/*
* Create a minor node for the specified volume.
*/
int
zvol_create_minor(zfs_cmd_t *zc)
{
char *name = zc->zc_name;
dev_t dev = zc->zc_dev;
zvol_state_t *zv;
objset_t *os;
uint64_t volsize;
minor_t minor = 0;
struct pathname linkpath;
int ds_mode = DS_MODE_PRIMARY;
vnode_t *vp = NULL;
char *devpath;
size_t devpathlen = strlen(ZVOL_FULL_DEV_DIR) + 1 + strlen(name) + 1;
char chrbuf[30], blkbuf[30];
int error;
mutex_enter(&zvol_state_lock);
if ((zv = zvol_minor_lookup(name)) != NULL) {
mutex_exit(&zvol_state_lock);
return (EEXIST);
}
if (strchr(name, '@') != 0)
ds_mode |= DS_MODE_READONLY;
error = dmu_objset_open(name, DMU_OST_ZVOL, ds_mode, &os);
if (error) {
mutex_exit(&zvol_state_lock);
return (error);
}
error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
if (error) {
dmu_objset_close(os);
mutex_exit(&zvol_state_lock);
return (error);
}
/*
* If there's an existing /dev/zvol symlink, try to use the
* same minor number we used last time.
*/
devpath = kmem_alloc(devpathlen, KM_SLEEP);
(void) sprintf(devpath, "%s/%s", ZVOL_FULL_DEV_DIR, name);
error = lookupname(devpath, UIO_SYSSPACE, NO_FOLLOW, NULL, &vp);
kmem_free(devpath, devpathlen);
if (error == 0 && vp->v_type != VLNK)
error = EINVAL;
if (error == 0) {
pn_alloc(&linkpath);
error = pn_getsymlink(vp, &linkpath, kcred);
if (error == 0) {
char *ms = strstr(linkpath.pn_path, ZVOL_PSEUDO_DEV);
if (ms != NULL) {
ms += strlen(ZVOL_PSEUDO_DEV);
minor = stoi(&ms);
}
}
pn_free(&linkpath);
}
if (vp != NULL)
VN_RELE(vp);
/*
* If we found a minor but it's already in use, we must pick a new one.
*/
if (minor != 0 && ddi_get_soft_state(zvol_state, minor) != NULL)
minor = 0;
if (minor == 0)
minor = zvol_minor_alloc();
if (minor == 0) {
dmu_objset_close(os);
mutex_exit(&zvol_state_lock);
return (ENXIO);
}
if (ddi_soft_state_zalloc(zvol_state, minor) != DDI_SUCCESS) {
dmu_objset_close(os);
mutex_exit(&zvol_state_lock);
return (EAGAIN);
}
(void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, name);
(void) sprintf(chrbuf, "%uc,raw", minor);
if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
ddi_soft_state_free(zvol_state, minor);
dmu_objset_close(os);
mutex_exit(&zvol_state_lock);
return (EAGAIN);
}
(void) sprintf(blkbuf, "%uc", minor);
if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
ddi_remove_minor_node(zfs_dip, chrbuf);
ddi_soft_state_free(zvol_state, minor);
dmu_objset_close(os);
mutex_exit(&zvol_state_lock);
return (EAGAIN);
}
zv = ddi_get_soft_state(zvol_state, minor);
(void) strcpy(zv->zv_name, name);
zv->zv_min_bs = DEV_BSHIFT;
zv->zv_minor = minor;
zv->zv_volsize = volsize;
zv->zv_objset = os;
zv->zv_mode = ds_mode;
zv->zv_zilog = zil_open(os, NULL);
rw_init(&zv->zv_dslock, NULL, RW_DEFAULT, NULL);
zil_replay(os, zv, &zv->zv_txg_assign, zvol_replay_vector, NULL);
zvol_size_changed(zv, dev);
/* XXX this should handle the possible i/o error */
VERIFY(dsl_prop_register(dmu_objset_ds(zv->zv_objset),
"readonly", zvol_readonly_changed_cb, zv) == 0);
zvol_minors++;
mutex_exit(&zvol_state_lock);
return (0);
}
int
smb_pathname(smb_request_t *sr, char *path, int flags,
smb_node_t *root_node, smb_node_t *cur_node, smb_node_t **dir_node,
smb_node_t **ret_node, cred_t *cred)
{
char *component, *real_name, *namep;
pathname_t pn, rpn, upn, link_pn;
smb_node_t *dnode, *fnode;
smb_attr_t attr;
vnode_t *rootvp, *vp;
size_t pathleft;
int err = 0;
int nlink = 0;
int local_flags;
uint32_t abe_flag = 0;
char namebuf[MAXNAMELEN];
if (path == NULL)
return (EINVAL);
ASSERT(root_node);
ASSERT(cur_node);
ASSERT(ret_node);
*ret_node = NULL;
if (dir_node)
*dir_node = NULL;
(void) pn_alloc(&upn);
if ((err = pn_set(&upn, path)) != 0) {
(void) pn_free(&upn);
return (err);
}
if (SMB_TREE_SUPPORTS_ABE(sr))
abe_flag = SMB_ABE;
(void) pn_alloc(&pn);
(void) pn_alloc(&rpn);
component = kmem_alloc(MAXNAMELEN, KM_SLEEP);
real_name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
fnode = NULL;
dnode = cur_node;
smb_node_ref(dnode);
rootvp = root_node->vp;
while ((pathleft = pn_pathleft(&upn)) != 0) {
if (fnode) {
smb_node_release(dnode);
dnode = fnode;
fnode = NULL;
}
if ((err = pn_getcomponent(&upn, component)) != 0)
break;
if ((namep = smb_pathname_catia_v5tov4(sr, component,
namebuf, sizeof (namebuf))) == NULL) {
err = EILSEQ;
break;
}
if ((err = pn_set(&pn, namep)) != 0)
break;
local_flags = flags & FIGNORECASE;
err = smb_pathname_lookup(&pn, &rpn, local_flags,
&vp, rootvp, dnode->vp, &attr, cred);
if (err) {
if (!SMB_TREE_SUPPORTS_SHORTNAMES(sr) ||
!smb_maybe_mangled(component))
break;
if ((err = smb_unmangle(dnode, component,
real_name, MAXNAMELEN, abe_flag)) != 0)
break;
if ((namep = smb_pathname_catia_v5tov4(sr, real_name,
namebuf, sizeof (namebuf))) == NULL) {
err = EILSEQ;
break;
}
if ((err = pn_set(&pn, namep)) != 0)
break;
local_flags = 0;
err = smb_pathname_lookup(&pn, &rpn, local_flags,
&vp, rootvp, dnode->vp, &attr, cred);
if (err)
break;
}
/*
* This check MUST be done before symlink check
* since a reparse point is of type VLNK but should
* not be handled like a regular symlink.
*/
if (attr.sa_dosattr & FILE_ATTRIBUTE_REPARSE_POINT) {
err = EREMOTE;
VN_RELE(vp);
break;
}
if ((vp->v_type == VLNK) &&
((flags & FOLLOW) || pn_pathleft(&upn))) {
if (++nlink > MAXSYMLINKS) {
err = ELOOP;
VN_RELE(vp);
break;
}
(void) pn_alloc(&link_pn);
err = pn_getsymlink(vp, &link_pn, cred);
VN_RELE(vp);
if (err == 0) {
if (pn_pathleft(&link_pn) == 0)
(void) pn_set(&link_pn, ".");
err = pn_insert(&upn, &link_pn,
strlen(component));
}
pn_free(&link_pn);
if (err)
break;
if (upn.pn_pathlen == 0) {
err = ENOENT;
break;
}
if (upn.pn_path[0] == '/') {
fnode = root_node;
smb_node_ref(fnode);
}
if (pn_fixslash(&upn))
flags |= FOLLOW;
} else {
if (flags & FIGNORECASE) {
if (strcmp(rpn.pn_path, "/") != 0)
pn_setlast(&rpn);
namep = rpn.pn_path;
} else {
namep = pn.pn_path;
}
namep = smb_pathname_catia_v4tov5(sr, namep,
namebuf, sizeof (namebuf));
fnode = smb_node_lookup(sr, NULL, cred, vp, namep,
dnode, NULL);
VN_RELE(vp);
if (fnode == NULL) {
err = ENOMEM;
break;
}
}
while (upn.pn_path[0] == '/') {
upn.pn_path++;
upn.pn_pathlen--;
}
}
if ((pathleft) && (err == ENOENT))
err = ENOTDIR;
if (err) {
if (fnode)
smb_node_release(fnode);
if (dnode)
smb_node_release(dnode);
} else {
*ret_node = fnode;
if (dir_node)
*dir_node = dnode;
else
smb_node_release(dnode);
}
kmem_free(component, MAXNAMELEN);
kmem_free(real_name, MAXNAMELEN);
(void) pn_free(&pn);
(void) pn_free(&rpn);
(void) pn_free(&upn);
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);
}
