[exim.git] / doc / doc-misc / Ext-mbx-locking
1 UNIX Advisory File Locking Implications on c-client
2 Mark Crispin, 28 November 1995
12 Advisory locking is a mechanism by which cooperating processes
13 can signal to each other their usage of a resource and whether or not
14 that usage is critical. It is not a mechanism to protect against
15 processes which do not cooperate in the locking.
17 The most basic form of locking involves a counter. This counter
18 is -1 when the resource is available. If a process wants the lock, it
19 executes an atomic increment-and-test-if-zero. If the value is zero,
20 the process has the lock and can execute the critical code that needs
21 exclusive usage of a resource. When it is finished, it sets the lock
22 back to -1. In C terms:
24 while (++lock) /* try to get lock */
25 invoke_other_threads (); /* failed, try again */
26 .
27 . /* critical code here */
28 .
29 lock = -1; /* release lock */
31 This particular form of locking appears most commonly in
32 multi-threaded applications such as operating system kernels. It
33 makes several presumptions:
34 (1) it is alright to keep testing the lock (no overflow)
35 (2) the critical resource is single-access only
36 (3) there is shared writeable memory between the two threads
37 (4) the threads can be trusted to release the lock when finished
39 In applications programming on multi-user systems, most commonly
40 the other threads are in an entirely different process, which may even
41 be logged in as a different user. Few operating systems offer shared
42 writeable memory between such processes.
44 A means of communicating this is by use of a file with a mutually
45 agreed upon name. A binary semaphore can be passed by means of the
46 existence or non-existence of that file, provided that there is an
47 atomic means to create a file if and only if that file does not exist.
48 In C terms:
50 /* try to get lock */
51 while ((fd = open ("lockfile",O_WRONLY|O_CREAT|O_EXCL,0666)) < 0)
52 sleep (1); /* failed, try again */
53 close (fd); /* got the lock */
54 .
55 . /* critical code here */
56 .
57 unlink ("lockfile"); /* release lock */
59 This form of locking makes fewer presumptions, but it still is
60 guilty of presumptions (2) and (4) above. Presumption (2) limits the
61 ability to have processes sharing a resource in a non-conflicting
62 fashion (e.g. reading from a file). Presumption (4) leads to
63 deadlocks should the process crash while it has a resource locked.
65 Most modern operating systems provide a resource locking system
66 call that has none of these presumptions. In particular, a mechanism
67 is provided for identifying shared locks as opposed to exclusive
68 locks. A shared lock permits other processes to obtain a shared lock,
69 but denies exclusive locks. In other words:
71 current state want shared want exclusive
72 ------------- ----------- --------------
73 unlocked YES YES
74 locked shared YES NO
75 locked exclusive NO NO
77 Furthermore, the operating system automatically relinquishes all
78 locks held by that process when it terminates.
80 A useful operation is the ability to upgrade a shared lock to
81 exclusive (provided there are no other shared users of the lock) and
82 to downgrade an exclusive lock to shared. It is important that at no
83 time is the lock ever removed; a process upgrading to exclusive must
84 not relinquish its shared lock.
86 Most commonly, the resources being locked are files. Shared
87 locks are particularly important with files; multiple simultaneous
88 processes can read from a file, but only one can safely write at a
89 time. Some writes may be safer than others; an append to the end of
90 the file is safer than changing existing file data. In turn, changing
91 a file record in place is safer than rewriting the file with an
92 entirely different structure.
97 In the oldest versions of UNIX, the use of a semaphore lockfile
98 was the only available form of locking. Advisory locking system calls
99 were not added to UNIX until after the BSD vs. System V split. Both
100 of these system calls deal with file resources only.
102 Most systems only have one or the other form of locking. AIX
103 emulates the BSD form of locking as a jacket into the System V form.
104 Ultrix and OSF/1 implement both forms.
105 \f
106 BSD
108 BSD added the flock() system call. It offers capabilities to
109 acquire shared lock, acquire exclusive lock, and unlock. Optionally,
110 the process can request an immediate error return instead of blocking
111 when the lock is unavailable.
116 flock() advertises that it permits upgrading of shared locks to
117 exclusive and downgrading of exclusive locks to shared, but it does so
118 by releasing the former lock and then trying to acquire the new lock.
119 This creates a window of vulnerability in which another process can
120 grab the exclusive lock. Therefore, this capability is not useful,
121 although many programmers have been deluded by incautious reading of
122 the flock() man page to believe otherwise. This problem can be
123 programmed around, once the programmer is aware of it.
125 flock() always returns as if it succeeded on NFS files, when in
126 fact it is a no-op. There is no way around this.
128 Leaving aside these two problems, flock() works remarkably well,
129 and has shown itself to be robust and trustworthy.
130 \f
133 System V added new functions to the fnctl() system call, and a
134 simple interface through the lockf() subroutine. This was
135 subsequently included in POSIX. Both offer the facility to apply the
136 lock to a particular region of the file instead of to the entire file.
137 lockf() only supports exclusive locks, and calls fcntl() internally;
138 hence it won't be discussed further.
140 Functionally, fcntl() locking is a superset of flock(); it is
141 possible to implement a flock() emulator using fcntl(), with one minor
142 exception: it is not possible to acquire an exclusive lock if the file
143 is not open for write.
145 The fcntl() locking functions are: query lock station of a file
146 region, lock/unlock a region, and lock/unlock a region and block until
147 have the lock. The locks may be shared or exclusive. By means of the
148 statd and lockd daemons, fcntl() locking is available on NFS files.
150 When statd is started at system boot, it reads its /etc/state
151 file (which contains the number of times it has been invoked) and
152 /etc/sm directory (which contains a list of all remote sites which are
153 client or server locking with this site), and notifies the statd on
154 each of these systems that it has been restarted. Each statd then
155 notifies the local lockd of the restart of that system.
157 lockd receives fcntl() requests for NFS files. It communicates
158 with the lockd at the server and requests it to apply the lock, and
159 with the statd to request it for notification when the server goes
160 down. It blocks until all these requests are completed.
162 There is quite a mythos about fcntl() locking.
164 One religion holds that fcntl() locking is the best thing since
165 sliced bread, and that programs which use flock() should be converted
166 to fcntl() so that NFS locking will work. However, as noted above,
167 very few systems support both calls, so such an exercise is pointless
168 except on Ultrix and OSF/1.
170 Another religion, which I adhere to, has the opposite viewpoint.
175 For all of the hairy code to do individual section locking of a
176 file, it's clear that the designers of fcntl() locking never
177 considered some very basic locking operations. It's as if all they
178 knew about locking they got out of some CS textbook with not
179 investigation of real-world needs.
181 It is not possible to acquire an exclusive lock unless the file
182 is open for write. You could have append with shared read, and thus
183 you could have a case in which a read-only access may need to go
184 exclusive. This problem can be programmed around once the programmer
185 is aware of it.
187 If the file is opened on another file designator in the same
188 process, the file is unlocked even if no attempt is made to do any
189 form of locking on the second designator. This is a very bad bug. It
190 means that an application must keep track of all the files that it has
191 opened and locked.
193 If there is no statd/lockd on the NFS server, fcntl() will hang
194 forever waiting for them to appear. This is a bad bug. It means that
195 any attempt to lock on a server that doesn't run these daemons will
196 hang. There is no way for an application to request flock() style
197 ``try to lock, but no-op if the mechanism ain't there''.
199 There is a rumor to the effect that fcntl() will hang forever on
200 local files too if there is no local statd/lockd. These daemons are
201 running on mailer.u, although they appear not to have much CPU time.
202 A useful experiment would be to kill them and see if imapd is affected
203 in any way, but I decline to do so without an OK from UCS! ;-) If
204 killing statd/lockd can be done without breaking fcntl() on local
205 files, this would become one of the primary means of dealing with this
206 problem.
208 The statd and lockd daemons have quite a reputation for extreme
209 fragility. There have been numerous reports about the locking
210 mechanism being wedged on a systemwide or even clusterwide basis,
211 requiring a reboot to clear. It is rumored that this wedge, once it
212 happens, also blocks local locking. Presumably killing and restarting
213 statd would suffice to clear the wedge, but I haven't verified this.
215 There appears to be a limit to how many locks may be in use at a
216 time on the system, although the documentation only mentions it in
217 passing. On some of their systems, UCS has increased lockd's ``size
218 of the socket buffer'', whatever that means.
219 \f
222 c-client uses flock(). On System V systems, flock() is simulated
223 by an emulator that calls fcntl(). This emulator is provided by some
224 systems (e.g. AIX), or uses c-client's flock.c module.
229 Locking in the traditional UNIX formats was largely dictated by
230 the status quo in other applications; however, additional protection
231 is added against inadvertantly running multiple instances of a
232 c-client application on the same mail file.
234 (1) c-client attempts to create a .lock file (mail file name with
235 ``.lock'' appended) whenever it reads from, or writes to, the mail
236 file. This is an exclusive lock, and is held only for short periods
237 of time while c-client is actually doing the I/O. There is a 5-minute
238 timeout for this lock, after which it is broken on the presumption
239 that it is a stale lock. If it can not create the .lock file due to
240 an EACCES (protection failure) error, it once silently proceeded
241 without this lock; this was for systems which protect /usr/spool/mail
242 from unprivileged processes creating files. Today, c-client reports
243 an error unless it is built otherwise. The purpose of this lock is to
244 prevent against unfavorable interactions with mail delivery.
246 (2) c-client applies a shared flock() to the mail file whenever
247 it reads from the mail file, and an exclusive flock() whenever it
248 writes to the mail file. This lock is freed as soon as it finishes
249 reading. The purpose of this lock is to prevent against unfavorable
250 interactions with mail delivery.
252 (3) c-client applies an exclusive flock() to a file on /tmp
253 (whose name represents the device and inode number of the file) when
254 it opens the mail file. This lock is maintained throughout the
255 session, although c-client has a feature (called ``kiss of death'')
256 which permits c-client to forcibly and irreversibly seize the lock
257 from a cooperating c-client application that surrenders the lock on
258 demand. The purpose of this lock is to prevent against unfavorable
259 interactions with other instances of c-client (rewriting the mail
260 file).
262 Mail delivery daemons use lock (1), (2), or both. Lock (1) works
263 over NFS; lock (2) is the only one that works on sites that protect
264 /usr/spool/mail against unprivileged file creation. Prudent mail
265 delivery daemons use both forms of locking, and of course so does
266 c-client.
268 If only lock (2) is used, then multiple processes can read from
269 the mail file simultaneously, although in real life this doesn't
270 really change things. The normal state of locks (1) and (2) is
271 unlocked except for very brief periods.
276 The design of the locking mechanism of these formats was
277 motivated by a design to enable multiple simultaneous read/write
278 access. It is almost the reverse of how locking works with
279 bezerk/mmdf.
281 (1) c-client applies a shared flock() to the mail file when it
282 opens the mail file. It upgrades this lock to exclusive whenever it
283 tries to expunge the mail file. Because of the flock() bug that
284 upgrading a lock actually releases it, it will not do so until it has
285 acquired an exclusive lock (2) first. The purpose of this lock is to
286 prevent against expunge taking place while some other c-client has the
287 mail file open (and thus knows where all the messages are).
289 (2) c-client applies a shared flock() to a file on /tmp (whose
290 name represents the device and inode number of the file) when it
291 parses the mail file. It applies an exclusive flock() to this file
292 when it appends new mail to the mail file, as well as before it
293 attempts to upgrade lock (1) to exclusive. The purpose of this lock
294 is to prevent against data being appended while some other c-client is
295 parsing mail in the file (to prevent reading of incomplete messages).
296 It also protects against the lock-releasing timing race on lock (1).
297 \f
300 In a perfect world, locking works. You are protected against
301 unfavorable interactions with the mailer and against your own mistake
302 by running more than one instance of your mail reader. In tenex/mtx
303 formats, you have the additional benefit that multiple simultaneous
304 read/write access works, with the sole restriction being that you
305 can't expunge if there are any sharers of the mail file.
307 If the mail file is NFS-mounted, then flock() locking is a silent
308 no-op. This is the way BSD implements flock(), and c-client's
309 emulation of flock() through fcntl() tests for NFS files and
310 duplicates this functionality. There is no locking protection for
311 tenex/mtx mail files at all, and only protection against the mailer
312 for bezerk/mmdf mail files. This has been the accepted state of
313 affairs on UNIX for many sad years.
315 If you can not create .lock files, it should not affect locking,
316 since the flock() locks suffice for all protection. This is, however,
317 not true if the mailer does not check for flock() locking, or if the
318 the mail file is NFS-mounted.
320 What this means is that there is *no* locking protection at all
321 in the case of a client using an NFS-mounted /usr/spool/mail that does
322 not permit file creation by unprivileged programs. It is impossible,
323 under these circumstances, for an unprivileged program to do anything
324 about it. Worse, if EACCES errors on .lock file creation are no-op'ed
325 , the user won't even know about it. This is arguably a site
326 configuration error.
328 The problem with not being able to create .lock files exists on
329 System V as well, but the failure modes for flock() -- which is
330 implemented via fcntl() -- are different.
332 On System V, if the mail file is NFS-mounted and either the
333 client or the server lacks a functioning statd/lockd pair, then the
334 lock attempt would have hung forever if it weren't for the fact that
335 c-client tests for NFS and no-ops the flock() emulator in this case.
336 Systemwide or clusterwide failures of statd/lockd have been known to
337 occur which cause all locks in all processes to hang (including
338 local?). Without the special NFS test made by c-client, there would
339 be no way to request BSD-style no-op behavior, nor is there any way to
340 determine that this is happening other than the system being hung.
342 The additional locking introduced by c-client was shown to cause
343 much more stress on the System V locking mechanism than has
344 traditionally been placed upon it. If it was stressed too far, all
345 hell broke loose. Fortunately, this is now past history.
346 \f
349 c-client based applications have a reasonable chance of winning
350 as long as you don't use NFS for remote access to mail files. That's
351 what IMAP is for, after all. It is, however, very important to
352 realize that you can *not* use the lock-upgrade feature by itself
353 because it releases the lock as an interim step -- you need to have
354 lock-upgrading guarded by another lock.
356 If you have the misfortune of using System V, you are likely to
357 run into problems sooner or later having to do with statd/lockd. You
358 basically end up with one of three unsatisfactory choices:
359 1) Grit your teeth and live with it.
360 2) Try to make it work:
361 a) avoid NFS access so as not to stress statd/lockd.
362 b) try to understand the code in statd/lockd and hack it
363 to be more robust.
364 c) hunt out the system limit of locks, if there is one,
365 and increase it. Figure on at least two locks per
366 simultaneous imapd process and four locks per Pine
367 process. Better yet, make the limit be 10 times the
368 maximum number of processes.
369 d) increase the socket buffer (-S switch to lockd) if
370 it is offered. I don't know what this actually does,
371 but giving lockd more resources to do its work can't
372 hurt. Maybe.
373 3) Decide that it can't possibly work, and turn off the
374 fcntl() calls in your program.
375 4) If nuking statd/lockd can be done without breaking local
376 locking, then do so. This would make SVR4 have the same
377 limitations as BSD locking, with a couple of additional
378 bugs.
379 5) Check for NFS, and don't do the fcntl() in the NFS case.
380 This is what c-client does.
382 Note that if you are going to use NFS to access files on a server
383 which does not have statd/lockd running, your only choice is (3), (4),
384 or (5). Here again, IMAP can bail you out.
386 These problems aren't unique to c-client applications; they have
387 also been reported with Elm, Mediamail, and other email tools.
389 Of the other two SVR4 locking bugs:
391 Programmer awareness is necessary to deal with the bug that you
392 can not get an exclusive lock unless the file is open for write. I
393 believe that c-client has fixed all of these cases.
395 The problem about opening a second designator smashing any
396 current locks on the file has not been addressed satisfactorily yet.
397 This is not an easy problem to deal with, especially in c-client which
398 really doesn't know what other files/streams may be open by Pine.
400 Aren't you so happy that you bought an System V system?