Provided by: manpages-dev_6.03-1_all bug

NAME

       fcntl - manipulate file descriptor

LIBRARY

       Standard C library (libc, -lc)

SYNOPSIS

       #include <fcntl.h>

       int fcntl(int fd, int cmd, ... /* arg */ );

DESCRIPTION

       fcntl()  performs  one  of  the operations described below on the open file descriptor fd.
       The operation is determined by cmd.

       fcntl() can take an optional third argument.  Whether or not this argument is required  is
       determined  by cmd.  The required argument type is indicated in parentheses after each cmd
       name (in most cases, the required type is int, and we identify the argument using the name
       arg), or void is specified if the argument is not required.

       Certain  of  the  operations  below  are  supported  only  since a particular Linux kernel
       version.  The preferred method of checking whether the host kernel supports  a  particular
       operation  is  to invoke fcntl() with the desired cmd value and then test whether the call
       failed with EINVAL, indicating that the kernel does not recognize this value.

   Duplicating a file descriptor
       F_DUPFD (int)
              Duplicate  the  file  descriptor  fd  using  the  lowest-numbered  available   file
              descriptor  greater  than  or  equal to arg.  This is different from dup2(2), which
              uses exactly the file descriptor specified.

              On success, the new file descriptor is returned.

              See dup(2) for further details.

       F_DUPFD_CLOEXEC (int; since Linux 2.6.24)
              As for F_DUPFD, but additionally set the close-on-exec flag for the duplicate  file
              descriptor.   Specifying this flag permits a program to avoid an additional fcntl()
              F_SETFD operation to set the FD_CLOEXEC flag.  For an explanation of why this  flag
              is useful, see the description of O_CLOEXEC in open(2).

   File descriptor flags
       The following commands manipulate the flags associated with a file descriptor.  Currently,
       only one such flag is defined: FD_CLOEXEC, the close-on-exec flag.  If the FD_CLOEXEC  bit
       is  set,  the  file descriptor will automatically be closed during a successful execve(2).
       (If the execve(2) fails, the file descriptor is left open.)  If the FD_CLOEXEC bit is  not
       set, the file descriptor will remain open across an execve(2).

       F_GETFD (void)
              Return (as the function result) the file descriptor flags; arg is ignored.

       F_SETFD (int)
              Set the file descriptor flags to the value specified by arg.

       In multithreaded programs, using fcntl() F_SETFD to set the close-on-exec flag at the same
       time as another thread performs a fork(2) plus execve(2) is vulnerable to a race condition
       that  may  unintentionally  leak  the file descriptor to the program executed in the child
       process.  See the discussion of the O_CLOEXEC flag in open(2) for details and a remedy  to
       the problem.

   File status flags
       Each open file description has certain associated status flags, initialized by open(2) and
       possibly  modified  by  fcntl().   Duplicated  file   descriptors   (made   with   dup(2),
       fcntl(F_DUPFD), fork(2), etc.) refer to the same open file description, and thus share the
       same file status flags.

       The file status flags and their semantics are described in open(2).

       F_GETFL (void)
              Return (as the function result) the file access mode and the file status flags; arg
              is ignored.

       F_SETFL (int)
              Set  the  file  status  flags  to  the  value  specified  by arg.  File access mode
              (O_RDONLY, O_WRONLY, O_RDWR)  and  file  creation  flags  (i.e.,  O_CREAT,  O_EXCL,
              O_NOCTTY,  O_TRUNC) in arg are ignored.  On Linux, this command can change only the
              O_APPEND, O_ASYNC, O_DIRECT, O_NOATIME, and O_NONBLOCK flags.  It is  not  possible
              to change the O_DSYNC and O_SYNC flags; see BUGS, below.

   Advisory record locking
       Linux  implements traditional ("process-associated") UNIX record locks, as standardized by
       POSIX.  For a Linux-specific alternative with better semantics, see the discussion of open
       file description locks below.

       F_SETLK, F_SETLKW, and F_GETLK are used to acquire, release, and test for the existence of
       record locks (also known as byte-range, file-segment, or file-region  locks).   The  third
       argument,  lock,  is  a  pointer to a structure that has at least the following fields (in
       unspecified order).

           struct flock {
               ...
               short l_type;    /* Type of lock: F_RDLCK,
                                   F_WRLCK, F_UNLCK */
               short l_whence;  /* How to interpret l_start:
                                   SEEK_SET, SEEK_CUR, SEEK_END */
               off_t l_start;   /* Starting offset for lock */
               off_t l_len;     /* Number of bytes to lock */
               pid_t l_pid;     /* PID of process blocking our lock
                                   (set by F_GETLK and F_OFD_GETLK) */
               ...
           };

       The l_whence, l_start, and l_len fields of this structure specify the range  of  bytes  we
       wish  to  lock.   Bytes  past  the end of the file may be locked, but not bytes before the
       start of the file.

       l_start is the starting offset for the lock, and is interpreted relative  to  either:  the
       start  of  the  file  (if  l_whence  is SEEK_SET); the current file offset (if l_whence is
       SEEK_CUR); or the end of the file (if l_whence is SEEK_END).   In  the  final  two  cases,
       l_start  can be a negative number provided the offset does not lie before the start of the
       file.

       l_len specifies the number of bytes to be locked.  If l_len is positive, then the range to
       be  locked  covers  bytes  l_start  up to and including l_start+l_len-1.  Specifying 0 for
       l_len has the special meaning: lock all  bytes  starting  at  the  location  specified  by
       l_whence and l_start through to the end of file, no matter how large the file grows.

       POSIX.1-2001  allows  (but does not require) an implementation to support a negative l_len
       value; if l_len is negative, the interval described by lock covers bytes l_start+l_len  up
       to and including l_start-1.  This is supported since Linux 2.4.21 and Linux 2.5.49.

       The  l_type  field  can  be  used to place a read (F_RDLCK) or a write (F_WRLCK) lock on a
       file.  Any number of processes may hold a read lock (shared lock) on a  file  region,  but
       only  one  process may hold a write lock (exclusive lock).  An exclusive lock excludes all
       other locks, both shared and exclusive.  A single process can hold only one type  of  lock
       on  a file region; if a new lock is applied to an already-locked region, then the existing
       lock is converted to  the  new  lock  type.   (Such  conversions  may  involve  splitting,
       shrinking, or coalescing with an existing lock if the byte range specified by the new lock
       does not precisely coincide with the range of the existing lock.)

       F_SETLK (struct flock *)
              Acquire a lock (when l_type is F_RDLCK or F_WRLCK) or release a lock  (when  l_type
              is  F_UNLCK)  on  the bytes specified by the l_whence, l_start, and l_len fields of
              lock.  If a conflicting lock is held by another process, this call returns  -1  and
              sets  errno  to  EACCES or EAGAIN.  (The error returned in this case differs across
              implementations, so POSIX  requires  a  portable  application  to  check  for  both
              errors.)

       F_SETLKW (struct flock *)
              As  for  F_SETLK, but if a conflicting lock is held on the file, then wait for that
              lock to be released.  If a signal  is  caught  while  waiting,  then  the  call  is
              interrupted  and  (after the signal handler has returned) returns immediately (with
              return value -1 and errno set to EINTR; see signal(7)).

       F_GETLK (struct flock *)
              On input to this call, lock describes a lock we would like to place  on  the  file.
              If  the  lock  could  be  placed,  fcntl()  does not actually place it, but returns
              F_UNLCK in the l_type field of lock and leaves the other fields  of  the  structure
              unchanged.

              If  one  or  more  incompatible  locks  would  prevent this lock being placed, then
              fcntl() returns details about one of those locks in the l_type, l_whence,  l_start,
              and  l_len  fields  of  lock.   If  the conflicting lock is a traditional (process-
              associated) record lock, then the l_pid field is set to  the  PID  of  the  process
              holding  that lock.  If the conflicting lock is an open file description lock, then
              l_pid is set to -1.  Note that the returned information may already be out of  date
              by the time the caller inspects it.

       In  order  to  place  a read lock, fd must be open for reading.  In order to place a write
       lock, fd must be open for writing.  To place both types of lock, open a file read-write.

       When placing locks with F_SETLKW, the  kernel  detects  deadlocks,  whereby  two  or  more
       processes  have their lock requests mutually blocked by locks held by the other processes.
       For example, suppose process A holds a write lock on byte 100 of a  file,  and  process  B
       holds  a  write  lock on byte 200.  If each process then attempts to lock the byte already
       locked by the other  process  using  F_SETLKW,  then,  without  deadlock  detection,  both
       processes  would  remain blocked indefinitely.  When the kernel detects such deadlocks, it
       causes one of the blocking lock requests to immediately fail with the  error  EDEADLK;  an
       application  that encounters such an error should release some of its locks to allow other
       applications to proceed before attempting regain the locks  that  it  requires.   Circular
       deadlocks  involving more than two processes are also detected.  Note, however, that there
       are limitations to the kernel's deadlock-detection algorithm; see BUGS.

       As well as being removed by an explicit F_UNLCK, record locks are  automatically  released
       when the process terminates.

       Record locks are not inherited by a child created via fork(2), but are preserved across an
       execve(2).

       Because of the buffering performed by the stdio(3) library, the use of record locking with
       routines in that package should be avoided; use read(2) and write(2) instead.

       The  record  locks  described  above are associated with the process (unlike the open file
       description locks described below).  This has some unfortunate consequences:

       •  If a process closes any file descriptor referring to a file, then all of the  process's
          locks  on  that  file  are  released, regardless of the file descriptor(s) on which the
          locks were obtained.  This is bad: it means that a process can lose its locks on a file
          such  as  /etc/passwd  or  /etc/mtab when for some reason a library function decides to
          open, read, and close the same file.

       •  The threads in a process share locks.  In other words, a  multithreaded  program  can't
          use  record  locking to ensure that threads don't simultaneously access the same region
          of a file.

       Open file description locks solve both of these problems.

   Open file description locks (non-POSIX)
       Open file description locks are advisory byte-range  locks  whose  operation  is  in  most
       respects  identical  to  the  traditional record locks described above.  This lock type is
       Linux-specific, and available since Linux 3.15.  (There is  a  proposal  with  the  Austin
       Group  to  include this lock type in the next revision of POSIX.1.)  For an explanation of
       open file descriptions, see open(2).

       The principal difference between the two lock types is  that  whereas  traditional  record
       locks  are  associated with a process, open file description locks are associated with the
       open file description on which they are acquired, much like locks acquired with  flock(2).
       Consequently  (and  unlike traditional advisory record locks), open file description locks
       are inherited across fork(2) (and clone(2) with CLONE_FILES), and are  only  automatically
       released  on the last close of the open file description, instead of being released on any
       close of the file.

       Conflicting lock combinations (i.e., a read lock and a write  lock  or  two  write  locks)
       where one lock is an open file description lock and the other is a traditional record lock
       conflict even when they are acquired by the same process on the same file descriptor.

       Open file description locks placed via the same open file description (i.e., via the  same
       file  descriptor,  or  via  a duplicate of the file descriptor created by fork(2), dup(2),
       fcntl() F_DUPFD, and so on) are always compatible: if a new lock is placed on  an  already
       locked  region,  then  the  existing  lock  is  converted  to  the  new  lock type.  (Such
       conversions may result in splitting, shrinking, or coalescing with  an  existing  lock  as
       discussed above.)

       On  the other hand, open file description locks may conflict with each other when they are
       acquired via different open file descriptions.   Thus,  the  threads  in  a  multithreaded
       program  can  use  open  file  description locks to synchronize access to a file region by
       having each thread perform its own  open(2)  on  the  file  and  applying  locks  via  the
       resulting file descriptor.

       As  with  traditional advisory locks, the third argument to fcntl(), lock, is a pointer to
       an flock structure.  By contrast with traditional record locks, the l_pid  field  of  that
       structure must be set to zero when using the commands described below.

       The commands for working with open file description locks are analogous to those used with
       traditional locks:

       F_OFD_SETLK (struct flock *)
              Acquire an open file description lock  (when  l_type  is  F_RDLCK  or  F_WRLCK)  or
              release  an  open  file  description  lock  (when  l_type  is F_UNLCK) on the bytes
              specified by the l_whence, l_start, and l_len fields of  lock.   If  a  conflicting
              lock is held by another process, this call returns -1 and sets errno to EAGAIN.

       F_OFD_SETLKW (struct flock *)
              As  for  F_OFD_SETLK,  but if a conflicting lock is held on the file, then wait for
              that lock to be released.  If a signal is caught while waiting, then  the  call  is
              interrupted  and  (after the signal handler has returned) returns immediately (with
              return value -1 and errno set to EINTR; see signal(7)).

       F_OFD_GETLK (struct flock *)
              On input to this call, lock describes an open file description lock we  would  like
              to place on the file.  If the lock could be placed, fcntl() does not actually place
              it, but returns F_UNLCK in the l_type field of lock and leaves the other fields  of
              the structure unchanged.  If one or more incompatible locks would prevent this lock
              being placed, then details about one of these  locks  are  returned  via  lock,  as
              described above for F_GETLK.

       In  the  current  implementation,  no  deadlock  detection  is  performed  for  open  file
       description locks.  (This contrasts with process-associated record locks,  for  which  the
       kernel does perform deadlock detection.)

   Mandatory locking
       Warning:  the  Linux  implementation  of mandatory locking is unreliable.  See BUGS below.
       Because of these bugs, and the fact that the feature is believed to be little used,  since
       Linux   4.5,  mandatory  locking  has  been  made  an  optional  feature,  governed  by  a
       configuration option (CONFIG_MANDATORY_FILE_LOCKING).  This feature is no longer supported
       at all in Linux 5.15 and above.

       By  default,  both traditional (process-associated) and open file description record locks
       are advisory.  Advisory locks are not enforced and are  useful  only  between  cooperating
       processes.

       Both  lock  types  can also be mandatory.  Mandatory locks are enforced for all processes.
       If a process tries to perform an incompatible access (e.g., read(2) or write(2)) on a file
       region  that  has an incompatible mandatory lock, then the result depends upon whether the
       O_NONBLOCK flag is enabled for its open file description.  If the O_NONBLOCK flag  is  not
       enabled,  then the system call is blocked until the lock is removed or converted to a mode
       that is compatible with the access.  If the O_NONBLOCK flag is enabled,  then  the  system
       call fails with the error EAGAIN.

       To  make  use of mandatory locks, mandatory locking must be enabled both on the filesystem
       that contains the file to be locked, and on the file itself.  Mandatory locking is enabled
       on  a  filesystem  using  the  "-o  mand"  option to mount(8), or the MS_MANDLOCK flag for
       mount(2).  Mandatory locking is enabled on a file by disabling group execute permission on
       the file and enabling the set-group-ID permission bit (see chmod(1) and chmod(2)).

       Mandatory  locking  is  not specified by POSIX.  Some other systems also support mandatory
       locking, although the details of how to enable it vary across systems.

   Lost locks
       When an advisory lock is obtained on a networked filesystem such as  NFS  it  is  possible
       that the lock might get lost.  This may happen due to administrative action on the server,
       or due to a network partition (i.e., loss of network connectivity with the  server)  which
       lasts long enough for the server to assume that the client is no longer functioning.

       When  the  filesystem  determines  that  a  lock has been lost, future read(2) or write(2)
       requests may fail with the error EIO.  This error will persist until the lock  is  removed
       or  the  file  descriptor  is  closed.   Since Linux 3.12, this happens at least for NFSv4
       (including all minor versions).

       Some versions of UNIX send a signal (SIGLOST) in this circumstance.  Linux does not define
       this signal, and does not provide any asynchronous notification of lost locks.

   Managing signals
       F_GETOWN,  F_SETOWN,  F_GETOWN_EX,  F_SETOWN_EX, F_GETSIG, and F_SETSIG are used to manage
       I/O availability signals:

       F_GETOWN (void)
              Return (as the function result) the  process  ID  or  process  group  ID  currently
              receiving  SIGIO  and SIGURG signals for events on file descriptor fd.  Process IDs
              are returned as positive values; process group IDs are returned as negative  values
              (but see BUGS below).  arg is ignored.

       F_SETOWN (int)
              Set  the  process ID or process group ID that will receive SIGIO and SIGURG signals
              for events on the file descriptor fd.  The target process or process  group  ID  is
              specified  in  arg.  A process ID is specified as a positive value; a process group
              ID is specified as a negative value.  Most commonly, the calling process  specifies
              itself as the owner (that is, arg is specified as getpid(2)).

              As  well  as  setting the file descriptor owner, one must also enable generation of
              signals on the file descriptor.  This is done by using the fcntl() F_SETFL  command
              to  set the O_ASYNC file status flag on the file descriptor.  Subsequently, a SIGIO
              signal is sent whenever input or output becomes possible on  the  file  descriptor.
              The  fcntl() F_SETSIG command can be used to obtain delivery of a signal other than
              SIGIO.

              Sending a signal to the owner process (group) specified by F_SETOWN is  subject  to
              the same permissions checks as are described for kill(2), where the sending process
              is the one that employs F_SETOWN (but see BUGS below).  If  this  permission  check
              fails, then the signal is silently discarded.  Note: The F_SETOWN operation records
              the caller's credentials at the time of the fcntl() call, and  it  is  these  saved
              credentials that are used for the permission checks.

              If  the  file descriptor fd refers to a socket, F_SETOWN also selects the recipient
              of SIGURG signals that are delivered when out-of-band data arrives on that  socket.
              (SIGURG  is sent in any situation where select(2) would report the socket as having
              an "exceptional condition".)

              The following was true in Linux 2.6.x up to and including Linux 2.6.11:

                     If a nonzero value is given to F_SETSIG in a multithreaded  process  running
                     with  a  threading  library that supports thread groups (e.g., NPTL), then a
                     positive value given to F_SETOWN has a different meaning: instead of being a
                     process  ID  identifying  a  whole  process, it is a thread ID identifying a
                     specific thread within a process.  Consequently, it may be necessary to pass
                     F_SETOWN  the  result  of  gettid(2)  instead  of  getpid(2) to get sensible
                     results when F_SETSIG is used.  (In current Linux threading implementations,
                     a  main thread's thread ID is the same as its process ID.  This means that a
                     single-threaded program can equally  use  gettid(2)  or  getpid(2)  in  this
                     scenario.)   Note,  however,  that  the  statements in this paragraph do not
                     apply to the SIGURG signal generated for out-of-band data on a socket:  this
                     signal  is  always sent to either a process or a process group, depending on
                     the value given to F_SETOWN.

              The above behavior was accidentally dropped in Linux 2.6.12, and won't be restored.
              From  Linux  2.6.32 onward, use F_SETOWN_EX to target SIGIO and SIGURG signals at a
              particular thread.

       F_GETOWN_EX (struct f_owner_ex *) (since Linux 2.6.32)
              Return the current  file  descriptor  owner  settings  as  defined  by  a  previous
              F_SETOWN_EX  operation.  The information is returned in the structure pointed to by
              arg, which has the following form:

                  struct f_owner_ex {
                      int   type;
                      pid_t pid;
                  };

              The  type  field  will  have  one  of  the  values  F_OWNER_TID,  F_OWNER_PID,   or
              F_OWNER_PGRP.   The  pid  field  is  a  positive  integer representing a thread ID,
              process ID, or process group ID.  See F_SETOWN_EX for more details.

       F_SETOWN_EX (struct f_owner_ex *) (since Linux 2.6.32)
              This operation performs a similar task to F_SETOWN.  It allows the caller to direct
              I/O  availability  signals  to  a  specific thread, process, or process group.  The
              caller specifies the target of signals via arg, which is a pointer to a  f_owner_ex
              structure.  The type field has one of the following values, which define how pid is
              interpreted:

              F_OWNER_TID
                     Send the signal to the thread whose thread ID (the value returned by a  call
                     to clone(2) or gettid(2)) is specified in pid.

              F_OWNER_PID
                     Send the signal to the process whose ID is specified in pid.

              F_OWNER_PGRP
                     Send  the  signal  to the process group whose ID is specified in pid.  (Note
                     that, unlike with F_SETOWN, a process group ID is specified  as  a  positive
                     value here.)

       F_GETSIG (void)
              Return  (as  the  function  result)  the  signal  sent when input or output becomes
              possible.  A value of zero means SIGIO is sent.  Any other value (including  SIGIO)
              is  the  signal  sent instead, and in this case additional info is available to the
              signal handler if installed with SA_SIGINFO.  arg is ignored.

       F_SETSIG (int)
              Set the signal sent when input or output becomes possible to  the  value  given  in
              arg.   A  value  of  zero  means to send the default SIGIO signal.  Any other value
              (including SIGIO) is the signal to send instead, and in this case  additional  info
              is available to the signal handler if installed with SA_SIGINFO.

              By  using  F_SETSIG  with  a  nonzero  value, and setting SA_SIGINFO for the signal
              handler (see sigaction(2)), extra information about I/O events  is  passed  to  the
              handler  in  a  siginfo_t  structure.  If the si_code field indicates the source is
              SI_SIGIO, the si_fd field gives the file  descriptor  associated  with  the  event.
              Otherwise,  there  is  no  indication  which  file descriptors are pending, and you
              should use the usual mechanisms (select(2), poll(2), read(2)  with  O_NONBLOCK  set
              etc.) to determine which file descriptors are available for I/O.

              Note  that  the  file  descriptor  provided  in si_fd is the one that was specified
              during the F_SETSIG operation.  This can lead to an unusual corner  case.   If  the
              file descriptor is duplicated (dup(2) or similar), and the original file descriptor
              is closed, then I/O events will continue to be generated, but the si_fd field  will
              contain the number of the now closed file descriptor.

              By  selecting  a  real  time signal (value >= SIGRTMIN), multiple I/O events may be
              queued using the same signal numbers.  (Queuing is dependent on available  memory.)
              Extra  information  is  available  if  SA_SIGINFO is set for the signal handler, as
              above.

              Note that Linux imposes a limit on the number of  real-time  signals  that  may  be
              queued  to a process (see getrlimit(2) and signal(7)) and if this limit is reached,
              then the kernel reverts to delivering SIGIO, and this signal is  delivered  to  the
              entire process rather than to a specific thread.

       Using  these  mechanisms,  a  program  can  implement fully asynchronous I/O without using
       select(2) or poll(2) most of the time.

       The use of O_ASYNC is specific to BSD and Linux.  The only use of  F_GETOWN  and  F_SETOWN
       specified  in  POSIX.1  is  in  conjunction  with the use of the SIGURG signal on sockets.
       (POSIX does not specify  the  SIGIO  signal.)   F_GETOWN_EX,  F_SETOWN_EX,  F_GETSIG,  and
       F_SETSIG are Linux-specific.  POSIX has asynchronous I/O and the aio_sigevent structure to
       achieve similar things; these are also available in Linux as part of  the  GNU  C  Library
       (glibc).

   Leases
       F_SETLEASE  and  F_GETLEASE  (Linux  2.4  onward)  are  used to establish a new lease, and
       retrieve the current lease,  on  the  open  file  description  referred  to  by  the  file
       descriptor  fd.   A  file lease provides a mechanism whereby the process holding the lease
       (the "lease holder") is notified (via delivery of a signal) when  a  process  (the  "lease
       breaker") tries to open(2) or truncate(2) the file referred to by that file descriptor.

       F_SETLEASE (int)
              Set  or remove a file lease according to which of the following values is specified
              in the integer arg:

              F_RDLCK
                     Take out a read lease.  This will cause the calling process to  be  notified
                     when  the  file  is opened for writing or is truncated.  A read lease can be
                     placed only on a file descriptor that is opened read-only.

              F_WRLCK
                     Take out a write lease.  This will cause the caller to be notified when  the
                     file is opened for reading or writing or is truncated.  A write lease may be
                     placed on a file only if there are no other open file  descriptors  for  the
                     file.

              F_UNLCK
                     Remove our lease from the file.

       Leases  are  associated  with  an  open  file  description (see open(2)).  This means that
       duplicate file descriptors (created by, for example, fork(2) or dup(2)) refer to the  same
       lease,  and  this  lease  may  be  modified  or  released  using any of these descriptors.
       Furthermore, the lease is released by either an explicit F_UNLCK operation on any of these
       duplicate file descriptors, or when all such file descriptors have been closed.

       Leases  may  be  taken  out only on regular files.  An unprivileged process may take out a
       lease only on a file whose UID (owner) matches the  filesystem  UID  of  the  process.   A
       process with the CAP_LEASE capability may take out leases on arbitrary files.

       F_GETLEASE (void)
              Indicates what type of lease is associated with the file descriptor fd by returning
              either F_RDLCK, F_WRLCK, or F_UNLCK, indicating, respectively, a  read  lease  ,  a
              write lease, or no lease.  arg is ignored.

       When  a  process  (the  "lease breaker") performs an open(2) or truncate(2) that conflicts
       with a lease established via F_SETLEASE, the system call is blocked by the kernel and  the
       kernel  notifies  the  lease  holder by sending it a signal (SIGIO by default).  The lease
       holder should respond to receipt of this signal by doing whatever cleanup is  required  in
       preparation for the file to be accessed by another process (e.g., flushing cached buffers)
       and then either remove or downgrade its lease.   A  lease  is  removed  by  performing  an
       F_SETLEASE command specifying arg as F_UNLCK.  If the lease holder currently holds a write
       lease on the file, and the lease breaker is opening the  file  for  reading,  then  it  is
       sufficient  for  the lease holder to downgrade the lease to a read lease.  This is done by
       performing an F_SETLEASE command specifying arg as F_RDLCK.

       If the lease holder fails to downgrade or remove the lease within the  number  of  seconds
       specified in /proc/sys/fs/lease-break-time, then the kernel forcibly removes or downgrades
       the lease holder's lease.

       Once a lease break has been initiated, F_GETLEASE returns the target  lease  type  (either
       F_RDLCK  or  F_UNLCK,  depending on what would be compatible with the lease breaker) until
       the lease holder voluntarily downgrades or removes the lease or the kernel  forcibly  does
       so after the lease break timer expires.

       Once  the  lease  has been voluntarily or forcibly removed or downgraded, and assuming the
       lease breaker has not unblocked its system call, the kernel permits  the  lease  breaker's
       system call to proceed.

       If  the lease breaker's blocked open(2) or truncate(2) is interrupted by a signal handler,
       then the system call fails with the error EINTR,  but  the  other  steps  still  occur  as
       described  above.   If the lease breaker is killed by a signal while blocked in open(2) or
       truncate(2), then the other steps still occur as described above.  If  the  lease  breaker
       specifies  the  O_NONBLOCK flag when calling open(2), then the call immediately fails with
       the error EWOULDBLOCK, but the other steps still occur as described above.

       The default signal used to notify the lease holder is SIGIO, but this can be changed using
       the  F_SETSIG command to fcntl().  If a F_SETSIG command is performed (even one specifying
       SIGIO), and the signal handler is established using  SA_SIGINFO,  then  the  handler  will
       receive a siginfo_t structure as its second argument, and the si_fd field of this argument
       will hold the file descriptor of the  leased  file  that  has  been  accessed  by  another
       process.  (This is useful if the caller holds leases against multiple files.)

   File and directory change notification (dnotify)
       F_NOTIFY (int)
              (Linux 2.4 onward) Provide notification when the directory referred to by fd or any
              of the files that it contains is changed.  The events to be notified are  specified
              in  arg,  which  is  a  bit  mask  specified  by ORing together zero or more of the
              following bits:

              DN_ACCESS
                     A file was accessed (read(2), pread(2), readv(2), and similar)
              DN_MODIFY
                     A  file  was  modified   (write(2),   pwrite(2),   writev(2),   truncate(2),
                     ftruncate(2), and similar).
              DN_CREATE
                     A   file  was  created  (open(2),  creat(2),  mknod(2),  mkdir(2),  link(2),
                     symlink(2), rename(2) into this directory).
              DN_DELETE
                     A file was unlinked (unlink(2), rename(2) to another directory, rmdir(2)).
              DN_RENAME
                     A file was renamed within this directory (rename(2)).
              DN_ATTRIB
                     The attributes  of  a  file  were  changed  (chown(2),  chmod(2),  utime(2),
                     utimensat(2), and similar).

              (In  order  to obtain these definitions, the _GNU_SOURCE feature test macro must be
              defined before including any header files.)

              Directory  notifications  are  normally  "one-shot",  and  the   application   must
              reregister  to  receive  further  notifications.  Alternatively, if DN_MULTISHOT is
              included in arg, then notification will remain in effect until explicitly removed.

              A series of F_NOTIFY requests is cumulative, with the events in arg being added  to
              the set already monitored.  To disable notification of all events, make an F_NOTIFY
              call specifying arg as 0.

              Notification occurs via delivery of a signal.  The default  signal  is  SIGIO,  but
              this can be changed using the F_SETSIG command to fcntl().  (Note that SIGIO is one
              of the nonqueuing standard signals; switching to the  use  of  a  real-time  signal
              means  that  multiple  notifications  can be queued to the process.)  In the latter
              case, the signal handler receives a siginfo_t structure as its second argument  (if
              the handler was established using SA_SIGINFO) and the si_fd field of this structure
              contains  the  file  descriptor  which  generated  the  notification  (useful  when
              establishing notification on multiple directories).

              Especially  when  using  DN_MULTISHOT,  a  real  time  signal  should  be  used for
              notification, so that multiple notifications can be queued.

              NOTE: New applications should use the  inotify  interface  (available  since  Linux
              2.6.13),  which  provides  a much superior interface for obtaining notifications of
              filesystem events.  See inotify(7).

   Changing the capacity of a pipe
       F_SETPIPE_SZ (int; since Linux 2.6.35)
              Change the capacity of the pipe referred to by fd to be at  least  arg  bytes.   An
              unprivileged  process  can adjust the pipe capacity to any value between the system
              page size and  the  limit  defined  in  /proc/sys/fs/pipe-max-size  (see  proc(5)).
              Attempts  to  set  the pipe capacity below the page size are silently rounded up to
              the page size.  Attempts by an unprivileged process to set the pipe capacity  above
              the limit in /proc/sys/fs/pipe-max-size yield the error EPERM; a privileged process
              (CAP_SYS_RESOURCE) can override the limit.

              When allocating the buffer for the pipe, the kernel may use a capacity larger  than
              arg, if that is convenient for the implementation.  (In the current implementation,
              the allocation is the next higher power-of-two page-size multiple of the  requested
              size.)   The  actual  capacity  (in  bytes) that is set is returned as the function
              result.

              Attempting to set the pipe  capacity  smaller  than  the  amount  of  buffer  space
              currently used to store data produces the error EBUSY.

              Note that because of the way the pages of the pipe buffer are employed when data is
              written to the pipe, the number of bytes that can be written may be less  than  the
              nominal size, depending on the size of the writes.

       F_GETPIPE_SZ (void; since Linux 2.6.35)
              Return (as the function result) the capacity of the pipe referred to by fd.

   File Sealing
       File seals limit the set of allowed operations on a given file.  For each seal that is set
       on a file, a specific set of operations will fail with EPERM on this  file  from  now  on.
       The  file  is  said  to  be  sealed.   The default set of seals depends on the type of the
       underlying file and filesystem.  For an overview of file  sealing,  a  discussion  of  its
       purpose, and some code examples, see memfd_create(2).

       Currently, file seals can be applied only to a file descriptor returned by memfd_create(2)
       (if the MFD_ALLOW_SEALING was employed).  On other  filesystems,  all  fcntl()  operations
       that operate on seals will return EINVAL.

       Seals  are  a property of an inode.  Thus, all open file descriptors referring to the same
       inode share the same set of seals.  Furthermore, seals can never be removed, only added.

       F_ADD_SEALS (int; since Linux 3.17)
              Add the seals given in the bit-mask argument arg to the set of seals of  the  inode
              referred  to  by the file descriptor fd.  Seals cannot be removed again.  Once this
              call succeeds, the seals are enforced by the kernel immediately.   If  the  current
              set of seals includes F_SEAL_SEAL (see below), then this call will be rejected with
              EPERM.  Adding a seal that is already set is a no-op, in case  F_SEAL_SEAL  is  not
              set already.  In order to place a seal, the file descriptor fd must be writable.

       F_GET_SEALS (void; since Linux 3.17)
              Return  (as  the function result) the current set of seals of the inode referred to
              by fd.  If no seals are set, 0 is returned.  If the file does not support  sealing,
              -1 is returned and errno is set to EINVAL.

       The following seals are available:

       F_SEAL_SEAL
              If  this  seal  is set, any further call to fcntl() with F_ADD_SEALS fails with the
              error EPERM.  Therefore, this seal prevents any modifications to the set  of  seals
              itself.   If  the  initial  set  of seals of a file includes F_SEAL_SEAL, then this
              effectively causes the set of seals to be constant and locked.

       F_SEAL_SHRINK
              If this seal is set, the file in question cannot be reduced in size.  This  affects
              open(2) with the O_TRUNC flag as well as truncate(2) and ftruncate(2).  Those calls
              fail with EPERM if you try to shrink the file in  question.   Increasing  the  file
              size is still possible.

       F_SEAL_GROW
              If  this  seal  is set, the size of the file in question cannot be increased.  This
              affects write(2) beyond  the  end  of  the  file,  truncate(2),  ftruncate(2),  and
              fallocate(2).   These  calls  fail  with EPERM if you use them to increase the file
              size.  If you keep the size or shrink it, those calls still work as expected.

       F_SEAL_WRITE
              If this seal is set, you cannot  modify  the  contents  of  the  file.   Note  that
              shrinking  or  growing  the  size of the file is still possible and allowed.  Thus,
              this seal is normally used in combination with one of the other seals.   This  seal
              affects    write(2)    and    fallocate(2)    (only   in   combination   with   the
              FALLOC_FL_PUNCH_HOLE flag).  Those calls fail with  EPERM  if  this  seal  is  set.
              Furthermore, trying to create new shared, writable memory-mappings via mmap(2) will
              also fail with EPERM.

              Using the F_ADD_SEALS operation to set the F_SEAL_WRITE seal fails  with  EBUSY  if
              any writable, shared mapping exists.  Such mappings must be unmapped before you can
              add  this  seal.   Furthermore,  if  there  are  any  asynchronous  I/O  operations
              (io_submit(2)) pending on the file, all outstanding writes will be discarded.

       F_SEAL_FUTURE_WRITE (since Linux 5.1)
              The  effect  of  this seal is similar to F_SEAL_WRITE, but the contents of the file
              can still be modified via shared writable mappings that were created prior  to  the
              seal  being  set.   Any  attempt  to  create a new writable mapping on the file via
              mmap(2) will fail with EPERM.  Likewise, an  attempt  to  write  to  the  file  via
              write(2) will fail with EPERM.

              Using  this  seal,  one  process can create a memory buffer that it can continue to
              modify while sharing that buffer on a "read-only" basis with other processes.

   File read/write hints
       Write lifetime hints can be used to inform the kernel about the relative expected lifetime
       of writes on a given inode or via a particular open file description.  (See open(2) for an
       explanation of open file descriptions.)  In this context, the term "write lifetime"  means
       the expected time the data will live on media, before being overwritten or erased.

       An  application  may use the different hint values specified below to separate writes into
       different write classes, so that multiple  users  or  applications  running  on  a  single
       storage  back-end can aggregate their I/O patterns in a consistent manner.  However, there
       are no functional semantics implied by these flags, and different I/O classes can use  the
       write lifetime hints in arbitrary ways, so long as the hints are used consistently.

       The following operations can be applied to the file descriptor, fd:

       F_GET_RW_HINT (uint64_t *; since Linux 4.13)
              Returns  the  value  of  the  read/write  hint associated with the underlying inode
              referred to by fd.

       F_SET_RW_HINT (uint64_t *; since Linux 4.13)
              Sets the read/write hint value associated with the underlying inode referred to  by
              fd.   This  hint  persists until either it is explicitly modified or the underlying
              filesystem is unmounted.

       F_GET_FILE_RW_HINT (uint64_t *; since Linux 4.13)
              Returns the value of the read/write hint associated with the open file  description
              referred to by fd.

       F_SET_FILE_RW_HINT (uint64_t *; since Linux 4.13)
              Sets  the  read/write hint value associated with the open file description referred
              to by fd.

       If an open file description has not been assigned a read/write hint, then it shall use the
       value assigned to the inode, if any.

       The following read/write hints are valid since Linux 4.13:

       RWH_WRITE_LIFE_NOT_SET
              No specific hint has been set.  This is the default value.

       RWH_WRITE_LIFE_NONE
              No specific write lifetime is associated with this file or inode.

       RWH_WRITE_LIFE_SHORT
              Data  written to this inode or via this open file description is expected to have a
              short lifetime.

       RWH_WRITE_LIFE_MEDIUM
              Data written to this inode or via this open file description is expected to have  a
              lifetime longer than data written with RWH_WRITE_LIFE_SHORT.

       RWH_WRITE_LIFE_LONG
              Data  written to this inode or via this open file description is expected to have a
              lifetime longer than data written with RWH_WRITE_LIFE_MEDIUM.

       RWH_WRITE_LIFE_EXTREME
              Data written to this inode or via this open file description is expected to have  a
              lifetime longer than data written with RWH_WRITE_LIFE_LONG.

       All  the  write-specific  hints  are  relative  to  each other, and no individual absolute
       meaning should be attributed to them.

RETURN VALUE

       For a successful call, the return value depends on the operation:

       F_DUPFD
              The new file descriptor.

       F_GETFD
              Value of file descriptor flags.

       F_GETFL
              Value of file status flags.

       F_GETLEASE
              Type of lease held on file descriptor.

       F_GETOWN
              Value of file descriptor owner.

       F_GETSIG
              Value of signal sent when read or write becomes possible, or zero  for  traditional
              SIGIO behavior.

       F_GETPIPE_SZ, F_SETPIPE_SZ
              The pipe capacity.

       F_GET_SEALS
              A  bit  mask  identifying the seals that have been set for the inode referred to by
              fd.

       All other commands
              Zero.

       On error, -1 is returned, and errno is set to indicate the error.

ERRORS

       EACCES or EAGAIN
              Operation is prohibited by locks held by other processes.

       EAGAIN The operation is prohibited because the file  has  been  memory-mapped  by  another
              process.

       EBADF  fd is not an open file descriptor

       EBADF  cmd is F_SETLK or F_SETLKW and the file descriptor open mode doesn't match with the
              type of lock requested.

       EBUSY  cmd is F_SETPIPE_SZ and the new pipe capacity specified in arg is smaller than  the
              amount of buffer space currently used to store data in the pipe.

       EBUSY  cmd  is F_ADD_SEALS, arg includes F_SEAL_WRITE, and there exists a writable, shared
              mapping on the file referred to by fd.

       EDEADLK
              It was detected that the specified F_SETLKW command would cause a deadlock.

       EFAULT lock is outside your accessible address space.

       EINTR  cmd is F_SETLKW or F_OFD_SETLKW and the operation was interrupted by a signal;  see
              signal(7).

       EINTR  cmd  is  F_GETLK,  F_SETLK,  F_OFD_GETLK,  or  F_OFD_SETLK,  and  the operation was
              interrupted by a signal before the lock was checked or acquired.  Most likely  when
              locking a remote file (e.g., locking over NFS), but can sometimes happen locally.

       EINVAL The value specified in cmd is not recognized by this kernel.

       EINVAL cmd is F_ADD_SEALS and arg includes an unrecognized sealing bit.

       EINVAL cmd  is F_ADD_SEALS or F_GET_SEALS and the filesystem containing the inode referred
              to by fd does not support sealing.

       EINVAL cmd is F_DUPFD and arg is negative or is greater than the maximum  allowable  value
              (see the discussion of RLIMIT_NOFILE in getrlimit(2)).

       EINVAL cmd is F_SETSIG and arg is not an allowable signal number.

       EINVAL cmd  is  F_OFD_SETLK,  F_OFD_SETLKW, or F_OFD_GETLK, and l_pid was not specified as
              zero.

       EMFILE cmd is F_DUPFD and the per-process limit on the number of open file descriptors has
              been reached.

       ENOLCK Too  many  segment  locks  open,  lock  table is full, or a remote locking protocol
              failed (e.g., locking over NFS).

       ENOTDIR
              F_NOTIFY was specified in cmd, but fd does not refer to a directory.

       EPERM  cmd is F_SETPIPE_SZ and the soft or hard user pipe  limit  has  been  reached;  see
              pipe(7).

       EPERM  Attempted  to  clear the O_APPEND flag on a file that has the append-only attribute
              set.

       EPERM  cmd was F_ADD_SEALS, but fd was not open for writing or the current set of seals on
              the file already includes F_SEAL_SEAL.

STANDARDS

       SVr4,  4.3BSD,  POSIX.1-2001.   Only  the  operations  F_DUPFD, F_GETFD, F_SETFD, F_GETFL,
       F_SETFL, F_GETLK, F_SETLK, and F_SETLKW are specified in POSIX.1-2001.

       F_GETOWN and F_SETOWN are specified in POSIX.1-2001.  (To get  their  definitions,  define
       either  _XOPEN_SOURCE  with  the  value  500 or greater, or _POSIX_C_SOURCE with the value
       200809L or greater.)

       F_DUPFD_CLOEXEC  is  specified  in  POSIX.1-2008.   (To  get   this   definition,   define
       _POSIX_C_SOURCE  with the value 200809L or greater, or _XOPEN_SOURCE with the value 700 or
       greater.)

       F_GETOWN_EX,  F_SETOWN_EX,  F_SETPIPE_SZ,  F_GETPIPE_SZ,  F_GETSIG,  F_SETSIG,   F_NOTIFY,
       F_GETLEASE,  and  F_SETLEASE  are Linux-specific.  (Define the _GNU_SOURCE macro to obtain
       these definitions.)

       F_OFD_SETLK, F_OFD_SETLKW,  and  F_OFD_GETLK  are  Linux-specific  (and  one  must  define
       _GNU_SOURCE  to obtain their definitions), but work is being done to have them included in
       the next version of POSIX.1.

       F_ADD_SEALS and F_GET_SEALS are Linux-specific.

NOTES

       The errors returned by dup2(2) are different from those returned by F_DUPFD.

   File locking
       The original Linux fcntl() system call was not designed to handle large file  offsets  (in
       the flock structure).  Consequently, an fcntl64() system call was added in Linux 2.4.  The
       newer  system  call  employs  a  different  structure  for  file  locking,  flock64,   and
       corresponding  commands, F_GETLK64, F_SETLK64, and F_SETLKW64.  However, these details can
       be ignored by applications using  glibc,  whose  fcntl()  wrapper  function  transparently
       employs the more recent system call where it is available.

   Record locks
       Since  Linux 2.0, there is no interaction between the types of lock placed by flock(2) and
       fcntl().

       Several systems have more fields in  struct  flock  such  as,  for  example,  l_sysid  (to
       identify  the  machine  where  the lock is held).  Clearly, l_pid alone is not going to be
       very useful if the process holding the lock may live on a  different  machine;  on  Linux,
       while present on some architectures (such as MIPS32), this field is not used.

       The  original  Linux fcntl() system call was not designed to handle large file offsets (in
       the flock structure).  Consequently, an fcntl64() system call was added in Linux 2.4.  The
       newer   system  call  employs  a  different  structure  for  file  locking,  flock64,  and
       corresponding commands, F_GETLK64, F_SETLK64, and F_SETLKW64.  However, these details  can
       be  ignored  by  applications  using  glibc,  whose fcntl() wrapper function transparently
       employs the more recent system call where it is available.

   Record locking and NFS
       Before Linux 3.12, if an NFSv4 client loses contact with the server for a period  of  time
       (defined  as  more than 90 seconds with no communication), it might lose and regain a lock
       without ever being aware of the fact.  (The period of time after which contact is  assumed
       lost  is  known  as the NFSv4 leasetime.  On a Linux NFS server, this can be determined by
       looking at /proc/fs/nfsd/nfsv4leasetime, which  expresses  the  period  in  seconds.   The
       default  value  for  this  file  is 90.)  This scenario potentially risks data corruption,
       since another process might acquire a lock in the intervening period and perform file I/O.

       Since Linux 3.12, if an NFSv4 client loses contact with the server, any I/O to the file by
       a  process  which "thinks" it holds a lock will fail until that process closes and reopens
       the file.  A kernel parameter, nfs.recover_lost_locks, can be  set  to  1  to  obtain  the
       pre-3.12  behavior,  whereby the client will attempt to recover lost locks when contact is
       reestablished with the server.  Because of the attendant risk  of  data  corruption,  this
       parameter defaults to 0 (disabled).

BUGS

   F_SETFL
       It  is  not  possible  to use F_SETFL to change the state of the O_DSYNC and O_SYNC flags.
       Attempts to change the state of these flags are silently ignored.

   F_GETOWN
       A limitation of the Linux system call conventions on  some  architectures  (notably  i386)
       means  that if a (negative) process group ID to be returned by F_GETOWN falls in the range
       -1 to -4095, then the return value is wrongly interpreted by glibc  as  an  error  in  the
       system  call;  that is, the return value of fcntl() will be -1, and errno will contain the
       (positive) process  group  ID.   The  Linux-specific  F_GETOWN_EX  operation  avoids  this
       problem.   Since  glibc  2.11,  glibc  makes  the  kernel  F_GETOWN  problem  invisible by
       implementing F_GETOWN using F_GETOWN_EX.

   F_SETOWN
       In Linux 2.4 and earlier, there is bug that can occur when an  unprivileged  process  uses
       F_SETOWN  to specify the owner of a socket file descriptor as a process (group) other than
       the caller.  In this case, fcntl() can return -1 with errno set to EPERM,  even  when  the
       owner  process  (group) is one that the caller has permission to send signals to.  Despite
       this error return, the file descriptor owner is set, and  signals  will  be  sent  to  the
       owner.

   Deadlock detection
       The  deadlock-detection  algorithm  employed  by  the  kernel  when  dealing with F_SETLKW
       requests can yield both false negatives (failures to detect deadlocks, leaving  a  set  of
       deadlocked  processes blocked indefinitely) and false positives (EDEADLK errors when there
       is no deadlock).  For example, the kernel limits the lock depth of its  dependency  search
       to  10  steps,  meaning  that  circular  deadlock chains that exceed that size will not be
       detected.  In addition, the kernel may falsely  indicate  a  deadlock  when  two  or  more
       processes  created  using  the  clone(2)  CLONE_FILES flag place locks that appear (to the
       kernel) to conflict.

   Mandatory locking
       The Linux implementation of mandatory locking is subject to race conditions  which  render
       it  unreliable:  a  write(2)  call  that  overlaps  with  a lock may modify data after the
       mandatory lock is acquired; a read(2) call that overlaps with a lock may detect changes to
       data  that  were  made  only after a write lock was acquired.  Similar races exist between
       mandatory locks and mmap(2).  It is therefore inadvisable to rely on mandatory locking.

SEE ALSO

       dup2(2), flock(2), open(2), socket(2), lockf(3), capabilities(7),  feature_test_macros(7),
       lslocks(8)

       locks.txt,  mandatory-locking.txt,  and  dnotify.txt  in the Linux kernel source directory
       Documentation/filesystems/  (on  older  kernels,  these  files  are  directly  under   the
       Documentation/ directory, and mandatory-locking.txt is called mandatory.txt)