1EPOLL(7) Linux Programmer's Manual EPOLL(7)
2
6 epoll - I/O event notification facility
7
9 #include <sys/epoll.h>
10
12 The epoll API performs a similar task to poll(2): monitoring multiple
13 file descriptors to see if I/O is possible on any of them. The epoll
14 API can be used either as an edge-triggered or a level-triggered inter‐
15 face and scales well to large numbers of watched file descriptors. The
16 following system calls are provided to create and manage an epoll
17 instance:
18 * epoll_create(2) creates an epoll instance and returns a file
20 descriptor referring to that instance. (The more recent epoll_cre‐
21 ate1(2) extends the functionality of epoll_create(2).)
22 * Interest in particular file descriptors is then registered via
24 epoll_ctl(2). The set of file descriptors currently registered on
25 an epoll instance is sometimes called an epoll set.
26 * epoll_wait(2) waits for I/O events, blocking the calling thread if
28 no events are currently available.
29 Level-triggered and edge-triggered
31 The epoll event distribution interface is able to behave both as edge-
32 triggered (ET) and as level-triggered (LT). The difference between the
33 two mechanisms can be described as follows. Suppose that this scenario
34 happens:
35 1. The file descriptor that represents the read side of a pipe (rfd) is
37 registered on the epoll instance.
38 2. A pipe writer writes 2 kB of data on the write side of the pipe.
40 3. A call to epoll_wait(2) is done that will return rfd as a ready file
42 descriptor.
43 4. The pipe reader reads 1 kB of data from rfd.
45 5. A call to epoll_wait(2) is done.
47 If the rfd file descriptor has been added to the epoll interface using
49 the EPOLLET (edge-triggered) flag, the call to epoll_wait(2) done in
50 step 5 will probably hang despite the available data still present in
51 the file input buffer; meanwhile the remote peer might be expecting a
52 response based on the data it already sent. The reason for this is
53 that edge-triggered mode delivers events only when changes occur on the
54 monitored file descriptor. So, in step 5 the caller might end up wait‐
55 ing for some data that is already present inside the input buffer. In
56 the above example, an event on rfd will be generated because of the
57 write done in 2 and the event is consumed in 3. Since the read opera‐
58 tion done in 4 does not consume the whole buffer data, the call to
59 epoll_wait(2) done in step 5 might block indefinitely.
60 An application that employs the EPOLLET flag should use nonblocking
62 file descriptors to avoid having a blocking read or write starve a task
63 that is handling multiple file descriptors. The suggested way to use
64 epoll as an edge-triggered (EPOLLET) interface is as follows:
65 i with nonblocking file descriptors; and
67 ii by waiting for an event only after read(2) or write(2)
69 return EAGAIN.
70 By contrast, when used as a level-triggered interface (the default,
72 when EPOLLET is not specified), epoll is simply a faster poll(2), and
73 can be used wherever the latter is used since it shares the same seman‐
74 tics.
75 Since even with edge-triggered epoll, multiple events can be generated
77 upon receipt of multiple chunks of data, the caller has the option to
78 specify the EPOLLONESHOT flag, to tell epoll to disable the associated
79 file descriptor after the receipt of an event with epoll_wait(2). When
80 the EPOLLONESHOT flag is specified, it is the caller's responsibility
81 to rearm the file descriptor using epoll_ctl(2) with EPOLL_CTL_MOD.
82 /proc interfaces
84 The following interfaces can be used to limit the amount of kernel mem‐
85 ory consumed by epoll:
86 /proc/sys/fs/epoll/max_user_watches (since Linux 2.6.28)
88 This specifies a limit on the total number of file descriptors
89 that a user can register across all epoll instances on the sys‐
90 tem. The limit is per real user ID. Each registered file
91 descriptor costs roughly 90 bytes on a 32-bit kernel, and
92 roughly 160 bytes on a 64-bit kernel. Currently, the default
93 value for max_user_watches is 1/25 (4%) of the available low
94 memory, divided by the registration cost in bytes.
95 Example for suggested usage
97 While the usage of epoll when employed as a level-triggered interface
98 does have the same semantics as poll(2), the edge-triggered usage
99 requires more clarification to avoid stalls in the application event
100 loop. In this example, listener is a nonblocking socket on which lis‐
101 ten(2) has been called. The function do_use_fd() uses the new ready
102 file descriptor until EAGAIN is returned by either read(2) or write(2).
103 An event-driven state machine application should, after having received
104 EAGAIN, record its current state so that at the next call to
105 do_use_fd() it will continue to read(2) or write(2) from where it
106 stopped before.
107 #define MAX_EVENTS 10
109 struct epoll_event ev, events[MAX_EVENTS];
110 int listen_sock, conn_sock, nfds, epollfd;
111 /* Set up listening socket, 'listen_sock' (socket(),
113 bind(), listen()) */
114 epollfd = epoll_create(10);
116 if (epollfd == -1) {
117 perror("epoll_create");
118 exit(EXIT_FAILURE);
119 }
120 ev.events = EPOLLIN;
122 ev.data.fd = listen_sock;
123 if (epoll_ctl(epollfd, EPOLL_CTL_ADD, listen_sock, &ev) == -1) {
124 perror("epoll_ctl: listen_sock");
125 exit(EXIT_FAILURE);
126 }
127 for (;;) {
129 nfds = epoll_wait(epollfd, events, MAX_EVENTS, -1);
130 if (nfds == -1) {
131 perror("epoll_pwait");
132 exit(EXIT_FAILURE);
133 }
134 for (n = 0; n < nfds; ++n) {
136 if (events[n].data.fd == listen_sock) {
137 conn_sock = accept(listen_sock,
138 (struct sockaddr *) &local, &addrlen);
139 if (conn_sock == -1) {
140 perror("accept");
141 exit(EXIT_FAILURE);
142 }
143 setnonblocking(conn_sock);
144 ev.events = EPOLLIN | EPOLLET;
145 ev.data.fd = conn_sock;
146 if (epoll_ctl(epollfd, EPOLL_CTL_ADD, conn_sock,
147 &ev) == -1) {
148 perror("epoll_ctl: conn_sock");
149 exit(EXIT_FAILURE);
150 }
151 } else {
152 do_use_fd(events[n].data.fd);
153 }
154 }
155 }
156 When used as an edge-triggered interface, for performance reasons, it
158 is possible to add the file descriptor inside the epoll interface
159 (EPOLL_CTL_ADD) once by specifying (EPOLLIN|EPOLLOUT). This allows you
160 to avoid continuously switching between EPOLLIN and EPOLLOUT calling
161 epoll_ctl(2) with EPOLL_CTL_MOD.
162 Questions and answers
164 Q0 What is the key used to distinguish the file descriptors registered
165 in an epoll set?
166 A0 The key is the combination of the file descriptor number and the
168 open file description (also known as an "open file handle", the
169 kernel's internal representation of an open file).
170 Q1 What happens if you register the same file descriptor on an epoll
172 instance twice?
173 A1 You will probably get EEXIST. However, it is possible to add a
175 duplicate (dup(2), dup2(2), fcntl(2) F_DUPFD) descriptor to the
176 same epoll instance. This can be a useful technique for filtering
177 events, if the duplicate file descriptors are registered with dif‐
178 ferent events masks.
179 Q2 Can two epoll instances wait for the same file descriptor? If so,
181 are events reported to both epoll file descriptors?
182 A2 Yes, and events would be reported to both. However, careful pro‐
184 gramming may be needed to do this correctly.
185 Q3 Is the epoll file descriptor itself poll/epoll/selectable?
187 A3 Yes. If an epoll file descriptor has events waiting then it will
189 indicate as being readable.
190 Q4 What happens if one attempts to put an epoll file descriptor into
192 its own file descriptor set?
193 A4 The epoll_ctl(2) call will fail (EINVAL). However, you can add an
195 epoll file descriptor inside another epoll file descriptor set.
196 Q5 Can I send an epoll file descriptor over a UNIX domain socket to
198 another process?
199 A5 Yes, but it does not make sense to do this, since the receiving
201 process would not have copies of the file descriptors in the epoll
202 set.
203 Q6 Will closing a file descriptor cause it to be removed from all
205 epoll sets automatically?
206 A6 Yes, but be aware of the following point. A file descriptor is a
208 reference to an open file description (see open(2)). Whenever a
209 descriptor is duplicated via dup(2), dup2(2), fcntl(2) F_DUPFD, or
210 fork(2), a new file descriptor referring to the same open file
211 description is created. An open file description continues to
212 exist until all file descriptors referring to it have been closed.
213 A file descriptor is removed from an epoll set only after all the
214 file descriptors referring to the underlying open file description
215 have been closed (or before if the descriptor is explicitly removed
216 using epoll_ctl(2) EPOLL_CTL_DEL). This means that even after a
217 file descriptor that is part of an epoll set has been closed,
218 events may be reported for that file descriptor if other file
219 descriptors referring to the same underlying file description
220 remain open.
221 Q7 If more than one event occurs between epoll_wait(2) calls, are they
223 combined or reported separately?
224 A7 They will be combined.
226 Q8 Does an operation on a file descriptor affect the already collected
228 but not yet reported events?
229 A8 You can do two operations on an existing file descriptor. Remove
231 would be meaningless for this case. Modify will reread available
232 I/O.
233 Q9 Do I need to continuously read/write a file descriptor until EAGAIN
235 when using the EPOLLET flag (edge-triggered behavior) ?
236 A9 Receiving an event from epoll_wait(2) should suggest to you that
238 such file descriptor is ready for the requested I/O operation. You
239 must consider it ready until the next (nonblocking) read/write
240 yields EAGAIN. When and how you will use the file descriptor is
241 entirely up to you.
242 For packet/token-oriented files (e.g., datagram socket, terminal in
244 canonical mode), the only way to detect the end of the read/write
245 I/O space is to continue to read/write until EAGAIN.
246 For stream-oriented files (e.g., pipe, FIFO, stream socket), the
248 condition that the read/write I/O space is exhausted can also be
249 detected by checking the amount of data read from / written to the
250 target file descriptor. For example, if you call read(2) by asking
251 to read a certain amount of data and read(2) returns a lower number
252 of bytes, you can be sure of having exhausted the read I/O space
253 for the file descriptor. The same is true when writing using
254 write(2). (Avoid this latter technique if you cannot guarantee
255 that the monitored file descriptor always refers to a stream-ori‐
256 ented file.)
257 Possible pitfalls and ways to avoid them
259 o Starvation (edge-triggered)
260 If there is a large amount of I/O space, it is possible that by trying
262 to drain it the other files will not get processed causing starvation.
263 (This problem is not specific to epoll.)
264 The solution is to maintain a ready list and mark the file descriptor
266 as ready in its associated data structure, thereby allowing the appli‐
267 cation to remember which files need to be processed but still round
268 robin amongst all the ready files. This also supports ignoring subse‐
269 quent events you receive for file descriptors that are already ready.
270 o If using an event cache...
272 If you use an event cache or store all the file descriptors returned
274 from epoll_wait(2), then make sure to provide a way to mark its closure
275 dynamically (i.e., caused by a previous event's processing). Suppose
276 you receive 100 events from epoll_wait(2), and in event #47 a condition
277 causes event #13 to be closed. If you remove the structure and
278 close(2) the file descriptor for event #13, then your event cache might
279 still say there are events waiting for that file descriptor causing
280 confusion.
281 One solution for this is to call, during the processing of event 47,
283 epoll_ctl(EPOLL_CTL_DEL) to delete file descriptor 13 and close(2),
284 then mark its associated data structure as removed and link it to a
285 cleanup list. If you find another event for file descriptor 13 in your
286 batch processing, you will discover the file descriptor had been previ‐
287 ously removed and there will be no confusion.
288
290 The epoll API was introduced in Linux kernel 2.5.44. Support was added
291 to glibc in version 2.3.2.
292
294 The epoll API is Linux-specific. Some other systems provide similar
295 mechanisms, for example, FreeBSD has kqueue, and Solaris has /dev/poll.
296
298 epoll_create(2), epoll_create1(2), epoll_ctl(2), epoll_wait(2)
299
301 This page is part of release 3.53 of the Linux man-pages project. A
302 description of the project, and information about reporting bugs, can
303 be found at http://www.kernel.org/doc/man-pages/.
304Linux 2012-04-17 EPOLL(7)