/* File descriptors managment and switching */ #ifdef HAVE_CONFIG_H #include "config.h" #endif #include #include #include /* FreeBSD FD_ZERO() macro calls bzero() */ #ifdef __GNU__ /* For GNU Hurd bug workaround in set_handlers() */ #include /* OS/2 needs this after sys/types.h */ #endif #include #ifdef HAVE_SYS_WAIT_H #include #endif #ifdef HAVE_UNISTD_H #include #endif /* This must be here, thanks to BSD. */ #ifdef HAVE_INTTYPES_H #include /* OMG */ #endif #if defined(HAVE_POLL_H) && defined(HAVE_POLL) && !defined(INTERIX) && !defined(__HOS_AIX__) #define USE_POLL #include #endif #if defined(HAVE_LIBEV) && !defined(OPENVMS) && !defined(DOS) #ifdef HAVE_LIBEV_EVENT_H #include #elif defined(HAVE_EVENT_H) #include #endif #define USE_LIBEVENT #endif #if (defined(HAVE_EVENT_H) || defined(HAVE_EV_EVENT_H) || defined(HAVE_LIBEV_EVENT_H)) && defined(HAVE_LIBEVENT) && !defined(OPENVMS) && !defined(DOS) #if defined(HAVE_EVENT_H) #include #elif defined(HAVE_EV_EVENT_H) #include #endif #define USE_LIBEVENT #endif #ifdef HAVE_SYS_SELECT_H #include #endif #define EINTRLOOPX(ret_, call_, x_) \ do { \ (ret_) = (call_); \ } while ((ret_) == (x_) && errno == EINTR) #define EINTRLOOP(ret_, call_) EINTRLOOPX(ret_, call_, -1) #include "elinks.h" #include "intl/gettext/libintl.h" #include "main/main.h" #include "main/select.h" #include "main/timer.h" #include "osdep/signals.h" #include "terminal/terminal.h" #include "util/error.h" #include "util/memory.h" #include "util/time.h" #ifndef FD_SETSIZE #define FD_SETSIZE 1024 #endif /* #define DEBUG_CALLS */ static int n_threads = 0; struct thread { select_handler_T read_func; select_handler_T write_func; select_handler_T error_func; void *data; #ifdef USE_LIBEVENT struct event *read_event; struct event *write_event; #endif }; #ifdef CONFIG_OS_WIN32 /* CreatePipe produces big numbers for handles */ #undef FD_SETSIZE #define FD_SETSIZE 4096 #endif static struct thread *threads = NULL; static fd_set w_read; static fd_set w_write; static fd_set w_error; static fd_set x_read; static fd_set x_write; static fd_set x_error; static int w_max; int get_file_handles_count(void) { int i = 0, j; for (j = 0; j < FD_SETSIZE; j++) if (threads[j].read_func || threads[j].write_func || threads[j].error_func) i++; return i; } struct bottom_half { LIST_HEAD(struct bottom_half); select_handler_T fn; void *data; }; static INIT_LIST_OF(struct bottom_half, bottom_halves); int register_bottom_half_do(select_handler_T fn, void *data) { struct bottom_half *bh; foreach (bh, bottom_halves) if (bh->fn == fn && bh->data == data) return 0; bh = mem_alloc(sizeof(*bh)); if (!bh) return -1; bh->fn = fn; bh->data = data; add_to_list(bottom_halves, bh); return 0; } void check_bottom_halves(void) { while (!list_empty(bottom_halves)) { struct bottom_half *bh = bottom_halves.prev; select_handler_T fn = bh->fn; void *data = bh->data; del_from_list(bh); mem_free(bh); fn(data); } } static void restrict_fds(void) { #if defined(RLIMIT_OFILE) && !defined(RLIMIT_NOFILE) #define RLIMIT_NOFILE RLIMIT_OFILE #endif #if defined(HAVE_GETRLIMIT) && defined(HAVE_SETRLIMIT) && defined(RLIMIT_NOFILE) struct rlimit limit; int rs; EINTRLOOP(rs, getrlimit(RLIMIT_NOFILE, &limit)); if (rs) goto skip_limit; if (limit.rlim_cur > FD_SETSIZE) { limit.rlim_cur = FD_SETSIZE; EINTRLOOP(rs, setrlimit(RLIMIT_NOFILE, &limit)); } skip_limit:; #endif } #ifdef USE_LIBEVENT int event_enabled = 0; #ifndef HAVE_EVENT_GET_STRUCT_EVENT_SIZE #define sizeof_struct_event sizeof(struct event) #else #define sizeof_struct_event (event_get_struct_event_size()) #endif static inline struct event *timer_event(struct timer *tm) { return (struct event *)((unsigned char *)tm - sizeof_struct_event); } #ifdef HAVE_EVENT_BASE_SET struct event_base *event_base; #endif static void event_callback(int h, short ev, void *data) { #ifndef EV_PERSIST if (event_add((struct event *)data, NULL) == -1) elinks_internal("ERROR: event_add failed: %s", strerror(errno)); #endif if (!(ev & EV_READ) == !(ev & EV_WRITE)) elinks_internal("event_callback: invalid flags %d on handle %d", (int)ev, h); if (ev & EV_READ) { #if defined(HAVE_LIBEV) /* Old versions of libev badly interact with fork and fire * events spuriously. */ if (ev_version_major() < 4 && !can_read(h)) return; #endif threads[h].read_func(threads[h].data); } else if (ev & EV_WRITE) { #if defined(HAVE_LIBEV) /* Old versions of libev badly interact with fork and fire * events spuriously. */ if (ev_version_major() < 4 && !can_write(h)) return; #endif threads[h].write_func(threads[h].data); } check_bottom_halves(); } static void set_event_for_action(int h, void (*func)(void *), struct event **evptr, short evtype) { if (func) { if (!*evptr) { #ifdef EV_PERSIST evtype |= EV_PERSIST; #endif *evptr = mem_alloc(sizeof_struct_event); event_set(*evptr, h, evtype, event_callback, *evptr); #ifdef HAVE_EVENT_BASE_SET if (event_base_set(event_base, *evptr) == -1) elinks_internal("ERROR: event_base_set failed: %s, handle %d", strerror(errno), h); #endif } if (event_add(*evptr, NULL) == -1) elinks_internal("ERROR: event_add failed: %s, handle %d", strerror(errno), h); } else { if (*evptr) { if (event_del(*evptr) == -1) elinks_internal("ERROR: event_del failed: %s, handle %d", strerror(errno), h); } } } static void set_events_for_handle(int h) { set_event_for_action(h, threads[h].read_func, &threads[h].read_event, EV_READ); set_event_for_action(h, threads[h].write_func, &threads[h].write_event, EV_WRITE); } static void enable_libevent(void) { int i; if (get_cmd_opt_bool("no-libevent")) return; #if !defined(NO_FORK_ON_EXIT) && defined(HAVE_KQUEUE) && !defined(HAVE_EVENT_REINIT) /* kqueue doesn't work after fork */ if (!F) return; #endif #if defined(HAVE_EVENT_CONFIG_SET_FLAG) { struct event_config *cfg; cfg = event_config_new(); if (!cfg) return; if (event_config_set_flag(cfg, EVENT_BASE_FLAG_NOLOCK) == -1) { event_config_free(cfg); return; } event_base = event_base_new_with_config(cfg); event_config_free(cfg); if (!event_base) return; } #elif defined(HAVE_EVENT_BASE_NEW) event_base = event_base_new(); if (!event_base) return; #elif defined(HAVE_EVENT_BASE_SET) event_base = event_init(); if (!event_base) return; #else event_init(); #endif event_enabled = 1; for (i = 0; i < w_max; i++) set_events_for_handle(i); /* foreach(tm, timers) set_event_for_timer(tm); */ set_events_for_timer(); } static void terminate_libevent(void) { int i; if (event_enabled) { for (i = 0; i < n_threads; i++) { set_event_for_action(i, NULL, &threads[i].read_event, EV_READ); if (threads[i].read_event) mem_free(threads[i].read_event); set_event_for_action(i, NULL, &threads[i].write_event, EV_WRITE); if (threads[i].write_event) mem_free(threads[i].write_event); } #ifdef HAVE_EVENT_BASE_FREE event_base_free(event_base); #endif event_enabled = 0; } } static void do_event_loop(int flags) { int e; #ifdef HAVE_EVENT_BASE_SET e = event_base_loop(event_base, flags); #else e = event_loop(flags); #endif if (e == -1) elinks_internal("ERROR: event_base_loop failed: %s", strerror(errno)); } #endif select_handler_T get_handler(int fd, enum select_handler_type tp) { #ifndef CONFIG_OS_WIN32 assertm(fd >= 0 && fd < FD_SETSIZE, "get_handler: handle %d >= FD_SETSIZE %d", fd, FD_SETSIZE); if_assert_failed return NULL; #endif switch (tp) { case SELECT_HANDLER_READ: return threads[fd].read_func; case SELECT_HANDLER_WRITE: return threads[fd].write_func; case SELECT_HANDLER_ERROR: return threads[fd].error_func; case SELECT_HANDLER_DATA: return threads[fd].data; } INTERNAL("get_handler: bad type %d", tp); return NULL; } void set_handlers(int fd, select_handler_T read_func, select_handler_T write_func, select_handler_T error_func, void *data) { #ifndef CONFIG_OS_WIN32 assertm(fd >= 0 && fd < FD_SETSIZE, "set_handlers: handle %d >= FD_SETSIZE %d", fd, FD_SETSIZE); if_assert_failed return; #endif #ifdef __GNU__ /* GNU Hurd pflocal bug : * If ELinks does a select() where the initial exceptfds set * includes a pipe that is not listed in the other fd_sets, * then select() always reports an exception there. That * makes Elinks think the pipe has failed and close it. * To work around this bug, do not monitor exceptions for * pipes on the Hurd. */ if (error_func) { struct stat st; if (fstat(fd, &st) == 0 && S_ISFIFO(st.st_mode)) error_func = NULL; } #endif /* __GNU__ */ #if defined(USE_POLL) && defined(USE_LIBEVENT) if (!event_enabled) #endif if (fd >= (int)FD_SETSIZE) { elinks_internal("too big handle %d", fd); return; } if (fd >= n_threads) { threads = mem_realloc(threads, (fd + 1) * sizeof(struct thread)); memset(threads + n_threads, 0, (fd + 1 - n_threads) * sizeof(struct thread)); n_threads = fd + 1; } threads[fd].read_func = read_func; threads[fd].write_func = write_func; threads[fd].error_func = error_func; threads[fd].data = data; if (read_func || write_func || error_func) { if (fd >= w_max) w_max = fd + 1; } else if (fd == w_max - 1) { int i; for (i = fd - 1; i >= 0; i--) if (FD_ISSET(i, &w_read) || FD_ISSET(i, &w_write) || FD_ISSET(i, &w_error)) break; w_max = i + 1; } #ifdef USE_LIBEVENT if (event_enabled) { set_events_for_handle(fd); return; } #endif if (read_func) { FD_SET(fd, &w_read); } else { FD_CLR(fd, &w_read); FD_CLR(fd, &x_read); } if (write_func) { FD_SET(fd, &w_write); } else { FD_CLR(fd, &w_write); FD_CLR(fd, &x_write); } if (error_func) { FD_SET(fd, &w_error); } else { FD_CLR(fd, &w_error); FD_CLR(fd, &x_error); } } void select_loop(void (*init)(void)) { timeval_T last_time; int select_errors = 0; clear_signal_mask_and_handlers(); FD_ZERO(&w_read); FD_ZERO(&w_write); FD_ZERO(&w_error); w_max = 0; timeval_now(&last_time); #ifdef SIGPIPE signal(SIGPIPE, SIG_IGN); #endif init(); check_bottom_halves(); #ifdef USE_LIBEVENT enable_libevent(); #if defined(USE_POLL) if (!event_enabled) { restrict_fds(); } #endif if (event_enabled) { while (!program.terminate) { check_signals(); if (1 /*(!F)*/) { do_event_loop(EVLOOP_NONBLOCK); check_signals(); redraw_all_terminals(); } if (program.terminate) break; do_event_loop(EVLOOP_ONCE); } } else #endif while (!program.terminate) { struct timeval *timeout = NULL; int n, i, has_timer; timeval_T t; check_signals(); check_timers(&last_time); redraw_all_terminals(); memcpy(&x_read, &w_read, sizeof(fd_set)); memcpy(&x_write, &w_write, sizeof(fd_set)); memcpy(&x_error, &w_error, sizeof(fd_set)); if (program.terminate) break; has_timer = get_next_timer_time(&t); if (!w_max && !has_timer) break; critical_section = 1; if (check_signals()) { critical_section = 0; continue; } #if 0 { int i; printf("\nR:"); for (i = 0; i < 256; i++) if (FD_ISSET(i, &x_read)) printf("%d,", i); printf("\nW:"); for (i = 0; i < 256; i++) if (FD_ISSET(i, &x_write)) printf("%d,", i); printf("\nE:"); for (i = 0; i < 256; i++) if (FD_ISSET(i, &x_error)) printf("%d,", i); fflush(stdout); } #endif if (has_timer) { /* Be sure timeout is not negative. */ timeval_limit_to_zero_or_one(&t); timeout = (struct timeval *) &t; } n = select(w_max, &x_read, &x_write, &x_error, timeout); if (n < 0) { /* The following calls (especially gettext) * might change errno. */ const int errno_from_select = errno; critical_section = 0; uninstall_alarm(); if (errno_from_select != EINTR) { ERROR(gettext("The call to %s failed: %d (%s)"), "select()", errno_from_select, (unsigned char *) strerror(errno_from_select)); if (++select_errors > 10) /* Infinite loop prevention. */ INTERNAL(gettext("%d select() failures."), select_errors); } continue; } select_errors = 0; critical_section = 0; uninstall_alarm(); check_signals(); /*printf("sel: %d\n", n);*/ check_timers(&last_time); i = -1; while (n > 0 && ++i < w_max) { int k = 0; #if 0 printf("C %d : %d,%d,%d\n", i, FD_ISSET(i, &w_read), FD_ISSET(i, &w_write), FD_ISSET(i, &w_error)); printf("A %d : %d,%d,%d\n", i, FD_ISSET(i, &x_read), FD_ISSET(i, &x_write), FD_ISSET(i, &x_error)); #endif if (FD_ISSET(i, &x_read)) { if (threads[i].read_func) { threads[i].read_func(threads[i].data); check_bottom_halves(); } k = 1; } if (FD_ISSET(i, &x_write)) { if (threads[i].write_func) { threads[i].write_func(threads[i].data); check_bottom_halves(); } k = 1; } if (FD_ISSET(i, &x_error)) { if (threads[i].error_func) { threads[i].error_func(threads[i].data); check_bottom_halves(); } k = 1; } n -= k; } } } static int can_read_or_write(int fd, int write) { #if defined(USE_POLL) struct pollfd p; int rs; p.fd = fd; p.events = !write ? POLLIN : POLLOUT; EINTRLOOP(rs, poll(&p, 1, 0)); if (rs < 0) elinks_internal("ERROR: poll for %s (%d) failed: %s", !write ? "read" : "write", fd, strerror(errno)); if (!rs) return 0; if (p.revents & POLLNVAL) elinks_internal("ERROR: poll for %s (%d) failed: %s", !write ? "read" : "write", fd, strerror(errno)); return 1; #else struct timeval tv = {0, 0}; fd_set fds; fd_set *rfds = NULL; fd_set *wfds = NULL; FD_ZERO(&fds); FD_SET(fd, &fds); if (write) wfds = &fds; else rfds = &fds; return select(fd + 1, rfds, wfds, NULL, &tv); #endif } int can_read(int fd) { return can_read_or_write(fd, 0); } int can_write(int fd) { return can_read_or_write(fd, 1); } void terminate_select(void) { #ifdef USE_LIBEVENT terminate_libevent(); #endif mem_free_if(threads); }