Request Handlers

(You can use the Index to directly jump to a specific handler).

class pyfuse3.Operations

This class defines the request handler methods that an pyfuse3 file system may implement. If a particular request handler has not been implemented, it must raise FUSEError with an errorcode of errno.ENOSYS. Further requests of this type will then be handled directly by the FUSE kernel module without calling the handler again.

The only exception that request handlers are allowed to raise is FUSEError. This will cause the specified errno to be returned by the syscall that is being handled.

It is recommended that file systems are derived from this class and only overwrite the handlers that they actually implement. (The methods defined in this class all just raise FUSEError(ENOSYS) or do nothing).

supports_dot_lookup = True

If set, indicates that the filesystem supports lookup of the . and .. entries. This is required if the file system will be shared over NFS.

enable_writeback_cache = True

Enables write-caching in the kernel if available. This means that individual write request may be buffered and merged in the kernel before they are send to the filesystem.

enable_acl = False

Enable ACL support. When enabled, the kernel will cache and have responsibility for enforcing ACLs. ACL will be stored as xattrs and passed to userspace, which is responsible for updating the ACLs in the filesystem, keeping the file mode in sync with the ACL, and ensuring inheritance of default ACLs when new filesystem nodes are created. Note that this requires that the file system is able to parse and interpret the xattr representation of ACLs.

Enabling this feature implicitly turns on the default_permissions option.
coroutine access(inode, mode, ctx)

Check if requesting process has mode rights on inode.

ctx will be a RequestContext instance.

The method must return a boolean value.

If the default_permissions mount option is given, this method is not called.

When implementing this method, the get_sup_groups function may be useful.

coroutine create(parent_inode, name, mode, flags, ctx)

Create a file with permissions mode and open it with flags

ctx will be a RequestContext instance.

The method must return a tuple of the form (fi, attr), where fi is a FileInfo instance handle like the one returned by open and attr is an EntryAttributes instance with the attributes of the newly created directory entry.

(Successful) execution of this handler increases the lookup count for the returned inode by one.

coroutine flush(fh)

Handle close() syscall.

fh will by an integer filehandle returned by a prior open or create call.

This method is called whenever a file descriptor is closed. It may be called multiple times for the same open file (e.g. if the file handle has been duplicated).

coroutine forget(inode_list)

Decrease lookup counts for inodes in inode_list

inode_list is a list of (inode, nlookup) tuples. This method should reduce the lookup count for each inode by nlookup.

If the lookup count reaches zero, the inode is currently not known to the kernel. In this case, the file system will typically check if there are still directory entries referring to this inode and, if not, remove the inode.

If the file system is unmounted, it may not have received forget calls to bring all lookup counts to zero. The filesystem needs to take care to clean up inodes that at that point still have non-zero lookup count (e.g. by explicitly calling forget with the current lookup count for every such inode after main has returned).

This method must not raise any exceptions (not even FUSEError), since it is not handling a particular client request.

coroutine fsync(fh, datasync)

Flush buffers for open file fh

If datasync is true, only the file contents should be flushed (in contrast to the metadata about the file).

fh will by an integer filehandle returned by a prior open or create call.

coroutine fsyncdir(fh, datasync)

Flush buffers for open directory fh

If datasync is true, only the directory contents should be flushed (in contrast to metadata about the directory itself).

coroutine getattr(inode, ctx)

Get attributes for inode

ctx will be a RequestContext instance.

This method should return an EntryAttributes instance with the attributes of inode. The entry_timeout attribute is ignored in this context.

coroutine getxattr(inode, name, ctx)

Return extended attribute name of inode

ctx will be a RequestContext instance.

If the attribute does not exist, the method must raise FUSEError with an error code of ENOATTR. name will be of type bytes, but is guaranteed not to contain zero-bytes (\0).

init()

Initialize operations

This method will be called just before the file system starts handling requests. It must not raise any exceptions (not even FUSEError), since it is not handling a particular client request.

Create directory entry name in parent_inode refering to inode.

ctx will be a RequestContext instance.

The method must return an EntryAttributes instance with the attributes of the newly created directory entry.

(Successful) execution of this handler increases the lookup count for the returned inode by one.

coroutine listxattr(inode, ctx)

Get list of extended attributes for inode

ctx will be a RequestContext instance.

This method must return a sequence of bytes objects. The objects must not include zero-bytes (\0).

coroutine lookup(parent_inode, name, ctx)

Look up a directory entry by name and get its attributes.

This method should return an EntryAttributes instance for the directory entry name in the directory with inode parent_inode.

If there is no such entry, the method should either return an EntryAttributes instance with zero st_ino value (in which case the negative lookup will be cached as specified by entry_timeout), or it should raise FUSEError with an errno of errno.ENOENT (in this case the negative result will not be cached).

ctx will be a RequestContext instance.

The file system must be able to handle lookups for . and .., no matter if these entries are returned by readdir or not.

(Successful) execution of this handler increases the lookup count for the returned inode by one.

coroutine mkdir(parent_inode, name, mode, ctx)

Create a directory

This method must create a new directory name with mode mode in the directory with inode parent_inode. ctx will be a RequestContext instance.

This method must return an EntryAttributes instance with the attributes of the newly created directory entry.

(Successful) execution of this handler increases the lookup count for the returned inode by one.

coroutine mknod(parent_inode, name, mode, rdev, ctx)

Create (possibly special) file

This method must create a (special or regular) file name in the directory with inode parent_inode. Whether the file is special or regular is determined by its mode. If the file is neither a block nor character device, rdev can be ignored. ctx will be a RequestContext instance.

The method must return an EntryAttributes instance with the attributes of the newly created directory entry.

(Successful) execution of this handler increases the lookup count for the returned inode by one.

coroutine open(inode, flags, ctx)

Open a inode inode with flags.

ctx will be a RequestContext instance.

flags will be a bitwise or of the open flags described in the open(2) manpage and defined in the os module (with the exception of O_CREAT, O_EXCL, O_NOCTTY and O_TRUNC)

This method must return a FileInfo instance. The FileInfo.fh field must contain an integer file handle, which will be passed to the read, write, flush, fsync and release methods to identify the open file. The FileInfo instance may also have relevant configuration attributes set; see the FileInfo documentation for more information.

coroutine opendir(inode, ctx)

Open the directory with inode inode

ctx will be a RequestContext instance.

This method should return an integer file handle. The file handle will be passed to the readdir, fsyncdir and releasedir methods to identify the directory.

coroutine read(fh, off, size)

Read size bytes from fh at position off

fh will by an integer filehandle returned by a prior open or create call.

This function should return exactly the number of bytes requested except on EOF or error, otherwise the rest of the data will be substituted with zeroes.

coroutine readdir(fh, start_id, token)

Read entries in open directory fh.

This method should list the contents of directory fh (as returned by a prior opendir call), starting at the entry identified by start_id.

Instead of returning the directory entries directly, the method must call readdir_reply for each directory entry. If readdir_reply returns True, the file system must increase the lookup count for the provided directory entry by one and call readdir_reply again for the next entry (if any). If readdir_reply returns False, the lookup count must not be increased and the method should return without further calls to readdir_reply.

The start_id parameter will be either zero (in which case listing should begin with the first entry) or it will correspond to a value that was previously passed by the file system to the readdir_reply function in the next_id parameter.

If entries are added or removed during a readdir cycle, they may or may not be returned. However, they must not cause other entries to be skipped or returned more than once.

. and .. entries may be included but are not required. However, if they are reported the filesystem must not increase the lookup count for the corresponding inodes (even if readdir_reply returns True).

Return target of symbolic link inode.

ctx will be a RequestContext instance.

coroutine release(fh)

Release open file

This method will be called when the last file descriptor of fh has been closed, i.e. when the file is no longer opened by any client process.

fh will by an integer filehandle returned by a prior open or create call. Once release has been called, no future requests for fh will be received (until the value is re-used in the return value of another open or create call).

This method may return an error by raising FUSEError, but the error will be discarded because there is no corresponding client request.

coroutine releasedir(fh)

Release open directory

This method will be called exactly once for each opendir call. After fh has been released, no further readdir requests will be received for it (until it is opened again with opendir).

coroutine removexattr(inode, name, ctx)

Remove extended attribute name of inode

ctx will be a RequestContext instance.

If the attribute does not exist, the method must raise FUSEError with an error code of ENOATTR. name will be of type bytes, but is guaranteed not to contain zero-bytes (\0).

coroutine rename(parent_inode_old, name_old, parent_inode_new, name_new, flags, ctx)

Rename a directory entry.

This method must rename name_old in the directory with inode parent_inode_old to name_new in the directory with inode parent_inode_new. If name_new already exists, it should be overwritten.

flags may be RENAME_EXCHANGE or RENAME_NOREPLACE. If RENAME_NOREPLACE is specified, the filesystem must not overwrite name_new if it exists and return an error instead. If RENAME_EXCHANGE is specified, the filesystem must atomically exchange the two files, i.e. both must exist and neither may be deleted.

ctx will be a RequestContext instance.

Let the inode associated with name_old in parent_inode_old be inode_moved, and the inode associated with name_new in parent_inode_new (if it exists) be called inode_deref.

If inode_deref exists and has a non-zero lookup count, or if there are other directory entries referring to inode_deref), the file system must update only the directory entry for name_new to point to inode_moved instead of inode_deref. (Potential) removal of inode_deref (containing the previous contents of name_new) must be deferred to the forget method to be carried out when the lookup count reaches zero (and of course only if at that point there are no more directory entries associated with inode_deref either).

coroutine rmdir(parent_inode, name, ctx)

Remove directory name

This method must remove the directory name from the direcory with inode parent_inode. ctx will be a RequestContext instance. If there are still entries in the directory, the method should raise FUSEError(errno.ENOTEMPTY).

If the inode associated with name (i.e., not the parent_inode) has a non-zero lookup count, the file system must remove only the directory entry (so that future calls to readdir for parent_inode will no longer include name, but e.g. calls to getattr for file’s inode still succeed). Removal of the associated inode holding the directory contents and metadata must be deferred to the forget method to be carried out when the lookup count reaches zero.

(Since hard links to directories are not allowed by POSIX, this method is not required to check if there are still other directory entries refering to the same inode. This conveniently avoids the ambigiouties associated with the . and .. entries).

coroutine setattr(inode, attr, fields, fh, ctx)

Change attributes of inode

fields will be an SetattrFields instance that specifies which attributes are to be updated. attr will be an EntryAttributes instance for inode that contains the new values for changed attributes, and undefined values for all other attributes.

Most file systems will additionally set the st_ctime_ns attribute to the current time (to indicate that the inode metadata was changed).

If the syscall that is being processed received a file descriptor argument (like e.g. ftruncate(2) or fchmod(2)), fh will be the file handle returned by the corresponding call to the open handler. If the syscall was path based (like e.g. truncate(2) or chmod(2)), fh will be None.

ctx will be a RequestContext instance.

The method should return an EntryAttributes instance (containing both the changed and unchanged values).

coroutine setxattr(inode, name, value, ctx)

Set extended attribute name of inode to value.

ctx will be a RequestContext instance.

The attribute may or may not exist already. Both name and value will be of type bytes. name is guaranteed not to contain zero-bytes (\0).

stacktrace()

Asynchronous debugging

This method will be called when the fuse_stacktrace extended attribute is set on the mountpoint. The default implementation logs the current stack trace of every running Python thread. This can be quite useful to debug file system deadlocks.

coroutine statfs(ctx)

Get file system statistics

ctx will be a RequestContext instance.

The method must return an appropriately filled StatvfsData instance.

Create a symbolic link

This method must create a symbolink link named name in the directory with inode parent_inode, pointing to target. ctx will be a RequestContext instance.

The method must return an EntryAttributes instance with the attributes of the newly created directory entry.

(Successful) execution of this handler increases the lookup count for the returned inode by one.

Remove a (possibly special) file

This method must remove the (special or regular) file name from the direcory with inode parent_inode. ctx will be a RequestContext instance.

If the inode associated with file (i.e., not the parent_inode) has a non-zero lookup count, or if there are still other directory entries referring to this inode (due to hardlinks), the file system must remove only the directory entry (so that future calls to readdir for parent_inode will no longer include name, but e.g. calls to getattr for file’s inode still succeed). (Potential) removal of the associated inode with the file contents and metadata must be deferred to the forget method to be carried out when the lookup count reaches zero (and of course only if at that point there are no more directory entries associated with the inode either).

coroutine write(fh, off, buf)

Write buf into fh at off

fh will by an integer filehandle returned by a prior open or create call.

This method must return the number of bytes written. However, unless the file system has been mounted with the direct_io option, the file system must always write all the provided data (i.e., return len(buf)).