man7/inotify.7

   1 '\" t
   2 .\" Copyright (C) 2006 Michael Kerrisk <mtk.manpages@gmail.com>
   3 .\"
   4 .\" %%%LICENSE_START(VERBATIM)
   5 .\" Permission is granted to make and distribute verbatim copies of this
   6 .\" manual provided the copyright notice and this permission notice are
   7 .\" preserved on all copies.
   8 .\"
   9 .\" Permission is granted to copy and distribute modified versions of this
  10 .\" manual under the conditions for verbatim copying, provided that the
  11 .\" entire resulting derived work is distributed under the terms of a
  12 .\" permission notice identical to this one.
  13 .\"
  14 .\" Since the Linux kernel and libraries are constantly changing, this
  15 .\" manual page may be incorrect or out-of-date.  The author(s) assume no
  16 .\" responsibility for errors or omissions, or for damages resulting from
  17 .\" the use of the information contained herein.  The author(s) may not
  18 .\" have taken the same level of care in the production of this manual,
  19 .\" which is licensed free of charge, as they might when working
  20 .\" professionally.
  21 .\"
  22 .\" Formatted or processed versions of this manual, if unaccompanied by
  23 .\" the source, must acknowledge the copyright and authors of this work.
  24 .\" %%%LICENSE_END
  25 .\"
  26 .TH INOTIFY 7 2014-03-28 "Linux" "Linux Programmer's Manual"
  27 .SH NAME
  28 inotify \- monitoring filesystem events
  29 .SH DESCRIPTION
  30 The
  31 .I inotify
  32 API provides a mechanism for monitoring filesystem events.
  33 Inotify can be used to monitor individual files,
  34 or to monitor directories.
  35 When a directory is monitored, inotify will return events
  36 for the directory itself, and for files inside the directory.
  37
  38 The following system calls are used with this API:
  39 .BR inotify_init (2)
  40 (or
  41 .BR inotify_init1 (2)),
  42 .BR inotify_add_watch (2),
  43 .BR inotify_rm_watch (2),
  44 .BR read (2),
  45 and
  46 .BR close (2).
  47
  48 .BR inotify_init (2)
  49 creates an inotify instance and returns a file descriptor
  50 referring to the inotify instance.
  51 The more recent
  52 .BR inotify_init1 (2)
  53 is like
  54 .BR inotify_init (2),
  55 but provides some extra functionality.
  56
  57 .BR inotify_add_watch (2)
  58 manipulates the "watch list" associated with an inotify instance.
  59 Each item ("watch") in the watch list specifies the pathname of
  60 a file or directory,
  61 along with some set of events that the kernel should monitor for the
  62 file referred to by that pathname.
  63 .BR inotify_add_watch (2)
  64 either creates a new watch item, or modifies an existing watch.
  65 Each watch has a unique "watch descriptor", an integer
  66 returned by
  67 .BR inotify_add_watch (2)
  68 when the watch is created.
  69
  70 .BR inotify_rm_watch (2)
  71 removes an item from an inotify watch list.
  72
  73 When all file descriptors referring to an inotify
  74 instance have been closed,
  75 the underlying object and its resources are
  76 freed for reuse by the kernel;
  77 all associated watches are automatically freed.
  78
  79 To determine what events have occurred, an application
  80 .BR read (2)s
  81 from the inotify file descriptor.
  82 If no events have so far occurred, then,
  83 assuming a blocking file descriptor,
  84 .BR read (2)
  85 will block until at least one event occurs
  86 (unless interrupted by a signal,
  87 in which case the call fails with the error
  88 .BR EINTR ;
  89 see
  90 .BR signal (7)).
  91
  92 Each successful
  93 .BR read (2)
  94 returns a buffer containing one or more of the following structures:
  95 .in +4n
  96 .nf
  97
  98 struct inotify_event {
  99     int      wd;       /* Watch descriptor */
 100 .\" FIXME . The type of the 'wd' field should probably be "int32_t".
 101 .\" I submitted a patch to fix this.  See the LKML thread
 102 .\" "[patch] Fix type errors in inotify interfaces", 18 Nov 2008
 103 .\" Glibc bug filed: http://sources.redhat.com/bugzilla/show_bug.cgi?id=7040
 104     uint32_t mask;     /* Mask of events */
 105     uint32_t cookie;   /* Unique cookie associating related
 106                           events (for rename(2)) */
 107     uint32_t len;      /* Size of \fIname\fP field */
 108     char     name[];   /* Optional null-terminated name */
 109 };
 110 .fi
 111 .in
 112
 113 .I wd
 114 identifies the watch for which this event occurs.
 115 It is one of the watch descriptors returned by a previous call to
 116 .BR inotify_add_watch (2).
 117
 118 .I mask
 119 contains bits that describe the event that occurred (see below).
 120
 121 .I cookie
 122 is a unique integer that connects related events.
 123 Currently this is used only for rename events, and
 124 allows the resulting pair of
 125 .B IN_MOVED_FROM
 126 and
 127 .B IN_MOVED_TO
 128 events to be connected by the application.
 129 For all other event types,
 130 .I cookie
 131 is set to 0.
 132
 133 The
 134 .I name
 135 field is present only when an event is returned
 136 for a file inside a watched directory;
 137 it identifies the file pathname relative to the watched directory.
 138 This pathname is null-terminated,
 139 and may include further null bytes (\(aq\\0\(aq) to align subsequent reads to a
 140 suitable address boundary.
 141
 142 The
 143 .I len
 144 field counts all of the bytes in
 145 .IR name ,
 146 including the null bytes;
 147 the length of each
 148 .I inotify_event
 149 structure is thus
 150 .IR "sizeof(struct inotify_event)+len" .
 151
 152 The behavior when the buffer given to
 153 .BR read (2)
 154 is too small to return information about the next event depends
 155 on the kernel version: in kernels before 2.6.21,
 156 .BR read (2)
 157 returns 0; since kernel 2.6.21,
 158 .BR read (2)
 159 fails with the error
 160 .BR EINVAL .
 161 Specifying a buffer of size
 162
 163     sizeof(struct inotify_event) + NAME_MAX + 1
 164
 165 will be sufficient to read at least one event.
 166 .SS inotify events
 167 The
 168 .BR inotify_add_watch (2)
 169 .I mask
 170 argument and the
 171 .I mask
 172 field of the
 173 .I inotify_event
 174 structure returned when
 175 .BR read (2)ing
 176 an inotify file descriptor are both bit masks identifying
 177 inotify events.
 178 The following bits can be specified in
 179 .I mask
 180 when calling
 181 .BR inotify_add_watch (2)
 182 and may be returned in the
 183 .I mask
 184 field returned by
 185 .BR read (2):
 186 .TP
 187 .B IN_ACCESS
 188 File was accessed (read,
 189 .BR execve (2))
 190 (*).
 191 .TP
 192 .B IN_ATTRIB
 193 Metadata changed\(emfor example, permissions, timestamps, extended attributes,
 194 link count (since Linux 2.6.25), UID, or GID. (*).
 195 .TP
 196 .B IN_CLOSE_WRITE
 197 File opened for writing was closed (*).
 198 .TP
 199 .B IN_CLOSE_NOWRITE
 200 File not opened for writing was closed (*).
 201 .TP
 202 .B IN_CREATE
 203 File/directory created in watched directory (*).
 204 .TP
 205 .B IN_DELETE
 206 File/directory deleted from watched directory (*).
 207 .TP
 208 .B IN_DELETE_SELF
 209 Watched file/directory was itself deleted.
 210 (This event also occurs if an object is moved to another filesystem,
 211 since
 212 .BR mv (1)
 213 in effect copies the file to the other filesystem and
 214 then deletes it from original filesystem.)
 215 .TP
 216 .B IN_MODIFY
 217 File was modified (*).
 218 .TP
 219 .B IN_MOVE_SELF
 220 Watched file/directory was itself moved.
 221 .TP
 222 .B IN_MOVED_FROM
 223 Generated for the directory containing the old filename
 224 when a file is renamed (*).
 225 .TP
 226 .B IN_MOVED_TO
 227 Generated for the directory containing the new filename
 228 when a file is renamed (*).
 229 .TP
 230 .B IN_OPEN
 231 File was opened (*).
 232 .PP
 233 When monitoring a directory,
 234 the events marked with an asterisk (*) above can occur for
 235 files in the directory, in which case the
 236 .I name
 237 field in the returned
 238 .I inotify_event
 239 structure identifies the name of the file within the directory.
 240 .PP
 241 The
 242 .B IN_ALL_EVENTS
 243 macro is defined as a bit mask of all of the above events.
 244 This macro can be used as the
 245 .I mask
 246 argument when calling
 247 .BR inotify_add_watch (2).
 248
 249 Two additional convenience macros are defined:
 250 .TP
 251 .BR IN_MOVE
 252 Equates to
 253 .BR "IN_MOVED_FROM | IN_MOVED_TO" .
 254 .TP
 255 .BR IN_CLOSE
 256 Equates to
 257 .BR "IN_CLOSE_WRITE | IN_CLOSE_NOWRITE" .
 258 .PP
 259 The following further bits can be specified in
 260 .I mask
 261 when calling
 262 .BR inotify_add_watch (2):
 263 .TP
 264 .BR IN_DONT_FOLLOW " (since Linux 2.6.15)"
 265 Don't dereference
 266 .I pathname
 267 if it is a symbolic link.
 268 .TP
 269 .BR IN_EXCL_UNLINK " (since Linux 2.6.36)"
 270 .\" commit 8c1934c8d70b22ca8333b216aec6c7d09fdbd6a6
 271 By default, when watching events on the children of a directory,
 272 events are generated for children even after they have been unlinked
 273 from the directory.
 274 This can result in large numbers of uninteresting events for
 275 some applications (e.g., if watching
 276 .IR /tmp ,
 277 in which many applications create temporary files whose
 278 names are immediately unlinked).
 279 Specifying
 280 .B IN_EXCL_UNLINK
 281 changes the default behavior,
 282 so that events are not generated for children after
 283 they have been unlinked from the watched directory.
 284 .TP
 285 .B IN_MASK_ADD
 286 Add (OR) events to watch mask for this pathname if
 287 it already exists (instead of replacing mask).
 288 .TP
 289 .B IN_ONESHOT
 290 Monitor
 291 .I pathname
 292 for one event, then remove from
 293 watch list.
 294 .TP
 295 .BR IN_ONLYDIR " (since Linux 2.6.15)"
 296 Only watch
 297 .I pathname
 298 if it is a directory.
 299 .PP
 300 The following bits may be set in the
 301 .I mask
 302 field returned by
 303 .BR read (2):
 304 .TP
 305 .B IN_IGNORED
 306 Watch was removed explicitly
 307 .RB ( inotify_rm_watch (2))
 308 or automatically (file was deleted, or filesystem was unmounted).
 309 See also BUGS.
 310 .TP
 311 .B IN_ISDIR
 312 Subject of this event is a directory.
 313 .TP
 314 .B IN_Q_OVERFLOW
 315 Event queue overflowed
 316 .RI ( wd
 317 is \-1 for this event).
 318 .TP
 319 .B IN_UNMOUNT
 320 Filesystem containing watched object was unmounted.
 321 .SS /proc interfaces
 322 The following interfaces can be used to limit the amount of
 323 kernel memory consumed by inotify:
 324 .TP
 325 .I /proc/sys/fs/inotify/max_queued_events
 326 The value in this file is used when an application calls
 327 .BR inotify_init (2)
 328 to set an upper limit on the number of events that can be
 329 queued to the corresponding inotify instance.
 330 Events in excess of this limit are dropped, but an
 331 .B IN_Q_OVERFLOW
 332 event is always generated.
 333 .TP
 334 .I /proc/sys/fs/inotify/max_user_instances
 335 This specifies an upper limit on the number of inotify instances
 336 that can be created per real user ID.
 337 .TP
 338 .I /proc/sys/fs/inotify/max_user_watches
 339 This specifies an upper limit on the number of watches
 340 that can be created per real user ID.
 341 .SH VERSIONS
 342 Inotify was merged into the 2.6.13 Linux kernel.
 343 The required library interfaces were added to glibc in version 2.4.
 344 .RB ( IN_DONT_FOLLOW ,
 345 .BR IN_MASK_ADD ,
 346 and
 347 .B IN_ONLYDIR
 348 were added in version 2.5.)
 349 .SH CONFORMING TO
 350 The inotify API is Linux-specific.
 351 .SH NOTES
 352 Inotify file descriptors can be monitored using
 353 .BR select (2),
 354 .BR poll (2),
 355 and
 356 .BR epoll (7).
 357 When an event is available, the file descriptor indicates as readable.
 358
 359 Since Linux 2.6.25,
 360 signal-driven I/O notification is available for inotify file descriptors;
 361 see the discussion of
 362 .B F_SETFL
 363 (for setting the
 364 .B O_ASYNC
 365 flag),
 366 .BR F_SETOWN ,
 367 and
 368 .B F_SETSIG
 369 in
 370 .BR fcntl (2).
 371 The
 372 .I siginfo_t
 373 structure (described in
 374 .BR sigaction (2))
 375 that is passed to the signal handler has the following fields set:
 376 .IR si_fd
 377 is set to the inotify file descriptor number;
 378 .IR si_signo
 379 is set to the signal number;
 380 .IR si_code
 381 is set to
 382 .BR POLL_IN ;
 383 and
 384 .B POLLIN
 385 is set in
 386 .IR si_band .
 387
 388 If successive output inotify events produced on the
 389 inotify file descriptor are identical (same
 390 .IR wd ,
 391 .IR mask ,
 392 .IR cookie ,
 393 and
 394 .IR name ),
 395 then they are coalesced into a single event if the
 396 older event has not yet been read (but see BUGS).
 397 This reduces the amount of kernel memory required for the event queue,
 398 but also means that an application can't use inotify to reliably count
 399 file events.
 400
 401 The events returned by reading from an inotify file descriptor
 402 form an ordered queue.
 403 Thus, for example, it is guaranteed that when renaming from
 404 one directory to another, events will be produced in the
 405 correct order on the inotify file descriptor.
 406
 407 The
 408 .B FIONREAD
 409 .BR ioctl (2)
 410 returns the number of bytes available to read from an
 411 inotify file descriptor.
 412 .SS Limitations and caveats
 413 Inotify monitoring of directories is not recursive:
 414 to monitor subdirectories under a directory,
 415 additional watches must be created.
 416 This can take a significant amount time for large directory trees.
 417
 418 The inotify API provides no information about the user or process that
 419 triggered the inotify event.
 420 In particular, there is no easy
 421 way for a process that is monitoring events via inotify
 422 to distinguish events that it triggers
 423 itself from those that are triggered by other processes.
 424
 425 Note that the event queue can overflow.
 426 In this case, events are lost.
 427 Robust applications should handle the possibility of
 428 lost events gracefully.
 429
 430 The inotify API identifies affected files by filename.
 431 However, by the time an application processes an inotify event,
 432 the filename may already have been deleted or renamed.
 433
 434 If monitoring an entire directory subtree,
 435 and a new subdirectory is created in that tree or an existing directory
 436 is renamed into that tree,
 437 be aware that by the time you create a watch for the new subdirectory,
 438 new files (and subdirectories) may already exist inside the subdirectory.
 439 Therefore, you may want to scan the contents of the subdirectory
 440 immediately after adding the watch (and, if desired,
 441 recursively add watches for any subdirectories that it contains).
 442
 443 The inotify applications identifies events via watch descriptors.
 444 It is the application's responsibility to cache a mapping
 445 (if one is needed) between watch descriptors and pathnames.
 446 Be aware that directory renamings may affect multiple cached pathnames.
 447 .SH BUGS
 448 .\" FIXME kernel commit 611da04f7a31b2208e838be55a42c7a1310ae321
 449 .\" implies that unmount events were buggy 2.6.11 to 2.6.36
 450 .\"
 451 In kernels before 2.6.16, the
 452 .B IN_ONESHOT
 453 .I mask
 454 flag does not work.
 455
 456 As originally designed and implemented, the
 457 .B IN_ONESHOT
 458 flag did not cause an
 459 .B IN_IGNORED
 460 event to be generated when the watch was dropped after one event.
 461 However, as an unintended effect of other changes,
 462 since Linux 2.6.36, an
 463 .B IN_IGNORED
 464 event is generated in this case.
 465
 466 Before kernel 2.6.25,
 467 .\" commit 1c17d18e3775485bf1e0ce79575eb637a94494a2
 468 the kernel code that was intended to coalesce successive identical events
 469 (i.e., the two most recent events could potentially be coalesced
 470 if the older had not yet been read)
 471 instead checked if the most recent event could be coalesced with the
 472 .I oldest
 473 unread event.
 474 .SH SEE ALSO
 475 .BR inotifywait (1),
 476 .BR inotifywatch (1),
 477 .BR inotify_add_watch (2),
 478 .BR inotify_init (2),
 479 .BR inotify_init1 (2),
 480 .BR inotify_rm_watch (2),
 481 .BR read (2),
 482 .BR stat (2)
 483
 484 .IR Documentation/filesystems/inotify.txt
 485 in the Linux kernel source tree