static inline int make_file_safe(struct vfs_fsal_obj_handle *dir_hdl,
const struct req_op_context *opctx,
int dir_fd, const char *name, mode_t unix_mode,
uid_t user, gid_t group,
struct vfs_fsal_obj_handle **hdl)
{
int retval;
struct stat stat;
vfs_file_handle_t *fh;
vfs_alloc_handle(fh);
retval = fchownat(dir_fd, name, user, group, AT_SYMLINK_NOFOLLOW);
if (retval < 0)
goto fileerr;
/* now that it is owned properly, set accessible mode */
retval = fchmodat(dir_fd, name, unix_mode, 0);
if (retval < 0)
goto fileerr;
retval = vfs_name_to_handle(dir_fd, dir_hdl->obj_handle.fs, name, fh);
if (retval < 0)
goto fileerr;
retval = fstatat(dir_fd, name, &stat, AT_SYMLINK_NOFOLLOW);
if (retval < 0)
goto fileerr;
/* allocate an obj_handle and fill it up */
*hdl = alloc_handle(dir_fd, fh, dir_hdl->obj_handle.fs, &stat,
dir_hdl->handle, name, opctx->fsal_export);
if (*hdl == NULL)
return ENOMEM;
return 0;
fileerr:
retval = errno;
return retval;
}
fsal_status_t vfs_open2(struct fsal_obj_handle *obj_hdl,
struct state_t *state,
fsal_openflags_t openflags,
enum fsal_create_mode createmode,
const char *name,
struct attrlist *attrib_set,
fsal_verifier_t verifier,
struct fsal_obj_handle **new_obj,
struct attrlist *attrs_out,
bool *caller_perm_check)
{
int posix_flags = 0;
int fd, dir_fd;
int retval = 0;
mode_t unix_mode;
fsal_status_t status = {0, 0};
struct vfs_fd *my_fd = NULL;
struct vfs_fsal_obj_handle *myself, *hdl = NULL;
struct stat stat;
vfs_file_handle_t *fh = NULL;
bool truncated;
bool created = false;
if (state != NULL)
my_fd = (struct vfs_fd *)(state + 1);
myself = container_of(obj_hdl, struct vfs_fsal_obj_handle, obj_handle);
LogAttrlist(COMPONENT_FSAL, NIV_FULL_DEBUG,
"attrib_set ", attrib_set, false);
fsal2posix_openflags(openflags, &posix_flags);
truncated = (posix_flags & O_TRUNC) != 0;
LogFullDebug(COMPONENT_FSAL,
truncated ? "Truncate" : "No truncate");
if (createmode >= FSAL_EXCLUSIVE) {
/* Now fixup attrs for verifier if exclusive create */
set_common_verifier(attrib_set, verifier);
}
if (name == NULL) {
/* This is an open by handle */
struct vfs_fsal_obj_handle *myself;
myself = container_of(obj_hdl,
struct vfs_fsal_obj_handle,
obj_handle);
if (obj_hdl->fsal != obj_hdl->fs->fsal) {
LogDebug(COMPONENT_FSAL,
"FSAL %s operation for handle belonging to FSAL %s, return EXDEV",
obj_hdl->fsal->name, obj_hdl->fs->fsal->name);
return fsalstat(posix2fsal_error(EXDEV), EXDEV);
}
if (state != NULL) {
/* Prepare to take the share reservation, but only if we
* are called with a valid state (if state is NULL the
* caller is a stateless create such as NFS v3 CREATE).
*/
/* This can block over an I/O operation. */
PTHREAD_RWLOCK_wrlock(&obj_hdl->lock);
/* Check share reservation conflicts. */
status = check_share_conflict(&myself->u.file.share,
openflags,
false);
if (FSAL_IS_ERROR(status)) {
PTHREAD_RWLOCK_unlock(&obj_hdl->lock);
return status;
}
/* Take the share reservation now by updating the
* counters.
*/
update_share_counters(&myself->u.file.share,
FSAL_O_CLOSED,
openflags);
PTHREAD_RWLOCK_unlock(&obj_hdl->lock);
} else {
/* We need to use the global fd to continue, and take
* the lock to protect it.
*/
my_fd = &hdl->u.file.fd;
PTHREAD_RWLOCK_wrlock(&obj_hdl->lock);
}
status = vfs_open_my_fd(myself, openflags, posix_flags, my_fd);
if (FSAL_IS_ERROR(status)) {
if (state == NULL) {
/* Release the lock taken above, and return
* since there is nothing to undo.
*/
PTHREAD_RWLOCK_unlock(&obj_hdl->lock);
return status;
} else {
/* Error - need to release the share */
goto undo_share;
}
}
if (createmode >= FSAL_EXCLUSIVE || truncated) {
/* Refresh the attributes */
struct stat stat;
retval = fstat(my_fd->fd, &stat);
if (retval == 0) {
LogFullDebug(COMPONENT_FSAL,
"New size = %" PRIx64,
stat.st_size);
} else {
if (errno == EBADF)
errno = ESTALE;
status = fsalstat(posix2fsal_error(errno),
errno);
}
/* Now check verifier for exclusive, but not for
* FSAL_EXCLUSIVE_9P.
*/
if (!FSAL_IS_ERROR(status) &&
createmode >= FSAL_EXCLUSIVE &&
createmode != FSAL_EXCLUSIVE_9P &&
!check_verifier_stat(&stat, verifier)) {
/* Verifier didn't match, return EEXIST */
status =
fsalstat(posix2fsal_error(EEXIST), EEXIST);
}
}
if (state == NULL) {
/* If no state, release the lock taken above and return
* status.
*/
PTHREAD_RWLOCK_unlock(&obj_hdl->lock);
return status;
}
if (!FSAL_IS_ERROR(status)) {
/* Return success. */
return status;
}
(void) vfs_close_my_fd(my_fd);
undo_share:
/* Can only get here with state not NULL and an error */
/* On error we need to release our share reservation
* and undo the update of the share counters.
* This can block over an I/O operation.
*/
PTHREAD_RWLOCK_wrlock(&obj_hdl->lock);
update_share_counters(&myself->u.file.share,
openflags,
FSAL_O_CLOSED);
PTHREAD_RWLOCK_unlock(&obj_hdl->lock);
return status;
}
/* In this path where we are opening by name, we can't check share
* reservation yet since we don't have an object_handle yet. If we
* indeed create the object handle (there is no race with another
* open by name), then there CAN NOT be a share conflict, otherwise
* the share conflict will be resolved when the object handles are
* merged.
*/
#ifdef ENABLE_VFS_DEBUG_ACL
if (createmode != FSAL_NO_CREATE) {
/* Need to ammend attributes for inherited ACL, these will be
* set later. We also need to test for permission to create
* since there might be an ACL.
*/
struct attrlist attrs;
fsal_accessflags_t access_type;
access_type = FSAL_MODE_MASK_SET(FSAL_W_OK) |
FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_ADD_FILE);
status = obj_hdl->obj_ops.test_access(obj_hdl, access_type,
NULL, NULL, false);
if (FSAL_IS_ERROR(status))
return status;
fsal_prepare_attrs(&attrs, ATTR_ACL);
status = obj_hdl->obj_ops.getattrs(obj_hdl, &attrs);
if (FSAL_IS_ERROR(status))
return status;
status.major = fsal_inherit_acls(attrib_set, attrs.acl,
FSAL_ACE_FLAG_FILE_INHERIT);
/* Done with the attrs */
fsal_release_attrs(&attrs);
if (FSAL_IS_ERROR(status))
return status;
}
#endif /* ENABLE_VFS_DEBUG_ACL */
if (createmode != FSAL_NO_CREATE) {
/* Now add in O_CREAT and O_EXCL. */
posix_flags |= O_CREAT;
/* And if we are at least FSAL_GUARDED, do an O_EXCL create. */
if (createmode >= FSAL_GUARDED)
posix_flags |= O_EXCL;
/* Fetch the mode attribute to use in the openat system call. */
unix_mode = fsal2unix_mode(attrib_set->mode) &
~op_ctx->fsal_export->exp_ops.fs_umask(op_ctx->fsal_export);
/* Don't set the mode if we later set the attributes */
FSAL_UNSET_MASK(attrib_set->mask, ATTR_MODE);
}
if (createmode == FSAL_UNCHECKED && (attrib_set->mask != 0)) {
/* If we have FSAL_UNCHECKED and want to set more attributes
* than the mode, we attempt an O_EXCL create first, if that
* succeeds, then we will be allowed to set the additional
* attributes, otherwise, we don't know we created the file
* and this can NOT set the attributes.
*/
posix_flags |= O_EXCL;
}
dir_fd = vfs_fsal_open(myself, O_PATH | O_NOACCESS, &status.major);
if (dir_fd < 0)
return fsalstat(status.major, -dir_fd);
/** @todo: not sure what this accomplishes... */
retval = vfs_stat_by_handle(dir_fd, &stat);
if (retval < 0) {
retval = errno;
status = fsalstat(posix2fsal_error(retval), retval);
goto direrr;
}
/* Become the user because we are creating an object in this dir.
*/
if (createmode != FSAL_NO_CREATE)
fsal_set_credentials(op_ctx->creds);
if ((posix_flags & O_CREAT) != 0)
fd = openat(dir_fd, name, posix_flags, unix_mode);
else
fd = openat(dir_fd, name, posix_flags);
if (fd == -1 && errno == EEXIST && createmode == FSAL_UNCHECKED) {
/* We tried to create O_EXCL to set attributes and failed.
* Remove O_EXCL and retry. We still try O_CREAT again just in
* case file disappears out from under us.
*
* Note that because we have dropped O_EXCL, later on we will
* not assume we created the file, and thus will not set
* additional attributes. We don't need to separately track
* the condition of not wanting to set attributes.
*/
posix_flags &= ~O_EXCL;
fd = openat(dir_fd, name, posix_flags, unix_mode);
}
/* Preserve errno */
retval = errno;
/* If we were creating, restore credentials now. */
if (createmode != FSAL_NO_CREATE)
fsal_restore_ganesha_credentials();
if (fd < 0) {
status = fsalstat(posix2fsal_error(retval), retval);
goto direrr;
}
/* Remember if we were responsible for creating the file.
* Note that in an UNCHECKED retry we MIGHT have re-created the
* file and won't remember that. Oh well, so in that rare case we
* leak a partially created file if we have a subsequent error in here.
* Also notify caller to do permission check if we DID NOT create the
* file. Note it IS possible in the case of a race between an UNCHECKED
* open and an external unlink, we did create the file, but we will
* still force a permission check. Of course that permission check
* SHOULD succeed since we also won't set the mode the caller requested
* and the default file create permissions SHOULD allow the owner
* read/write.
*/
created = (posix_flags & O_EXCL) != 0;
*caller_perm_check = !created;
vfs_alloc_handle(fh);
retval = vfs_name_to_handle(dir_fd, obj_hdl->fs, name, fh);
if (retval < 0) {
retval = errno;
status = fsalstat(posix2fsal_error(retval), retval);
goto fileerr;
}
retval = fstat(fd, &stat);
if (retval < 0) {
retval = errno;
status = fsalstat(posix2fsal_error(retval), retval);
goto fileerr;
}
/* allocate an obj_handle and fill it up */
hdl = alloc_handle(dir_fd, fh, obj_hdl->fs, &stat, myself->handle, name,
op_ctx->fsal_export);
if (hdl == NULL) {
status = fsalstat(posix2fsal_error(ENOMEM), ENOMEM);
goto fileerr;
}
/* If we didn't have a state above, use the global fd. At this point,
* since we just created the global fd, no one else can have a
* reference to it, and thus we can mamnipulate unlocked which is
* handy since we can then call setattr2 which WILL take the lock
* without a double locking deadlock.
*/
if (my_fd == NULL)
my_fd = &hdl->u.file.fd;
my_fd->fd = fd;
my_fd->openflags = openflags;
*new_obj = &hdl->obj_handle;
if (created && attrib_set->mask != 0) {
/* Set attributes using our newly opened file descriptor as the
* share_fd if there are any left to set (mode and truncate
* have already been handled).
*
* Note that we only set the attributes if we were responsible
* for creating the file and we have attributes to set.
*
* Note if we have ENABLE_VFS_DEBUG_ACL an inherited ACL might
* be part of the attributes we are setting here.
*/
status = (*new_obj)->obj_ops.setattr2(*new_obj,
false,
state,
attrib_set);
if (FSAL_IS_ERROR(status)) {
/* Release the handle we just allocated. */
(*new_obj)->obj_ops.release(*new_obj);
*new_obj = NULL;
goto fileerr;
}
if (attrs_out != NULL) {
status = (*new_obj)->obj_ops.getattrs(*new_obj,
attrs_out);
if (FSAL_IS_ERROR(status) &&
(attrs_out->mask & ATTR_RDATTR_ERR) == 0) {
/* Get attributes failed and caller expected
* to get the attributes. Otherwise continue
* with attrs_out indicating ATTR_RDATTR_ERR.
*/
goto fileerr;
}
}
} else if (attrs_out != NULL) {
/* Since we haven't set any attributes other than what was set
* on create (if we even created), just use the stat results
* we used to create the fsal_obj_handle.
*/
posix2fsal_attributes(&stat, attrs_out);
/* Make sure ATTR_RDATTR_ERR is cleared on success. */
attrs_out->mask &= ~ATTR_RDATTR_ERR;
}
close(dir_fd);
if (state != NULL) {
/* Prepare to take the share reservation, but only if we are
* called with a valid state (if state is NULL the caller is
* a stateless create such as NFS v3 CREATE).
*/
/* This can block over an I/O operation. */
PTHREAD_RWLOCK_wrlock(&(*new_obj)->lock);
/* Take the share reservation now by updating the counters. */
update_share_counters(&hdl->u.file.share,
FSAL_O_CLOSED,
openflags);
PTHREAD_RWLOCK_unlock(&(*new_obj)->lock);
}
return fsalstat(ERR_FSAL_NO_ERROR, 0);
fileerr:
close(fd);
/* Delete the file if we actually created it. */
if (created)
unlinkat(dir_fd, name, 0);
direrr:
close(dir_fd);
return fsalstat(posix2fsal_error(retval), retval);
}
static fsal_status_t makesymlink(struct fsal_obj_handle *dir_hdl,
const char *name, const char *link_path,
struct attrlist *attrib,
struct fsal_obj_handle **handle)
{
struct vfs_fsal_obj_handle *myself, *hdl;
int dir_fd = -1;
struct stat stat;
fsal_errors_t fsal_error = ERR_FSAL_NO_ERROR;
int retval = 0;
int flags = O_PATH | O_NOACCESS;
vfs_file_handle_t *fh = NULL;
vfs_alloc_handle(fh);
LogDebug(COMPONENT_FSAL, "create %s", name);
*handle = NULL; /* poison it first */
if (!dir_hdl->obj_ops.handle_is(dir_hdl, DIRECTORY)) {
LogCrit(COMPONENT_FSAL,
"Parent handle is not a directory. hdl = 0x%p",
dir_hdl);
return fsalstat(ERR_FSAL_NOTDIR, 0);
}
myself = container_of(dir_hdl, struct vfs_fsal_obj_handle, obj_handle);
if (dir_hdl->fsal != dir_hdl->fs->fsal) {
LogDebug(COMPONENT_FSAL,
"FSAL %s operation for handle belonging to FSAL %s, return EXDEV",
dir_hdl->fsal->name,
dir_hdl->fs->fsal != NULL
? dir_hdl->fs->fsal->name
: "(none)");
retval = EXDEV;
goto hdlerr;
}
dir_fd = vfs_fsal_open(myself, flags, &fsal_error);
if (dir_fd < 0)
return fsalstat(fsal_error, -dir_fd);
flags |= O_NOFOLLOW; /* BSD needs O_NOFOLLOW for
* fhopen() of symlinks */
retval = vfs_stat_by_handle(dir_fd, myself->handle, &stat, flags);
if (retval < 0) {
retval = errno;
goto direrr;
}
/* Become the user because we are creating an object in this dir.
*/
fsal_set_credentials(op_ctx->creds);
retval = symlinkat(link_path, dir_fd, name);
if (retval < 0) {
retval = errno;
fsal_restore_ganesha_credentials();
goto direrr;
}
fsal_restore_ganesha_credentials();
retval = vfs_name_to_handle(dir_fd, dir_hdl->fs, name, fh);
if (retval < 0) {
retval = errno;
goto linkerr;
}
/* now get attributes info,
* being careful to get the link, not the target */
retval = fstatat(dir_fd, name, &stat, AT_SYMLINK_NOFOLLOW);
if (retval < 0) {
retval = errno;
goto linkerr;
}
/* allocate an obj_handle and fill it up */
hdl = alloc_handle(dir_fd, fh, dir_hdl->fs, &stat, NULL, name,
op_ctx->fsal_export);
if (hdl == NULL) {
retval = ENOMEM;
goto linkerr;
}
*handle = &hdl->obj_handle;
close(dir_fd);
return fsalstat(ERR_FSAL_NO_ERROR, 0);
linkerr:
unlinkat(dir_fd, name, 0);
direrr:
close(dir_fd);
hdlerr:
if (retval == ENOENT)
fsal_error = ERR_FSAL_STALE;
else
fsal_error = posix2fsal_error(retval);
return fsalstat(fsal_error, retval);
}
static fsal_status_t makedir(struct fsal_obj_handle *dir_hdl,
const char *name, struct attrlist *attrib,
struct fsal_obj_handle **handle)
{
struct vfs_fsal_obj_handle *myself, *hdl;
int dir_fd;
struct stat stat;
mode_t unix_mode;
fsal_errors_t fsal_error = ERR_FSAL_NO_ERROR;
int retval = 0;
int flags = O_PATH | O_NOACCESS;
vfs_file_handle_t *fh = NULL;
vfs_alloc_handle(fh);
LogDebug(COMPONENT_FSAL, "create %s", name);
*handle = NULL; /* poison it */
if (!dir_hdl->obj_ops.handle_is(dir_hdl, DIRECTORY)) {
LogCrit(COMPONENT_FSAL,
"Parent handle is not a directory. hdl = 0x%p",
dir_hdl);
return fsalstat(ERR_FSAL_NOTDIR, 0);
}
myself = container_of(dir_hdl, struct vfs_fsal_obj_handle, obj_handle);
if (dir_hdl->fsal != dir_hdl->fs->fsal) {
LogDebug(COMPONENT_FSAL,
"FSAL %s operation for handle belonging to FSAL %s, return EXDEV",
dir_hdl->fsal->name,
dir_hdl->fs->fsal != NULL
? dir_hdl->fs->fsal->name
: "(none)");
retval = EXDEV;
goto hdlerr;
}
unix_mode = fsal2unix_mode(attrib->mode)
& ~op_ctx->fsal_export->exp_ops.fs_umask(op_ctx->fsal_export);
dir_fd = vfs_fsal_open(myself, flags, &fsal_error);
if (dir_fd < 0)
return fsalstat(fsal_error, -dir_fd);
retval = vfs_stat_by_handle(dir_fd, myself->handle, &stat, flags);
if (retval < 0) {
retval = errno;
goto direrr;
}
/* Become the user because we are creating an object in this dir.
*/
fsal_set_credentials(op_ctx->creds);
retval = mkdirat(dir_fd, name, unix_mode);
if (retval < 0) {
retval = errno;
fsal_restore_ganesha_credentials();
goto direrr;
}
fsal_restore_ganesha_credentials();
retval = vfs_name_to_handle(dir_fd, dir_hdl->fs, name, fh);
if (retval < 0) {
retval = errno;
goto fileerr;
}
retval = fstatat(dir_fd, name, &stat, AT_SYMLINK_NOFOLLOW);
if (retval < 0) {
retval = errno;
goto fileerr;
}
/* allocate an obj_handle and fill it up */
hdl = alloc_handle(dir_fd, fh, dir_hdl->fs, &stat,
myself->handle, name,
op_ctx->fsal_export);
if (hdl == NULL) {
retval = ENOMEM;
goto fileerr;
}
*handle = &hdl->obj_handle;
close(dir_fd);
return fsalstat(ERR_FSAL_NO_ERROR, 0);
fileerr:
unlinkat(dir_fd, name, 0);
direrr:
close(dir_fd);
hdlerr:
fsal_error = posix2fsal_error(retval);
return fsalstat(fsal_error, retval);
}
static fsal_status_t lookup(struct fsal_obj_handle *parent,
const char *path, struct fsal_obj_handle **handle)
{
struct vfs_fsal_obj_handle *parent_hdl, *hdl;
fsal_errors_t fsal_error = ERR_FSAL_NO_ERROR;
int retval, dirfd;
struct stat stat;
vfs_file_handle_t *fh = NULL;
vfs_alloc_handle(fh);
fsal_dev_t dev;
struct fsal_filesystem *fs;
bool xfsal = false;
*handle = NULL; /* poison it first */
parent_hdl =
container_of(parent, struct vfs_fsal_obj_handle, obj_handle);
if (!parent->obj_ops.handle_is(parent, DIRECTORY)) {
LogCrit(COMPONENT_FSAL,
"Parent handle is not a directory. hdl = 0x%p", parent);
return fsalstat(ERR_FSAL_NOTDIR, 0);
}
if (parent->fsal != parent->fs->fsal) {
LogDebug(COMPONENT_FSAL,
"FSAL %s operation for handle belonging to FSAL %s, return EXDEV",
parent->fsal->name,
parent->fs->fsal != NULL
? parent->fs->fsal->name
: "(none)");
retval = EXDEV;
goto hdlerr;
}
fs = parent->fs;
dirfd = vfs_fsal_open(parent_hdl, O_PATH | O_NOACCESS, &fsal_error);
if (dirfd < 0)
return fsalstat(fsal_error, -dirfd);
retval = fstatat(dirfd, path, &stat, AT_SYMLINK_NOFOLLOW);
if (retval < 0) {
retval = errno;
goto direrr;
}
dev = posix2fsal_devt(stat.st_dev);
if ((dev.minor != parent_hdl->dev.minor) ||
(dev.major != parent_hdl->dev.major)) {
/* XDEV */
fs = lookup_dev(&dev);
if (fs == NULL) {
LogDebug(COMPONENT_FSAL,
"Lookup of %s crosses filesystem boundary to "
"unknown file system dev=%"PRIu64".%"PRIu64,
path, dev.major, dev.minor);
retval = EXDEV;
goto direrr;
}
if (fs->fsal != parent->fsal) {
xfsal = true;
LogDebug(COMPONENT_FSAL,
"Lookup of %s crosses filesystem boundary to file system %s into FSAL %s",
path, fs->path,
fs->fsal != NULL
? fs->fsal->name
: "(none)");
} else {
LogDebug(COMPONENT_FSAL,
"Lookup of %s crosses filesystem boundary to file system %s",
path, fs->path);
}
}
if (xfsal || vfs_name_to_handle(dirfd, fs, path, fh) < 0) {
retval = errno;
if (((retval == ENOTTY) ||
(retval == EOPNOTSUPP) ||
(retval == ENOTSUP) ||
xfsal) &&
(fs != parent->fs)) {
/* Crossed device into territory not handled by
* this FSAL (XFS or VFS). Need to invent a handle.
* The made up handle will be JUST the fsid, we
* do not expect to see the handle on the wire, and
* this handle will not be valid for any form of this
* FSAL.
*/
LogDebug(COMPONENT_FSAL,
"vfs_name_to_handle %s, inventing FSAL %s handle for FSAL %s filesystem %s",
xfsal ? "skipped" : "failed",
parent->fsal->name,
fs->fsal != NULL
? fs->fsal->name
: "(none)",
path);
retval = vfs_encode_dummy_handle(fh, fs);
if (retval < 0) {
retval = errno;
goto direrr;
}
retval = 0;
} else {
/* Some other error */
goto direrr;
}
}
/* allocate an obj_handle and fill it up */
hdl = alloc_handle(dirfd, fh, fs, &stat, parent_hdl->handle, path,
op_ctx->fsal_export);
close(dirfd);
if (hdl == NULL) {
retval = ENOMEM;
goto hdlerr;
}
*handle = &hdl->obj_handle;
return fsalstat(ERR_FSAL_NO_ERROR, 0);
direrr:
close(dirfd);
hdlerr:
fsal_error = posix2fsal_error(retval);
return fsalstat(fsal_error, retval);
}
