/*
* Return true if the supplied vnode has a sub-directory exported.
*/
int
has_visible(struct exportinfo *exi, vnode_t *vp)
{
struct exp_visible *visp;
fid_t fid;
bool_t vp_is_exported;
vp_is_exported = VN_CMP(vp, exi->exi_vp);
/*
* An exported root vnode has a sub-dir shared if it has a visible list.
* i.e. if it does not have a visible list, then there is no node in
* this filesystem leads to any other shared node.
*/
if (vp_is_exported && (vp->v_flag & VROOT))
return (exi->exi_visible ? 1 : 0);
/*
* Only the exportinfo of a fs root node may have a visible list.
* Either it is a pseudo root node, or a real exported root node.
*/
exi = get_root_export(exi);
if (!exi->exi_visible)
return (0);
/* Get the fid of the vnode */
bzero(&fid, sizeof (fid));
fid.fid_len = MAXFIDSZ;
if (vop_fid_pseudo(vp, &fid) != 0) {
return (0);
}
/*
* See if vp is in the visible list of the root node exportinfo.
*/
for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
if (EQFID(&fid, &visp->vis_fid)) {
/*
* If vp is an exported non-root node with only 1 path
* count (for itself), it indicates no sub-dir shared
* using this vp as a path.
*/
if (vp_is_exported && visp->vis_count < 2)
break;
return (1);
}
}
return (0);
}
/*
* Returns true if the supplied vnode is the
* directory of an export point.
*/
int
nfs_exported(struct exportinfo *exi, vnode_t *vp)
{
struct exp_visible *visp;
fid_t fid;
/*
* First check to see if vp is the export root
* This check required for the case of lookup ..
* where .. is a V_ROOT vnode and a pseudo exportroot.
* Pseudo export root objects do not have an entry
* in the visible list even though every V_ROOT
* pseudonode is visible. It is safe to compare
* vp here because pseudo_exportfs put a hold on
* it when exi_vp was initialized.
*
* Note: VN_CMP() won't match for LOFS shares, but they're
* handled below w/EQFID/EQFSID.
*/
if (VN_CMP(vp, exi->exi_vp))
return (1);
/* Get the fid of the vnode */
bzero(&fid, sizeof (fid));
fid.fid_len = MAXFIDSZ;
if (vop_fid_pseudo(vp, &fid) != 0)
return (0);
if (EQFID(&fid, &exi->exi_fid) &&
EQFSID(&vp->v_vfsp->vfs_fsid, &exi->exi_fsid)) {
return (1);
}
/* See if it matches any fid in the visible list */
for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
if (EQFID(&fid, &visp->vis_fid))
return (visp->vis_exported);
}
return (0);
}
/*
* See if vp is in use by the accounting system on any zone. This does a deep
* comparison of vnodes such that a file and a lofs "shadow" node of it will
* appear to be the same.
*
* If 'compare_vfs' is true, the function will do a comparison of vfs_t's
* instead (ie, is the vfs_t on which the vnode resides in use by the
* accounting system in any zone).
*
* Returns 1 if found (in use), 0 otherwise.
*/
static int
acct_find(vnode_t *vp, boolean_t compare_vfs)
{
struct acct_globals *ag;
vnode_t *realvp;
ASSERT(MUTEX_HELD(&acct_list_lock));
ASSERT(vp != NULL);
if (VOP_REALVP(vp, &realvp, NULL))
realvp = vp;
for (ag = list_head(&acct_list); ag != NULL;
ag = list_next(&acct_list, ag)) {
vnode_t *racctvp;
boolean_t found = B_FALSE;
mutex_enter(&ag->aclock);
if (ag->acctvp == NULL) {
mutex_exit(&ag->aclock);
continue;
}
if (VOP_REALVP(ag->acctvp, &racctvp, NULL))
racctvp = ag->acctvp;
if (compare_vfs) {
if (racctvp->v_vfsp == realvp->v_vfsp)
found = B_TRUE;
} else {
if (VN_CMP(realvp, racctvp))
found = B_TRUE;
}
mutex_exit(&ag->aclock);
if (found)
return (1);
}
return (0);
}
/*
* Returns true if the supplied vnode is visible
* in this export. If vnode is visible, return
* vis_exported in expseudo.
*/
int
nfs_visible(struct exportinfo *exi, vnode_t *vp, int *expseudo)
{
struct exp_visible *visp;
fid_t fid;
/*
* First check to see if vp is export root.
*
* A pseudo export root can never be exported
* (it would be a real export then); however,
* it is always visible. If a pseudo root object
* was exported by server admin, then the entire
* pseudo exportinfo (and all visible entries) would
* be destroyed. A pseudo exportinfo only exists
* to provide access to real (descendant) export(s).
*
* Previously, rootdir was special cased here; however,
* the export root special case handles the rootdir
* case also.
*/
if (VN_CMP(vp, exi->exi_vp)) {
*expseudo = 0;
return (1);
}
/*
* Only a PSEUDO node has a visible list or an exported VROOT
* node may have a visible list.
*/
if (! PSEUDO(exi))
exi = get_root_export(exi);
/* Get the fid of the vnode */
bzero(&fid, sizeof (fid));
fid.fid_len = MAXFIDSZ;
if (vop_fid_pseudo(vp, &fid) != 0) {
*expseudo = 0;
return (0);
}
/*
* We can't trust VN_CMP() above because of LOFS.
* Even though VOP_CMP will do the right thing for LOFS
* objects, VN_CMP will short circuit out early when the
* vnode ops ptrs are different. Just in case we're dealing
* with LOFS, compare exi_fid/fsid here.
*
* expseudo is not set because this is not an export
*/
if (EQFID(&exi->exi_fid, &fid) &&
EQFSID(&exi->exi_fsid, &vp->v_vfsp->vfs_fsid)) {
*expseudo = 0;
return (1);
}
/* See if it matches any fid in the visible list */
for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
if (EQFID(&fid, &visp->vis_fid)) {
*expseudo = visp->vis_exported;
return (1);
}
}
*expseudo = 0;
return (0);
}
/*
* This function checks the path to a new export to
* check whether all the pathname components are
* exported. It works by climbing the file tree one
* component at a time via "..", crossing mountpoints
* if necessary until an export entry is found, or the
* system root is reached.
*
* If an unexported mountpoint is found, then
* a new pseudo export is added and the pathname from
* the mountpoint down to the export is added to the
* visible list for the new pseudo export. If an existing
* pseudo export is found, then the pathname is added
* to its visible list.
*
* Note that there's some tests for exportdir.
* The exportinfo entry that's passed as a parameter
* is that of the real export and exportdir is set
* for this case.
*
* Here is an example of a possible setup:
*
* () - a new fs; fs mount point
* EXPORT - a real exported node
* PSEUDO - a pseudo node
* vis - visible list
* f# - security flavor#
* (f#) - security flavor# propagated from its descendents
* "" - covered vnode
*
*
* /
* |
* (a) PSEUDO (f1,f2)
* | vis: b,b,"c","n"
* |
* b
* ---------|------------------
* | |
* (c) EXPORT,f1(f2) (n) PSEUDO (f1,f2)
* | vis: "e","d" | vis: m,m,,p,q,"o"
* | |
* ------------------ -------------------
* | | | | |
* (d) (e) f m EXPORT,f1(f2) p
* EXPORT EXPORT | |
* f1 f2 | |
* | | |
* j (o) EXPORT,f2 q EXPORT f2
*
*/
int
treeclimb_export(struct exportinfo *exip)
{
vnode_t *dvp, *vp;
fid_t fid;
int error;
int exportdir;
struct exportinfo *exi = NULL;
struct exportinfo *new_exi = exip;
struct exp_visible *visp;
struct exp_visible *vis_head = NULL;
struct vattr va;
treenode_t *tree_head = NULL;
ASSERT(RW_WRITE_HELD(&exported_lock));
vp = exip->exi_vp;
VN_HOLD(vp);
exportdir = 1;
for (;;) {
bzero(&fid, sizeof (fid));
fid.fid_len = MAXFIDSZ;
error = vop_fid_pseudo(vp, &fid);
if (error)
break;
if (! exportdir) {
/*
* Check if this exportroot is a VROOT dir. If so,
* then attach the pseudonodes. If not, then
* continue .. traversal until we hit a VROOT
* export (pseudo or real).
*/
exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
if (exi != NULL && vp->v_flag & VROOT) {
/*
* Found an export info
*
* Extend the list of visible
* directories whether it's a pseudo
* or a real export.
*/
more_visible(exi, tree_head);
break; /* and climb no further */
}
}
/*
* If at the root of the filesystem, need
* to traverse across the mountpoint
* and continue the climb on the mounted-on
* filesystem.
*/
if (vp->v_flag & VROOT) {
if (! exportdir) {
/*
* Found the root directory of a filesystem
* that isn't exported. Need to export
* this as a pseudo export so that an NFS v4
* client can do lookups in it.
*/
new_exi = pseudo_exportfs(vp, &fid, vis_head,
NULL);
vis_head = NULL;
}
if (VN_CMP(vp, rootdir)) {
/* at system root */
/*
* If sharing "/", new_exi is shared exportinfo
* (exip). Otherwise, new_exi is exportinfo
* created in pseudo_exportfs() above.
*/
ns_root = tree_prepend_node(tree_head, 0,
new_exi);
break;
}
vp = untraverse(vp);
exportdir = 0;
continue;
}
/*
* Do a getattr to obtain the nodeid (inode num)
* for this vnode.
*/
va.va_mask = AT_NODEID;
error = VOP_GETATTR(vp, &va, 0, CRED(), NULL);
if (error)
break;
/*
* Add this directory fid to visible list
*/
visp = kmem_alloc(sizeof (*visp), KM_SLEEP);
VN_HOLD(vp);
visp->vis_vp = vp;
visp->vis_fid = fid; /* structure copy */
visp->vis_ino = va.va_nodeid;
visp->vis_count = 1;
visp->vis_exported = exportdir;
visp->vis_secinfo = NULL;
visp->vis_seccnt = 0;
visp->vis_next = vis_head;
vis_head = visp;
/*
* Will set treenode's pointer to exportinfo to
* 1. shared exportinfo (exip) - if first visit here
* 2. freshly allocated pseudo export (if any)
* 3. null otherwise
*/
tree_head = tree_prepend_node(tree_head, visp, new_exi);
new_exi = NULL;
/*
* Now, do a ".." to find parent dir of vp.
*/
error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, CRED(),
NULL, NULL, NULL);
if (error == ENOTDIR && exportdir) {
dvp = exip->exi_dvp;
ASSERT(dvp != NULL);
VN_HOLD(dvp);
error = 0;
}
if (error)
break;
exportdir = 0;
VN_RELE(vp);
vp = dvp;
}
VN_RELE(vp);
/*
* We can have set error due to error in:
* 1. vop_fid_pseudo()
* 2. VOP_GETATTR()
* 3. VOP_LOOKUP()
* We must free pseudo exportinfos, visibles and treenodes.
* Visibles are referenced from treenode_t::tree_vis and
* exportinfo_t::exi_visible. To avoid double freeing, only
* exi_visible pointer is used, via exi_rele(), for the clean-up.
*/
if (error) {
/* Free unconnected visibles, if there are any. */
if (vis_head)
free_visible(vis_head);
/* Connect unconnected exportinfo, if there is any. */
if (new_exi && new_exi != exip)
tree_head = tree_prepend_node(tree_head, 0, new_exi);
while (tree_head) {
treenode_t *t2 = tree_head;
exportinfo_t *e = tree_head->tree_exi;
/* exip will be freed in exportfs() */
if (e && e != exip) {
exp_kstats_delete(e->exi_kstats);
avl_remove(&exi_id_tree, e);
export_unlink(e);
exi_rele(e);
}
tree_head = tree_head->tree_child_first;
kmem_free(t2, sizeof (*t2));
}
}
return (error);
}
static int
chdirec(vnode_t *vp, int ischroot, int do_traverse)
{
int error;
vnode_t *oldvp;
proc_t *pp = curproc;
vnode_t **vpp;
refstr_t *cwd;
int newcwd = 1;
if (vp->v_type != VDIR) {
error = ENOTDIR;
goto bad;
}
if (error = VOP_ACCESS(vp, VEXEC, 0, CRED(), NULL))
goto bad;
/*
* The VOP_ACCESS() may have covered 'vp' with a new filesystem,
* if 'vp' is an autoFS vnode. Traverse the mountpoint so
* that we don't end up with a covered current directory.
*/
if (vn_mountedvfs(vp) != NULL && do_traverse) {
if (error = traverse(&vp))
goto bad;
}
/*
* Special chroot semantics: chroot is allowed if privileged
* or if the target is really a loopback mount of the root (or
* root of the zone) as determined by comparing dev and inode
* numbers
*/
if (ischroot) {
struct vattr tattr;
struct vattr rattr;
vnode_t *zonevp = curproc->p_zone->zone_rootvp;
tattr.va_mask = AT_FSID|AT_NODEID;
if (error = VOP_GETATTR(vp, &tattr, 0, CRED(), NULL))
goto bad;
rattr.va_mask = AT_FSID|AT_NODEID;
if (error = VOP_GETATTR(zonevp, &rattr, 0, CRED(), NULL))
goto bad;
if ((tattr.va_fsid != rattr.va_fsid ||
tattr.va_nodeid != rattr.va_nodeid) &&
(error = secpolicy_chroot(CRED())) != 0)
goto bad;
vpp = &PTOU(pp)->u_rdir;
} else {
vpp = &PTOU(pp)->u_cdir;
}
if (audit_active) /* update abs cwd/root path see c2audit.c */
audit_chdirec(vp, vpp);
mutex_enter(&pp->p_lock);
/*
* This bit of logic prevents us from overwriting u_cwd if we are
* changing to the same directory. We set the cwd to NULL so that we
* don't try to do the lookup on the next call to getcwd().
*/
if (!ischroot && *vpp != NULL && vp != NULL && VN_CMP(*vpp, vp))
newcwd = 0;
oldvp = *vpp;
*vpp = vp;
if ((cwd = PTOU(pp)->u_cwd) != NULL && newcwd)
PTOU(pp)->u_cwd = NULL;
mutex_exit(&pp->p_lock);
if (cwd && newcwd)
refstr_rele(cwd);
if (oldvp)
VN_RELE(oldvp);
return (0);
bad:
VN_RELE(vp);
return (error);
}
/*
* Given a directory, return the full, resolved path. This looks up "..",
* searches for the given vnode in the parent, appends the component, etc. It
* is used to implement vnodetopath() and getcwd() when the cached path fails.
*/
static int
dirtopath(vnode_t *vrootp, vnode_t *vp, char *buf, size_t buflen, int flags,
cred_t *cr)
{
pathname_t pn, rpn, emptypn;
vnode_t *cmpvp, *pvp = NULL;
vnode_t *startvp = vp;
int err = 0, vprivs;
size_t complen;
char *dbuf;
dirent64_t *dp;
char *bufloc;
size_t dlen = DIRENT64_RECLEN(MAXPATHLEN);
refstr_t *mntpt;
/* Operation only allowed on directories */
ASSERT(vp->v_type == VDIR);
/* We must have at least enough space for "/" */
if (buflen < 2)
return (ENAMETOOLONG);
/* Start at end of string with terminating null */
bufloc = &buf[buflen - 1];
*bufloc = '\0';
pn_alloc(&pn);
pn_alloc(&rpn);
dbuf = kmem_alloc(dlen, KM_SLEEP);
bzero(&emptypn, sizeof (emptypn));
/*
* Begin with an additional reference on vp. This will be decremented
* during the loop.
*/
VN_HOLD(vp);
for (;;) {
/*
* Return if we've reached the root. If the buffer is empty,
* return '/'. We explicitly don't use vn_compare(), since it
* compares the real vnodes. A lofs mount of '/' would produce
* incorrect results otherwise.
*/
if (VN_CMP(vrootp, vp)) {
if (*bufloc == '\0')
*--bufloc = '/';
break;
}
/*
* If we've reached the VFS root, something has gone wrong. We
* should have reached the root in the above check. The only
* explantation is that 'vp' is not contained withing the given
* root, in which case we return EPERM.
*/
if (VN_CMP(rootdir, vp)) {
err = EPERM;
goto out;
}
/*
* Shortcut: see if this vnode is a mountpoint. If so,
* grab the path information from the vfs_t.
*/
if (vp->v_flag & VROOT) {
mntpt = vfs_getmntpoint(vp->v_vfsp);
if ((err = pn_set(&pn, (char *)refstr_value(mntpt)))
== 0) {
refstr_rele(mntpt);
rpn.pn_path = rpn.pn_buf;
/*
* Ensure the mountpoint still exists.
*/
VN_HOLD(vrootp);
if (vrootp != rootdir)
VN_HOLD(vrootp);
if (lookuppnvp(&pn, &rpn, flags, NULL,
&cmpvp, vrootp, vrootp, cr) == 0) {
if (VN_CMP(vp, cmpvp)) {
VN_RELE(cmpvp);
complen = strlen(rpn.pn_path);
bufloc -= complen;
if (bufloc < buf) {
err = ERANGE;
goto out;
}
bcopy(rpn.pn_path, bufloc,
complen);
break;
} else {
VN_RELE(cmpvp);
}
}
} else {
refstr_rele(mntpt);
}
}
/*
* Shortcut: see if this vnode has correct v_path. If so,
* we have the work done.
*/
mutex_enter(&vp->v_lock);
if (vp->v_path != NULL) {
if ((err = pn_set(&pn, vp->v_path)) == 0) {
mutex_exit(&vp->v_lock);
rpn.pn_path = rpn.pn_buf;
/*
* Ensure the v_path pointing to correct vnode
*/
VN_HOLD(vrootp);
if (vrootp != rootdir)
VN_HOLD(vrootp);
if (lookuppnvp(&pn, &rpn, flags, NULL,
&cmpvp, vrootp, vrootp, cr) == 0) {
if (VN_CMP(vp, cmpvp)) {
VN_RELE(cmpvp);
complen = strlen(rpn.pn_path);
bufloc -= complen;
if (bufloc < buf) {
err = ERANGE;
goto out;
}
bcopy(rpn.pn_path, bufloc,
complen);
break;
} else {
VN_RELE(cmpvp);
}
}
} else {
mutex_exit(&vp->v_lock);
}
} else {
mutex_exit(&vp->v_lock);
}
/*
* Shortcuts failed, search for this vnode in its parent. If
* this is a mountpoint, then get the vnode underneath.
*/
if (vp->v_flag & VROOT)
vp = vn_under(vp);
if ((err = VOP_LOOKUP(vp, "..", &pvp, &emptypn, 0, vrootp, cr,
NULL, NULL, NULL)) != 0)
goto out;
/*
* With extended attributes, it's possible for a directory to
* have a parent that is a regular file. Check for that here.
*/
if (pvp->v_type != VDIR) {
err = ENOTDIR;
goto out;
}
/*
* If this is true, something strange has happened. This is
* only true if we are the root of a filesystem, which should
* have been caught by the check above.
*/
if (VN_CMP(pvp, vp)) {
err = ENOENT;
goto out;
}
/*
* Check if we have read and search privilege so, that
* we can lookup the path in the directory
*/
vprivs = (flags & LOOKUP_CHECKREAD) ? VREAD | VEXEC : VEXEC;
if ((err = VOP_ACCESS(pvp, vprivs, 0, cr, NULL)) != 0) {
goto out;
}
/*
* Try to obtain the path component from dnlc cache
* before searching through the directory.
*/
if ((cmpvp = dnlc_reverse_lookup(vp, dbuf, dlen)) != NULL) {
/*
* If we got parent vnode as a result,
* then the answered path is correct.
*/
if (VN_CMP(cmpvp, pvp)) {
VN_RELE(cmpvp);
complen = strlen(dbuf);
bufloc -= complen;
if (bufloc <= buf) {
err = ENAMETOOLONG;
goto out;
}
bcopy(dbuf, bufloc, complen);
/* Prepend a slash to the current path */
*--bufloc = '/';
/* And continue with the next component */
VN_RELE(vp);
vp = pvp;
pvp = NULL;
continue;
} else {
VN_RELE(cmpvp);
}
}
/*
* Search the parent directory for the entry corresponding to
* this vnode.
*/
if ((err = dirfindvp(vrootp, pvp, vp, cr, dbuf, dlen, &dp))
!= 0)
goto out;
complen = strlen(dp->d_name);
bufloc -= complen;
if (bufloc <= buf) {
err = ENAMETOOLONG;
goto out;
}
bcopy(dp->d_name, bufloc, complen);
/* Prepend a slash to the current path. */
*--bufloc = '/';
/* And continue with the next component */
VN_RELE(vp);
vp = pvp;
pvp = NULL;
}
/*
* Place the path at the beginning of the buffer.
*/
if (bufloc != buf)
ovbcopy(bufloc, buf, buflen - (bufloc - buf));
out:
/*
* If the error was ESTALE and the current directory to look in
* was the root for this lookup, the root for a mounted file
* system, or the starting directory for lookups, then
* return ENOENT instead of ESTALE. In this case, no recovery
* is possible by the higher level. If ESTALE was returned for
* some intermediate directory along the path, then recovery
* is potentially possible and retrying from the higher level
* will either correct the situation by purging stale cache
* entries or eventually get back to the point where no recovery
* is possible.
*/
if (err == ESTALE &&
(VN_CMP(vp, vrootp) || (vp->v_flag & VROOT) || vp == startvp))
err = ENOENT;
kmem_free(dbuf, dlen);
VN_RELE(vp);
if (pvp)
VN_RELE(pvp);
pn_free(&pn);
pn_free(&rpn);
return (err);
}
/*
* Starting at current directory, translate pathname pnp to end.
* Leave pathname of final component in pnp, return the vnode
* for the final component in *compvpp, and return the vnode
* for the parent of the final component in dirvpp.
*
* This is the central routine in pathname translation and handles
* multiple components in pathnames, separating them at /'s. It also
* implements mounted file systems and processes symbolic links.
*
* vp is the vnode where the directory search should start.
*
* Reference counts: vp must be held prior to calling this function. rootvp
* should only be held if rootvp != rootdir.
*/
int
lookuppnvp(
struct pathname *pnp, /* pathname to lookup */
struct pathname *rpnp, /* if non-NULL, return resolved path */
int flags, /* follow symlinks */
vnode_t **dirvpp, /* ptr for parent vnode */
vnode_t **compvpp, /* ptr for entry vnode */
vnode_t *rootvp, /* rootvp */
vnode_t *vp, /* directory to start search at */
cred_t *cr) /* user's credential */
{
vnode_t *cvp; /* current component vp */
char component[MAXNAMELEN]; /* buffer for component (incl null) */
int error;
int nlink;
int lookup_flags;
struct pathname presrvd; /* case preserved name */
struct pathname *pp = NULL;
vnode_t *startvp;
vnode_t *zonevp = curproc->p_zone->zone_rootvp; /* zone root */
int must_be_directory = 0;
boolean_t retry_with_kcred;
uint32_t auditing = AU_AUDITING();
CPU_STATS_ADDQ(CPU, sys, namei, 1);
nlink = 0;
cvp = NULL;
if (rpnp)
rpnp->pn_pathlen = 0;
lookup_flags = dirvpp ? LOOKUP_DIR : 0;
if (flags & FIGNORECASE) {
lookup_flags |= FIGNORECASE;
pn_alloc(&presrvd);
pp = &presrvd;
}
if (auditing)
audit_anchorpath(pnp, vp == rootvp);
/*
* Eliminate any trailing slashes in the pathname.
* If there are any, we must follow all symlinks.
* Also, we must guarantee that the last component is a directory.
*/
if (pn_fixslash(pnp)) {
flags |= FOLLOW;
must_be_directory = 1;
}
startvp = vp;
next:
retry_with_kcred = B_FALSE;
/*
* Make sure we have a directory.
*/
if (vp->v_type != VDIR) {
error = ENOTDIR;
goto bad;
}
if (rpnp && VN_CMP(vp, rootvp))
(void) pn_set(rpnp, "/");
/*
* Process the next component of the pathname.
*/
if (error = pn_getcomponent(pnp, component)) {
goto bad;
}
/*
* Handle "..": two special cases.
* 1. If we're at the root directory (e.g. after chroot or
* zone_enter) then change ".." to "." so we can't get
* out of this subtree.
* 2. If this vnode is the root of a mounted file system,
* then replace it with the vnode that was mounted on
* so that we take the ".." in the other file system.
*/
if (component[0] == '.' && component[1] == '.' && component[2] == 0) {
checkforroot:
if (VN_CMP(vp, rootvp) || VN_CMP(vp, zonevp)) {
component[1] = '\0';
} else if (vp->v_flag & VROOT) {
vfs_t *vfsp;
cvp = vp;
/*
* While we deal with the vfs pointer from the vnode
* the filesystem could have been forcefully unmounted
* and the vnode's v_vfsp could have been invalidated
* by VFS_UNMOUNT. Hence, we cache v_vfsp and use it
* with vfs_rlock_wait/vfs_unlock.
* It is safe to use the v_vfsp even it is freed by
* VFS_UNMOUNT because vfs_rlock_wait/vfs_unlock
* do not dereference v_vfsp. It is just used as a
* magic cookie.
* One more corner case here is the memory getting
* reused for another vfs structure. In this case
* lookuppnvp's vfs_rlock_wait will succeed, domount's
* vfs_lock will fail and domount will bail out with an
* error (EBUSY).
*/
vfsp = cvp->v_vfsp;
/*
* This lock is used to synchronize
* mounts/unmounts and lookups.
* Threads doing mounts/unmounts hold the
* writers version vfs_lock_wait().
*/
vfs_rlock_wait(vfsp);
/*
* If this vnode is on a file system that
* has been forcibly unmounted,
* we can't proceed. Cancel this operation
* and return EIO.
*
* vfs_vnodecovered is NULL if unmounted.
* Currently, nfs uses VFS_UNMOUNTED to
* check if it's a forced-umount. Keep the
* same checking here as well even though it
* may not be needed.
*/
if (((vp = cvp->v_vfsp->vfs_vnodecovered) == NULL) ||
(cvp->v_vfsp->vfs_flag & VFS_UNMOUNTED)) {
vfs_unlock(vfsp);
VN_RELE(cvp);
if (pp)
pn_free(pp);
return (EIO);
}
VN_HOLD(vp);
vfs_unlock(vfsp);
VN_RELE(cvp);
cvp = NULL;
/*
* Crossing mount points. For eg: We are doing
* a lookup of ".." for file systems root vnode
* mounted here, and VOP_LOOKUP() (with covered vnode)
* will be on underlying file systems mount point
* vnode. Set retry_with_kcred flag as we might end
* up doing VOP_LOOKUP() with kcred if required.
*/
retry_with_kcred = B_TRUE;
goto checkforroot;
}
}
/*
* LOOKUP_CHECKREAD is a private flag used by vnodetopath() to indicate
* that we need to have read permission on every directory in the entire
* path. This is used to ensure that a forward-lookup of a cached value
* has the same effect as a reverse-lookup when the cached value cannot
* be found.
*/
if ((flags & LOOKUP_CHECKREAD) &&
(error = VOP_ACCESS(vp, VREAD, 0, cr, NULL)) != 0)
goto bad;
/*
* Perform a lookup in the current directory.
*/
error = VOP_LOOKUP(vp, component, &cvp, pnp, lookup_flags,
rootvp, cr, NULL, NULL, pp);
/*
* Retry with kcred - If crossing mount points & error is EACCES.
*
* If we are crossing mount points here and doing ".." lookup,
* VOP_LOOKUP() might fail if the underlying file systems
* mount point has no execute permission. In cases like these,
* we retry VOP_LOOKUP() by giving as much privilage as possible
* by passing kcred credentials.
*
* In case of hierarchical file systems, passing kcred still may
* or may not work.
* For eg: UFS FS --> Mount NFS FS --> Again mount UFS on some
* directory inside NFS FS.
*/
if ((error == EACCES) && retry_with_kcred)
error = VOP_LOOKUP(vp, component, &cvp, pnp, lookup_flags,
rootvp, zone_kcred(), NULL, NULL, pp);
if (error) {
cvp = NULL;
/*
* On error, return hard error if
* (a) we're not at the end of the pathname yet, or
* (b) the caller didn't want the parent directory, or
* (c) we failed for some reason other than a missing entry.
*/
if (pn_pathleft(pnp) || dirvpp == NULL || error != ENOENT)
goto bad;
if (auditing) { /* directory access */
if (error = audit_savepath(pnp, vp, vp, error, cr))
goto bad_noaudit;
}
pn_setlast(pnp);
/*
* We inform the caller that the desired entry must be
* a directory by adding a '/' to the component name.
*/
if (must_be_directory && (error = pn_addslash(pnp)) != 0)
goto bad;
*dirvpp = vp;
if (compvpp != NULL)
*compvpp = NULL;
if (rootvp != rootdir)
VN_RELE(rootvp);
if (pp)
pn_free(pp);
return (0);
}
/*
* Traverse mount points.
* XXX why don't we need to hold a read lock here (call vn_vfsrlock)?
* What prevents a concurrent update to v_vfsmountedhere?
* Possible answer: if mounting, we might not see the mount
* if it is concurrently coming into existence, but that's
* really not much different from the thread running a bit slower.
* If unmounting, we may get into traverse() when we shouldn't,
* but traverse() will catch this case for us.
* (For this to work, fetching v_vfsmountedhere had better
* be atomic!)
*/
if (vn_mountedvfs(cvp) != NULL) {
if ((error = traverse(&cvp)) != 0)
goto bad;
}
/*
* If we hit a symbolic link and there is more path to be
* translated or this operation does not wish to apply
* to a link, then place the contents of the link at the
* front of the remaining pathname.
*/
if (cvp->v_type == VLNK && ((flags & FOLLOW) || pn_pathleft(pnp))) {
struct pathname linkpath;
if (++nlink > MAXSYMLINKS) {
error = ELOOP;
goto bad;
}
pn_alloc(&linkpath);
if (error = pn_getsymlink(cvp, &linkpath, cr)) {
pn_free(&linkpath);
goto bad;
}
if (auditing)
audit_symlink(pnp, &linkpath);
if (pn_pathleft(&linkpath) == 0)
(void) pn_set(&linkpath, ".");
error = pn_insert(pnp, &linkpath, strlen(component));
pn_free(&linkpath);
if (error)
goto bad;
VN_RELE(cvp);
cvp = NULL;
if (pnp->pn_pathlen == 0) {
error = ENOENT;
goto bad;
}
if (pnp->pn_path[0] == '/') {
do {
pnp->pn_path++;
pnp->pn_pathlen--;
} while (pnp->pn_path[0] == '/');
VN_RELE(vp);
vp = rootvp;
VN_HOLD(vp);
}
if (auditing)
audit_anchorpath(pnp, vp == rootvp);
if (pn_fixslash(pnp)) {
flags |= FOLLOW;
must_be_directory = 1;
}
goto next;
}
/*
* If rpnp is non-NULL, remember the resolved path name therein.
* Do not include "." components. Collapse occurrences of
* "previous/..", so long as "previous" is not itself "..".
* Exhausting rpnp results in error ENAMETOOLONG.
*/
if (rpnp && strcmp(component, ".") != 0) {
size_t len;
if (strcmp(component, "..") == 0 &&
rpnp->pn_pathlen != 0 &&
!((rpnp->pn_pathlen > 2 &&
strncmp(rpnp->pn_path+rpnp->pn_pathlen-3, "/..", 3) == 0) ||
(rpnp->pn_pathlen == 2 &&
strncmp(rpnp->pn_path, "..", 2) == 0))) {
while (rpnp->pn_pathlen &&
rpnp->pn_path[rpnp->pn_pathlen-1] != '/')
rpnp->pn_pathlen--;
if (rpnp->pn_pathlen > 1)
rpnp->pn_pathlen--;
rpnp->pn_path[rpnp->pn_pathlen] = '\0';
} else {
if (rpnp->pn_pathlen != 0 &&
rpnp->pn_path[rpnp->pn_pathlen-1] != '/')
rpnp->pn_path[rpnp->pn_pathlen++] = '/';
if (flags & FIGNORECASE) {
/*
* Return the case-preserved name
* within the resolved path.
*/
error = copystr(pp->pn_buf,
rpnp->pn_path + rpnp->pn_pathlen,
rpnp->pn_bufsize - rpnp->pn_pathlen, &len);
} else {
error = copystr(component,
rpnp->pn_path + rpnp->pn_pathlen,
rpnp->pn_bufsize - rpnp->pn_pathlen, &len);
}
if (error) /* copystr() returns ENAMETOOLONG */
goto bad;
rpnp->pn_pathlen += (len - 1);
ASSERT(rpnp->pn_bufsize > rpnp->pn_pathlen);
}
}
/*
* If no more components, return last directory (if wanted) and
* last component (if wanted).
*/
if (pn_pathleft(pnp) == 0) {
/*
* If there was a trailing slash in the pathname,
* make sure the last component is a directory.
*/
if (must_be_directory && cvp->v_type != VDIR) {
error = ENOTDIR;
goto bad;
}
if (dirvpp != NULL) {
/*
* Check that we have the real parent and not
* an alias of the last component.
*/
if (vn_compare(vp, cvp)) {
if (auditing)
(void) audit_savepath(pnp, cvp, vp,
EINVAL, cr);
pn_setlast(pnp);
VN_RELE(vp);
VN_RELE(cvp);
if (rootvp != rootdir)
VN_RELE(rootvp);
if (pp)
pn_free(pp);
return (EINVAL);
}
*dirvpp = vp;
} else
VN_RELE(vp);
if (auditing)
(void) audit_savepath(pnp, cvp, vp, 0, cr);
if (pnp->pn_path == pnp->pn_buf)
(void) pn_set(pnp, ".");
else
pn_setlast(pnp);
if (rpnp) {
if (VN_CMP(cvp, rootvp))
(void) pn_set(rpnp, "/");
else if (rpnp->pn_pathlen == 0)
(void) pn_set(rpnp, ".");
}
if (compvpp != NULL)
*compvpp = cvp;
else
VN_RELE(cvp);
if (rootvp != rootdir)
VN_RELE(rootvp);
if (pp)
pn_free(pp);
return (0);
}
/*
* Skip over slashes from end of last component.
*/
while (pnp->pn_path[0] == '/') {
pnp->pn_path++;
pnp->pn_pathlen--;
}
/*
* Searched through another level of directory:
* release previous directory handle and save new (result
* of lookup) as current directory.
*/
VN_RELE(vp);
vp = cvp;
cvp = NULL;
goto next;
bad:
if (auditing) /* reached end of path */
(void) audit_savepath(pnp, cvp, vp, error, cr);
bad_noaudit:
/*
* Error. Release vnodes and return.
*/
if (cvp)
VN_RELE(cvp);
/*
* If the error was ESTALE and the current directory to look in
* was the root for this lookup, the root for a mounted file
* system, or the starting directory for lookups, then
* return ENOENT instead of ESTALE. In this case, no recovery
* is possible by the higher level. If ESTALE was returned for
* some intermediate directory along the path, then recovery
* is potentially possible and retrying from the higher level
* will either correct the situation by purging stale cache
* entries or eventually get back to the point where no recovery
* is possible.
*/
if (error == ESTALE &&
(VN_CMP(vp, rootvp) || (vp->v_flag & VROOT) || vp == startvp))
error = ENOENT;
VN_RELE(vp);
if (rootvp != rootdir)
VN_RELE(rootvp);
if (pp)
pn_free(pp);
return (error);
}
int
dogetcwd(char *buf, size_t buflen)
{
int ret;
vnode_t *vp;
vnode_t *compvp;
refstr_t *cwd, *oldcwd;
const char *value;
pathname_t rpnp, pnp;
proc_t *p = curproc;
/*
* Check to see if there is a cached version of the cwd. If so, lookup
* the cached value and make sure it is the same vnode.
*/
mutex_enter(&p->p_lock);
if ((cwd = PTOU(p)->u_cwd) != NULL)
refstr_hold(cwd);
vp = PTOU(p)->u_cdir;
VN_HOLD(vp);
mutex_exit(&p->p_lock);
/*
* Make sure we have permission to access the current directory.
*/
if ((ret = VOP_ACCESS(vp, VEXEC, 0, CRED(), NULL)) != 0) {
if (cwd != NULL)
refstr_rele(cwd);
VN_RELE(vp);
return (ret);
}
if (cwd) {
value = refstr_value(cwd);
if ((ret = pn_get((char *)value, UIO_SYSSPACE, &pnp)) != 0) {
refstr_rele(cwd);
VN_RELE(vp);
return (ret);
}
pn_alloc(&rpnp);
if (lookuppn(&pnp, &rpnp, NO_FOLLOW, NULL, &compvp) == 0) {
if (VN_CMP(vp, compvp) &&
strcmp(value, rpnp.pn_path) == 0) {
VN_RELE(compvp);
VN_RELE(vp);
pn_free(&pnp);
pn_free(&rpnp);
if (strlen(value) + 1 > buflen) {
refstr_rele(cwd);
return (ENAMETOOLONG);
}
bcopy(value, buf, strlen(value) + 1);
refstr_rele(cwd);
return (0);
}
VN_RELE(compvp);
}
pn_free(&rpnp);
pn_free(&pnp);
refstr_rele(cwd);
}
ret = vnodetopath_common(NULL, vp, buf, buflen, CRED(),
LOOKUP_CHECKREAD);
VN_RELE(vp);
/*
* Store the new cwd and replace the existing cached copy.
*/
if (ret == 0)
cwd = refstr_alloc(buf);
else
cwd = NULL;
mutex_enter(&p->p_lock);
oldcwd = PTOU(p)->u_cwd;
PTOU(p)->u_cwd = cwd;
mutex_exit(&p->p_lock);
if (oldcwd)
refstr_rele(oldcwd);
return (ret);
}
/************************************************************************
* iumfs_free_node()
*
* 指定された vnode および iumnode を解放する
*
* 1. iumnode に関連づいたリソースを解放
* 2. iumnode 構造体を解放
* 3. vnode 構造体を解放
*
* これが呼ばれるのは、iumfs_inactive() もしくは iumfs_unmount() 経由の
* iumfs_free_all_node() だけ。つまり、v_count が 1（未参照状態）である
* 事が確かな場合だけ。
*
* 引数:
*
* vp: 解放する vnode 構造体のポインタ
* cr : システムコールを呼び出したユーザのクレデンシャル
*
* 戻り値:
* 無し
*
************************************************************************/
void
iumfs_free_node(vnode_t *vp, struct cred *cr)
{
iumnode_t *inp; // ファイルシステム型依存のノード情報(iumnode構造体)
vnode_t *rootvp; // ファイルシステムのルートディレクトリの vnode。
iumfs_t *iumfsp; // ファイルシステム型依存のプライベートデータ構造体
vfs_t *vfsp; // ファイルシステム構造体
int err;
DEBUG_PRINT((CE_CONT, "iumfs_free_node is called\n"));
iumfsp = VNODE2IUMFS(vp);
vfsp = VNODE2VFS(vp);
inp = VNODE2IUMNODE(vp);
DEBUG_PRINT((CE_CONT, "iumfs_free_node: vnode=%p, vp->v_count=%d\n", vp, vp->v_count));
/*
* 最初にノードリンクリストから iumnode をはずす。誰かが利用中(EBUSY)だったらリターン。
* 仮にノードリストに入っていなかったとしても、（ありえないはずだが) vnode のフリーは行う。
*/
if((err = iumfs_remove_node_from_list(vfsp, vp)) != 0){
if (err == ENOENT)
cmn_err(CE_CONT, "iumfs_free_node: can't find vnode in the list. Free it anyway.\n");
else
return;
}
// debug 用
rootvp = VNODE2ROOT(vp);
if (rootvp != NULL && VN_CMP(rootvp, vp) != 0) {
DEBUG_PRINT((CE_CONT, "iumfs_free_node: rootvnode is being freed\n"));
mutex_enter(&(iumfsp->iumfs_lock));
iumfsp->rootvnode = NULL;
mutex_exit(&(iumfsp->iumfs_lock));
}
/*
* もし iumnode にデータ（ディレクトリエントリ等）
* を含んでいたらそれらも解放する。
*/
if (inp->data != NULL) {
kmem_free(inp->data, inp->dlen);
}
/*
* この vnode に関連した page を無効にする
*/
err = pvn_vplist_dirty(vp, 0, iumfs_putapage, B_INVAL, cr);
DEBUG_PRINT((CE_CONT, "iumfs_free_node: pvn_vplist_dirty returned with (%d)\n", err));
if (vn_has_cached_data(vp)) {
cmn_err(CE_WARN, "iumfs_free_node: vnode still have cached\n");
}
// iumnode を解放
mutex_destroy(&(inp)->i_dlock);
rw_destroy(&(inp)->i_listlock);
kmem_free(inp, sizeof (iumnode_t));
// vnode を解放
#ifdef SOL10
vn_free(vp);
#else
mutex_destroy(&(vp)->v_lock);
kmem_free(vp, sizeof (vnode_t));
#endif
DEBUG_PRINT((CE_CONT, "iumfs_free_node: return\n"));
return;
}
