_DEBUG, "NA->knconf->protofmly \"%s\"", kncp->knc_protofmly);
|
|
plog(XLOG_DEBUG, "NA->knconf->proto \"%s\"", kncp->knc_proto);
|
|
plog(XLOG_DEBUG, "NA->knconf->rdev %lu", (u_long) kncp->knc_rdev);
|
|
/* don't print knconf->unused field */
|
|
#else /* not HAVE_TRANSPORT_TYPE_TLI */
|
|
# ifdef NFS_ARGS_T_ADDR_IS_POINTER
|
|
sap = (struct sockaddr_in *) nap->addr;
|
|
# else /* not NFS_ARGS_T_ADDR_IS_POINTER */
|
|
sap = (struct sockaddr_in *) &nap->addr;
|
|
# endif /* not NFS_ARGS_T_ADDR_IS_POINTER */
|
|
plog(XLOG_DEBUG, "NA->addr {sockaddr_in} (len=%d) = \"%s\"",
|
|
(int) sizeof(struct sockaddr_in),
|
|
get_hex_string(sizeof(struct sockaddr_in), (const char *)sap));
|
|
#ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
|
|
/* as per POSIX, sin_len need not be set (used internally by kernel) */
|
|
plog(XLOG_DEBUG, "NA->addr.sin_len = %d", sap->sin_len);
|
|
#endif /* HAVE_STRUCT_SOCKADDR_SA_LEN */
|
|
plog(XLOG_DEBUG, "NA->addr.sin_family = %d", sap->sin_family);
|
|
plog(XLOG_DEBUG, "NA->addr.sin_port = %d", sap->sin_port);
|
|
plog(XLOG_DEBUG, "NA->addr.sin_addr = \"%s\"",
|
|
get_hex_string(sizeof(struct in_addr), (const char *) &sap->sin_addr));
|
|
#endif /* not HAVE_TRANSPORT_TYPE_TLI */
|
|
#ifdef HAVE_NFS_ARGS_T_ADDRLEN
|
|
plog(XLOG_DEBUG, "NA->addrlen = %d", nap->addrlen);
|
|
#endif /* ifdef HAVE_NFS_ARGS_T_ADDRLEN */
|
|
|
|
plog(XLOG_DEBUG, "NA->hostname = \"%s\"", nap->hostname ? nap->hostname : "null");
|
|
#ifdef HAVE_NFS_ARGS_T_NAMLEN
|
|
plog(XLOG_DEBUG, "NA->namlen = %d", nap->namlen);
|
|
#endif /* HAVE_NFS_ARGS_T_NAMLEN */
|
|
|
|
#ifdef MNT2_NFS_OPT_FSNAME
|
|
plog(XLOG_DEBUG, "NA->fsname = \"%s\"", nap->fsname ? nap->fsname : "null");
|
|
#endif /* MNT2_NFS_OPT_FSNAME */
|
|
|
|
#ifdef HAVE_NFS_ARGS_T_FHSIZE
|
|
plog(XLOG_DEBUG, "NA->fhsize = %d", nap->fhsize);
|
|
fhlen = nap->fhsize;
|
|
#endif /* HAVE_NFS_ARGS_T_FHSIZE */
|
|
#ifdef HAVE_NFS_ARGS_T_FH_LEN
|
|
plog(XLOG_DEBUG, "NA->fh_len = %d", nap->fh_len);
|
|
fhlen = nap->fh_len;
|
|
#endif /* HAVE_NFS_ARGS_T_FH_LEN */
|
|
|
|
/*
|
|
* XXX: need to figure out how to correctly print file handles,
|
|
* since some times they are pointers, and sometimes the real structure
|
|
* is stored in nfs_args. Even if it is a pointer, it can be the actual
|
|
* char[] array, or a structure containing multiple fields.
|
|
*/
|
|
plog(XLOG_DEBUG, "NA->filehandle = \"%s\"",
|
|
get_hex_string(fhlen, (const char *) &nap->NFS_FH_FIELD));
|
|
|
|
#ifdef HAVE_NFS_ARGS_T_SOTYPE
|
|
plog(XLOG_DEBUG, "NA->sotype = %d", nap->sotype);
|
|
#endif /* HAVE_NFS_ARGS_T_SOTYPE */
|
|
#ifdef HAVE_NFS_ARGS_T_PROTO
|
|
plog(XLOG_DEBUG, "NA->proto = %d", (int) nap->proto);
|
|
#endif /* HAVE_NFS_ARGS_T_PROTO */
|
|
#ifdef HAVE_NFS_ARGS_T_VERSION
|
|
plog(XLOG_DEBUG, "NA->version = %d", nap->version);
|
|
#endif /* HAVE_NFS_ARGS_T_VERSION */
|
|
|
|
plog(XLOG_DEBUG, "NA->flags = 0x%x", (int) nap->flags);
|
|
|
|
plog(XLOG_DEBUG, "NA->rsize = %d", (int) nap->rsize);
|
|
plog(XLOG_DEBUG, "NA->wsize = %d", (int) nap->wsize);
|
|
#ifdef HAVE_NFS_ARGS_T_BSIZE
|
|
plog(XLOG_DEBUG, "NA->bsize = %d", nap->bsize);
|
|
#endif /* HAVE_NFS_ARGS_T_BSIZE */
|
|
plog(XLOG_DEBUG, "NA->timeo = %d", (int) nap->timeo);
|
|
plog(XLOG_DEBUG, "NA->retrans = %d", (int) nap->retrans);
|
|
|
|
#ifdef HAVE_NFS_ARGS_T_ACREGMIN
|
|
plog(XLOG_DEBUG, "NA->acregmin = %d", (int) nap->acregmin);
|
|
plog(XLOG_DEBUG, "NA->acregmax = %d", (int) nap->acregmax);
|
|
plog(XLOG_DEBUG, "NA->acdirmin = %d", (int) nap->acdirmin);
|
|
plog(XLOG_DEBUG, "NA->acdirmax = %d", (int) nap->acdirmax);
|
|
#endif /* HAVE_NFS_ARGS_T_ACREGMIN */
|
|
#ifdef MNTTAB_OPT_SYMTTL
|
|
plog(XLOG_DEBUG, "NA->symttl = %d", nap->symttl);
|
|
#endif /* MNTTAB_OPT_SYMTTL */
|
|
#ifdef MNTTAB_OPT_PG_THRESH
|
|
plog(XLOG_DEBUG, "NA->pg_thresh = %d", nap->pg_thresh);
|
|
#endif /* MNTTAB_OPT_PG_THRESH */
|
|
|
|
#ifdef MNT2_NFS_OPT_BIODS
|
|
plog(XLOG_DEBUG, "NA->biods = %d", nap->biods);
|
|
#endif /* MNT2_NFS_OPT_BIODS */
|
|
|
|
}
|
|
#endif /* DEBUG */
|