/*
* Return a znode for the extended attribute directory for zp.
* ** If the directory does not already exist, it is created **
*
* IN: zp - znode to obtain attribute directory from
* cr - credentials of caller
* flags - flags from the VOP_LOOKUP call
*
* OUT: xipp - pointer to extended attribute znode
*
* RETURN: 0 on success
* error number on failure
*/
int
zfs_get_xattrdir(znode_t *zp, struct inode **xipp, cred_t *cr, int flags)
{
zfsvfs_t *zfsvfs = ZTOZSB(zp);
znode_t *xzp;
zfs_dirlock_t *dl;
vattr_t va;
int error;
top:
error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR, NULL, NULL);
if (error)
return (error);
if (xzp != NULL) {
*xipp = ZTOI(xzp);
zfs_dirent_unlock(dl);
return (0);
}
if (!(flags & CREATE_XATTR_DIR)) {
zfs_dirent_unlock(dl);
return (SET_ERROR(ENOENT));
}
if (zfs_is_readonly(zfsvfs)) {
zfs_dirent_unlock(dl);
return (SET_ERROR(EROFS));
}
/*
* The ability to 'create' files in an attribute
* directory comes from the write_xattr permission on the base file.
*
* The ability to 'search' an attribute directory requires
* read_xattr permission on the base file.
*
* Once in a directory the ability to read/write attributes
* is controlled by the permissions on the attribute file.
*/
va.va_mask = ATTR_MODE | ATTR_UID | ATTR_GID;
va.va_mode = S_IFDIR | S_ISVTX | 0777;
zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
va.va_dentry = NULL;
error = zfs_make_xattrdir(zp, &va, xipp, cr);
zfs_dirent_unlock(dl);
if (error == ERESTART) {
/* NB: we already did dmu_tx_wait() if necessary */
goto top;
}
return (error);
}
/*
* Return a znode for the extended attribute directory for zp.
* ** If the directory does not already exist, it is created **
*
* IN: zp - znode to obtain attribute directory from
* cr - credentials of caller
*
* OUT: xzpp - pointer to extended attribute znode
*
* RETURN: 0 on success
* error number on failure
*/
int
zfs_get_xattrdir(znode_t *zp, vnode_t **xvpp, cred_t *cr)
{
zfsvfs_t *zfsvfs = zp->z_zfsvfs;
znode_t *xzp;
zfs_dirlock_t *dl;
vattr_t va;
int error;
top:
error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR);
if (error)
return (error);
if (xzp != NULL) {
*xvpp = ZTOV(xzp);
zfs_dirent_unlock(dl);
return (0);
}
ASSERT(zp->z_phys->zp_xattr == 0);
if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
zfs_dirent_unlock(dl);
return (EROFS);
}
/*
* The ability to 'create' files in an attribute
* directory comes from the write_xattr permission on the base file.
*
* The ability to 'search' an attribute directory requires
* read_xattr permission on the base file.
*
* Once in a directory the ability to read/write attributes
* is controlled by the permissions on the attribute file.
*/
va.va_mask = AT_TYPE | AT_MODE | AT_UID | AT_GID;
va.va_type = VDIR;
va.va_mode = S_IFDIR | S_ISVTX | 0777;
va.va_uid = (uid_t)zp->z_phys->zp_uid;
va.va_gid = (gid_t)zp->z_phys->zp_gid;
error = zfs_make_xattrdir(zp, &va, xvpp, cr);
zfs_dirent_unlock(dl);
if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
/* NB: we already did dmu_tx_wait() if necessary */
goto top;
}
return (error);
}
/*
* Look up an entry in a directory.
*
* NOTE: '.' and '..' are handled as special cases because
* no directory entries are actually stored for them. If this is
* the root of a filesystem, then '.zfs' is also treated as a
* special pseudo-directory.
*/
int
zfs_dirlook(znode_t *dzp, char *name, struct inode **ipp, int flags,
int *deflg, pathname_t *rpnp)
{
zfs_dirlock_t *dl;
znode_t *zp;
int error = 0;
uint64_t parent;
if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
*ipp = ZTOI(dzp);
igrab(*ipp);
} else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
zfs_sb_t *zsb = ZTOZSB(dzp);
/*
* If we are a snapshot mounted under .zfs, return
* the vp for the snapshot directory.
*/
if ((error = sa_lookup(dzp->z_sa_hdl,
SA_ZPL_PARENT(zsb), &parent, sizeof (parent))) != 0)
return (error);
#ifdef HAVE_SNAPSHOT
if (parent == dzp->z_id && zsb->z_parent != zsb) {
error = zfsctl_root_lookup(zsb->z_parent->z_ctldir,
"snapshot", ipp, NULL, 0, NULL, kcred,
NULL, NULL, NULL);
return (error);
}
#endif /* HAVE_SNAPSHOT */
rw_enter(&dzp->z_parent_lock, RW_READER);
error = zfs_zget(zsb, parent, &zp);
if (error == 0)
*ipp = ZTOI(zp);
rw_exit(&dzp->z_parent_lock);
#ifdef HAVE_SNAPSHOT
} else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) {
*ipp = zfsctl_root(dzp);
#endif /* HAVE_SNAPSHOT */
} else {
int zf;
zf = ZEXISTS | ZSHARED;
if (flags & FIGNORECASE)
zf |= ZCILOOK;
error = zfs_dirent_lock(&dl, dzp, name, &zp, zf, deflg, rpnp);
if (error == 0) {
*ipp = ZTOI(zp);
zfs_dirent_unlock(dl);
dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
}
rpnp = NULL;
}
if ((flags & FIGNORECASE) && rpnp && !error)
(void) strlcpy(rpnp->pn_buf, name, rpnp->pn_bufsize);
return (error);
}
/*
* Look up an entry in a directory.
*
* NOTE: '.' and '..' are handled as special cases because
* no directory entries are actually stored for them. If this is
* the root of a filesystem, then '.zfs' is also treated as a
* special pseudo-directory.
*/
int
zfs_dirlook(znode_t *dzp, char *name, vnode_t **vpp, int flags,
int *deflg, pathname_t *rpnp)
{
zfs_dirlock_t *dl;
znode_t *zp;
int error = 0;
if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
*vpp = ZTOV(dzp);
VN_HOLD(*vpp);
} else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
/*
* If we are a snapshot mounted under .zfs, return
* the vp for the snapshot directory.
*/
if (dzp->z_phys->zp_parent == dzp->z_id &&
zfsvfs->z_parent != zfsvfs) {
error = zfsctl_root_lookup(zfsvfs->z_parent->z_ctldir,
"snapshot", vpp, NULL, 0, NULL, kcred,
NULL, NULL, NULL);
return (error);
}
rw_enter(&dzp->z_parent_lock, RW_READER);
error = zfs_zget(zfsvfs, dzp->z_phys->zp_parent, &zp);
if (error == 0)
*vpp = ZTOV(zp);
rw_exit(&dzp->z_parent_lock);
} else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) {
*vpp = zfsctl_root(dzp);
} else {
int zf;
zf = ZEXISTS | ZSHARED;
if (flags & FIGNORECASE)
zf |= ZCILOOK;
error = zfs_dirent_lock(&dl, dzp, name, &zp, zf, deflg, rpnp);
if (error == 0) {
*vpp = ZTOV(zp);
zfs_dirent_unlock(dl);
dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
}
rpnp = NULL;
}
if ((flags & FIGNORECASE) && rpnp && !error)
(void) strlcpy(rpnp->pn_buf, name, rpnp->pn_bufsize);
return (error);
}
zfs_dirlook(znode_t *dzp, char *name, vnode_t **vpp)
#endif
{
zfs_dirlock_t *dl;
znode_t *zp;
int error = 0;
uint64_t parent;
#ifdef __APPLE__
char *name = cnp->cn_nameptr;
#endif
if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
*vpp = ZTOV(dzp);
VN_HOLD(*vpp);
} else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
/*
* If we are a snapshot mounted under .zfs, return
* the vp for the snapshot directory.
*/
if ((error = sa_lookup(dzp->z_sa_hdl,
SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
return (error);
if (parent == dzp->z_id && zfsvfs->z_parent != zfsvfs) {
error = zfsctl_root_lookup(zfsvfs->z_parent->z_ctldir,
"snapshot", vpp, NULL, 0, NULL, kcred
/*, NULL, NULL, NULL*/);
return (error);
}
rw_enter(&dzp->z_parent_lock, RW_READER);
error = zfs_zget(zfsvfs, parent, &zp);
if (error == 0)
*vpp = ZTOV(zp);
rw_exit(&dzp->z_parent_lock);
} else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) {
*vpp = zfsctl_root(dzp);
} else {
#ifdef __APPLE__
error = zfs_dirent_lock(&dl, dzp, cnp, &zp, ZEXISTS | ZSHARED);
#else
error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS | ZSHARED);
#endif
if (error == 0) {
*vpp = ZTOV(zp);
zfs_dirent_unlock(dl);
dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
}
}
return (error);
}
/*
* Look up an entry in a directory.
*
* NOTE: '.' and '..' are handled as special cases because
* no directory entries are actually stored for them. If this is
* the root of a filesystem, then '.zfs' is also treated as a
* special pseudo-directory.
*/
int
zfs_dirlook(znode_t *dzp, char *name, vnode_t **vpp)
{
zfs_dirlock_t *dl;
znode_t *zp;
int error = 0;
if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
*vpp = ZTOV(dzp);
VN_HOLD(*vpp);
} else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
/*
* If we are a snapshot mounted under .zfs, return
* the vp for the snapshot directory.
*/
if (dzp->z_phys->zp_parent == dzp->z_id &&
zfsvfs->z_parent != zfsvfs) {
error = zfsctl_root_lookup(zfsvfs->z_parent->z_ctldir,
"snapshot", vpp, NULL, 0, NULL, kcred);
return (error);
}
rw_enter(&dzp->z_parent_lock, RW_READER);
error = zfs_zget(zfsvfs, dzp->z_phys->zp_parent, &zp);
if (error == 0)
*vpp = ZTOV(zp);
rw_exit(&dzp->z_parent_lock);
} else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) {
*vpp = zfsctl_root(dzp);
} else {
error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS | ZSHARED);
if (error == 0) {
*vpp = ZTOV(zp);
zfs_dirent_unlock(dl);
dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
}
}
return (error);
}
/*
* Lock a directory entry. A dirlock on <dzp, name> protects that name
* in dzp's directory zap object. As long as you hold a dirlock, you can
* assume two things: (1) dzp cannot be reaped, and (2) no other thread
* can change the zap entry for (i.e. link or unlink) this name.
*
* Input arguments:
* dzp - znode for directory
* name - name of entry to lock
* flag - ZNEW: if the entry already exists, fail with EEXIST.
* ZEXISTS: if the entry does not exist, fail with ENOENT.
* ZSHARED: allow concurrent access with other ZSHARED callers.
* ZXATTR: we want dzp's xattr directory
* ZCILOOK: On a mixed sensitivity file system,
* this lookup should be case-insensitive.
* ZCIEXACT: On a purely case-insensitive file system,
* this lookup should be case-sensitive.
* ZRENAMING: we are locking for renaming, force narrow locks
* ZHAVELOCK: Don't grab the z_name_lock for this call. The
* current thread already holds it.
*
* Output arguments:
* zpp - pointer to the znode for the entry (NULL if there isn't one)
* dlpp - pointer to the dirlock for this entry (NULL on error)
* direntflags - (case-insensitive lookup only)
* flags if multiple case-sensitive matches exist in directory
* realpnp - (case-insensitive lookup only)
* actual name matched within the directory
*
* Return value: 0 on success or errno on failure.
*
* NOTE: Always checks for, and rejects, '.' and '..'.
* NOTE: For case-insensitive file systems we take wide locks (see below),
* but return znode pointers to a single match.
*/
int
zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name, znode_t **zpp,
int flag, int *direntflags, pathname_t *realpnp)
{
zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
zfs_dirlock_t *dl;
boolean_t update;
boolean_t exact;
uint64_t zoid;
vnode_t *vp = NULL;
int error = 0;
int cmpflags;
*zpp = NULL;
*dlpp = NULL;
/*
* Verify that we are not trying to lock '.', '..', or '.zfs'
*/
if ((name[0] == '.' &&
(name[1] == '\0' || (name[1] == '.' && name[2] == '\0'))) ||
(zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0))
return (EEXIST);
/*
* Case sensitivity and normalization preferences are set when
* the file system is created. These are stored in the
* zfsvfs->z_case and zfsvfs->z_norm fields. These choices
* affect what vnodes can be cached in the DNLC, how we
* perform zap lookups, and the "width" of our dirlocks.
*
* A normal dirlock locks a single name. Note that with
* normalization a name can be composed multiple ways, but
* when normalized, these names all compare equal. A wide
* dirlock locks multiple names. We need these when the file
* system is supporting mixed-mode access. It is sometimes
* necessary to lock all case permutations of file name at
* once so that simultaneous case-insensitive/case-sensitive
* behaves as rationally as possible.
*/
/*
* Decide if exact matches should be requested when performing
* a zap lookup on file systems supporting case-insensitive
* access.
*/
exact =
((zfsvfs->z_case == ZFS_CASE_INSENSITIVE) && (flag & ZCIEXACT)) ||
((zfsvfs->z_case == ZFS_CASE_MIXED) && !(flag & ZCILOOK));
/*
* Only look in or update the DNLC if we are looking for the
* name on a file system that does not require normalization
* or case folding. We can also look there if we happen to be
* on a non-normalizing, mixed sensitivity file system IF we
* are looking for the exact name.
*
* Maybe can add TO-UPPERed version of name to dnlc in ci-only
* case for performance improvement?
*/
update = !zfsvfs->z_norm ||
((zfsvfs->z_case == ZFS_CASE_MIXED) &&
!(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER) && !(flag & ZCILOOK));
/*
* ZRENAMING indicates we are in a situation where we should
* take narrow locks regardless of the file system's
* preferences for normalizing and case folding. This will
* prevent us deadlocking trying to grab the same wide lock
* twice if the two names happen to be case-insensitive
* matches.
*/
if (flag & ZRENAMING)
cmpflags = 0;
else
cmpflags = zfsvfs->z_norm;
/*
* Wait until there are no locks on this name.
*
* Don't grab the the lock if it is already held. However, cannot
* have both ZSHARED and ZHAVELOCK together.
*/
ASSERT(!(flag & ZSHARED) || !(flag & ZHAVELOCK));
if (!(flag & ZHAVELOCK))
rw_enter(&dzp->z_name_lock, RW_READER);
mutex_enter(&dzp->z_lock);
for (;;) {
if (dzp->z_unlinked) {
mutex_exit(&dzp->z_lock);
if (!(flag & ZHAVELOCK))
rw_exit(&dzp->z_name_lock);
return (ENOENT);
}
for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next) {
if ((u8_strcmp(name, dl->dl_name, 0, cmpflags,
U8_UNICODE_LATEST, &error) == 0) || error != 0)
break;
}
if (error != 0) {
mutex_exit(&dzp->z_lock);
if (!(flag & ZHAVELOCK))
rw_exit(&dzp->z_name_lock);
return (ENOENT);
}
if (dl == NULL) {
/*
* Allocate a new dirlock and add it to the list.
*/
dl = kmem_alloc(sizeof (zfs_dirlock_t), KM_SLEEP);
cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL);
dl->dl_name = name;
dl->dl_sharecnt = 0;
dl->dl_namelock = 0;
dl->dl_namesize = 0;
dl->dl_dzp = dzp;
dl->dl_next = dzp->z_dirlocks;
dzp->z_dirlocks = dl;
break;
}
if ((flag & ZSHARED) && dl->dl_sharecnt != 0)
break;
cv_wait(&dl->dl_cv, &dzp->z_lock);
}
/*
* If the z_name_lock was NOT held for this dirlock record it.
*/
if (flag & ZHAVELOCK)
dl->dl_namelock = 1;
if ((flag & ZSHARED) && ++dl->dl_sharecnt > 1 && dl->dl_namesize == 0) {
/*
* We're the second shared reference to dl. Make a copy of
* dl_name in case the first thread goes away before we do.
* Note that we initialize the new name before storing its
* pointer into dl_name, because the first thread may load
* dl->dl_name at any time. He'll either see the old value,
* which is his, or the new shared copy; either is OK.
*/
dl->dl_namesize = strlen(dl->dl_name) + 1;
name = kmem_alloc(dl->dl_namesize, KM_SLEEP);
bcopy(dl->dl_name, name, dl->dl_namesize);
dl->dl_name = name;
}
mutex_exit(&dzp->z_lock);
/*
* We have a dirlock on the name. (Note that it is the dirlock,
* not the dzp's z_lock, that protects the name in the zap object.)
* See if there's an object by this name; if so, put a hold on it.
*/
if (flag & ZXATTR) {
error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &zoid,
sizeof (zoid));
if (error == 0)
error = (zoid == 0 ? ENOENT : 0);
} else {
if (update)
vp = dnlc_lookup(ZTOV(dzp), name);
if (vp == DNLC_NO_VNODE) {
VN_RELE(vp);
error = ENOENT;
} else if (vp) {
if (flag & ZNEW) {
zfs_dirent_unlock(dl);
VN_RELE(vp);
return (EEXIST);
}
*dlpp = dl;
*zpp = VTOZ(vp);
return (0);
} else {
error = zfs_match_find(zfsvfs, dzp, name, exact,
update, direntflags, realpnp, &zoid);
}
}
if (error) {
if (error != ENOENT || (flag & ZEXISTS)) {
zfs_dirent_unlock(dl);
return (error);
}
} else {
if (flag & ZNEW) {
zfs_dirent_unlock(dl);
return (EEXIST);
}
error = zfs_zget(zfsvfs, zoid, zpp);
if (error) {
zfs_dirent_unlock(dl);
return (error);
}
if (!(flag & ZXATTR) && update)
dnlc_update(ZTOV(dzp), name, ZTOV(*zpp));
}
*dlpp = dl;
return (0);
}
/*
* Return a znode for the extended attribute directory for zp.
* ** If the directory does not already exist, it is created **
*
* IN: zp - znode to obtain attribute directory from
* cr - credentials of caller
* flags - flags from the VOP_LOOKUP call
*
* OUT: xzpp - pointer to extended attribute znode
*
* RETURN: 0 on success
* error number on failure
*/
int
zfs_get_xattrdir(znode_t *zp, vnode_t **xvpp, cred_t *cr, int flags)
{
zfsvfs_t *zfsvfs = zp->z_zfsvfs;
znode_t *xzp;
zfs_dirlock_t *dl;
vattr_t va;
int error;
top:
error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR, NULL, NULL);
if (error)
return (error);
if (xzp != NULL) {
*xvpp = ZTOV(xzp);
zfs_dirent_unlock(dl);
return (0);
}
if (!(flags & CREATE_XATTR_DIR)) {
zfs_dirent_unlock(dl);
#ifdef illumos
return (SET_ERROR(ENOENT));
#else
return (SET_ERROR(ENOATTR));
#endif
}
if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
zfs_dirent_unlock(dl);
return (SET_ERROR(EROFS));
}
/*
* The ability to 'create' files in an attribute
* directory comes from the write_xattr permission on the base file.
*
* The ability to 'search' an attribute directory requires
* read_xattr permission on the base file.
*
* Once in a directory the ability to read/write attributes
* is controlled by the permissions on the attribute file.
*/
va.va_mask = AT_TYPE | AT_MODE | AT_UID | AT_GID;
va.va_type = VDIR;
va.va_mode = S_IFDIR | S_ISVTX | 0777;
zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
error = zfs_make_xattrdir(zp, &va, xvpp, cr);
zfs_dirent_unlock(dl);
if (error == ERESTART) {
/* NB: we already did dmu_tx_wait() if necessary */
goto top;
}
if (error == 0)
VOP_UNLOCK(*xvpp, 0);
return (error);
}
/*
* Lock a directory entry. A dirlock on <dzp, name> protects that name
* in dzp's directory zap object. As long as you hold a dirlock, you can
* assume two things: (1) dzp cannot be reaped, and (2) no other thread
* can change the zap entry for (i.e. link or unlink) this name.
*
* Input arguments:
* dzp - znode for directory
* name - name of entry to lock
* flag - ZNEW: if the entry already exists, fail with EEXIST.
* ZEXISTS: if the entry does not exist, fail with ENOENT.
* ZSHARED: allow concurrent access with other ZSHARED callers.
* ZXATTR: we want dzp's xattr directory
*
* Output arguments:
* zpp - pointer to the znode for the entry (NULL if there isn't one)
* dlpp - pointer to the dirlock for this entry (NULL on error)
*
* Return value: 0 on success or errno on failure.
*
* NOTE: Always checks for, and rejects, '.' and '..'.
*/
int
zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name, znode_t **zpp,
int flag)
{
zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
zfs_dirlock_t *dl;
uint64_t zoid;
int error;
vnode_t *vp;
*zpp = NULL;
*dlpp = NULL;
/*
* Verify that we are not trying to lock '.', '..', or '.zfs'
*/
if (name[0] == '.' &&
(name[1] == '\0' || (name[1] == '.' && name[2] == '\0')) ||
zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0)
return (EEXIST);
/*
* Wait until there are no locks on this name.
*/
mutex_enter(&dzp->z_lock);
for (;;) {
if (dzp->z_reap) {
mutex_exit(&dzp->z_lock);
return (ENOENT);
}
for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next)
if (strcmp(name, dl->dl_name) == 0)
break;
if (dl == NULL) {
/*
* Allocate a new dirlock and add it to the list.
*/
dl = kmem_alloc(sizeof (zfs_dirlock_t), KM_SLEEP);
cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL);
dl->dl_name = name;
dl->dl_sharecnt = 0;
dl->dl_namesize = 0;
dl->dl_dzp = dzp;
dl->dl_next = dzp->z_dirlocks;
dzp->z_dirlocks = dl;
break;
}
if ((flag & ZSHARED) && dl->dl_sharecnt != 0)
break;
cv_wait(&dl->dl_cv, &dzp->z_lock);
}
if ((flag & ZSHARED) && ++dl->dl_sharecnt > 1 && dl->dl_namesize == 0) {
/*
* We're the second shared reference to dl. Make a copy of
* dl_name in case the first thread goes away before we do.
* Note that we initialize the new name before storing its
* pointer into dl_name, because the first thread may load
* dl->dl_name at any time. He'll either see the old value,
* which is his, or the new shared copy; either is OK.
*/
dl->dl_namesize = strlen(dl->dl_name) + 1;
name = kmem_alloc(dl->dl_namesize, KM_SLEEP);
bcopy(dl->dl_name, name, dl->dl_namesize);
dl->dl_name = name;
}
mutex_exit(&dzp->z_lock);
/*
* We have a dirlock on the name. (Note that it is the dirlock,
* not the dzp's z_lock, that protects the name in the zap object.)
* See if there's an object by this name; if so, put a hold on it.
*/
if (flag & ZXATTR) {
zoid = dzp->z_phys->zp_xattr;
error = (zoid == 0 ? ENOENT : 0);
} else {
vp = dnlc_lookup(ZTOV(dzp), name);
if (vp == DNLC_NO_VNODE) {
VN_RELE(vp);
error = ENOENT;
} else if (vp) {
if (flag & ZNEW) {
zfs_dirent_unlock(dl);
VN_RELE(vp);
return (EEXIST);
}
*dlpp = dl;
*zpp = VTOZ(vp);
return (0);
} else {
error = zap_lookup(zfsvfs->z_os, dzp->z_id, name,
8, 1, &zoid);
if (error == ENOENT)
dnlc_update(ZTOV(dzp), name, DNLC_NO_VNODE);
}
}
if (error) {
if (error != ENOENT || (flag & ZEXISTS)) {
zfs_dirent_unlock(dl);
return (error);
}
} else {
if (flag & ZNEW) {
zfs_dirent_unlock(dl);
return (EEXIST);
}
error = zfs_zget(zfsvfs, zoid, zpp);
if (error) {
zfs_dirent_unlock(dl);
return (error);
}
if (!(flag & ZXATTR))
dnlc_update(ZTOV(dzp), name, ZTOV(*zpp));
}
*dlpp = dl;
return (0);
}
zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name, znode_t **zpp,
int flag)
#endif
{
zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
zfs_dirlock_t *dl;
uint64_t zoid;
int error;
vnode_t *vp;
#ifdef __APPLE__
char *name;
u_int8_t *nfc_name = NULL; /* NFC form of name */
int nfc_namesize = 0;
#endif
*zpp = NULL;
*dlpp = NULL;
#ifdef __APPLE__
/* Note: cnp will be NULL for ZXATTR case */
name = cnp ? cnp->cn_nameptr : "";
if (cnp)
ASSERT(name[cnp->cn_namelen] == '\0');
#endif
/*
* Verify that we are not trying to lock '.', '..', or '.zfs'
*/
if ((name[0] == '.') &&
((name[1] == '\0') || ((name[1] == '.') && (name[2] == '\0'))) ||
zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0)
return (EEXIST);
#ifdef __APPLE__
/*
* Mac OS X: store non-ascii names in UTF-8 NFC (pre-composed) on disk.
*
* The NFC name ptr is stored in dl->dl_name (allocated here)
* and its freed by zfs_dirent_unlock (since dl_namesize != 0).
*
* Since NFC size will not expand, we can allocate the same sized buffer.
*/
if (!is_ascii_str(name)) {
size_t outlen;
nfc_namesize = strlen(name) + 1;
nfc_name = kmem_alloc(nfc_namesize, KM_SLEEP);
if (utf8_normalizestr((const u_int8_t *)name, nfc_namesize, nfc_name,
&outlen, nfc_namesize, UTF_PRECOMPOSED) == 0) {
/* Normalization succeeded, switch to NFC name. */
name = (char *)nfc_name;
} else {
/* Normalization failed, just use input name as-is. */
kmem_free(nfc_name, nfc_namesize);
nfc_name = NULL;
}
}
#endif
/*
* Wait until there are no locks on this name.
*/
rw_enter(&dzp->z_name_lock, RW_READER);
mutex_enter(&dzp->z_lock);
for (;;) {
if (dzp->z_unlinked) {
mutex_exit(&dzp->z_lock);
rw_exit(&dzp->z_name_lock);
#ifdef __APPLE__
/* Release any unused NFC name before returning */
if (nfc_name) {
kmem_free(nfc_name, nfc_namesize);
}
#endif
return (ENOENT);
}
for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next)
if (strcmp(name, dl->dl_name) == 0)
break;
if (dl == NULL) {
/*
* Allocate a new dirlock and add it to the list.
*/
dl = kmem_alloc(sizeof (zfs_dirlock_t), KM_SLEEP);
cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL);
dl->dl_name = name;
dl->dl_sharecnt = 0;
dl->dl_namesize = 0;
dl->dl_dzp = dzp;
dl->dl_next = dzp->z_dirlocks;
dzp->z_dirlocks = dl;
#ifdef __APPLE__
/*
* Keep the NFC name around in dir lock by tagging it
* (setting nfc_namesize).
*/
if (nfc_name) {
dl->dl_namesize = nfc_namesize;
nfc_name = NULL; /* its now part of the dir lock */
}
#endif
break;
}
if ((flag & ZSHARED) && dl->dl_sharecnt != 0)
break;
cv_wait(&dl->dl_cv, &dzp->z_lock);
dl=NULL;
}
#ifdef __APPLE__
/*
* Release any unused NFC name (ie if we found a pre-existing lock entry)
*/
if (nfc_name) {
kmem_free(nfc_name, nfc_namesize);
nfc_name = NULL;
}
#endif
if ((flag & ZSHARED) && ++dl->dl_sharecnt > 1 && dl->dl_namesize == 0) {
/*
* We're the second shared reference to dl. Make a copy of
* dl_name in case the first thread goes away before we do.
* Note that we initialize the new name before storing its
* pointer into dl_name, because the first thread may load
* dl->dl_name at any time. He'll either see the old value,
* which is his, or the new shared copy; either is OK.
*/
dl->dl_namesize = strlen(dl->dl_name) + 1;
name = kmem_alloc(dl->dl_namesize, KM_SLEEP);
bcopy(dl->dl_name, name, dl->dl_namesize);
dl->dl_name = name;
}
mutex_exit(&dzp->z_lock);
/*
* We have a dirlock on the name. (Note that it is the dirlock,
* not the dzp's z_lock, that protects the name in the zap object.)
* See if there's an object by this name; if so, put a hold on it.
*/
if (flag & ZXATTR) {
error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &zoid,
sizeof (zoid));
if (error == 0)
error = (zoid == 0 ? ENOENT : 0);
} else {
#ifdef __APPLE__
/*
* Lookup an entry in the vnode name cache
*
* If the lookup succeeds, the vnode is returned in *vpp,
* and a status of -1 is returned.
*
* If the lookup determines that the name does not exist
* (negative caching), a status of ENOENT is returned.
*
* If the lookup fails, a status of zero is returned.
*/
switch ( cache_lookup(ZTOV(dzp), &vp, cnp) ) {
case -1:
break;
case ENOENT:
vp = DNLC_NO_VNODE;
break;
default:
vp = NULLVP;
}
#else
vp = dnlc_lookup(ZTOV(dzp), name);
#endif /* __APPLE__ */
if (vp == DNLC_NO_VNODE) {
VN_RELE(vp);
error = ENOENT;
} else if (vp) {
if (flag & ZNEW) {
zfs_dirent_unlock(dl);
VN_RELE(vp);
return (EEXIST);
}
*dlpp = dl;
*zpp = VTOZ(vp);
return (0);
} else {
error = zap_lookup(zfsvfs->z_os, dzp->z_id, name,
8, 1, &zoid);
zoid = ZFS_DIRENT_OBJ(zoid);
if (error == ENOENT)
#ifdef __APPLE__
/*
* Add a negative entry into the VFS name cache
*/
if ((flag & ZNEW) == 0 &&
(dzp->z_pflags & ZFS_XATTR) == 0 &&
(cnp) &&
(cnp->cn_flags & MAKEENTRY) &&
(cnp->cn_nameiop != CREATE) &&
(cnp->cn_nameiop != RENAME)) {
cache_enter(ZTOV(dzp), NULLVP, cnp);
}
#else
dnlc_update(ZTOV(dzp), name, DNLC_NO_VNODE);
#endif /* __APPLE__ */
}
}
if (error) {
if (error != ENOENT || (flag & ZEXISTS)) {
zfs_dirent_unlock(dl);
return (error);
}
} else {
if (flag & ZNEW) {
zfs_dirent_unlock(dl);
return (EEXIST);
}
//error = zfs_zget_sans_vnode(zfsvfs, zoid, zpp);
error = zfs_zget(zfsvfs, zoid, zpp);
if (error) {
zfs_dirent_unlock(dl);
return (error);
} else {
// Should this be here?
//printf("zfs_dir attach 1\n");
//zfs_attach_vnode(*zpp);
}
if (!(flag & ZXATTR))
#ifdef __APPLE__
if (cnp && cnp->cn_flags & MAKEENTRY)
cache_enter(ZTOV(dzp), ZTOV(*zpp), cnp);
#else
dnlc_update(ZTOV(dzp), name, ZTOV(*zpp));
#endif /* __APPLE__ */
}
*dlpp = dl;
return (0);
}
/*
* Return a znode for the extended attribute directory for zp.
* ** If the directory does not already exist, it is created **
*
* IN: zp - znode to obtain attribute directory from
* cr - credentials of caller
* flags - flags from the VOP_LOOKUP call
*
* OUT: xzpp - pointer to extended attribute znode
*
* RETURN: 0 on success
* error number on failure
*/
int
zfs_get_xattrdir(znode_t *zp, vnode_t **xvpp, cred_t *cr, int flags)
{
zfsvfs_t *zfsvfs = zp->z_zfsvfs;
znode_t *xzp;
zfs_dirlock_t *dl;
vattr_t va;
int error;
//printf("zfs_get_xattrdir\n");
top:
#ifdef __APPLE__
error = zfs_dirent_lock(&dl, zp, NULL, &xzp, ZXATTR);
#else
error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR);
#endif
if (error)
return (error);
if (xzp != NULL) {
*xvpp = ZTOV(xzp);
zfs_dirent_unlock(dl);
return (0);
}
//ASSERT(zp->z_phys->zp_xattr == 0);
if (!(flags & CREATE_XATTR_DIR)) {
zfs_dirent_unlock(dl);
return (ENOENT);
}
#ifdef __APPLE__
if (vfs_isrdonly(zfsvfs->z_vfs))
#else
if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY)
#endif
{
zfs_dirent_unlock(dl);
return (EROFS);
}
/*
* The ability to 'create' files in an attribute
* directory comes from the write_xattr permission on the base file.
*
* The ability to 'search' an attribute directory requires
* read_xattr permission on the base file.
*
* Once in a directory the ability to read/write attributes
* is controlled by the permissions on the attribute file.
*/
va.va_mask = AT_TYPE | AT_MODE | AT_UID | AT_GID;
va.va_type = VDIR;
va.va_mode = S_IFDIR | S_ISVTX | 0777;
zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
error = zfs_make_xattrdir(zp, &va, xvpp, cr);
zfs_dirent_unlock(dl);
if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
/* NB: we already did dmu_tx_wait() if necessary */
goto top;
}
return (error);
}
/*
* Lookup/Create an extended attribute entry.
*
* Input arguments:
* dzp - znode for hidden attribute directory
* name - name of attribute
* flag - ZNEW: if the entry already exists, fail with EEXIST.
* ZEXISTS: if the entry does not exist, fail with ENOENT.
*
* Output arguments:
* vpp - pointer to the vnode for the entry (NULL if there isn't one)
*
* Return value: 0 on success or errno value on failure.
*/
int
zfs_obtain_xattr(znode_t *dzp, const char *name, mode_t mode, cred_t *cr,
vnode_t **vpp, int flag)
{
znode_t *xzp = NULL;
zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
zilog_t *zilog;
zfs_dirlock_t *dl;
dmu_tx_t *tx;
struct vnode_attr vattr;
int error;
struct componentname cn;
zfs_acl_ids_t acl_ids;
/* zfs_dirent_lock() expects a component name */
bzero(&cn, sizeof (cn));
cn.cn_nameiop = LOOKUP;
cn.cn_flags = ISLASTCN;
cn.cn_nameptr = (char *)name;
cn.cn_namelen = strlen(name);
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(dzp);
zilog = zfsvfs->z_log;
VATTR_INIT(&vattr);
VATTR_SET(&vattr, va_type, VREG);
VATTR_SET(&vattr, va_mode, mode & ~S_IFMT);
if ((error = zfs_acl_ids_create(dzp, 0,
&vattr, cr, NULL, &acl_ids)) != 0) {
ZFS_EXIT(zfsvfs);
return (error);
}
top:
/* Lock the attribute entry name. */
if ( (error = zfs_dirent_lock(&dl, dzp, (char *)name, &xzp, flag,
NULL, &cn)) ) {
goto out;
}
/* If the name already exists, we're done. */
if (xzp != NULL) {
zfs_dirent_unlock(dl);
goto out;
}
tx = dmu_tx_create(zfsvfs->z_os);
dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
//dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
//dmu_tx_hold_bonus(tx, dzp->z_id);
dmu_tx_hold_zap(tx, dzp->z_id, TRUE, (char *)name);
#if 1 // FIXME
if (dzp->z_pflags & ZFS_INHERIT_ACE) {
dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
}
#endif
zfs_sa_upgrade_txholds(tx, dzp);
error = dmu_tx_assign(tx, TXG_NOWAIT);
if (error) {
zfs_dirent_unlock(dl);
if (error == ERESTART) {
dmu_tx_wait(tx);
dmu_tx_abort(tx);
goto top;
}
dmu_tx_abort(tx);
goto out;
}
zfs_mknode(dzp, &vattr, tx, cr, 0, &xzp, &acl_ids);
/*
ASSERT(xzp->z_id == zoid);
*/
(void) zfs_link_create(dl, xzp, tx, ZNEW);
zfs_log_create(zilog, tx, TX_CREATE, dzp, xzp, (char *)name,
NULL /* vsecp */, 0 /*acl_ids.z_fuidp*/, &vattr);
zfs_acl_ids_free(&acl_ids);
dmu_tx_commit(tx);
zfs_znode_wait_vnode(xzp);
zfs_dirent_unlock(dl);
out:
if (error == EEXIST)
error = ENOATTR;
if (xzp)
*vpp = ZTOV(xzp);
ZFS_EXIT(zfsvfs);
return (error);
}
