1af93fea7SJake Freeland /*- 2af93fea7SJake Freeland * SPDX-License-Identifier: BSD-2-Clause 3af93fea7SJake Freeland * 4af93fea7SJake Freeland * Copyright (c) 2014 Dmitry Chagin <dchagin@FreeBSD.org> 5af93fea7SJake Freeland * Copyright (c) 2023 Jake Freeland <jfree@FreeBSD.org> 6af93fea7SJake Freeland * 7af93fea7SJake Freeland * Redistribution and use in source and binary forms, with or without 8af93fea7SJake Freeland * modification, are permitted provided that the following conditions 9af93fea7SJake Freeland * are met: 10af93fea7SJake Freeland * 1. Redistributions of source code must retain the above copyright 11af93fea7SJake Freeland * notice, this list of conditions and the following disclaimer. 12af93fea7SJake Freeland * 2. Redistributions in binary form must reproduce the above copyright 13af93fea7SJake Freeland * notice, this list of conditions and the following disclaimer in the 14af93fea7SJake Freeland * documentation and/or other materials provided with the distribution. 15af93fea7SJake Freeland * 16af93fea7SJake Freeland * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17af93fea7SJake Freeland * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18af93fea7SJake Freeland * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19af93fea7SJake Freeland * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20af93fea7SJake Freeland * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21af93fea7SJake Freeland * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22af93fea7SJake Freeland * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23af93fea7SJake Freeland * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24af93fea7SJake Freeland * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25af93fea7SJake Freeland * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26af93fea7SJake Freeland * SUCH DAMAGE. 27af93fea7SJake Freeland */ 28af93fea7SJake Freeland 29af93fea7SJake Freeland #include <sys/param.h> 30af93fea7SJake Freeland #include <sys/systm.h> 31af93fea7SJake Freeland #include <sys/callout.h> 32af93fea7SJake Freeland #include <sys/fcntl.h> 33af93fea7SJake Freeland #include <sys/file.h> 34af93fea7SJake Freeland #include <sys/filedesc.h> 35af93fea7SJake Freeland #include <sys/filio.h> 36af93fea7SJake Freeland #include <sys/kernel.h> 37af93fea7SJake Freeland #include <sys/lock.h> 38af93fea7SJake Freeland #include <sys/malloc.h> 39af93fea7SJake Freeland #include <sys/mount.h> 40af93fea7SJake Freeland #include <sys/mutex.h> 41af93fea7SJake Freeland #include <sys/poll.h> 42af93fea7SJake Freeland #include <sys/proc.h> 43af93fea7SJake Freeland #include <sys/queue.h> 44af93fea7SJake Freeland #include <sys/selinfo.h> 45af93fea7SJake Freeland #include <sys/stat.h> 4602f534b5SMateusz Guzik #include <sys/sx.h> 4718cb4223SJohn Baldwin #include <sys/syscallsubr.h> 48af93fea7SJake Freeland #include <sys/sysctl.h> 49af93fea7SJake Freeland #include <sys/sysent.h> 50af93fea7SJake Freeland #include <sys/sysproto.h> 51af93fea7SJake Freeland #include <sys/timerfd.h> 52af93fea7SJake Freeland #include <sys/timespec.h> 53af93fea7SJake Freeland #include <sys/uio.h> 54af93fea7SJake Freeland #include <sys/user.h> 55af93fea7SJake Freeland 56af93fea7SJake Freeland #include <security/audit/audit.h> 57af93fea7SJake Freeland 58af93fea7SJake Freeland static MALLOC_DEFINE(M_TIMERFD, "timerfd", "timerfd structures"); 5902f534b5SMateusz Guzik 60f4296cfbSMateusz Guzik static struct mtx timerfd_list_lock; 6102f534b5SMateusz Guzik static LIST_HEAD(, timerfd) timerfd_list; 62f4296cfbSMateusz Guzik MTX_SYSINIT(timerfd, &timerfd_list_lock, "timerfd_list_lock", MTX_DEF); 6302f534b5SMateusz Guzik 64af93fea7SJake Freeland static struct unrhdr64 tfdino_unr; 65af93fea7SJake Freeland 66af93fea7SJake Freeland #define TFD_NOJUMP 0 /* Realtime clock has not jumped. */ 67af93fea7SJake Freeland #define TFD_READ 1 /* Jumped, tfd has been read since. */ 68af93fea7SJake Freeland #define TFD_ZREAD 2 /* Jumped backwards, CANCEL_ON_SET=false. */ 69af93fea7SJake Freeland #define TFD_CANCELED 4 /* Jumped, CANCEL_ON_SET=true. */ 70af93fea7SJake Freeland #define TFD_JUMPED (TFD_ZREAD | TFD_CANCELED) 71af93fea7SJake Freeland 72a1f50615SJake Freeland /* 73a1f50615SJake Freeland * One structure allocated per timerfd descriptor. 74a1f50615SJake Freeland * 75a1f50615SJake Freeland * Locking semantics: 76a1f50615SJake Freeland * (t) locked by tfd_lock mtx 77a1f50615SJake Freeland * (l) locked by timerfd_list_lock sx 78a1f50615SJake Freeland * (c) const until freeing 79a1f50615SJake Freeland */ 80af93fea7SJake Freeland struct timerfd { 81af93fea7SJake Freeland /* User specified. */ 82a1f50615SJake Freeland struct itimerspec tfd_time; /* (t) tfd timer */ 83a1f50615SJake Freeland clockid_t tfd_clockid; /* (c) timing base */ 84a1f50615SJake Freeland int tfd_flags; /* (c) creation flags */ 85a1f50615SJake Freeland int tfd_timflags; /* (t) timer flags */ 86af93fea7SJake Freeland 87af93fea7SJake Freeland /* Used internally. */ 88a1f50615SJake Freeland timerfd_t tfd_count; /* (t) expiration count since read */ 89a1f50615SJake Freeland bool tfd_expired; /* (t) true upon initial expiration */ 90a1f50615SJake Freeland struct mtx tfd_lock; /* tfd mtx lock */ 91a1f50615SJake Freeland struct callout tfd_callout; /* (t) expiration notification */ 92a1f50615SJake Freeland struct selinfo tfd_sel; /* (t) I/O alerts */ 93a1f50615SJake Freeland struct timespec tfd_boottim; /* (t) cached boottime */ 94a1f50615SJake Freeland int tfd_jumped; /* (t) timer jump status */ 95a1f50615SJake Freeland LIST_ENTRY(timerfd) entry; /* (l) entry in list */ 96af93fea7SJake Freeland 97af93fea7SJake Freeland /* For stat(2). */ 98a1f50615SJake Freeland ino_t tfd_ino; /* (c) inode number */ 99a1f50615SJake Freeland struct timespec tfd_atim; /* (t) time of last read */ 100a1f50615SJake Freeland struct timespec tfd_mtim; /* (t) time of last settime */ 101a1f50615SJake Freeland struct timespec tfd_birthtim; /* (c) creation time */ 102af93fea7SJake Freeland }; 103af93fea7SJake Freeland 104af93fea7SJake Freeland static void 105af93fea7SJake Freeland timerfd_init(void *data) 106af93fea7SJake Freeland { 107af93fea7SJake Freeland new_unrhdr64(&tfdino_unr, 1); 108af93fea7SJake Freeland } 109af93fea7SJake Freeland 110af93fea7SJake Freeland SYSINIT(timerfd, SI_SUB_VFS, SI_ORDER_ANY, timerfd_init, NULL); 111af93fea7SJake Freeland 112af93fea7SJake Freeland static inline void 113af93fea7SJake Freeland timerfd_getboottime(struct timespec *ts) 114af93fea7SJake Freeland { 115af93fea7SJake Freeland struct timeval tv; 116a1f50615SJake Freeland 117af93fea7SJake Freeland getboottime(&tv); 118af93fea7SJake Freeland TIMEVAL_TO_TIMESPEC(&tv, ts); 119af93fea7SJake Freeland } 120af93fea7SJake Freeland 121af93fea7SJake Freeland /* 122af93fea7SJake Freeland * Call when a discontinuous jump has occured in CLOCK_REALTIME and 123af93fea7SJake Freeland * update timerfd's cached boottime. A jump can be triggered using 124af93fea7SJake Freeland * functions like clock_settime(2) or settimeofday(2). 125af93fea7SJake Freeland * 126af93fea7SJake Freeland * Timer is marked TFD_CANCELED if TFD_TIMER_CANCEL_ON_SET is set 127af93fea7SJake Freeland * and the realtime clock jumps. 128af93fea7SJake Freeland * Timer is marked TFD_ZREAD if TFD_TIMER_CANCEL_ON_SET is not set, 129af93fea7SJake Freeland * but the realtime clock jumps backwards. 130af93fea7SJake Freeland */ 131af93fea7SJake Freeland void 132af93fea7SJake Freeland timerfd_jumped(void) 133af93fea7SJake Freeland { 134af93fea7SJake Freeland struct timerfd *tfd; 135af93fea7SJake Freeland struct timespec boottime, diff; 136af93fea7SJake Freeland 1375eab5230SMateusz Guzik if (LIST_EMPTY(&timerfd_list)) 1385eab5230SMateusz Guzik return; 1395eab5230SMateusz Guzik 140af93fea7SJake Freeland timerfd_getboottime(&boottime); 141f4296cfbSMateusz Guzik mtx_lock(&timerfd_list_lock); 14202f534b5SMateusz Guzik LIST_FOREACH(tfd, &timerfd_list, entry) { 143af93fea7SJake Freeland mtx_lock(&tfd->tfd_lock); 144af93fea7SJake Freeland if (tfd->tfd_clockid != CLOCK_REALTIME || 145af93fea7SJake Freeland (tfd->tfd_timflags & TFD_TIMER_ABSTIME) == 0 || 146af93fea7SJake Freeland timespeccmp(&boottime, &tfd->tfd_boottim, ==)) { 147af93fea7SJake Freeland mtx_unlock(&tfd->tfd_lock); 148af93fea7SJake Freeland continue; 149af93fea7SJake Freeland } 150af93fea7SJake Freeland 151af93fea7SJake Freeland if (callout_active(&tfd->tfd_callout)) { 152af93fea7SJake Freeland if ((tfd->tfd_timflags & TFD_TIMER_CANCEL_ON_SET) != 0) 153af93fea7SJake Freeland tfd->tfd_jumped = TFD_CANCELED; 154af93fea7SJake Freeland else if (timespeccmp(&boottime, &tfd->tfd_boottim, <)) 155af93fea7SJake Freeland tfd->tfd_jumped = TFD_ZREAD; 156af93fea7SJake Freeland 157af93fea7SJake Freeland /* 158af93fea7SJake Freeland * Do not reschedule callout when 159af93fea7SJake Freeland * inside interval time loop. 160af93fea7SJake Freeland */ 161af93fea7SJake Freeland if (!tfd->tfd_expired) { 162af93fea7SJake Freeland timespecsub(&boottime, 163af93fea7SJake Freeland &tfd->tfd_boottim, &diff); 164af93fea7SJake Freeland timespecsub(&tfd->tfd_time.it_value, 165af93fea7SJake Freeland &diff, &tfd->tfd_time.it_value); 166af93fea7SJake Freeland if (callout_stop(&tfd->tfd_callout) == 1) { 167af93fea7SJake Freeland callout_schedule_sbt(&tfd->tfd_callout, 168af93fea7SJake Freeland tstosbt(tfd->tfd_time.it_value), 169af93fea7SJake Freeland 0, C_ABSOLUTE); 170af93fea7SJake Freeland } 171af93fea7SJake Freeland } 172af93fea7SJake Freeland } 173af93fea7SJake Freeland 174af93fea7SJake Freeland tfd->tfd_boottim = boottime; 175af93fea7SJake Freeland mtx_unlock(&tfd->tfd_lock); 176af93fea7SJake Freeland } 177f4296cfbSMateusz Guzik mtx_unlock(&timerfd_list_lock); 178af93fea7SJake Freeland } 179af93fea7SJake Freeland 180af93fea7SJake Freeland static int 181af93fea7SJake Freeland timerfd_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 182af93fea7SJake Freeland int flags, struct thread *td) 183af93fea7SJake Freeland { 184af93fea7SJake Freeland struct timerfd *tfd = fp->f_data; 185af93fea7SJake Freeland timerfd_t count; 186af93fea7SJake Freeland int error = 0; 187af93fea7SJake Freeland 188af93fea7SJake Freeland if (uio->uio_resid < sizeof(timerfd_t)) 189af93fea7SJake Freeland return (EINVAL); 190af93fea7SJake Freeland 191af93fea7SJake Freeland mtx_lock(&tfd->tfd_lock); 192af93fea7SJake Freeland retry: 193af93fea7SJake Freeland getnanotime(&tfd->tfd_atim); 194af93fea7SJake Freeland if ((tfd->tfd_jumped & TFD_JUMPED) != 0) { 195af93fea7SJake Freeland if (tfd->tfd_jumped == TFD_CANCELED) 196af93fea7SJake Freeland error = ECANCELED; 197af93fea7SJake Freeland tfd->tfd_jumped = TFD_READ; 198af93fea7SJake Freeland tfd->tfd_count = 0; 199af93fea7SJake Freeland mtx_unlock(&tfd->tfd_lock); 200af93fea7SJake Freeland return (error); 201af93fea7SJake Freeland } else { 202af93fea7SJake Freeland tfd->tfd_jumped = TFD_NOJUMP; 203af93fea7SJake Freeland } 204af93fea7SJake Freeland if (tfd->tfd_count == 0) { 205af93fea7SJake Freeland if ((fp->f_flag & FNONBLOCK) != 0) { 206af93fea7SJake Freeland mtx_unlock(&tfd->tfd_lock); 207af93fea7SJake Freeland return (EAGAIN); 208af93fea7SJake Freeland } 209af93fea7SJake Freeland td->td_rtcgen = atomic_load_acq_int(&rtc_generation); 210af93fea7SJake Freeland error = mtx_sleep(&tfd->tfd_count, &tfd->tfd_lock, 211af93fea7SJake Freeland PCATCH, "tfdrd", 0); 212af93fea7SJake Freeland if (error == 0) { 213af93fea7SJake Freeland goto retry; 214af93fea7SJake Freeland } else { 215af93fea7SJake Freeland mtx_unlock(&tfd->tfd_lock); 216af93fea7SJake Freeland return (error); 217af93fea7SJake Freeland } 218af93fea7SJake Freeland } 219af93fea7SJake Freeland 220af93fea7SJake Freeland count = tfd->tfd_count; 221af93fea7SJake Freeland tfd->tfd_count = 0; 222af93fea7SJake Freeland mtx_unlock(&tfd->tfd_lock); 223af93fea7SJake Freeland error = uiomove(&count, sizeof(timerfd_t), uio); 224af93fea7SJake Freeland 225af93fea7SJake Freeland return (error); 226af93fea7SJake Freeland } 227af93fea7SJake Freeland 228af93fea7SJake Freeland static int 229af93fea7SJake Freeland timerfd_ioctl(struct file *fp, u_long cmd, void *data, 230af93fea7SJake Freeland struct ucred *active_cred, struct thread *td) 231af93fea7SJake Freeland { 232af93fea7SJake Freeland switch (cmd) { 233af93fea7SJake Freeland case FIOASYNC: 234af93fea7SJake Freeland if (*(int *)data != 0) 235af93fea7SJake Freeland atomic_set_int(&fp->f_flag, FASYNC); 236af93fea7SJake Freeland else 237af93fea7SJake Freeland atomic_clear_int(&fp->f_flag, FASYNC); 238af93fea7SJake Freeland return (0); 239af93fea7SJake Freeland case FIONBIO: 240af93fea7SJake Freeland if (*(int *)data != 0) 241af93fea7SJake Freeland atomic_set_int(&fp->f_flag, FNONBLOCK); 242af93fea7SJake Freeland else 243af93fea7SJake Freeland atomic_clear_int(&fp->f_flag, FNONBLOCK); 244af93fea7SJake Freeland return (0); 245af93fea7SJake Freeland } 246af93fea7SJake Freeland return (ENOTTY); 247af93fea7SJake Freeland } 248af93fea7SJake Freeland 249af93fea7SJake Freeland static int 250af93fea7SJake Freeland timerfd_poll(struct file *fp, int events, struct ucred *active_cred, 251af93fea7SJake Freeland struct thread *td) 252af93fea7SJake Freeland { 253af93fea7SJake Freeland struct timerfd *tfd = fp->f_data; 254af93fea7SJake Freeland int revents = 0; 255af93fea7SJake Freeland 256af93fea7SJake Freeland mtx_lock(&tfd->tfd_lock); 257af93fea7SJake Freeland if ((events & (POLLIN | POLLRDNORM)) != 0 && 258af93fea7SJake Freeland tfd->tfd_count > 0 && tfd->tfd_jumped != TFD_READ) 259af93fea7SJake Freeland revents |= events & (POLLIN | POLLRDNORM); 260af93fea7SJake Freeland if (revents == 0) 261af93fea7SJake Freeland selrecord(td, &tfd->tfd_sel); 262af93fea7SJake Freeland mtx_unlock(&tfd->tfd_lock); 263af93fea7SJake Freeland 264af93fea7SJake Freeland return (revents); 265af93fea7SJake Freeland } 266af93fea7SJake Freeland 267af93fea7SJake Freeland static void 268af93fea7SJake Freeland filt_timerfddetach(struct knote *kn) 269af93fea7SJake Freeland { 270af93fea7SJake Freeland struct timerfd *tfd = kn->kn_hook; 271af93fea7SJake Freeland 272af93fea7SJake Freeland mtx_lock(&tfd->tfd_lock); 273af93fea7SJake Freeland knlist_remove(&tfd->tfd_sel.si_note, kn, 1); 274af93fea7SJake Freeland mtx_unlock(&tfd->tfd_lock); 275af93fea7SJake Freeland } 276af93fea7SJake Freeland 277af93fea7SJake Freeland static int 278af93fea7SJake Freeland filt_timerfdread(struct knote *kn, long hint) 279af93fea7SJake Freeland { 280af93fea7SJake Freeland struct timerfd *tfd = kn->kn_hook; 281af93fea7SJake Freeland 282a1f50615SJake Freeland mtx_assert(&tfd->tfd_lock, MA_OWNED); 283a1f50615SJake Freeland kn->kn_data = (int64_t)tfd->tfd_count; 284af93fea7SJake Freeland return (tfd->tfd_count > 0); 285af93fea7SJake Freeland } 286af93fea7SJake Freeland 287af93fea7SJake Freeland static struct filterops timerfd_rfiltops = { 288af93fea7SJake Freeland .f_isfd = 1, 289af93fea7SJake Freeland .f_detach = filt_timerfddetach, 290af93fea7SJake Freeland .f_event = filt_timerfdread, 291af93fea7SJake Freeland }; 292af93fea7SJake Freeland 293af93fea7SJake Freeland static int 294af93fea7SJake Freeland timerfd_kqfilter(struct file *fp, struct knote *kn) 295af93fea7SJake Freeland { 296af93fea7SJake Freeland struct timerfd *tfd = fp->f_data; 297af93fea7SJake Freeland 298af93fea7SJake Freeland if (kn->kn_filter != EVFILT_READ) 299af93fea7SJake Freeland return (EINVAL); 300af93fea7SJake Freeland 301af93fea7SJake Freeland kn->kn_fop = &timerfd_rfiltops; 302af93fea7SJake Freeland kn->kn_hook = tfd; 303af93fea7SJake Freeland knlist_add(&tfd->tfd_sel.si_note, kn, 0); 304af93fea7SJake Freeland 305af93fea7SJake Freeland return (0); 306af93fea7SJake Freeland } 307af93fea7SJake Freeland 308af93fea7SJake Freeland static int 309af93fea7SJake Freeland timerfd_stat(struct file *fp, struct stat *sb, struct ucred *active_cred) 310af93fea7SJake Freeland { 311af93fea7SJake Freeland struct timerfd *tfd = fp->f_data; 312af93fea7SJake Freeland 313af93fea7SJake Freeland bzero(sb, sizeof(*sb)); 314af93fea7SJake Freeland sb->st_nlink = fp->f_count - 1; 315af93fea7SJake Freeland sb->st_uid = fp->f_cred->cr_uid; 316af93fea7SJake Freeland sb->st_gid = fp->f_cred->cr_gid; 317af93fea7SJake Freeland sb->st_blksize = PAGE_SIZE; 318af93fea7SJake Freeland mtx_lock(&tfd->tfd_lock); 319af93fea7SJake Freeland sb->st_atim = tfd->tfd_atim; 320af93fea7SJake Freeland sb->st_mtim = tfd->tfd_mtim; 321af93fea7SJake Freeland mtx_unlock(&tfd->tfd_lock); 322a1f50615SJake Freeland sb->st_ctim = sb->st_mtim; 323a1f50615SJake Freeland sb->st_ino = tfd->tfd_ino; 324a1f50615SJake Freeland sb->st_birthtim = tfd->tfd_birthtim; 325af93fea7SJake Freeland 326af93fea7SJake Freeland return (0); 327af93fea7SJake Freeland } 328af93fea7SJake Freeland 329af93fea7SJake Freeland static int 330af93fea7SJake Freeland timerfd_close(struct file *fp, struct thread *td) 331af93fea7SJake Freeland { 332af93fea7SJake Freeland struct timerfd *tfd = fp->f_data; 333af93fea7SJake Freeland 334f4296cfbSMateusz Guzik mtx_lock(&timerfd_list_lock); 33502f534b5SMateusz Guzik LIST_REMOVE(tfd, entry); 336f4296cfbSMateusz Guzik mtx_unlock(&timerfd_list_lock); 33702f534b5SMateusz Guzik 338af93fea7SJake Freeland callout_drain(&tfd->tfd_callout); 339af93fea7SJake Freeland seldrain(&tfd->tfd_sel); 340af93fea7SJake Freeland knlist_destroy(&tfd->tfd_sel.si_note); 341af93fea7SJake Freeland mtx_destroy(&tfd->tfd_lock); 342af93fea7SJake Freeland free(tfd, M_TIMERFD); 343af93fea7SJake Freeland fp->f_ops = &badfileops; 344af93fea7SJake Freeland 345af93fea7SJake Freeland return (0); 346af93fea7SJake Freeland } 347af93fea7SJake Freeland 348af93fea7SJake Freeland static int 349af93fea7SJake Freeland timerfd_fill_kinfo(struct file *fp, struct kinfo_file *kif, 350af93fea7SJake Freeland struct filedesc *fdp) 351af93fea7SJake Freeland { 352af93fea7SJake Freeland struct timerfd *tfd = fp->f_data; 353af93fea7SJake Freeland 354af93fea7SJake Freeland kif->kf_type = KF_TYPE_TIMERFD; 355af93fea7SJake Freeland kif->kf_un.kf_timerfd.kf_timerfd_clockid = tfd->tfd_clockid; 356af93fea7SJake Freeland kif->kf_un.kf_timerfd.kf_timerfd_flags = tfd->tfd_flags; 357af93fea7SJake Freeland kif->kf_un.kf_timerfd.kf_timerfd_addr = (uintptr_t)tfd; 358af93fea7SJake Freeland 359af93fea7SJake Freeland return (0); 360af93fea7SJake Freeland } 361af93fea7SJake Freeland 362af93fea7SJake Freeland static struct fileops timerfdops = { 363af93fea7SJake Freeland .fo_read = timerfd_read, 364af93fea7SJake Freeland .fo_write = invfo_rdwr, 365af93fea7SJake Freeland .fo_truncate = invfo_truncate, 366af93fea7SJake Freeland .fo_ioctl = timerfd_ioctl, 367af93fea7SJake Freeland .fo_poll = timerfd_poll, 368af93fea7SJake Freeland .fo_kqfilter = timerfd_kqfilter, 369af93fea7SJake Freeland .fo_stat = timerfd_stat, 370af93fea7SJake Freeland .fo_close = timerfd_close, 371af93fea7SJake Freeland .fo_chmod = invfo_chmod, 372af93fea7SJake Freeland .fo_chown = invfo_chown, 373af93fea7SJake Freeland .fo_sendfile = invfo_sendfile, 374af93fea7SJake Freeland .fo_fill_kinfo = timerfd_fill_kinfo, 375f28526e9SKonstantin Belousov .fo_cmp = file_kcmp_generic, 376af93fea7SJake Freeland .fo_flags = DFLAG_PASSABLE, 377af93fea7SJake Freeland }; 378af93fea7SJake Freeland 379af93fea7SJake Freeland static void 380af93fea7SJake Freeland timerfd_curval(struct timerfd *tfd, struct itimerspec *old_value) 381af93fea7SJake Freeland { 382af93fea7SJake Freeland struct timespec curr_value; 383af93fea7SJake Freeland 384a1f50615SJake Freeland mtx_assert(&tfd->tfd_lock, MA_OWNED); 385af93fea7SJake Freeland *old_value = tfd->tfd_time; 386af93fea7SJake Freeland if (timespecisset(&tfd->tfd_time.it_value)) { 387af93fea7SJake Freeland nanouptime(&curr_value); 388af93fea7SJake Freeland timespecsub(&tfd->tfd_time.it_value, &curr_value, 389af93fea7SJake Freeland &old_value->it_value); 390af93fea7SJake Freeland } 391af93fea7SJake Freeland } 392af93fea7SJake Freeland 393af93fea7SJake Freeland static void 394af93fea7SJake Freeland timerfd_expire(void *arg) 395af93fea7SJake Freeland { 396af93fea7SJake Freeland struct timerfd *tfd = (struct timerfd *)arg; 397af93fea7SJake Freeland struct timespec uptime; 398af93fea7SJake Freeland 399af93fea7SJake Freeland ++tfd->tfd_count; 400af93fea7SJake Freeland tfd->tfd_expired = true; 401af93fea7SJake Freeland if (timespecisset(&tfd->tfd_time.it_interval)) { 402af93fea7SJake Freeland /* Count missed events. */ 403af93fea7SJake Freeland nanouptime(&uptime); 404af93fea7SJake Freeland if (timespeccmp(&uptime, &tfd->tfd_time.it_value, >)) { 405af93fea7SJake Freeland timespecsub(&uptime, &tfd->tfd_time.it_value, &uptime); 406af93fea7SJake Freeland tfd->tfd_count += tstosbt(uptime) / 407af93fea7SJake Freeland tstosbt(tfd->tfd_time.it_interval); 408af93fea7SJake Freeland } 409af93fea7SJake Freeland timespecadd(&tfd->tfd_time.it_value, 410af93fea7SJake Freeland &tfd->tfd_time.it_interval, &tfd->tfd_time.it_value); 411af93fea7SJake Freeland callout_schedule_sbt(&tfd->tfd_callout, 412af93fea7SJake Freeland tstosbt(tfd->tfd_time.it_value), 413af93fea7SJake Freeland 0, C_ABSOLUTE); 414af93fea7SJake Freeland } else { 415af93fea7SJake Freeland /* Single shot timer. */ 416af93fea7SJake Freeland callout_deactivate(&tfd->tfd_callout); 417af93fea7SJake Freeland timespecclear(&tfd->tfd_time.it_value); 418af93fea7SJake Freeland } 419af93fea7SJake Freeland 420af93fea7SJake Freeland wakeup(&tfd->tfd_count); 421af93fea7SJake Freeland selwakeup(&tfd->tfd_sel); 422af93fea7SJake Freeland KNOTE_LOCKED(&tfd->tfd_sel.si_note, 0); 423af93fea7SJake Freeland } 424af93fea7SJake Freeland 425af93fea7SJake Freeland int 426af93fea7SJake Freeland kern_timerfd_create(struct thread *td, int clockid, int flags) 427af93fea7SJake Freeland { 428af93fea7SJake Freeland struct file *fp; 429af93fea7SJake Freeland struct timerfd *tfd; 4305eab5230SMateusz Guzik int error, fd, fflags; 431af93fea7SJake Freeland 432af93fea7SJake Freeland AUDIT_ARG_VALUE(clockid); 433af93fea7SJake Freeland AUDIT_ARG_FFLAGS(flags); 434af93fea7SJake Freeland 435af93fea7SJake Freeland if (clockid != CLOCK_REALTIME && clockid != CLOCK_MONOTONIC) 436af93fea7SJake Freeland return (EINVAL); 437af93fea7SJake Freeland if ((flags & ~(TFD_CLOEXEC | TFD_NONBLOCK)) != 0) 438af93fea7SJake Freeland return (EINVAL); 4395eab5230SMateusz Guzik 4405eab5230SMateusz Guzik fflags = FREAD; 441af93fea7SJake Freeland if ((flags & TFD_CLOEXEC) != 0) 442af93fea7SJake Freeland fflags |= O_CLOEXEC; 4435eab5230SMateusz Guzik if ((flags & TFD_NONBLOCK) != 0) 4445eab5230SMateusz Guzik fflags |= FNONBLOCK; 445af93fea7SJake Freeland 44602f534b5SMateusz Guzik error = falloc(td, &fp, &fd, fflags); 44702f534b5SMateusz Guzik if (error != 0) 44802f534b5SMateusz Guzik return (error); 44902f534b5SMateusz Guzik 450af93fea7SJake Freeland tfd = malloc(sizeof(*tfd), M_TIMERFD, M_WAITOK | M_ZERO); 451af93fea7SJake Freeland tfd->tfd_clockid = (clockid_t)clockid; 452af93fea7SJake Freeland tfd->tfd_flags = flags; 453af93fea7SJake Freeland tfd->tfd_ino = alloc_unr64(&tfdino_unr); 454af93fea7SJake Freeland mtx_init(&tfd->tfd_lock, "timerfd", NULL, MTX_DEF); 455af93fea7SJake Freeland callout_init_mtx(&tfd->tfd_callout, &tfd->tfd_lock, 0); 456af93fea7SJake Freeland knlist_init_mtx(&tfd->tfd_sel.si_note, &tfd->tfd_lock); 457af93fea7SJake Freeland timerfd_getboottime(&tfd->tfd_boottim); 458af93fea7SJake Freeland getnanotime(&tfd->tfd_birthtim); 459f4296cfbSMateusz Guzik mtx_lock(&timerfd_list_lock); 46002f534b5SMateusz Guzik LIST_INSERT_HEAD(&timerfd_list, tfd, entry); 461f4296cfbSMateusz Guzik mtx_unlock(&timerfd_list_lock); 462af93fea7SJake Freeland 463af93fea7SJake Freeland finit(fp, fflags, DTYPE_TIMERFD, tfd, &timerfdops); 46402f534b5SMateusz Guzik 465af93fea7SJake Freeland fdrop(fp, td); 466af93fea7SJake Freeland 467af93fea7SJake Freeland td->td_retval[0] = fd; 468af93fea7SJake Freeland return (0); 469af93fea7SJake Freeland } 470af93fea7SJake Freeland 471af93fea7SJake Freeland int 472af93fea7SJake Freeland kern_timerfd_gettime(struct thread *td, int fd, struct itimerspec *curr_value) 473af93fea7SJake Freeland { 474af93fea7SJake Freeland struct file *fp; 475af93fea7SJake Freeland struct timerfd *tfd; 476af93fea7SJake Freeland int error; 477af93fea7SJake Freeland 478af93fea7SJake Freeland error = fget(td, fd, &cap_write_rights, &fp); 479af93fea7SJake Freeland if (error != 0) 480af93fea7SJake Freeland return (error); 481a1f50615SJake Freeland if (fp->f_type != DTYPE_TIMERFD) { 482af93fea7SJake Freeland fdrop(fp, td); 483af93fea7SJake Freeland return (EINVAL); 484af93fea7SJake Freeland } 485a1f50615SJake Freeland tfd = fp->f_data; 486af93fea7SJake Freeland 487af93fea7SJake Freeland mtx_lock(&tfd->tfd_lock); 488af93fea7SJake Freeland timerfd_curval(tfd, curr_value); 489af93fea7SJake Freeland mtx_unlock(&tfd->tfd_lock); 490af93fea7SJake Freeland 491af93fea7SJake Freeland fdrop(fp, td); 492af93fea7SJake Freeland return (0); 493af93fea7SJake Freeland } 494af93fea7SJake Freeland 495af93fea7SJake Freeland int 496af93fea7SJake Freeland kern_timerfd_settime(struct thread *td, int fd, int flags, 497af93fea7SJake Freeland const struct itimerspec *new_value, struct itimerspec *old_value) 498af93fea7SJake Freeland { 499af93fea7SJake Freeland struct file *fp; 500af93fea7SJake Freeland struct timerfd *tfd; 501af93fea7SJake Freeland struct timespec ts; 502af93fea7SJake Freeland int error = 0; 503af93fea7SJake Freeland 504af93fea7SJake Freeland if ((flags & ~(TFD_TIMER_ABSTIME | TFD_TIMER_CANCEL_ON_SET)) != 0) 505af93fea7SJake Freeland return (EINVAL); 506af93fea7SJake Freeland if (!timespecvalid_interval(&new_value->it_value) || 507af93fea7SJake Freeland !timespecvalid_interval(&new_value->it_interval)) 508af93fea7SJake Freeland return (EINVAL); 509af93fea7SJake Freeland 510af93fea7SJake Freeland error = fget(td, fd, &cap_write_rights, &fp); 511af93fea7SJake Freeland if (error != 0) 512af93fea7SJake Freeland return (error); 513a1f50615SJake Freeland if (fp->f_type != DTYPE_TIMERFD) { 514af93fea7SJake Freeland fdrop(fp, td); 515af93fea7SJake Freeland return (EINVAL); 516af93fea7SJake Freeland } 517a1f50615SJake Freeland tfd = fp->f_data; 518af93fea7SJake Freeland 519af93fea7SJake Freeland mtx_lock(&tfd->tfd_lock); 520af93fea7SJake Freeland getnanotime(&tfd->tfd_mtim); 521af93fea7SJake Freeland tfd->tfd_timflags = flags; 522af93fea7SJake Freeland 523af93fea7SJake Freeland /* Store old itimerspec, if applicable. */ 524af93fea7SJake Freeland if (old_value != NULL) 525af93fea7SJake Freeland timerfd_curval(tfd, old_value); 526af93fea7SJake Freeland 527af93fea7SJake Freeland /* Set new expiration. */ 528af93fea7SJake Freeland tfd->tfd_time = *new_value; 529af93fea7SJake Freeland if (timespecisset(&tfd->tfd_time.it_value)) { 530af93fea7SJake Freeland if ((flags & TFD_TIMER_ABSTIME) == 0) { 531af93fea7SJake Freeland nanouptime(&ts); 532af93fea7SJake Freeland timespecadd(&tfd->tfd_time.it_value, &ts, 533af93fea7SJake Freeland &tfd->tfd_time.it_value); 534af93fea7SJake Freeland } else if (tfd->tfd_clockid == CLOCK_REALTIME) { 535af93fea7SJake Freeland /* ECANCELED if unread jump is pending. */ 536af93fea7SJake Freeland if (tfd->tfd_jumped == TFD_CANCELED) 537af93fea7SJake Freeland error = ECANCELED; 538af93fea7SJake Freeland /* Convert from CLOCK_REALTIME to CLOCK_BOOTTIME. */ 539af93fea7SJake Freeland timespecsub(&tfd->tfd_time.it_value, &tfd->tfd_boottim, 540af93fea7SJake Freeland &tfd->tfd_time.it_value); 541af93fea7SJake Freeland } 542af93fea7SJake Freeland callout_reset_sbt(&tfd->tfd_callout, 543af93fea7SJake Freeland tstosbt(tfd->tfd_time.it_value), 544af93fea7SJake Freeland 0, timerfd_expire, tfd, C_ABSOLUTE); 545af93fea7SJake Freeland } else { 546af93fea7SJake Freeland callout_stop(&tfd->tfd_callout); 547af93fea7SJake Freeland } 548af93fea7SJake Freeland tfd->tfd_count = 0; 549af93fea7SJake Freeland tfd->tfd_expired = false; 550af93fea7SJake Freeland tfd->tfd_jumped = TFD_NOJUMP; 551af93fea7SJake Freeland mtx_unlock(&tfd->tfd_lock); 552af93fea7SJake Freeland 553af93fea7SJake Freeland fdrop(fp, td); 554af93fea7SJake Freeland return (error); 555af93fea7SJake Freeland } 556af93fea7SJake Freeland 557af93fea7SJake Freeland int 558af93fea7SJake Freeland sys_timerfd_create(struct thread *td, struct timerfd_create_args *uap) 559af93fea7SJake Freeland { 560af93fea7SJake Freeland return (kern_timerfd_create(td, uap->clockid, uap->flags)); 561af93fea7SJake Freeland } 562af93fea7SJake Freeland 563af93fea7SJake Freeland int 564af93fea7SJake Freeland sys_timerfd_gettime(struct thread *td, struct timerfd_gettime_args *uap) 565af93fea7SJake Freeland { 566af93fea7SJake Freeland struct itimerspec curr_value; 567af93fea7SJake Freeland int error; 568af93fea7SJake Freeland 569af93fea7SJake Freeland error = kern_timerfd_gettime(td, uap->fd, &curr_value); 570af93fea7SJake Freeland if (error == 0) 571af93fea7SJake Freeland error = copyout(&curr_value, uap->curr_value, 572af93fea7SJake Freeland sizeof(curr_value)); 573af93fea7SJake Freeland 574af93fea7SJake Freeland return (error); 575af93fea7SJake Freeland } 576af93fea7SJake Freeland 577af93fea7SJake Freeland int 578af93fea7SJake Freeland sys_timerfd_settime(struct thread *td, struct timerfd_settime_args *uap) 579af93fea7SJake Freeland { 580af93fea7SJake Freeland struct itimerspec new_value, old_value; 581af93fea7SJake Freeland int error; 582af93fea7SJake Freeland 583af93fea7SJake Freeland error = copyin(uap->new_value, &new_value, sizeof(new_value)); 584af93fea7SJake Freeland if (error != 0) 585af93fea7SJake Freeland return (error); 586af93fea7SJake Freeland if (uap->old_value == NULL) { 587af93fea7SJake Freeland error = kern_timerfd_settime(td, uap->fd, uap->flags, 588af93fea7SJake Freeland &new_value, NULL); 589af93fea7SJake Freeland } else { 590af93fea7SJake Freeland error = kern_timerfd_settime(td, uap->fd, uap->flags, 591af93fea7SJake Freeland &new_value, &old_value); 592af93fea7SJake Freeland if (error == 0) 593af93fea7SJake Freeland error = copyout(&old_value, uap->old_value, 594af93fea7SJake Freeland sizeof(old_value)); 595af93fea7SJake Freeland } 596af93fea7SJake Freeland return (error); 597af93fea7SJake Freeland } 598