2 * FUSE: Filesystem in Userspace
3 * Copyright (C) 2001-2007 Miklos Szeredi <miklos@szeredi.hu>
5 * This program can be distributed under the terms of the GNU LGPLv2.
6 * See the file COPYING.LIB.
11 #if !defined(FUSE_H_) && !defined(FUSE_LOWLEVEL_H_)
13 "Never include <fuse_common.h> directly; use <fuse.h> or <fuse_lowlevel.h> instead."
16 #ifndef FUSE_COMMON_H_
17 #define FUSE_COMMON_H_
22 /** Major version of FUSE library interface */
23 #define FUSE_MAJOR_VERSION 3
25 /** Minor version of FUSE library interface */
26 #define FUSE_MINOR_VERSION 2
28 #define FUSE_MAKE_VERSION(maj, min) ((maj) * 10 + (min))
29 #define FUSE_VERSION FUSE_MAKE_VERSION(FUSE_MAJOR_VERSION, FUSE_MINOR_VERSION)
32 * Information about an open file.
34 * File Handles are created by the open, opendir, and create methods and closed
35 * by the release and releasedir methods. Multiple file handles may be
36 * concurrently open for the same file. Generally, a client will create one
37 * file handle per file descriptor, though in some cases multiple file
38 * descriptors can share a single file handle.
40 struct fuse_file_info
{
41 /** Open flags. Available in open() and release() */
45 * In case of a write operation indicates if this was caused
46 * by a delayed write from the page cache. If so, then the
47 * context's pid, uid, and gid fields will not be valid, and
48 * the *fh* value may not match the *fh* value that would
49 * have been sent with the corresponding individual write
50 * requests if write caching had been disabled.
52 unsigned int writepage
:1;
54 /** Can be filled in by open, to use direct I/O on this file. */
55 unsigned int direct_io
:1;
58 * Can be filled in by open. It signals the kernel that any
59 * currently cached file data (ie., data that the filesystem
60 * provided the last time the file was open) need not be
61 * invalidated. Has no effect when set in other contexts (in
62 * particular it does nothing when set by opendir()).
64 unsigned int keep_cache
:1;
67 * Indicates a flush operation. Set in flush operation, also
68 * maybe set in highlevel lock operation and lowlevel release
74 * Can be filled in by open, to indicate that the file is not
77 unsigned int nonseekable
:1;
80 * Indicates that flock locks for this file should be
81 * released. If set, lock_owner shall contain a valid value.
82 * May only be set in ->release().
84 unsigned int flock_release
:1;
87 * Can be filled in by opendir. It signals the kernel to
88 * enable caching of entries returned by readdir(). Has no
89 * effect when set in other contexts (in particular it does
90 * nothing when set by open()).
92 unsigned int cache_readdir
:1;
94 /* Indicates that suid/sgid bits should be removed upon write */
95 unsigned int kill_priv
:1;
98 /** Padding. Reserved for future use*/
99 unsigned int padding
:24;
100 unsigned int padding2
:32;
103 * File handle id. May be filled in by filesystem in create,
104 * open, and opendir(). Available in most other file operations on the
109 /** Lock owner id. Available in locking operations and flush */
113 * Requested poll events. Available in ->poll. Only set on kernels
114 * which support it. If unsupported, this field is set to zero.
116 uint32_t poll_events
;
120 * Capability bits for 'fuse_conn_info.capable' and 'fuse_conn_info.want'
124 * Indicates that the filesystem supports asynchronous read requests.
126 * If this capability is not requested/available, the kernel will
127 * ensure that there is at most one pending read request per
128 * file-handle at any time, and will attempt to order read requests by
131 * This feature is enabled by default when supported by the kernel.
133 #define FUSE_CAP_ASYNC_READ (1 << 0)
136 * Indicates that the filesystem supports "remote" locking.
138 * This feature is enabled by default when supported by the kernel,
139 * and if getlk() and setlk() handlers are implemented.
141 #define FUSE_CAP_POSIX_LOCKS (1 << 1)
144 * Indicates that the filesystem supports the O_TRUNC open flag. If
145 * disabled, and an application specifies O_TRUNC, fuse first calls
146 * truncate() and then open() with O_TRUNC filtered out.
148 * This feature is enabled by default when supported by the kernel.
150 #define FUSE_CAP_ATOMIC_O_TRUNC (1 << 3)
153 * Indicates that the filesystem supports lookups of "." and "..".
155 * This feature is disabled by default.
157 #define FUSE_CAP_EXPORT_SUPPORT (1 << 4)
160 * Indicates that the kernel should not apply the umask to the
161 * file mode on create operations.
163 * This feature is disabled by default.
165 #define FUSE_CAP_DONT_MASK (1 << 6)
168 * Indicates that libfuse should try to use splice() when writing to
169 * the fuse device. This may improve performance.
171 * This feature is disabled by default.
173 #define FUSE_CAP_SPLICE_WRITE (1 << 7)
176 * Indicates that libfuse should try to move pages instead of copying when
177 * writing to / reading from the fuse device. This may improve performance.
179 * This feature is disabled by default.
181 #define FUSE_CAP_SPLICE_MOVE (1 << 8)
184 * Indicates that libfuse should try to use splice() when reading from
185 * the fuse device. This may improve performance.
187 * This feature is enabled by default when supported by the kernel and
188 * if the filesystem implements a write_buf() handler.
190 #define FUSE_CAP_SPLICE_READ (1 << 9)
193 * If set, the calls to flock(2) will be emulated using POSIX locks and must
194 * then be handled by the filesystem's setlock() handler.
196 * If not set, flock(2) calls will be handled by the FUSE kernel module
197 * internally (so any access that does not go through the kernel cannot be taken
200 * This feature is enabled by default when supported by the kernel and
201 * if the filesystem implements a flock() handler.
203 #define FUSE_CAP_FLOCK_LOCKS (1 << 10)
206 * Indicates that the filesystem supports ioctl's on directories.
208 * This feature is enabled by default when supported by the kernel.
210 #define FUSE_CAP_IOCTL_DIR (1 << 11)
213 * Traditionally, while a file is open the FUSE kernel module only
214 * asks the filesystem for an update of the file's attributes when a
215 * client attempts to read beyond EOF. This is unsuitable for
216 * e.g. network filesystems, where the file contents may change
217 * without the kernel knowing about it.
219 * If this flag is set, FUSE will check the validity of the attributes
220 * on every read. If the attributes are no longer valid (i.e., if the
221 * *attr_timeout* passed to fuse_reply_attr() or set in `struct
222 * fuse_entry_param` has passed), it will first issue a `getattr`
223 * request. If the new mtime differs from the previous value, any
224 * cached file *contents* will be invalidated as well.
226 * This flag should always be set when available. If all file changes
227 * go through the kernel, *attr_timeout* should be set to a very large
228 * number to avoid unnecessary getattr() calls.
230 * This feature is enabled by default when supported by the kernel.
232 #define FUSE_CAP_AUTO_INVAL_DATA (1 << 12)
235 * Indicates that the filesystem supports readdirplus.
237 * This feature is enabled by default when supported by the kernel and if the
238 * filesystem implements a readdirplus() handler.
240 #define FUSE_CAP_READDIRPLUS (1 << 13)
243 * Indicates that the filesystem supports adaptive readdirplus.
245 * If FUSE_CAP_READDIRPLUS is not set, this flag has no effect.
247 * If FUSE_CAP_READDIRPLUS is set and this flag is not set, the kernel
248 * will always issue readdirplus() requests to retrieve directory
251 * If FUSE_CAP_READDIRPLUS is set and this flag is set, the kernel
252 * will issue both readdir() and readdirplus() requests, depending on
253 * how much information is expected to be required.
255 * As of Linux 4.20, the algorithm is as follows: when userspace
256 * starts to read directory entries, issue a READDIRPLUS request to
257 * the filesystem. If any entry attributes have been looked up by the
258 * time userspace requests the next batch of entries continue with
259 * READDIRPLUS, otherwise switch to plain READDIR. This will reasult
260 * in eg plain "ls" triggering READDIRPLUS first then READDIR after
261 * that because it doesn't do lookups. "ls -l" should result in all
262 * READDIRPLUS, except if dentries are already cached.
264 * This feature is enabled by default when supported by the kernel and
265 * if the filesystem implements both a readdirplus() and a readdir()
268 #define FUSE_CAP_READDIRPLUS_AUTO (1 << 14)
271 * Indicates that the filesystem supports asynchronous direct I/O submission.
273 * If this capability is not requested/available, the kernel will ensure that
274 * there is at most one pending read and one pending write request per direct
275 * I/O file-handle at any time.
277 * This feature is enabled by default when supported by the kernel.
279 #define FUSE_CAP_ASYNC_DIO (1 << 15)
282 * Indicates that writeback caching should be enabled. This means that
283 * individual write request may be buffered and merged in the kernel
284 * before they are send to the filesystem.
286 * This feature is disabled by default.
288 #define FUSE_CAP_WRITEBACK_CACHE (1 << 16)
291 * Indicates support for zero-message opens. If this flag is set in
292 * the `capable` field of the `fuse_conn_info` structure, then the
293 * filesystem may return `ENOSYS` from the open() handler to indicate
294 * success. Further attempts to open files will be handled in the
295 * kernel. (If this flag is not set, returning ENOSYS will be treated
296 * as an error and signaled to the caller).
298 * Setting (or unsetting) this flag in the `want` field has *no
301 #define FUSE_CAP_NO_OPEN_SUPPORT (1 << 17)
304 * Indicates support for parallel directory operations. If this flag
305 * is unset, the FUSE kernel module will ensure that lookup() and
306 * readdir() requests are never issued concurrently for the same
309 * This feature is enabled by default when supported by the kernel.
311 #define FUSE_CAP_PARALLEL_DIROPS (1 << 18)
314 * Indicates support for POSIX ACLs.
316 * If this feature is enabled, the kernel will cache and have
317 * responsibility for enforcing ACLs. ACL will be stored as xattrs and
318 * passed to userspace, which is responsible for updating the ACLs in
319 * the filesystem, keeping the file mode in sync with the ACL, and
320 * ensuring inheritance of default ACLs when new filesystem nodes are
321 * created. Note that this requires that the file system is able to
322 * parse and interpret the xattr representation of ACLs.
324 * Enabling this feature implicitly turns on the
325 * ``default_permissions`` mount option (even if it was not passed to
328 * This feature is disabled by default.
330 #define FUSE_CAP_POSIX_ACL (1 << 19)
333 * Indicates that the filesystem is responsible for unsetting
334 * setuid and setgid bits when a file is written, truncated, or
335 * its owner is changed.
337 * This feature is enabled by default when supported by the kernel.
339 #define FUSE_CAP_HANDLE_KILLPRIV (1 << 20)
342 * Indicates support for zero-message opendirs. If this flag is set in
343 * the `capable` field of the `fuse_conn_info` structure, then the filesystem
344 * may return `ENOSYS` from the opendir() handler to indicate success. Further
345 * opendir and releasedir messages will be handled in the kernel. (If this
346 * flag is not set, returning ENOSYS will be treated as an error and signalled
349 * Setting (or unsetting) this flag in the `want` field has *no effect*.
351 #define FUSE_CAP_NO_OPENDIR_SUPPORT (1 << 24)
354 * Indicates that the kernel supports the FUSE_ATTR_SUBMOUNT flag.
356 * Setting (or unsetting) this flag in the `want` field has *no effect*.
358 #define FUSE_CAP_SUBMOUNTS (1 << 27)
361 * Indicates that the filesystem is responsible for clearing
362 * security.capability xattr and clearing setuid and setgid bits. Following
364 * - clear "security.capability" on write, truncate and chown unconditionally
365 * - clear suid/sgid if following is true. Note, sgid is cleared only if
366 * group executable bit is set.
367 * o setattr has FATTR_SIZE and FATTR_KILL_SUIDGID set.
368 * o setattr has FATTR_UID or FATTR_GID
369 * o open has O_TRUNC and FUSE_OPEN_KILL_SUIDGID
370 * o create has O_TRUNC and FUSE_OPEN_KILL_SUIDGID flag set.
371 * o write has FUSE_WRITE_KILL_SUIDGID
373 #define FUSE_CAP_HANDLE_KILLPRIV_V2 (1 << 28)
376 * Indicates that file server supports extended struct fuse_setxattr_in
378 #define FUSE_CAP_SETXATTR_EXT (1 << 29)
381 * Indicates that file server supports creating file security context
383 #define FUSE_CAP_SECURITY_CTX (1ULL << 32)
388 * FUSE_IOCTL_COMPAT: 32bit compat ioctl on 64bit machine
389 * FUSE_IOCTL_UNRESTRICTED: not restricted to well-formed ioctls, retry allowed
390 * FUSE_IOCTL_RETRY: retry with new iovecs
391 * FUSE_IOCTL_DIR: is a directory
393 * FUSE_IOCTL_MAX_IOV: maximum of in_iovecs + out_iovecs
395 #define FUSE_IOCTL_COMPAT (1 << 0)
396 #define FUSE_IOCTL_UNRESTRICTED (1 << 1)
397 #define FUSE_IOCTL_RETRY (1 << 2)
398 #define FUSE_IOCTL_DIR (1 << 4)
400 #define FUSE_IOCTL_MAX_IOV 256
403 * Connection information, passed to the ->init() method
405 * Some of the elements are read-write, these can be changed to
406 * indicate the value requested by the filesystem. The requested
407 * value must usually be smaller than the indicated value.
409 struct fuse_conn_info
{
411 * Major version of the protocol (read-only)
413 unsigned proto_major
;
416 * Minor version of the protocol (read-only)
418 unsigned proto_minor
;
421 * Maximum size of the write buffer
426 * Maximum size of read requests. A value of zero indicates no
427 * limit. However, even if the filesystem does not specify a
428 * limit, the maximum size of read requests will still be
429 * limited by the kernel.
431 * NOTE: For the time being, the maximum size of read requests
432 * must be set both here *and* passed to fuse_session_new()
433 * using the ``-o max_read=<n>`` mount option. At some point
434 * in the future, specifying the mount option will no longer
442 unsigned max_readahead
;
445 * Capability flags that the kernel supports (read-only)
450 * Capability flags that the filesystem wants to enable.
452 * libfuse attempts to initialize this field with
453 * reasonable default values before calling the init() handler.
458 * Maximum number of pending "background" requests. A
459 * background request is any type of request for which the
460 * total number is not limited by other means. As of kernel
461 * 4.8, only two types of requests fall into this category:
463 * 1. Read-ahead requests
464 * 2. Asynchronous direct I/O requests
466 * Read-ahead requests are generated (if max_readahead is
467 * non-zero) by the kernel to preemptively fill its caches
468 * when it anticipates that userspace will soon read more
471 * Asynchronous direct I/O requests are generated if
472 * FUSE_CAP_ASYNC_DIO is enabled and userspace submits a large
473 * direct I/O request. In this case the kernel will internally
474 * split it up into multiple smaller requests and submit them
475 * to the filesystem concurrently.
477 * Note that the following requests are *not* background
478 * requests: writeback requests (limited by the kernel's
479 * flusher algorithm), regular (i.e., synchronous and
480 * buffered) userspace read/write requests (limited to one per
481 * thread), asynchronous read requests (Linux's io_submit(2)
482 * call actually blocks, so these are also limited to one per
485 unsigned max_background
;
488 * Kernel congestion threshold parameter. If the number of pending
489 * background requests exceeds this number, the FUSE kernel module will
490 * mark the filesystem as "congested". This instructs the kernel to
491 * expect that queued requests will take some time to complete, and to
492 * adjust its algorithms accordingly (e.g. by putting a waiting thread
493 * to sleep instead of using a busy-loop).
495 unsigned congestion_threshold
;
498 * When FUSE_CAP_WRITEBACK_CACHE is enabled, the kernel is responsible
499 * for updating mtime and ctime when write requests are received. The
500 * updated values are passed to the filesystem with setattr() requests.
501 * However, if the filesystem does not support the full resolution of
502 * the kernel timestamps (nanoseconds), the mtime and ctime values used
503 * by kernel and filesystem will differ (and result in an apparent
504 * change of times after a cache flush).
506 * To prevent this problem, this variable can be used to inform the
507 * kernel about the timestamp granularity supported by the file-system.
508 * The value should be power of 10. The default is 1, i.e. full
509 * nano-second resolution. Filesystems supporting only second resolution
510 * should set this to 1000000000.
517 unsigned reserved
[22];
521 struct fuse_pollhandle
;
522 struct fuse_conn_info_opts
;
525 * This function parses several command-line options that can be used
526 * to override elements of struct fuse_conn_info. The pointer returned
527 * by this function should be passed to the
528 * fuse_apply_conn_info_opts() method by the file system's init()
531 * Before using this function, think twice if you really want these
532 * parameters to be adjustable from the command line. In most cases,
533 * they should be determined by the file system internally.
535 * The following options are recognized:
537 * -o max_write=N sets conn->max_write
538 * -o max_readahead=N sets conn->max_readahead
539 * -o max_background=N sets conn->max_background
540 * -o congestion_threshold=N sets conn->congestion_threshold
541 * -o async_read sets FUSE_CAP_ASYNC_READ in conn->want
542 * -o sync_read unsets FUSE_CAP_ASYNC_READ in conn->want
543 * -o atomic_o_trunc sets FUSE_CAP_ATOMIC_O_TRUNC in conn->want
544 * -o no_remote_lock Equivalent to -o
545 *no_remote_flock,no_remote_posix_lock -o no_remote_flock Unsets
546 *FUSE_CAP_FLOCK_LOCKS in conn->want -o no_remote_posix_lock Unsets
547 *FUSE_CAP_POSIX_LOCKS in conn->want -o [no_]splice_write (un-)sets
548 *FUSE_CAP_SPLICE_WRITE in conn->want -o [no_]splice_move (un-)sets
549 *FUSE_CAP_SPLICE_MOVE in conn->want -o [no_]splice_read (un-)sets
550 *FUSE_CAP_SPLICE_READ in conn->want -o [no_]auto_inval_data (un-)sets
551 *FUSE_CAP_AUTO_INVAL_DATA in conn->want -o readdirplus=no unsets
552 *FUSE_CAP_READDIRPLUS in conn->want -o readdirplus=yes sets
553 *FUSE_CAP_READDIRPLUS and unsets FUSE_CAP_READDIRPLUS_AUTO in conn->want -o
554 *readdirplus=auto sets FUSE_CAP_READDIRPLUS and FUSE_CAP_READDIRPLUS_AUTO
555 *in conn->want -o [no_]async_dio (un-)sets FUSE_CAP_ASYNC_DIO in
556 *conn->want -o [no_]writeback_cache (un-)sets FUSE_CAP_WRITEBACK_CACHE in
557 *conn->want -o time_gran=N sets conn->time_gran
559 * Known options will be removed from *args*, unknown options will be
560 * passed through unchanged.
562 * @param args argument vector (input+output)
563 * @return parsed options
565 struct fuse_conn_info_opts
*fuse_parse_conn_info_opts(struct fuse_args
*args
);
568 * This function applies the (parsed) parameters in *opts* to the
569 * *conn* pointer. It may modify the following fields: wants,
570 * max_write, max_readahead, congestion_threshold, max_background,
571 * time_gran. A field is only set (or unset) if the corresponding
572 * option has been explicitly set.
574 void fuse_apply_conn_info_opts(struct fuse_conn_info_opts
*opts
,
575 struct fuse_conn_info
*conn
);
578 * Go into the background
580 * @param foreground if true, stay in the foreground
581 * @return 0 on success, -1 on failure
583 int fuse_daemonize(int foreground
);
586 * Get the version of the library
588 * @return the version
590 int fuse_version(void);
593 * Get the full package version string of the library
595 * @return the package version
597 const char *fuse_pkgversion(void);
600 * Destroy poll handle
602 * @param ph the poll handle
604 void fuse_pollhandle_destroy(struct fuse_pollhandle
*ph
);
613 enum fuse_buf_flags
{
615 * Buffer contains a file descriptor
617 * If this flag is set, the .fd field is valid, otherwise the
618 * .mem fields is valid.
620 FUSE_BUF_IS_FD
= (1 << 1),
623 * Seek on the file descriptor
625 * If this flag is set then the .pos field is valid and is
626 * used to seek to the given offset before performing
627 * operation on file descriptor.
629 FUSE_BUF_FD_SEEK
= (1 << 2),
632 * Retry operation on file descriptor
634 * If this flag is set then retry operation on file descriptor
635 * until .size bytes have been copied or an error or EOF is
638 FUSE_BUF_FD_RETRY
= (1 << 3),
644 * Generic data buffer for I/O, extended attributes, etc... Data may
645 * be supplied as a memory pointer or as a file descriptor
649 * Size of data in bytes
656 enum fuse_buf_flags flags
;
661 * Used unless FUSE_BUF_IS_FD flag is set.
668 * Used if FUSE_BUF_IS_FD flag is set.
675 * Used if FUSE_BUF_FD_SEEK flag is set.
683 * An array of data buffers, each containing a memory pointer or a
686 * Allocate dynamically to add more than one buffer.
690 * Number of buffers in the array
695 * Index of current buffer within the array
700 * Current offset within the current buffer
707 struct fuse_buf buf
[1];
710 /* Initialize bufvec with a single buffer of given size */
711 #define FUSE_BUFVEC_INIT(size__) \
712 ((struct fuse_bufvec){ /* .count= */ 1, \
714 /* .off = */ 0, /* .buf = */ \
716 /* .size = */ (size__), \
717 /* .flags = */ (enum fuse_buf_flags)0, \
724 * Get total size of data in a fuse buffer vector
726 * @param bufv buffer vector
727 * @return size of data
729 size_t fuse_buf_size(const struct fuse_bufvec
*bufv
);
732 * Copy data from one buffer vector to another
734 * @param dst destination buffer vector
735 * @param src source buffer vector
736 * @return actual number of bytes copied or -errno on error
738 ssize_t
fuse_buf_copy(struct fuse_bufvec
*dst
, struct fuse_bufvec
*src
);
741 * Memory buffer iterator
744 struct fuse_mbuf_iter
{
751 * Total length, in bytes
756 * Offset from start of buffer
761 /* Initialize memory buffer iterator from a fuse_buf */
762 #define FUSE_MBUF_ITER_INIT(fbuf) \
763 ((struct fuse_mbuf_iter){ \
765 .size = fbuf->size, \
770 * Consume bytes from a memory buffer iterator
772 * @param iter memory buffer iterator
773 * @param len number of bytes to consume
774 * @return pointer to start of consumed bytes or
775 * NULL if advancing beyond end of buffer
777 void *fuse_mbuf_iter_advance(struct fuse_mbuf_iter
*iter
, size_t len
);
780 * Consume a NUL-terminated string from a memory buffer iterator
782 * @param iter memory buffer iterator
783 * @return pointer to the string or
784 * NULL if advancing beyond end of buffer or there is no NUL-terminator
786 const char *fuse_mbuf_iter_advance_str(struct fuse_mbuf_iter
*iter
);
792 * Exit session on HUP, TERM and INT signals and ignore PIPE signal
794 * Stores session in a global variable. May only be called once per
795 * process until fuse_remove_signal_handlers() is called.
797 * Once either of the POSIX signals arrives, the signal handler calls
798 * fuse_session_exit().
800 * @param se the session to exit
801 * @return 0 on success, -1 on failure
804 * fuse_remove_signal_handlers()
806 int fuse_set_signal_handlers(struct fuse_session
*se
);
809 * Restore default signal handlers
811 * Resets global session. After this fuse_set_signal_handlers() may
814 * @param se the same session as given in fuse_set_signal_handlers()
817 * fuse_set_signal_handlers()
819 void fuse_remove_signal_handlers(struct fuse_session
*se
);
822 * Compatibility stuff
825 #if !defined(FUSE_USE_VERSION) || FUSE_USE_VERSION < 30
826 #error only API version 30 or greater is supported
831 * This interface uses 64 bit off_t.
833 * On 32bit systems please add -D_FILE_OFFSET_BITS=64 to your compile flags!
835 QEMU_BUILD_BUG_ON(sizeof(off_t
) != 8);
837 #endif /* FUSE_COMMON_H_ */