diff --git a/sys/compat/linux/linux_emul.c b/sys/compat/linux/linux_emul.c index 1dfbe239ccc4..499bebe8926a 100644 --- a/sys/compat/linux/linux_emul.c +++ b/sys/compat/linux/linux_emul.c @@ -1,404 +1,383 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 1994-1996 Søren Schmidt * Copyright (c) 2006 Roman Divacky * Copyright (c) 2013 Dmitry Chagin * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if BYTE_ORDER == LITTLE_ENDIAN #define SHELLMAGIC 0x2123 /* #! */ #else #define SHELLMAGIC 0x2321 #endif /* * This returns reference to the thread emuldata entry (if found) * * Hold PROC_LOCK when referencing emuldata from other threads. */ struct linux_emuldata * em_find(struct thread *td) { struct linux_emuldata *em; em = td->td_emuldata; return (em); } /* * This returns reference to the proc pemuldata entry (if found) * * Hold PROC_LOCK when referencing proc pemuldata from other threads. * Hold LINUX_PEM_LOCK wher referencing pemuldata members. */ struct linux_pemuldata * pem_find(struct proc *p) { struct linux_pemuldata *pem; pem = p->p_emuldata; return (pem); } /* * Linux apps generally expect the soft open file limit to be set * to 1024, often iterating over all the file descriptors up to that * limit instead of using closefrom(2). Give them what they want, * unless there already is a resource limit in place. */ static void linux_set_default_openfiles(struct thread *td, struct proc *p) { struct rlimit rlim; int error; if (linux_default_openfiles < 0) return; PROC_LOCK(p); lim_rlimit_proc(p, RLIMIT_NOFILE, &rlim); PROC_UNLOCK(p); if (rlim.rlim_cur != rlim.rlim_max || rlim.rlim_cur <= linux_default_openfiles) return; rlim.rlim_cur = linux_default_openfiles; error = kern_proc_setrlimit(td, p, RLIMIT_NOFILE, &rlim); KASSERT(error == 0, ("kern_proc_setrlimit failed")); } /* * The default stack size limit in Linux is 8MB. */ static void linux_set_default_stacksize(struct thread *td, struct proc *p) { struct rlimit rlim; int error; if (linux_default_stacksize < 0) return; PROC_LOCK(p); lim_rlimit_proc(p, RLIMIT_STACK, &rlim); PROC_UNLOCK(p); if (rlim.rlim_cur != rlim.rlim_max || rlim.rlim_cur <= linux_default_stacksize) return; rlim.rlim_cur = linux_default_stacksize; error = kern_proc_setrlimit(td, p, RLIMIT_STACK, &rlim); KASSERT(error == 0, ("kern_proc_setrlimit failed")); } void linux_proc_init(struct thread *td, struct thread *newtd, int flags) { struct linux_emuldata *em; struct linux_pemuldata *pem; - struct epoll_emuldata *emd; struct proc *p; if (newtd != NULL) { p = newtd->td_proc; /* non-exec call */ em = malloc(sizeof(*em), M_TEMP, M_WAITOK | M_ZERO); if (flags & LINUX_CLONE_THREAD) { LINUX_CTR1(proc_init, "thread newtd(%d)", newtd->td_tid); em->em_tid = newtd->td_tid; } else { LINUX_CTR1(proc_init, "fork newtd(%d)", p->p_pid); em->em_tid = p->p_pid; pem = malloc(sizeof(*pem), M_LINUX, M_WAITOK | M_ZERO); sx_init(&pem->pem_sx, "lpemlk"); p->p_emuldata = pem; } newtd->td_emuldata = em; linux_set_default_openfiles(td, p); linux_set_default_stacksize(td, p); } else { p = td->td_proc; /* exec */ LINUX_CTR1(proc_init, "exec newtd(%d)", p->p_pid); /* lookup the old one */ em = em_find(td); KASSERT(em != NULL, ("proc_init: emuldata not found in exec case.\n")); em->em_tid = p->p_pid; em->flags = 0; em->robust_futexes = NULL; em->child_clear_tid = NULL; em->child_set_tid = NULL; - /* epoll should be destroyed in a case of exec. */ pem = pem_find(p); KASSERT(pem != NULL, ("proc_exit: proc emuldata not found.\n")); pem->persona = 0; - if (pem->epoll != NULL) { - emd = pem->epoll; - pem->epoll = NULL; - free(emd, M_EPOLL); - } } } void linux_on_exit(struct proc *p) { struct linux_pemuldata *pem; - struct epoll_emuldata *emd; struct thread *td = curthread; MPASS(SV_CURPROC_ABI() == SV_ABI_LINUX); LINUX_CTR3(proc_exit, "thread(%d) proc(%d) p %p", td->td_tid, p->p_pid, p); pem = pem_find(p); if (pem == NULL) return; (p->p_sysent->sv_thread_detach)(td); p->p_emuldata = NULL; - if (pem->epoll != NULL) { - emd = pem->epoll; - pem->epoll = NULL; - free(emd, M_EPOLL); - } - sx_destroy(&pem->pem_sx); free(pem, M_LINUX); } /* * If a Linux binary is exec'ing something, try this image activator * first. We override standard shell script execution in order to * be able to modify the interpreter path. We only do this if a Linux * binary is doing the exec, so we do not create an EXEC module for it. */ int linux_exec_imgact_try(struct image_params *imgp) { const char *head = (const char *)imgp->image_header; char *rpath; int error = -1; /* * The interpreter for shell scripts run from a Linux binary needs * to be located in /compat/linux if possible in order to recursively * maintain Linux path emulation. */ if (((const short *)head)[0] == SHELLMAGIC) { /* * Run our normal shell image activator. If it succeeds attempt * to use the alternate path for the interpreter. If an * alternate path is found, use our stringspace to store it. */ if ((error = exec_shell_imgact(imgp)) == 0) { linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc), imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0, AT_FDCWD); if (rpath != NULL) imgp->args->fname_buf = imgp->interpreter_name = rpath; } } return (error); } int linux_common_execve(struct thread *td, struct image_args *eargs) { struct linux_pemuldata *pem; - struct epoll_emuldata *emd; struct vmspace *oldvmspace; struct linux_emuldata *em; struct proc *p; int error; p = td->td_proc; error = pre_execve(td, &oldvmspace); if (error != 0) return (error); error = kern_execve(td, eargs, NULL, oldvmspace); post_execve(td, error, oldvmspace); if (error != EJUSTRETURN) return (error); /* * In a case of transition from Linux binary execing to * FreeBSD binary we destroy Linux emuldata thread & proc entries. */ if (SV_CURPROC_ABI() != SV_ABI_LINUX) { PROC_LOCK(p); em = em_find(td); KASSERT(em != NULL, ("proc_exec: thread emuldata not found.\n")); td->td_emuldata = NULL; pem = pem_find(p); KASSERT(pem != NULL, ("proc_exec: proc pemuldata not found.\n")); p->p_emuldata = NULL; PROC_UNLOCK(p); - if (pem->epoll != NULL) { - emd = pem->epoll; - pem->epoll = NULL; - free(emd, M_EPOLL); - } - free(em, M_TEMP); free(pem, M_LINUX); } return (EJUSTRETURN); } void linux_on_exec(struct proc *p, struct image_params *imgp) { struct thread *td; struct thread *othertd; #if defined(__amd64__) struct linux_pemuldata *pem; #endif td = curthread; MPASS((imgp->sysent->sv_flags & SV_ABI_MASK) == SV_ABI_LINUX); /* * When execing to Linux binary, we create Linux emuldata * thread entry. */ if (SV_PROC_ABI(p) == SV_ABI_LINUX) { /* * Process already was under Linuxolator * before exec. Update emuldata to reflect * single-threaded cleaned state after exec. */ linux_proc_init(td, NULL, 0); } else { /* * We are switching the process to Linux emulator. */ linux_proc_init(td, td, 0); /* * Create a transient td_emuldata for all suspended * threads, so that p->p_sysent->sv_thread_detach() == * linux_thread_detach() can find expected but unused * emuldata. */ FOREACH_THREAD_IN_PROC(td->td_proc, othertd) { if (othertd == td) continue; linux_proc_init(td, othertd, LINUX_CLONE_THREAD); } } #if defined(__amd64__) /* * An IA32 executable which has executable stack will have the * READ_IMPLIES_EXEC personality flag set automatically. */ if (SV_PROC_FLAG(td->td_proc, SV_ILP32) && imgp->stack_prot & VM_PROT_EXECUTE) { pem = pem_find(p); pem->persona |= LINUX_READ_IMPLIES_EXEC; } #endif } void linux_thread_dtor(struct thread *td) { struct linux_emuldata *em; em = em_find(td); if (em == NULL) return; td->td_emuldata = NULL; LINUX_CTR1(thread_dtor, "thread(%d)", em->em_tid); free(em, M_TEMP); } void linux_schedtail(struct thread *td) { struct linux_emuldata *em; struct proc *p; int error = 0; int *child_set_tid; p = td->td_proc; em = em_find(td); KASSERT(em != NULL, ("linux_schedtail: thread emuldata not found.\n")); child_set_tid = em->child_set_tid; if (child_set_tid != NULL) { error = copyout(&em->em_tid, child_set_tid, sizeof(em->em_tid)); LINUX_CTR4(schedtail, "thread(%d) %p stored %d error %d", td->td_tid, child_set_tid, em->em_tid, error); } else LINUX_CTR1(schedtail, "thread(%d)", em->em_tid); } diff --git a/sys/compat/linux/linux_emul.h b/sys/compat/linux/linux_emul.h index 1bbc69ad98be..de66a7a4c82a 100644 --- a/sys/compat/linux/linux_emul.h +++ b/sys/compat/linux/linux_emul.h @@ -1,83 +1,82 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2006 Roman Divacky * Copyright (c) 2013 Dmitry Chagin * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _LINUX_EMUL_H_ #define _LINUX_EMUL_H_ struct image_params; /* * modeled after similar structure in NetBSD * this will be extended as we need more functionality */ struct linux_emuldata { int *child_set_tid; /* in clone(): Child's TID to set on clone */ int *child_clear_tid;/* in clone(): Child's TID to clear on exit */ int flags; /* thread emuldata flags */ int em_tid; /* thread id */ struct linux_robust_list_head *robust_futexes; }; struct linux_emuldata *em_find(struct thread *); int linux_exec_imgact_try(struct image_params *); void linux_proc_init(struct thread *, struct thread *, int); void linux_on_exit(struct proc *); void linux_schedtail(struct thread *); void linux_on_exec(struct proc *, struct image_params *); void linux_thread_dtor(struct thread *); void linux_thread_detach(struct thread *); int linux_common_execve(struct thread *, struct image_args *); /* process emuldata flags */ #define LINUX_XDEPR_REQUEUEOP 0x00000001 /* uses deprecated futex REQUEUE op*/ #define LINUX_XUNSUP_EPOLL 0x00000002 /* unsupported epoll events */ #define LINUX_XUNSUP_FUTEXPIOP 0x00000004 /* uses unsupported pi futex */ struct linux_pemuldata { uint32_t flags; /* process emuldata flags */ struct sx pem_sx; /* lock for this struct */ - void *epoll; /* epoll data */ uint32_t persona; /* process execution domain */ uint32_t ptrace_flags; /* used by ptrace(2) */ }; #define LINUX_PEM_XLOCK(p) sx_xlock(&(p)->pem_sx) #define LINUX_PEM_XUNLOCK(p) sx_xunlock(&(p)->pem_sx) #define LINUX_PEM_SLOCK(p) sx_slock(&(p)->pem_sx) #define LINUX_PEM_SUNLOCK(p) sx_sunlock(&(p)->pem_sx) struct linux_pemuldata *pem_find(struct proc *); #endif /* !_LINUX_EMUL_H_ */ diff --git a/sys/compat/linux/linux_event.c b/sys/compat/linux/linux_event.c index 54f6b083adf3..370bf4d6485f 100644 --- a/sys/compat/linux/linux_event.c +++ b/sys/compat/linux/linux_event.c @@ -1,1049 +1,983 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2007 Roman Divacky * Copyright (c) 2014 Dmitry Chagin * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef COMPAT_LINUX32 #include #include #else #include #include #endif #include #include #include #include #include -/* - * epoll defines 'struct epoll_event' with the field 'data' as 64 bits - * on all architectures. But on 32 bit architectures BSD 'struct kevent' only - * has 32 bit opaque pointer as 'udata' field. So we can't pass epoll supplied - * data verbatuim. Therefore we allocate 64-bit memory block to pass - * user supplied data for every file descriptor. - */ - typedef uint64_t epoll_udata_t; -struct epoll_emuldata { - uint32_t fdc; /* epoll udata max index */ - epoll_udata_t udata[1]; /* epoll user data vector */ -}; - -#define EPOLL_DEF_SZ 16 -#define EPOLL_SIZE(fdn) \ - (sizeof(struct epoll_emuldata)+(fdn) * sizeof(epoll_udata_t)) - struct epoll_event { uint32_t events; epoll_udata_t data; } #if defined(__amd64__) __attribute__((packed)) #endif ; #define LINUX_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) -static void epoll_fd_install(struct thread *td, int fd, epoll_udata_t udata); static int epoll_to_kevent(struct thread *td, int fd, struct epoll_event *l_event, struct kevent *kevent, int *nkevents); static void kevent_to_epoll(struct kevent *kevent, struct epoll_event *l_event); static int epoll_kev_copyout(void *arg, struct kevent *kevp, int count); static int epoll_kev_copyin(void *arg, struct kevent *kevp, int count); static int epoll_register_kevent(struct thread *td, struct file *epfp, int fd, int filter, unsigned int flags); static int epoll_fd_registered(struct thread *td, struct file *epfp, int fd); static int epoll_delete_all_events(struct thread *td, struct file *epfp, int fd); struct epoll_copyin_args { struct kevent *changelist; }; struct epoll_copyout_args { struct epoll_event *leventlist; struct proc *p; uint32_t count; int error; }; /* timerfd */ typedef uint64_t timerfd_t; static fo_rdwr_t timerfd_read; static fo_ioctl_t timerfd_ioctl; static fo_poll_t timerfd_poll; static fo_kqfilter_t timerfd_kqfilter; static fo_stat_t timerfd_stat; static fo_close_t timerfd_close; static fo_fill_kinfo_t timerfd_fill_kinfo; static struct fileops timerfdops = { .fo_read = timerfd_read, .fo_write = invfo_rdwr, .fo_truncate = invfo_truncate, .fo_ioctl = timerfd_ioctl, .fo_poll = timerfd_poll, .fo_kqfilter = timerfd_kqfilter, .fo_stat = timerfd_stat, .fo_close = timerfd_close, .fo_chmod = invfo_chmod, .fo_chown = invfo_chown, .fo_sendfile = invfo_sendfile, .fo_fill_kinfo = timerfd_fill_kinfo, .fo_flags = DFLAG_PASSABLE }; static void filt_timerfddetach(struct knote *kn); static int filt_timerfdread(struct knote *kn, long hint); static struct filterops timerfd_rfiltops = { .f_isfd = 1, .f_detach = filt_timerfddetach, .f_event = filt_timerfdread }; struct timerfd { clockid_t tfd_clockid; struct itimerspec tfd_time; struct callout tfd_callout; timerfd_t tfd_count; bool tfd_canceled; struct selinfo tfd_sel; struct mtx tfd_lock; }; static void linux_timerfd_expire(void *); static void linux_timerfd_curval(struct timerfd *, struct itimerspec *); -static void -epoll_fd_install(struct thread *td, int fd, epoll_udata_t udata) -{ - struct linux_pemuldata *pem; - struct epoll_emuldata *emd; - struct proc *p; - - p = td->td_proc; - - pem = pem_find(p); - KASSERT(pem != NULL, ("epoll proc emuldata not found.\n")); - - LINUX_PEM_XLOCK(pem); - if (pem->epoll == NULL) { - emd = malloc(EPOLL_SIZE(fd), M_EPOLL, M_WAITOK); - emd->fdc = fd; - pem->epoll = emd; - } else { - emd = pem->epoll; - if (fd > emd->fdc) { - emd = realloc(emd, EPOLL_SIZE(fd), M_EPOLL, M_WAITOK); - emd->fdc = fd; - pem->epoll = emd; - } - } - emd->udata[fd] = udata; - LINUX_PEM_XUNLOCK(pem); -} - static int epoll_create_common(struct thread *td, int flags) { - int error; - - error = kern_kqueue(td, flags, NULL); - if (error != 0) - return (error); - - epoll_fd_install(td, EPOLL_DEF_SZ, 0); - return (0); + return (kern_kqueue(td, flags, NULL)); } #ifdef LINUX_LEGACY_SYSCALLS int linux_epoll_create(struct thread *td, struct linux_epoll_create_args *args) { /* * args->size is unused. Linux just tests it * and then forgets it as well. */ if (args->size <= 0) return (EINVAL); return (epoll_create_common(td, 0)); } #endif int linux_epoll_create1(struct thread *td, struct linux_epoll_create1_args *args) { int flags; if ((args->flags & ~(LINUX_O_CLOEXEC)) != 0) return (EINVAL); flags = 0; if ((args->flags & LINUX_O_CLOEXEC) != 0) flags |= O_CLOEXEC; return (epoll_create_common(td, flags)); } /* Structure converting function from epoll to kevent. */ static int epoll_to_kevent(struct thread *td, int fd, struct epoll_event *l_event, struct kevent *kevent, int *nkevents) { uint32_t levents = l_event->events; struct linux_pemuldata *pem; struct proc *p; unsigned short kev_flags = EV_ADD | EV_ENABLE; /* flags related to how event is registered */ if ((levents & LINUX_EPOLLONESHOT) != 0) kev_flags |= EV_DISPATCH; if ((levents & LINUX_EPOLLET) != 0) kev_flags |= EV_CLEAR; if ((levents & LINUX_EPOLLERR) != 0) kev_flags |= EV_ERROR; if ((levents & LINUX_EPOLLRDHUP) != 0) kev_flags |= EV_EOF; /* flags related to what event is registered */ if ((levents & LINUX_EPOLL_EVRD) != 0) { - EV_SET(kevent++, fd, EVFILT_READ, kev_flags, 0, 0, 0); + EV_SET(kevent, fd, EVFILT_READ, kev_flags, 0, 0, 0); + kevent->ext[0] = l_event->data; + ++kevent; ++(*nkevents); } if ((levents & LINUX_EPOLL_EVWR) != 0) { - EV_SET(kevent++, fd, EVFILT_WRITE, kev_flags, 0, 0, 0); + EV_SET(kevent, fd, EVFILT_WRITE, kev_flags, 0, 0, 0); + kevent->ext[0] = l_event->data; + ++kevent; ++(*nkevents); } /* zero event mask is legal */ if ((levents & (LINUX_EPOLL_EVRD | LINUX_EPOLL_EVWR)) == 0) { EV_SET(kevent++, fd, EVFILT_READ, EV_ADD|EV_DISABLE, 0, 0, 0); ++(*nkevents); } if ((levents & ~(LINUX_EPOLL_EVSUP)) != 0) { p = td->td_proc; pem = pem_find(p); KASSERT(pem != NULL, ("epoll proc emuldata not found.\n")); - KASSERT(pem->epoll != NULL, ("epoll proc epolldata not found.\n")); LINUX_PEM_XLOCK(pem); if ((pem->flags & LINUX_XUNSUP_EPOLL) == 0) { pem->flags |= LINUX_XUNSUP_EPOLL; LINUX_PEM_XUNLOCK(pem); linux_msg(td, "epoll_ctl unsupported flags: 0x%x", levents); } else LINUX_PEM_XUNLOCK(pem); return (EINVAL); } return (0); } /* * Structure converting function from kevent to epoll. In a case * this is called on error in registration we store the error in * event->data and pick it up later in linux_epoll_ctl(). */ static void kevent_to_epoll(struct kevent *kevent, struct epoll_event *l_event) { + l_event->data = kevent->ext[0]; + if ((kevent->flags & EV_ERROR) != 0) { l_event->events = LINUX_EPOLLERR; return; } /* XXX EPOLLPRI, EPOLLHUP */ switch (kevent->filter) { case EVFILT_READ: l_event->events = LINUX_EPOLLIN; if ((kevent->flags & EV_EOF) != 0) l_event->events |= LINUX_EPOLLRDHUP; break; case EVFILT_WRITE: l_event->events = LINUX_EPOLLOUT; break; } } /* * Copyout callback used by kevent. This converts kevent * events to epoll events and copies them back to the * userspace. This is also called on error on registering * of the filter. */ static int epoll_kev_copyout(void *arg, struct kevent *kevp, int count) { struct epoll_copyout_args *args; - struct linux_pemuldata *pem; - struct epoll_emuldata *emd; struct epoll_event *eep; - int error, fd, i; + int error, i; args = (struct epoll_copyout_args*) arg; eep = malloc(sizeof(*eep) * count, M_EPOLL, M_WAITOK | M_ZERO); - pem = pem_find(args->p); - KASSERT(pem != NULL, ("epoll proc emuldata not found.\n")); - LINUX_PEM_SLOCK(pem); - emd = pem->epoll; - KASSERT(emd != NULL, ("epoll proc epolldata not found.\n")); - - for (i = 0; i < count; i++) { + for (i = 0; i < count; i++) kevent_to_epoll(&kevp[i], &eep[i]); - fd = kevp[i].ident; - KASSERT(fd <= emd->fdc, ("epoll user data vector" - " is too small.\n")); - eep[i].data = emd->udata[fd]; - } - LINUX_PEM_SUNLOCK(pem); - error = copyout(eep, args->leventlist, count * sizeof(*eep)); if (error == 0) { args->leventlist += count; args->count += count; } else if (args->error == 0) args->error = error; free(eep, M_EPOLL); return (error); } /* * Copyin callback used by kevent. This copies already * converted filters from kernel memory to the kevent * internal kernel memory. Hence the memcpy instead of * copyin. */ static int epoll_kev_copyin(void *arg, struct kevent *kevp, int count) { struct epoll_copyin_args *args; args = (struct epoll_copyin_args*) arg; memcpy(kevp, args->changelist, count * sizeof(*kevp)); args->changelist += count; return (0); } /* * Load epoll filter, convert it to kevent filter * and load it into kevent subsystem. */ int linux_epoll_ctl(struct thread *td, struct linux_epoll_ctl_args *args) { struct file *epfp, *fp; struct epoll_copyin_args ciargs; struct kevent kev[2]; struct kevent_copyops k_ops = { &ciargs, NULL, epoll_kev_copyin}; struct epoll_event le; cap_rights_t rights; int nchanges = 0; int error; if (args->op != LINUX_EPOLL_CTL_DEL) { error = copyin(args->event, &le, sizeof(le)); if (error != 0) return (error); } error = fget(td, args->epfd, cap_rights_init_one(&rights, CAP_KQUEUE_CHANGE), &epfp); if (error != 0) return (error); if (epfp->f_type != DTYPE_KQUEUE) { error = EINVAL; goto leave1; } /* Protect user data vector from incorrectly supplied fd. */ error = fget(td, args->fd, cap_rights_init_one(&rights, CAP_POLL_EVENT), &fp); if (error != 0) goto leave1; /* Linux disallows spying on himself */ if (epfp == fp) { error = EINVAL; goto leave0; } ciargs.changelist = kev; if (args->op != LINUX_EPOLL_CTL_DEL) { error = epoll_to_kevent(td, args->fd, &le, kev, &nchanges); if (error != 0) goto leave0; } switch (args->op) { case LINUX_EPOLL_CTL_MOD: error = epoll_delete_all_events(td, epfp, args->fd); if (error != 0) goto leave0; break; case LINUX_EPOLL_CTL_ADD: if (epoll_fd_registered(td, epfp, args->fd)) { error = EEXIST; goto leave0; } break; case LINUX_EPOLL_CTL_DEL: /* CTL_DEL means unregister this fd with this epoll */ error = epoll_delete_all_events(td, epfp, args->fd); goto leave0; default: error = EINVAL; goto leave0; } - epoll_fd_install(td, args->fd, le.data); - error = kern_kevent_fp(td, epfp, nchanges, 0, &k_ops, NULL); leave0: fdrop(fp, td); leave1: fdrop(epfp, td); return (error); } /* * Wait for a filter to be triggered on the epoll file descriptor. */ static int linux_epoll_wait_common(struct thread *td, int epfd, struct epoll_event *events, int maxevents, int timeout, sigset_t *uset) { struct epoll_copyout_args coargs; struct kevent_copyops k_ops = { &coargs, epoll_kev_copyout, NULL}; struct timespec ts, *tsp; cap_rights_t rights; struct file *epfp; sigset_t omask; int error; if (maxevents <= 0 || maxevents > LINUX_MAX_EVENTS) return (EINVAL); error = fget(td, epfd, cap_rights_init_one(&rights, CAP_KQUEUE_EVENT), &epfp); if (error != 0) return (error); if (epfp->f_type != DTYPE_KQUEUE) { error = EINVAL; goto leave; } if (uset != NULL) { error = kern_sigprocmask(td, SIG_SETMASK, uset, &omask, 0); if (error != 0) goto leave; td->td_pflags |= TDP_OLDMASK; /* * Make sure that ast() is called on return to * usermode and TDP_OLDMASK is cleared, restoring old * sigmask. */ thread_lock(td); td->td_flags |= TDF_ASTPENDING; thread_unlock(td); } coargs.leventlist = events; coargs.p = td->td_proc; coargs.count = 0; coargs.error = 0; /* * Linux epoll_wait(2) man page states that timeout of -1 causes caller * to block indefinitely. Real implementation does it if any negative * timeout value is passed. */ if (timeout >= 0) { /* Convert from milliseconds to timespec. */ ts.tv_sec = timeout / 1000; ts.tv_nsec = (timeout % 1000) * 1000000; tsp = &ts; } else { tsp = NULL; } error = kern_kevent_fp(td, epfp, 0, maxevents, &k_ops, tsp); if (error == 0 && coargs.error != 0) error = coargs.error; /* * kern_kevent might return ENOMEM which is not expected from epoll_wait. * Maybe we should translate that but I don't think it matters at all. */ if (error == 0) td->td_retval[0] = coargs.count; if (uset != NULL) error = kern_sigprocmask(td, SIG_SETMASK, &omask, NULL, 0); leave: fdrop(epfp, td); return (error); } #ifdef LINUX_LEGACY_SYSCALLS int linux_epoll_wait(struct thread *td, struct linux_epoll_wait_args *args) { return (linux_epoll_wait_common(td, args->epfd, args->events, args->maxevents, args->timeout, NULL)); } #endif int linux_epoll_pwait(struct thread *td, struct linux_epoll_pwait_args *args) { sigset_t mask, *pmask; l_sigset_t lmask; int error; if (args->mask != NULL) { if (args->sigsetsize != sizeof(l_sigset_t)) return (EINVAL); error = copyin(args->mask, &lmask, sizeof(l_sigset_t)); if (error != 0) return (error); linux_to_bsd_sigset(&lmask, &mask); pmask = &mask; } else pmask = NULL; return (linux_epoll_wait_common(td, args->epfd, args->events, args->maxevents, args->timeout, pmask)); } static int epoll_register_kevent(struct thread *td, struct file *epfp, int fd, int filter, unsigned int flags) { struct epoll_copyin_args ciargs; struct kevent kev; struct kevent_copyops k_ops = { &ciargs, NULL, epoll_kev_copyin}; ciargs.changelist = &kev; EV_SET(&kev, fd, filter, flags, 0, 0, 0); return (kern_kevent_fp(td, epfp, 1, 0, &k_ops, NULL)); } static int epoll_fd_registered(struct thread *td, struct file *epfp, int fd) { /* * Set empty filter flags to avoid accidental modification of already * registered events. In the case of event re-registration: * 1. If event does not exists kevent() does nothing and returns ENOENT * 2. If event does exists, it's enabled/disabled state is preserved * but fflags, data and udata fields are overwritten. So we can not * set socket lowats and store user's context pointer in udata. */ if (epoll_register_kevent(td, epfp, fd, EVFILT_READ, 0) != ENOENT || epoll_register_kevent(td, epfp, fd, EVFILT_WRITE, 0) != ENOENT) return (1); return (0); } static int epoll_delete_all_events(struct thread *td, struct file *epfp, int fd) { int error1, error2; error1 = epoll_register_kevent(td, epfp, fd, EVFILT_READ, EV_DELETE); error2 = epoll_register_kevent(td, epfp, fd, EVFILT_WRITE, EV_DELETE); /* return 0 if at least one result positive */ return (error1 == 0 ? 0 : error2); } #ifdef LINUX_LEGACY_SYSCALLS int linux_eventfd(struct thread *td, struct linux_eventfd_args *args) { struct specialfd_eventfd ae; bzero(&ae, sizeof(ae)); ae.initval = args->initval; return (kern_specialfd(td, SPECIALFD_EVENTFD, &ae)); } #endif int linux_eventfd2(struct thread *td, struct linux_eventfd2_args *args) { struct specialfd_eventfd ae; int flags; if ((args->flags & ~(LINUX_O_CLOEXEC | LINUX_O_NONBLOCK | LINUX_EFD_SEMAPHORE)) != 0) return (EINVAL); flags = 0; if ((args->flags & LINUX_O_CLOEXEC) != 0) flags |= EFD_CLOEXEC; if ((args->flags & LINUX_O_NONBLOCK) != 0) flags |= EFD_NONBLOCK; if ((args->flags & LINUX_EFD_SEMAPHORE) != 0) flags |= EFD_SEMAPHORE; bzero(&ae, sizeof(ae)); ae.flags = flags; ae.initval = args->initval; return (kern_specialfd(td, SPECIALFD_EVENTFD, &ae)); } int linux_timerfd_create(struct thread *td, struct linux_timerfd_create_args *args) { struct filedesc *fdp; struct timerfd *tfd; struct file *fp; clockid_t clockid; int fflags, fd, error; if ((args->flags & ~LINUX_TFD_CREATE_FLAGS) != 0) return (EINVAL); error = linux_to_native_clockid(&clockid, args->clockid); if (error != 0) return (error); if (clockid != CLOCK_REALTIME && clockid != CLOCK_MONOTONIC) return (EINVAL); fflags = 0; if ((args->flags & LINUX_TFD_CLOEXEC) != 0) fflags |= O_CLOEXEC; fdp = td->td_proc->p_fd; error = falloc(td, &fp, &fd, fflags); if (error != 0) return (error); tfd = malloc(sizeof(*tfd), M_EPOLL, M_WAITOK | M_ZERO); tfd->tfd_clockid = clockid; mtx_init(&tfd->tfd_lock, "timerfd", NULL, MTX_DEF); callout_init_mtx(&tfd->tfd_callout, &tfd->tfd_lock, 0); knlist_init_mtx(&tfd->tfd_sel.si_note, &tfd->tfd_lock); fflags = FREAD; if ((args->flags & LINUX_O_NONBLOCK) != 0) fflags |= FNONBLOCK; finit(fp, fflags, DTYPE_LINUXTFD, tfd, &timerfdops); fdrop(fp, td); td->td_retval[0] = fd; return (error); } static int timerfd_close(struct file *fp, struct thread *td) { struct timerfd *tfd; tfd = fp->f_data; if (fp->f_type != DTYPE_LINUXTFD || tfd == NULL) return (EINVAL); timespecclear(&tfd->tfd_time.it_value); timespecclear(&tfd->tfd_time.it_interval); mtx_lock(&tfd->tfd_lock); callout_drain(&tfd->tfd_callout); mtx_unlock(&tfd->tfd_lock); seldrain(&tfd->tfd_sel); knlist_destroy(&tfd->tfd_sel.si_note); fp->f_ops = &badfileops; mtx_destroy(&tfd->tfd_lock); free(tfd, M_EPOLL); return (0); } static int timerfd_read(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { struct timerfd *tfd; timerfd_t count; int error; tfd = fp->f_data; if (fp->f_type != DTYPE_LINUXTFD || tfd == NULL) return (EINVAL); if (uio->uio_resid < sizeof(timerfd_t)) return (EINVAL); error = 0; mtx_lock(&tfd->tfd_lock); retry: if (tfd->tfd_canceled) { tfd->tfd_count = 0; mtx_unlock(&tfd->tfd_lock); return (ECANCELED); } if (tfd->tfd_count == 0) { if ((fp->f_flag & FNONBLOCK) != 0) { mtx_unlock(&tfd->tfd_lock); return (EAGAIN); } error = mtx_sleep(&tfd->tfd_count, &tfd->tfd_lock, PCATCH, "ltfdrd", 0); if (error == 0) goto retry; } if (error == 0) { count = tfd->tfd_count; tfd->tfd_count = 0; mtx_unlock(&tfd->tfd_lock); error = uiomove(&count, sizeof(timerfd_t), uio); } else mtx_unlock(&tfd->tfd_lock); return (error); } static int timerfd_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { struct timerfd *tfd; int revents = 0; tfd = fp->f_data; if (fp->f_type != DTYPE_LINUXTFD || tfd == NULL) return (POLLERR); mtx_lock(&tfd->tfd_lock); if ((events & (POLLIN|POLLRDNORM)) && tfd->tfd_count > 0) revents |= events & (POLLIN|POLLRDNORM); if (revents == 0) selrecord(td, &tfd->tfd_sel); mtx_unlock(&tfd->tfd_lock); return (revents); } static int timerfd_kqfilter(struct file *fp, struct knote *kn) { struct timerfd *tfd; tfd = fp->f_data; if (fp->f_type != DTYPE_LINUXTFD || tfd == NULL) return (EINVAL); if (kn->kn_filter == EVFILT_READ) kn->kn_fop = &timerfd_rfiltops; else return (EINVAL); kn->kn_hook = tfd; knlist_add(&tfd->tfd_sel.si_note, kn, 0); return (0); } static void filt_timerfddetach(struct knote *kn) { struct timerfd *tfd = kn->kn_hook; mtx_lock(&tfd->tfd_lock); knlist_remove(&tfd->tfd_sel.si_note, kn, 1); mtx_unlock(&tfd->tfd_lock); } static int filt_timerfdread(struct knote *kn, long hint) { struct timerfd *tfd = kn->kn_hook; return (tfd->tfd_count > 0); } static int timerfd_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred, struct thread *td) { if (fp->f_data == NULL || fp->f_type != DTYPE_LINUXTFD) return (EINVAL); switch (cmd) { case FIONBIO: case FIOASYNC: return (0); } return (ENOTTY); } static int timerfd_stat(struct file *fp, struct stat *st, struct ucred *active_cred, struct thread *td) { return (ENXIO); } static int timerfd_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { kif->kf_type = KF_TYPE_UNKNOWN; return (0); } static void linux_timerfd_clocktime(struct timerfd *tfd, struct timespec *ts) { if (tfd->tfd_clockid == CLOCK_REALTIME) getnanotime(ts); else /* CLOCK_MONOTONIC */ getnanouptime(ts); } static void linux_timerfd_curval(struct timerfd *tfd, struct itimerspec *ots) { struct timespec cts; linux_timerfd_clocktime(tfd, &cts); *ots = tfd->tfd_time; if (ots->it_value.tv_sec != 0 || ots->it_value.tv_nsec != 0) { timespecsub(&ots->it_value, &cts, &ots->it_value); if (ots->it_value.tv_sec < 0 || (ots->it_value.tv_sec == 0 && ots->it_value.tv_nsec == 0)) { ots->it_value.tv_sec = 0; ots->it_value.tv_nsec = 1; } } } int linux_timerfd_gettime(struct thread *td, struct linux_timerfd_gettime_args *args) { struct l_itimerspec lots; struct itimerspec ots; struct timerfd *tfd; struct file *fp; int error; error = fget(td, args->fd, &cap_read_rights, &fp); if (error != 0) return (error); tfd = fp->f_data; if (fp->f_type != DTYPE_LINUXTFD || tfd == NULL) { error = EINVAL; goto out; } mtx_lock(&tfd->tfd_lock); linux_timerfd_curval(tfd, &ots); mtx_unlock(&tfd->tfd_lock); error = native_to_linux_itimerspec(&lots, &ots); if (error == 0) error = copyout(&lots, args->old_value, sizeof(lots)); out: fdrop(fp, td); return (error); } int linux_timerfd_settime(struct thread *td, struct linux_timerfd_settime_args *args) { struct l_itimerspec lots; struct itimerspec nts, ots; struct timespec cts, ts; struct timerfd *tfd; struct timeval tv; struct file *fp; int error; if ((args->flags & ~LINUX_TFD_SETTIME_FLAGS) != 0) return (EINVAL); error = copyin(args->new_value, &lots, sizeof(lots)); if (error != 0) return (error); error = linux_to_native_itimerspec(&nts, &lots); if (error != 0) return (error); error = fget(td, args->fd, &cap_write_rights, &fp); if (error != 0) return (error); tfd = fp->f_data; if (fp->f_type != DTYPE_LINUXTFD || tfd == NULL) { error = EINVAL; goto out; } mtx_lock(&tfd->tfd_lock); if (!timespecisset(&nts.it_value)) timespecclear(&nts.it_interval); if (args->old_value != NULL) linux_timerfd_curval(tfd, &ots); tfd->tfd_time = nts; tfd->tfd_count = 0; if (timespecisset(&nts.it_value)) { linux_timerfd_clocktime(tfd, &cts); ts = nts.it_value; if ((args->flags & LINUX_TFD_TIMER_ABSTIME) == 0) { timespecadd(&tfd->tfd_time.it_value, &cts, &tfd->tfd_time.it_value); } else { timespecsub(&ts, &cts, &ts); } TIMESPEC_TO_TIMEVAL(&tv, &ts); callout_reset(&tfd->tfd_callout, tvtohz(&tv), linux_timerfd_expire, tfd); tfd->tfd_canceled = false; } else { tfd->tfd_canceled = true; callout_stop(&tfd->tfd_callout); } mtx_unlock(&tfd->tfd_lock); if (args->old_value != NULL) { error = native_to_linux_itimerspec(&lots, &ots); if (error == 0) error = copyout(&lots, args->old_value, sizeof(lots)); } out: fdrop(fp, td); return (error); } static void linux_timerfd_expire(void *arg) { struct timespec cts, ts; struct timeval tv; struct timerfd *tfd; tfd = (struct timerfd *)arg; linux_timerfd_clocktime(tfd, &cts); if (timespeccmp(&cts, &tfd->tfd_time.it_value, >=)) { if (timespecisset(&tfd->tfd_time.it_interval)) timespecadd(&tfd->tfd_time.it_value, &tfd->tfd_time.it_interval, &tfd->tfd_time.it_value); else /* single shot timer */ timespecclear(&tfd->tfd_time.it_value); if (timespecisset(&tfd->tfd_time.it_value)) { timespecsub(&tfd->tfd_time.it_value, &cts, &ts); TIMESPEC_TO_TIMEVAL(&tv, &ts); callout_reset(&tfd->tfd_callout, tvtohz(&tv), linux_timerfd_expire, tfd); } tfd->tfd_count++; KNOTE_LOCKED(&tfd->tfd_sel.si_note, 0); selwakeup(&tfd->tfd_sel); wakeup(&tfd->tfd_count); } else if (timespecisset(&tfd->tfd_time.it_value)) { timespecsub(&tfd->tfd_time.it_value, &cts, &ts); TIMESPEC_TO_TIMEVAL(&tv, &ts); callout_reset(&tfd->tfd_callout, tvtohz(&tv), linux_timerfd_expire, tfd); } }