Another wish.

[exim.git] / doc / doc-misc / Ext-mbx-locking

	 UNIX Advisory File Locking Implications on c-client
		    Mark Crispin, 28 November 1995


	THIS DOCUMENT HAS BEEN UPDATED TO REFLECT THE CODE IN THE
	IMAP-4 TOOLKIT AS OF NOVEMBER 28, 1995.  SOME STATEMENTS
	IN THIS DOCUMENT DO NOT APPLY TO EARLIER VERSIONS OF THE
	IMAP TOOLKIT.

INTRODUCTION

     Advisory locking is a mechanism by which cooperating processes
can signal to each other their usage of a resource and whether or not
that usage is critical.  It is not a mechanism to protect against
processes which do not cooperate in the locking.

     The most basic form of locking involves a counter.  This counter
is -1 when the resource is available.  If a process wants the lock, it
executes an atomic increment-and-test-if-zero.  If the value is zero,
the process has the lock and can execute the critical code that needs
exclusive usage of a resource.  When it is finished, it sets the lock
back to -1.  In C terms:

  while (++lock)		/* try to get lock */
    invoke_other_threads ();	/* failed, try again */
   .
   .	/* critical code  here */
   .
  lock = -1;			/* release lock */

     This particular form of locking appears most commonly in
multi-threaded applications such as operating system kernels.  It
makes several presumptions:
 (1) it is alright to keep testing the lock (no overflow)
 (2) the critical resource is single-access only
 (3) there is shared writeable memory between the two threads
 (4) the threads can be trusted to release the lock when finished

     In applications programming on multi-user systems, most commonly
the other threads are in an entirely different process, which may even
be logged in as a different user.  Few operating systems offer shared
writeable memory between such processes.

     A means of communicating this is by use of a file with a mutually
agreed upon name.  A binary semaphore can be passed by means of the
existance or non-existance of that file, provided that there is an
atomic means to create a file if and only if that file does not exist.
In C terms:

				/* try to get lock */
  while ((fd = open ("lockfile",O_WRONLY|O_CREAT|O_EXCL,0666)) < 0)
    sleep (1);			/* failed, try again */
  close (fd);			/* got the lock */
   .
   .	/* critical code  here */
   .
  unlink ("lockfile");		/* release lock */

     This form of locking makes fewer presumptions, but it still is
guilty of presumptions (2) and (4) above.  Presumption (2) limits the
ability to have processes sharing a resource in a non-conflicting
fashion (e.g. reading from a file).  Presumption (4) leads to
deadlocks should the process crash while it has a resource locked.

     Most modern operating systems provide a resource locking system
call that has none of these presumptions.  In particular, a mechanism
is provided for identifying shared locks as opposed to exclusive
locks.  A shared lock permits other processes to obtain a shared lock,
but denies exclusive locks.  In other words:

	current state		want shared	want exclusive
	-------------		-----------	--------------
	 unlocked		 YES		 YES
	 locked shared		 YES		 NO
	 locked exclusive	 NO		 NO

     Furthermore, the operating system automatically relinquishes all
locks held by that process when it terminates.

     A useful operation is the ability to upgrade a shared lock to
exclusive (provided there are no other shared users of the lock) and
to downgrade an exclusive lock to shared.  It is important that at no
time is the lock ever removed; a process upgrading to exclusive must
not relenquish its shared lock.

     Most commonly, the resources being locked are files.  Shared
locks are particularly important with files; multiple simultaneous
processes can read from a file, but only one can safely write at a
time.  Some writes may be safer than others; an append to the end of
the file is safer than changing existing file data.  In turn, changing
a file record in place is safer than rewriting the file with an
entirely different structure.


FILE LOCKING ON UNIX

     In the oldest versions of UNIX, the use of a semaphore lockfile
was the only available form of locking.  Advisory locking system calls
were not added to UNIX until after the BSD vs. System V split.  Both
of these system calls deal with file resources only.

     Most systems only have one or the other form of locking.  AIX
emulates the BSD form of locking as a jacket into the System V form.
Ultrix and OSF/1 implement both forms.
\f
BSD

     BSD added the flock() system call.  It offers capabilities to
acquire shared lock, acquire exclusive lock, and unlock.  Optionally,
the process can request an immediate error return instead of blocking
when the lock is unavailable.


FLOCK() BUGS

     flock() advertises that it permits upgrading of shared locks to
exclusive and downgrading of exclusive locks to shared, but it does so
by releasing the former lock and then trying to acquire the new lock.
This creates a window of vulnerability in which another process can
grab the exclusive lock.  Therefore, this capability is not useful,
although many programmers have been deluded by incautious reading of
the flock() man page to believe otherwise.  This problem can be
programmed around, once the programmer is aware of it.

     flock() always returns as if it succeeded on NFS files, when in
fact it is a no-op.  There is no way around this.

     Leaving aside these two problems, flock() works remarkably well,
and has shown itself to be robust and trustworthy.
\f
SYSTEM V/POSIX

     System V added new functions to the fnctl() system call, and a
simple interface through the lockf() subroutine.  This was
subsequently included in POSIX.  Both offer the facility to apply the
lock to a particular region of the file instead of to the entire file.
lockf() only supports exclusive locks, and calls fcntl() internally;
hence it won't be discussed further.

     Functionally, fcntl() locking is a superset of flock(); it is
possible to implement a flock() emulator using fcntl(), with one minor
exception: it is not possible to acquire an exclusive lock if the file
is not open for write.

     The fcntl() locking functions are: query lock station of a file
region, lock/unlock a region, and lock/unlock a region and block until
have the lock.  The locks may be shared or exclusive.  By means of the
statd and lockd daemons, fcntl() locking is available on NFS files.

     When statd is started at system boot, it reads its /etc/state
file (which contains the number of times it has been invoked) and
/etc/sm directory (which contains a list of all remote sites which are
client or server locking with this site), and notifies the statd on
each of these systems that it has been restarted.  Each statd then
notifies the local lockd of the restart of that system.

     lockd receives fcntl() requests for NFS files.  It communicates
with the lockd at the server and requests it to apply the lock, and
with the statd to request it for notification when the server goes
down.  It blocks until all these requests are completed.

     There is quite a mythos about fcntl() locking.

     One religion holds that fcntl() locking is the best thing since
sliced bread, and that programs which use flock() should be converted
to fcntl() so that NFS locking will work.  However, as noted above,
very few systems support both calls, so such an exercise is pointless
except on Ultrix and OSF/1.

     Another religion, which I adhere to, has the opposite viewpoint.


FCNTL() BUGS

     For all of the hairy code to do individual section locking of a
file, it's clear that the designers of fcntl() locking never
considered some very basic locking operations.  It's as if all they
knew about locking they got out of some CS textbook with not
investigation of real-world needs.

     It is not possible to acquire an exclusive lock unless the file
is open for write.  You could have append with shared read, and thus
you could have a case in which a read-only access may need to go
exclusive.  This problem can be programmed around once the programmer
is aware of it.

     If the file is opened on another file designator in the same
process, the file is unlocked even if no attempt is made to do any
form of locking on the second designator.  This is a very bad bug.  It
means that an application must keep track of all the files that it has
opened and locked.

     If there is no statd/lockd on the NFS server, fcntl() will hang
forever waiting for them to appear.  This is a bad bug.  It means that
any attempt to lock on a server that doesn't run these daemons will
hang.  There is no way for an application to request flock() style
``try to lock, but no-op if the mechanism ain't there''.

     There is a rumor to the effect that fcntl() will hang forever on
local files too if there is no local statd/lockd.  These daemons are
running on mailer.u, although they appear not to have much CPU time.
A useful experiment would be to kill them and see if imapd is affected
in any way, but I decline to do so without an OK from UCS!  ;-) If
killing statd/lockd can be done without breaking fcntl() on local
files, this would become one of the primary means of dealing with this
problem.

     The statd and lockd daemons have quite a reputation for extreme
fragility.  There have been numerous reports about the locking
mechanism being wedged on a systemwide or even clusterwide basis,
requiring a reboot to clear.  It is rumored that this wedge, once it
happens, also blocks local locking.  Presumably killing and restarting
statd would suffice to clear the wedge, but I haven't verified this.

     There appears to be a limit to how many locks may be in use at a
time on the system, although the documentation only mentions it in
passing.  On some of their systems, UCS has increased lockd's ``size
of the socket buffer'', whatever that means.
\f
C-CLIENT USAGE

     c-client uses flock().  On System V systems, flock() is simulated
by an emulator that calls fcntl().  This emulator is provided by some
systems (e.g. AIX), or uses c-client's flock.c module.


BEZERK AND MMDF

     Locking in the traditional UNIX formats was largely dictated by
the status quo in other applications; however, additional protection
is added against inadvertantly running multiple instances of a
c-client application on the same mail file.

     (1) c-client attempts to create a .lock file (mail file name with
``.lock'' appended) whenever it reads from, or writes to, the mail
file.  This is an exclusive lock, and is held only for short periods
of time while c-client is actually doing the I/O.  There is a 5-minute
timeout for this lock, after which it is broken on the presumption
that it is a stale lock.  If it can not create the .lock file due to
an EACCES (protection failure) error, it once silently proceeded
without this lock; this was for systems which protect /usr/spool/mail
from unprivileged processes creating files.  Today, c-client reports
an error unless it is built otherwise.  The purpose of this lock is to
prevent against unfavorable interactions with mail delivery.

     (2) c-client applies a shared flock() to the mail file whenever
it reads from the mail file, and an exclusive flock() whenever it
writes to the mail file.  This lock is freed as soon as it finishes
reading.  The purpose of this lock is to prevent against unfavorable
interactions with mail delivery.

     (3) c-client applies an exclusive flock() to a file on /tmp
(whose name represents the device and inode number of the file) when
it opens the mail file.  This lock is maintained throughout the
session, although c-client has a feature (called ``kiss of death'')
which permits c-client to forcibly and irreversibly seize the lock
from a cooperating c-client application that surrenders the lock on
demand.  The purpose of this lock is to prevent against unfavorable
interactions with other instances of c-client (rewriting the mail
file).

     Mail delivery daemons use lock (1), (2), or both.  Lock (1) works
over NFS; lock (2) is the only one that works on sites that protect
/usr/spool/mail against unprivileged file creation.  Prudent mail
delivery daemons use both forms of locking, and of course so does
c-client.

     If only lock (2) is used, then multiple processes can read from
the mail file simultaneously, although in real life this doesn't
really change things.  The normal state of locks (1) and (2) is
unlocked except for very brief periods.


TENEX AND MTX

     The design of the locking mechanism of these formats was
motivated by a design to enable multiple simultaneous read/write
access.  It is almost the reverse of how locking works with
bezerk/mmdf.

     (1) c-client applies a shared flock() to the mail file when it
opens the mail file.  It upgrades this lock to exclusive whenever it
tries to expunge the mail file.  Because of the flock() bug that
upgrading a lock actually releases it, it will not do so until it has
acquired an exclusive lock (2) first.  The purpose of this lock is to
prevent against expunge taking place while some other c-client has the
mail file open (and thus knows where all the messages are).

     (2) c-client applies a shared flock() to a file on /tmp (whose
name represents the device and inode number of the file) when it
parses the mail file.  It applies an exclusive flock() to this file
when it appends new mail to the mail file, as well as before it
attempts to upgrade lock (1) to exclusive.  The purpose of this lock
is to prevent against data being appended while some other c-client is
parsing mail in the file (to prevent reading of incomplete messages).
It also protects against the lock-releasing timing race on lock (1).
\f
OBSERVATIONS

     In a perfect world, locking works.  You are protected against
unfavorable interactions with the mailer and against your own mistake
by running more than one instance of your mail reader.  In tenex/mtx
formats, you have the additional benefit that multiple simultaneous
read/write access works, with the sole restriction being that you
can't expunge if there are any sharers of the mail file.

     If the mail file is NFS-mounted, then flock() locking is a silent
no-op.  This is the way BSD implements flock(), and c-client's
emulation of flock() through fcntl() tests for NFS files and
duplicates this functionality.  There is no locking protection for
tenex/mtx mail files at all, and only protection against the mailer
for bezerk/mmdf mail files.  This has been the accepted state of
affairs on UNIX for many sad years.

     If you can not create .lock files, it should not affect locking,
since the flock() locks suffice for all protection.  This is, however,
not true if the mailer does not check for flock() locking, or if the
the mail file is NFS-mounted.

     What this means is that there is *no* locking protection at all
in the case of a client using an NFS-mounted /usr/spool/mail that does
not permit file creation by unprivileged programs.  It is impossible,
under these circumstances, for an unprivileged program to do anything
about it.  Worse, if EACCES errors on .lock file creation are no-op'ed
, the user won't even know about it.  This is arguably a site
configuration error.

     The problem with not being able to create .lock files exists on
System V as well, but the failure modes for flock() -- which is
implemented via fcntl() -- are different.

     On System V, if the mail file is NFS-mounted and either the
client or the server lacks a functioning statd/lockd pair, then the
lock attempt would have hung forever if it weren't for the fact that
c-client tests for NFS and no-ops the flock() emulator in this case.
Systemwide or clusterwide failures of statd/lockd have been known to
occur which cause all locks in all processes to hang (including
local?).  Without the special NFS test made by c-client, there would
be no way to request BSD-style no-op behavior, nor is there any way to
determine that this is happening other than the system being hung.

     The additional locking introduced by c-client was shown to cause
much more stress on the System V locking mechanism than has
traditionally been placed upon it.  If it was stressed too far, all
hell broke loose.  Fortunately, this is now past history.
\f
TRADEOFFS

     c-client based applications have a reasonable chance of winning
as long as you don't use NFS for remote access to mail files.  That's
what IMAP is for, after all.  It is, however, very important to
realize that you can *not* use the lock-upgrade feature by itself
because it releases the lock as an interim step -- you need to have
lock-upgrading guarded by another lock.

     If you have the misfortune of using System V, you are likely to
run into problems sooner or later having to do with statd/lockd.  You
basically end up with one of three unsatisfactory choices:
	1) Grit your teeth and live with it.
	2) Try to make it work:
	   a) avoid NFS access so as not to stress statd/lockd.
	   b) try to understand the code in statd/lockd and hack it
	      to be more robust.
	   c) hunt out the system limit of locks, if there is one,
	      and increase it.	Figure on at least two locks per
	      simultaneous imapd process and four locks per Pine
	      process.	Better yet, make the limit be 10 times the
	      maximum number of processes.
	   d) increase the socket buffer (-S switch to lockd) if
	      it is offered.  I don't know what this actually does,
	      but giving lockd more resources to do its work can't
	      hurt.  Maybe.
	3) Decide that it can't possibly work, and turn off the
	   fcntl() calls in your program.
	4) If nuking statd/lockd can be done without breaking local
	   locking, then do so.  This would make SVR4 have the same
	   limitations as BSD locking, with a couple of additional
	   bugs.
	5) Check for NFS, and don't do the fcntl() in the NFS case.
	   This is what c-client does.

     Note that if you are going to use NFS to access files on a server
which does not have statd/lockd running, your only choice is (3), (4),
or (5).  Here again, IMAP can bail you out.

     These problems aren't unique to c-client applications; they have
also been reported with Elm, Mediamail, and other email tools.

     Of the other two SVR4 locking bugs:

     Programmer awareness is necessary to deal with the bug that you
can not get an exclusive lock unless the file is open for write.  I
believe that c-client has fixed all of these cases.

     The problem about opening a second designator smashing any
current locks on the file has not been addressed satisfactorily yet.
This is not an easy problem to deal with, especially in c-client which
really doesn't know what other files/streams may be open by Pine.

     Aren't you so happy that you bought an System V system?