1 EPOLL(4) Linux Programmer's Manual EPOLL(4) 2 3 4 5 NAME 6 epoll - I/O event notification facility 7 8 SYNOPSIS 9 #include <sys/epoll.h> 10 11 DESCRIPTION 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 17 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 21 22 NOTES 23 The epoll event distribution interface is able to behave both as Edge 24 Triggered ( ET ) and Level Triggered ( LT ). The difference between ET 25 and LT event distribution mechanism can be described as follows. Sup- 26 pose that this scenario happens : 27 28 1 The file descriptor that represent the read side of a pipe ( RFD 29 ) is added inside the epoll device. 30 31 2 Pipe writer writes 2Kb of data on the write side of the pipe. 32 33 3 A call to epoll_wait(2) is done that will return RFD as ready 34 file descriptor. 35 36 4 The pipe reader reads 1Kb of data from RFD. 37 38 5 A call to epoll_wait(2) is done. 39 40 41 If the RFD file descriptor has been added to the epoll interface using 42 the EPOLLET flag, the call to epoll_wait(2) done in step 5 will proba- 43 bly hang because of the available data still present in the file input 44 buffers and the remote peer might be expecting a response based on the 45 data it already sent. The reason for this is that Edge Triggered event 46 distribution delivers events only when events happens on the monitored 47 file. So, in step 5 the caller might end up waiting for some data that 48 is already present inside the input buffer. In the above example, an 49 event on RFD will be generated because of the write done in 2 , and the 50 event is consumed in 3. Since the read operation done in 4 does not 51 consume the whole buffer data, the call to epoll_wait(2) done in step 5 52 might lock indefinitely. The epoll interface, when used with the EPOL- 53 LET flag ( Edge Triggered ) should use non-blocking file descriptors to 54 avoid having a blocking read or write starve the task that is handling 55 multiple file descriptors. The suggested way to use epoll as an Edge 56 Triggered ( EPOLLET ) interface is below, and possible pitfalls to 57 avoid follow. 58 59 i with non-blocking file descriptors 60 61 ii by going to wait for an event only after read(2) or write(2) 62 return EAGAIN 63 64 On the contrary, when used as a Level Triggered interface, epoll is by 65 all means a faster poll(2), and can be used wherever the latter is used 66 since it shares the same semantics. Since even with the Edge Triggered 67 epoll multiple events can be generated up on receival of multiple 68 chunks of data, the caller has the option to specify the EPOLLONESHOT 69 flag, to tell epoll to disable the associated file descriptor after the 70 receival of an event with epoll_wait(2). When the EPOLLONESHOT flag is 71 specified, it is caller responsibility to rearm the file descriptor 72 using epoll_ctl(2) with EPOLL_CTL_MOD. 73 74 75 EXAMPLE FOR SUGGESTED USAGE 76 While the usage of epoll when employed like a Level Triggered interface 77 does have the same semantics of poll(2), an Edge Triggered usage 78 requires more clarifiction to avoid stalls in the application event 79 loop. In this example, listener is a non-blocking socket on which lis- 80 ten(2) has been called. The function do_use_fd() uses the new ready 81 file descriptor until EAGAIN is returned by either read(2) or write(2). 82 An event driven state machine application should, after having received 83 EAGAIN, record its current state so that at the next call to 84 do_use_fd() it will continue to read(2) or write(2) from where it 85 stopped before. 86 87 struct epoll_event ev, *events; 88 89 for(;;) { 90 nfds = epoll_wait(kdpfd, events, maxevents, -1); 91 92 for(n = 0; n < nfds; ++n) { 93 if(events[n].data.fd == listener) { 94 client = accept(listener, (struct sockaddr *) &local, 95 &addrlen); 96 if(client < 0){ 97 perror("accept"); 98 continue; 99 } 100 setnonblocking(client); 101 ev.events = EPOLLIN | EPOLLET; 102 ev.data.fd = client; 103 if (epoll_ctl(kdpfd, EPOLL_CTL_ADD, client, &ev) < 0) { 104 fprintf(stderr, "epoll set insertion error: fd=%d0, 105 client); 106 return -1; 107 } 108 } 109 else 110 do_use_fd(events[n].data.fd); 111 } 112 } 113 114 When used as an Edge triggered interface, for performance reasons, it 115 is possible to add the file descriptor inside the epoll interface ( 116 EPOLL_CTL_ADD ) once by specifying ( EPOLLIN|EPOLLOUT ). This allows 117 you to avoid continuously switching between EPOLLIN and EPOLLOUT call- 118 ing epoll_ctl(2) with EPOLL_CTL_MOD. 119 120 121 QUESTIONS AND ANSWERS (from linux-kernel) 122 Q1 What happens if you add the same fd to an epoll_set twice? 123 124 A1 You will probably get EEXIST. However, it is possible that two 125 threads may add the same fd twice. This is a harmless condition. 126 127 Q2 Can two epoll sets wait for the same fd? If so, are events 128 reported to both epoll sets fds? 129 130 A2 Yes. However, it is not recommended. Yes it would be reported to 131 both. 132 133 Q3 Is the epoll fd itself poll/epoll/selectable? 134 135 A3 Yes. 136 137 Q4 What happens if the epoll fd is put into its own fd set? 138 139 A4 It will fail. However, you can add an epoll fd inside another 140 epoll fd set. 141 142 Q5 Can I send the epoll fd over a unix-socket to another process? 143 144 A5 No. 145 146 Q6 Will the close of an fd cause it to be removed from all epoll 147 sets automatically? 148 149 A6 Yes. 150 151 Q7 If more than one event comes in between epoll_wait(2) calls, are 152 they combined or reported separately? 153 154 A7 They will be combined. 155 156 Q8 Does an operation on an fd affect the already collected but not 157 yet reported events? 158 159 A8 You can do two operations on an existing fd. Remove would be 160 meaningless for this case. Modify will re-read available I/O. 161 162 Q9 Do I need to continuously read/write an fd until EAGAIN when 163 using the EPOLLET flag ( Edge Triggered behaviour ) ? 164 165 A9 No you don't. Receiving an event from epoll_wait(2) should sug- 166 gest to you that such file descriptor is ready for the requested 167 I/O operation. You have simply to consider it ready until you 168 will receive the next EAGAIN. When and how you will use such 169 file descriptor is entirely up to you. Also, the condition that 170 the read/write I/O space is exhausted can be detected by check- 171 ing the amount of data read/write from/to the target file 172 descriptor. For example, if you call read(2) by asking to read a 173 certain amount of data and read(2) returns a lower number of 174 bytes, you can be sure to have exhausted the read I/O space for 175 such file descriptor. Same is valid when writing using the 176 write(2) function. 177 178 179 POSSIBLE PITFALLS AND WAYS TO AVOID THEM 180 o Starvation ( Edge Triggered ) 181 182 If there is a large amount of I/O space, it is possible that by trying 183 to drain it the other files will not get processed causing starvation. 184 This is not specific to epoll. 185 186 187 The solution is to maintain a ready list and mark the file descriptor 188 as ready in its associated data structure, thereby allowing the appli- 189 cation to remember which files need to be processed but still round 190 robin amongst all the ready files. This also supports ignoring subse- 191 quent events you receive for fd's that are already ready. 192 193 194 195 o If using an event cache... 196 197 If you use an event cache or store all the fd's returned from 198 epoll_wait(2), then make sure to provide a way to mark its closure 199 dynamically (ie- caused by a previous event's processing). Suppose you 200 receive 100 events from epoll_wait(2), and in eventi #47 a condition 201 causes event #13 to be closed. If you remove the structure and close() 202 the fd for event #13, then your event cache might still say there are 203 events waiting for that fd causing confusion. 204 205 206 One solution for this is to call, during the processing of event 47, 207 epoll_ctl(EPOLL_CTL_DEL) to delete fd 13 and close(), then mark its 208 associated data structure as removed and link it to a cleanup list. If 209 you find another event for fd 13 in your batch processing, you will 210 discover the fd had been previously removed and there will be no confu- 211 sion. 212 213 214 215 CONFORMING TO 216 epoll(4) is a new API introduced in Linux kernel 2.5.44. Its interface 217 should be finalized in Linux kernel 2.5.66. 218 219 SEE ALSO 220 epoll_create(2) epoll_ctl(2) epoll_wait(2) 221 222 223 224 225 Linux 23 October 2002 EPOLL(4) 226
