1	Mandatory File Locking For The Linux Operating System
2
3		Andy Walker <andy@lysaker.kvaerner.no>
4
5			   15 April 1996
6
7
81. What is  mandatory locking?
9------------------------------
10
11Mandatory locking is kernel enforced file locking, as opposed to the more usual
12cooperative file locking used to guarantee sequential access to files among
13processes. File locks are applied using the flock() and fcntl() system calls
14(and the lockf() library routine which is a wrapper around fcntl().) It is
15normally a process' responsibility to check for locks on a file it wishes to
16update, before applying its own lock, updating the file and unlocking it again.
17The most commonly used example of this (and in the case of sendmail, the most
18troublesome) is access to a user's mailbox. The mail user agent and the mail
19transfer agent must guard against updating the mailbox at the same time, and
20prevent reading the mailbox while it is being updated.
21
22In a perfect world all processes would use and honour a cooperative, or
23"advisory" locking scheme. However, the world isn't perfect, and there's
24a lot of poorly written code out there.
25
26In trying to address this problem, the designers of System V UNIX came up
27with a "mandatory" locking scheme, whereby the operating system kernel would
28block attempts by a process to write to a file that another process holds a
29"read" -or- "shared" lock on, and block attempts to both read and write to a
30file that a process holds a "write " -or- "exclusive" lock on.
31
32The System V mandatory locking scheme was intended to have as little impact as
33possible on existing user code. The scheme is based on marking individual files
34as candidates for mandatory locking, and using the existing fcntl()/lockf()
35interface for applying locks just as if they were normal, advisory locks.
36
37Note 1: In saying "file" in the paragraphs above I am actually not telling
38the whole truth. System V locking is based on fcntl(). The granularity of
39fcntl() is such that it allows the locking of byte ranges in files, in addition
40to entire files, so the mandatory locking rules also have byte level
41granularity.
42
43Note 2: POSIX.1 does not specify any scheme for mandatory locking, despite
44borrowing the fcntl() locking scheme from System V. The mandatory locking
45scheme is defined by the System V Interface Definition (SVID) Version 3.
46
472. Marking a file for mandatory locking
48---------------------------------------
49
50A file is marked as a candidate for mandatory locking by setting the group-id
51bit in its file mode but removing the group-execute bit. This is an otherwise
52meaningless combination, and was chosen by the System V implementors so as not
53to break existing user programs.
54
55Note that the group-id bit is usually automatically cleared by the kernel when
56a setgid file is written to. This is a security measure. The kernel has been
57modified to recognize the special case of a mandatory lock candidate and to
58refrain from clearing this bit. Similarly the kernel has been modified not
59to run mandatory lock candidates with setgid privileges.
60
613. Available implementations
62----------------------------
63
64I have considered the implementations of mandatory locking available with
65SunOS 4.1.x, Solaris 2.x and HP-UX 9.x.
66
67Generally I have tried to make the most sense out of the behaviour exhibited
68by these three reference systems. There are many anomalies.
69
70All the reference systems reject all calls to open() for a file on which
71another process has outstanding mandatory locks. This is in direct
72contravention of SVID 3, which states that only calls to open() with the
73O_TRUNC flag set should be rejected. The Linux implementation follows the SVID
74definition, which is the "Right Thing", since only calls with O_TRUNC can
75modify the contents of the file.
76
77HP-UX even disallows open() with O_TRUNC for a file with advisory locks, not
78just mandatory locks. That would appear to contravene POSIX.1.
79
80mmap() is another interesting case. All the operating systems mentioned
81prevent mandatory locks from being applied to an mmap()'ed file, but  HP-UX
82also disallows advisory locks for such a file. SVID actually specifies the
83paranoid HP-UX behaviour.
84
85In my opinion only MAP_SHARED mappings should be immune from locking, and then
86only from mandatory locks - that is what is currently implemented.
87
88SunOS is so hopeless that it doesn't even honour the O_NONBLOCK flag for
89mandatory locks, so reads and writes to locked files always block when they
90should return EAGAIN.
91
92I'm afraid that this is such an esoteric area that the semantics described
93below are just as valid as any others, so long as the main points seem to
94agree.
95
964. Semantics
97------------
98
991. Mandatory locks can only be applied via the fcntl()/lockf() locking
100   interface - in other words the System V/POSIX interface. BSD style
101   locks using flock() never result in a mandatory lock.
102
1032. If a process has locked a region of a file with a mandatory read lock, then
104   other processes are permitted to read from that region. If any of these
105   processes attempts to write to the region it will block until the lock is
106   released, unless the process has opened the file with the O_NONBLOCK
107   flag in which case the system call will return immediately with the error
108   status EAGAIN.
109
1103. If a process has locked a region of a file with a mandatory write lock, all
111   attempts to read or write to that region block until the lock is released,
112   unless a process has opened the file with the O_NONBLOCK flag in which case
113   the system call will return immediately with the error status EAGAIN.
114
1154. Calls to open() with O_TRUNC, or to creat(), on a existing file that has
116   any mandatory locks owned by other processes will be rejected with the
117   error status EAGAIN.
118
1195. Attempts to apply a mandatory lock to a file that is memory mapped and
120   shared (via mmap() with MAP_SHARED) will be rejected with the error status
121   EAGAIN.
122
1236. Attempts to create a shared memory map of a file (via mmap() with MAP_SHARED)
124   that has any mandatory locks in effect will be rejected with the error status
125   EAGAIN.
126
1275. Which system calls are affected?
128-----------------------------------
129
130Those which modify a file's contents, not just the inode. That gives read(),
131write(), readv(), writev(), open(), creat(), mmap(), truncate() and
132ftruncate(). truncate() and ftruncate() are considered to be "write" actions
133for the purposes of mandatory locking.
134
135The affected region is usually defined as stretching from the current position
136for the total number of bytes read or written. For the truncate calls it is
137defined as the bytes of a file removed or added (we must also consider bytes
138added, as a lock can specify just "the whole file", rather than a specific
139range of bytes.)
140
141Note 3: I may have overlooked some system calls that need mandatory lock
142checking in my eagerness to get this code out the door. Please let me know, or
143better still fix the system calls yourself and submit a patch to me or Linus.
144
1456. Warning!
146-----------
147
148Not even root can override a mandatory lock, so runaway processes can wreak
149havoc if they lock crucial files. The way around it is to change the file
150permissions (remove the setgid bit) before trying to read or write to it.
151Of course, that might be a bit tricky if the system is hung :-(
152
153