[deliverable/linux.git] / Documentation / filesystems / inotify.txt

				   inotify
	    a powerful yet simple file change notification system


Document started 15 Mar 2005 by Robert Love <rml@novell.com>


(i) User Interface

Inotify is controlled by a set of three system calls and normal file I/O on a
returned file descriptor.

First step in using inotify is to initialise an inotify instance:

	int fd = inotify_init ();

Each instance is associated with a unique, ordered queue.

Change events are managed by "watches".  A watch is an (object,mask) pair where
the object is a file or directory and the mask is a bit mask of one or more
inotify events that the application wishes to receive.  See <linux/inotify.h>
for valid events.  A watch is referenced by a watch descriptor, or wd.

Watches are added via a path to the file.

Watches on a directory will return events on any files inside of the directory.

Adding a watch is simple:

	int wd = inotify_add_watch (fd, path, mask);

Where "fd" is the return value from inotify_init(), path is the path to the
object to watch, and mask is the watch mask (see <linux/inotify.h>).

You can update an existing watch in the same manner, by passing in a new mask.

An existing watch is removed via

	int ret = inotify_rm_watch (fd, wd);

Events are provided in the form of an inotify_event structure that is read(2)
from a given inotify instance.  The filename is of dynamic length and follows
the struct. It is of size len.  The filename is padded with null bytes to
ensure proper alignment.  This padding is reflected in len.

You can slurp multiple events by passing a large buffer, for example

	size_t len = read (fd, buf, BUF_LEN);

Where "buf" is a pointer to an array of "inotify_event" structures at least
BUF_LEN bytes in size.  The above example will return as many events as are
available and fit in BUF_LEN.

Each inotify instance fd is also select()- and poll()-able.

You can find the size of the current event queue via the standard FIONREAD
ioctl on the fd returned by inotify_init().

All watches are destroyed and cleaned up on close.


(ii)

Prototypes:

	int inotify_init (void);
	int inotify_add_watch (int fd, const char *path, __u32 mask);
	int inotify_rm_watch (int fd, __u32 mask);


(iii) Internal Kernel Implementation

Each inotify instance is associated with an inotify_device structure.

Each watch is associated with an inotify_watch structure.  Watches are chained
off of each associated device and each associated inode.

See fs/inotify.c for the locking and lifetime rules.


(iv) Rationale

Q: What is the design decision behind not tying the watch to the open fd of
   the watched object?

A: Watches are associated with an open inotify device, not an open file.
   This solves the primary problem with dnotify: keeping the file open pins
   the file and thus, worse, pins the mount.  Dnotify is therefore infeasible
   for use on a desktop system with removable media as the media cannot be
   unmounted.  Watching a file should not require that it be open.

Q: What is the design decision behind using an-fd-per-instance as opposed to
   an fd-per-watch?

A: An fd-per-watch quickly consumes more file descriptors than are allowed,
   more fd's than are feasible to manage, and more fd's than are optimally
   select()-able.  Yes, root can bump the per-process fd limit and yes, users
   can use epoll, but requiring both is a silly and extraneous requirement.
   A watch consumes less memory than an open file, separating the number
   spaces is thus sensible.  The current design is what user-space developers
   want: Users initialize inotify, once, and add n watches, requiring but one
   fd and no twiddling with fd limits.  Initializing an inotify instance two
   thousand times is silly.  If we can implement user-space's preferences 
   cleanly--and we can, the idr layer makes stuff like this trivial--then we 
   should.

   There are other good arguments.  With a single fd, there is a single
   item to block on, which is mapped to a single queue of events.  The single
   fd returns all watch events and also any potential out-of-band data.  If
   every fd was a separate watch,

   - There would be no way to get event ordering.  Events on file foo and
     file bar would pop poll() on both fd's, but there would be no way to tell
     which happened first.  A single queue trivially gives you ordering.  Such
     ordering is crucial to existing applications such as Beagle.  Imagine
     "mv a b ; mv b a" events without ordering.

   - We'd have to maintain n fd's and n internal queues with state,
     versus just one.  It is a lot messier in the kernel.  A single, linear
     queue is the data structure that makes sense.

   - User-space developers prefer the current API.  The Beagle guys, for
     example, love it.  Trust me, I asked.  It is not a surprise: Who'd want
     to manage and block on 1000 fd's via select?

   - No way to get out of band data.

   - 1024 is still too low.  ;-)

   When you talk about designing a file change notification system that
   scales to 1000s of directories, juggling 1000s of fd's just does not seem
   the right interface.  It is too heavy.

   Additionally, it _is_ possible to  more than one instance  and
   juggle more than one queue and thus more than one associated fd.  There
   need not be a one-fd-per-process mapping; it is one-fd-per-queue and a
   process can easily want more than one queue.

Q: Why the system call approach?

A: The poor user-space interface is the second biggest problem with dnotify.
   Signals are a terrible, terrible interface for file notification.  Or for
   anything, for that matter.  The ideal solution, from all perspectives, is a
   file descriptor-based one that allows basic file I/O and poll/select.
   Obtaining the fd and managing the watches could have been done either via a
   device file or a family of new system calls.  We decided to implement a
   family of system calls because that is the preffered approach for new kernel
   interfaces.  The only real difference was whether we wanted to use open(2)
   and ioctl(2) or a couple of new system calls.  System calls beat ioctls.
Commit	Line	Data
6f97933d RL	1	inotify
6f97933d RL	2	a powerful yet simple file change notification system
0eeca283 RL	3
	4
	5
	6	Document started 15 Mar 2005 by Robert Love <rml@novell.com>
	7
6f97933d	8
0eeca283 RL	9	(i) User Interface
0eeca283 RL	10
6f97933d RL	11	Inotify is controlled by a set of three system calls and normal file I/O on a
6f97933d RL	12	returned file descriptor.
0eeca283	13
6f97933d	14	First step in using inotify is to initialise an inotify instance:
0eeca283 RL	15
	16	int fd = inotify_init ();
	17
6f97933d RL	18	Each instance is associated with a unique, ordered queue.
6f97933d RL	19
0eeca283 RL	20	Change events are managed by "watches". A watch is an (object,mask) pair where
	21	the object is a file or directory and the mask is a bit mask of one or more
	22	inotify events that the application wishes to receive. See <linux/inotify.h>
	23	for valid events. A watch is referenced by a watch descriptor, or wd.
	24
	25	Watches are added via a path to the file.
	26
	27	Watches on a directory will return events on any files inside of the directory.
	28
6f97933d	29	Adding a watch is simple:
0eeca283 RL	30
	31	int wd = inotify_add_watch (fd, path, mask);
	32
6f97933d RL	33	Where "fd" is the return value from inotify_init(), path is the path to the
6f97933d RL	34	object to watch, and mask is the watch mask (see <linux/inotify.h>).
0eeca283 RL	35
	36	You can update an existing watch in the same manner, by passing in a new mask.
	37
6f97933d	38	An existing watch is removed via
0eeca283	39
6f97933d	40	int ret = inotify_rm_watch (fd, wd);
0eeca283 RL	41
0eeca283 RL	42	Events are provided in the form of an inotify_event structure that is read(2)
6f97933d RL	43	from a given inotify instance. The filename is of dynamic length and follows
	44	the struct. It is of size len. The filename is padded with null bytes to
	45	ensure proper alignment. This padding is reflected in len.
0eeca283 RL	46
	47	You can slurp multiple events by passing a large buffer, for example
	48
	49	size_t len = read (fd, buf, BUF_LEN);
	50
6f97933d RL	51	Where "buf" is a pointer to an array of "inotify_event" structures at least
	52	BUF_LEN bytes in size. The above example will return as many events as are
	53	available and fit in BUF_LEN.
0eeca283	54
6f97933d	55	Each inotify instance fd is also select()- and poll()-able.
0eeca283	56
6f97933d RL	57	You can find the size of the current event queue via the standard FIONREAD
6f97933d RL	58	ioctl on the fd returned by inotify_init().
0eeca283 RL	59
	60	All watches are destroyed and cleaned up on close.
	61
	62
6f97933d RL	63	(ii)
	64
	65	Prototypes:
	66
	67	int inotify_init (void);
	68	int inotify_add_watch (int fd, const char *path, __u32 mask);
	69	int inotify_rm_watch (int fd, __u32 mask);
	70
0eeca283	71
6f97933d RL	72	(iii) Internal Kernel Implementation
	73
	74	Each inotify instance is associated with an inotify_device structure.
0eeca283 RL	75
	76	Each watch is associated with an inotify_watch structure. Watches are chained
	77	off of each associated device and each associated inode.
	78
	79	See fs/inotify.c for the locking and lifetime rules.
	80
	81
6f97933d	82	(iv) Rationale
0eeca283 RL	83
	84	Q: What is the design decision behind not tying the watch to the open fd of
	85	the watched object?
	86
	87	A: Watches are associated with an open inotify device, not an open file.
	88	This solves the primary problem with dnotify: keeping the file open pins
	89	the file and thus, worse, pins the mount. Dnotify is therefore infeasible
	90	for use on a desktop system with removable media as the media cannot be
6f97933d	91	unmounted. Watching a file should not require that it be open.
0eeca283	92
6f97933d	93	Q: What is the design decision behind using an-fd-per-instance as opposed to
0eeca283 RL	94	an fd-per-watch?
	95
	96	A: An fd-per-watch quickly consumes more file descriptors than are allowed,
	97	more fd's than are feasible to manage, and more fd's than are optimally
	98	select()-able. Yes, root can bump the per-process fd limit and yes, users
	99	can use epoll, but requiring both is a silly and extraneous requirement.
	100	A watch consumes less memory than an open file, separating the number
	101	spaces is thus sensible. The current design is what user-space developers
6f97933d RL	102	want: Users initialize inotify, once, and add n watches, requiring but one
6f97933d RL	103	fd and no twiddling with fd limits. Initializing an inotify instance two
0eeca283 RL	104	thousand times is silly. If we can implement user-space's preferences
	105	cleanly--and we can, the idr layer makes stuff like this trivial--then we
	106	should.
	107
	108	There are other good arguments. With a single fd, there is a single
	109	item to block on, which is mapped to a single queue of events. The single
	110	fd returns all watch events and also any potential out-of-band data. If
	111	every fd was a separate watch,
	112
	113	- There would be no way to get event ordering. Events on file foo and
	114	file bar would pop poll() on both fd's, but there would be no way to tell
	115	which happened first. A single queue trivially gives you ordering. Such
	116	ordering is crucial to existing applications such as Beagle. Imagine
	117	"mv a b ; mv b a" events without ordering.
	118
	119	- We'd have to maintain n fd's and n internal queues with state,
	120	versus just one. It is a lot messier in the kernel. A single, linear
	121	queue is the data structure that makes sense.
	122
	123	- User-space developers prefer the current API. The Beagle guys, for
	124	example, love it. Trust me, I asked. It is not a surprise: Who'd want
	125	to manage and block on 1000 fd's via select?
	126
0eeca283 RL	127	- No way to get out of band data.
	128
	129	- 1024 is still too low. ;-)
	130
	131	When you talk about designing a file change notification system that
	132	scales to 1000s of directories, juggling 1000s of fd's just does not seem
	133	the right interface. It is too heavy.
	134
6f97933d RL	135	Additionally, it _is_ possible to more than one instance and
	136	juggle more than one queue and thus more than one associated fd. There
	137	need not be a one-fd-per-process mapping; it is one-fd-per-queue and a
	138	process can easily want more than one queue.
	139
0eeca283 RL	140	Q: Why the system call approach?
	141
	142	A: The poor user-space interface is the second biggest problem with dnotify.
	143	Signals are a terrible, terrible interface for file notification. Or for
	144	anything, for that matter. The ideal solution, from all perspectives, is a
	145	file descriptor-based one that allows basic file I/O and poll/select.
	146	Obtaining the fd and managing the watches could have been done either via a
	147	device file or a family of new system calls. We decided to implement a
	148	family of system calls because that is the preffered approach for new kernel
6f97933d RL	149	interfaces. The only real difference was whether we wanted to use open(2)
6f97933d RL	150	and ioctl(2) or a couple of new system calls. System calls beat ioctls.
0eeca283	151