{ "mode": "man", "parameter": "epoll", "section": "7", "url": "https://www.chedong.com/phpMan.php/man/epoll/7/json", "generated": "2026-11-08T04:12:27Z", "synopsis": "", "sections": { "NAME": { "content": "epoll - I/O event notification facility\n", "subsections": [] }, "SYNOPSIS": { "content": "", "subsections": [ { "name": "#include ", "content": "" } ] }, "DESCRIPTION": { "content": "The epoll API performs a similar task to poll(2): monitoring multiple file descriptors to see\nif I/O is possible on any of them. The epoll API can be used either as an edge-triggered or\na level-triggered interface and scales well to large numbers of watched file descriptors.\n\nThe central concept of the epoll API is the epoll instance, an in-kernel data structure\nwhich, from a user-space perspective, can be considered as a container for two lists:\n\n• The interest list (sometimes also called the epoll set): the set of file descriptors that\nthe process has registered an interest in monitoring.\n\n• The ready list: the set of file descriptors that are \"ready\" for I/O. The ready list is a\nsubset of (or, more precisely, a set of references to) the file descriptors in the inter‐\nest list. The ready list is dynamically populated by the kernel as a result of I/O activ‐\nity on those file descriptors.\n\nThe following system calls are provided to create and manage an epoll instance:\n\n• epollcreate(2) creates a new epoll instance and returns a file descriptor referring to\nthat instance. (The more recent epollcreate1(2) extends the functionality of epollcre‐\nate(2).)\n\n• Interest in particular file descriptors is then registered via epollctl(2), which adds\nitems to the interest list of the epoll instance.\n\n• epollwait(2) waits for I/O events, blocking the calling thread if no events are currently\navailable. (This system call can be thought of as fetching items from the ready list of\nthe epoll instance.)\n", "subsections": [ { "name": "Level-triggered and edge-triggered", "content": "The epoll event distribution interface is able to behave both as edge-triggered (ET) and as\nlevel-triggered (LT). The difference between the two mechanisms can be described as follows.\nSuppose that this scenario happens:\n\n(1) The file descriptor that represents the read side of a pipe (rfd) is registered on the\nepoll instance.\n\n(2) A pipe writer writes 2 kB of data on the write side of the pipe.\n\n(3) A call to epollwait(2) is done that will return rfd as a ready file descriptor.\n\n(4) The pipe reader reads 1 kB of data from rfd.\n\n(5) A call to epollwait(2) is done.\n\nIf the rfd file descriptor has been added to the epoll interface using the EPOLLET (edge-\ntriggered) flag, the call to epollwait(2) done in step 5 will probably hang despite the\navailable data still present in the file input buffer; meanwhile the remote peer might be ex‐\npecting a response based on the data it already sent. The reason for this is that edge-trig‐\ngered mode delivers events only when changes occur on the monitored file descriptor. So, in\nstep 5 the caller might end up waiting for some data that is already present inside the input\nbuffer. In the above example, an event on rfd will be generated because of the write done in\n2 and the event is consumed in 3. Since the read operation done in 4 does not consume the\nwhole buffer data, the call to epollwait(2) done in step 5 might block indefinitely.\n\nAn application that employs the EPOLLET flag should use nonblocking file descriptors to avoid\nhaving a blocking read or write starve a task that is handling multiple file descriptors.\nThe suggested way to use epoll as an edge-triggered (EPOLLET) interface is as follows:\n\n(1) with nonblocking file descriptors; and\n\n(2) by waiting for an event only after read(2) or write(2) return EAGAIN.\n\nBy contrast, when used as a level-triggered interface (the default, when EPOLLET is not spec‐\nified), epoll is simply a faster poll(2), and can be used wherever the latter is used since\nit shares the same semantics.\n\nSince even with edge-triggered epoll, multiple events can be generated upon receipt of multi‐\nple chunks of data, the caller has the option to specify the EPOLLONESHOT flag, to tell epoll\nto disable the associated file descriptor after the receipt of an event with epollwait(2).\nWhen the EPOLLONESHOT flag is specified, it is the caller's responsibility to rearm the file\ndescriptor using epollctl(2) with EPOLLCTLMOD.\n\nIf multiple threads (or processes, if child processes have inherited the epoll file descrip‐\ntor across fork(2)) are blocked in epollwait(2) waiting on the same epoll file descriptor\nand a file descriptor in the interest list that is marked for edge-triggered (EPOLLET) noti‐\nfication becomes ready, just one of the threads (or processes) is awoken from epollwait(2).\nThis provides a useful optimization for avoiding \"thundering herd\" wake-ups in some scenar‐\nios.\n" }, { "name": "Interaction with autosleep", "content": "If the system is in autosleep mode via /sys/power/autosleep and an event happens which wakes\nthe device from sleep, the device driver will keep the device awake only until that event is\nqueued. To keep the device awake until the event has been processed, it is necessary to use\nthe epollctl(2) EPOLLWAKEUP flag.\n\nWhen the EPOLLWAKEUP flag is set in the events field for a struct epollevent, the system\nwill be kept awake from the moment the event is queued, through the epollwait(2) call which\nreturns the event until the subsequent epollwait(2) call. If the event should keep the sys‐\ntem awake beyond that time, then a separate wakelock should be taken before the second\nepollwait(2) call.\n" }, { "name": "/proc interfaces", "content": "The following interfaces can be used to limit the amount of kernel memory consumed by epoll:\n\n/proc/sys/fs/epoll/maxuserwatches (since Linux 2.6.28)\nThis specifies a limit on the total number of file descriptors that a user can regis‐\nter across all epoll instances on the system. The limit is per real user ID. Each\nregistered file descriptor costs roughly 90 bytes on a 32-bit kernel, and roughly 160\nbytes on a 64-bit kernel. Currently, the default value for maxuserwatches is 1/25\n(4%) of the available low memory, divided by the registration cost in bytes.\n" }, { "name": "Example for suggested usage", "content": "While the usage of epoll when employed as a level-triggered interface does have the same se‐\nmantics as poll(2), the edge-triggered usage requires more clarification to avoid stalls in\nthe application event loop. In this example, listener is a nonblocking socket on which lis‐\nten(2) has been called. The function dousefd() uses the new ready file descriptor until\nEAGAIN is returned by either read(2) or write(2). An event-driven state machine application\nshould, after having received EAGAIN, record its current state so that at the next call to\ndousefd() it will continue to read(2) or write(2) from where it stopped before.\n\n#define MAXEVENTS 10\nstruct epollevent ev, events[MAXEVENTS];\nint listensock, connsock, nfds, epollfd;\n\n/* Code to set up listening socket, 'listensock',\n(socket(), bind(), listen()) omitted. */\n\nepollfd = epollcreate1(0);\nif (epollfd == -1) {\nperror(\"epollcreate1\");\nexit(EXITFAILURE);\n}\n\nev.events = EPOLLIN;\nev.data.fd = listensock;\nif (epollctl(epollfd, EPOLLCTLADD, listensock, &ev) == -1) {\nperror(\"epollctl: listensock\");\nexit(EXITFAILURE);\n}\n\nfor (;;) {\nnfds = epollwait(epollfd, events, MAXEVENTS, -1);\nif (nfds == -1) {\nperror(\"epollwait\");\nexit(EXITFAILURE);\n}\n\nfor (n = 0; n < nfds; ++n) {\nif (events[n].data.fd == listensock) {\nconnsock = accept(listensock,\n(struct sockaddr *) &addr, &addrlen);\nif (connsock == -1) {\nperror(\"accept\");\nexit(EXITFAILURE);\n}\nsetnonblocking(connsock);\nev.events = EPOLLIN | EPOLLET;\nev.data.fd = connsock;\nif (epollctl(epollfd, EPOLLCTLADD, connsock,\n&ev) == -1) {\nperror(\"epollctl: connsock\");\nexit(EXITFAILURE);\n}\n} else {\ndousefd(events[n].data.fd);\n}\n}\n}\n\nWhen used as an edge-triggered interface, for performance reasons, it is possible to add the\nfile descriptor inside the epoll interface (EPOLLCTLADD) once by specifying (EPOLLIN|EPOLL‐\nOUT). This allows you to avoid continuously switching between EPOLLIN and EPOLLOUT calling\nepollctl(2) with EPOLLCTLMOD.\n" }, { "name": "Questions and answers", "content": "• What is the key used to distinguish the file descriptors registered in an interest list?\n\nThe key is the combination of the file descriptor number and the open file description\n(also known as an \"open file handle\", the kernel's internal representation of an open\nfile).\n\n• What happens if you register the same file descriptor on an epoll instance twice?\n\nYou will probably get EEXIST. However, it is possible to add a duplicate (dup(2),\ndup2(2), fcntl(2) FDUPFD) file descriptor to the same epoll instance. This can be a use‐\nful technique for filtering events, if the duplicate file descriptors are registered with\ndifferent events masks.\n\n• Can two epoll instances wait for the same file descriptor? If so, are events reported to\nboth epoll file descriptors?\n\nYes, and events would be reported to both. However, careful programming may be needed to\ndo this correctly.\n\n• Is the epoll file descriptor itself poll/epoll/selectable?\n\nYes. If an epoll file descriptor has events waiting, then it will indicate as being read‐\nable.\n\n• What happens if one attempts to put an epoll file descriptor into its own file descriptor\nset?\n\nThe epollctl(2) call fails (EINVAL). However, you can add an epoll file descriptor in‐\nside another epoll file descriptor set.\n\n• Can I send an epoll file descriptor over a UNIX domain socket to another process?\n\nYes, but it does not make sense to do this, since the receiving process would not have\ncopies of the file descriptors in the interest list.\n\n• Will closing a file descriptor cause it to be removed from all epoll interest lists?\n\nYes, but be aware of the following point. A file descriptor is a reference to an open\nfile description (see open(2)). Whenever a file descriptor is duplicated via dup(2),\ndup2(2), fcntl(2) FDUPFD, or fork(2), a new file descriptor referring to the same open\nfile description is created. An open file description continues to exist until all file\ndescriptors referring to it have been closed.\n\nA file descriptor is removed from an interest list only after all the file descriptors re‐\nferring to the underlying open file description have been closed. This means that even\nafter a file descriptor that is part of an interest list has been closed, events may be\nreported for that file descriptor if other file descriptors referring to the same underly‐\ning file description remain open. To prevent this happening, the file descriptor must be\nexplicitly removed from the interest list (using epollctl(2) EPOLLCTLDEL) before it is\nduplicated. Alternatively, the application must ensure that all file descriptors are\nclosed (which may be difficult if file descriptors were duplicated behind the scenes by\nlibrary functions that used dup(2) or fork(2)).\n\n• If more than one event occurs between epollwait(2) calls, are they combined or reported\nseparately?\n\nThey will be combined.\n\n• Does an operation on a file descriptor affect the already collected but not yet reported\nevents?\n\nYou can do two operations on an existing file descriptor. Remove would be meaningless for\nthis case. Modify will reread available I/O.\n\n• Do I need to continuously read/write a file descriptor until EAGAIN when using the EPOLLET\nflag (edge-triggered behavior)?\n\nReceiving an event from epollwait(2) should suggest to you that such file descriptor is\nready for the requested I/O operation. You must consider it ready until the next (non‐\nblocking) read/write yields EAGAIN. When and how you will use the file descriptor is en‐\ntirely up to you.\n\nFor packet/token-oriented files (e.g., datagram socket, terminal in canonical mode), the\nonly way to detect the end of the read/write I/O space is to continue to read/write until\nEAGAIN.\n\nFor stream-oriented files (e.g., pipe, FIFO, stream socket), the condition that the\nread/write I/O space is exhausted can also be detected by checking the amount of data read\nfrom / written to the target file descriptor. For example, if you call read(2) by asking\nto read a certain amount of data and read(2) returns a lower number of bytes, you can be\nsure of having exhausted the read I/O space for the file descriptor. The same is true\nwhen writing using write(2). (Avoid this latter technique if you cannot guarantee that\nthe monitored file descriptor always refers to a stream-oriented file.)\n" }, { "name": "Possible pitfalls and ways to avoid them", "content": "• Starvation (edge-triggered)\n\nIf there is a large amount of I/O space, it is possible that by trying to drain it the\nother files will not get processed causing starvation. (This problem is not specific to\nepoll.)\n\nThe solution is to maintain a ready list and mark the file descriptor as ready in its as‐\nsociated data structure, thereby allowing the application to remember which files need to\nbe processed but still round robin amongst all the ready files. This also supports ignor‐\ning subsequent events you receive for file descriptors that are already ready.\n\n• If using an event cache...\n\nIf you use an event cache or store all the file descriptors returned from epollwait(2),\nthen make sure to provide a way to mark its closure dynamically (i.e., caused by a previ‐\nous event's processing). Suppose you receive 100 events from epollwait(2), and in event\n#47 a condition causes event #13 to be closed. If you remove the structure and close(2)\nthe file descriptor for event #13, then your event cache might still say there are events\nwaiting for that file descriptor causing confusion.\n\nOne solution for this is to call, during the processing of event 47,\nepollctl(EPOLLCTLDEL) to delete file descriptor 13 and close(2), then mark its associ‐\nated data structure as removed and link it to a cleanup list. If you find another event\nfor file descriptor 13 in your batch processing, you will discover the file descriptor had\nbeen previously removed and there will be no confusion.\n" } ] }, "VERSIONS": { "content": "Some other systems provide similar mechanisms; for example, FreeBSD has kqueue, and Solaris\nhas /dev/poll.\n", "subsections": [] }, "STANDARDS": { "content": "Linux.\n", "subsections": [] }, "HISTORY": { "content": "Linux 2.5.44. glibc 2.3.2.\n", "subsections": [] }, "NOTES": { "content": "The set of file descriptors that is being monitored via an epoll file descriptor can be\nviewed via the entry for the epoll file descriptor in the process's /proc/pid/fdinfo direc‐\ntory. See proc(5) for further details.\n\nThe kcmp(2) KCMPEPOLLTFD operation can be used to test whether a file descriptor is present\nin an epoll instance.\n", "subsections": [] }, "SEE ALSO": { "content": "epollcreate(2), epollcreate1(2), epollctl(2), epollwait(2), poll(2), select(2)\n\nLinux man-pages 6.7 2023-10-31 epoll(7)", "subsections": [] } }, "summary": "epoll - I/O event notification facility", "flags": [], "examples": [], "see_also": [ { "name": "epollcreate", "section": "2", "url": "https://www.chedong.com/phpMan.php/man/epollcreate/2/json" }, { "name": "epollcreate1", "section": "2", "url": "https://www.chedong.com/phpMan.php/man/epollcreate1/2/json" }, { "name": "epollctl", "section": "2", "url": "https://www.chedong.com/phpMan.php/man/epollctl/2/json" }, { "name": "epollwait", "section": "2", "url": "https://www.chedong.com/phpMan.php/man/epollwait/2/json" }, { "name": "poll", "section": "2", "url": "https://www.chedong.com/phpMan.php/man/poll/2/json" }, { "name": "select", "section": "2", "url": "https://www.chedong.com/phpMan.php/man/select/2/json" } ] }