1EPOLL(7) Linux Programmer's Manual EPOLL(7)
2
6 epoll - I/O event notification facility
7
9 #include <sys/epoll.h>
10
12 epoll is a variant of poll(2) that can be used either as Edge or Level
13 Triggered interface and scales well to large numbers of watched fds.
14 Three system calls are provided to set up and control an epoll set:
15 epoll_create(2), epoll_ctl(2), epoll_wait(2).
16 An epoll set is connected to a file descriptor created by epoll_cre‐
18 ate(2). Interest for certain file descriptors is then registered via
19 epoll_ctl(2). Finally, the actual wait is started by epoll_wait(2).
20
22 The epoll event distribution interface is able to behave both as Edge
23 Triggered ( ET ) and Level Triggered ( LT ). The difference between ET
24 and LT event distribution mechanism can be described as follows. Sup‐
25 pose that this scenario happens :
26 1 The file descriptor that represents the read side of a pipe (
28 RFD ) is added inside the epoll device.
29 2 Pipe writer writes 2Kb of data on the write side of the pipe.
31 3 A call to epoll_wait(2) is done that will return RFD as ready
33 file descriptor.
34 4 The pipe reader reads 1Kb of data from RFD.
36 5 A call to epoll_wait(2) is done.
38 If the RFD file descriptor has been added to the epoll interface using
40 the EPOLLET flag, the call to epoll_wait(2) done in step 5 will proba‐
41 bly hang because of the available data still present in the file input
42 buffers and the remote peer might be expecting a response based on the
43 data it already sent. The reason for this is that Edge Triggered event
44 distribution delivers events only when events happens on the monitored
45 file. So, in step 5 the caller might end up waiting for some data that
46 is already present inside the input buffer. In the above example, an
47 event on RFD will be generated because of the write done in 2 and the
48 event is consumed in 3. Since the read operation done in 4 does not
49 consume the whole buffer data, the call to epoll_wait(2) done in step 5
50 might lock indefinitely. The epoll interface, when used with the EPOL‐
51 LET flag ( Edge Triggered ) should use non-blocking file descriptors to
52 avoid having a blocking read or write starve the task that is handling
53 multiple file descriptors. The suggested way to use epoll as an Edge
54 Triggered (EPOLLET) interface is below, and possible pitfalls to avoid
55 follow.
56 i with non-blocking file descriptors
58 ii by going to wait for an event only after read(2) or
60 write(2) return EAGAIN
61 On the contrary, when used as a Level Triggered interface, epoll is by
63 all means a faster poll(2), and can be used wherever the latter is used
64 since it shares the same semantics. Since even with the Edge Triggered
65 epoll multiple events can be generated up on receipt of multiple chunks
66 of data, the caller has the option to specify the EPOLLONESHOT flag, to
67 tell epoll to disable the associated file descriptor after the receipt
68 of an event with epoll_wait(2). When the EPOLLONESHOT flag is speci‐
69 fied, it is caller responsibility to rearm the file descriptor using
70 epoll_ctl(2) with EPOLL_CTL_MOD.
71
73 While the usage of epoll when employed like a Level Triggered interface
74 does have the same semantics of poll(2), an Edge Triggered usage
75 requires more clarification to avoid stalls in the application event
76 loop. In this example, listener is a non-blocking socket on which lis‐
77 ten(2) has been called. The function do_use_fd() uses the new ready
78 file descriptor until EAGAIN is returned by either read(2) or write(2).
79 An event driven state machine application should, after having received
80 EAGAIN, record its current state so that at the next call to
81 do_use_fd() it will continue to read(2) or write(2) from where it
82 stopped before.
83 struct epoll_event ev, *events;
85 for(;;) {
87 nfds = epoll_wait(kdpfd, events, maxevents, -1);
88 for(n = 0; n < nfds; ++n) {
90 if(events[n].data.fd == listener) {
91 client = accept(listener, (struct sockaddr *) &local,
92 &addrlen);
93 if(client < 0){
94 perror("accept");
95 continue;
96 }
97 setnonblocking(client);
98 ev.events = EPOLLIN | EPOLLET;
99 ev.data.fd = client;
100 if (epoll_ctl(kdpfd, EPOLL_CTL_ADD, client, &ev) < 0) {
101 fprintf(stderr, "epoll set insertion error: fd=%d\n",
102 client);
103 return -1;
104 }
105 }
106 else
107 do_use_fd(events[n].data.fd);
108 }
109 }
110 When used as an Edge triggered interface, for performance reasons, it
112 is possible to add the file descriptor inside the epoll interface (
113 EPOLL_CTL_ADD ) once by specifying ( EPOLLIN|EPOLLOUT ). This allows
114 you to avoid continuously switching between EPOLLIN and EPOLLOUT call‐
115 ing epoll_ctl(2) with EPOLL_CTL_MOD.
116
119 Q1 What happens if you add the same fd to an epoll_set twice?
120 A1 You will probably get EEXIST. However, it is possible that two
122 threads may add the same fd twice. This is a harmless condition.
123 Q2 Can two epoll sets wait for the same fd? If so, are events
125 reported to both epoll sets fds?
126 A2 Yes. However, it is not recommended. Yes it would be reported to
128 both.
129 Q3 Is the epoll fd itself poll/epoll/selectable?
131 A3 Yes.
133 Q4 What happens if the epoll fd is put into its own fd set?
135 A4 It will fail. However, you can add an epoll fd inside another
137 epoll fd set.
138 Q5 Can I send the epoll fd over a unix-socket to another process?
140 A5 No.
142 Q6 Will the close of an fd cause it to be removed from all epoll
144 sets automatically?
145 A6 Yes.
147 Q7 If more than one event comes in between epoll_wait(2) calls, are
149 they combined or reported separately?
150 A7 They will be combined.
152 Q8 Does an operation on an fd affect the already collected but not
154 yet reported events?
155 A8 You can do two operations on an existing fd. Remove would be
157 meaningless for this case. Modify will re-read available I/O.
158 Q9 Do I need to continuously read/write an fd until EAGAIN when
160 using the EPOLLET flag ( Edge Triggered behaviour ) ?
161 A9 No you don't. Receiving an event from epoll_wait(2) should sug‐
163 gest to you that such file descriptor is ready for the requested
164 I/O operation. You have simply to consider it ready until you
165 will receive the next EAGAIN. When and how you will use such
166 file descriptor is entirely up to you. Also, the condition that
167 the read/write I/O space is exhausted can be detected by check‐
168 ing the amount of data read/write from/to the target file
169 descriptor. For example, if you call read(2) by asking to read a
170 certain amount of data and read(2) returns a lower number of
171 bytes, you can be sure to have exhausted the read I/O space for
172 such file descriptor. Same is valid when writing using the
173 write(2) function.
174
176 o Starvation ( Edge Triggered )
177 If there is a large amount of I/O space, it is possible that by trying
179 to drain it the other files will not get processed causing starvation.
180 This is not specific to epoll.
181 The solution is to maintain a ready list and mark the file descriptor
183 as ready in its associated data structure, thereby allowing the appli‐
184 cation to remember which files need to be processed but still round
185 robin amongst all the ready files. This also supports ignoring subse‐
186 quent events you receive for fd's that are already ready.
187 o If using an event cache...
189 If you use an event cache or store all the fd's returned from
191 epoll_wait(2), then make sure to provide a way to mark its closure
192 dynamically (ie- caused by a previous event's processing). Suppose you
193 receive 100 events from epoll_wait(2), and in event #47 a condition
194 causes event #13 to be closed. If you remove the structure and close()
195 the fd for event #13, then your event cache might still say there are
196 events waiting for that fd causing confusion.
197 One solution for this is to call, during the processing of event 47,
199 epoll_ctl(EPOLL_CTL_DEL) to delete fd 13 and close(), then mark its
200 associated data structure as removed and link it to a cleanup list. If
201 you find another event for fd 13 in your batch processing, you will
202 discover the fd had been previously removed and there will be no confu‐
203 sion.
204
206 The epoll API is Linux specific. Some other systems provide similar
207 mechanisms, e.g., FreeBSD has kqueue, and Solaris has /dev/poll.
208
210 epoll(7) is a new API introduced in Linux kernel 2.5.44. Its interface
211 should be finalized in Linux kernel 2.5.66.
212
214 epoll_create(2), epoll_ctl(2), epoll_wait(2)
215Linux 2002-10-23 EPOLL(7)