1FCNTL(2) Linux Programmer's Manual FCNTL(2)
2
6 fcntl - manipulate file descriptor
7
9 #include <unistd.h>
10 #include <fcntl.h>
11 int fcntl(int fd, int cmd, ... /* arg */ );
13
15 fcntl() performs one of the operations described below on the open file
16 descriptor fd. The operation is determined by cmd.
17 fcntl() can take an optional third argument. Whether or not this argu‐
19 ment is required is determined by cmd. The required argument type is
20 indicated in parentheses after each cmd name (in most cases, the
21 required type is int, and we identify the argument using the name arg),
22 or void is specified if the argument is not required.
23 Certain of the operations below are supported only since a particular
25 Linux kernel version. The preferred method of checking whether the
26 host kernel supports a particular operation is to invoke fcntl() with
27 the desired cmd value and then test whether the call failed with EIN‐
28 VAL, indicating that the kernel does not recognize this value.
29 Duplicating a file descriptor
31 F_DUPFD (int)
32 Duplicate the file descriptor fd using the lowest-numbered
33 available file descriptor greater than or equal to arg. This is
34 different from dup2(2), which uses exactly the file descriptor
35 specified.
36 On success, the new file descriptor is returned.
38 See dup(2) for further details.
40 F_DUPFD_CLOEXEC (int; since Linux 2.6.24)
42 As for F_DUPFD, but additionally set the close-on-exec flag for
43 the duplicate file descriptor. Specifying this flag permits a
44 program to avoid an additional fcntl() F_SETFD operation to set
45 the FD_CLOEXEC flag. For an explanation of why this flag is
46 useful, see the description of O_CLOEXEC in open(2).
47 File descriptor flags
49 The following commands manipulate the flags associated with a file
50 descriptor. Currently, only one such flag is defined: FD_CLOEXEC, the
51 close-on-exec flag. If the FD_CLOEXEC bit is set, the file descriptor
52 will automatically be closed during a successful execve(2). (If the
53 execve(2) fails, the file descriptor is left open.) If the FD_CLOEXEC
54 bit is not set, the file descriptor will remain open across an
55 execve(2).
56 F_GETFD (void)
58 Return (as the function result) the file descriptor flags; arg
59 is ignored.
60 F_SETFD (int)
62 Set the file descriptor flags to the value specified by arg.
63 In multithreaded programs, using fcntl() F_SETFD to set the close-on-
65 exec flag at the same time as another thread performs a fork(2) plus
66 execve(2) is vulnerable to a race condition that may unintentionally
67 leak the file descriptor to the program executed in the child process.
68 See the discussion of the O_CLOEXEC flag in open(2) for details and a
69 remedy to the problem.
70 File status flags
72 Each open file description has certain associated status flags, ini‐
73 tialized by open(2) and possibly modified by fcntl(). Duplicated file
74 descriptors (made with dup(2), fcntl(F_DUPFD), fork(2), etc.) refer to
75 the same open file description, and thus share the same file status
76 flags.
77 The file status flags and their semantics are described in open(2).
79 F_GETFL (void)
81 Return (as the function result) the file access mode and the
82 file status flags; arg is ignored.
83 F_SETFL (int)
85 Set the file status flags to the value specified by arg. File
86 access mode (O_RDONLY, O_WRONLY, O_RDWR) and file creation flags
87 (i.e., O_CREAT, O_EXCL, O_NOCTTY, O_TRUNC) in arg are ignored.
88 On Linux, this command can change only the O_APPEND, O_ASYNC,
89 O_DIRECT, O_NOATIME, and O_NONBLOCK flags. It is not possible
90 to change the O_DSYNC and O_SYNC flags; see BUGS, below.
91 Advisory record locking
93 Linux implements traditional ("process-associated") UNIX record locks,
94 as standardized by POSIX. For a Linux-specific alternative with better
95 semantics, see the discussion of open file description locks below.
96 F_SETLK, F_SETLKW, and F_GETLK are used to acquire, release, and test
98 for the existence of record locks (also known as byte-range, file-seg‐
99 ment, or file-region locks). The third argument, lock, is a pointer to
100 a structure that has at least the following fields (in unspecified
101 order).
102 struct flock {
104 ...
105 short l_type; /* Type of lock: F_RDLCK,
106 F_WRLCK, F_UNLCK */
107 short l_whence; /* How to interpret l_start:
108 SEEK_SET, SEEK_CUR, SEEK_END */
109 off_t l_start; /* Starting offset for lock */
110 off_t l_len; /* Number of bytes to lock */
111 pid_t l_pid; /* PID of process blocking our lock
112 (set by F_GETLK and F_OFD_GETLK) */
113 ...
114 };
115 The l_whence, l_start, and l_len fields of this structure specify the
117 range of bytes we wish to lock. Bytes past the end of the file may be
118 locked, but not bytes before the start of the file.
119 l_start is the starting offset for the lock, and is interpreted rela‐
121 tive to either: the start of the file (if l_whence is SEEK_SET); the
122 current file offset (if l_whence is SEEK_CUR); or the end of the file
123 (if l_whence is SEEK_END). In the final two cases, l_start can be a
124 negative number provided the offset does not lie before the start of
125 the file.
126 l_len specifies the number of bytes to be locked. If l_len is posi‐
128 tive, then the range to be locked covers bytes l_start up to and
129 including l_start+l_len-1. Specifying 0 for l_len has the special
130 meaning: lock all bytes starting at the location specified by l_whence
131 and l_start through to the end of file, no matter how large the file
132 grows.
133 POSIX.1-2001 allows (but does not require) an implementation to support
135 a negative l_len value; if l_len is negative, the interval described by
136 lock covers bytes l_start+l_len up to and including l_start-1. This is
137 supported by Linux since kernel versions 2.4.21 and 2.5.49.
138 The l_type field can be used to place a read (F_RDLCK) or a write
140 (F_WRLCK) lock on a file. Any number of processes may hold a read lock
141 (shared lock) on a file region, but only one process may hold a write
142 lock (exclusive lock). An exclusive lock excludes all other locks,
143 both shared and exclusive. A single process can hold only one type of
144 lock on a file region; if a new lock is applied to an already-locked
145 region, then the existing lock is converted to the new lock type.
146 (Such conversions may involve splitting, shrinking, or coalescing with
147 an existing lock if the byte range specified by the new lock does not
148 precisely coincide with the range of the existing lock.)
149 F_SETLK (struct flock *)
151 Acquire a lock (when l_type is F_RDLCK or F_WRLCK) or release a
152 lock (when l_type is F_UNLCK) on the bytes specified by the
153 l_whence, l_start, and l_len fields of lock. If a conflicting
154 lock is held by another process, this call returns -1 and sets
155 errno to EACCES or EAGAIN. (The error returned in this case
156 differs across implementations, so POSIX requires a portable
157 application to check for both errors.)
158 F_SETLKW (struct flock *)
160 As for F_SETLK, but if a conflicting lock is held on the file,
161 then wait for that lock to be released. If a signal is caught
162 while waiting, then the call is interrupted and (after the sig‐
163 nal handler has returned) returns immediately (with return value
164 -1 and errno set to EINTR; see signal(7)).
165 F_GETLK (struct flock *)
167 On input to this call, lock describes a lock we would like to
168 place on the file. If the lock could be placed, fcntl() does
169 not actually place it, but returns F_UNLCK in the l_type field
170 of lock and leaves the other fields of the structure unchanged.
171 If one or more incompatible locks would prevent this lock being
173 placed, then fcntl() returns details about one of those locks in
174 the l_type, l_whence, l_start, and l_len fields of lock. If the
175 conflicting lock is a traditional (process-associated) record
176 lock, then the l_pid field is set to the PID of the process
177 holding that lock. If the conflicting lock is an open file
178 description lock, then l_pid is set to -1. Note that the
179 returned information may already be out of date by the time the
180 caller inspects it.
181 In order to place a read lock, fd must be open for reading. In order
183 to place a write lock, fd must be open for writing. To place both
184 types of lock, open a file read-write.
185 When placing locks with F_SETLKW, the kernel detects deadlocks, whereby
187 two or more processes have their lock requests mutually blocked by
188 locks held by the other processes. For example, suppose process A
189 holds a write lock on byte 100 of a file, and process B holds a write
190 lock on byte 200. If each process then attempts to lock the byte
191 already locked by the other process using F_SETLKW, then, without dead‐
192 lock detection, both processes would remain blocked indefinitely. When
193 the kernel detects such deadlocks, it causes one of the blocking lock
194 requests to immediately fail with the error EDEADLK; an application
195 that encounters such an error should release some of its locks to allow
196 other applications to proceed before attempting regain the locks that
197 it requires. Circular deadlocks involving more than two processes are
198 also detected. Note, however, that there are limitations to the ker‐
199 nel's deadlock-detection algorithm; see BUGS.
200 As well as being removed by an explicit F_UNLCK, record locks are auto‐
202 matically released when the process terminates.
203 Record locks are not inherited by a child created via fork(2), but are
205 preserved across an execve(2).
206 Because of the buffering performed by the stdio(3) library, the use of
208 record locking with routines in that package should be avoided; use
209 read(2) and write(2) instead.
210 The record locks described above are associated with the process
212 (unlike the open file description locks described below). This has
213 some unfortunate consequences:
214 * If a process closes any file descriptor referring to a file, then
216 all of the process's locks on that file are released, regardless of
217 the file descriptor(s) on which the locks were obtained. This is
218 bad: it means that a process can lose its locks on a file such as
219 /etc/passwd or /etc/mtab when for some reason a library function
220 decides to open, read, and close the same file.
221 * The threads in a process share locks. In other words, a multi‐
223 threaded program can't use record locking to ensure that threads
224 don't simultaneously access the same region of a file.
225 Open file description locks solve both of these problems.
227 Open file description locks (non-POSIX)
229 Open file description locks are advisory byte-range locks whose opera‐
230 tion is in most respects identical to the traditional record locks
231 described above. This lock type is Linux-specific, and available since
232 Linux 3.15. (There is a proposal with the Austin Group to include this
233 lock type in the next revision of POSIX.1.) For an explanation of open
234 file descriptions, see open(2).
235 The principal difference between the two lock types is that whereas
237 traditional record locks are associated with a process, open file
238 description locks are associated with the open file description on
239 which they are acquired, much like locks acquired with flock(2). Con‐
240 sequently (and unlike traditional advisory record locks), open file
241 description locks are inherited across fork(2) (and clone(2) with
242 CLONE_FILES), and are only automatically released on the last close of
243 the open file description, instead of being released on any close of
244 the file.
245 Conflicting lock combinations (i.e., a read lock and a write lock or
247 two write locks) where one lock is an open file description lock and
248 the other is a traditional record lock conflict even when they are
249 acquired by the same process on the same file descriptor.
250 Open file description locks placed via the same open file description
252 (i.e., via the same file descriptor, or via a duplicate of the file
253 descriptor created by fork(2), dup(2), fcntl() F_DUPFD, and so on) are
254 always compatible: if a new lock is placed on an already locked region,
255 then the existing lock is converted to the new lock type. (Such con‐
256 versions may result in splitting, shrinking, or coalescing with an
257 existing lock as discussed above.)
258 On the other hand, open file description locks may conflict with each
260 other when they are acquired via different open file descriptions.
261 Thus, the threads in a multithreaded program can use open file descrip‐
262 tion locks to synchronize access to a file region by having each thread
263 perform its own open(2) on the file and applying locks via the result‐
264 ing file descriptor.
265 As with traditional advisory locks, the third argument to fcntl(),
267 lock, is a pointer to an flock structure. By contrast with traditional
268 record locks, the l_pid field of that structure must be set to zero
269 when using the commands described below.
270 The commands for working with open file description locks are analogous
272 to those used with traditional locks:
273 F_OFD_SETLK (struct flock *)
275 Acquire an open file description lock (when l_type is F_RDLCK or
276 F_WRLCK) or release an open file description lock (when l_type
277 is F_UNLCK) on the bytes specified by the l_whence, l_start, and
278 l_len fields of lock. If a conflicting lock is held by another
279 process, this call returns -1 and sets errno to EAGAIN.
280 F_OFD_SETLKW (struct flock *)
282 As for F_OFD_SETLK, but if a conflicting lock is held on the
283 file, then wait for that lock to be released. If a signal is
284 caught while waiting, then the call is interrupted and (after
285 the signal handler has returned) returns immediately (with
286 return value -1 and errno set to EINTR; see signal(7)).
287 F_OFD_GETLK (struct flock *)
289 On input to this call, lock describes an open file description
290 lock we would like to place on the file. If the lock could be
291 placed, fcntl() does not actually place it, but returns F_UNLCK
292 in the l_type field of lock and leaves the other fields of the
293 structure unchanged. If one or more incompatible locks would
294 prevent this lock being placed, then details about one of these
295 locks are returned via lock, as described above for F_GETLK.
296 In the current implementation, no deadlock detection is performed for
298 open file description locks. (This contrasts with process-associated
299 record locks, for which the kernel does perform deadlock detection.)
300 Mandatory locking
302 Warning: the Linux implementation of mandatory locking is unreliable.
303 See BUGS below. Because of these bugs, and the fact that the feature
304 is believed to be little used, since Linux 4.5, mandatory locking has
305 been made an optional feature, governed by a configuration option (CON‐
306 FIG_MANDATORY_FILE_LOCKING). This is an initial step toward removing
307 this feature completely.
308 By default, both traditional (process-associated) and open file
310 description record locks are advisory. Advisory locks are not enforced
311 and are useful only between cooperating processes.
312 Both lock types can also be mandatory. Mandatory locks are enforced
314 for all processes. If a process tries to perform an incompatible
315 access (e.g., read(2) or write(2)) on a file region that has an incom‐
316 patible mandatory lock, then the result depends upon whether the O_NON‐
317 BLOCK flag is enabled for its open file description. If the O_NONBLOCK
318 flag is not enabled, then the system call is blocked until the lock is
319 removed or converted to a mode that is compatible with the access. If
320 the O_NONBLOCK flag is enabled, then the system call fails with the
321 error EAGAIN.
322 To make use of mandatory locks, mandatory locking must be enabled both
324 on the filesystem that contains the file to be locked, and on the file
325 itself. Mandatory locking is enabled on a filesystem using the "-o
326 mand" option to mount(8), or the MS_MANDLOCK flag for mount(2). Manda‐
327 tory locking is enabled on a file by disabling group execute permission
328 on the file and enabling the set-group-ID permission bit (see chmod(1)
329 and chmod(2)).
330 Mandatory locking is not specified by POSIX. Some other systems also
332 support mandatory locking, although the details of how to enable it
333 vary across systems.
334 Lost locks
336 When an advisory lock is obtained on a networked filesystem such as NFS
337 it is possible that the lock might get lost. This may happen due to
338 administrative action on the server, or due to a network partition
339 (i.e., loss of network connectivity with the server) which lasts long
340 enough for the server to assume that the client is no longer function‐
341 ing.
342 When the filesystem determines that a lock has been lost, future
344 read(2) or write(2) requests may fail with the error EIO. This error
345 will persist until the lock is removed or the file descriptor is
346 closed. Since Linux 3.12, this happens at least for NFSv4 (including
347 all minor versions).
348 Some versions of UNIX send a signal (SIGLOST) in this circumstance.
350 Linux does not define this signal, and does not provide any asynchro‐
351 nous notification of lost locks.
352 Managing signals
354 F_GETOWN, F_SETOWN, F_GETOWN_EX, F_SETOWN_EX, F_GETSIG and F_SETSIG are
355 used to manage I/O availability signals:
356 F_GETOWN (void)
358 Return (as the function result) the process ID or process group
359 currently receiving SIGIO and SIGURG signals for events on file
360 descriptor fd. Process IDs are returned as positive values;
361 process group IDs are returned as negative values (but see BUGS
362 below). arg is ignored.
363 F_SETOWN (int)
365 Set the process ID or process group ID that will receive SIGIO
366 and SIGURG signals for events on the file descriptor fd. The
367 target process or process group ID is specified in arg. A
368 process ID is specified as a positive value; a process group ID
369 is specified as a negative value. Most commonly, the calling
370 process specifies itself as the owner (that is, arg is specified
371 as getpid(2)).
372 As well as setting the file descriptor owner, one must also
374 enable generation of signals on the file descriptor. This is
375 done by using the fcntl() F_SETFL command to set the O_ASYNC
376 file status flag on the file descriptor. Subsequently, a SIGIO
377 signal is sent whenever input or output becomes possible on the
378 file descriptor. The fcntl() F_SETSIG command can be used to
379 obtain delivery of a signal other than SIGIO.
380 Sending a signal to the owner process (group) specified by
382 F_SETOWN is subject to the same permissions checks as are
383 described for kill(2), where the sending process is the one that
384 employs F_SETOWN (but see BUGS below). If this permission check
385 fails, then the signal is silently discarded. Note: The
386 F_SETOWN operation records the caller's credentials at the time
387 of the fcntl() call, and it is these saved credentials that are
388 used for the permission checks.
389 If the file descriptor fd refers to a socket, F_SETOWN also
391 selects the recipient of SIGURG signals that are delivered when
392 out-of-band data arrives on that socket. (SIGURG is sent in any
393 situation where select(2) would report the socket as having an
394 "exceptional condition".)
395 The following was true in 2.6.x kernels up to and including ker‐
397 nel 2.6.11:
398 If a nonzero value is given to F_SETSIG in a multi‐
400 threaded process running with a threading library that
401 supports thread groups (e.g., NPTL), then a positive
402 value given to F_SETOWN has a different meaning: instead
403 of being a process ID identifying a whole process, it is
404 a thread ID identifying a specific thread within a
405 process. Consequently, it may be necessary to pass
406 F_SETOWN the result of gettid(2) instead of getpid(2) to
407 get sensible results when F_SETSIG is used. (In current
408 Linux threading implementations, a main thread's thread
409 ID is the same as its process ID. This means that a sin‐
410 gle-threaded program can equally use gettid(2) or get‐
411 pid(2) in this scenario.) Note, however, that the state‐
412 ments in this paragraph do not apply to the SIGURG signal
413 generated for out-of-band data on a socket: this signal
414 is always sent to either a process or a process group,
415 depending on the value given to F_SETOWN.
416 The above behavior was accidentally dropped in Linux 2.6.12, and
418 won't be restored. From Linux 2.6.32 onward, use F_SETOWN_EX to
419 target SIGIO and SIGURG signals at a particular thread.
420 F_GETOWN_EX (struct f_owner_ex *) (since Linux 2.6.32)
422 Return the current file descriptor owner settings as defined by
423 a previous F_SETOWN_EX operation. The information is returned
424 in the structure pointed to by arg, which has the following
425 form:
426 struct f_owner_ex {
428 int type;
429 pid_t pid;
430 };
431 The type field will have one of the values F_OWNER_TID,
433 F_OWNER_PID, or F_OWNER_PGRP. The pid field is a positive inte‐
434 ger representing a thread ID, process ID, or process group ID.
435 See F_SETOWN_EX for more details.
436 F_SETOWN_EX (struct f_owner_ex *) (since Linux 2.6.32)
438 This operation performs a similar task to F_SETOWN. It allows
439 the caller to direct I/O availability signals to a specific
440 thread, process, or process group. The caller specifies the
441 target of signals via arg, which is a pointer to a f_owner_ex
442 structure. The type field has one of the following values,
443 which define how pid is interpreted:
444 F_OWNER_TID
446 Send the signal to the thread whose thread ID (the value
447 returned by a call to clone(2) or gettid(2)) is specified
448 in pid.
449 F_OWNER_PID
451 Send the signal to the process whose ID is specified in
452 pid.
453 F_OWNER_PGRP
455 Send the signal to the process group whose ID is speci‐
456 fied in pid. (Note that, unlike with F_SETOWN, a process
457 group ID is specified as a positive value here.)
458 F_GETSIG (void)
460 Return (as the function result) the signal sent when input or
461 output becomes possible. A value of zero means SIGIO is sent.
462 Any other value (including SIGIO) is the signal sent instead,
463 and in this case additional info is available to the signal han‐
464 dler if installed with SA_SIGINFO. arg is ignored.
465 F_SETSIG (int)
467 Set the signal sent when input or output becomes possible to the
468 value given in arg. A value of zero means to send the default
469 SIGIO signal. Any other value (including SIGIO) is the signal
470 to send instead, and in this case additional info is available
471 to the signal handler if installed with SA_SIGINFO.
472 By using F_SETSIG with a nonzero value, and setting SA_SIGINFO
474 for the signal handler (see sigaction(2)), extra information
475 about I/O events is passed to the handler in a siginfo_t struc‐
476 ture. If the si_code field indicates the source is SI_SIGIO,
477 the si_fd field gives the file descriptor associated with the
478 event. Otherwise, there is no indication which file descriptors
479 are pending, and you should use the usual mechanisms (select(2),
480 poll(2), read(2) with O_NONBLOCK set etc.) to determine which
481 file descriptors are available for I/O.
482 Note that the file descriptor provided in si_fd is the one that
484 was specified during the F_SETSIG operation. This can lead to
485 an unusual corner case. If the file descriptor is duplicated
486 (dup(2) or similar), and the original file descriptor is closed,
487 then I/O events will continue to be generated, but the si_fd
488 field will contain the number of the now closed file descriptor.
489 By selecting a real time signal (value >= SIGRTMIN), multiple
491 I/O events may be queued using the same signal numbers. (Queu‐
492 ing is dependent on available memory.) Extra information is
493 available if SA_SIGINFO is set for the signal handler, as above.
494 Note that Linux imposes a limit on the number of real-time sig‐
496 nals that may be queued to a process (see getrlimit(2) and sig‐
497 nal(7)) and if this limit is reached, then the kernel reverts to
498 delivering SIGIO, and this signal is delivered to the entire
499 process rather than to a specific thread.
500 Using these mechanisms, a program can implement fully asynchronous I/O
502 without using select(2) or poll(2) most of the time.
503 The use of O_ASYNC is specific to BSD and Linux. The only use of
505 F_GETOWN and F_SETOWN specified in POSIX.1 is in conjunction with the
506 use of the SIGURG signal on sockets. (POSIX does not specify the SIGIO
507 signal.) F_GETOWN_EX, F_SETOWN_EX, F_GETSIG, and F_SETSIG are Linux-
508 specific. POSIX has asynchronous I/O and the aio_sigevent structure to
509 achieve similar things; these are also available in Linux as part of
510 the GNU C Library (Glibc).
511 Leases
513 F_SETLEASE and F_GETLEASE (Linux 2.4 onward) are used to establish a
514 new lease, and retrieve the current lease, on the open file description
515 referred to by the file descriptor fd. A file lease provides a mecha‐
516 nism whereby the process holding the lease (the "lease holder") is
517 notified (via delivery of a signal) when a process (the "lease
518 breaker") tries to open(2) or truncate(2) the file referred to by that
519 file descriptor.
520 F_SETLEASE (int)
522 Set or remove a file lease according to which of the following
523 values is specified in the integer arg:
524 F_RDLCK
526 Take out a read lease. This will cause the calling
527 process to be notified when the file is opened for writ‐
528 ing or is truncated. A read lease can be placed only on
529 a file descriptor that is opened read-only.
530 F_WRLCK
532 Take out a write lease. This will cause the caller to be
533 notified when the file is opened for reading or writing
534 or is truncated. A write lease may be placed on a file
535 only if there are no other open file descriptors for the
536 file.
537 F_UNLCK
539 Remove our lease from the file.
540 Leases are associated with an open file description (see open(2)).
542 This means that duplicate file descriptors (created by, for example,
543 fork(2) or dup(2)) refer to the same lease, and this lease may be modi‐
544 fied or released using any of these descriptors. Furthermore, the
545 lease is released by either an explicit F_UNLCK operation on any of
546 these duplicate file descriptors, or when all such file descriptors
547 have been closed.
548 Leases may be taken out only on regular files. An unprivileged process
550 may take out a lease only on a file whose UID (owner) matches the
551 filesystem UID of the process. A process with the CAP_LEASE capability
552 may take out leases on arbitrary files.
553 F_GETLEASE (void)
555 Indicates what type of lease is associated with the file
556 descriptor fd by returning either F_RDLCK, F_WRLCK, or F_UNLCK,
557 indicating, respectively, a read lease , a write lease, or no
558 lease. arg is ignored.
559 When a process (the "lease breaker") performs an open(2) or truncate(2)
561 that conflicts with a lease established via F_SETLEASE, the system call
562 is blocked by the kernel and the kernel notifies the lease holder by
563 sending it a signal (SIGIO by default). The lease holder should
564 respond to receipt of this signal by doing whatever cleanup is required
565 in preparation for the file to be accessed by another process (e.g.,
566 flushing cached buffers) and then either remove or downgrade its lease.
567 A lease is removed by performing an F_SETLEASE command specifying arg
568 as F_UNLCK. If the lease holder currently holds a write lease on the
569 file, and the lease breaker is opening the file for reading, then it is
570 sufficient for the lease holder to downgrade the lease to a read lease.
571 This is done by performing an F_SETLEASE command specifying arg as
572 F_RDLCK.
573 If the lease holder fails to downgrade or remove the lease within the
575 number of seconds specified in /proc/sys/fs/lease-break-time, then the
576 kernel forcibly removes or downgrades the lease holder's lease.
577 Once a lease break has been initiated, F_GETLEASE returns the target
579 lease type (either F_RDLCK or F_UNLCK, depending on what would be com‐
580 patible with the lease breaker) until the lease holder voluntarily
581 downgrades or removes the lease or the kernel forcibly does so after
582 the lease break timer expires.
583 Once the lease has been voluntarily or forcibly removed or downgraded,
585 and assuming the lease breaker has not unblocked its system call, the
586 kernel permits the lease breaker's system call to proceed.
587 If the lease breaker's blocked open(2) or truncate(2) is interrupted by
589 a signal handler, then the system call fails with the error EINTR, but
590 the other steps still occur as described above. If the lease breaker
591 is killed by a signal while blocked in open(2) or truncate(2), then the
592 other steps still occur as described above. If the lease breaker spec‐
593 ifies the O_NONBLOCK flag when calling open(2), then the call immedi‐
594 ately fails with the error EWOULDBLOCK, but the other steps still occur
595 as described above.
596 The default signal used to notify the lease holder is SIGIO, but this
598 can be changed using the F_SETSIG command to fcntl(). If a F_SETSIG
599 command is performed (even one specifying SIGIO), and the signal han‐
600 dler is established using SA_SIGINFO, then the handler will receive a
601 siginfo_t structure as its second argument, and the si_fd field of this
602 argument will hold the file descriptor of the leased file that has been
603 accessed by another process. (This is useful if the caller holds
604 leases against multiple files.)
605 File and directory change notification (dnotify)
607 F_NOTIFY (int)
608 (Linux 2.4 onward) Provide notification when the directory
609 referred to by fd or any of the files that it contains is
610 changed. The events to be notified are specified in arg, which
611 is a bit mask specified by ORing together zero or more of the
612 following bits:
613 DN_ACCESS A file was accessed (read(2), pread(2), readv(2),
615 and similar)
616 DN_MODIFY A file was modified (write(2), pwrite(2), writev(2),
617 truncate(2), ftruncate(2), and similar).
618 DN_CREATE A file was created (open(2), creat(2), mknod(2),
619 mkdir(2), link(2), symlink(2), rename(2) into this
620 directory).
621 DN_DELETE A file was unlinked (unlink(2), rename(2) to another
622 directory, rmdir(2)).
623 DN_RENAME A file was renamed within this directory
624 (rename(2)).
625 DN_ATTRIB The attributes of a file were changed (chown(2),
626 chmod(2), utime(2), utimensat(2), and similar).
627 (In order to obtain these definitions, the _GNU_SOURCE feature
629 test macro must be defined before including any header files.)
630 Directory notifications are normally "one-shot", and the appli‐
632 cation must reregister to receive further notifications. Alter‐
633 natively, if DN_MULTISHOT is included in arg, then notification
634 will remain in effect until explicitly removed.
635 A series of F_NOTIFY requests is cumulative, with the events in
637 arg being added to the set already monitored. To disable noti‐
638 fication of all events, make an F_NOTIFY call specifying arg as
639 0.
640 Notification occurs via delivery of a signal. The default sig‐
642 nal is SIGIO, but this can be changed using the F_SETSIG command
643 to fcntl(). (Note that SIGIO is one of the nonqueuing standard
644 signals; switching to the use of a real-time signal means that
645 multiple notifications can be queued to the process.) In the
646 latter case, the signal handler receives a siginfo_t structure
647 as its second argument (if the handler was established using
648 SA_SIGINFO) and the si_fd field of this structure contains the
649 file descriptor which generated the notification (useful when
650 establishing notification on multiple directories).
651 Especially when using DN_MULTISHOT, a real time signal should be
653 used for notification, so that multiple notifications can be
654 queued.
655 NOTE: New applications should use the inotify interface (avail‐
657 able since kernel 2.6.13), which provides a much superior inter‐
658 face for obtaining notifications of filesystem events. See ino‐
659 tify(7).
660 Changing the capacity of a pipe
662 F_SETPIPE_SZ (int; since Linux 2.6.35)
663 Change the capacity of the pipe referred to by fd to be at least
664 arg bytes. An unprivileged process can adjust the pipe capacity
665 to any value between the system page size and the limit defined
666 in /proc/sys/fs/pipe-max-size (see proc(5)). Attempts to set
667 the pipe capacity below the page size are silently rounded up to
668 the page size. Attempts by an unprivileged process to set the
669 pipe capacity above the limit in /proc/sys/fs/pipe-max-size
670 yield the error EPERM; a privileged process (CAP_SYS_RESOURCE)
671 can override the limit.
672 When allocating the buffer for the pipe, the kernel may use a
674 capacity larger than arg, if that is convenient for the imple‐
675 mentation. (In the current implementation, the allocation is
676 the next higher power-of-two page-size multiple of the requested
677 size.) The actual capacity (in bytes) that is set is returned
678 as the function result.
679 Attempting to set the pipe capacity smaller than the amount of
681 buffer space currently used to store data produces the error
682 EBUSY.
683 Note that because of the way the pages of the pipe buffer are
685 employed when data is written to the pipe, the number of bytes
686 that can be written may be less than the nominal size, depending
687 on the size of the writes.
688 F_GETPIPE_SZ (void; since Linux 2.6.35)
690 Return (as the function result) the capacity of the pipe
691 referred to by fd.
692 File Sealing
694 File seals limit the set of allowed operations on a given file. For
695 each seal that is set on a file, a specific set of operations will fail
696 with EPERM on this file from now on. The file is said to be sealed.
697 The default set of seals depends on the type of the underlying file and
698 filesystem. For an overview of file sealing, a discussion of its pur‐
699 pose, and some code examples, see memfd_create(2).
700 Currently, file seals can be applied only to a file descriptor returned
702 by memfd_create(2) (if the MFD_ALLOW_SEALING was employed). On other
703 filesystems, all fcntl() operations that operate on seals will return
704 EINVAL.
705 Seals are a property of an inode. Thus, all open file descriptors
707 referring to the same inode share the same set of seals. Furthermore,
708 seals can never be removed, only added.
709 F_ADD_SEALS (int; since Linux 3.17)
711 Add the seals given in the bit-mask argument arg to the set of
712 seals of the inode referred to by the file descriptor fd. Seals
713 cannot be removed again. Once this call succeeds, the seals are
714 enforced by the kernel immediately. If the current set of seals
715 includes F_SEAL_SEAL (see below), then this call will be
716 rejected with EPERM. Adding a seal that is already set is a no-
717 op, in case F_SEAL_SEAL is not set already. In order to place a
718 seal, the file descriptor fd must be writable.
719 F_GET_SEALS (void; since Linux 3.17)
721 Return (as the function result) the current set of seals of the
722 inode referred to by fd. If no seals are set, 0 is returned.
723 If the file does not support sealing, -1 is returned and errno
724 is set to EINVAL.
725 The following seals are available:
727 F_SEAL_SEAL
729 If this seal is set, any further call to fcntl() with
730 F_ADD_SEALS fails with the error EPERM. Therefore, this seal
731 prevents any modifications to the set of seals itself. If the
732 initial set of seals of a file includes F_SEAL_SEAL, then this
733 effectively causes the set of seals to be constant and locked.
734 F_SEAL_SHRINK
736 If this seal is set, the file in question cannot be reduced in
737 size. This affects open(2) with the O_TRUNC flag as well as
738 truncate(2) and ftruncate(2). Those calls fail with EPERM if
739 you try to shrink the file in question. Increasing the file
740 size is still possible.
741 F_SEAL_GROW
743 If this seal is set, the size of the file in question cannot be
744 increased. This affects write(2) beyond the end of the file,
745 truncate(2), ftruncate(2), and fallocate(2). These calls fail
746 with EPERM if you use them to increase the file size. If you
747 keep the size or shrink it, those calls still work as expected.
748 F_SEAL_WRITE
750 If this seal is set, you cannot modify the contents of the file.
751 Note that shrinking or growing the size of the file is still
752 possible and allowed. Thus, this seal is normally used in com‐
753 bination with one of the other seals. This seal affects
754 write(2) and fallocate(2) (only in combination with the FAL‐
755 LOC_FL_PUNCH_HOLE flag). Those calls fail with EPERM if this
756 seal is set. Furthermore, trying to create new shared, writable
757 memory-mappings via mmap(2) will also fail with EPERM.
758 Using the F_ADD_SEALS operation to set the F_SEAL_WRITE seal
760 fails with EBUSY if any writable, shared mapping exists. Such
761 mappings must be unmapped before you can add this seal. Fur‐
762 thermore, if there are any asynchronous I/O operations (io_sub‐
763 mit(2)) pending on the file, all outstanding writes will be dis‐
764 carded.
765 File read/write hints
767 Write lifetime hints can be used to inform the kernel about the rela‐
768 tive expected lifetime of writes on a given inode or via a particular
769 open file description. (See open(2) for an explanation of open file
770 descriptions.) In this context, the term "write lifetime" means the
771 expected time the data will live on media, before being overwritten or
772 erased.
773 An application may use the different hint values specified below to
775 separate writes into different write classes, so that multiple users or
776 applications running on a single storage back-end can aggregate their
777 I/O patterns in a consistent manner. However, there are no functional
778 semantics implied by these flags, and different I/O classes can use the
779 write lifetime hints in arbitrary ways, so long as the hints are used
780 consistently.
781 The following operations can be applied to the file descriptor, fd:
783 F_GET_RW_HINT (uint64_t *; since Linux 4.13)
785 Returns the value of the read/write hint associated with the
786 underlying inode referred to by fd.
787 F_SET_RW_HINT (uint64_t *; since Linux 4.13)
789 Sets the read/write hint value associated with the underlying
790 inode referred to by fd. This hint persists until either it is
791 explicitly modified or the underlying filesystem is unmounted.
792 F_GET_FILE_RW_HINT (uint64_t *; since Linux 4.13)
794 Returns the value of the read/write hint associated with the
795 open file description referred to by fd.
796 F_SET_FILE_RW_HINT (uint64_t *; since Linux 4.13)
798 Sets the read/write hint value associated with the open file
799 description referred to by fd.
800 If an open file description has not been assigned a read/write hint,
802 then it shall use the value assigned to the inode, if any.
803 The following read/write hints are valid since Linux 4.13:
805 RWH_WRITE_LIFE_NOT_SET
807 No specific hint has been set. This is the default value.
808 RWH_WRITE_LIFE_NONE
810 No specific write lifetime is associated with this file or
811 inode.
812 RWH_WRITE_LIFE_SHORT
814 Data written to this inode or via this open file description is
815 expected to have a short lifetime.
816 RWH_WRITE_LIFE_MEDIUM
818 Data written to this inode or via this open file description is
819 expected to have a lifetime longer than data written with
820 RWH_WRITE_LIFE_SHORT.
821 RWH_WRITE_LIFE_LONG
823 Data written to this inode or via this open file description is
824 expected to have a lifetime longer than data written with
825 RWH_WRITE_LIFE_MEDIUM.
826 RWH_WRITE_LIFE_EXTREME
828 Data written to this inode or via this open file description is
829 expected to have a lifetime longer than data written with
830 RWH_WRITE_LIFE_LONG.
831 All the write-specific hints are relative to each other, and no indi‐
833 vidual absolute meaning should be attributed to them.
834
836 For a successful call, the return value depends on the operation:
837 F_DUPFD The new file descriptor.
839 F_GETFD Value of file descriptor flags.
841 F_GETFL Value of file status flags.
843 F_GETLEASE
845 Type of lease held on file descriptor.
846 F_GETOWN Value of file descriptor owner.
848 F_GETSIG Value of signal sent when read or write becomes possible, or
850 zero for traditional SIGIO behavior.
851 F_GETPIPE_SZ, F_SETPIPE_SZ
853 The pipe capacity.
854 F_GET_SEALS
856 A bit mask identifying the seals that have been set for the
857 inode referred to by fd.
858 All other commands
860 Zero.
861 On error, -1 is returned, and errno is set appropriately.
863
865 EACCES or EAGAIN
866 Operation is prohibited by locks held by other processes.
867 EAGAIN The operation is prohibited because the file has been memory-
869 mapped by another process.
870 EBADF fd is not an open file descriptor
872 EBADF cmd is F_SETLK or F_SETLKW and the file descriptor open mode
874 doesn't match with the type of lock requested.
875 EBUSY cmd is F_SETPIPE_SZ and the new pipe capacity specified in arg
877 is smaller than the amount of buffer space currently used to
878 store data in the pipe.
879 EBUSY cmd is F_ADD_SEALS, arg includes F_SEAL_WRITE, and there exists
881 a writable, shared mapping on the file referred to by fd.
882 EDEADLK
884 It was detected that the specified F_SETLKW command would cause
885 a deadlock.
886 EFAULT lock is outside your accessible address space.
888 EINTR cmd is F_SETLKW or F_OFD_SETLKW and the operation was inter‐
890 rupted by a signal; see signal(7).
891 EINTR cmd is F_GETLK, F_SETLK, F_OFD_GETLK, or F_OFD_SETLK, and the
893 operation was interrupted by a signal before the lock was
894 checked or acquired. Most likely when locking a remote file
895 (e.g., locking over NFS), but can sometimes happen locally.
896 EINVAL The value specified in cmd is not recognized by this kernel.
898 EINVAL cmd is F_ADD_SEALS and arg includes an unrecognized sealing bit.
900 EINVAL cmd is F_ADD_SEALS or F_GET_SEALS and the filesystem containing
902 the inode referred to by fd does not support sealing.
903 EINVAL cmd is F_DUPFD and arg is negative or is greater than the maxi‐
905 mum allowable value (see the discussion of RLIMIT_NOFILE in
906 getrlimit(2)).
907 EINVAL cmd is F_SETSIG and arg is not an allowable signal number.
909 EINVAL cmd is F_OFD_SETLK, F_OFD_SETLKW, or F_OFD_GETLK, and l_pid was
911 not specified as zero.
912 EMFILE cmd is F_DUPFD and the per-process limit on the number of open
914 file descriptors has been reached.
915 ENOLCK Too many segment locks open, lock table is full, or a remote
917 locking protocol failed (e.g., locking over NFS).
918 ENOTDIR
920 F_NOTIFY was specified in cmd, but fd does not refer to a direc‐
921 tory.
922 EPERM cmd is F_SETPIPE_SZ and the soft or hard user pipe limit has
924 been reached; see pipe(7).
925 EPERM Attempted to clear the O_APPEND flag on a file that has the
927 append-only attribute set.
928 EPERM cmd was F_ADD_SEALS, but fd was not open for writing or the cur‐
930 rent set of seals on the file already includes F_SEAL_SEAL.
931
933 SVr4, 4.3BSD, POSIX.1-2001. Only the operations F_DUPFD, F_GETFD,
934 F_SETFD, F_GETFL, F_SETFL, F_GETLK, F_SETLK, and F_SETLKW are specified
935 in POSIX.1-2001.
936 F_GETOWN and F_SETOWN are specified in POSIX.1-2001. (To get their
938 definitions, define either _XOPEN_SOURCE with the value 500 or greater,
939 or _POSIX_C_SOURCE with the value 200809L or greater.)
940 F_DUPFD_CLOEXEC is specified in POSIX.1-2008. (To get this definition,
942 define _POSIX_C_SOURCE with the value 200809L or greater, or
943 _XOPEN_SOURCE with the value 700 or greater.)
944 F_GETOWN_EX, F_SETOWN_EX, F_SETPIPE_SZ, F_GETPIPE_SZ, F_GETSIG, F_SET‐
946 SIG, F_NOTIFY, F_GETLEASE, and F_SETLEASE are Linux-specific. (Define
947 the _GNU_SOURCE macro to obtain these definitions.)
948 F_OFD_SETLK, F_OFD_SETLKW, and F_OFD_GETLK are Linux-specific (and one
950 must define _GNU_SOURCE to obtain their definitions), but work is being
951 done to have them included in the next version of POSIX.1.
952 F_ADD_SEALS and F_GET_SEALS are Linux-specific.
954
956 The errors returned by dup2(2) are different from those returned by
957 F_DUPFD.
958 File locking
960 The original Linux fcntl() system call was not designed to handle large
961 file offsets (in the flock structure). Consequently, an fcntl64() sys‐
962 tem call was added in Linux 2.4. The newer system call employs a dif‐
963 ferent structure for file locking, flock64, and corresponding commands,
964 F_GETLK64, F_SETLK64, and F_SETLKW64. However, these details can be
965 ignored by applications using glibc, whose fcntl() wrapper function
966 transparently employs the more recent system call where it is avail‐
967 able.
968 Record locks
970 Since kernel 2.0, there is no interaction between the types of lock
971 placed by flock(2) and fcntl().
972 Several systems have more fields in struct flock such as, for example,
974 l_sysid (to identify the machine where the lock is held). Clearly,
975 l_pid alone is not going to be very useful if the process holding the
976 lock may live on a different machine; on Linux, while present on some
977 architectures (such as MIPS32), this field is not used.
978 The original Linux fcntl() system call was not designed to handle large
980 file offsets (in the flock structure). Consequently, an fcntl64() sys‐
981 tem call was added in Linux 2.4. The newer system call employs a dif‐
982 ferent structure for file locking, flock64, and corresponding commands,
983 F_GETLK64, F_SETLK64, and F_SETLKW64. However, these details can be
984 ignored by applications using glibc, whose fcntl() wrapper function
985 transparently employs the more recent system call where it is avail‐
986 able.
987 Record locking and NFS
989 Before Linux 3.12, if an NFSv4 client loses contact with the server for
990 a period of time (defined as more than 90 seconds with no communica‐
991 tion), it might lose and regain a lock without ever being aware of the
992 fact. (The period of time after which contact is assumed lost is known
993 as the NFSv4 leasetime. On a Linux NFS server, this can be determined
994 by looking at /proc/fs/nfsd/nfsv4leasetime, which expresses the period
995 in seconds. The default value for this file is 90.) This scenario
996 potentially risks data corruption, since another process might acquire
997 a lock in the intervening period and perform file I/O.
998 Since Linux 3.12, if an NFSv4 client loses contact with the server, any
1000 I/O to the file by a process which "thinks" it holds a lock will fail
1001 until that process closes and reopens the file. A kernel parameter,
1002 nfs.recover_lost_locks, can be set to 1 to obtain the pre-3.12 behav‐
1003 ior, whereby the client will attempt to recover lost locks when contact
1004 is reestablished with the server. Because of the attendant risk of
1005 data corruption, this parameter defaults to 0 (disabled).
1006
1008 F_SETFL
1009 It is not possible to use F_SETFL to change the state of the O_DSYNC
1010 and O_SYNC flags. Attempts to change the state of these flags are
1011 silently ignored.
1012 F_GETOWN
1014 A limitation of the Linux system call conventions on some architectures
1015 (notably i386) means that if a (negative) process group ID to be
1016 returned by F_GETOWN falls in the range -1 to -4095, then the return
1017 value is wrongly interpreted by glibc as an error in the system call;
1018 that is, the return value of fcntl() will be -1, and errno will contain
1019 the (positive) process group ID. The Linux-specific F_GETOWN_EX opera‐
1020 tion avoids this problem. Since glibc version 2.11, glibc makes the
1021 kernel F_GETOWN problem invisible by implementing F_GETOWN using
1022 F_GETOWN_EX.
1023 F_SETOWN
1025 In Linux 2.4 and earlier, there is bug that can occur when an unprivi‐
1026 leged process uses F_SETOWN to specify the owner of a socket file
1027 descriptor as a process (group) other than the caller. In this case,
1028 fcntl() can return -1 with errno set to EPERM, even when the owner
1029 process (group) is one that the caller has permission to send signals
1030 to. Despite this error return, the file descriptor owner is set, and
1031 signals will be sent to the owner.
1032 Deadlock detection
1034 The deadlock-detection algorithm employed by the kernel when dealing
1035 with F_SETLKW requests can yield both false negatives (failures to
1036 detect deadlocks, leaving a set of deadlocked processes blocked indefi‐
1037 nitely) and false positives (EDEADLK errors when there is no deadlock).
1038 For example, the kernel limits the lock depth of its dependency search
1039 to 10 steps, meaning that circular deadlock chains that exceed that
1040 size will not be detected. In addition, the kernel may falsely indi‐
1041 cate a deadlock when two or more processes created using the clone(2)
1042 CLONE_FILES flag place locks that appear (to the kernel) to conflict.
1043 Mandatory locking
1045 The Linux implementation of mandatory locking is subject to race condi‐
1046 tions which render it unreliable: a write(2) call that overlaps with a
1047 lock may modify data after the mandatory lock is acquired; a read(2)
1048 call that overlaps with a lock may detect changes to data that were
1049 made only after a write lock was acquired. Similar races exist between
1050 mandatory locks and mmap(2). It is therefore inadvisable to rely on
1051 mandatory locking.
1052
1054 dup2(2), flock(2), open(2), socket(2), lockf(3), capabilities(7), fea‐
1055 ture_test_macros(7), lslocks(8)
1056 locks.txt, mandatory-locking.txt, and dnotify.txt in the Linux kernel
1058 source directory Documentation/filesystems/ (on older kernels, these
1059 files are directly under the Documentation/ directory, and mandatory-
1060 locking.txt is called mandatory.txt)
1061
1063 This page is part of release 5.04 of the Linux man-pages project. A
1064 description of the project, information about reporting bugs, and the
1065 latest version of this page, can be found at
1066 https://www.kernel.org/doc/man-pages/.
1067Linux 2019-03-06 FCNTL(2)