diff --git a/sys/amd64/sgx/sgx_linux.c b/sys/amd64/sgx/sgx_linux.c index 60ebec0e0278..c2975b0d2c84 100644 --- a/sys/amd64/sgx/sgx_linux.c +++ b/sys/amd64/sgx/sgx_linux.c @@ -1,115 +1,116 @@ /*- * Copyright (c) 2017 Ruslan Bukin * All rights reserved. * * This software was developed by BAE Systems, the University of Cambridge * Computer Laboratory, and Memorial University under DARPA/AFRL contract * FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent Computing * (TC) research program. * * 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 #define SGX_LINUX_IOCTL_MIN (SGX_IOC_ENCLAVE_CREATE & 0xffff) #define SGX_LINUX_IOCTL_MAX (SGX_IOC_ENCLAVE_INIT & 0xffff) static int sgx_linux_ioctl(struct thread *td, struct linux_ioctl_args *args) { uint8_t data[SGX_IOCTL_MAX_DATA_LEN]; cap_rights_t rights; struct file *fp; u_long cmd; int error; int len; - error = fget(td, args->fd, cap_rights_init(&rights, CAP_IOCTL), &fp); + error = fget(td, args->fd, cap_rights_init_one(&rights, CAP_IOCTL), + &fp); if (error != 0) return (error); cmd = args->cmd; args->cmd &= ~(LINUX_IOC_IN | LINUX_IOC_OUT); if ((cmd & LINUX_IOC_IN) != 0) args->cmd |= IOC_IN; if ((cmd & LINUX_IOC_OUT) != 0) args->cmd |= IOC_OUT; len = IOCPARM_LEN(cmd); if (len > SGX_IOCTL_MAX_DATA_LEN) { error = EINVAL; goto out; } if ((cmd & LINUX_IOC_IN) != 0) { error = copyin((void *)args->arg, data, len); if (error != 0) goto out; } error = fo_ioctl(fp, args->cmd, (caddr_t)data, td->td_ucred, td); out: fdrop(fp, td); return (error); } static struct linux_ioctl_handler sgx_linux_handler = { sgx_linux_ioctl, SGX_LINUX_IOCTL_MIN, SGX_LINUX_IOCTL_MAX, }; SYSINIT(sgx_linux_register, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_ioctl_register_handler, &sgx_linux_handler); SYSUNINIT(sgx_linux_unregister, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_ioctl_unregister_handler, &sgx_linux_handler); static int sgx_linux_modevent(module_t mod, int type, void *data) { return (0); } DEV_MODULE(sgx_linux, sgx_linux_modevent, NULL); MODULE_DEPEND(sgx_linux, linux64, 1, 1, 1); diff --git a/sys/compat/freebsd32/freebsd32_ioctl.c b/sys/compat/freebsd32/freebsd32_ioctl.c index db6946246b90..b3ed457f7132 100644 --- a/sys/compat/freebsd32/freebsd32_ioctl.c +++ b/sys/compat/freebsd32/freebsd32_ioctl.c @@ -1,247 +1,247 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2008 David E. O'Brien * 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. * 3. Neither the name of the author nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * 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 #include #include #include CTASSERT(sizeof(struct mem_range_op32) == 12); static int freebsd32_ioctl_memrange(struct thread *td, struct freebsd32_ioctl_args *uap, struct file *fp) { struct mem_range_op mro; struct mem_range_op32 mro32; int error; u_long com; if ((error = copyin(uap->data, &mro32, sizeof(mro32))) != 0) return (error); PTRIN_CP(mro32, mro, mo_desc); CP(mro32, mro, mo_arg[0]); CP(mro32, mro, mo_arg[1]); com = 0; switch (uap->com) { case MEMRANGE_GET32: com = MEMRANGE_GET; break; case MEMRANGE_SET32: com = MEMRANGE_SET; break; default: panic("%s: unknown MEMRANGE %#x", __func__, uap->com); } if ((error = fo_ioctl(fp, com, (caddr_t)&mro, td->td_ucred, td)) != 0) return (error); if ( (com & IOC_OUT) ) { CP(mro, mro32, mo_arg[0]); CP(mro, mro32, mo_arg[1]); error = copyout(&mro32, uap->data, sizeof(mro32)); } return (error); } static int freebsd32_ioctl_barmmap(struct thread *td, struct freebsd32_ioctl_args *uap, struct file *fp) { struct pci_bar_mmap32 pbm32; struct pci_bar_mmap pbm; int error; error = copyin(uap->data, &pbm32, sizeof(pbm32)); if (error != 0) return (error); PTRIN_CP(pbm32, pbm, pbm_map_base); CP(pbm32, pbm, pbm_sel); CP(pbm32, pbm, pbm_reg); CP(pbm32, pbm, pbm_flags); CP(pbm32, pbm, pbm_memattr); pbm.pbm_bar_length = PAIR32TO64(uint64_t, pbm32.pbm_bar_length); error = fo_ioctl(fp, PCIOCBARMMAP, (caddr_t)&pbm, td->td_ucred, td); if (error == 0) { PTROUT_CP(pbm, pbm32, pbm_map_base); CP(pbm, pbm32, pbm_map_length); #if BYTE_ORDER == LITTLE_ENDIAN pbm32.pbm_bar_length1 = pbm.pbm_bar_length; pbm32.pbm_bar_length2 = pbm.pbm_bar_length >> 32; #else pbm32.pbm_bar_length1 = pbm.pbm_bar_length >> 32; pbm32.pbm_bar_length2 = pbm.pbm_bar_length; #endif CP(pbm, pbm32, pbm_bar_off); error = copyout(&pbm32, uap->data, sizeof(pbm32)); } return (error); } static int freebsd32_ioctl_sg(struct thread *td, struct freebsd32_ioctl_args *uap, struct file *fp) { struct sg_io_hdr io; struct sg_io_hdr32 io32; int error; if ((error = copyin(uap->data, &io32, sizeof(io32))) != 0) return (error); CP(io32, io, interface_id); CP(io32, io, dxfer_direction); CP(io32, io, cmd_len); CP(io32, io, mx_sb_len); CP(io32, io, iovec_count); CP(io32, io, dxfer_len); PTRIN_CP(io32, io, dxferp); PTRIN_CP(io32, io, cmdp); PTRIN_CP(io32, io, sbp); CP(io32, io, timeout); CP(io32, io, flags); CP(io32, io, pack_id); PTRIN_CP(io32, io, usr_ptr); CP(io32, io, status); CP(io32, io, masked_status); CP(io32, io, msg_status); CP(io32, io, sb_len_wr); CP(io32, io, host_status); CP(io32, io, driver_status); CP(io32, io, resid); CP(io32, io, duration); CP(io32, io, info); if ((error = fo_ioctl(fp, SG_IO, (caddr_t)&io, td->td_ucred, td)) != 0) return (error); CP(io, io32, interface_id); CP(io, io32, dxfer_direction); CP(io, io32, cmd_len); CP(io, io32, mx_sb_len); CP(io, io32, iovec_count); CP(io, io32, dxfer_len); PTROUT_CP(io, io32, dxferp); PTROUT_CP(io, io32, cmdp); PTROUT_CP(io, io32, sbp); CP(io, io32, timeout); CP(io, io32, flags); CP(io, io32, pack_id); PTROUT_CP(io, io32, usr_ptr); CP(io, io32, status); CP(io, io32, masked_status); CP(io, io32, msg_status); CP(io, io32, sb_len_wr); CP(io, io32, host_status); CP(io, io32, driver_status); CP(io, io32, resid); CP(io, io32, duration); CP(io, io32, info); error = copyout(&io32, uap->data, sizeof(io32)); return (error); } int freebsd32_ioctl(struct thread *td, struct freebsd32_ioctl_args *uap) { struct ioctl_args ap /*{ int fd; u_long com; caddr_t data; }*/ ; struct file *fp; cap_rights_t rights; int error; - error = fget(td, uap->fd, cap_rights_init(&rights, CAP_IOCTL), &fp); + error = fget(td, uap->fd, cap_rights_init_one(&rights, CAP_IOCTL), &fp); if (error != 0) return (error); if ((fp->f_flag & (FREAD | FWRITE)) == 0) { fdrop(fp, td); return (EBADF); } switch (uap->com) { case MEMRANGE_GET32: /* FALLTHROUGH */ case MEMRANGE_SET32: error = freebsd32_ioctl_memrange(td, uap, fp); break; case SG_IO_32: error = freebsd32_ioctl_sg(td, uap, fp); break; case PCIOCBARMMAP_32: error = freebsd32_ioctl_barmmap(td, uap, fp); break; default: fdrop(fp, td); ap.fd = uap->fd; ap.com = uap->com; PTRIN_CP(*uap, ap, data); return sys_ioctl(td, &ap); } fdrop(fp, td); return (error); } diff --git a/sys/compat/freebsd32/freebsd32_misc.c b/sys/compat/freebsd32/freebsd32_misc.c index 14afd433d9f1..b7db1c4468d7 100644 --- a/sys/compat/freebsd32/freebsd32_misc.c +++ b/sys/compat/freebsd32/freebsd32_misc.c @@ -1,3833 +1,3833 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2002 Doug Rabson * 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_inet.h" #include "opt_inet6.h" #include "opt_ktrace.h" #define __ELF_WORD_SIZE 32 #ifdef COMPAT_FREEBSD11 #define _WANT_FREEBSD11_KEVENT #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* Must come after sys/malloc.h */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* Must come after sys/selinfo.h */ #include /* Must come after sys/selinfo.h */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #ifdef INET #include #endif #include #include #include #include #include #include #include #include #ifdef __amd64__ #include #endif #include #include #include #include #include #include #include FEATURE(compat_freebsd_32bit, "Compatible with 32-bit FreeBSD"); struct ptrace_io_desc32 { int piod_op; uint32_t piod_offs; uint32_t piod_addr; uint32_t piod_len; }; struct ptrace_sc_ret32 { uint32_t sr_retval[2]; int sr_error; }; struct ptrace_vm_entry32 { int pve_entry; int pve_timestamp; uint32_t pve_start; uint32_t pve_end; uint32_t pve_offset; u_int pve_prot; u_int pve_pathlen; int32_t pve_fileid; u_int pve_fsid; uint32_t pve_path; }; #ifdef __amd64__ CTASSERT(sizeof(struct timeval32) == 8); CTASSERT(sizeof(struct timespec32) == 8); CTASSERT(sizeof(struct itimerval32) == 16); CTASSERT(sizeof(struct bintime32) == 12); #endif CTASSERT(sizeof(struct statfs32) == 256); #ifdef __amd64__ CTASSERT(sizeof(struct rusage32) == 72); #endif CTASSERT(sizeof(struct sigaltstack32) == 12); #ifdef __amd64__ CTASSERT(sizeof(struct kevent32) == 56); #else CTASSERT(sizeof(struct kevent32) == 64); #endif CTASSERT(sizeof(struct iovec32) == 8); CTASSERT(sizeof(struct msghdr32) == 28); #ifdef __amd64__ CTASSERT(sizeof(struct stat32) == 208); CTASSERT(sizeof(struct freebsd11_stat32) == 96); #endif CTASSERT(sizeof(struct sigaction32) == 24); static int freebsd32_kevent_copyout(void *arg, struct kevent *kevp, int count); static int freebsd32_kevent_copyin(void *arg, struct kevent *kevp, int count); static int freebsd32_user_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags, const struct timespec32 *ua_rqtp, struct timespec32 *ua_rmtp); void freebsd32_rusage_out(const struct rusage *s, struct rusage32 *s32) { TV_CP(*s, *s32, ru_utime); TV_CP(*s, *s32, ru_stime); CP(*s, *s32, ru_maxrss); CP(*s, *s32, ru_ixrss); CP(*s, *s32, ru_idrss); CP(*s, *s32, ru_isrss); CP(*s, *s32, ru_minflt); CP(*s, *s32, ru_majflt); CP(*s, *s32, ru_nswap); CP(*s, *s32, ru_inblock); CP(*s, *s32, ru_oublock); CP(*s, *s32, ru_msgsnd); CP(*s, *s32, ru_msgrcv); CP(*s, *s32, ru_nsignals); CP(*s, *s32, ru_nvcsw); CP(*s, *s32, ru_nivcsw); } int freebsd32_wait4(struct thread *td, struct freebsd32_wait4_args *uap) { int error, status; struct rusage32 ru32; struct rusage ru, *rup; if (uap->rusage != NULL) rup = &ru; else rup = NULL; error = kern_wait(td, uap->pid, &status, uap->options, rup); if (error) return (error); if (uap->status != NULL) error = copyout(&status, uap->status, sizeof(status)); if (uap->rusage != NULL && error == 0) { freebsd32_rusage_out(&ru, &ru32); error = copyout(&ru32, uap->rusage, sizeof(ru32)); } return (error); } int freebsd32_wait6(struct thread *td, struct freebsd32_wait6_args *uap) { struct wrusage32 wru32; struct __wrusage wru, *wrup; struct siginfo32 si32; struct __siginfo si, *sip; int error, status; if (uap->wrusage != NULL) wrup = &wru; else wrup = NULL; if (uap->info != NULL) { sip = &si; bzero(sip, sizeof(*sip)); } else sip = NULL; error = kern_wait6(td, uap->idtype, PAIR32TO64(id_t, uap->id), &status, uap->options, wrup, sip); if (error != 0) return (error); if (uap->status != NULL) error = copyout(&status, uap->status, sizeof(status)); if (uap->wrusage != NULL && error == 0) { freebsd32_rusage_out(&wru.wru_self, &wru32.wru_self); freebsd32_rusage_out(&wru.wru_children, &wru32.wru_children); error = copyout(&wru32, uap->wrusage, sizeof(wru32)); } if (uap->info != NULL && error == 0) { siginfo_to_siginfo32 (&si, &si32); error = copyout(&si32, uap->info, sizeof(si32)); } return (error); } #ifdef COMPAT_FREEBSD4 static void copy_statfs(struct statfs *in, struct statfs32 *out) { statfs_scale_blocks(in, INT32_MAX); bzero(out, sizeof(*out)); CP(*in, *out, f_bsize); out->f_iosize = MIN(in->f_iosize, INT32_MAX); CP(*in, *out, f_blocks); CP(*in, *out, f_bfree); CP(*in, *out, f_bavail); out->f_files = MIN(in->f_files, INT32_MAX); out->f_ffree = MIN(in->f_ffree, INT32_MAX); CP(*in, *out, f_fsid); CP(*in, *out, f_owner); CP(*in, *out, f_type); CP(*in, *out, f_flags); out->f_syncwrites = MIN(in->f_syncwrites, INT32_MAX); out->f_asyncwrites = MIN(in->f_asyncwrites, INT32_MAX); strlcpy(out->f_fstypename, in->f_fstypename, MFSNAMELEN); strlcpy(out->f_mntonname, in->f_mntonname, min(MNAMELEN, FREEBSD4_MNAMELEN)); out->f_syncreads = MIN(in->f_syncreads, INT32_MAX); out->f_asyncreads = MIN(in->f_asyncreads, INT32_MAX); strlcpy(out->f_mntfromname, in->f_mntfromname, min(MNAMELEN, FREEBSD4_MNAMELEN)); } #endif #ifdef COMPAT_FREEBSD4 int freebsd4_freebsd32_getfsstat(struct thread *td, struct freebsd4_freebsd32_getfsstat_args *uap) { struct statfs *buf, *sp; struct statfs32 stat32; size_t count, size, copycount; int error; count = uap->bufsize / sizeof(struct statfs32); size = count * sizeof(struct statfs); error = kern_getfsstat(td, &buf, size, &count, UIO_SYSSPACE, uap->mode); if (size > 0) { sp = buf; copycount = count; while (copycount > 0 && error == 0) { copy_statfs(sp, &stat32); error = copyout(&stat32, uap->buf, sizeof(stat32)); sp++; uap->buf++; copycount--; } free(buf, M_STATFS); } if (error == 0) td->td_retval[0] = count; return (error); } #endif #ifdef COMPAT_FREEBSD10 int freebsd10_freebsd32_pipe(struct thread *td, struct freebsd10_freebsd32_pipe_args *uap) { return (freebsd10_pipe(td, (struct freebsd10_pipe_args*)uap)); } #endif int freebsd32_sigaltstack(struct thread *td, struct freebsd32_sigaltstack_args *uap) { struct sigaltstack32 s32; struct sigaltstack ss, oss, *ssp; int error; if (uap->ss != NULL) { error = copyin(uap->ss, &s32, sizeof(s32)); if (error) return (error); PTRIN_CP(s32, ss, ss_sp); CP(s32, ss, ss_size); CP(s32, ss, ss_flags); ssp = &ss; } else ssp = NULL; error = kern_sigaltstack(td, ssp, &oss); if (error == 0 && uap->oss != NULL) { PTROUT_CP(oss, s32, ss_sp); CP(oss, s32, ss_size); CP(oss, s32, ss_flags); error = copyout(&s32, uap->oss, sizeof(s32)); } return (error); } /* * Custom version of exec_copyin_args() so that we can translate * the pointers. */ int freebsd32_exec_copyin_args(struct image_args *args, const char *fname, enum uio_seg segflg, u_int32_t *argv, u_int32_t *envv) { char *argp, *envp; u_int32_t *p32, arg; int error; bzero(args, sizeof(*args)); if (argv == NULL) return (EFAULT); /* * Allocate demand-paged memory for the file name, argument, and * environment strings. */ error = exec_alloc_args(args); if (error != 0) return (error); /* * Copy the file name. */ error = exec_args_add_fname(args, fname, segflg); if (error != 0) goto err_exit; /* * extract arguments first */ p32 = argv; for (;;) { error = copyin(p32++, &arg, sizeof(arg)); if (error) goto err_exit; if (arg == 0) break; argp = PTRIN(arg); error = exec_args_add_arg(args, argp, UIO_USERSPACE); if (error != 0) goto err_exit; } /* * extract environment strings */ if (envv) { p32 = envv; for (;;) { error = copyin(p32++, &arg, sizeof(arg)); if (error) goto err_exit; if (arg == 0) break; envp = PTRIN(arg); error = exec_args_add_env(args, envp, UIO_USERSPACE); if (error != 0) goto err_exit; } } return (0); err_exit: exec_free_args(args); return (error); } int freebsd32_execve(struct thread *td, struct freebsd32_execve_args *uap) { struct image_args eargs; struct vmspace *oldvmspace; int error; error = pre_execve(td, &oldvmspace); if (error != 0) return (error); error = freebsd32_exec_copyin_args(&eargs, uap->fname, UIO_USERSPACE, uap->argv, uap->envv); if (error == 0) error = kern_execve(td, &eargs, NULL, oldvmspace); post_execve(td, error, oldvmspace); AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); return (error); } int freebsd32_fexecve(struct thread *td, struct freebsd32_fexecve_args *uap) { struct image_args eargs; struct vmspace *oldvmspace; int error; error = pre_execve(td, &oldvmspace); if (error != 0) return (error); error = freebsd32_exec_copyin_args(&eargs, NULL, UIO_SYSSPACE, uap->argv, uap->envv); if (error == 0) { eargs.fd = uap->fd; error = kern_execve(td, &eargs, NULL, oldvmspace); } post_execve(td, error, oldvmspace); AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); return (error); } int freebsd32_mknodat(struct thread *td, struct freebsd32_mknodat_args *uap) { return (kern_mknodat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode, PAIR32TO64(dev_t, uap->dev))); } int freebsd32_mprotect(struct thread *td, struct freebsd32_mprotect_args *uap) { int prot; prot = uap->prot; #if defined(__amd64__) if (i386_read_exec && (prot & PROT_READ) != 0) prot |= PROT_EXEC; #endif return (kern_mprotect(td, (uintptr_t)PTRIN(uap->addr), uap->len, prot)); } int freebsd32_mmap(struct thread *td, struct freebsd32_mmap_args *uap) { int prot; prot = uap->prot; #if defined(__amd64__) if (i386_read_exec && (prot & PROT_READ)) prot |= PROT_EXEC; #endif return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, uap->flags, uap->fd, PAIR32TO64(off_t, uap->pos))); } #ifdef COMPAT_FREEBSD6 int freebsd6_freebsd32_mmap(struct thread *td, struct freebsd6_freebsd32_mmap_args *uap) { int prot; prot = uap->prot; #if defined(__amd64__) if (i386_read_exec && (prot & PROT_READ)) prot |= PROT_EXEC; #endif return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, uap->flags, uap->fd, PAIR32TO64(off_t, uap->pos))); } #endif int freebsd32_setitimer(struct thread *td, struct freebsd32_setitimer_args *uap) { struct itimerval itv, oitv, *itvp; struct itimerval32 i32; int error; if (uap->itv != NULL) { error = copyin(uap->itv, &i32, sizeof(i32)); if (error) return (error); TV_CP(i32, itv, it_interval); TV_CP(i32, itv, it_value); itvp = &itv; } else itvp = NULL; error = kern_setitimer(td, uap->which, itvp, &oitv); if (error || uap->oitv == NULL) return (error); TV_CP(oitv, i32, it_interval); TV_CP(oitv, i32, it_value); return (copyout(&i32, uap->oitv, sizeof(i32))); } int freebsd32_getitimer(struct thread *td, struct freebsd32_getitimer_args *uap) { struct itimerval itv; struct itimerval32 i32; int error; error = kern_getitimer(td, uap->which, &itv); if (error || uap->itv == NULL) return (error); TV_CP(itv, i32, it_interval); TV_CP(itv, i32, it_value); return (copyout(&i32, uap->itv, sizeof(i32))); } int freebsd32_select(struct thread *td, struct freebsd32_select_args *uap) { struct timeval32 tv32; struct timeval tv, *tvp; int error; if (uap->tv != NULL) { error = copyin(uap->tv, &tv32, sizeof(tv32)); if (error) return (error); CP(tv32, tv, tv_sec); CP(tv32, tv, tv_usec); tvp = &tv; } else tvp = NULL; /* * XXX Do pointers need PTRIN()? */ return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp, sizeof(int32_t) * 8)); } int freebsd32_pselect(struct thread *td, struct freebsd32_pselect_args *uap) { struct timespec32 ts32; struct timespec ts; struct timeval tv, *tvp; sigset_t set, *uset; int error; if (uap->ts != NULL) { error = copyin(uap->ts, &ts32, sizeof(ts32)); if (error != 0) return (error); CP(ts32, ts, tv_sec); CP(ts32, ts, tv_nsec); TIMESPEC_TO_TIMEVAL(&tv, &ts); tvp = &tv; } else tvp = NULL; if (uap->sm != NULL) { error = copyin(uap->sm, &set, sizeof(set)); if (error != 0) return (error); uset = &set; } else uset = NULL; /* * XXX Do pointers need PTRIN()? */ error = kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp, uset, sizeof(int32_t) * 8); return (error); } /* * Copy 'count' items into the destination list pointed to by uap->eventlist. */ static int freebsd32_kevent_copyout(void *arg, struct kevent *kevp, int count) { struct freebsd32_kevent_args *uap; struct kevent32 ks32[KQ_NEVENTS]; uint64_t e; int i, j, error; KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); uap = (struct freebsd32_kevent_args *)arg; for (i = 0; i < count; i++) { CP(kevp[i], ks32[i], ident); CP(kevp[i], ks32[i], filter); CP(kevp[i], ks32[i], flags); CP(kevp[i], ks32[i], fflags); #if BYTE_ORDER == LITTLE_ENDIAN ks32[i].data1 = kevp[i].data; ks32[i].data2 = kevp[i].data >> 32; #else ks32[i].data1 = kevp[i].data >> 32; ks32[i].data2 = kevp[i].data; #endif PTROUT_CP(kevp[i], ks32[i], udata); for (j = 0; j < nitems(kevp->ext); j++) { e = kevp[i].ext[j]; #if BYTE_ORDER == LITTLE_ENDIAN ks32[i].ext64[2 * j] = e; ks32[i].ext64[2 * j + 1] = e >> 32; #else ks32[i].ext64[2 * j] = e >> 32; ks32[i].ext64[2 * j + 1] = e; #endif } } error = copyout(ks32, uap->eventlist, count * sizeof *ks32); if (error == 0) uap->eventlist += count; return (error); } /* * Copy 'count' items from the list pointed to by uap->changelist. */ static int freebsd32_kevent_copyin(void *arg, struct kevent *kevp, int count) { struct freebsd32_kevent_args *uap; struct kevent32 ks32[KQ_NEVENTS]; uint64_t e; int i, j, error; KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); uap = (struct freebsd32_kevent_args *)arg; error = copyin(uap->changelist, ks32, count * sizeof *ks32); if (error) goto done; uap->changelist += count; for (i = 0; i < count; i++) { CP(ks32[i], kevp[i], ident); CP(ks32[i], kevp[i], filter); CP(ks32[i], kevp[i], flags); CP(ks32[i], kevp[i], fflags); kevp[i].data = PAIR32TO64(uint64_t, ks32[i].data); PTRIN_CP(ks32[i], kevp[i], udata); for (j = 0; j < nitems(kevp->ext); j++) { #if BYTE_ORDER == LITTLE_ENDIAN e = ks32[i].ext64[2 * j + 1]; e <<= 32; e += ks32[i].ext64[2 * j]; #else e = ks32[i].ext64[2 * j]; e <<= 32; e += ks32[i].ext64[2 * j + 1]; #endif kevp[i].ext[j] = e; } } done: return (error); } int freebsd32_kevent(struct thread *td, struct freebsd32_kevent_args *uap) { struct timespec32 ts32; struct timespec ts, *tsp; struct kevent_copyops k_ops = { .arg = uap, .k_copyout = freebsd32_kevent_copyout, .k_copyin = freebsd32_kevent_copyin, }; #ifdef KTRACE struct kevent32 *eventlist = uap->eventlist; #endif int error; if (uap->timeout) { error = copyin(uap->timeout, &ts32, sizeof(ts32)); if (error) return (error); CP(ts32, ts, tv_sec); CP(ts32, ts, tv_nsec); tsp = &ts; } else tsp = NULL; #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT_ARRAY)) ktrstructarray("kevent32", UIO_USERSPACE, uap->changelist, uap->nchanges, sizeof(struct kevent32)); #endif error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents, &k_ops, tsp); #ifdef KTRACE if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY)) ktrstructarray("kevent32", UIO_USERSPACE, eventlist, td->td_retval[0], sizeof(struct kevent32)); #endif return (error); } #ifdef COMPAT_FREEBSD11 static int freebsd32_kevent11_copyout(void *arg, struct kevent *kevp, int count) { struct freebsd11_freebsd32_kevent_args *uap; struct kevent32_freebsd11 ks32[KQ_NEVENTS]; int i, error; KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); uap = (struct freebsd11_freebsd32_kevent_args *)arg; for (i = 0; i < count; i++) { CP(kevp[i], ks32[i], ident); CP(kevp[i], ks32[i], filter); CP(kevp[i], ks32[i], flags); CP(kevp[i], ks32[i], fflags); CP(kevp[i], ks32[i], data); PTROUT_CP(kevp[i], ks32[i], udata); } error = copyout(ks32, uap->eventlist, count * sizeof *ks32); if (error == 0) uap->eventlist += count; return (error); } /* * Copy 'count' items from the list pointed to by uap->changelist. */ static int freebsd32_kevent11_copyin(void *arg, struct kevent *kevp, int count) { struct freebsd11_freebsd32_kevent_args *uap; struct kevent32_freebsd11 ks32[KQ_NEVENTS]; int i, j, error; KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); uap = (struct freebsd11_freebsd32_kevent_args *)arg; error = copyin(uap->changelist, ks32, count * sizeof *ks32); if (error) goto done; uap->changelist += count; for (i = 0; i < count; i++) { CP(ks32[i], kevp[i], ident); CP(ks32[i], kevp[i], filter); CP(ks32[i], kevp[i], flags); CP(ks32[i], kevp[i], fflags); CP(ks32[i], kevp[i], data); PTRIN_CP(ks32[i], kevp[i], udata); for (j = 0; j < nitems(kevp->ext); j++) kevp[i].ext[j] = 0; } done: return (error); } int freebsd11_freebsd32_kevent(struct thread *td, struct freebsd11_freebsd32_kevent_args *uap) { struct timespec32 ts32; struct timespec ts, *tsp; struct kevent_copyops k_ops = { .arg = uap, .k_copyout = freebsd32_kevent11_copyout, .k_copyin = freebsd32_kevent11_copyin, }; #ifdef KTRACE struct kevent32_freebsd11 *eventlist = uap->eventlist; #endif int error; if (uap->timeout) { error = copyin(uap->timeout, &ts32, sizeof(ts32)); if (error) return (error); CP(ts32, ts, tv_sec); CP(ts32, ts, tv_nsec); tsp = &ts; } else tsp = NULL; #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT_ARRAY)) ktrstructarray("kevent32_freebsd11", UIO_USERSPACE, uap->changelist, uap->nchanges, sizeof(struct kevent32_freebsd11)); #endif error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents, &k_ops, tsp); #ifdef KTRACE if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY)) ktrstructarray("kevent32_freebsd11", UIO_USERSPACE, eventlist, td->td_retval[0], sizeof(struct kevent32_freebsd11)); #endif return (error); } #endif int freebsd32_gettimeofday(struct thread *td, struct freebsd32_gettimeofday_args *uap) { struct timeval atv; struct timeval32 atv32; struct timezone rtz; int error = 0; if (uap->tp) { microtime(&atv); CP(atv, atv32, tv_sec); CP(atv, atv32, tv_usec); error = copyout(&atv32, uap->tp, sizeof (atv32)); } if (error == 0 && uap->tzp != NULL) { rtz.tz_minuteswest = 0; rtz.tz_dsttime = 0; error = copyout(&rtz, uap->tzp, sizeof (rtz)); } return (error); } int freebsd32_getrusage(struct thread *td, struct freebsd32_getrusage_args *uap) { struct rusage32 s32; struct rusage s; int error; error = kern_getrusage(td, uap->who, &s); if (error == 0) { freebsd32_rusage_out(&s, &s32); error = copyout(&s32, uap->rusage, sizeof(s32)); } return (error); } static void ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl, struct ptrace_lwpinfo32 *pl32) { bzero(pl32, sizeof(*pl32)); pl32->pl_lwpid = pl->pl_lwpid; pl32->pl_event = pl->pl_event; pl32->pl_flags = pl->pl_flags; pl32->pl_sigmask = pl->pl_sigmask; pl32->pl_siglist = pl->pl_siglist; siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo); strcpy(pl32->pl_tdname, pl->pl_tdname); pl32->pl_child_pid = pl->pl_child_pid; pl32->pl_syscall_code = pl->pl_syscall_code; pl32->pl_syscall_narg = pl->pl_syscall_narg; } static void ptrace_sc_ret_to32(const struct ptrace_sc_ret *psr, struct ptrace_sc_ret32 *psr32) { bzero(psr32, sizeof(*psr32)); psr32->sr_retval[0] = psr->sr_retval[0]; psr32->sr_retval[1] = psr->sr_retval[1]; psr32->sr_error = psr->sr_error; } int freebsd32_ptrace(struct thread *td, struct freebsd32_ptrace_args *uap) { union { struct ptrace_io_desc piod; struct ptrace_lwpinfo pl; struct ptrace_vm_entry pve; struct dbreg32 dbreg; struct fpreg32 fpreg; struct reg32 reg; register_t args[nitems(td->td_sa.args)]; struct ptrace_sc_ret psr; int ptevents; } r; union { struct ptrace_io_desc32 piod; struct ptrace_lwpinfo32 pl; struct ptrace_vm_entry32 pve; uint32_t args[nitems(td->td_sa.args)]; struct ptrace_sc_ret32 psr; } r32; void *addr; int data, error = 0, i; AUDIT_ARG_PID(uap->pid); AUDIT_ARG_CMD(uap->req); AUDIT_ARG_VALUE(uap->data); addr = &r; data = uap->data; switch (uap->req) { case PT_GET_EVENT_MASK: case PT_GET_SC_ARGS: case PT_GET_SC_RET: break; case PT_LWPINFO: if (uap->data > sizeof(r32.pl)) return (EINVAL); /* * Pass size of native structure in 'data'. Truncate * if necessary to avoid siginfo. */ data = sizeof(r.pl); if (uap->data < offsetof(struct ptrace_lwpinfo32, pl_siginfo) + sizeof(struct siginfo32)) data = offsetof(struct ptrace_lwpinfo, pl_siginfo); break; case PT_GETREGS: bzero(&r.reg, sizeof(r.reg)); break; case PT_GETFPREGS: bzero(&r.fpreg, sizeof(r.fpreg)); break; case PT_GETDBREGS: bzero(&r.dbreg, sizeof(r.dbreg)); break; case PT_SETREGS: error = copyin(uap->addr, &r.reg, sizeof(r.reg)); break; case PT_SETFPREGS: error = copyin(uap->addr, &r.fpreg, sizeof(r.fpreg)); break; case PT_SETDBREGS: error = copyin(uap->addr, &r.dbreg, sizeof(r.dbreg)); break; case PT_SET_EVENT_MASK: if (uap->data != sizeof(r.ptevents)) error = EINVAL; else error = copyin(uap->addr, &r.ptevents, uap->data); break; case PT_IO: error = copyin(uap->addr, &r32.piod, sizeof(r32.piod)); if (error) break; CP(r32.piod, r.piod, piod_op); PTRIN_CP(r32.piod, r.piod, piod_offs); PTRIN_CP(r32.piod, r.piod, piod_addr); CP(r32.piod, r.piod, piod_len); break; case PT_VM_ENTRY: error = copyin(uap->addr, &r32.pve, sizeof(r32.pve)); if (error) break; CP(r32.pve, r.pve, pve_entry); CP(r32.pve, r.pve, pve_timestamp); CP(r32.pve, r.pve, pve_start); CP(r32.pve, r.pve, pve_end); CP(r32.pve, r.pve, pve_offset); CP(r32.pve, r.pve, pve_prot); CP(r32.pve, r.pve, pve_pathlen); CP(r32.pve, r.pve, pve_fileid); CP(r32.pve, r.pve, pve_fsid); PTRIN_CP(r32.pve, r.pve, pve_path); break; default: addr = uap->addr; break; } if (error) return (error); error = kern_ptrace(td, uap->req, uap->pid, addr, data); if (error) return (error); switch (uap->req) { case PT_VM_ENTRY: CP(r.pve, r32.pve, pve_entry); CP(r.pve, r32.pve, pve_timestamp); CP(r.pve, r32.pve, pve_start); CP(r.pve, r32.pve, pve_end); CP(r.pve, r32.pve, pve_offset); CP(r.pve, r32.pve, pve_prot); CP(r.pve, r32.pve, pve_pathlen); CP(r.pve, r32.pve, pve_fileid); CP(r.pve, r32.pve, pve_fsid); error = copyout(&r32.pve, uap->addr, sizeof(r32.pve)); break; case PT_IO: CP(r.piod, r32.piod, piod_len); error = copyout(&r32.piod, uap->addr, sizeof(r32.piod)); break; case PT_GETREGS: error = copyout(&r.reg, uap->addr, sizeof(r.reg)); break; case PT_GETFPREGS: error = copyout(&r.fpreg, uap->addr, sizeof(r.fpreg)); break; case PT_GETDBREGS: error = copyout(&r.dbreg, uap->addr, sizeof(r.dbreg)); break; case PT_GET_EVENT_MASK: /* NB: The size in uap->data is validated in kern_ptrace(). */ error = copyout(&r.ptevents, uap->addr, uap->data); break; case PT_LWPINFO: ptrace_lwpinfo_to32(&r.pl, &r32.pl); error = copyout(&r32.pl, uap->addr, uap->data); break; case PT_GET_SC_ARGS: for (i = 0; i < nitems(r.args); i++) r32.args[i] = (uint32_t)r.args[i]; error = copyout(r32.args, uap->addr, MIN(uap->data, sizeof(r32.args))); break; case PT_GET_SC_RET: ptrace_sc_ret_to32(&r.psr, &r32.psr); error = copyout(&r32.psr, uap->addr, MIN(uap->data, sizeof(r32.psr))); break; } return (error); } int freebsd32_copyinuio(struct iovec32 *iovp, u_int iovcnt, struct uio **uiop) { struct iovec32 iov32; struct iovec *iov; struct uio *uio; u_int iovlen; int error, i; *uiop = NULL; if (iovcnt > UIO_MAXIOV) return (EINVAL); iovlen = iovcnt * sizeof(struct iovec); uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK); iov = (struct iovec *)(uio + 1); for (i = 0; i < iovcnt; i++) { error = copyin(&iovp[i], &iov32, sizeof(struct iovec32)); if (error) { free(uio, M_IOV); return (error); } iov[i].iov_base = PTRIN(iov32.iov_base); iov[i].iov_len = iov32.iov_len; } uio->uio_iov = iov; uio->uio_iovcnt = iovcnt; uio->uio_segflg = UIO_USERSPACE; uio->uio_offset = -1; uio->uio_resid = 0; for (i = 0; i < iovcnt; i++) { if (iov->iov_len > INT_MAX - uio->uio_resid) { free(uio, M_IOV); return (EINVAL); } uio->uio_resid += iov->iov_len; iov++; } *uiop = uio; return (0); } int freebsd32_readv(struct thread *td, struct freebsd32_readv_args *uap) { struct uio *auio; int error; error = freebsd32_copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = kern_readv(td, uap->fd, auio); free(auio, M_IOV); return (error); } int freebsd32_writev(struct thread *td, struct freebsd32_writev_args *uap) { struct uio *auio; int error; error = freebsd32_copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = kern_writev(td, uap->fd, auio); free(auio, M_IOV); return (error); } int freebsd32_preadv(struct thread *td, struct freebsd32_preadv_args *uap) { struct uio *auio; int error; error = freebsd32_copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = kern_preadv(td, uap->fd, auio, PAIR32TO64(off_t,uap->offset)); free(auio, M_IOV); return (error); } int freebsd32_pwritev(struct thread *td, struct freebsd32_pwritev_args *uap) { struct uio *auio; int error; error = freebsd32_copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = kern_pwritev(td, uap->fd, auio, PAIR32TO64(off_t,uap->offset)); free(auio, M_IOV); return (error); } int freebsd32_copyiniov(struct iovec32 *iovp32, u_int iovcnt, struct iovec **iovp, int error) { struct iovec32 iov32; struct iovec *iov; u_int iovlen; int i; *iovp = NULL; if (iovcnt > UIO_MAXIOV) return (error); iovlen = iovcnt * sizeof(struct iovec); iov = malloc(iovlen, M_IOV, M_WAITOK); for (i = 0; i < iovcnt; i++) { error = copyin(&iovp32[i], &iov32, sizeof(struct iovec32)); if (error) { free(iov, M_IOV); return (error); } iov[i].iov_base = PTRIN(iov32.iov_base); iov[i].iov_len = iov32.iov_len; } *iovp = iov; return (0); } static int freebsd32_copyinmsghdr(struct msghdr32 *msg32, struct msghdr *msg) { struct msghdr32 m32; int error; error = copyin(msg32, &m32, sizeof(m32)); if (error) return (error); msg->msg_name = PTRIN(m32.msg_name); msg->msg_namelen = m32.msg_namelen; msg->msg_iov = PTRIN(m32.msg_iov); msg->msg_iovlen = m32.msg_iovlen; msg->msg_control = PTRIN(m32.msg_control); msg->msg_controllen = m32.msg_controllen; msg->msg_flags = m32.msg_flags; return (0); } static int freebsd32_copyoutmsghdr(struct msghdr *msg, struct msghdr32 *msg32) { struct msghdr32 m32; int error; m32.msg_name = PTROUT(msg->msg_name); m32.msg_namelen = msg->msg_namelen; m32.msg_iov = PTROUT(msg->msg_iov); m32.msg_iovlen = msg->msg_iovlen; m32.msg_control = PTROUT(msg->msg_control); m32.msg_controllen = msg->msg_controllen; m32.msg_flags = msg->msg_flags; error = copyout(&m32, msg32, sizeof(m32)); return (error); } #ifndef __mips__ #define FREEBSD32_ALIGNBYTES (sizeof(int) - 1) #else #define FREEBSD32_ALIGNBYTES (sizeof(long) - 1) #endif #define FREEBSD32_ALIGN(p) \ (((u_long)(p) + FREEBSD32_ALIGNBYTES) & ~FREEBSD32_ALIGNBYTES) #define FREEBSD32_CMSG_SPACE(l) \ (FREEBSD32_ALIGN(sizeof(struct cmsghdr)) + FREEBSD32_ALIGN(l)) #define FREEBSD32_CMSG_DATA(cmsg) ((unsigned char *)(cmsg) + \ FREEBSD32_ALIGN(sizeof(struct cmsghdr))) static size_t freebsd32_cmsg_convert(const struct cmsghdr *cm, void *data, socklen_t datalen) { size_t copylen; union { struct timespec32 ts; struct timeval32 tv; struct bintime32 bt; } tmp32; union { struct timespec ts; struct timeval tv; struct bintime bt; } *in; in = data; copylen = 0; switch (cm->cmsg_level) { case SOL_SOCKET: switch (cm->cmsg_type) { case SCM_TIMESTAMP: TV_CP(*in, tmp32, tv); copylen = sizeof(tmp32.tv); break; case SCM_BINTIME: BT_CP(*in, tmp32, bt); copylen = sizeof(tmp32.bt); break; case SCM_REALTIME: case SCM_MONOTONIC: TS_CP(*in, tmp32, ts); copylen = sizeof(tmp32.ts); break; default: break; } default: break; } if (copylen == 0) return (datalen); KASSERT((datalen >= copylen), ("corrupted cmsghdr")); bcopy(&tmp32, data, copylen); return (copylen); } static int freebsd32_copy_msg_out(struct msghdr *msg, struct mbuf *control) { struct cmsghdr *cm; void *data; socklen_t clen, datalen, datalen_out, oldclen; int error; caddr_t ctlbuf; int len, maxlen, copylen; struct mbuf *m; error = 0; len = msg->msg_controllen; maxlen = msg->msg_controllen; msg->msg_controllen = 0; ctlbuf = msg->msg_control; for (m = control; m != NULL && len > 0; m = m->m_next) { cm = mtod(m, struct cmsghdr *); clen = m->m_len; while (cm != NULL) { if (sizeof(struct cmsghdr) > clen || cm->cmsg_len > clen) { error = EINVAL; break; } data = CMSG_DATA(cm); datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; datalen_out = freebsd32_cmsg_convert(cm, data, datalen); /* * Copy out the message header. Preserve the native * message size in case we need to inspect the message * contents later. */ copylen = sizeof(struct cmsghdr); if (len < copylen) { msg->msg_flags |= MSG_CTRUNC; m_dispose_extcontrolm(m); goto exit; } oldclen = cm->cmsg_len; cm->cmsg_len = FREEBSD32_ALIGN(sizeof(struct cmsghdr)) + datalen_out; error = copyout(cm, ctlbuf, copylen); cm->cmsg_len = oldclen; if (error != 0) goto exit; ctlbuf += FREEBSD32_ALIGN(copylen); len -= FREEBSD32_ALIGN(copylen); copylen = datalen_out; if (len < copylen) { msg->msg_flags |= MSG_CTRUNC; m_dispose_extcontrolm(m); break; } /* Copy out the message data. */ error = copyout(data, ctlbuf, copylen); if (error) goto exit; ctlbuf += FREEBSD32_ALIGN(copylen); len -= FREEBSD32_ALIGN(copylen); if (CMSG_SPACE(datalen) < clen) { clen -= CMSG_SPACE(datalen); cm = (struct cmsghdr *) ((caddr_t)cm + CMSG_SPACE(datalen)); } else { clen = 0; cm = NULL; } msg->msg_controllen += FREEBSD32_CMSG_SPACE(datalen_out); } } if (len == 0 && m != NULL) { msg->msg_flags |= MSG_CTRUNC; m_dispose_extcontrolm(m); } exit: return (error); } int freebsd32_recvmsg(td, uap) struct thread *td; struct freebsd32_recvmsg_args /* { int s; struct msghdr32 *msg; int flags; } */ *uap; { struct msghdr msg; struct msghdr32 m32; struct iovec *uiov, *iov; struct mbuf *control = NULL; struct mbuf **controlp; int error; error = copyin(uap->msg, &m32, sizeof(m32)); if (error) return (error); error = freebsd32_copyinmsghdr(uap->msg, &msg); if (error) return (error); error = freebsd32_copyiniov(PTRIN(m32.msg_iov), m32.msg_iovlen, &iov, EMSGSIZE); if (error) return (error); msg.msg_flags = uap->flags; uiov = msg.msg_iov; msg.msg_iov = iov; controlp = (msg.msg_control != NULL) ? &control : NULL; error = kern_recvit(td, uap->s, &msg, UIO_USERSPACE, controlp); if (error == 0) { msg.msg_iov = uiov; if (control != NULL) error = freebsd32_copy_msg_out(&msg, control); else msg.msg_controllen = 0; if (error == 0) error = freebsd32_copyoutmsghdr(&msg, uap->msg); } free(iov, M_IOV); if (control != NULL) { if (error != 0) m_dispose_extcontrolm(control); m_freem(control); } return (error); } /* * Copy-in the array of control messages constructed using alignment * and padding suitable for a 32-bit environment and construct an * mbuf using alignment and padding suitable for a 64-bit kernel. * The alignment and padding are defined indirectly by CMSG_DATA(), * CMSG_SPACE() and CMSG_LEN(). */ static int freebsd32_copyin_control(struct mbuf **mp, caddr_t buf, u_int buflen) { struct cmsghdr *cm; struct mbuf *m; void *in, *in1, *md; u_int msglen, outlen; int error; if (buflen > MCLBYTES) return (EINVAL); in = malloc(buflen, M_TEMP, M_WAITOK); error = copyin(buf, in, buflen); if (error != 0) goto out; /* * Make a pass over the input buffer to determine the amount of space * required for 64 bit-aligned copies of the control messages. */ in1 = in; outlen = 0; while (buflen > 0) { if (buflen < sizeof(*cm)) { error = EINVAL; break; } cm = (struct cmsghdr *)in1; if (cm->cmsg_len < FREEBSD32_ALIGN(sizeof(*cm))) { error = EINVAL; break; } msglen = FREEBSD32_ALIGN(cm->cmsg_len); if (msglen > buflen || msglen < cm->cmsg_len) { error = EINVAL; break; } buflen -= msglen; in1 = (char *)in1 + msglen; outlen += CMSG_ALIGN(sizeof(*cm)) + CMSG_ALIGN(msglen - FREEBSD32_ALIGN(sizeof(*cm))); } if (error == 0 && outlen > MCLBYTES) { /* * XXXMJ This implies that the upper limit on 32-bit aligned * control messages is less than MCLBYTES, and so we are not * perfectly compatible. However, there is no platform * guarantee that mbuf clusters larger than MCLBYTES can be * allocated. */ error = EINVAL; } if (error != 0) goto out; m = m_get2(outlen, M_WAITOK, MT_CONTROL, 0); m->m_len = outlen; md = mtod(m, void *); /* * Make a second pass over input messages, copying them into the output * buffer. */ in1 = in; while (outlen > 0) { /* Copy the message header and align the length field. */ cm = md; memcpy(cm, in1, sizeof(*cm)); msglen = cm->cmsg_len - FREEBSD32_ALIGN(sizeof(*cm)); cm->cmsg_len = CMSG_ALIGN(sizeof(*cm)) + msglen; /* Copy the message body. */ in1 = (char *)in1 + FREEBSD32_ALIGN(sizeof(*cm)); md = (char *)md + CMSG_ALIGN(sizeof(*cm)); memcpy(md, in1, msglen); in1 = (char *)in1 + FREEBSD32_ALIGN(msglen); md = (char *)md + CMSG_ALIGN(msglen); KASSERT(outlen >= CMSG_ALIGN(sizeof(*cm)) + CMSG_ALIGN(msglen), ("outlen %u underflow, msglen %u", outlen, msglen)); outlen -= CMSG_ALIGN(sizeof(*cm)) + CMSG_ALIGN(msglen); } *mp = m; out: free(in, M_TEMP); return (error); } int freebsd32_sendmsg(struct thread *td, struct freebsd32_sendmsg_args *uap) { struct msghdr msg; struct msghdr32 m32; struct iovec *iov; struct mbuf *control = NULL; struct sockaddr *to = NULL; int error; error = copyin(uap->msg, &m32, sizeof(m32)); if (error) return (error); error = freebsd32_copyinmsghdr(uap->msg, &msg); if (error) return (error); error = freebsd32_copyiniov(PTRIN(m32.msg_iov), m32.msg_iovlen, &iov, EMSGSIZE); if (error) return (error); msg.msg_iov = iov; if (msg.msg_name != NULL) { error = getsockaddr(&to, msg.msg_name, msg.msg_namelen); if (error) { to = NULL; goto out; } msg.msg_name = to; } if (msg.msg_control) { if (msg.msg_controllen < sizeof(struct cmsghdr)) { error = EINVAL; goto out; } error = freebsd32_copyin_control(&control, msg.msg_control, msg.msg_controllen); if (error) goto out; msg.msg_control = NULL; msg.msg_controllen = 0; } error = kern_sendit(td, uap->s, &msg, uap->flags, control, UIO_USERSPACE); out: free(iov, M_IOV); if (to) free(to, M_SONAME); return (error); } int freebsd32_recvfrom(struct thread *td, struct freebsd32_recvfrom_args *uap) { struct msghdr msg; struct iovec aiov; int error; if (uap->fromlenaddr) { error = copyin(PTRIN(uap->fromlenaddr), &msg.msg_namelen, sizeof(msg.msg_namelen)); if (error) return (error); } else { msg.msg_namelen = 0; } msg.msg_name = PTRIN(uap->from); msg.msg_iov = &aiov; msg.msg_iovlen = 1; aiov.iov_base = PTRIN(uap->buf); aiov.iov_len = uap->len; msg.msg_control = NULL; msg.msg_flags = uap->flags; error = kern_recvit(td, uap->s, &msg, UIO_USERSPACE, NULL); if (error == 0 && uap->fromlenaddr) error = copyout(&msg.msg_namelen, PTRIN(uap->fromlenaddr), sizeof (msg.msg_namelen)); return (error); } int freebsd32_settimeofday(struct thread *td, struct freebsd32_settimeofday_args *uap) { struct timeval32 tv32; struct timeval tv, *tvp; struct timezone tz, *tzp; int error; if (uap->tv) { error = copyin(uap->tv, &tv32, sizeof(tv32)); if (error) return (error); CP(tv32, tv, tv_sec); CP(tv32, tv, tv_usec); tvp = &tv; } else tvp = NULL; if (uap->tzp) { error = copyin(uap->tzp, &tz, sizeof(tz)); if (error) return (error); tzp = &tz; } else tzp = NULL; return (kern_settimeofday(td, tvp, tzp)); } int freebsd32_utimes(struct thread *td, struct freebsd32_utimes_args *uap) { struct timeval32 s32[2]; struct timeval s[2], *sp; int error; if (uap->tptr != NULL) { error = copyin(uap->tptr, s32, sizeof(s32)); if (error) return (error); CP(s32[0], s[0], tv_sec); CP(s32[0], s[0], tv_usec); CP(s32[1], s[1], tv_sec); CP(s32[1], s[1], tv_usec); sp = s; } else sp = NULL; return (kern_utimesat(td, AT_FDCWD, uap->path, UIO_USERSPACE, sp, UIO_SYSSPACE)); } int freebsd32_lutimes(struct thread *td, struct freebsd32_lutimes_args *uap) { struct timeval32 s32[2]; struct timeval s[2], *sp; int error; if (uap->tptr != NULL) { error = copyin(uap->tptr, s32, sizeof(s32)); if (error) return (error); CP(s32[0], s[0], tv_sec); CP(s32[0], s[0], tv_usec); CP(s32[1], s[1], tv_sec); CP(s32[1], s[1], tv_usec); sp = s; } else sp = NULL; return (kern_lutimes(td, uap->path, UIO_USERSPACE, sp, UIO_SYSSPACE)); } int freebsd32_futimes(struct thread *td, struct freebsd32_futimes_args *uap) { struct timeval32 s32[2]; struct timeval s[2], *sp; int error; if (uap->tptr != NULL) { error = copyin(uap->tptr, s32, sizeof(s32)); if (error) return (error); CP(s32[0], s[0], tv_sec); CP(s32[0], s[0], tv_usec); CP(s32[1], s[1], tv_sec); CP(s32[1], s[1], tv_usec); sp = s; } else sp = NULL; return (kern_futimes(td, uap->fd, sp, UIO_SYSSPACE)); } int freebsd32_futimesat(struct thread *td, struct freebsd32_futimesat_args *uap) { struct timeval32 s32[2]; struct timeval s[2], *sp; int error; if (uap->times != NULL) { error = copyin(uap->times, s32, sizeof(s32)); if (error) return (error); CP(s32[0], s[0], tv_sec); CP(s32[0], s[0], tv_usec); CP(s32[1], s[1], tv_sec); CP(s32[1], s[1], tv_usec); sp = s; } else sp = NULL; return (kern_utimesat(td, uap->fd, uap->path, UIO_USERSPACE, sp, UIO_SYSSPACE)); } int freebsd32_futimens(struct thread *td, struct freebsd32_futimens_args *uap) { struct timespec32 ts32[2]; struct timespec ts[2], *tsp; int error; if (uap->times != NULL) { error = copyin(uap->times, ts32, sizeof(ts32)); if (error) return (error); CP(ts32[0], ts[0], tv_sec); CP(ts32[0], ts[0], tv_nsec); CP(ts32[1], ts[1], tv_sec); CP(ts32[1], ts[1], tv_nsec); tsp = ts; } else tsp = NULL; return (kern_futimens(td, uap->fd, tsp, UIO_SYSSPACE)); } int freebsd32_utimensat(struct thread *td, struct freebsd32_utimensat_args *uap) { struct timespec32 ts32[2]; struct timespec ts[2], *tsp; int error; if (uap->times != NULL) { error = copyin(uap->times, ts32, sizeof(ts32)); if (error) return (error); CP(ts32[0], ts[0], tv_sec); CP(ts32[0], ts[0], tv_nsec); CP(ts32[1], ts[1], tv_sec); CP(ts32[1], ts[1], tv_nsec); tsp = ts; } else tsp = NULL; return (kern_utimensat(td, uap->fd, uap->path, UIO_USERSPACE, tsp, UIO_SYSSPACE, uap->flag)); } int freebsd32_adjtime(struct thread *td, struct freebsd32_adjtime_args *uap) { struct timeval32 tv32; struct timeval delta, olddelta, *deltap; int error; if (uap->delta) { error = copyin(uap->delta, &tv32, sizeof(tv32)); if (error) return (error); CP(tv32, delta, tv_sec); CP(tv32, delta, tv_usec); deltap = δ } else deltap = NULL; error = kern_adjtime(td, deltap, &olddelta); if (uap->olddelta && error == 0) { CP(olddelta, tv32, tv_sec); CP(olddelta, tv32, tv_usec); error = copyout(&tv32, uap->olddelta, sizeof(tv32)); } return (error); } #ifdef COMPAT_FREEBSD4 int freebsd4_freebsd32_statfs(struct thread *td, struct freebsd4_freebsd32_statfs_args *uap) { struct statfs32 s32; struct statfs *sp; int error; sp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_statfs(td, uap->path, UIO_USERSPACE, sp); if (error == 0) { copy_statfs(sp, &s32); error = copyout(&s32, uap->buf, sizeof(s32)); } free(sp, M_STATFS); return (error); } #endif #ifdef COMPAT_FREEBSD4 int freebsd4_freebsd32_fstatfs(struct thread *td, struct freebsd4_freebsd32_fstatfs_args *uap) { struct statfs32 s32; struct statfs *sp; int error; sp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_fstatfs(td, uap->fd, sp); if (error == 0) { copy_statfs(sp, &s32); error = copyout(&s32, uap->buf, sizeof(s32)); } free(sp, M_STATFS); return (error); } #endif #ifdef COMPAT_FREEBSD4 int freebsd4_freebsd32_fhstatfs(struct thread *td, struct freebsd4_freebsd32_fhstatfs_args *uap) { struct statfs32 s32; struct statfs *sp; fhandle_t fh; int error; if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0) return (error); sp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_fhstatfs(td, fh, sp); if (error == 0) { copy_statfs(sp, &s32); error = copyout(&s32, uap->buf, sizeof(s32)); } free(sp, M_STATFS); return (error); } #endif int freebsd32_pread(struct thread *td, struct freebsd32_pread_args *uap) { return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, PAIR32TO64(off_t, uap->offset))); } int freebsd32_pwrite(struct thread *td, struct freebsd32_pwrite_args *uap) { return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, PAIR32TO64(off_t, uap->offset))); } #ifdef COMPAT_43 int ofreebsd32_lseek(struct thread *td, struct ofreebsd32_lseek_args *uap) { return (kern_lseek(td, uap->fd, uap->offset, uap->whence)); } #endif int freebsd32_lseek(struct thread *td, struct freebsd32_lseek_args *uap) { int error; off_t pos; error = kern_lseek(td, uap->fd, PAIR32TO64(off_t, uap->offset), uap->whence); /* Expand the quad return into two parts for eax and edx */ pos = td->td_uretoff.tdu_off; td->td_retval[RETVAL_LO] = pos & 0xffffffff; /* %eax */ td->td_retval[RETVAL_HI] = pos >> 32; /* %edx */ return error; } int freebsd32_truncate(struct thread *td, struct freebsd32_truncate_args *uap) { return (kern_truncate(td, uap->path, UIO_USERSPACE, PAIR32TO64(off_t, uap->length))); } int freebsd32_ftruncate(struct thread *td, struct freebsd32_ftruncate_args *uap) { return (kern_ftruncate(td, uap->fd, PAIR32TO64(off_t, uap->length))); } #ifdef COMPAT_43 int ofreebsd32_getdirentries(struct thread *td, struct ofreebsd32_getdirentries_args *uap) { struct ogetdirentries_args ap; int error; long loff; int32_t loff_cut; ap.fd = uap->fd; ap.buf = uap->buf; ap.count = uap->count; ap.basep = NULL; error = kern_ogetdirentries(td, &ap, &loff); if (error == 0) { loff_cut = loff; error = copyout(&loff_cut, uap->basep, sizeof(int32_t)); } return (error); } #endif #if defined(COMPAT_FREEBSD11) int freebsd11_freebsd32_getdirentries(struct thread *td, struct freebsd11_freebsd32_getdirentries_args *uap) { long base; int32_t base32; int error; error = freebsd11_kern_getdirentries(td, uap->fd, uap->buf, uap->count, &base, NULL); if (error) return (error); if (uap->basep != NULL) { base32 = base; error = copyout(&base32, uap->basep, sizeof(int32_t)); } return (error); } int freebsd11_freebsd32_getdents(struct thread *td, struct freebsd11_freebsd32_getdents_args *uap) { struct freebsd11_freebsd32_getdirentries_args ap; ap.fd = uap->fd; ap.buf = uap->buf; ap.count = uap->count; ap.basep = NULL; return (freebsd11_freebsd32_getdirentries(td, &ap)); } #endif /* COMPAT_FREEBSD11 */ #ifdef COMPAT_FREEBSD6 /* versions with the 'int pad' argument */ int freebsd6_freebsd32_pread(struct thread *td, struct freebsd6_freebsd32_pread_args *uap) { return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, PAIR32TO64(off_t, uap->offset))); } int freebsd6_freebsd32_pwrite(struct thread *td, struct freebsd6_freebsd32_pwrite_args *uap) { return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, PAIR32TO64(off_t, uap->offset))); } int freebsd6_freebsd32_lseek(struct thread *td, struct freebsd6_freebsd32_lseek_args *uap) { int error; off_t pos; error = kern_lseek(td, uap->fd, PAIR32TO64(off_t, uap->offset), uap->whence); /* Expand the quad return into two parts for eax and edx */ pos = *(off_t *)(td->td_retval); td->td_retval[RETVAL_LO] = pos & 0xffffffff; /* %eax */ td->td_retval[RETVAL_HI] = pos >> 32; /* %edx */ return error; } int freebsd6_freebsd32_truncate(struct thread *td, struct freebsd6_freebsd32_truncate_args *uap) { return (kern_truncate(td, uap->path, UIO_USERSPACE, PAIR32TO64(off_t, uap->length))); } int freebsd6_freebsd32_ftruncate(struct thread *td, struct freebsd6_freebsd32_ftruncate_args *uap) { return (kern_ftruncate(td, uap->fd, PAIR32TO64(off_t, uap->length))); } #endif /* COMPAT_FREEBSD6 */ struct sf_hdtr32 { uint32_t headers; int hdr_cnt; uint32_t trailers; int trl_cnt; }; static int freebsd32_do_sendfile(struct thread *td, struct freebsd32_sendfile_args *uap, int compat) { struct sf_hdtr32 hdtr32; struct sf_hdtr hdtr; struct uio *hdr_uio, *trl_uio; struct file *fp; cap_rights_t rights; struct iovec32 *iov32; off_t offset, sbytes; int error; offset = PAIR32TO64(off_t, uap->offset); if (offset < 0) return (EINVAL); hdr_uio = trl_uio = NULL; if (uap->hdtr != NULL) { error = copyin(uap->hdtr, &hdtr32, sizeof(hdtr32)); if (error) goto out; PTRIN_CP(hdtr32, hdtr, headers); CP(hdtr32, hdtr, hdr_cnt); PTRIN_CP(hdtr32, hdtr, trailers); CP(hdtr32, hdtr, trl_cnt); if (hdtr.headers != NULL) { iov32 = PTRIN(hdtr32.headers); error = freebsd32_copyinuio(iov32, hdtr32.hdr_cnt, &hdr_uio); if (error) goto out; #ifdef COMPAT_FREEBSD4 /* * In FreeBSD < 5.0 the nbytes to send also included * the header. If compat is specified subtract the * header size from nbytes. */ if (compat) { if (uap->nbytes > hdr_uio->uio_resid) uap->nbytes -= hdr_uio->uio_resid; else uap->nbytes = 0; } #endif } if (hdtr.trailers != NULL) { iov32 = PTRIN(hdtr32.trailers); error = freebsd32_copyinuio(iov32, hdtr32.trl_cnt, &trl_uio); if (error) goto out; } } AUDIT_ARG_FD(uap->fd); if ((error = fget_read(td, uap->fd, - cap_rights_init(&rights, CAP_PREAD), &fp)) != 0) + cap_rights_init_one(&rights, CAP_PREAD), &fp)) != 0) goto out; error = fo_sendfile(fp, uap->s, hdr_uio, trl_uio, offset, uap->nbytes, &sbytes, uap->flags, td); fdrop(fp, td); if (uap->sbytes != NULL) copyout(&sbytes, uap->sbytes, sizeof(off_t)); out: if (hdr_uio) free(hdr_uio, M_IOV); if (trl_uio) free(trl_uio, M_IOV); return (error); } #ifdef COMPAT_FREEBSD4 int freebsd4_freebsd32_sendfile(struct thread *td, struct freebsd4_freebsd32_sendfile_args *uap) { return (freebsd32_do_sendfile(td, (struct freebsd32_sendfile_args *)uap, 1)); } #endif int freebsd32_sendfile(struct thread *td, struct freebsd32_sendfile_args *uap) { return (freebsd32_do_sendfile(td, uap, 0)); } static void copy_stat(struct stat *in, struct stat32 *out) { CP(*in, *out, st_dev); CP(*in, *out, st_ino); CP(*in, *out, st_mode); CP(*in, *out, st_nlink); CP(*in, *out, st_uid); CP(*in, *out, st_gid); CP(*in, *out, st_rdev); TS_CP(*in, *out, st_atim); TS_CP(*in, *out, st_mtim); TS_CP(*in, *out, st_ctim); CP(*in, *out, st_size); CP(*in, *out, st_blocks); CP(*in, *out, st_blksize); CP(*in, *out, st_flags); CP(*in, *out, st_gen); TS_CP(*in, *out, st_birthtim); out->st_padding0 = 0; out->st_padding1 = 0; #ifdef __STAT32_TIME_T_EXT out->st_atim_ext = 0; out->st_mtim_ext = 0; out->st_ctim_ext = 0; out->st_btim_ext = 0; #endif bzero(out->st_spare, sizeof(out->st_spare)); } #ifdef COMPAT_43 static void copy_ostat(struct stat *in, struct ostat32 *out) { bzero(out, sizeof(*out)); CP(*in, *out, st_dev); CP(*in, *out, st_ino); CP(*in, *out, st_mode); CP(*in, *out, st_nlink); CP(*in, *out, st_uid); CP(*in, *out, st_gid); CP(*in, *out, st_rdev); out->st_size = MIN(in->st_size, INT32_MAX); TS_CP(*in, *out, st_atim); TS_CP(*in, *out, st_mtim); TS_CP(*in, *out, st_ctim); CP(*in, *out, st_blksize); CP(*in, *out, st_blocks); CP(*in, *out, st_flags); CP(*in, *out, st_gen); } #endif #ifdef COMPAT_43 int ofreebsd32_stat(struct thread *td, struct ofreebsd32_stat_args *uap) { struct stat sb; struct ostat32 sb32; int error; error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error) return (error); copy_ostat(&sb, &sb32); error = copyout(&sb32, uap->ub, sizeof (sb32)); return (error); } #endif int freebsd32_fstat(struct thread *td, struct freebsd32_fstat_args *uap) { struct stat ub; struct stat32 ub32; int error; error = kern_fstat(td, uap->fd, &ub); if (error) return (error); copy_stat(&ub, &ub32); error = copyout(&ub32, uap->ub, sizeof(ub32)); return (error); } #ifdef COMPAT_43 int ofreebsd32_fstat(struct thread *td, struct ofreebsd32_fstat_args *uap) { struct stat ub; struct ostat32 ub32; int error; error = kern_fstat(td, uap->fd, &ub); if (error) return (error); copy_ostat(&ub, &ub32); error = copyout(&ub32, uap->ub, sizeof(ub32)); return (error); } #endif int freebsd32_fstatat(struct thread *td, struct freebsd32_fstatat_args *uap) { struct stat ub; struct stat32 ub32; int error; error = kern_statat(td, uap->flag, uap->fd, uap->path, UIO_USERSPACE, &ub, NULL); if (error) return (error); copy_stat(&ub, &ub32); error = copyout(&ub32, uap->buf, sizeof(ub32)); return (error); } #ifdef COMPAT_43 int ofreebsd32_lstat(struct thread *td, struct ofreebsd32_lstat_args *uap) { struct stat sb; struct ostat32 sb32; int error; error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error) return (error); copy_ostat(&sb, &sb32); error = copyout(&sb32, uap->ub, sizeof (sb32)); return (error); } #endif int freebsd32_fhstat(struct thread *td, struct freebsd32_fhstat_args *uap) { struct stat sb; struct stat32 sb32; struct fhandle fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); if (error != 0) return (error); error = kern_fhstat(td, fh, &sb); if (error != 0) return (error); copy_stat(&sb, &sb32); error = copyout(&sb32, uap->sb, sizeof (sb32)); return (error); } #if defined(COMPAT_FREEBSD11) extern int ino64_trunc_error; static int freebsd11_cvtstat32(struct stat *in, struct freebsd11_stat32 *out) { CP(*in, *out, st_ino); if (in->st_ino != out->st_ino) { switch (ino64_trunc_error) { default: case 0: break; case 1: return (EOVERFLOW); case 2: out->st_ino = UINT32_MAX; break; } } CP(*in, *out, st_nlink); if (in->st_nlink != out->st_nlink) { switch (ino64_trunc_error) { default: case 0: break; case 1: return (EOVERFLOW); case 2: out->st_nlink = UINT16_MAX; break; } } out->st_dev = in->st_dev; if (out->st_dev != in->st_dev) { switch (ino64_trunc_error) { default: break; case 1: return (EOVERFLOW); } } CP(*in, *out, st_mode); CP(*in, *out, st_uid); CP(*in, *out, st_gid); out->st_rdev = in->st_rdev; if (out->st_rdev != in->st_rdev) { switch (ino64_trunc_error) { default: break; case 1: return (EOVERFLOW); } } TS_CP(*in, *out, st_atim); TS_CP(*in, *out, st_mtim); TS_CP(*in, *out, st_ctim); CP(*in, *out, st_size); CP(*in, *out, st_blocks); CP(*in, *out, st_blksize); CP(*in, *out, st_flags); CP(*in, *out, st_gen); TS_CP(*in, *out, st_birthtim); out->st_lspare = 0; bzero((char *)&out->st_birthtim + sizeof(out->st_birthtim), sizeof(*out) - offsetof(struct freebsd11_stat32, st_birthtim) - sizeof(out->st_birthtim)); return (0); } int freebsd11_freebsd32_stat(struct thread *td, struct freebsd11_freebsd32_stat_args *uap) { struct stat sb; struct freebsd11_stat32 sb32; int error; error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); error = freebsd11_cvtstat32(&sb, &sb32); if (error == 0) error = copyout(&sb32, uap->ub, sizeof (sb32)); return (error); } int freebsd11_freebsd32_fstat(struct thread *td, struct freebsd11_freebsd32_fstat_args *uap) { struct stat sb; struct freebsd11_stat32 sb32; int error; error = kern_fstat(td, uap->fd, &sb); if (error != 0) return (error); error = freebsd11_cvtstat32(&sb, &sb32); if (error == 0) error = copyout(&sb32, uap->ub, sizeof (sb32)); return (error); } int freebsd11_freebsd32_fstatat(struct thread *td, struct freebsd11_freebsd32_fstatat_args *uap) { struct stat sb; struct freebsd11_stat32 sb32; int error; error = kern_statat(td, uap->flag, uap->fd, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); error = freebsd11_cvtstat32(&sb, &sb32); if (error == 0) error = copyout(&sb32, uap->buf, sizeof (sb32)); return (error); } int freebsd11_freebsd32_lstat(struct thread *td, struct freebsd11_freebsd32_lstat_args *uap) { struct stat sb; struct freebsd11_stat32 sb32; int error; error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); error = freebsd11_cvtstat32(&sb, &sb32); if (error == 0) error = copyout(&sb32, uap->ub, sizeof (sb32)); return (error); } int freebsd11_freebsd32_fhstat(struct thread *td, struct freebsd11_freebsd32_fhstat_args *uap) { struct stat sb; struct freebsd11_stat32 sb32; struct fhandle fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); if (error != 0) return (error); error = kern_fhstat(td, fh, &sb); if (error != 0) return (error); error = freebsd11_cvtstat32(&sb, &sb32); if (error == 0) error = copyout(&sb32, uap->sb, sizeof (sb32)); return (error); } #endif int freebsd32___sysctl(struct thread *td, struct freebsd32___sysctl_args *uap) { int error, name[CTL_MAXNAME]; size_t j, oldlen; uint32_t tmp; if (uap->namelen > CTL_MAXNAME || uap->namelen < 2) return (EINVAL); error = copyin(uap->name, name, uap->namelen * sizeof(int)); if (error) return (error); if (uap->oldlenp) { error = fueword32(uap->oldlenp, &tmp); oldlen = tmp; } else { oldlen = 0; } if (error != 0) return (EFAULT); error = userland_sysctl(td, name, uap->namelen, uap->old, &oldlen, 1, uap->new, uap->newlen, &j, SCTL_MASK32); if (error) return (error); if (uap->oldlenp) suword32(uap->oldlenp, j); return (0); } int freebsd32___sysctlbyname(struct thread *td, struct freebsd32___sysctlbyname_args *uap) { size_t oldlen, rv; int error; uint32_t tmp; if (uap->oldlenp != NULL) { error = fueword32(uap->oldlenp, &tmp); oldlen = tmp; } else { error = oldlen = 0; } if (error != 0) return (EFAULT); error = kern___sysctlbyname(td, uap->name, uap->namelen, uap->old, &oldlen, uap->new, uap->newlen, &rv, SCTL_MASK32, 1); if (error != 0) return (error); if (uap->oldlenp != NULL) error = suword32(uap->oldlenp, rv); return (error); } int freebsd32_jail(struct thread *td, struct freebsd32_jail_args *uap) { uint32_t version; int error; struct jail j; error = copyin(uap->jail, &version, sizeof(uint32_t)); if (error) return (error); switch (version) { case 0: { /* FreeBSD single IPv4 jails. */ struct jail32_v0 j32_v0; bzero(&j, sizeof(struct jail)); error = copyin(uap->jail, &j32_v0, sizeof(struct jail32_v0)); if (error) return (error); CP(j32_v0, j, version); PTRIN_CP(j32_v0, j, path); PTRIN_CP(j32_v0, j, hostname); j.ip4s = htonl(j32_v0.ip_number); /* jail_v0 is host order */ break; } case 1: /* * Version 1 was used by multi-IPv4 jail implementations * that never made it into the official kernel. */ return (EINVAL); case 2: /* JAIL_API_VERSION */ { /* FreeBSD multi-IPv4/IPv6,noIP jails. */ struct jail32 j32; error = copyin(uap->jail, &j32, sizeof(struct jail32)); if (error) return (error); CP(j32, j, version); PTRIN_CP(j32, j, path); PTRIN_CP(j32, j, hostname); PTRIN_CP(j32, j, jailname); CP(j32, j, ip4s); CP(j32, j, ip6s); PTRIN_CP(j32, j, ip4); PTRIN_CP(j32, j, ip6); break; } default: /* Sci-Fi jails are not supported, sorry. */ return (EINVAL); } return (kern_jail(td, &j)); } int freebsd32_jail_set(struct thread *td, struct freebsd32_jail_set_args *uap) { struct uio *auio; int error; /* Check that we have an even number of iovecs. */ if (uap->iovcnt & 1) return (EINVAL); error = freebsd32_copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = kern_jail_set(td, auio, uap->flags); free(auio, M_IOV); return (error); } int freebsd32_jail_get(struct thread *td, struct freebsd32_jail_get_args *uap) { struct iovec32 iov32; struct uio *auio; int error, i; /* Check that we have an even number of iovecs. */ if (uap->iovcnt & 1) return (EINVAL); error = freebsd32_copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = kern_jail_get(td, auio, uap->flags); if (error == 0) for (i = 0; i < uap->iovcnt; i++) { PTROUT_CP(auio->uio_iov[i], iov32, iov_base); CP(auio->uio_iov[i], iov32, iov_len); error = copyout(&iov32, uap->iovp + i, sizeof(iov32)); if (error != 0) break; } free(auio, M_IOV); return (error); } int freebsd32_sigaction(struct thread *td, struct freebsd32_sigaction_args *uap) { struct sigaction32 s32; struct sigaction sa, osa, *sap; int error; if (uap->act) { error = copyin(uap->act, &s32, sizeof(s32)); if (error) return (error); sa.sa_handler = PTRIN(s32.sa_u); CP(s32, sa, sa_flags); CP(s32, sa, sa_mask); sap = &sa; } else sap = NULL; error = kern_sigaction(td, uap->sig, sap, &osa, 0); if (error == 0 && uap->oact != NULL) { s32.sa_u = PTROUT(osa.sa_handler); CP(osa, s32, sa_flags); CP(osa, s32, sa_mask); error = copyout(&s32, uap->oact, sizeof(s32)); } return (error); } #ifdef COMPAT_FREEBSD4 int freebsd4_freebsd32_sigaction(struct thread *td, struct freebsd4_freebsd32_sigaction_args *uap) { struct sigaction32 s32; struct sigaction sa, osa, *sap; int error; if (uap->act) { error = copyin(uap->act, &s32, sizeof(s32)); if (error) return (error); sa.sa_handler = PTRIN(s32.sa_u); CP(s32, sa, sa_flags); CP(s32, sa, sa_mask); sap = &sa; } else sap = NULL; error = kern_sigaction(td, uap->sig, sap, &osa, KSA_FREEBSD4); if (error == 0 && uap->oact != NULL) { s32.sa_u = PTROUT(osa.sa_handler); CP(osa, s32, sa_flags); CP(osa, s32, sa_mask); error = copyout(&s32, uap->oact, sizeof(s32)); } return (error); } #endif #ifdef COMPAT_43 struct osigaction32 { u_int32_t sa_u; osigset_t sa_mask; int sa_flags; }; #define ONSIG 32 int ofreebsd32_sigaction(struct thread *td, struct ofreebsd32_sigaction_args *uap) { struct osigaction32 s32; struct sigaction sa, osa, *sap; int error; if (uap->signum <= 0 || uap->signum >= ONSIG) return (EINVAL); if (uap->nsa) { error = copyin(uap->nsa, &s32, sizeof(s32)); if (error) return (error); sa.sa_handler = PTRIN(s32.sa_u); CP(s32, sa, sa_flags); OSIG2SIG(s32.sa_mask, sa.sa_mask); sap = &sa; } else sap = NULL; error = kern_sigaction(td, uap->signum, sap, &osa, KSA_OSIGSET); if (error == 0 && uap->osa != NULL) { s32.sa_u = PTROUT(osa.sa_handler); CP(osa, s32, sa_flags); SIG2OSIG(osa.sa_mask, s32.sa_mask); error = copyout(&s32, uap->osa, sizeof(s32)); } return (error); } int ofreebsd32_sigprocmask(struct thread *td, struct ofreebsd32_sigprocmask_args *uap) { sigset_t set, oset; int error; OSIG2SIG(uap->mask, set); error = kern_sigprocmask(td, uap->how, &set, &oset, SIGPROCMASK_OLD); SIG2OSIG(oset, td->td_retval[0]); return (error); } int ofreebsd32_sigpending(struct thread *td, struct ofreebsd32_sigpending_args *uap) { struct proc *p = td->td_proc; sigset_t siglist; PROC_LOCK(p); siglist = p->p_siglist; SIGSETOR(siglist, td->td_siglist); PROC_UNLOCK(p); SIG2OSIG(siglist, td->td_retval[0]); return (0); } struct sigvec32 { u_int32_t sv_handler; int sv_mask; int sv_flags; }; int ofreebsd32_sigvec(struct thread *td, struct ofreebsd32_sigvec_args *uap) { struct sigvec32 vec; struct sigaction sa, osa, *sap; int error; if (uap->signum <= 0 || uap->signum >= ONSIG) return (EINVAL); if (uap->nsv) { error = copyin(uap->nsv, &vec, sizeof(vec)); if (error) return (error); sa.sa_handler = PTRIN(vec.sv_handler); OSIG2SIG(vec.sv_mask, sa.sa_mask); sa.sa_flags = vec.sv_flags; sa.sa_flags ^= SA_RESTART; sap = &sa; } else sap = NULL; error = kern_sigaction(td, uap->signum, sap, &osa, KSA_OSIGSET); if (error == 0 && uap->osv != NULL) { vec.sv_handler = PTROUT(osa.sa_handler); SIG2OSIG(osa.sa_mask, vec.sv_mask); vec.sv_flags = osa.sa_flags; vec.sv_flags &= ~SA_NOCLDWAIT; vec.sv_flags ^= SA_RESTART; error = copyout(&vec, uap->osv, sizeof(vec)); } return (error); } int ofreebsd32_sigblock(struct thread *td, struct ofreebsd32_sigblock_args *uap) { sigset_t set, oset; OSIG2SIG(uap->mask, set); kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0); SIG2OSIG(oset, td->td_retval[0]); return (0); } int ofreebsd32_sigsetmask(struct thread *td, struct ofreebsd32_sigsetmask_args *uap) { sigset_t set, oset; OSIG2SIG(uap->mask, set); kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0); SIG2OSIG(oset, td->td_retval[0]); return (0); } int ofreebsd32_sigsuspend(struct thread *td, struct ofreebsd32_sigsuspend_args *uap) { sigset_t mask; OSIG2SIG(uap->mask, mask); return (kern_sigsuspend(td, mask)); } struct sigstack32 { u_int32_t ss_sp; int ss_onstack; }; int ofreebsd32_sigstack(struct thread *td, struct ofreebsd32_sigstack_args *uap) { struct sigstack32 s32; struct sigstack nss, oss; int error = 0, unss; if (uap->nss != NULL) { error = copyin(uap->nss, &s32, sizeof(s32)); if (error) return (error); nss.ss_sp = PTRIN(s32.ss_sp); CP(s32, nss, ss_onstack); unss = 1; } else { unss = 0; } oss.ss_sp = td->td_sigstk.ss_sp; oss.ss_onstack = sigonstack(cpu_getstack(td)); if (unss) { td->td_sigstk.ss_sp = nss.ss_sp; td->td_sigstk.ss_size = 0; td->td_sigstk.ss_flags |= (nss.ss_onstack & SS_ONSTACK); td->td_pflags |= TDP_ALTSTACK; } if (uap->oss != NULL) { s32.ss_sp = PTROUT(oss.ss_sp); CP(oss, s32, ss_onstack); error = copyout(&s32, uap->oss, sizeof(s32)); } return (error); } #endif int freebsd32_nanosleep(struct thread *td, struct freebsd32_nanosleep_args *uap) { return (freebsd32_user_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME, uap->rqtp, uap->rmtp)); } int freebsd32_clock_nanosleep(struct thread *td, struct freebsd32_clock_nanosleep_args *uap) { int error; error = freebsd32_user_clock_nanosleep(td, uap->clock_id, uap->flags, uap->rqtp, uap->rmtp); return (kern_posix_error(td, error)); } static int freebsd32_user_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags, const struct timespec32 *ua_rqtp, struct timespec32 *ua_rmtp) { struct timespec32 rmt32, rqt32; struct timespec rmt, rqt; int error, error2; error = copyin(ua_rqtp, &rqt32, sizeof(rqt32)); if (error) return (error); CP(rqt32, rqt, tv_sec); CP(rqt32, rqt, tv_nsec); error = kern_clock_nanosleep(td, clock_id, flags, &rqt, &rmt); if (error == EINTR && ua_rmtp != NULL && (flags & TIMER_ABSTIME) == 0) { CP(rmt, rmt32, tv_sec); CP(rmt, rmt32, tv_nsec); error2 = copyout(&rmt32, ua_rmtp, sizeof(rmt32)); if (error2 != 0) error = error2; } return (error); } int freebsd32_clock_gettime(struct thread *td, struct freebsd32_clock_gettime_args *uap) { struct timespec ats; struct timespec32 ats32; int error; error = kern_clock_gettime(td, uap->clock_id, &ats); if (error == 0) { CP(ats, ats32, tv_sec); CP(ats, ats32, tv_nsec); error = copyout(&ats32, uap->tp, sizeof(ats32)); } return (error); } int freebsd32_clock_settime(struct thread *td, struct freebsd32_clock_settime_args *uap) { struct timespec ats; struct timespec32 ats32; int error; error = copyin(uap->tp, &ats32, sizeof(ats32)); if (error) return (error); CP(ats32, ats, tv_sec); CP(ats32, ats, tv_nsec); return (kern_clock_settime(td, uap->clock_id, &ats)); } int freebsd32_clock_getres(struct thread *td, struct freebsd32_clock_getres_args *uap) { struct timespec ts; struct timespec32 ts32; int error; if (uap->tp == NULL) return (0); error = kern_clock_getres(td, uap->clock_id, &ts); if (error == 0) { CP(ts, ts32, tv_sec); CP(ts, ts32, tv_nsec); error = copyout(&ts32, uap->tp, sizeof(ts32)); } return (error); } int freebsd32_ktimer_create(struct thread *td, struct freebsd32_ktimer_create_args *uap) { struct sigevent32 ev32; struct sigevent ev, *evp; int error, id; if (uap->evp == NULL) { evp = NULL; } else { evp = &ev; error = copyin(uap->evp, &ev32, sizeof(ev32)); if (error != 0) return (error); error = convert_sigevent32(&ev32, &ev); if (error != 0) return (error); } error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1); if (error == 0) { error = copyout(&id, uap->timerid, sizeof(int)); if (error != 0) kern_ktimer_delete(td, id); } return (error); } int freebsd32_ktimer_settime(struct thread *td, struct freebsd32_ktimer_settime_args *uap) { struct itimerspec32 val32, oval32; struct itimerspec val, oval, *ovalp; int error; error = copyin(uap->value, &val32, sizeof(val32)); if (error != 0) return (error); ITS_CP(val32, val); ovalp = uap->ovalue != NULL ? &oval : NULL; error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp); if (error == 0 && uap->ovalue != NULL) { ITS_CP(oval, oval32); error = copyout(&oval32, uap->ovalue, sizeof(oval32)); } return (error); } int freebsd32_ktimer_gettime(struct thread *td, struct freebsd32_ktimer_gettime_args *uap) { struct itimerspec32 val32; struct itimerspec val; int error; error = kern_ktimer_gettime(td, uap->timerid, &val); if (error == 0) { ITS_CP(val, val32); error = copyout(&val32, uap->value, sizeof(val32)); } return (error); } int freebsd32_clock_getcpuclockid2(struct thread *td, struct freebsd32_clock_getcpuclockid2_args *uap) { clockid_t clk_id; int error; error = kern_clock_getcpuclockid2(td, PAIR32TO64(id_t, uap->id), uap->which, &clk_id); if (error == 0) error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t)); return (error); } int freebsd32_thr_new(struct thread *td, struct freebsd32_thr_new_args *uap) { struct thr_param32 param32; struct thr_param param; int error; if (uap->param_size < 0 || uap->param_size > sizeof(struct thr_param32)) return (EINVAL); bzero(¶m, sizeof(struct thr_param)); bzero(¶m32, sizeof(struct thr_param32)); error = copyin(uap->param, ¶m32, uap->param_size); if (error != 0) return (error); param.start_func = PTRIN(param32.start_func); param.arg = PTRIN(param32.arg); param.stack_base = PTRIN(param32.stack_base); param.stack_size = param32.stack_size; param.tls_base = PTRIN(param32.tls_base); param.tls_size = param32.tls_size; param.child_tid = PTRIN(param32.child_tid); param.parent_tid = PTRIN(param32.parent_tid); param.flags = param32.flags; param.rtp = PTRIN(param32.rtp); param.spare[0] = PTRIN(param32.spare[0]); param.spare[1] = PTRIN(param32.spare[1]); param.spare[2] = PTRIN(param32.spare[2]); return (kern_thr_new(td, ¶m)); } int freebsd32_thr_suspend(struct thread *td, struct freebsd32_thr_suspend_args *uap) { struct timespec32 ts32; struct timespec ts, *tsp; int error; error = 0; tsp = NULL; if (uap->timeout != NULL) { error = copyin((const void *)uap->timeout, (void *)&ts32, sizeof(struct timespec32)); if (error != 0) return (error); ts.tv_sec = ts32.tv_sec; ts.tv_nsec = ts32.tv_nsec; tsp = &ts; } return (kern_thr_suspend(td, tsp)); } void siginfo_to_siginfo32(const siginfo_t *src, struct siginfo32 *dst) { bzero(dst, sizeof(*dst)); dst->si_signo = src->si_signo; dst->si_errno = src->si_errno; dst->si_code = src->si_code; dst->si_pid = src->si_pid; dst->si_uid = src->si_uid; dst->si_status = src->si_status; dst->si_addr = (uintptr_t)src->si_addr; dst->si_value.sival_int = src->si_value.sival_int; dst->si_timerid = src->si_timerid; dst->si_overrun = src->si_overrun; } #ifndef _FREEBSD32_SYSPROTO_H_ struct freebsd32_sigqueue_args { pid_t pid; int signum; /* union sigval32 */ int value; }; #endif int freebsd32_sigqueue(struct thread *td, struct freebsd32_sigqueue_args *uap) { union sigval sv; /* * On 32-bit ABIs, sival_int and sival_ptr are the same. * On 64-bit little-endian ABIs, the low bits are the same. * In 64-bit big-endian ABIs, sival_int overlaps with * sival_ptr's HIGH bits. We choose to support sival_int * rather than sival_ptr in this case as it seems to be * more common. */ bzero(&sv, sizeof(sv)); sv.sival_int = uap->value; return (kern_sigqueue(td, uap->pid, uap->signum, &sv)); } int freebsd32_sigtimedwait(struct thread *td, struct freebsd32_sigtimedwait_args *uap) { struct timespec32 ts32; struct timespec ts; struct timespec *timeout; sigset_t set; ksiginfo_t ksi; struct siginfo32 si32; int error; if (uap->timeout) { error = copyin(uap->timeout, &ts32, sizeof(ts32)); if (error) return (error); ts.tv_sec = ts32.tv_sec; ts.tv_nsec = ts32.tv_nsec; timeout = &ts; } else timeout = NULL; error = copyin(uap->set, &set, sizeof(set)); if (error) return (error); error = kern_sigtimedwait(td, set, &ksi, timeout); if (error) return (error); if (uap->info) { siginfo_to_siginfo32(&ksi.ksi_info, &si32); error = copyout(&si32, uap->info, sizeof(struct siginfo32)); } if (error == 0) td->td_retval[0] = ksi.ksi_signo; return (error); } /* * MPSAFE */ int freebsd32_sigwaitinfo(struct thread *td, struct freebsd32_sigwaitinfo_args *uap) { ksiginfo_t ksi; struct siginfo32 si32; sigset_t set; int error; error = copyin(uap->set, &set, sizeof(set)); if (error) return (error); error = kern_sigtimedwait(td, set, &ksi, NULL); if (error) return (error); if (uap->info) { siginfo_to_siginfo32(&ksi.ksi_info, &si32); error = copyout(&si32, uap->info, sizeof(struct siginfo32)); } if (error == 0) td->td_retval[0] = ksi.ksi_signo; return (error); } int freebsd32_cpuset_setid(struct thread *td, struct freebsd32_cpuset_setid_args *uap) { return (kern_cpuset_setid(td, uap->which, PAIR32TO64(id_t, uap->id), uap->setid)); } int freebsd32_cpuset_getid(struct thread *td, struct freebsd32_cpuset_getid_args *uap) { return (kern_cpuset_getid(td, uap->level, uap->which, PAIR32TO64(id_t, uap->id), uap->setid)); } int freebsd32_cpuset_getaffinity(struct thread *td, struct freebsd32_cpuset_getaffinity_args *uap) { return (kern_cpuset_getaffinity(td, uap->level, uap->which, PAIR32TO64(id_t,uap->id), uap->cpusetsize, uap->mask)); } int freebsd32_cpuset_setaffinity(struct thread *td, struct freebsd32_cpuset_setaffinity_args *uap) { return (kern_cpuset_setaffinity(td, uap->level, uap->which, PAIR32TO64(id_t,uap->id), uap->cpusetsize, uap->mask)); } int freebsd32_cpuset_getdomain(struct thread *td, struct freebsd32_cpuset_getdomain_args *uap) { return (kern_cpuset_getdomain(td, uap->level, uap->which, PAIR32TO64(id_t,uap->id), uap->domainsetsize, uap->mask, uap->policy)); } int freebsd32_cpuset_setdomain(struct thread *td, struct freebsd32_cpuset_setdomain_args *uap) { return (kern_cpuset_setdomain(td, uap->level, uap->which, PAIR32TO64(id_t,uap->id), uap->domainsetsize, uap->mask, uap->policy)); } int freebsd32_nmount(struct thread *td, struct freebsd32_nmount_args /* { struct iovec *iovp; unsigned int iovcnt; int flags; } */ *uap) { struct uio *auio; uint64_t flags; int error; /* * Mount flags are now 64-bits. On 32-bit archtectures only * 32-bits are passed in, but from here on everything handles * 64-bit flags correctly. */ flags = uap->flags; AUDIT_ARG_FFLAGS(flags); /* * Filter out MNT_ROOTFS. We do not want clients of nmount() in * userspace to set this flag, but we must filter it out if we want * MNT_UPDATE on the root file system to work. * MNT_ROOTFS should only be set by the kernel when mounting its * root file system. */ flags &= ~MNT_ROOTFS; /* * check that we have an even number of iovec's * and that we have at least two options. */ if ((uap->iovcnt & 1) || (uap->iovcnt < 4)) return (EINVAL); error = freebsd32_copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = vfs_donmount(td, flags, auio); free(auio, M_IOV); return error; } #if 0 int freebsd32_xxx(struct thread *td, struct freebsd32_xxx_args *uap) { struct yyy32 *p32, s32; struct yyy *p = NULL, s; struct xxx_arg ap; int error; if (uap->zzz) { error = copyin(uap->zzz, &s32, sizeof(s32)); if (error) return (error); /* translate in */ p = &s; } error = kern_xxx(td, p); if (error) return (error); if (uap->zzz) { /* translate out */ error = copyout(&s32, p32, sizeof(s32)); } return (error); } #endif int syscall32_module_handler(struct module *mod, int what, void *arg) { return (kern_syscall_module_handler(freebsd32_sysent, mod, what, arg)); } int syscall32_helper_register(struct syscall_helper_data *sd, int flags) { return (kern_syscall_helper_register(freebsd32_sysent, sd, flags)); } int syscall32_helper_unregister(struct syscall_helper_data *sd) { return (kern_syscall_helper_unregister(freebsd32_sysent, sd)); } int freebsd32_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) { int argc, envc, i; u_int32_t *vectp; char *stringp; uintptr_t destp, ustringp; struct freebsd32_ps_strings *arginfo; char canary[sizeof(long) * 8]; int32_t pagesizes32[MAXPAGESIZES]; size_t execpath_len; int error, szsigcode; /* * Calculate string base and vector table pointers. * Also deal with signal trampoline code for this exec type. */ if (imgp->execpath != NULL && imgp->auxargs != NULL) execpath_len = strlen(imgp->execpath) + 1; else execpath_len = 0; arginfo = (struct freebsd32_ps_strings *)curproc->p_sysent-> sv_psstrings; imgp->ps_strings = arginfo; if (imgp->proc->p_sysent->sv_sigcode_base == 0) szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); else szsigcode = 0; destp = (uintptr_t)arginfo; /* * install sigcode */ if (szsigcode != 0) { destp -= szsigcode; destp = rounddown2(destp, sizeof(uint32_t)); error = copyout(imgp->proc->p_sysent->sv_sigcode, (void *)destp, szsigcode); if (error != 0) return (error); } /* * Copy the image path for the rtld. */ if (execpath_len != 0) { destp -= execpath_len; imgp->execpathp = (void *)destp; error = copyout(imgp->execpath, imgp->execpathp, execpath_len); if (error != 0) return (error); } /* * Prepare the canary for SSP. */ arc4rand(canary, sizeof(canary), 0); destp -= sizeof(canary); imgp->canary = (void *)destp; error = copyout(canary, imgp->canary, sizeof(canary)); if (error != 0) return (error); imgp->canarylen = sizeof(canary); /* * Prepare the pagesizes array. */ for (i = 0; i < MAXPAGESIZES; i++) pagesizes32[i] = (uint32_t)pagesizes[i]; destp -= sizeof(pagesizes32); destp = rounddown2(destp, sizeof(uint32_t)); imgp->pagesizes = (void *)destp; error = copyout(pagesizes32, imgp->pagesizes, sizeof(pagesizes32)); if (error != 0) return (error); imgp->pagesizeslen = sizeof(pagesizes32); /* * Allocate room for the argument and environment strings. */ destp -= ARG_MAX - imgp->args->stringspace; destp = rounddown2(destp, sizeof(uint32_t)); ustringp = destp; exec_stackgap(imgp, &destp); if (imgp->auxargs) { /* * Allocate room on the stack for the ELF auxargs * array. It has up to AT_COUNT entries. */ destp -= AT_COUNT * sizeof(Elf32_Auxinfo); destp = rounddown2(destp, sizeof(uint32_t)); } vectp = (uint32_t *)destp; /* * Allocate room for the argv[] and env vectors including the * terminating NULL pointers. */ vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; /* * vectp also becomes our initial stack base */ *stack_base = (uintptr_t)vectp; stringp = imgp->args->begin_argv; argc = imgp->args->argc; envc = imgp->args->envc; /* * Copy out strings - arguments and environment. */ error = copyout(stringp, (void *)ustringp, ARG_MAX - imgp->args->stringspace); if (error != 0) return (error); /* * Fill in "ps_strings" struct for ps, w, etc. */ imgp->argv = vectp; if (suword32(&arginfo->ps_argvstr, (u_int32_t)(intptr_t)vectp) != 0 || suword32(&arginfo->ps_nargvstr, argc) != 0) return (EFAULT); /* * Fill in argument portion of vector table. */ for (; argc > 0; --argc) { if (suword32(vectp++, ustringp) != 0) return (EFAULT); while (*stringp++ != 0) ustringp++; ustringp++; } /* a null vector table pointer separates the argp's from the envp's */ if (suword32(vectp++, 0) != 0) return (EFAULT); imgp->envv = vectp; if (suword32(&arginfo->ps_envstr, (u_int32_t)(intptr_t)vectp) != 0 || suword32(&arginfo->ps_nenvstr, envc) != 0) return (EFAULT); /* * Fill in environment portion of vector table. */ for (; envc > 0; --envc) { if (suword32(vectp++, ustringp) != 0) return (EFAULT); while (*stringp++ != 0) ustringp++; ustringp++; } /* end of vector table is a null pointer */ if (suword32(vectp, 0) != 0) return (EFAULT); if (imgp->auxargs) { vectp++; error = imgp->sysent->sv_copyout_auxargs(imgp, (uintptr_t)vectp); if (error != 0) return (error); } return (0); } int freebsd32_kldstat(struct thread *td, struct freebsd32_kldstat_args *uap) { struct kld_file_stat *stat; struct kld32_file_stat *stat32; int error, version; if ((error = copyin(&uap->stat->version, &version, sizeof(version))) != 0) return (error); if (version != sizeof(struct kld32_file_stat_1) && version != sizeof(struct kld32_file_stat)) return (EINVAL); stat = malloc(sizeof(*stat), M_TEMP, M_WAITOK | M_ZERO); stat32 = malloc(sizeof(*stat32), M_TEMP, M_WAITOK | M_ZERO); error = kern_kldstat(td, uap->fileid, stat); if (error == 0) { bcopy(&stat->name[0], &stat32->name[0], sizeof(stat->name)); CP(*stat, *stat32, refs); CP(*stat, *stat32, id); PTROUT_CP(*stat, *stat32, address); CP(*stat, *stat32, size); bcopy(&stat->pathname[0], &stat32->pathname[0], sizeof(stat->pathname)); stat32->version = version; error = copyout(stat32, uap->stat, version); } free(stat, M_TEMP); free(stat32, M_TEMP); return (error); } int freebsd32_posix_fallocate(struct thread *td, struct freebsd32_posix_fallocate_args *uap) { int error; error = kern_posix_fallocate(td, uap->fd, PAIR32TO64(off_t, uap->offset), PAIR32TO64(off_t, uap->len)); return (kern_posix_error(td, error)); } int freebsd32_posix_fadvise(struct thread *td, struct freebsd32_posix_fadvise_args *uap) { int error; error = kern_posix_fadvise(td, uap->fd, PAIR32TO64(off_t, uap->offset), PAIR32TO64(off_t, uap->len), uap->advice); return (kern_posix_error(td, error)); } int convert_sigevent32(struct sigevent32 *sig32, struct sigevent *sig) { CP(*sig32, *sig, sigev_notify); switch (sig->sigev_notify) { case SIGEV_NONE: break; case SIGEV_THREAD_ID: CP(*sig32, *sig, sigev_notify_thread_id); /* FALLTHROUGH */ case SIGEV_SIGNAL: CP(*sig32, *sig, sigev_signo); PTRIN_CP(*sig32, *sig, sigev_value.sival_ptr); break; case SIGEV_KEVENT: CP(*sig32, *sig, sigev_notify_kqueue); CP(*sig32, *sig, sigev_notify_kevent_flags); PTRIN_CP(*sig32, *sig, sigev_value.sival_ptr); break; default: return (EINVAL); } return (0); } int freebsd32_procctl(struct thread *td, struct freebsd32_procctl_args *uap) { void *data; union { struct procctl_reaper_status rs; struct procctl_reaper_pids rp; struct procctl_reaper_kill rk; } x; union { struct procctl_reaper_pids32 rp; } x32; int error, error1, flags, signum; if (uap->com >= PROC_PROCCTL_MD_MIN) return (cpu_procctl(td, uap->idtype, PAIR32TO64(id_t, uap->id), uap->com, PTRIN(uap->data))); switch (uap->com) { case PROC_ASLR_CTL: case PROC_PROTMAX_CTL: case PROC_SPROTECT: case PROC_STACKGAP_CTL: case PROC_TRACE_CTL: case PROC_TRAPCAP_CTL: error = copyin(PTRIN(uap->data), &flags, sizeof(flags)); if (error != 0) return (error); data = &flags; break; case PROC_REAP_ACQUIRE: case PROC_REAP_RELEASE: if (uap->data != NULL) return (EINVAL); data = NULL; break; case PROC_REAP_STATUS: data = &x.rs; break; case PROC_REAP_GETPIDS: error = copyin(uap->data, &x32.rp, sizeof(x32.rp)); if (error != 0) return (error); CP(x32.rp, x.rp, rp_count); PTRIN_CP(x32.rp, x.rp, rp_pids); data = &x.rp; break; case PROC_REAP_KILL: error = copyin(uap->data, &x.rk, sizeof(x.rk)); if (error != 0) return (error); data = &x.rk; break; case PROC_ASLR_STATUS: case PROC_PROTMAX_STATUS: case PROC_STACKGAP_STATUS: case PROC_TRACE_STATUS: case PROC_TRAPCAP_STATUS: data = &flags; break; case PROC_PDEATHSIG_CTL: error = copyin(uap->data, &signum, sizeof(signum)); if (error != 0) return (error); data = &signum; break; case PROC_PDEATHSIG_STATUS: data = &signum; break; default: return (EINVAL); } error = kern_procctl(td, uap->idtype, PAIR32TO64(id_t, uap->id), uap->com, data); switch (uap->com) { case PROC_REAP_STATUS: if (error == 0) error = copyout(&x.rs, uap->data, sizeof(x.rs)); break; case PROC_REAP_KILL: error1 = copyout(&x.rk, uap->data, sizeof(x.rk)); if (error == 0) error = error1; break; case PROC_ASLR_STATUS: case PROC_PROTMAX_STATUS: case PROC_STACKGAP_STATUS: case PROC_TRACE_STATUS: case PROC_TRAPCAP_STATUS: if (error == 0) error = copyout(&flags, uap->data, sizeof(flags)); break; case PROC_PDEATHSIG_STATUS: if (error == 0) error = copyout(&signum, uap->data, sizeof(signum)); break; } return (error); } int freebsd32_fcntl(struct thread *td, struct freebsd32_fcntl_args *uap) { long tmp; switch (uap->cmd) { /* * Do unsigned conversion for arg when operation * interprets it as flags or pointer. */ case F_SETLK_REMOTE: case F_SETLKW: case F_SETLK: case F_GETLK: case F_SETFD: case F_SETFL: case F_OGETLK: case F_OSETLK: case F_OSETLKW: tmp = (unsigned int)(uap->arg); break; default: tmp = uap->arg; break; } return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, tmp)); } int freebsd32_ppoll(struct thread *td, struct freebsd32_ppoll_args *uap) { struct timespec32 ts32; struct timespec ts, *tsp; sigset_t set, *ssp; int error; if (uap->ts != NULL) { error = copyin(uap->ts, &ts32, sizeof(ts32)); if (error != 0) return (error); CP(ts32, ts, tv_sec); CP(ts32, ts, tv_nsec); tsp = &ts; } else tsp = NULL; if (uap->set != NULL) { error = copyin(uap->set, &set, sizeof(set)); if (error != 0) return (error); ssp = &set; } else ssp = NULL; return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp)); } int freebsd32_sched_rr_get_interval(struct thread *td, struct freebsd32_sched_rr_get_interval_args *uap) { struct timespec ts; struct timespec32 ts32; int error; error = kern_sched_rr_get_interval(td, uap->pid, &ts); if (error == 0) { CP(ts, ts32, tv_sec); CP(ts, ts32, tv_nsec); error = copyout(&ts32, uap->interval, sizeof(ts32)); } return (error); } static void timex_to_32(struct timex32 *dst, struct timex *src) { CP(*src, *dst, modes); CP(*src, *dst, offset); CP(*src, *dst, freq); CP(*src, *dst, maxerror); CP(*src, *dst, esterror); CP(*src, *dst, status); CP(*src, *dst, constant); CP(*src, *dst, precision); CP(*src, *dst, tolerance); CP(*src, *dst, ppsfreq); CP(*src, *dst, jitter); CP(*src, *dst, shift); CP(*src, *dst, stabil); CP(*src, *dst, jitcnt); CP(*src, *dst, calcnt); CP(*src, *dst, errcnt); CP(*src, *dst, stbcnt); } static void timex_from_32(struct timex *dst, struct timex32 *src) { CP(*src, *dst, modes); CP(*src, *dst, offset); CP(*src, *dst, freq); CP(*src, *dst, maxerror); CP(*src, *dst, esterror); CP(*src, *dst, status); CP(*src, *dst, constant); CP(*src, *dst, precision); CP(*src, *dst, tolerance); CP(*src, *dst, ppsfreq); CP(*src, *dst, jitter); CP(*src, *dst, shift); CP(*src, *dst, stabil); CP(*src, *dst, jitcnt); CP(*src, *dst, calcnt); CP(*src, *dst, errcnt); CP(*src, *dst, stbcnt); } int freebsd32_ntp_adjtime(struct thread *td, struct freebsd32_ntp_adjtime_args *uap) { struct timex tx; struct timex32 tx32; int error, retval; error = copyin(uap->tp, &tx32, sizeof(tx32)); if (error == 0) { timex_from_32(&tx, &tx32); error = kern_ntp_adjtime(td, &tx, &retval); if (error == 0) { timex_to_32(&tx32, &tx); error = copyout(&tx32, uap->tp, sizeof(tx32)); if (error == 0) td->td_retval[0] = retval; } } return (error); } diff --git a/sys/compat/linux/linux_event.c b/sys/compat/linux/linux_event.c index b4b4be1f7b49..dfb4588392cc 100644 --- a/sys/compat/linux/linux_event.c +++ b/sys/compat/linux/linux_event.c @@ -1,1047 +1,1048 @@ /*- * 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); } #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); ++(*nkevents); } if ((levents & LINUX_EPOLL_EVWR) != 0) { EV_SET(kevent++, fd, EVFILT_WRITE, kev_flags, 0, 0, 0); ++(*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) { 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; 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++) { 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(&rights, CAP_KQUEUE_CHANGE), &epfp); + 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(&rights, CAP_POLL_EVENT), &fp); + 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(&rights, CAP_KQUEUE_EVENT), &epfp); + 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; 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); } } diff --git a/sys/dev/aac/aac_linux.c b/sys/dev/aac/aac_linux.c index 02babb5c0576..3f2d8a827441 100644 --- a/sys/dev/aac/aac_linux.c +++ b/sys/dev/aac/aac_linux.c @@ -1,96 +1,97 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2002 Scott Long * 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("$FreeBSD$"); /* * Linux ioctl handler for the aac device driver */ #include #include #include #include #include #include #include #include #ifdef __amd64__ #include #include #else #include #include #endif #include /* There are multiple ioctl number ranges that need to be handled */ #define AAC_LINUX_IOCTL_MIN 0x0000 #define AAC_LINUX_IOCTL_MAX 0x21ff static linux_ioctl_function_t aac_linux_ioctl; static struct linux_ioctl_handler aac_linux_handler = {aac_linux_ioctl, AAC_LINUX_IOCTL_MIN, AAC_LINUX_IOCTL_MAX}; SYSINIT (aac_linux_register, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_ioctl_register_handler, &aac_linux_handler); SYSUNINIT(aac_linux_unregister, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_ioctl_unregister_handler, &aac_linux_handler); static int aac_linux_modevent(module_t mod, int type, void *data) { /* Do we care about any specific load/unload actions? */ return (0); } DEV_MODULE(aac_linux, aac_linux_modevent, NULL); MODULE_DEPEND(aac_linux, linux, 1, 1, 1); static int aac_linux_ioctl(struct thread *td, struct linux_ioctl_args *args) { cap_rights_t rights; struct file *fp; u_long cmd; int error; - error = fget(td, args->fd, cap_rights_init(&rights, CAP_IOCTL), &fp); + error = fget(td, args->fd, cap_rights_init_one(&rights, CAP_IOCTL), + &fp); if (error != 0) return (error); cmd = args->cmd; /* * Pass the ioctl off to our standard handler. */ error = (fo_ioctl(fp, cmd, (caddr_t)args->arg, td->td_ucred, td)); fdrop(fp, td); return (error); } diff --git a/sys/dev/aacraid/aacraid_linux.c b/sys/dev/aacraid/aacraid_linux.c index b58b8bacd7f3..7592a0d5a8cb 100644 --- a/sys/dev/aacraid/aacraid_linux.c +++ b/sys/dev/aacraid/aacraid_linux.c @@ -1,100 +1,100 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2002 Scott Long * Copyright (c) 2002-2010 Adaptec, Inc. * Copyright (c) 2010-2012 PMC-Sierra, Inc. * 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("$FreeBSD$"); /* * Linux ioctl handler for the aac device driver */ #include #include #include #include #include #include #include #include #ifdef __amd64__ #include #include #else #include #include #endif #include /* There are multiple ioctl number ranges that need to be handled */ #define AAC_LINUX_IOCTL_MIN 0x0000 #define AAC_LINUX_IOCTL_MAX 0x21ff static linux_ioctl_function_t aacraid_linux_ioctl; static struct linux_ioctl_handler aacraid_linux_handler = {aacraid_linux_ioctl, AAC_LINUX_IOCTL_MIN, AAC_LINUX_IOCTL_MAX}; SYSINIT (aacraid_linux_register, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_ioctl_register_handler, &aacraid_linux_handler); SYSUNINIT(aacraid_linux_unregister, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_ioctl_unregister_handler, &aacraid_linux_handler); static int aacraid_linux_modevent(module_t mod, int type, void *data) { /* Do we care about any specific load/unload actions? */ return (0); } DEV_MODULE(aacraid_linux, aacraid_linux_modevent, NULL); MODULE_DEPEND(aacraid_linux, linux, 1, 1, 1); static int aacraid_linux_ioctl(struct thread *td, struct linux_ioctl_args *args) { struct file *fp; cap_rights_t rights; u_long cmd; int error; if ((error = fget(td, args->fd, - cap_rights_init(&rights, CAP_IOCTL), + cap_rights_init_one(&rights, CAP_IOCTL), &fp)) != 0) { return (error); } cmd = args->cmd; /* * Pass the ioctl off to our standard handler. */ error = (fo_ioctl(fp, cmd, (caddr_t)args->arg, td->td_ucred, td)); fdrop(fp, td); return (error); } diff --git a/sys/dev/amr/amr_linux.c b/sys/dev/amr/amr_linux.c index 967907cc6846..9e0a5cac693e 100644 --- a/sys/dev/amr/amr_linux.c +++ b/sys/dev/amr/amr_linux.c @@ -1,87 +1,87 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2005 Paul Saab * 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 #if defined(__amd64__) /* Assume amd64 wants 32 bit Linux */ #include #include #else #include #include #endif #include /* There are multiple ioctl number ranges that need to be handled */ #define AMR_LINUX_IOCTL_MIN 0x6d00 #define AMR_LINUX_IOCTL_MAX 0x6d01 static linux_ioctl_function_t amr_linux_ioctl; static struct linux_ioctl_handler amr_linux_handler = {amr_linux_ioctl, AMR_LINUX_IOCTL_MIN, AMR_LINUX_IOCTL_MAX}; SYSINIT (amr_register, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_ioctl_register_handler, &amr_linux_handler); SYSUNINIT(amr_unregister, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_ioctl_unregister_handler, &amr_linux_handler); static int amr_linux_modevent(module_t mod, int cmd, void *data) { return (0); } DEV_MODULE(amr_linux, amr_linux_modevent, NULL); MODULE_DEPEND(amr, linux, 1, 1, 1); static int amr_linux_ioctl(struct thread *p, struct linux_ioctl_args *args) { cap_rights_t rights; struct file *fp; int error; - error = fget(p, args->fd, cap_rights_init(&rights, CAP_IOCTL), &fp); + error = fget(p, args->fd, cap_rights_init_one(&rights, CAP_IOCTL), &fp); if (error != 0) return (error); error = fo_ioctl(fp, args->cmd, (caddr_t)args->arg, p->td_ucred, p); fdrop(fp, p); return (error); } diff --git a/sys/dev/cxgbe/cxgbei/icl_cxgbei.c b/sys/dev/cxgbe/cxgbei/icl_cxgbei.c index f5d4f03476cb..5588d3cb9511 100644 --- a/sys/dev/cxgbe/cxgbei/icl_cxgbei.c +++ b/sys/dev/cxgbe/cxgbei/icl_cxgbei.c @@ -1,1098 +1,1098 @@ /*- * Copyright (c) 2012 The FreeBSD Foundation * Copyright (c) 2015 Chelsio Communications, Inc. * All rights reserved. * * This software was developed by Edward Tomasz Napierala under sponsorship * from the FreeBSD Foundation. * * 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. * */ /* * cxgbei implementation of iSCSI Common Layer kobj(9) interface. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #ifdef TCP_OFFLOAD #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "common/common.h" #include "common/t4_tcb.h" #include "tom/t4_tom.h" #include "cxgbei.h" SYSCTL_NODE(_kern_icl, OID_AUTO, cxgbei, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "Chelsio iSCSI offload"); static int coalesce = 1; SYSCTL_INT(_kern_icl_cxgbei, OID_AUTO, coalesce, CTLFLAG_RWTUN, &coalesce, 0, "Try to coalesce PDUs before sending"); static int partial_receive_len = 128 * 1024; SYSCTL_INT(_kern_icl_cxgbei, OID_AUTO, partial_receive_len, CTLFLAG_RWTUN, &partial_receive_len, 0, "Minimum read size for partially received " "data segment"); static int sendspace = 1048576; SYSCTL_INT(_kern_icl_cxgbei, OID_AUTO, sendspace, CTLFLAG_RWTUN, &sendspace, 0, "Default send socket buffer size"); static int recvspace = 1048576; SYSCTL_INT(_kern_icl_cxgbei, OID_AUTO, recvspace, CTLFLAG_RWTUN, &recvspace, 0, "Default receive socket buffer size"); static uma_zone_t prsv_zone; static volatile u_int icl_cxgbei_ncons; #define ICL_CONN_LOCK(X) mtx_lock(X->ic_lock) #define ICL_CONN_UNLOCK(X) mtx_unlock(X->ic_lock) #define ICL_CONN_LOCK_ASSERT(X) mtx_assert(X->ic_lock, MA_OWNED) #define ICL_CONN_LOCK_ASSERT_NOT(X) mtx_assert(X->ic_lock, MA_NOTOWNED) struct icl_pdu *icl_cxgbei_new_pdu(int); void icl_cxgbei_new_pdu_set_conn(struct icl_pdu *, struct icl_conn *); static icl_conn_new_pdu_t icl_cxgbei_conn_new_pdu; icl_conn_pdu_free_t icl_cxgbei_conn_pdu_free; static icl_conn_pdu_data_segment_length_t icl_cxgbei_conn_pdu_data_segment_length; static icl_conn_pdu_append_data_t icl_cxgbei_conn_pdu_append_data; static icl_conn_pdu_get_data_t icl_cxgbei_conn_pdu_get_data; static icl_conn_pdu_queue_t icl_cxgbei_conn_pdu_queue; static icl_conn_handoff_t icl_cxgbei_conn_handoff; static icl_conn_free_t icl_cxgbei_conn_free; static icl_conn_close_t icl_cxgbei_conn_close; static icl_conn_task_setup_t icl_cxgbei_conn_task_setup; static icl_conn_task_done_t icl_cxgbei_conn_task_done; static icl_conn_transfer_setup_t icl_cxgbei_conn_transfer_setup; static icl_conn_transfer_done_t icl_cxgbei_conn_transfer_done; static kobj_method_t icl_cxgbei_methods[] = { KOBJMETHOD(icl_conn_new_pdu, icl_cxgbei_conn_new_pdu), KOBJMETHOD(icl_conn_pdu_free, icl_cxgbei_conn_pdu_free), KOBJMETHOD(icl_conn_pdu_data_segment_length, icl_cxgbei_conn_pdu_data_segment_length), KOBJMETHOD(icl_conn_pdu_append_data, icl_cxgbei_conn_pdu_append_data), KOBJMETHOD(icl_conn_pdu_get_data, icl_cxgbei_conn_pdu_get_data), KOBJMETHOD(icl_conn_pdu_queue, icl_cxgbei_conn_pdu_queue), KOBJMETHOD(icl_conn_handoff, icl_cxgbei_conn_handoff), KOBJMETHOD(icl_conn_free, icl_cxgbei_conn_free), KOBJMETHOD(icl_conn_close, icl_cxgbei_conn_close), KOBJMETHOD(icl_conn_task_setup, icl_cxgbei_conn_task_setup), KOBJMETHOD(icl_conn_task_done, icl_cxgbei_conn_task_done), KOBJMETHOD(icl_conn_transfer_setup, icl_cxgbei_conn_transfer_setup), KOBJMETHOD(icl_conn_transfer_done, icl_cxgbei_conn_transfer_done), { 0, 0 } }; DEFINE_CLASS(icl_cxgbei, icl_cxgbei_methods, sizeof(struct icl_cxgbei_conn)); void icl_cxgbei_conn_pdu_free(struct icl_conn *ic, struct icl_pdu *ip) { #ifdef INVARIANTS struct icl_cxgbei_pdu *icp = ip_to_icp(ip); #endif MPASS(icp->icp_signature == CXGBEI_PDU_SIGNATURE); MPASS(ic == ip->ip_conn); MPASS(ip->ip_bhs_mbuf != NULL); m_freem(ip->ip_ahs_mbuf); m_freem(ip->ip_data_mbuf); m_freem(ip->ip_bhs_mbuf); /* storage for icl_cxgbei_pdu itself */ #ifdef DIAGNOSTIC if (__predict_true(ic != NULL)) refcount_release(&ic->ic_outstanding_pdus); #endif } struct icl_pdu * icl_cxgbei_new_pdu(int flags) { struct icl_cxgbei_pdu *icp; struct icl_pdu *ip; struct mbuf *m; uintptr_t a; m = m_gethdr(flags, MT_DATA); if (__predict_false(m == NULL)) return (NULL); a = roundup2(mtod(m, uintptr_t), _Alignof(struct icl_cxgbei_pdu)); icp = (struct icl_cxgbei_pdu *)a; bzero(icp, sizeof(*icp)); icp->icp_signature = CXGBEI_PDU_SIGNATURE; ip = &icp->ip; ip->ip_bhs_mbuf = m; a = roundup2((uintptr_t)(icp + 1), _Alignof(struct iscsi_bhs *)); ip->ip_bhs = (struct iscsi_bhs *)a; #ifdef INVARIANTS /* Everything must fit entirely in the mbuf. */ a = (uintptr_t)(ip->ip_bhs + 1); MPASS(a <= (uintptr_t)m + MSIZE); #endif bzero(ip->ip_bhs, sizeof(*ip->ip_bhs)); m->m_data = (void *)ip->ip_bhs; m->m_len = sizeof(struct iscsi_bhs); m->m_pkthdr.len = m->m_len; return (ip); } void icl_cxgbei_new_pdu_set_conn(struct icl_pdu *ip, struct icl_conn *ic) { ip->ip_conn = ic; #ifdef DIAGNOSTIC refcount_acquire(&ic->ic_outstanding_pdus); #endif } /* * Allocate icl_pdu with empty BHS to fill up by the caller. */ static struct icl_pdu * icl_cxgbei_conn_new_pdu(struct icl_conn *ic, int flags) { struct icl_pdu *ip; ip = icl_cxgbei_new_pdu(flags); if (__predict_false(ip == NULL)) return (NULL); icl_cxgbei_new_pdu_set_conn(ip, ic); return (ip); } static size_t icl_pdu_data_segment_length(const struct icl_pdu *request) { uint32_t len = 0; len += request->ip_bhs->bhs_data_segment_len[0]; len <<= 8; len += request->ip_bhs->bhs_data_segment_len[1]; len <<= 8; len += request->ip_bhs->bhs_data_segment_len[2]; return (len); } size_t icl_cxgbei_conn_pdu_data_segment_length(struct icl_conn *ic, const struct icl_pdu *request) { return (icl_pdu_data_segment_length(request)); } static struct mbuf * finalize_pdu(struct icl_cxgbei_conn *icc, struct icl_cxgbei_pdu *icp) { struct icl_pdu *ip = &icp->ip; uint8_t ulp_submode, padding; struct mbuf *m, *last; struct iscsi_bhs *bhs; /* * Fix up the data segment mbuf first. */ m = ip->ip_data_mbuf; ulp_submode = icc->ulp_submode; if (m) { last = m_last(m); /* * Round up the data segment to a 4B boundary. Pad with 0 if * necessary. There will definitely be room in the mbuf. */ padding = roundup2(ip->ip_data_len, 4) - ip->ip_data_len; if (padding) { bzero(mtod(last, uint8_t *) + last->m_len, padding); last->m_len += padding; } } else { MPASS(ip->ip_data_len == 0); ulp_submode &= ~ULP_CRC_DATA; padding = 0; } /* * Now the header mbuf that has the BHS. */ m = ip->ip_bhs_mbuf; MPASS(m->m_pkthdr.len == sizeof(struct iscsi_bhs)); MPASS(m->m_len == sizeof(struct iscsi_bhs)); bhs = ip->ip_bhs; bhs->bhs_data_segment_len[2] = ip->ip_data_len; bhs->bhs_data_segment_len[1] = ip->ip_data_len >> 8; bhs->bhs_data_segment_len[0] = ip->ip_data_len >> 16; /* "Convert" PDU to mbuf chain. Do not use icp/ip after this. */ m->m_pkthdr.len = sizeof(struct iscsi_bhs) + ip->ip_data_len + padding; m->m_next = ip->ip_data_mbuf; set_mbuf_ulp_submode(m, ulp_submode); #ifdef INVARIANTS bzero(icp, sizeof(*icp)); #endif #ifdef DIAGNOSTIC refcount_release(&icc->ic.ic_outstanding_pdus); #endif return (m); } int icl_cxgbei_conn_pdu_append_data(struct icl_conn *ic, struct icl_pdu *ip, const void *addr, size_t len, int flags) { struct mbuf *m; #ifdef INVARIANTS struct icl_cxgbei_pdu *icp = ip_to_icp(ip); #endif MPASS(icp->icp_signature == CXGBEI_PDU_SIGNATURE); MPASS(ic == ip->ip_conn); KASSERT(len > 0, ("%s: len is %jd", __func__, (intmax_t)len)); m = ip->ip_data_mbuf; if (m == NULL) { m = m_getjcl(M_NOWAIT, MT_DATA, 0, MJUM16BYTES); if (__predict_false(m == NULL)) return (ENOMEM); ip->ip_data_mbuf = m; } if (__predict_true(m_append(m, len, addr) != 0)) { ip->ip_data_len += len; MPASS(ip->ip_data_len <= ic->ic_max_data_segment_length); return (0); } else { if (flags & M_WAITOK) { CXGBE_UNIMPLEMENTED("fail safe append"); } ip->ip_data_len = m_length(m, NULL); return (1); } } void icl_cxgbei_conn_pdu_get_data(struct icl_conn *ic, struct icl_pdu *ip, size_t off, void *addr, size_t len) { struct icl_cxgbei_pdu *icp = ip_to_icp(ip); if (icp->icp_flags & ICPF_RX_DDP) return; /* data is DDP'ed, no need to copy */ m_copydata(ip->ip_data_mbuf, off, len, addr); } void icl_cxgbei_conn_pdu_queue(struct icl_conn *ic, struct icl_pdu *ip) { struct icl_cxgbei_conn *icc = ic_to_icc(ic); struct icl_cxgbei_pdu *icp = ip_to_icp(ip); struct socket *so = ic->ic_socket; struct toepcb *toep = icc->toep; struct inpcb *inp; struct mbuf *m; MPASS(ic == ip->ip_conn); MPASS(ip->ip_bhs_mbuf != NULL); /* The kernel doesn't generate PDUs with AHS. */ MPASS(ip->ip_ahs_mbuf == NULL && ip->ip_ahs_len == 0); ICL_CONN_LOCK_ASSERT(ic); /* NOTE: sowriteable without so_snd lock is a mostly harmless race. */ if (ic->ic_disconnecting || so == NULL || !sowriteable(so)) { icl_cxgbei_conn_pdu_free(ic, ip); return; } m = finalize_pdu(icc, icp); M_ASSERTPKTHDR(m); MPASS((m->m_pkthdr.len & 3) == 0); /* * Do not get inp from toep->inp as the toepcb might have detached * already. */ inp = sotoinpcb(so); INP_WLOCK(inp); if (__predict_false(inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) || __predict_false((toep->flags & TPF_ATTACHED) == 0)) m_freem(m); else { mbufq_enqueue(&toep->ulp_pduq, m); t4_push_pdus(icc->sc, toep, 0); } INP_WUNLOCK(inp); } static struct icl_conn * icl_cxgbei_new_conn(const char *name, struct mtx *lock) { struct icl_cxgbei_conn *icc; struct icl_conn *ic; refcount_acquire(&icl_cxgbei_ncons); icc = (struct icl_cxgbei_conn *)kobj_create(&icl_cxgbei_class, M_CXGBE, M_WAITOK | M_ZERO); icc->icc_signature = CXGBEI_CONN_SIGNATURE; STAILQ_INIT(&icc->rcvd_pdus); ic = &icc->ic; ic->ic_lock = lock; /* XXXNP: review. Most of these icl_conn fields aren't really used */ STAILQ_INIT(&ic->ic_to_send); cv_init(&ic->ic_send_cv, "icl_cxgbei_tx"); cv_init(&ic->ic_receive_cv, "icl_cxgbei_rx"); #ifdef DIAGNOSTIC refcount_init(&ic->ic_outstanding_pdus, 0); #endif /* This is a stop-gap value that will be corrected during handoff. */ ic->ic_max_data_segment_length = 16384; ic->ic_name = name; ic->ic_offload = "cxgbei"; ic->ic_unmapped = false; CTR2(KTR_CXGBE, "%s: icc %p", __func__, icc); return (ic); } void icl_cxgbei_conn_free(struct icl_conn *ic) { struct icl_cxgbei_conn *icc = ic_to_icc(ic); MPASS(icc->icc_signature == CXGBEI_CONN_SIGNATURE); CTR2(KTR_CXGBE, "%s: icc %p", __func__, icc); cv_destroy(&ic->ic_send_cv); cv_destroy(&ic->ic_receive_cv); kobj_delete((struct kobj *)icc, M_CXGBE); refcount_release(&icl_cxgbei_ncons); } static int icl_cxgbei_setsockopt(struct icl_conn *ic, struct socket *so, int sspace, int rspace) { struct sockopt opt; int error, one = 1, ss, rs; ss = max(sendspace, sspace); rs = max(recvspace, rspace); error = soreserve(so, ss, rs); if (error != 0) { icl_cxgbei_conn_close(ic); return (error); } SOCKBUF_LOCK(&so->so_snd); so->so_snd.sb_flags |= SB_AUTOSIZE; SOCKBUF_UNLOCK(&so->so_snd); SOCKBUF_LOCK(&so->so_rcv); so->so_rcv.sb_flags |= SB_AUTOSIZE; SOCKBUF_UNLOCK(&so->so_rcv); /* * Disable Nagle. */ bzero(&opt, sizeof(opt)); opt.sopt_dir = SOPT_SET; opt.sopt_level = IPPROTO_TCP; opt.sopt_name = TCP_NODELAY; opt.sopt_val = &one; opt.sopt_valsize = sizeof(one); error = sosetopt(so, &opt); if (error != 0) { icl_cxgbei_conn_close(ic); return (error); } return (0); } /* * Request/response structure used to find out the adapter offloading a socket. */ struct find_ofld_adapter_rr { struct socket *so; struct adapter *sc; /* result */ }; static void find_offload_adapter(struct adapter *sc, void *arg) { struct find_ofld_adapter_rr *fa = arg; struct socket *so = fa->so; struct tom_data *td = sc->tom_softc; struct tcpcb *tp; struct inpcb *inp; /* Non-TCP were filtered out earlier. */ MPASS(so->so_proto->pr_protocol == IPPROTO_TCP); if (fa->sc != NULL) return; /* Found already. */ if (td == NULL) return; /* TOE not enabled on this adapter. */ inp = sotoinpcb(so); INP_WLOCK(inp); if ((inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) == 0) { tp = intotcpcb(inp); if (tp->t_flags & TF_TOE && tp->tod == &td->tod) fa->sc = sc; /* Found. */ } INP_WUNLOCK(inp); } /* XXXNP: move this to t4_tom. */ static void send_iscsi_flowc_wr(struct adapter *sc, struct toepcb *toep, int maxlen) { struct wrqe *wr; struct fw_flowc_wr *flowc; const u_int nparams = 1; u_int flowclen; struct ofld_tx_sdesc *txsd = &toep->txsd[toep->txsd_pidx]; flowclen = sizeof(*flowc) + nparams * sizeof(struct fw_flowc_mnemval); wr = alloc_wrqe(roundup2(flowclen, 16), toep->ofld_txq); if (wr == NULL) { /* XXX */ panic("%s: allocation failure.", __func__); } flowc = wrtod(wr); memset(flowc, 0, wr->wr_len); flowc->op_to_nparams = htobe32(V_FW_WR_OP(FW_FLOWC_WR) | V_FW_FLOWC_WR_NPARAMS(nparams)); flowc->flowid_len16 = htonl(V_FW_WR_LEN16(howmany(flowclen, 16)) | V_FW_WR_FLOWID(toep->tid)); flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_TXDATAPLEN_MAX; flowc->mnemval[0].val = htobe32(maxlen); txsd->tx_credits = howmany(flowclen, 16); txsd->plen = 0; KASSERT(toep->tx_credits >= txsd->tx_credits && toep->txsd_avail > 0, ("%s: not enough credits (%d)", __func__, toep->tx_credits)); toep->tx_credits -= txsd->tx_credits; if (__predict_false(++toep->txsd_pidx == toep->txsd_total)) toep->txsd_pidx = 0; toep->txsd_avail--; t4_wrq_tx(sc, wr); } static void set_ulp_mode_iscsi(struct adapter *sc, struct toepcb *toep, int hcrc, int dcrc) { uint64_t val = ULP_MODE_ISCSI; if (hcrc) val |= ULP_CRC_HEADER << 4; if (dcrc) val |= ULP_CRC_DATA << 4; CTR4(KTR_CXGBE, "%s: tid %u, ULP_MODE_ISCSI, CRC hdr=%d data=%d", __func__, toep->tid, hcrc, dcrc); t4_set_tcb_field(sc, toep->ctrlq, toep, W_TCB_ULP_TYPE, V_TCB_ULP_TYPE(M_TCB_ULP_TYPE) | V_TCB_ULP_RAW(M_TCB_ULP_RAW), val, 0, 0); } /* * XXXNP: Who is responsible for cleaning up the socket if this returns with an * error? Review all error paths. * * XXXNP: What happens to the socket's fd reference if the operation is * successful, and how does that affect the socket's life cycle? */ int icl_cxgbei_conn_handoff(struct icl_conn *ic, int fd) { struct icl_cxgbei_conn *icc = ic_to_icc(ic); struct cxgbei_data *ci; struct find_ofld_adapter_rr fa; struct file *fp; struct socket *so; struct inpcb *inp; struct tcpcb *tp; struct toepcb *toep; cap_rights_t rights; int error; MPASS(icc->icc_signature == CXGBEI_CONN_SIGNATURE); ICL_CONN_LOCK_ASSERT_NOT(ic); /* * Steal the socket from userland. */ error = fget(curthread, fd, - cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp); + cap_rights_init_one(&rights, CAP_SOCK_CLIENT), &fp); if (error != 0) return (error); if (fp->f_type != DTYPE_SOCKET) { fdrop(fp, curthread); return (EINVAL); } so = fp->f_data; if (so->so_type != SOCK_STREAM || so->so_proto->pr_protocol != IPPROTO_TCP) { fdrop(fp, curthread); return (EINVAL); } ICL_CONN_LOCK(ic); if (ic->ic_socket != NULL) { ICL_CONN_UNLOCK(ic); fdrop(fp, curthread); return (EBUSY); } ic->ic_disconnecting = false; ic->ic_socket = so; fp->f_ops = &badfileops; fp->f_data = NULL; fdrop(fp, curthread); ICL_CONN_UNLOCK(ic); /* Find the adapter offloading this socket. */ fa.sc = NULL; fa.so = so; t4_iterate(find_offload_adapter, &fa); if (fa.sc == NULL) return (EINVAL); icc->sc = fa.sc; ci = icc->sc->iscsi_ulp_softc; inp = sotoinpcb(so); INP_WLOCK(inp); tp = intotcpcb(inp); if (inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) error = EBUSY; else { /* * socket could not have been "unoffloaded" if here. */ MPASS(tp->t_flags & TF_TOE); MPASS(tp->tod != NULL); MPASS(tp->t_toe != NULL); toep = tp->t_toe; MPASS(toep->vi->adapter == icc->sc); icc->toep = toep; icc->cwt = cxgbei_select_worker_thread(icc); /* * We maintain the _send_ DSL in this field just to have a * convenient way to assert that the kernel never sends * oversized PDUs. This field is otherwise unused in the driver * or the kernel. */ ic->ic_max_data_segment_length = ci->max_tx_pdu_len - ISCSI_BHS_SIZE; icc->ulp_submode = 0; if (ic->ic_header_crc32c) { icc->ulp_submode |= ULP_CRC_HEADER; ic->ic_max_data_segment_length -= ISCSI_HEADER_DIGEST_SIZE; } if (ic->ic_data_crc32c) { icc->ulp_submode |= ULP_CRC_DATA; ic->ic_max_data_segment_length -= ISCSI_DATA_DIGEST_SIZE; } so->so_options |= SO_NO_DDP; toep->params.ulp_mode = ULP_MODE_ISCSI; toep->ulpcb = icc; send_iscsi_flowc_wr(icc->sc, toep, ci->max_tx_pdu_len); set_ulp_mode_iscsi(icc->sc, toep, ic->ic_header_crc32c, ic->ic_data_crc32c); error = 0; } INP_WUNLOCK(inp); if (error == 0) { error = icl_cxgbei_setsockopt(ic, so, ci->max_tx_pdu_len, ci->max_rx_pdu_len); } return (error); } void icl_cxgbei_conn_close(struct icl_conn *ic) { struct icl_cxgbei_conn *icc = ic_to_icc(ic); struct icl_pdu *ip; struct socket *so; struct sockbuf *sb; struct inpcb *inp; struct toepcb *toep = icc->toep; MPASS(icc->icc_signature == CXGBEI_CONN_SIGNATURE); ICL_CONN_LOCK_ASSERT_NOT(ic); ICL_CONN_LOCK(ic); so = ic->ic_socket; if (ic->ic_disconnecting || so == NULL) { CTR4(KTR_CXGBE, "%s: icc %p (disconnecting = %d), so %p", __func__, icc, ic->ic_disconnecting, so); ICL_CONN_UNLOCK(ic); return; } ic->ic_disconnecting = true; /* These are unused in this driver right now. */ MPASS(STAILQ_EMPTY(&ic->ic_to_send)); MPASS(ic->ic_receive_pdu == NULL); #ifdef DIAGNOSTIC KASSERT(ic->ic_outstanding_pdus == 0, ("destroying session with %d outstanding PDUs", ic->ic_outstanding_pdus)); #endif ICL_CONN_UNLOCK(ic); CTR3(KTR_CXGBE, "%s: tid %d, icc %p", __func__, toep ? toep->tid : -1, icc); inp = sotoinpcb(so); sb = &so->so_rcv; INP_WLOCK(inp); if (toep != NULL) { /* NULL if connection was never offloaded. */ toep->ulpcb = NULL; mbufq_drain(&toep->ulp_pduq); SOCKBUF_LOCK(sb); if (icc->rx_flags & RXF_ACTIVE) { volatile u_int *p = &icc->rx_flags; SOCKBUF_UNLOCK(sb); INP_WUNLOCK(inp); while (*p & RXF_ACTIVE) pause("conclo", 1); INP_WLOCK(inp); SOCKBUF_LOCK(sb); } while (!STAILQ_EMPTY(&icc->rcvd_pdus)) { ip = STAILQ_FIRST(&icc->rcvd_pdus); STAILQ_REMOVE_HEAD(&icc->rcvd_pdus, ip_next); icl_cxgbei_conn_pdu_free(ic, ip); } SOCKBUF_UNLOCK(sb); } INP_WUNLOCK(inp); ICL_CONN_LOCK(ic); ic->ic_socket = NULL; ICL_CONN_UNLOCK(ic); /* * XXXNP: we should send RST instead of FIN when PDUs held in various * queues were purged instead of delivered reliably but soabort isn't * really general purpose and wouldn't do the right thing here. */ soclose(so); } int icl_cxgbei_conn_task_setup(struct icl_conn *ic, struct icl_pdu *ip, struct ccb_scsiio *csio, uint32_t *ittp, void **arg) { struct icl_cxgbei_conn *icc = ic_to_icc(ic); struct toepcb *toep = icc->toep; struct adapter *sc = icc->sc; struct cxgbei_data *ci = sc->iscsi_ulp_softc; struct ppod_region *pr = &ci->pr; struct ppod_reservation *prsv; uint32_t itt; int rc = 0; /* This is for the offload driver's state. Must not be set already. */ MPASS(arg != NULL); MPASS(*arg == NULL); if ((csio->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_IN || csio->dxfer_len < ci->ddp_threshold) { no_ddp: /* * No DDP for this I/O. Allocate an ITT (based on the one * passed in) that cannot be a valid hardware DDP tag in the * iSCSI region. */ itt = *ittp & M_PPOD_TAG; itt = V_PPOD_TAG(itt) | pr->pr_invalid_bit; *ittp = htobe32(itt); MPASS(*arg == NULL); /* State is maintained for DDP only. */ if (rc != 0) counter_u64_add(ci->ddp_setup_error, 1); return (0); } /* * Reserve resources for DDP, update the itt that should be used in the * PDU, and save DDP specific state for this I/O in *arg. */ prsv = uma_zalloc(prsv_zone, M_NOWAIT); if (prsv == NULL) { rc = ENOMEM; goto no_ddp; } /* XXX add support for all CAM_DATA_ types */ MPASS((csio->ccb_h.flags & CAM_DATA_MASK) == CAM_DATA_VADDR); rc = t4_alloc_page_pods_for_buf(pr, (vm_offset_t)csio->data_ptr, csio->dxfer_len, prsv); if (rc != 0) { uma_zfree(prsv_zone, prsv); goto no_ddp; } rc = t4_write_page_pods_for_buf(sc, toep->ofld_txq, toep->tid, prsv, (vm_offset_t)csio->data_ptr, csio->dxfer_len); if (rc != 0) { t4_free_page_pods(prsv); uma_zfree(prsv_zone, prsv); goto no_ddp; } *ittp = htobe32(prsv->prsv_tag); *arg = prsv; counter_u64_add(ci->ddp_setup_ok, 1); return (0); } void icl_cxgbei_conn_task_done(struct icl_conn *ic, void *arg) { if (arg != NULL) { struct ppod_reservation *prsv = arg; t4_free_page_pods(prsv); uma_zfree(prsv_zone, prsv); } } /* XXXNP: PDU should be passed in as parameter, like on the initiator. */ #define io_to_request_pdu(io) ((io)->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr) #define io_to_ppod_reservation(io) ((io)->io_hdr.ctl_private[CTL_PRIV_FRONTEND2].ptr) int icl_cxgbei_conn_transfer_setup(struct icl_conn *ic, union ctl_io *io, uint32_t *tttp, void **arg) { struct icl_cxgbei_conn *icc = ic_to_icc(ic); struct toepcb *toep = icc->toep; struct ctl_scsiio *ctsio = &io->scsiio; struct adapter *sc = icc->sc; struct cxgbei_data *ci = sc->iscsi_ulp_softc; struct ppod_region *pr = &ci->pr; struct ppod_reservation *prsv; uint32_t ttt; int xferlen, rc = 0, alias; /* This is for the offload driver's state. Must not be set already. */ MPASS(arg != NULL); MPASS(*arg == NULL); if (ctsio->ext_data_filled == 0) { int first_burst; struct icl_pdu *ip = io_to_request_pdu(io); vm_offset_t buf; #ifdef INVARIANTS struct icl_cxgbei_pdu *icp = ip_to_icp(ip); MPASS(icp->icp_signature == CXGBEI_PDU_SIGNATURE); MPASS(ic == ip->ip_conn); MPASS(ip->ip_bhs_mbuf != NULL); #endif first_burst = icl_pdu_data_segment_length(ip); /* * Note that ICL calls conn_transfer_setup even if the first * burst had everything and there's nothing left to transfer. */ MPASS(ctsio->kern_data_len >= first_burst); xferlen = ctsio->kern_data_len; if (xferlen - first_burst < ci->ddp_threshold) { no_ddp: /* * No DDP for this transfer. Allocate a TTT (based on * the one passed in) that cannot be a valid hardware * DDP tag in the iSCSI region. */ ttt = *tttp & M_PPOD_TAG; ttt = V_PPOD_TAG(ttt) | pr->pr_invalid_bit; *tttp = htobe32(ttt); MPASS(io_to_ppod_reservation(io) == NULL); if (rc != 0) counter_u64_add(ci->ddp_setup_error, 1); return (0); } if (ctsio->kern_sg_entries == 0) buf = (vm_offset_t)ctsio->kern_data_ptr; else if (ctsio->kern_sg_entries == 1) { struct ctl_sg_entry *sgl = (void *)ctsio->kern_data_ptr; MPASS(sgl->len == xferlen); buf = (vm_offset_t)sgl->addr; } else { rc = EAGAIN; /* XXX implement */ goto no_ddp; } /* * Reserve resources for DDP, update the ttt that should be used * in the PDU, and save DDP specific state for this I/O. */ MPASS(io_to_ppod_reservation(io) == NULL); prsv = uma_zalloc(prsv_zone, M_NOWAIT); if (prsv == NULL) { rc = ENOMEM; goto no_ddp; } rc = t4_alloc_page_pods_for_buf(pr, buf, xferlen, prsv); if (rc != 0) { uma_zfree(prsv_zone, prsv); goto no_ddp; } rc = t4_write_page_pods_for_buf(sc, toep->ofld_txq, toep->tid, prsv, buf, xferlen); if (rc != 0) { t4_free_page_pods(prsv); uma_zfree(prsv_zone, prsv); goto no_ddp; } *tttp = htobe32(prsv->prsv_tag); io_to_ppod_reservation(io) = prsv; *arg = ctsio; counter_u64_add(ci->ddp_setup_ok, 1); return (0); } /* * In the middle of an I/O. A non-NULL page pod reservation indicates * that a DDP buffer is being used for the I/O. */ prsv = io_to_ppod_reservation(ctsio); if (prsv == NULL) goto no_ddp; alias = (prsv->prsv_tag & pr->pr_alias_mask) >> pr->pr_alias_shift; alias++; prsv->prsv_tag &= ~pr->pr_alias_mask; prsv->prsv_tag |= alias << pr->pr_alias_shift & pr->pr_alias_mask; *tttp = htobe32(prsv->prsv_tag); *arg = ctsio; return (0); } void icl_cxgbei_conn_transfer_done(struct icl_conn *ic, void *arg) { struct ctl_scsiio *ctsio = arg; if (ctsio != NULL && ctsio->kern_data_len == ctsio->ext_data_filled) { struct ppod_reservation *prsv; prsv = io_to_ppod_reservation(ctsio); MPASS(prsv != NULL); t4_free_page_pods(prsv); uma_zfree(prsv_zone, prsv); } } static void cxgbei_limits(struct adapter *sc, void *arg) { struct icl_drv_limits *idl = arg; struct cxgbei_data *ci; int max_dsl; if (begin_synchronized_op(sc, NULL, HOLD_LOCK, "t4lims") != 0) return; if (uld_active(sc, ULD_ISCSI)) { ci = sc->iscsi_ulp_softc; MPASS(ci != NULL); /* * AHS is not supported by the kernel so we'll not account for * it either in our PDU len -> data segment len conversions. */ max_dsl = ci->max_rx_pdu_len - ISCSI_BHS_SIZE - ISCSI_HEADER_DIGEST_SIZE - ISCSI_DATA_DIGEST_SIZE; if (idl->idl_max_recv_data_segment_length > max_dsl) idl->idl_max_recv_data_segment_length = max_dsl; max_dsl = ci->max_tx_pdu_len - ISCSI_BHS_SIZE - ISCSI_HEADER_DIGEST_SIZE - ISCSI_DATA_DIGEST_SIZE; if (idl->idl_max_send_data_segment_length > max_dsl) idl->idl_max_send_data_segment_length = max_dsl; } end_synchronized_op(sc, LOCK_HELD); } static int icl_cxgbei_limits(struct icl_drv_limits *idl) { /* Maximum allowed by the RFC. cxgbei_limits will clip them. */ idl->idl_max_recv_data_segment_length = (1 << 24) - 1; idl->idl_max_send_data_segment_length = (1 << 24) - 1; /* These are somewhat arbitrary. */ idl->idl_max_burst_length = 2 * 1024 * 1024; idl->idl_first_burst_length = 8192; t4_iterate(cxgbei_limits, idl); return (0); } int icl_cxgbei_mod_load(void) { int rc; /* * Space to track pagepod reservations. */ prsv_zone = uma_zcreate("Pagepod reservations", sizeof(struct ppod_reservation), NULL, NULL, NULL, NULL, UMA_ALIGN_CACHE, 0); refcount_init(&icl_cxgbei_ncons, 0); rc = icl_register("cxgbei", false, -100, icl_cxgbei_limits, icl_cxgbei_new_conn); return (rc); } int icl_cxgbei_mod_unload(void) { if (icl_cxgbei_ncons != 0) return (EBUSY); icl_unregister("cxgbei", false); uma_zdestroy(prsv_zone); return (0); } #endif diff --git a/sys/dev/filemon/filemon_wrapper.c b/sys/dev/filemon/filemon_wrapper.c index 52dd3bb483eb..4e64d8254f8d 100644 --- a/sys/dev/filemon/filemon_wrapper.c +++ b/sys/dev/filemon/filemon_wrapper.c @@ -1,460 +1,460 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2011, David E. O'Brien. * Copyright (c) 2009-2011, Juniper Networks, Inc. * Copyright (c) 2015-2016, EMC Corp. * 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 JUNIPER NETWORKS 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 JUNIPER NETWORKS 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 static void filemon_output_event(struct filemon *filemon, const char *fmt, ...) __printflike(2, 3); static eventhandler_tag filemon_exec_tag; static eventhandler_tag filemon_exit_tag; static eventhandler_tag filemon_fork_tag; static void filemon_output(struct filemon *filemon, char *msg, size_t len) { struct uio auio; struct iovec aiov; int error; if (filemon->fp == NULL) return; aiov.iov_base = msg; aiov.iov_len = len; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_resid = len; auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_WRITE; auio.uio_td = curthread; auio.uio_offset = (off_t) -1; if (filemon->fp->f_type == DTYPE_VNODE) bwillwrite(); error = fo_write(filemon->fp, &auio, filemon->cred, 0, curthread); if (error != 0 && filemon->error == 0) filemon->error = error; } static void filemon_output_event(struct filemon *filemon, const char *fmt, ...) { va_list ap; size_t len; va_start(ap, fmt); len = vsnprintf(filemon->msgbufr, sizeof(filemon->msgbufr), fmt, ap); va_end(ap); /* The event is truncated but still worth logging. */ if (len >= sizeof(filemon->msgbufr)) len = sizeof(filemon->msgbufr) - 1; filemon_output(filemon, filemon->msgbufr, len); } static int filemon_wrapper_chdir(struct thread *td, struct chdir_args *uap) { int error, ret; struct filemon *filemon; if ((ret = sys_chdir(td, uap)) == 0) { if ((filemon = filemon_proc_get(curproc)) != NULL) { if ((error = copyinstr(uap->path, filemon->fname1, sizeof(filemon->fname1), NULL)) != 0) { filemon->error = error; goto copyfail; } filemon_output_event(filemon, "C %d %s\n", curproc->p_pid, filemon->fname1); copyfail: filemon_drop(filemon); } } return (ret); } static void filemon_event_process_exec(void *arg __unused, struct proc *p, struct image_params *imgp) { struct filemon *filemon; if ((filemon = filemon_proc_get(p)) != NULL) { filemon_output_event(filemon, "E %d %s\n", p->p_pid, imgp->execpath != NULL ? imgp->execpath : ""); /* If the credentials changed then cease tracing. */ if (imgp->newcred != NULL && imgp->credential_setid && priv_check_cred(filemon->cred, PRIV_DEBUG_DIFFCRED) != 0) { /* * It may have changed to NULL already, but * will not be re-attached by anything else. */ if (p->p_filemon != NULL) { KASSERT(p->p_filemon == filemon, ("%s: proc %p didn't have expected" " filemon %p", __func__, p, filemon)); filemon_proc_drop(p); } } filemon_drop(filemon); } } static void _filemon_wrapper_openat(struct thread *td, const char *upath, int flags, int fd) { int error; struct file *fp; struct filemon *filemon; char *atpath, *freepath; cap_rights_t rights; if ((filemon = filemon_proc_get(curproc)) != NULL) { atpath = ""; freepath = NULL; fp = NULL; if ((error = copyinstr(upath, filemon->fname1, sizeof(filemon->fname1), NULL)) != 0) { filemon->error = error; goto copyfail; } if (filemon->fname1[0] != '/' && fd != AT_FDCWD) { /* * rats - we cannot do too much about this. * the trace should show a dir we read * recently.. output an A record as a clue * until we can do better. * XXX: This may be able to come out with * the namecache lookup now. */ filemon_output_event(filemon, "A %d %s\n", curproc->p_pid, filemon->fname1); /* * Try to resolve the path from the vnode using the * namecache. It may be inaccurate, but better * than nothing. */ if (getvnode(td, fd, - cap_rights_init(&rights, CAP_LOOKUP), &fp) == 0) { + cap_rights_init_one(&rights, CAP_LOOKUP), &fp) == 0) { vn_fullpath(fp->f_vnode, &atpath, &freepath); } } if (flags & O_RDWR) { /* * We'll get the W record below, but need * to also output an R to distinguish from * O_WRONLY. */ filemon_output_event(filemon, "R %d %s%s%s\n", curproc->p_pid, atpath, atpath[0] != '\0' ? "/" : "", filemon->fname1); } filemon_output_event(filemon, "%c %d %s%s%s\n", (flags & O_ACCMODE) ? 'W':'R', curproc->p_pid, atpath, atpath[0] != '\0' ? "/" : "", filemon->fname1); copyfail: filemon_drop(filemon); if (fp != NULL) fdrop(fp, td); free(freepath, M_TEMP); } } static int filemon_wrapper_open(struct thread *td, struct open_args *uap) { int ret; if ((ret = sys_open(td, uap)) == 0) _filemon_wrapper_openat(td, uap->path, uap->flags, AT_FDCWD); return (ret); } static int filemon_wrapper_openat(struct thread *td, struct openat_args *uap) { int ret; if ((ret = sys_openat(td, uap)) == 0) _filemon_wrapper_openat(td, uap->path, uap->flag, uap->fd); return (ret); } static int filemon_wrapper_rename(struct thread *td, struct rename_args *uap) { int error, ret; struct filemon *filemon; if ((ret = sys_rename(td, uap)) == 0) { if ((filemon = filemon_proc_get(curproc)) != NULL) { if (((error = copyinstr(uap->from, filemon->fname1, sizeof(filemon->fname1), NULL)) != 0) || ((error = copyinstr(uap->to, filemon->fname2, sizeof(filemon->fname2), NULL)) != 0)) { filemon->error = error; goto copyfail; } filemon_output_event(filemon, "M %d '%s' '%s'\n", curproc->p_pid, filemon->fname1, filemon->fname2); copyfail: filemon_drop(filemon); } } return (ret); } static void _filemon_wrapper_link(struct thread *td, const char *upath1, const char *upath2) { struct filemon *filemon; int error; if ((filemon = filemon_proc_get(curproc)) != NULL) { if (((error = copyinstr(upath1, filemon->fname1, sizeof(filemon->fname1), NULL)) != 0) || ((error = copyinstr(upath2, filemon->fname2, sizeof(filemon->fname2), NULL)) != 0)) { filemon->error = error; goto copyfail; } filemon_output_event(filemon, "L %d '%s' '%s'\n", curproc->p_pid, filemon->fname1, filemon->fname2); copyfail: filemon_drop(filemon); } } static int filemon_wrapper_link(struct thread *td, struct link_args *uap) { int ret; if ((ret = sys_link(td, uap)) == 0) _filemon_wrapper_link(td, uap->path, uap->link); return (ret); } static int filemon_wrapper_symlink(struct thread *td, struct symlink_args *uap) { int ret; if ((ret = sys_symlink(td, uap)) == 0) _filemon_wrapper_link(td, uap->path, uap->link); return (ret); } static int filemon_wrapper_linkat(struct thread *td, struct linkat_args *uap) { int ret; if ((ret = sys_linkat(td, uap)) == 0) _filemon_wrapper_link(td, uap->path1, uap->path2); return (ret); } static void filemon_event_process_exit(void *arg __unused, struct proc *p) { struct filemon *filemon; if ((filemon = filemon_proc_get(p)) != NULL) { filemon_output_event(filemon, "X %d %d %d\n", p->p_pid, p->p_xexit, p->p_xsig); /* * filemon_untrack_processes() may have dropped this p_filemon * already while in filemon_proc_get() before acquiring the * filemon lock. */ KASSERT(p->p_filemon == NULL || p->p_filemon == filemon, ("%s: p %p was attached while exiting, expected " "filemon %p or NULL", __func__, p, filemon)); if (p->p_filemon == filemon) filemon_proc_drop(p); filemon_drop(filemon); } } static int filemon_wrapper_unlink(struct thread *td, struct unlink_args *uap) { int error, ret; struct filemon *filemon; if ((ret = sys_unlink(td, uap)) == 0) { if ((filemon = filemon_proc_get(curproc)) != NULL) { if ((error = copyinstr(uap->path, filemon->fname1, sizeof(filemon->fname1), NULL)) != 0) { filemon->error = error; goto copyfail; } filemon_output_event(filemon, "D %d %s\n", curproc->p_pid, filemon->fname1); copyfail: filemon_drop(filemon); } } return (ret); } static void filemon_event_process_fork(void *arg __unused, struct proc *p1, struct proc *p2, int flags __unused) { struct filemon *filemon; if ((filemon = filemon_proc_get(p1)) != NULL) { filemon_output_event(filemon, "F %d %d\n", p1->p_pid, p2->p_pid); /* * filemon_untrack_processes() or * filemon_ioctl(FILEMON_SET_PID) may have changed the parent's * p_filemon while in filemon_proc_get() before acquiring the * filemon lock. Only inherit if the parent is still traced by * this filemon. */ if (p1->p_filemon == filemon) { PROC_LOCK(p2); /* * It may have been attached to already by a new * filemon. */ if (p2->p_filemon == NULL) { p2->p_filemon = filemon_acquire(filemon); ++filemon->proccnt; } PROC_UNLOCK(p2); } filemon_drop(filemon); } } static void filemon_wrapper_install(void) { sysent[SYS_chdir].sy_call = (sy_call_t *) filemon_wrapper_chdir; sysent[SYS_open].sy_call = (sy_call_t *) filemon_wrapper_open; sysent[SYS_openat].sy_call = (sy_call_t *) filemon_wrapper_openat; sysent[SYS_rename].sy_call = (sy_call_t *) filemon_wrapper_rename; sysent[SYS_unlink].sy_call = (sy_call_t *) filemon_wrapper_unlink; sysent[SYS_link].sy_call = (sy_call_t *) filemon_wrapper_link; sysent[SYS_symlink].sy_call = (sy_call_t *) filemon_wrapper_symlink; sysent[SYS_linkat].sy_call = (sy_call_t *) filemon_wrapper_linkat; #if defined(COMPAT_FREEBSD32) freebsd32_sysent[FREEBSD32_SYS_chdir].sy_call = (sy_call_t *) filemon_wrapper_chdir; freebsd32_sysent[FREEBSD32_SYS_open].sy_call = (sy_call_t *) filemon_wrapper_open; freebsd32_sysent[FREEBSD32_SYS_openat].sy_call = (sy_call_t *) filemon_wrapper_openat; freebsd32_sysent[FREEBSD32_SYS_rename].sy_call = (sy_call_t *) filemon_wrapper_rename; freebsd32_sysent[FREEBSD32_SYS_unlink].sy_call = (sy_call_t *) filemon_wrapper_unlink; freebsd32_sysent[FREEBSD32_SYS_link].sy_call = (sy_call_t *) filemon_wrapper_link; freebsd32_sysent[FREEBSD32_SYS_symlink].sy_call = (sy_call_t *) filemon_wrapper_symlink; freebsd32_sysent[FREEBSD32_SYS_linkat].sy_call = (sy_call_t *) filemon_wrapper_linkat; #endif /* COMPAT_FREEBSD32 */ filemon_exec_tag = EVENTHANDLER_REGISTER(process_exec, filemon_event_process_exec, NULL, EVENTHANDLER_PRI_LAST); filemon_exit_tag = EVENTHANDLER_REGISTER(process_exit, filemon_event_process_exit, NULL, EVENTHANDLER_PRI_LAST); filemon_fork_tag = EVENTHANDLER_REGISTER(process_fork, filemon_event_process_fork, NULL, EVENTHANDLER_PRI_LAST); } static void filemon_wrapper_deinstall(void) { sysent[SYS_chdir].sy_call = (sy_call_t *)sys_chdir; sysent[SYS_open].sy_call = (sy_call_t *)sys_open; sysent[SYS_openat].sy_call = (sy_call_t *)sys_openat; sysent[SYS_rename].sy_call = (sy_call_t *)sys_rename; sysent[SYS_unlink].sy_call = (sy_call_t *)sys_unlink; sysent[SYS_link].sy_call = (sy_call_t *)sys_link; sysent[SYS_symlink].sy_call = (sy_call_t *)sys_symlink; sysent[SYS_linkat].sy_call = (sy_call_t *)sys_linkat; #if defined(COMPAT_FREEBSD32) freebsd32_sysent[FREEBSD32_SYS_chdir].sy_call = (sy_call_t *)sys_chdir; freebsd32_sysent[FREEBSD32_SYS_open].sy_call = (sy_call_t *)sys_open; freebsd32_sysent[FREEBSD32_SYS_openat].sy_call = (sy_call_t *)sys_openat; freebsd32_sysent[FREEBSD32_SYS_rename].sy_call = (sy_call_t *)sys_rename; freebsd32_sysent[FREEBSD32_SYS_unlink].sy_call = (sy_call_t *)sys_unlink; freebsd32_sysent[FREEBSD32_SYS_link].sy_call = (sy_call_t *)sys_link; freebsd32_sysent[FREEBSD32_SYS_symlink].sy_call = (sy_call_t *)sys_symlink; freebsd32_sysent[FREEBSD32_SYS_linkat].sy_call = (sy_call_t *)sys_linkat; #endif /* COMPAT_FREEBSD32 */ EVENTHANDLER_DEREGISTER(process_exec, filemon_exec_tag); EVENTHANDLER_DEREGISTER(process_exit, filemon_exit_tag); EVENTHANDLER_DEREGISTER(process_fork, filemon_fork_tag); } diff --git a/sys/dev/ipmi/ipmi_linux.c b/sys/dev/ipmi/ipmi_linux.c index 136712e8cfb5..4b33c5b014d8 100644 --- a/sys/dev/ipmi/ipmi_linux.c +++ b/sys/dev/ipmi/ipmi_linux.c @@ -1,118 +1,119 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2009 IronPort Systems Inc. * 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("$FreeBSD$"); /* * Linux ioctl handler for the ipmi device driver */ #include #include #include #include #include #include #include #include #ifdef __amd64__ #include #include #else #include #include #endif #include #include #include /* There are multiple ioctl number ranges that need to be handled */ #define IPMI_LINUX_IOCTL_MIN 0x690b #define IPMI_LINUX_IOCTL_MAX 0x6915 /* Linux versions of ioctl's */ #define L_IPMICTL_RECEIVE_MSG_TRUNC _IOWR(IPMI_IOC_MAGIC, 11, struct ipmi_recv) #define L_IPMICTL_RECEIVE_MSG _IOWR(IPMI_IOC_MAGIC, 12, struct ipmi_recv) #define L_IPMICTL_SEND_COMMAND _IOW(IPMI_IOC_MAGIC, 13, struct ipmi_req) #define L_IPMICTL_REGISTER_FOR_CMD _IOW(IPMI_IOC_MAGIC, 14, struct ipmi_cmdspec) #define L_IPMICTL_UNREGISTER_FOR_CMD _IOW(IPMI_IOC_MAGIC, 15, struct ipmi_cmdspec) #define L_IPMICTL_SET_GETS_EVENTS_CMD _IOW(IPMI_IOC_MAGIC, 16, int) #define L_IPMICTL_SET_MY_ADDRESS_CMD _IOW(IPMI_IOC_MAGIC, 17, unsigned int) #define L_IPMICTL_GET_MY_ADDRESS_CMD _IOW(IPMI_IOC_MAGIC, 18, unsigned int) #define L_IPMICTL_SET_MY_LUN_CMD _IOW(IPMI_IOC_MAGIC, 19, unsigned int) #define L_IPMICTL_GET_MY_LUN_CMD _IOW(IPMI_IOC_MAGIC, 20, unsigned int) static linux_ioctl_function_t ipmi_linux_ioctl; static struct linux_ioctl_handler ipmi_linux_handler = {ipmi_linux_ioctl, IPMI_LINUX_IOCTL_MIN, IPMI_LINUX_IOCTL_MAX}; SYSINIT (ipmi_linux_register, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_ioctl_register_handler, &ipmi_linux_handler); SYSUNINIT(ipmi_linux_unregister, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_ioctl_unregister_handler, &ipmi_linux_handler); static int ipmi_linux_modevent(module_t mod, int type, void *data) { /* Do we care about any specific load/unload actions? */ return (0); } DEV_MODULE(ipmi_linux, ipmi_linux_modevent, NULL); MODULE_DEPEND(ipmi_linux, linux, 1, 1, 1); static int ipmi_linux_ioctl(struct thread *td, struct linux_ioctl_args *args) { cap_rights_t rights; struct file *fp; u_long cmd; int error; - error = fget(td, args->fd, cap_rights_init(&rights, CAP_IOCTL), &fp); + error = fget(td, args->fd, cap_rights_init_one(&rights, CAP_IOCTL), + &fp); if (error != 0) return (error); cmd = args->cmd; switch(cmd) { case L_IPMICTL_GET_MY_ADDRESS_CMD: cmd = IPMICTL_GET_MY_ADDRESS_CMD; break; case L_IPMICTL_GET_MY_LUN_CMD: cmd = IPMICTL_GET_MY_LUN_CMD; break; } /* * Pass the ioctl off to our standard handler. */ error = (fo_ioctl(fp, cmd, (caddr_t)args->arg, td->td_ucred, td)); fdrop(fp, td); return (error); } diff --git a/sys/dev/iscsi/icl_soft.c b/sys/dev/iscsi/icl_soft.c index f8bcbb9eaa1c..bd4d9db81bb0 100644 --- a/sys/dev/iscsi/icl_soft.c +++ b/sys/dev/iscsi/icl_soft.c @@ -1,1635 +1,1635 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2012 The FreeBSD Foundation * * This software was developed by Edward Tomasz Napierala under sponsorship * from the FreeBSD Foundation. * * 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. * */ /* * Software implementation of iSCSI Common Layer kobj(9) interface. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include struct icl_soft_pdu { struct icl_pdu ip; /* soft specific stuff goes here. */ u_int ref_cnt; icl_pdu_cb cb; int error; }; static int coalesce = 1; SYSCTL_INT(_kern_icl, OID_AUTO, coalesce, CTLFLAG_RWTUN, &coalesce, 0, "Try to coalesce PDUs before sending"); static int partial_receive_len = 128 * 1024; SYSCTL_INT(_kern_icl, OID_AUTO, partial_receive_len, CTLFLAG_RWTUN, &partial_receive_len, 0, "Minimum read size for partially received " "data segment"); static int sendspace = 1048576; SYSCTL_INT(_kern_icl, OID_AUTO, sendspace, CTLFLAG_RWTUN, &sendspace, 0, "Default send socket buffer size"); static int recvspace = 1048576; SYSCTL_INT(_kern_icl, OID_AUTO, recvspace, CTLFLAG_RWTUN, &recvspace, 0, "Default receive socket buffer size"); static MALLOC_DEFINE(M_ICL_SOFT, "icl_soft", "iSCSI software backend"); static uma_zone_t icl_soft_pdu_zone; static volatile u_int icl_ncons; #define ICL_CONN_LOCK(X) mtx_lock(X->ic_lock) #define ICL_CONN_UNLOCK(X) mtx_unlock(X->ic_lock) #define ICL_CONN_LOCK_ASSERT(X) mtx_assert(X->ic_lock, MA_OWNED) #define ICL_CONN_LOCK_ASSERT_NOT(X) mtx_assert(X->ic_lock, MA_NOTOWNED) STAILQ_HEAD(icl_pdu_stailq, icl_pdu); static icl_conn_new_pdu_t icl_soft_conn_new_pdu; static icl_conn_pdu_free_t icl_soft_conn_pdu_free; static icl_conn_pdu_data_segment_length_t icl_soft_conn_pdu_data_segment_length; static icl_conn_pdu_append_data_t icl_soft_conn_pdu_append_data; static icl_conn_pdu_get_data_t icl_soft_conn_pdu_get_data; static icl_conn_pdu_queue_t icl_soft_conn_pdu_queue; static icl_conn_pdu_queue_cb_t icl_soft_conn_pdu_queue_cb; static icl_conn_handoff_t icl_soft_conn_handoff; static icl_conn_free_t icl_soft_conn_free; static icl_conn_close_t icl_soft_conn_close; static icl_conn_task_setup_t icl_soft_conn_task_setup; static icl_conn_task_done_t icl_soft_conn_task_done; static icl_conn_transfer_setup_t icl_soft_conn_transfer_setup; static icl_conn_transfer_done_t icl_soft_conn_transfer_done; #ifdef ICL_KERNEL_PROXY static icl_conn_connect_t icl_soft_conn_connect; #endif static kobj_method_t icl_soft_methods[] = { KOBJMETHOD(icl_conn_new_pdu, icl_soft_conn_new_pdu), KOBJMETHOD(icl_conn_pdu_free, icl_soft_conn_pdu_free), KOBJMETHOD(icl_conn_pdu_data_segment_length, icl_soft_conn_pdu_data_segment_length), KOBJMETHOD(icl_conn_pdu_append_data, icl_soft_conn_pdu_append_data), KOBJMETHOD(icl_conn_pdu_get_data, icl_soft_conn_pdu_get_data), KOBJMETHOD(icl_conn_pdu_queue, icl_soft_conn_pdu_queue), KOBJMETHOD(icl_conn_pdu_queue_cb, icl_soft_conn_pdu_queue_cb), KOBJMETHOD(icl_conn_handoff, icl_soft_conn_handoff), KOBJMETHOD(icl_conn_free, icl_soft_conn_free), KOBJMETHOD(icl_conn_close, icl_soft_conn_close), KOBJMETHOD(icl_conn_task_setup, icl_soft_conn_task_setup), KOBJMETHOD(icl_conn_task_done, icl_soft_conn_task_done), KOBJMETHOD(icl_conn_transfer_setup, icl_soft_conn_transfer_setup), KOBJMETHOD(icl_conn_transfer_done, icl_soft_conn_transfer_done), #ifdef ICL_KERNEL_PROXY KOBJMETHOD(icl_conn_connect, icl_soft_conn_connect), #endif { 0, 0 } }; DEFINE_CLASS(icl_soft, icl_soft_methods, sizeof(struct icl_conn)); static void icl_conn_fail(struct icl_conn *ic) { if (ic->ic_socket == NULL) return; /* * XXX */ ic->ic_socket->so_error = EDOOFUS; (ic->ic_error)(ic); } static struct mbuf * icl_conn_receive(struct icl_conn *ic, size_t len) { struct uio uio; struct socket *so; struct mbuf *m; int error, flags; so = ic->ic_socket; memset(&uio, 0, sizeof(uio)); uio.uio_resid = len; flags = MSG_DONTWAIT; error = soreceive(so, NULL, &uio, &m, NULL, &flags); if (error != 0) { ICL_DEBUG("soreceive error %d", error); return (NULL); } if (uio.uio_resid != 0) { m_freem(m); ICL_DEBUG("short read"); return (NULL); } return (m); } static int icl_conn_receive_buf(struct icl_conn *ic, void *buf, size_t len) { struct iovec iov[1]; struct uio uio; struct socket *so; int error, flags; so = ic->ic_socket; memset(&uio, 0, sizeof(uio)); iov[0].iov_base = buf; iov[0].iov_len = len; uio.uio_iov = iov; uio.uio_iovcnt = 1; uio.uio_offset = 0; uio.uio_resid = len; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; flags = MSG_DONTWAIT; error = soreceive(so, NULL, &uio, NULL, NULL, &flags); if (error != 0) { ICL_DEBUG("soreceive error %d", error); return (-1); } if (uio.uio_resid != 0) { ICL_DEBUG("short read"); return (-1); } return (0); } static void icl_soft_conn_pdu_free(struct icl_conn *ic, struct icl_pdu *ip) { struct icl_soft_pdu *isp = (struct icl_soft_pdu *)ip; KASSERT(isp->ref_cnt == 0, ("freeing active PDU")); m_freem(ip->ip_bhs_mbuf); m_freem(ip->ip_ahs_mbuf); m_freem(ip->ip_data_mbuf); uma_zfree(icl_soft_pdu_zone, isp); #ifdef DIAGNOSTIC refcount_release(&ic->ic_outstanding_pdus); #endif } static void icl_soft_pdu_call_cb(struct icl_pdu *ip) { struct icl_soft_pdu *isp = (struct icl_soft_pdu *)ip; if (isp->cb != NULL) isp->cb(ip, isp->error); #ifdef DIAGNOSTIC refcount_release(&ip->ip_conn->ic_outstanding_pdus); #endif uma_zfree(icl_soft_pdu_zone, isp); } static void icl_soft_pdu_done(struct icl_pdu *ip, int error) { struct icl_soft_pdu *isp = (struct icl_soft_pdu *)ip; if (error != 0) isp->error = error; m_freem(ip->ip_bhs_mbuf); ip->ip_bhs_mbuf = NULL; m_freem(ip->ip_ahs_mbuf); ip->ip_ahs_mbuf = NULL; m_freem(ip->ip_data_mbuf); ip->ip_data_mbuf = NULL; if (atomic_fetchadd_int(&isp->ref_cnt, -1) == 1) icl_soft_pdu_call_cb(ip); } static void icl_soft_mbuf_done(struct mbuf *mb) { struct icl_soft_pdu *isp = (struct icl_soft_pdu *)mb->m_ext.ext_arg1; icl_soft_pdu_call_cb(&isp->ip); } /* * Allocate icl_pdu with empty BHS to fill up by the caller. */ struct icl_pdu * icl_soft_conn_new_pdu(struct icl_conn *ic, int flags) { struct icl_soft_pdu *isp; struct icl_pdu *ip; #ifdef DIAGNOSTIC refcount_acquire(&ic->ic_outstanding_pdus); #endif isp = uma_zalloc(icl_soft_pdu_zone, flags | M_ZERO); if (isp == NULL) { ICL_WARN("failed to allocate soft PDU"); #ifdef DIAGNOSTIC refcount_release(&ic->ic_outstanding_pdus); #endif return (NULL); } ip = &isp->ip; ip->ip_conn = ic; CTASSERT(sizeof(struct iscsi_bhs) <= MHLEN); ip->ip_bhs_mbuf = m_gethdr(flags, MT_DATA); if (ip->ip_bhs_mbuf == NULL) { ICL_WARN("failed to allocate BHS mbuf"); icl_soft_conn_pdu_free(ic, ip); return (NULL); } ip->ip_bhs = mtod(ip->ip_bhs_mbuf, struct iscsi_bhs *); memset(ip->ip_bhs, 0, sizeof(struct iscsi_bhs)); ip->ip_bhs_mbuf->m_len = sizeof(struct iscsi_bhs); return (ip); } static int icl_pdu_ahs_length(const struct icl_pdu *request) { return (request->ip_bhs->bhs_total_ahs_len * 4); } static size_t icl_pdu_data_segment_length(const struct icl_pdu *request) { uint32_t len = 0; len += request->ip_bhs->bhs_data_segment_len[0]; len <<= 8; len += request->ip_bhs->bhs_data_segment_len[1]; len <<= 8; len += request->ip_bhs->bhs_data_segment_len[2]; return (len); } size_t icl_soft_conn_pdu_data_segment_length(struct icl_conn *ic, const struct icl_pdu *request) { return (icl_pdu_data_segment_length(request)); } static void icl_pdu_set_data_segment_length(struct icl_pdu *response, uint32_t len) { response->ip_bhs->bhs_data_segment_len[2] = len; response->ip_bhs->bhs_data_segment_len[1] = len >> 8; response->ip_bhs->bhs_data_segment_len[0] = len >> 16; } static size_t icl_pdu_padding(const struct icl_pdu *ip) { if ((ip->ip_data_len % 4) != 0) return (4 - (ip->ip_data_len % 4)); return (0); } static size_t icl_pdu_size(const struct icl_pdu *response) { size_t len; KASSERT(response->ip_ahs_len == 0, ("responding with AHS")); len = sizeof(struct iscsi_bhs) + response->ip_data_len + icl_pdu_padding(response); if (response->ip_conn->ic_header_crc32c) len += ISCSI_HEADER_DIGEST_SIZE; if (response->ip_data_len != 0 && response->ip_conn->ic_data_crc32c) len += ISCSI_DATA_DIGEST_SIZE; return (len); } static int icl_pdu_receive_bhs(struct icl_pdu *request, size_t *availablep) { if (icl_conn_receive_buf(request->ip_conn, request->ip_bhs, sizeof(struct iscsi_bhs))) { ICL_DEBUG("failed to receive BHS"); return (-1); } *availablep -= sizeof(struct iscsi_bhs); return (0); } static int icl_pdu_receive_ahs(struct icl_pdu *request, size_t *availablep) { request->ip_ahs_len = icl_pdu_ahs_length(request); if (request->ip_ahs_len == 0) return (0); request->ip_ahs_mbuf = icl_conn_receive(request->ip_conn, request->ip_ahs_len); if (request->ip_ahs_mbuf == NULL) { ICL_DEBUG("failed to receive AHS"); return (-1); } *availablep -= request->ip_ahs_len; return (0); } static uint32_t icl_mbuf_to_crc32c(const struct mbuf *m0) { uint32_t digest = 0xffffffff; const struct mbuf *m; for (m = m0; m != NULL; m = m->m_next) digest = calculate_crc32c(digest, mtod(m, const void *), m->m_len); digest = digest ^ 0xffffffff; return (digest); } static int icl_pdu_check_header_digest(struct icl_pdu *request, size_t *availablep) { uint32_t received_digest, valid_digest; if (request->ip_conn->ic_header_crc32c == false) return (0); CTASSERT(sizeof(received_digest) == ISCSI_HEADER_DIGEST_SIZE); if (icl_conn_receive_buf(request->ip_conn, &received_digest, ISCSI_HEADER_DIGEST_SIZE)) { ICL_DEBUG("failed to receive header digest"); return (-1); } *availablep -= ISCSI_HEADER_DIGEST_SIZE; /* Temporary attach AHS to BHS to calculate header digest. */ request->ip_bhs_mbuf->m_next = request->ip_ahs_mbuf; valid_digest = icl_mbuf_to_crc32c(request->ip_bhs_mbuf); request->ip_bhs_mbuf->m_next = NULL; if (received_digest != valid_digest) { ICL_WARN("header digest check failed; got 0x%x, " "should be 0x%x", received_digest, valid_digest); return (-1); } return (0); } /* * Return the number of bytes that should be waiting in the receive socket * before icl_pdu_receive_data_segment() gets called. */ static size_t icl_pdu_data_segment_receive_len(const struct icl_pdu *request) { size_t len; len = icl_pdu_data_segment_length(request); if (len == 0) return (0); /* * Account for the parts of data segment already read from * the socket buffer. */ KASSERT(len > request->ip_data_len, ("len <= request->ip_data_len")); len -= request->ip_data_len; /* * Don't always wait for the full data segment to be delivered * to the socket; this might badly affect performance due to * TCP window scaling. */ if (len > partial_receive_len) { #if 0 ICL_DEBUG("need %zd bytes of data, limiting to %zd", len, partial_receive_len)); #endif len = partial_receive_len; return (len); } /* * Account for padding. Note that due to the way code is written, * the icl_pdu_receive_data_segment() must always receive padding * along with the last part of data segment, because it would be * impossible to tell whether we've already received the full data * segment including padding, or without it. */ if ((len % 4) != 0) len += 4 - (len % 4); #if 0 ICL_DEBUG("need %zd bytes of data", len)); #endif return (len); } static int icl_pdu_receive_data_segment(struct icl_pdu *request, size_t *availablep, bool *more_neededp) { struct icl_conn *ic; size_t len, padding = 0; struct mbuf *m; ic = request->ip_conn; *more_neededp = false; ic->ic_receive_len = 0; len = icl_pdu_data_segment_length(request); if (len == 0) return (0); if ((len % 4) != 0) padding = 4 - (len % 4); /* * Account for already received parts of data segment. */ KASSERT(len > request->ip_data_len, ("len <= request->ip_data_len")); len -= request->ip_data_len; if (len + padding > *availablep) { /* * Not enough data in the socket buffer. Receive as much * as we can. Don't receive padding, since, obviously, it's * not the end of data segment yet. */ #if 0 ICL_DEBUG("limited from %zd to %zd", len + padding, *availablep - padding)); #endif len = *availablep - padding; *more_neededp = true; padding = 0; } /* * Must not try to receive padding without at least one byte * of actual data segment. */ if (len > 0) { m = icl_conn_receive(request->ip_conn, len + padding); if (m == NULL) { ICL_DEBUG("failed to receive data segment"); return (-1); } if (request->ip_data_mbuf == NULL) request->ip_data_mbuf = m; else m_cat(request->ip_data_mbuf, m); request->ip_data_len += len; *availablep -= len + padding; } else ICL_DEBUG("len 0"); if (*more_neededp) ic->ic_receive_len = icl_pdu_data_segment_receive_len(request); return (0); } static int icl_pdu_check_data_digest(struct icl_pdu *request, size_t *availablep) { uint32_t received_digest, valid_digest; if (request->ip_conn->ic_data_crc32c == false) return (0); if (request->ip_data_len == 0) return (0); CTASSERT(sizeof(received_digest) == ISCSI_DATA_DIGEST_SIZE); if (icl_conn_receive_buf(request->ip_conn, &received_digest, ISCSI_DATA_DIGEST_SIZE)) { ICL_DEBUG("failed to receive data digest"); return (-1); } *availablep -= ISCSI_DATA_DIGEST_SIZE; /* * Note that ip_data_mbuf also contains padding; since digest * calculation is supposed to include that, we iterate over * the entire ip_data_mbuf chain, not just ip_data_len bytes of it. */ valid_digest = icl_mbuf_to_crc32c(request->ip_data_mbuf); if (received_digest != valid_digest) { ICL_WARN("data digest check failed; got 0x%x, " "should be 0x%x", received_digest, valid_digest); return (-1); } return (0); } /* * Somewhat contrary to the name, this attempts to receive only one * "part" of PDU at a time; call it repeatedly until it returns non-NULL. */ static struct icl_pdu * icl_conn_receive_pdu(struct icl_conn *ic, size_t *availablep) { struct icl_pdu *request; struct socket *so; size_t len; int error; bool more_needed; so = ic->ic_socket; if (ic->ic_receive_state == ICL_CONN_STATE_BHS) { KASSERT(ic->ic_receive_pdu == NULL, ("ic->ic_receive_pdu != NULL")); request = icl_soft_conn_new_pdu(ic, M_NOWAIT); if (request == NULL) { ICL_DEBUG("failed to allocate PDU; " "dropping connection"); icl_conn_fail(ic); return (NULL); } ic->ic_receive_pdu = request; } else { KASSERT(ic->ic_receive_pdu != NULL, ("ic->ic_receive_pdu == NULL")); request = ic->ic_receive_pdu; } if (*availablep < ic->ic_receive_len) { #if 0 ICL_DEBUG("not enough data; need %zd, " "have %zd", ic->ic_receive_len, *availablep); #endif return (NULL); } switch (ic->ic_receive_state) { case ICL_CONN_STATE_BHS: //ICL_DEBUG("receiving BHS"); error = icl_pdu_receive_bhs(request, availablep); if (error != 0) { ICL_DEBUG("failed to receive BHS; " "dropping connection"); break; } /* * We don't enforce any limit for AHS length; * its length is stored in 8 bit field. */ len = icl_pdu_data_segment_length(request); if (len > ic->ic_max_data_segment_length) { ICL_WARN("received data segment " "length %zd is larger than negotiated " "MaxDataSegmentLength %zd; " "dropping connection", len, ic->ic_max_data_segment_length); error = EINVAL; break; } ic->ic_receive_state = ICL_CONN_STATE_AHS; ic->ic_receive_len = icl_pdu_ahs_length(request); break; case ICL_CONN_STATE_AHS: //ICL_DEBUG("receiving AHS"); error = icl_pdu_receive_ahs(request, availablep); if (error != 0) { ICL_DEBUG("failed to receive AHS; " "dropping connection"); break; } ic->ic_receive_state = ICL_CONN_STATE_HEADER_DIGEST; if (ic->ic_header_crc32c == false) ic->ic_receive_len = 0; else ic->ic_receive_len = ISCSI_HEADER_DIGEST_SIZE; break; case ICL_CONN_STATE_HEADER_DIGEST: //ICL_DEBUG("receiving header digest"); error = icl_pdu_check_header_digest(request, availablep); if (error != 0) { ICL_DEBUG("header digest failed; " "dropping connection"); break; } ic->ic_receive_state = ICL_CONN_STATE_DATA; ic->ic_receive_len = icl_pdu_data_segment_receive_len(request); break; case ICL_CONN_STATE_DATA: //ICL_DEBUG("receiving data segment"); error = icl_pdu_receive_data_segment(request, availablep, &more_needed); if (error != 0) { ICL_DEBUG("failed to receive data segment;" "dropping connection"); break; } if (more_needed) break; ic->ic_receive_state = ICL_CONN_STATE_DATA_DIGEST; if (request->ip_data_len == 0 || ic->ic_data_crc32c == false) ic->ic_receive_len = 0; else ic->ic_receive_len = ISCSI_DATA_DIGEST_SIZE; break; case ICL_CONN_STATE_DATA_DIGEST: //ICL_DEBUG("receiving data digest"); error = icl_pdu_check_data_digest(request, availablep); if (error != 0) { ICL_DEBUG("data digest failed; " "dropping connection"); break; } /* * We've received complete PDU; reset the receive state machine * and return the PDU. */ ic->ic_receive_state = ICL_CONN_STATE_BHS; ic->ic_receive_len = sizeof(struct iscsi_bhs); ic->ic_receive_pdu = NULL; return (request); default: panic("invalid ic_receive_state %d\n", ic->ic_receive_state); } if (error != 0) { /* * Don't free the PDU; it's pointed to by ic->ic_receive_pdu * and will get freed in icl_soft_conn_close(). */ icl_conn_fail(ic); } return (NULL); } static void icl_conn_receive_pdus(struct icl_conn *ic, size_t available) { struct icl_pdu *response; struct socket *so; so = ic->ic_socket; /* * This can never happen; we're careful to only mess with ic->ic_socket * pointer when the send/receive threads are not running. */ KASSERT(so != NULL, ("NULL socket")); for (;;) { if (ic->ic_disconnecting) return; if (so->so_error != 0) { ICL_DEBUG("connection error %d; " "dropping connection", so->so_error); icl_conn_fail(ic); return; } /* * Loop until we have a complete PDU or there is not enough * data in the socket buffer. */ if (available < ic->ic_receive_len) { #if 0 ICL_DEBUG("not enough data; have %zd, " "need %zd", available, ic->ic_receive_len); #endif return; } response = icl_conn_receive_pdu(ic, &available); if (response == NULL) continue; if (response->ip_ahs_len > 0) { ICL_WARN("received PDU with unsupported " "AHS; opcode 0x%x; dropping connection", response->ip_bhs->bhs_opcode); icl_soft_conn_pdu_free(ic, response); icl_conn_fail(ic); return; } (ic->ic_receive)(response); } } static void icl_receive_thread(void *arg) { struct icl_conn *ic; size_t available; struct socket *so; ic = arg; so = ic->ic_socket; for (;;) { if (ic->ic_disconnecting) { //ICL_DEBUG("terminating"); break; } /* * Set the low watermark, to be checked by * soreadable() in icl_soupcall_receive() * to avoid unnecessary wakeups until there * is enough data received to read the PDU. */ SOCKBUF_LOCK(&so->so_rcv); available = sbavail(&so->so_rcv); if (available < ic->ic_receive_len) { so->so_rcv.sb_lowat = ic->ic_receive_len; cv_wait(&ic->ic_receive_cv, &so->so_rcv.sb_mtx); } else so->so_rcv.sb_lowat = so->so_rcv.sb_hiwat + 1; SOCKBUF_UNLOCK(&so->so_rcv); icl_conn_receive_pdus(ic, available); } ICL_CONN_LOCK(ic); ic->ic_receive_running = false; cv_signal(&ic->ic_send_cv); ICL_CONN_UNLOCK(ic); kthread_exit(); } static int icl_soupcall_receive(struct socket *so, void *arg, int waitflag) { struct icl_conn *ic; if (!soreadable(so)) return (SU_OK); ic = arg; cv_signal(&ic->ic_receive_cv); return (SU_OK); } static int icl_pdu_finalize(struct icl_pdu *request) { size_t padding, pdu_len; uint32_t digest, zero = 0; int ok; struct icl_conn *ic; ic = request->ip_conn; icl_pdu_set_data_segment_length(request, request->ip_data_len); pdu_len = icl_pdu_size(request); if (ic->ic_header_crc32c) { digest = icl_mbuf_to_crc32c(request->ip_bhs_mbuf); ok = m_append(request->ip_bhs_mbuf, sizeof(digest), (void *)&digest); if (ok != 1) { ICL_WARN("failed to append header digest"); return (1); } } if (request->ip_data_len != 0) { padding = icl_pdu_padding(request); if (padding > 0) { ok = m_append(request->ip_data_mbuf, padding, (void *)&zero); if (ok != 1) { ICL_WARN("failed to append padding"); return (1); } } if (ic->ic_data_crc32c) { digest = icl_mbuf_to_crc32c(request->ip_data_mbuf); ok = m_append(request->ip_data_mbuf, sizeof(digest), (void *)&digest); if (ok != 1) { ICL_WARN("failed to append data digest"); return (1); } } m_cat(request->ip_bhs_mbuf, request->ip_data_mbuf); request->ip_data_mbuf = NULL; } request->ip_bhs_mbuf->m_pkthdr.len = pdu_len; return (0); } static void icl_conn_send_pdus(struct icl_conn *ic, struct icl_pdu_stailq *queue) { struct icl_pdu *request, *request2; struct socket *so; long available, size, size2; int coalesced, error; ICL_CONN_LOCK_ASSERT_NOT(ic); so = ic->ic_socket; SOCKBUF_LOCK(&so->so_snd); /* * Check how much space do we have for transmit. We can't just * call sosend() and retry when we get EWOULDBLOCK or EMSGSIZE, * as it always frees the mbuf chain passed to it, even in case * of error. */ available = sbspace(&so->so_snd); /* * Notify the socket upcall that we don't need wakeups * for the time being. */ so->so_snd.sb_lowat = so->so_snd.sb_hiwat + 1; SOCKBUF_UNLOCK(&so->so_snd); while (!STAILQ_EMPTY(queue)) { request = STAILQ_FIRST(queue); size = icl_pdu_size(request); if (available < size) { /* * Set the low watermark, to be checked by * sowriteable() in icl_soupcall_send() * to avoid unnecessary wakeups until there * is enough space for the PDU to fit. */ SOCKBUF_LOCK(&so->so_snd); available = sbspace(&so->so_snd); if (available < size) { #if 1 ICL_DEBUG("no space to send; " "have %ld, need %ld", available, size); #endif so->so_snd.sb_lowat = max(size, so->so_snd.sb_hiwat / 8); SOCKBUF_UNLOCK(&so->so_snd); return; } SOCKBUF_UNLOCK(&so->so_snd); } STAILQ_REMOVE_HEAD(queue, ip_next); error = icl_pdu_finalize(request); if (error != 0) { ICL_DEBUG("failed to finalize PDU; " "dropping connection"); icl_soft_pdu_done(request, EIO); icl_conn_fail(ic); return; } if (coalesce) { coalesced = 1; for (;;) { request2 = STAILQ_FIRST(queue); if (request2 == NULL) break; size2 = icl_pdu_size(request2); if (available < size + size2) break; STAILQ_REMOVE_HEAD(queue, ip_next); error = icl_pdu_finalize(request2); if (error != 0) { ICL_DEBUG("failed to finalize PDU; " "dropping connection"); icl_soft_pdu_done(request, EIO); icl_soft_pdu_done(request2, EIO); icl_conn_fail(ic); return; } m_cat(request->ip_bhs_mbuf, request2->ip_bhs_mbuf); request2->ip_bhs_mbuf = NULL; request->ip_bhs_mbuf->m_pkthdr.len += size2; size += size2; STAILQ_REMOVE_AFTER(queue, request, ip_next); icl_soft_pdu_done(request2, 0); coalesced++; } #if 0 if (coalesced > 1) { ICL_DEBUG("coalesced %d PDUs into %ld bytes", coalesced, size); } #endif } available -= size; error = sosend(so, NULL, NULL, request->ip_bhs_mbuf, NULL, MSG_DONTWAIT, curthread); request->ip_bhs_mbuf = NULL; /* Sosend consumes the mbuf. */ if (error != 0) { ICL_DEBUG("failed to send PDU, error %d; " "dropping connection", error); icl_soft_pdu_done(request, error); icl_conn_fail(ic); return; } icl_soft_pdu_done(request, 0); } } static void icl_send_thread(void *arg) { struct icl_conn *ic; struct icl_pdu_stailq queue; ic = arg; STAILQ_INIT(&queue); ICL_CONN_LOCK(ic); for (;;) { for (;;) { /* * If the local queue is empty, populate it from * the main one. This way the icl_conn_send_pdus() * can go through all the queued PDUs without holding * any locks. */ if (STAILQ_EMPTY(&queue)) STAILQ_SWAP(&ic->ic_to_send, &queue, icl_pdu); ic->ic_check_send_space = false; ICL_CONN_UNLOCK(ic); icl_conn_send_pdus(ic, &queue); ICL_CONN_LOCK(ic); /* * The icl_soupcall_send() was called since the last * call to sbspace(); go around; */ if (ic->ic_check_send_space) continue; /* * Local queue is empty, but we still have PDUs * in the main one; go around. */ if (STAILQ_EMPTY(&queue) && !STAILQ_EMPTY(&ic->ic_to_send)) continue; /* * There might be some stuff in the local queue, * which didn't get sent due to not having enough send * space. Wait for socket upcall. */ break; } if (ic->ic_disconnecting) { //ICL_DEBUG("terminating"); break; } cv_wait(&ic->ic_send_cv, ic->ic_lock); } /* * We're exiting; move PDUs back to the main queue, so they can * get freed properly. At this point ordering doesn't matter. */ STAILQ_CONCAT(&ic->ic_to_send, &queue); ic->ic_send_running = false; cv_signal(&ic->ic_send_cv); ICL_CONN_UNLOCK(ic); kthread_exit(); } static int icl_soupcall_send(struct socket *so, void *arg, int waitflag) { struct icl_conn *ic; if (!sowriteable(so)) return (SU_OK); ic = arg; ICL_CONN_LOCK(ic); ic->ic_check_send_space = true; ICL_CONN_UNLOCK(ic); cv_signal(&ic->ic_send_cv); return (SU_OK); } static int icl_soft_conn_pdu_append_data(struct icl_conn *ic, struct icl_pdu *request, const void *addr, size_t len, int flags) { struct icl_soft_pdu *isp = (struct icl_soft_pdu *)request; struct mbuf *mb, *newmb; size_t copylen, off = 0; KASSERT(len > 0, ("len == 0")); if (flags & ICL_NOCOPY) { newmb = m_get(flags & ~ICL_NOCOPY, MT_DATA); if (newmb == NULL) { ICL_WARN("failed to allocate mbuf"); return (ENOMEM); } newmb->m_flags |= M_RDONLY; m_extaddref(newmb, __DECONST(char *, addr), len, &isp->ref_cnt, icl_soft_mbuf_done, isp, NULL); newmb->m_len = len; } else { newmb = m_getm2(NULL, len, flags, MT_DATA, 0); if (newmb == NULL) { ICL_WARN("failed to allocate mbuf for %zd bytes", len); return (ENOMEM); } for (mb = newmb; mb != NULL; mb = mb->m_next) { copylen = min(M_TRAILINGSPACE(mb), len - off); memcpy(mtod(mb, char *), (const char *)addr + off, copylen); mb->m_len = copylen; off += copylen; } KASSERT(off == len, ("%s: off != len", __func__)); } if (request->ip_data_mbuf == NULL) { request->ip_data_mbuf = newmb; request->ip_data_len = len; } else { m_cat(request->ip_data_mbuf, newmb); request->ip_data_len += len; } return (0); } void icl_soft_conn_pdu_get_data(struct icl_conn *ic, struct icl_pdu *ip, size_t off, void *addr, size_t len) { m_copydata(ip->ip_data_mbuf, off, len, addr); } static void icl_soft_conn_pdu_queue(struct icl_conn *ic, struct icl_pdu *ip) { icl_soft_conn_pdu_queue_cb(ic, ip, NULL); } static void icl_soft_conn_pdu_queue_cb(struct icl_conn *ic, struct icl_pdu *ip, icl_pdu_cb cb) { struct icl_soft_pdu *isp = (struct icl_soft_pdu *)ip; ICL_CONN_LOCK_ASSERT(ic); isp->ref_cnt++; isp->cb = cb; if (ic->ic_disconnecting || ic->ic_socket == NULL) { ICL_DEBUG("icl_pdu_queue on closed connection"); icl_soft_pdu_done(ip, ENOTCONN); return; } if (!STAILQ_EMPTY(&ic->ic_to_send)) { STAILQ_INSERT_TAIL(&ic->ic_to_send, ip, ip_next); /* * If the queue is not empty, someone else had already * signaled the send thread; no need to do that again, * just return. */ return; } STAILQ_INSERT_TAIL(&ic->ic_to_send, ip, ip_next); cv_signal(&ic->ic_send_cv); } static struct icl_conn * icl_soft_new_conn(const char *name, struct mtx *lock) { struct icl_conn *ic; refcount_acquire(&icl_ncons); ic = (struct icl_conn *)kobj_create(&icl_soft_class, M_ICL_SOFT, M_WAITOK | M_ZERO); STAILQ_INIT(&ic->ic_to_send); ic->ic_lock = lock; cv_init(&ic->ic_send_cv, "icl_tx"); cv_init(&ic->ic_receive_cv, "icl_rx"); #ifdef DIAGNOSTIC refcount_init(&ic->ic_outstanding_pdus, 0); #endif ic->ic_max_data_segment_length = ICL_MAX_DATA_SEGMENT_LENGTH; ic->ic_name = name; ic->ic_offload = "None"; ic->ic_unmapped = false; return (ic); } void icl_soft_conn_free(struct icl_conn *ic) { #ifdef DIAGNOSTIC KASSERT(ic->ic_outstanding_pdus == 0, ("destroying session with %d outstanding PDUs", ic->ic_outstanding_pdus)); #endif cv_destroy(&ic->ic_send_cv); cv_destroy(&ic->ic_receive_cv); kobj_delete((struct kobj *)ic, M_ICL_SOFT); refcount_release(&icl_ncons); } static int icl_conn_start(struct icl_conn *ic) { size_t minspace; struct sockopt opt; int error, one = 1; ICL_CONN_LOCK(ic); /* * XXX: Ugly hack. */ if (ic->ic_socket == NULL) { ICL_CONN_UNLOCK(ic); return (EINVAL); } ic->ic_receive_state = ICL_CONN_STATE_BHS; ic->ic_receive_len = sizeof(struct iscsi_bhs); ic->ic_disconnecting = false; ICL_CONN_UNLOCK(ic); /* * For sendspace, this is required because the current code cannot * send a PDU in pieces; thus, the minimum buffer size is equal * to the maximum PDU size. "+4" is to account for possible padding. * * What we should actually do here is to use autoscaling, but set * some minimal buffer size to "minspace". I don't know a way to do * that, though. */ minspace = sizeof(struct iscsi_bhs) + ic->ic_max_data_segment_length + ISCSI_HEADER_DIGEST_SIZE + ISCSI_DATA_DIGEST_SIZE + 4; if (sendspace < minspace) { ICL_WARN("kern.icl.sendspace too low; must be at least %zd", minspace); sendspace = minspace; } if (recvspace < minspace) { ICL_WARN("kern.icl.recvspace too low; must be at least %zd", minspace); recvspace = minspace; } error = soreserve(ic->ic_socket, sendspace, recvspace); if (error != 0) { ICL_WARN("soreserve failed with error %d", error); icl_soft_conn_close(ic); return (error); } ic->ic_socket->so_snd.sb_flags |= SB_AUTOSIZE; ic->ic_socket->so_rcv.sb_flags |= SB_AUTOSIZE; /* * Disable Nagle. */ bzero(&opt, sizeof(opt)); opt.sopt_dir = SOPT_SET; opt.sopt_level = IPPROTO_TCP; opt.sopt_name = TCP_NODELAY; opt.sopt_val = &one; opt.sopt_valsize = sizeof(one); error = sosetopt(ic->ic_socket, &opt); if (error != 0) { ICL_WARN("disabling TCP_NODELAY failed with error %d", error); icl_soft_conn_close(ic); return (error); } /* * Register socket upcall, to get notified about incoming PDUs * and free space to send outgoing ones. */ SOCKBUF_LOCK(&ic->ic_socket->so_snd); soupcall_set(ic->ic_socket, SO_SND, icl_soupcall_send, ic); SOCKBUF_UNLOCK(&ic->ic_socket->so_snd); SOCKBUF_LOCK(&ic->ic_socket->so_rcv); soupcall_set(ic->ic_socket, SO_RCV, icl_soupcall_receive, ic); SOCKBUF_UNLOCK(&ic->ic_socket->so_rcv); /* * Start threads. */ ICL_CONN_LOCK(ic); ic->ic_send_running = ic->ic_receive_running = true; ICL_CONN_UNLOCK(ic); error = kthread_add(icl_send_thread, ic, NULL, NULL, 0, 0, "%stx", ic->ic_name); if (error != 0) { ICL_WARN("kthread_add(9) failed with error %d", error); ICL_CONN_LOCK(ic); ic->ic_send_running = ic->ic_receive_running = false; cv_signal(&ic->ic_send_cv); ICL_CONN_UNLOCK(ic); icl_soft_conn_close(ic); return (error); } error = kthread_add(icl_receive_thread, ic, NULL, NULL, 0, 0, "%srx", ic->ic_name); if (error != 0) { ICL_WARN("kthread_add(9) failed with error %d", error); ICL_CONN_LOCK(ic); ic->ic_receive_running = false; cv_signal(&ic->ic_send_cv); ICL_CONN_UNLOCK(ic); icl_soft_conn_close(ic); return (error); } return (0); } int icl_soft_conn_handoff(struct icl_conn *ic, int fd) { struct file *fp; struct socket *so; cap_rights_t rights; int error; ICL_CONN_LOCK_ASSERT_NOT(ic); #ifdef ICL_KERNEL_PROXY /* * We're transitioning to Full Feature phase, and we don't * really care. */ if (fd == 0) { ICL_CONN_LOCK(ic); if (ic->ic_socket == NULL) { ICL_CONN_UNLOCK(ic); ICL_WARN("proxy handoff without connect"); return (EINVAL); } ICL_CONN_UNLOCK(ic); return (0); } #endif /* * Steal the socket from userland. */ error = fget(curthread, fd, - cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp); + cap_rights_init_one(&rights, CAP_SOCK_CLIENT), &fp); if (error != 0) return (error); if (fp->f_type != DTYPE_SOCKET) { fdrop(fp, curthread); return (EINVAL); } so = fp->f_data; if (so->so_type != SOCK_STREAM) { fdrop(fp, curthread); return (EINVAL); } ICL_CONN_LOCK(ic); if (ic->ic_socket != NULL) { ICL_CONN_UNLOCK(ic); fdrop(fp, curthread); return (EBUSY); } ic->ic_socket = fp->f_data; fp->f_ops = &badfileops; fp->f_data = NULL; fdrop(fp, curthread); ICL_CONN_UNLOCK(ic); error = icl_conn_start(ic); return (error); } void icl_soft_conn_close(struct icl_conn *ic) { struct icl_pdu *pdu; struct socket *so; ICL_CONN_LOCK(ic); /* * Wake up the threads, so they can properly terminate. */ ic->ic_disconnecting = true; while (ic->ic_receive_running || ic->ic_send_running) { cv_signal(&ic->ic_receive_cv); cv_signal(&ic->ic_send_cv); cv_wait(&ic->ic_send_cv, ic->ic_lock); } /* Some other thread could close the connection same time. */ so = ic->ic_socket; if (so == NULL) { ICL_CONN_UNLOCK(ic); return; } ic->ic_socket = NULL; /* * Deregister socket upcalls. */ ICL_CONN_UNLOCK(ic); SOCKBUF_LOCK(&so->so_snd); if (so->so_snd.sb_upcall != NULL) soupcall_clear(so, SO_SND); SOCKBUF_UNLOCK(&so->so_snd); SOCKBUF_LOCK(&so->so_rcv); if (so->so_rcv.sb_upcall != NULL) soupcall_clear(so, SO_RCV); SOCKBUF_UNLOCK(&so->so_rcv); soclose(so); ICL_CONN_LOCK(ic); if (ic->ic_receive_pdu != NULL) { //ICL_DEBUG("freeing partially received PDU"); icl_soft_conn_pdu_free(ic, ic->ic_receive_pdu); ic->ic_receive_pdu = NULL; } /* * Remove any outstanding PDUs from the send queue. */ while (!STAILQ_EMPTY(&ic->ic_to_send)) { pdu = STAILQ_FIRST(&ic->ic_to_send); STAILQ_REMOVE_HEAD(&ic->ic_to_send, ip_next); icl_soft_pdu_done(pdu, ENOTCONN); } KASSERT(STAILQ_EMPTY(&ic->ic_to_send), ("destroying session with non-empty send queue")); ICL_CONN_UNLOCK(ic); } int icl_soft_conn_task_setup(struct icl_conn *ic, struct icl_pdu *ip, struct ccb_scsiio *csio, uint32_t *task_tagp, void **prvp) { return (0); } void icl_soft_conn_task_done(struct icl_conn *ic, void *prv) { } int icl_soft_conn_transfer_setup(struct icl_conn *ic, union ctl_io *io, uint32_t *transfer_tag, void **prvp) { return (0); } void icl_soft_conn_transfer_done(struct icl_conn *ic, void *prv) { } static int icl_soft_limits(struct icl_drv_limits *idl) { idl->idl_max_recv_data_segment_length = 128 * 1024; idl->idl_max_send_data_segment_length = 128 * 1024; idl->idl_max_burst_length = 262144; idl->idl_first_burst_length = 65536; return (0); } #ifdef ICL_KERNEL_PROXY int icl_soft_conn_connect(struct icl_conn *ic, int domain, int socktype, int protocol, struct sockaddr *from_sa, struct sockaddr *to_sa) { return (icl_soft_proxy_connect(ic, domain, socktype, protocol, from_sa, to_sa)); } int icl_soft_handoff_sock(struct icl_conn *ic, struct socket *so) { int error; ICL_CONN_LOCK_ASSERT_NOT(ic); if (so->so_type != SOCK_STREAM) return (EINVAL); ICL_CONN_LOCK(ic); if (ic->ic_socket != NULL) { ICL_CONN_UNLOCK(ic); return (EBUSY); } ic->ic_socket = so; ICL_CONN_UNLOCK(ic); error = icl_conn_start(ic); return (error); } #endif /* ICL_KERNEL_PROXY */ static int icl_soft_load(void) { int error; icl_soft_pdu_zone = uma_zcreate("icl_soft_pdu", sizeof(struct icl_soft_pdu), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); refcount_init(&icl_ncons, 0); /* * The reason we call this "none" is that to the user, * it's known as "offload driver"; "offload driver: soft" * doesn't make much sense. */ error = icl_register("none", false, 0, icl_soft_limits, icl_soft_new_conn); KASSERT(error == 0, ("failed to register")); #if defined(ICL_KERNEL_PROXY) && 0 /* * Debugging aid for kernel proxy functionality. */ error = icl_register("proxytest", true, 0, icl_soft_limits, icl_soft_new_conn); KASSERT(error == 0, ("failed to register")); #endif return (error); } static int icl_soft_unload(void) { if (icl_ncons != 0) return (EBUSY); icl_unregister("none", false); #if defined(ICL_KERNEL_PROXY) && 0 icl_unregister("proxytest", true); #endif uma_zdestroy(icl_soft_pdu_zone); return (0); } static int icl_soft_modevent(module_t mod, int what, void *arg) { switch (what) { case MOD_LOAD: return (icl_soft_load()); case MOD_UNLOAD: return (icl_soft_unload()); default: return (EINVAL); } } moduledata_t icl_soft_data = { "icl_soft", icl_soft_modevent, 0 }; DECLARE_MODULE(icl_soft, icl_soft_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE); MODULE_DEPEND(icl_soft, icl, 1, 1, 1); MODULE_VERSION(icl_soft, 1); diff --git a/sys/dev/iscsi_initiator/iscsi.c b/sys/dev/iscsi_initiator/iscsi.c index 19c45f87d132..ce190c5c892a 100644 --- a/sys/dev/iscsi_initiator/iscsi.c +++ b/sys/dev/iscsi_initiator/iscsi.c @@ -1,880 +1,880 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2005-2011 Daniel Braniss * 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. * */ /* | $Id: iscsi.c 752 2009-08-20 11:23:28Z danny $ */ #include __FBSDID("$FreeBSD$"); #include "opt_iscsi_initiator.h" #include #include #ifdef DO_EVENTHANDLER #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static char *iscsi_driver_version = "2.3.1"; static struct isc_softc *isc; MALLOC_DEFINE(M_ISCSI, "iSCSI", "iSCSI driver"); MALLOC_DEFINE(M_ISCSIBUF, "iSCbuf", "iSCSI buffers"); static MALLOC_DEFINE(M_TMP, "iSCtmp", "iSCSI tmp"); #ifdef ISCSI_INITIATOR_DEBUG int iscsi_debug = ISCSI_INITIATOR_DEBUG; SYSCTL_INT(_debug, OID_AUTO, iscsi_initiator, CTLFLAG_RW, &iscsi_debug, 0, "iSCSI driver debug flag"); struct mtx iscsi_dbg_mtx; #endif static int max_sessions = MAX_SESSIONS; SYSCTL_INT(_net, OID_AUTO, iscsi_initiator_max_sessions, CTLFLAG_RDTUN, &max_sessions, 0, "Max sessions allowed"); static int max_pdus = MAX_PDUS; SYSCTL_INT(_net, OID_AUTO, iscsi_initiator_max_pdus, CTLFLAG_RDTUN, &max_pdus, 0, "Max PDU pool"); static char isid[6+1] = { 0x80, 'D', 'I', 'B', '0', '0', 0 }; static int i_create_session(struct cdev *dev, int *ndev); static int i_ping(struct cdev *dev); static int i_send(struct cdev *dev, caddr_t arg, struct thread *td); static int i_recv(struct cdev *dev, caddr_t arg, struct thread *td); static int i_setsoc(isc_session_t *sp, int fd, struct thread *td); static int i_fullfeature(struct cdev *dev, int flag); static d_open_t iscsi_open; static d_close_t iscsi_close; static d_ioctl_t iscsi_ioctl; #ifdef ISCSI_INITIATOR_DEBUG static d_read_t iscsi_read; #endif static struct cdevsw iscsi_cdevsw = { .d_version = D_VERSION, .d_open = iscsi_open, .d_close = iscsi_close, .d_ioctl = iscsi_ioctl, #ifdef ISCSI_INITIATOR_DEBUG .d_read = iscsi_read, #endif .d_name = "iSCSI", }; static int iscsi_open(struct cdev *dev, int flags, int otype, struct thread *td) { debug_called(8); debug(7, "dev=%d", dev2unit(dev)); if(dev2unit(dev) > max_sessions) { // should not happen return ENODEV; } return 0; } static int iscsi_close(struct cdev *dev, int flag, int otyp, struct thread *td) { isc_session_t *sp; debug_called(8); debug(3, "session=%d flag=%x", dev2unit(dev), flag); if(dev2unit(dev) == max_sessions) { return 0; } sp = dev->si_drv2; if(sp != NULL) { sdebug(3, "sp->flags=%x", sp->flags ); /* | if still in full phase, this probably means | that something went really bad. | it could be a result from 'shutdown', in which case | we will ignore it (so buffers can be flushed). | the problem is that there is no way of differentiating | between a shutdown procedure and 'iscontrol' dying. */ if(sp->flags & ISC_FFPHASE) // delay in case this is a shutdown. tsleep(sp, PRIBIO, "isc-cls", 60*hz); ism_stop(sp); } debug(2, "done"); return 0; } static int iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg, int mode, struct thread *td) { struct isc_softc *sc; isc_session_t *sp; isc_opt_t *opt; int error; debug_called(8); error = 0; if(dev2unit(dev) == max_sessions) { /* | non Session commands */ sc = dev->si_drv1; if(sc == NULL) return ENXIO; switch(cmd) { case ISCSISETSES: error = i_create_session(dev, (int *)arg); if(error == 0) break; default: error = ENXIO; } return error; } /* | session commands */ sp = dev->si_drv2; if(sp == NULL) return ENXIO; sdebug(6, "dev=%d cmd=%d", dev2unit(dev), (int)(cmd & 0xff)); switch(cmd) { case ISCSISETSOC: error = i_setsoc(sp, *(u_int *)arg, td); break; case ISCSISETOPT: opt = (isc_opt_t *)arg; error = i_setopt(sp, opt); break; case ISCSISEND: error = i_send(dev, arg, td); break; case ISCSIRECV: error = i_recv(dev, arg, td); break; case ISCSIPING: error = i_ping(dev); break; case ISCSISTART: error = sp->soc == NULL? ENOTCONN: i_fullfeature(dev, 1); if(error == 0) { sp->proc = td->td_proc; SYSCTL_ADD_INT(&sp->clist, SYSCTL_CHILDREN(sp->oid), OID_AUTO, "pid", CTLFLAG_RD, &sp->proc->p_pid, sizeof(pid_t), "control process id"); } break; case ISCSIRESTART: error = sp->soc == NULL? ENOTCONN: i_fullfeature(dev, 2); break; case ISCSISTOP: error = i_fullfeature(dev, 0); break; case ISCSISIGNAL: { int sig = *(int *)arg; if(sig < 0 || sig > _SIG_MAXSIG) error = EINVAL; else sp->signal = sig; break; } case ISCSIGETCAM: { iscsi_cam_t *cp = (iscsi_cam_t *)arg; error = ic_getCamVals(sp, cp); break; } default: error = ENOIOCTL; } return error; } static int iscsi_read(struct cdev *dev, struct uio *uio, int ioflag) { #ifdef ISCSI_INITIATOR_DEBUG struct isc_softc *sc; isc_session_t *sp; pduq_t *pq; char buf[1024]; sc = dev->si_drv1; sp = dev->si_drv2; if(dev2unit(dev) == max_sessions) { sprintf(buf, "/----- Session ------/\n"); uiomove(buf, strlen(buf), uio); int i = 0; TAILQ_FOREACH(sp, &sc->isc_sess, sp_link) { if(uio->uio_resid == 0) return 0; sprintf(buf, "%03d] '%s' '%s'\n", i++, sp->opt.targetAddress, sp->opt.targetName); uiomove(buf, strlen(buf), uio); } sprintf(buf, "free npdu_alloc=%d, npdu_max=%d\n", sc->npdu_alloc, sc->npdu_max); uiomove(buf, strlen(buf), uio); } else { int i = 0; struct socket *so = sp->soc; #define pukeit(i, pq) do {\ sprintf(buf, "%03d] %06x %02x %06x %06x %jd\n",\ i, ntohl(pq->pdu.ipdu.bhs.CmdSN),\ pq->pdu.ipdu.bhs.opcode, ntohl(pq->pdu.ipdu.bhs.itt),\ ntohl(pq->pdu.ipdu.bhs.ExpStSN),\ (intmax_t)pq->ts.sec);\ } while(0) sprintf(buf, "%d/%d /---- hld -----/\n", sp->stats.nhld, sp->stats.max_hld); uiomove(buf, strlen(buf), uio); TAILQ_FOREACH(pq, &sp->hld, pq_link) { if(uio->uio_resid == 0) return 0; pukeit(i, pq); i++; uiomove(buf, strlen(buf), uio); } sprintf(buf, "%d/%d /---- rsp -----/\n", sp->stats.nrsp, sp->stats.max_rsp); uiomove(buf, strlen(buf), uio); i = 0; TAILQ_FOREACH(pq, &sp->rsp, pq_link) { if(uio->uio_resid == 0) return 0; pukeit(i, pq); i++; uiomove(buf, strlen(buf), uio); } sprintf(buf, "%d/%d /---- csnd -----/\n", sp->stats.ncsnd, sp->stats.max_csnd); i = 0; uiomove(buf, strlen(buf), uio); TAILQ_FOREACH(pq, &sp->csnd, pq_link) { if(uio->uio_resid == 0) return 0; pukeit(i, pq); i++; uiomove(buf, strlen(buf), uio); } sprintf(buf, "%d/%d /---- wsnd -----/\n", sp->stats.nwsnd, sp->stats.max_wsnd); i = 0; uiomove(buf, strlen(buf), uio); TAILQ_FOREACH(pq, &sp->wsnd, pq_link) { if(uio->uio_resid == 0) return 0; pukeit(i, pq); i++; uiomove(buf, strlen(buf), uio); } sprintf(buf, "%d/%d /---- isnd -----/\n", sp->stats.nisnd, sp->stats.max_isnd); i = 0; uiomove(buf, strlen(buf), uio); TAILQ_FOREACH(pq, &sp->isnd, pq_link) { if(uio->uio_resid == 0) return 0; pukeit(i, pq); i++; uiomove(buf, strlen(buf), uio); } sprintf(buf, "/---- Stats ---/\n"); uiomove(buf, strlen(buf), uio); sprintf(buf, "recv=%d sent=%d\n", sp->stats.nrecv, sp->stats.nsent); uiomove(buf, strlen(buf), uio); sprintf(buf, "flags=%x pdus: alloc=%d max=%d\n", sp->flags, sc->npdu_alloc, sc->npdu_max); uiomove(buf, strlen(buf), uio); sprintf(buf, "cws=%d last cmd=%x exp=%x max=%x stat=%x itt=%x\n", sp->cws, sp->sn.cmd, sp->sn.expCmd, sp->sn.maxCmd, sp->sn.stat, sp->sn.itt); uiomove(buf, strlen(buf), uio); sprintf(buf, "/---- socket -----/\nso_count=%d so_state=%x\n", so->so_count, so->so_state); uiomove(buf, strlen(buf), uio); } #endif return 0; } static int i_ping(struct cdev *dev) { return 0; } /* | low level I/O */ static int i_setsoc(isc_session_t *sp, int fd, struct thread *td) { cap_rights_t rights; int error = 0; if(sp->soc != NULL) isc_stop_receiver(sp); - error = getsock_cap(td, fd, cap_rights_init(&rights, CAP_SOCK_CLIENT), - &sp->fp, NULL, NULL); + error = getsock_cap(td, fd, cap_rights_init_one(&rights, CAP_SOCK_CLIENT), + &sp->fp, NULL, NULL); if(error) return error; sp->soc = sp->fp->f_data; sp->td = td; isc_start_receiver(sp); return error; } static int i_send(struct cdev *dev, caddr_t arg, struct thread *td) { isc_session_t *sp = dev->si_drv2; caddr_t bp; pduq_t *pq; pdu_t *pp; int n, error; debug_called(8); if(sp->soc == NULL) return ENOTCONN; if((pq = pdu_alloc(sp->isc, M_NOWAIT)) == NULL) return EAGAIN; pp = &pq->pdu; pq->pdu = *(pdu_t *)arg; if((error = i_prepPDU(sp, pq)) != 0) goto out; bp = NULL; if((pq->len - sizeof(union ipdu_u)) > 0) { pq->buf = bp = malloc(pq->len - sizeof(union ipdu_u), M_ISCSIBUF, M_NOWAIT); if(pq->buf == NULL) { error = EAGAIN; goto out; } } else pq->buf = NULL; // just in case? sdebug(2, "len=%d ahs_len=%d ds_len=%d buf=%zu@%p", pq->len, pp->ahs_len, pp->ds_len, pq->len - sizeof(union ipdu_u), bp); if(pp->ahs_len) { // XXX: never tested, looks suspicious n = pp->ahs_len; error = copyin(pp->ahs_addr, bp, n); if(error != 0) { sdebug(3, "copyin ahs: error=%d", error); goto out; } pp->ahs_addr = (ahs_t *)bp; bp += n; } if(pp->ds_len) { n = pp->ds_len; error = copyin(pp->ds_addr, bp, n); if(error != 0) { sdebug(3, "copyin ds: error=%d", error); goto out; } pp->ds_addr = bp; bp += n; while(n & 03) { n++; *bp++ = 0; } } error = isc_qout(sp, pq); if(error == 0) wakeup(&sp->flags); // XXX: to 'push' proc_out ... out: if(error) pdu_free(sp->isc, pq); return error; } static int i_recv(struct cdev *dev, caddr_t arg, struct thread *td) { isc_session_t *sp = dev->si_drv2; pduq_t *pq; pdu_t *pp, *up; caddr_t bp; int error, mustfree, cnt; size_t need, have, n; debug_called(8); if(sp == NULL) return EIO; if(sp->soc == NULL) return ENOTCONN; cnt = 6; // XXX: maybe the user can request a time out? mtx_lock(&sp->rsp_mtx); while((pq = TAILQ_FIRST(&sp->rsp)) == NULL) { msleep(&sp->rsp, &sp->rsp_mtx, PRIBIO, "isc_rsp", hz*10); if(cnt-- == 0) break; // XXX: for now, needs work } if(pq != NULL) { sp->stats.nrsp--; TAILQ_REMOVE(&sp->rsp, pq, pq_link); } mtx_unlock(&sp->rsp_mtx); sdebug(6, "cnt=%d", cnt); if(pq == NULL) { error = ENOTCONN; sdebug(3, "error=%d sp->flags=%x ", error, sp->flags); return error; } up = (pdu_t *)arg; pp = &pq->pdu; up->ipdu = pp->ipdu; n = 0; up->ds_len = 0; up->ahs_len = 0; error = 0; if(pq->mp) { u_int len; // Grr... len = 0; if(pp->ahs_len) { len += pp->ahs_len; } if(pp->ds_len) { len += pp->ds_len; } mustfree = 0; if(len > pq->mp->m_len) { mustfree++; bp = malloc(len, M_TMP, M_WAITOK); sdebug(4, "need mbufcopy: %d", len); i_mbufcopy(pq->mp, bp, len); } else bp = mtod(pq->mp, caddr_t); if(pp->ahs_len) { need = pp->ahs_len; n = MIN(up->ahs_size, need); error = copyout(bp, (caddr_t)up->ahs_addr, n); up->ahs_len = n; bp += need; } if(!error && pp->ds_len) { need = pp->ds_len; if((have = up->ds_size) == 0) { have = up->ahs_size - n; up->ds_addr = (caddr_t)up->ahs_addr + n; } n = MIN(have, need); error = copyout(bp, (caddr_t)up->ds_addr, n); up->ds_len = n; } if(mustfree) free(bp, M_TMP); } sdebug(6, "len=%d ahs_len=%d ds_len=%d", pq->len, pp->ahs_len, pp->ds_len); pdu_free(sp->isc, pq); return error; } static int i_fullfeature(struct cdev *dev, int flag) { isc_session_t *sp = dev->si_drv2; int error; sdebug(2, "flag=%d", flag); error = 0; switch(flag) { case 0: // stop sp->flags &= ~ISC_FFPHASE; break; case 1: // start sp->flags |= ISC_FFPHASE; error = ic_init(sp); break; case 2: // restart sp->flags |= ISC_FFPHASE; ism_restart(sp); break; } return error; } static int i_create_session(struct cdev *dev, int *ndev) { struct isc_softc *sc = dev->si_drv1; isc_session_t *sp; int error, n; debug_called(8); sp = malloc(sizeof(isc_session_t), M_ISCSI, M_WAITOK | M_ZERO); if(sp == NULL) return ENOMEM; sx_xlock(&sc->unit_sx); if((n = alloc_unr(sc->unit)) < 0) { sx_unlock(&sc->unit_sx); free(sp, M_ISCSI); xdebug("too many sessions!"); return EPERM; } sx_unlock(&sc->unit_sx); mtx_lock(&sc->isc_mtx); TAILQ_INSERT_TAIL(&sc->isc_sess, sp, sp_link); isc->nsess++; mtx_unlock(&sc->isc_mtx); sp->dev = make_dev(&iscsi_cdevsw, n, UID_ROOT, GID_WHEEL, 0600, "iscsi%d", n); *ndev = sp->sid = n; sp->isc = sc; sp->dev->si_drv1 = sc; sp->dev->si_drv2 = sp; sp->opt.maxRecvDataSegmentLength = 8192; sp->opt.maxXmitDataSegmentLength = 8192; sp->opt.maxBurstLength = 65536; // 64k sp->opt.maxluns = ISCSI_MAX_LUNS; error = ism_start(sp); return error; } #ifdef notused static void iscsi_counters(isc_session_t *sp) { int h, r, s; pduq_t *pq; #define _puke(i, pq) do {\ debug(2, "%03d] %06x %02x %x %ld %jd %x\n",\ i, ntohl( pq->pdu.ipdu.bhs.CmdSN), \ pq->pdu.ipdu.bhs.opcode, ntohl(pq->pdu.ipdu.bhs.itt),\ (long)pq->ts.sec, pq->ts.frac, pq->flags);\ } while(0) h = r = s = 0; TAILQ_FOREACH(pq, &sp->hld, pq_link) { _puke(h, pq); h++; } TAILQ_FOREACH(pq, &sp->rsp, pq_link) r++; TAILQ_FOREACH(pq, &sp->csnd, pq_link) s++; TAILQ_FOREACH(pq, &sp->wsnd, pq_link) s++; TAILQ_FOREACH(pq, &sp->isnd, pq_link) s++; debug(2, "hld=%d rsp=%d snd=%d", h, r, s); } #endif static void iscsi_shutdown(void *v) { struct isc_softc *sc = v; isc_session_t *sp; int n; debug_called(8); if(sc == NULL) { xdebug("sc is NULL!"); return; } #ifdef DO_EVENTHANDLER if(sc->eh == NULL) debug(2, "sc->eh is NULL"); else { EVENTHANDLER_DEREGISTER(shutdown_pre_sync, sc->eh); debug(2, "done n=%d", sc->nsess); } #endif n = 0; TAILQ_FOREACH(sp, &sc->isc_sess, sp_link) { debug(2, "%2d] sp->flags=0x%08x", n, sp->flags); n++; } debug(2, "done"); } static void free_pdus(struct isc_softc *sc) { debug_called(8); if(sc->pdu_zone != NULL) { uma_zdestroy(sc->pdu_zone); sc->pdu_zone = NULL; } } static int iscsi_start(void) { debug_called(8); isc = malloc(sizeof(struct isc_softc), M_ISCSI, M_ZERO|M_WAITOK); mtx_init(&isc->isc_mtx, "iscsi-isc", NULL, MTX_DEF); TAILQ_INIT(&isc->isc_sess); /* | now init the free pdu list */ isc->pdu_zone = uma_zcreate("pdu", sizeof(pduq_t), NULL, NULL, NULL, NULL, 0, 0); uma_zone_set_max(isc->pdu_zone, max_pdus); isc->unit = new_unrhdr(0, max_sessions-1, NULL); sx_init(&isc->unit_sx, "iscsi sx"); #ifdef DO_EVENTHANDLER if((isc->eh = EVENTHANDLER_REGISTER(shutdown_pre_sync, iscsi_shutdown, sc, SHUTDOWN_PRI_DEFAULT-1)) == NULL) xdebug("shutdown event registration failed\n"); #endif /* | sysctl stuff */ sysctl_ctx_init(&isc->clist); isc->oid = SYSCTL_ADD_NODE(&isc->clist, SYSCTL_STATIC_CHILDREN(_net), OID_AUTO, "iscsi_initiator", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "iSCSI Subsystem"); SYSCTL_ADD_STRING(&isc->clist, SYSCTL_CHILDREN(isc->oid), OID_AUTO, "driver_version", CTLFLAG_RD, iscsi_driver_version, 0, "iscsi driver version"); SYSCTL_ADD_STRING(&isc->clist, SYSCTL_CHILDREN(isc->oid), OID_AUTO, "isid", CTLFLAG_RW, isid, 6+1, "initiator part of the Session Identifier"); SYSCTL_ADD_INT(&isc->clist, SYSCTL_CHILDREN(isc->oid), OID_AUTO, "sessions", CTLFLAG_RD, &isc->nsess, sizeof(isc->nsess), "number of active session"); #ifdef ISCSI_INITIATOR_DEBUG mtx_init(&iscsi_dbg_mtx, "iscsi_dbg", NULL, MTX_DEF); #endif isc->dev = make_dev_credf(MAKEDEV_CHECKNAME, &iscsi_cdevsw, max_sessions, NULL, UID_ROOT, GID_WHEEL, 0600, "iscsi"); if (isc->dev == NULL) { xdebug("iscsi_initiator: make_dev_credf failed"); return (EEXIST); } isc->dev->si_drv1 = isc; printf("iscsi: version %s\n", iscsi_driver_version); return (0); } /* | Notes: | unload SHOULD fail if there is activity | activity: there is/are active session/s */ static void iscsi_stop(void) { isc_session_t *sp, *sp_tmp; debug_called(8); /* | go through all the sessions | Note: close should have done this ... */ TAILQ_FOREACH_SAFE(sp, &isc->isc_sess, sp_link, sp_tmp) { //XXX: check for activity ... ism_stop(sp); } mtx_destroy(&isc->isc_mtx); sx_destroy(&isc->unit_sx); free_pdus(isc); if(isc->dev) destroy_dev(isc->dev); if(sysctl_ctx_free(&isc->clist)) xdebug("sysctl_ctx_free failed"); iscsi_shutdown(isc); // XXX: check EVENTHANDLER_ ... #ifdef ISCSI_INITIATOR_DEBUG mtx_destroy(&iscsi_dbg_mtx); #endif free(isc, M_ISCSI); } static int iscsi_modevent(module_t mod, int what, void *arg) { int error = 0; debug_called(8); switch(what) { case MOD_LOAD: error = iscsi_start(); break; case MOD_QUIESCE: if(isc->nsess) { xdebug("iscsi module busy(nsess=%d), cannot unload", isc->nsess); log(LOG_ERR, "iscsi module busy, cannot unload"); } return isc->nsess; case MOD_SHUTDOWN: break; case MOD_UNLOAD: iscsi_stop(); break; default: break; } return (error); } moduledata_t iscsi_mod = { "iscsi_initiator", (modeventhand_t) iscsi_modevent, 0 }; #ifdef ISCSI_ROOT static void iscsi_rootconf(void) { #if 0 nfs_setup_diskless(); if (nfs_diskless_valid) rootdevnames[0] = "nfs:"; #endif printf("** iscsi_rootconf **\n"); } SYSINIT(cpu_rootconf1, SI_SUB_ROOT_CONF, SI_ORDER_FIRST, iscsi_rootconf, NULL) #endif DECLARE_MODULE(iscsi_initiator, iscsi_mod, SI_SUB_DRIVERS, SI_ORDER_MIDDLE); MODULE_DEPEND(iscsi_initiator, cam, 1, 1, 1); diff --git a/sys/dev/mfi/mfi_linux.c b/sys/dev/mfi/mfi_linux.c index 1452b80c280f..892b4f7a9fba 100644 --- a/sys/dev/mfi/mfi_linux.c +++ b/sys/dev/mfi/mfi_linux.c @@ -1,109 +1,109 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2006 IronPort Systems * 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 #if defined(__amd64__) /* Assume amd64 wants 32 bit Linux */ #include #include #else #include #include #endif #include #include #include #include /* There are multiple ioctl number ranges that need to be handled */ #define MFI_LINUX_IOCTL_MIN 0x4d00 #define MFI_LINUX_IOCTL_MAX 0x4d04 static linux_ioctl_function_t mfi_linux_ioctl; static struct linux_ioctl_handler mfi_linux_handler = {mfi_linux_ioctl, MFI_LINUX_IOCTL_MIN, MFI_LINUX_IOCTL_MAX}; SYSINIT (mfi_register, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_ioctl_register_handler, &mfi_linux_handler); SYSUNINIT(mfi_unregister, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_ioctl_unregister_handler, &mfi_linux_handler); static struct linux_device_handler mfi_device_handler = { "mfi", "megaraid_sas", "mfi0", "megaraid_sas_ioctl_node", -1, 0, 1}; SYSINIT (mfi_register2, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_device_register_handler, &mfi_device_handler); SYSUNINIT(mfi_unregister2, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_device_unregister_handler, &mfi_device_handler); static int mfi_linux_modevent(module_t mod, int cmd, void *data) { return (0); } DEV_MODULE(mfi_linux, mfi_linux_modevent, NULL); MODULE_DEPEND(mfi, linux, 1, 1, 1); static int mfi_linux_ioctl(struct thread *p, struct linux_ioctl_args *args) { cap_rights_t rights; struct file *fp; int error; u_long cmd = args->cmd; switch (cmd) { case MFI_LINUX_CMD: cmd = MFI_LINUX_CMD_2; break; case MFI_LINUX_SET_AEN: cmd = MFI_LINUX_SET_AEN_2; break; } - error = fget(p, args->fd, cap_rights_init(&rights, CAP_IOCTL), &fp); + error = fget(p, args->fd, cap_rights_init_one(&rights, CAP_IOCTL), &fp); if (error != 0) return (error); error = fo_ioctl(fp, cmd, (caddr_t)args->arg, p->td_ucred, p); fdrop(fp, p); return (error); } diff --git a/sys/dev/mrsas/mrsas_linux.c b/sys/dev/mrsas/mrsas_linux.c index 1794f977a21d..11ce3c3f8d54 100644 --- a/sys/dev/mrsas/mrsas_linux.c +++ b/sys/dev/mrsas/mrsas_linux.c @@ -1,141 +1,141 @@ /* * Copyright (c) 2015, AVAGO Tech. All rights reserved. Author: Kashyap Desai, * Copyright (c) 2014, LSI Corp. All rights reserved. Author: Kashyap Desai, * Sibananda Sahu Support: freebsdraid@avagotech.com * * 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. 3. Neither the name of the * nor the names of its contributors may be used to endorse or * promote products derived from this software without specific prior written * permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS 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 COPYRIGHT HOLDER 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. * * The views and conclusions contained in the software and documentation are * those of the authors and should not be interpreted as representing * official policies,either expressed or implied, of the FreeBSD Project. * * Send feedback to: Mail to: AVAGO TECHNOLOGIES, 1621 * Barber Lane, Milpitas, CA 95035 ATTN: MegaRaid FreeBSD * */ #include __FBSDID("$FreeBSD$"); #include #include #if (__FreeBSD_version >= 1001511) #include #elif (__FreeBSD_version > 900000) #include #endif #include #include #include #include #include #include #if defined(__amd64__) /* Assume amd64 wants 32 bit Linux */ #include #include #else #include #include #endif #include #include #include #undef COMPAT_FREEBSD32 #include /* There are multiple ioctl number ranges that need to be handled */ #define MRSAS_LINUX_IOCTL_MIN 0x4d00 #define MRSAS_LINUX_IOCTL_MAX 0x4d01 static linux_ioctl_function_t mrsas_linux_ioctl; static struct linux_ioctl_handler mrsas_linux_handler = {mrsas_linux_ioctl, MRSAS_LINUX_IOCTL_MIN, MRSAS_LINUX_IOCTL_MAX}; SYSINIT(mrsas_register, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_ioctl_register_handler, &mrsas_linux_handler); SYSUNINIT(mrsas_unregister, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_ioctl_unregister_handler, &mrsas_linux_handler); static struct linux_device_handler mrsas_device_handler = {"mrsas", "megaraid_sas", "mrsas0", "megaraid_sas_ioctl_node", -1, 0, 1}; SYSINIT(mrsas_register2, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_device_register_handler, &mrsas_device_handler); SYSUNINIT(mrsas_unregister2, SI_SUB_KLD, SI_ORDER_MIDDLE, linux_device_unregister_handler, &mrsas_device_handler); static int mrsas_linux_modevent(module_t mod __unused, int cmd __unused, void *data __unused) { return (0); } /* * mrsas_linux_ioctl: linux emulator IOCtl commands entry point. * * This function is the entry point for IOCtls from linux binaries. * It calls the mrsas_ioctl function for processing * depending on the IOCTL command received. */ static int mrsas_linux_ioctl(struct thread *p, struct linux_ioctl_args *args) { #if (__FreeBSD_version >= 1000000) cap_rights_t rights; #endif struct file *fp; int error; u_long cmd = args->cmd; if (cmd != MRSAS_LINUX_CMD32) { error = ENOTSUP; goto END; } #if (__FreeBSD_version >= 1000000) - error = fget(p, args->fd, cap_rights_init(&rights, CAP_IOCTL), &fp); + error = fget(p, args->fd, cap_rights_init_one(&rights, CAP_IOCTL), &fp); #elif (__FreeBSD_version <= 900000) error = fget(p, args->fd, &fp); #else /* For FreeBSD version greater than * 9.0.0 but less than 10.0.0 */ error = fget(p, args->fd, CAP_IOCTL, &fp); #endif if (error != 0) goto END; error = fo_ioctl(fp, cmd, (caddr_t)args->arg, p->td_ucred, p); fdrop(fp, p); END: return (error); } DEV_MODULE(mrsas_linux, mrsas_linux_modevent, NULL); MODULE_DEPEND(mrsas, linux, 1, 1, 1); diff --git a/sys/dev/tdfx/tdfx_linux.c b/sys/dev/tdfx/tdfx_linux.c index e20763b37647..922ff5478a42 100644 --- a/sys/dev/tdfx/tdfx_linux.c +++ b/sys/dev/tdfx/tdfx_linux.c @@ -1,92 +1,92 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2006 The FreeBSD Project * 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 LINUX_IOCTL_SET(tdfx, LINUX_IOCTL_TDFX_MIN, LINUX_IOCTL_TDFX_MAX); /* * Linux emulation IOCTL for /dev/tdfx */ static int linux_ioctl_tdfx(struct thread *td, struct linux_ioctl_args* args) { cap_rights_t rights; int error = 0; u_long cmd = args->cmd & 0xffff; /* The structure passed to ioctl has two shorts, one int and one void*. */ char d_pio[2*sizeof(short) + sizeof(int) + sizeof(void*)]; struct file *fp; - error = fget(td, args->fd, cap_rights_init(&rights, CAP_IOCTL), &fp); + error = fget(td, args->fd, cap_rights_init_one(&rights, CAP_IOCTL), &fp); if (error != 0) return (error); /* We simply copy the data and send it right to ioctl */ copyin((caddr_t)args->arg, &d_pio, sizeof(d_pio)); error = fo_ioctl(fp, cmd, (caddr_t)&d_pio, td->td_ucred, td); fdrop(fp, td); return error; } static int tdfx_linux_modevent(struct module *mod __unused, int what, void *arg __unused) { switch (what) { case MOD_LOAD: case MOD_UNLOAD: return (0); } return (EOPNOTSUPP); } static moduledata_t tdfx_linux_mod = { "tdfx_linux", tdfx_linux_modevent, 0 }; /* As in SYSCALL_MODULE */ DECLARE_MODULE(tdfx_linux, tdfx_linux_mod, SI_SUB_DRIVERS, SI_ORDER_MIDDLE); MODULE_VERSION(tdfx_linux, 1); MODULE_DEPEND(tdfx_linux, tdfx, 1, 1, 1); MODULE_DEPEND(tdfx_linux, linux, 1, 1, 1); diff --git a/sys/fs/fdescfs/fdesc_vnops.c b/sys/fs/fdescfs/fdesc_vnops.c index ba13aaf1ce85..1271b50e6e94 100644 --- a/sys/fs/fdescfs/fdesc_vnops.c +++ b/sys/fs/fdescfs/fdesc_vnops.c @@ -1,660 +1,660 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1992, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software donated to Berkeley by * Jan-Simon Pendry. * * 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. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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. * * @(#)fdesc_vnops.c 8.9 (Berkeley) 1/21/94 * * $FreeBSD$ */ /* * /dev/fd Filesystem */ #include #include #include #include #include #include #include /* boottime */ #include #include #include #include /* Must come after sys/malloc.h */ #include #include #include #include #include #include #include #include #define NFDCACHE 4 #define FD_NHASH(ix) \ (&fdhashtbl[(ix) & fdhash]) static LIST_HEAD(fdhashhead, fdescnode) *fdhashtbl; static u_long fdhash; struct mtx fdesc_hashmtx; static vop_getattr_t fdesc_getattr; static vop_lookup_t fdesc_lookup; static vop_open_t fdesc_open; static vop_pathconf_t fdesc_pathconf; static vop_readdir_t fdesc_readdir; static vop_readlink_t fdesc_readlink; static vop_reclaim_t fdesc_reclaim; static vop_setattr_t fdesc_setattr; static struct vop_vector fdesc_vnodeops = { .vop_default = &default_vnodeops, .vop_access = VOP_NULL, .vop_getattr = fdesc_getattr, .vop_lookup = fdesc_lookup, .vop_open = fdesc_open, .vop_pathconf = fdesc_pathconf, .vop_readdir = fdesc_readdir, .vop_readlink = fdesc_readlink, .vop_reclaim = fdesc_reclaim, .vop_setattr = fdesc_setattr, }; VFS_VOP_VECTOR_REGISTER(fdesc_vnodeops); static void fdesc_insmntque_dtr(struct vnode *, void *); static void fdesc_remove_entry(struct fdescnode *); /* * Initialise cache headers */ int fdesc_init(struct vfsconf *vfsp) { mtx_init(&fdesc_hashmtx, "fdescfs_hash", NULL, MTX_DEF); fdhashtbl = hashinit(NFDCACHE, M_CACHE, &fdhash); return (0); } /* * Uninit ready for unload. */ int fdesc_uninit(struct vfsconf *vfsp) { hashdestroy(fdhashtbl, M_CACHE, fdhash); mtx_destroy(&fdesc_hashmtx); return (0); } /* * If allocating vnode fails, call this. */ static void fdesc_insmntque_dtr(struct vnode *vp, void *arg) { vgone(vp); vput(vp); } /* * Remove an entry from the hash if it exists. */ static void fdesc_remove_entry(struct fdescnode *fd) { struct fdhashhead *fc; struct fdescnode *fd2; fc = FD_NHASH(fd->fd_ix); mtx_lock(&fdesc_hashmtx); LIST_FOREACH(fd2, fc, fd_hash) { if (fd == fd2) { LIST_REMOVE(fd, fd_hash); break; } } mtx_unlock(&fdesc_hashmtx); } int fdesc_allocvp(fdntype ftype, unsigned fd_fd, int ix, struct mount *mp, struct vnode **vpp) { struct fdescmount *fmp; struct fdhashhead *fc; struct fdescnode *fd, *fd2; struct vnode *vp, *vp2; struct thread *td; int error; td = curthread; fc = FD_NHASH(ix); loop: mtx_lock(&fdesc_hashmtx); /* * If a forced unmount is progressing, we need to drop it. The flags are * protected by the hashmtx. */ fmp = mp->mnt_data; if (fmp == NULL || fmp->flags & FMNT_UNMOUNTF) { mtx_unlock(&fdesc_hashmtx); return (-1); } LIST_FOREACH(fd, fc, fd_hash) { if (fd->fd_ix == ix && fd->fd_vnode->v_mount == mp) { /* Get reference to vnode in case it's being free'd */ vp = fd->fd_vnode; VI_LOCK(vp); mtx_unlock(&fdesc_hashmtx); if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) goto loop; *vpp = vp; return (0); } } mtx_unlock(&fdesc_hashmtx); fd = malloc(sizeof(struct fdescnode), M_TEMP, M_WAITOK); error = getnewvnode("fdescfs", mp, &fdesc_vnodeops, &vp); if (error) { free(fd, M_TEMP); return (error); } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vp->v_data = fd; fd->fd_vnode = vp; fd->fd_type = ftype; fd->fd_fd = fd_fd; fd->fd_ix = ix; if (ftype == Fdesc && fmp->flags & FMNT_LINRDLNKF) vp->v_vflag |= VV_READLINK; error = insmntque1(vp, mp, fdesc_insmntque_dtr, NULL); if (error != 0) { *vpp = NULLVP; return (error); } /* Make sure that someone didn't beat us when inserting the vnode. */ mtx_lock(&fdesc_hashmtx); /* * If a forced unmount is progressing, we need to drop it. The flags are * protected by the hashmtx. */ fmp = mp->mnt_data; if (fmp == NULL || fmp->flags & FMNT_UNMOUNTF) { mtx_unlock(&fdesc_hashmtx); vgone(vp); vput(vp); *vpp = NULLVP; return (-1); } LIST_FOREACH(fd2, fc, fd_hash) { if (fd2->fd_ix == ix && fd2->fd_vnode->v_mount == mp) { /* Get reference to vnode in case it's being free'd */ vp2 = fd2->fd_vnode; VI_LOCK(vp2); mtx_unlock(&fdesc_hashmtx); error = vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK); /* Someone beat us, dec use count and wait for reclaim */ vgone(vp); vput(vp); /* If we didn't get it, return no vnode. */ if (error) vp2 = NULLVP; *vpp = vp2; return (error); } } /* If we came here, we can insert it safely. */ LIST_INSERT_HEAD(fc, fd, fd_hash); mtx_unlock(&fdesc_hashmtx); *vpp = vp; return (0); } struct fdesc_get_ino_args { fdntype ftype; unsigned fd_fd; int ix; struct file *fp; struct thread *td; }; static int fdesc_get_ino_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **rvp) { struct fdesc_get_ino_args *a; int error; a = arg; error = fdesc_allocvp(a->ftype, a->fd_fd, a->ix, mp, rvp); fdrop(a->fp, a->td); return (error); } /* * vp is the current namei directory * ndp is the name to locate in that directory... */ static int fdesc_lookup(struct vop_lookup_args *ap) { struct vnode **vpp = ap->a_vpp; struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; char *pname = cnp->cn_nameptr; struct thread *td = cnp->cn_thread; struct file *fp; struct fdesc_get_ino_args arg; int nlen = cnp->cn_namelen; u_int fd, fd1; int error; struct vnode *fvp; if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { error = EROFS; goto bad; } if (cnp->cn_namelen == 1 && *pname == '.') { *vpp = dvp; VREF(dvp); return (0); } if (VTOFDESC(dvp)->fd_type != Froot) { error = ENOTDIR; goto bad; } fd = 0; /* the only time a leading 0 is acceptable is if it's "0" */ if (*pname == '0' && nlen != 1) { error = ENOENT; goto bad; } while (nlen--) { if (*pname < '0' || *pname > '9') { error = ENOENT; goto bad; } fd1 = 10 * fd + *pname++ - '0'; if (fd1 < fd) { error = ENOENT; goto bad; } fd = fd1; } /* * No rights to check since 'fp' isn't actually used. */ if ((error = fget(td, fd, &cap_no_rights, &fp)) != 0) goto bad; /* Check if we're looking up ourselves. */ if (VTOFDESC(dvp)->fd_ix == FD_DESC + fd) { /* * In case we're holding the last reference to the file, the dvp * will be re-acquired. */ vhold(dvp); VOP_UNLOCK(dvp); fdrop(fp, td); /* Re-aquire the lock afterwards. */ vn_lock(dvp, LK_RETRY | LK_EXCLUSIVE); vdrop(dvp); fvp = dvp; if (VN_IS_DOOMED(dvp)) error = ENOENT; } else { /* * Unlock our root node (dvp) when doing this, since we might * deadlock since the vnode might be locked by another thread * and the root vnode lock will be obtained afterwards (in case * we're looking up the fd of the root vnode), which will be the * opposite lock order. Vhold the root vnode first so we don't * lose it. */ arg.ftype = Fdesc; arg.fd_fd = fd; arg.ix = FD_DESC + fd; arg.fp = fp; arg.td = td; error = vn_vget_ino_gen(dvp, fdesc_get_ino_alloc, &arg, LK_EXCLUSIVE, &fvp); } if (error) goto bad; *vpp = fvp; return (0); bad: *vpp = NULL; return (error); } static int fdesc_open(struct vop_open_args *ap) { struct vnode *vp = ap->a_vp; if (VTOFDESC(vp)->fd_type == Froot) return (0); /* * XXX Kludge: set td->td_proc->p_dupfd to contain the value of the file * descriptor being sought for duplication. The error return ensures * that the vnode for this device will be released by vn_open. Open * will detect this special error and take the actions in dupfdopen. * Other callers of vn_open or VOP_OPEN will simply report the * error. */ ap->a_td->td_dupfd = VTOFDESC(vp)->fd_fd; /* XXX */ return (ENODEV); } static int fdesc_pathconf(struct vop_pathconf_args *ap) { struct vnode *vp = ap->a_vp; int error; switch (ap->a_name) { case _PC_NAME_MAX: *ap->a_retval = NAME_MAX; return (0); case _PC_LINK_MAX: if (VTOFDESC(vp)->fd_type == Froot) *ap->a_retval = 2; else *ap->a_retval = 1; return (0); default: if (VTOFDESC(vp)->fd_type == Froot) return (vop_stdpathconf(ap)); vref(vp); VOP_UNLOCK(vp); error = kern_fpathconf(curthread, VTOFDESC(vp)->fd_fd, ap->a_name, ap->a_retval); vn_lock(vp, LK_SHARED | LK_RETRY); vunref(vp); return (error); } } static int fdesc_getattr(struct vop_getattr_args *ap) { struct vnode *vp = ap->a_vp; struct vattr *vap = ap->a_vap; struct timeval boottime; getboottime(&boottime); vap->va_mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH; vap->va_fileid = VTOFDESC(vp)->fd_ix; vap->va_uid = 0; vap->va_gid = 0; vap->va_blocksize = DEV_BSIZE; vap->va_atime.tv_sec = boottime.tv_sec; vap->va_atime.tv_nsec = 0; vap->va_mtime = vap->va_atime; vap->va_ctime = vap->va_mtime; vap->va_gen = 0; vap->va_flags = 0; vap->va_bytes = 0; vap->va_filerev = 0; switch (VTOFDESC(vp)->fd_type) { case Froot: vap->va_type = VDIR; vap->va_nlink = 2; vap->va_size = DEV_BSIZE; vap->va_rdev = NODEV; break; case Fdesc: vap->va_type = (vp->v_vflag & VV_READLINK) == 0 ? VCHR : VLNK; vap->va_nlink = 1; vap->va_size = 0; vap->va_rdev = makedev(0, vap->va_fileid); break; default: panic("fdesc_getattr"); break; } vp->v_type = vap->va_type; return (0); } static int fdesc_setattr(struct vop_setattr_args *ap) { struct vattr *vap = ap->a_vap; struct vnode *vp; struct mount *mp; struct file *fp; struct thread *td = curthread; cap_rights_t rights; unsigned fd; int error; /* * Can't mess with the root vnode */ if (VTOFDESC(ap->a_vp)->fd_type == Froot) return (EACCES); fd = VTOFDESC(ap->a_vp)->fd_fd; /* * Allow setattr where there is an underlying vnode. */ error = getvnode(td, fd, - cap_rights_init(&rights, CAP_EXTATTR_SET), &fp); + cap_rights_init_one(&rights, CAP_EXTATTR_SET), &fp); if (error) { /* * getvnode() returns EINVAL if the file descriptor is not * backed by a vnode. Silently drop all changes except * chflags(2) in this case. */ if (error == EINVAL) { if (vap->va_flags != VNOVAL) error = EOPNOTSUPP; else error = 0; } return (error); } vp = fp->f_vnode; if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) == 0) { vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); error = VOP_SETATTR(vp, ap->a_vap, ap->a_cred); VOP_UNLOCK(vp); vn_finished_write(mp); } fdrop(fp, td); return (error); } #define UIO_MX _GENERIC_DIRLEN(10) /* number of symbols in INT_MAX printout */ static int fdesc_readdir(struct vop_readdir_args *ap) { struct fdescmount *fmp; struct uio *uio = ap->a_uio; struct filedesc *fdp; struct dirent d; struct dirent *dp = &d; int error, i, off, fcnt; if (VTOFDESC(ap->a_vp)->fd_type != Froot) panic("fdesc_readdir: not dir"); fmp = VFSTOFDESC(ap->a_vp->v_mount); if (ap->a_ncookies != NULL) *ap->a_ncookies = 0; off = (int)uio->uio_offset; if (off != uio->uio_offset || off < 0 || (u_int)off % UIO_MX != 0 || uio->uio_resid < UIO_MX) return (EINVAL); i = (u_int)off / UIO_MX; fdp = uio->uio_td->td_proc->p_fd; error = 0; fcnt = i - 2; /* The first two nodes are `.' and `..' */ FILEDESC_SLOCK(fdp); while (i < fdp->fd_nfiles + 2 && uio->uio_resid >= UIO_MX) { bzero((caddr_t)dp, UIO_MX); switch (i) { case 0: /* `.' */ case 1: /* `..' */ dp->d_fileno = i + FD_ROOT; dp->d_namlen = i + 1; dp->d_reclen = UIO_MX; bcopy("..", dp->d_name, dp->d_namlen); dp->d_type = DT_DIR; dirent_terminate(dp); break; default: if (fdp->fd_ofiles[fcnt].fde_file == NULL) break; dp->d_namlen = sprintf(dp->d_name, "%d", fcnt); dp->d_reclen = UIO_MX; dp->d_type = (fmp->flags & FMNT_LINRDLNKF) == 0 ? DT_CHR : DT_LNK; dp->d_fileno = i + FD_DESC; dirent_terminate(dp); break; } /* NOTE: d_off is the offset of the *next* entry. */ dp->d_off = UIO_MX * (i + 1); if (dp->d_namlen != 0) { /* * And ship to userland */ FILEDESC_SUNLOCK(fdp); error = uiomove(dp, UIO_MX, uio); if (error) goto done; FILEDESC_SLOCK(fdp); } i++; fcnt++; } FILEDESC_SUNLOCK(fdp); done: uio->uio_offset = i * UIO_MX; return (error); } static int fdesc_reclaim(struct vop_reclaim_args *ap) { struct vnode *vp; struct fdescnode *fd; vp = ap->a_vp; fd = VTOFDESC(vp); fdesc_remove_entry(fd); free(vp->v_data, M_TEMP); vp->v_data = NULL; return (0); } static int fdesc_readlink(struct vop_readlink_args *va) { struct vnode *vp, *vn; struct thread *td; struct uio *uio; struct file *fp; char *freepath, *fullpath; size_t pathlen; int lockflags, fd_fd; int error; freepath = NULL; vn = va->a_vp; if (VTOFDESC(vn)->fd_type != Fdesc) panic("fdesc_readlink: not fdescfs link"); fd_fd = ((struct fdescnode *)vn->v_data)->fd_fd; lockflags = VOP_ISLOCKED(vn); VOP_UNLOCK(vn); td = curthread; error = fget_cap(td, fd_fd, &cap_no_rights, &fp, NULL); if (error != 0) goto out; switch (fp->f_type) { case DTYPE_VNODE: vp = fp->f_vnode; error = vn_fullpath(vp, &fullpath, &freepath); break; default: fullpath = "anon_inode:[unknown]"; break; } if (error == 0) { uio = va->a_uio; pathlen = strlen(fullpath); error = uiomove(fullpath, pathlen, uio); } if (freepath != NULL) free(freepath, M_TEMP); fdrop(fp, td); out: vn_lock(vn, lockflags | LK_RETRY); return (error); } diff --git a/sys/fs/nfsclient/nfs_clport.c b/sys/fs/nfsclient/nfs_clport.c index 730877b6eb6f..9c0ac5c4364e 100644 --- a/sys/fs/nfsclient/nfs_clport.c +++ b/sys/fs/nfsclient/nfs_clport.c @@ -1,1420 +1,1420 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * 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. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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_inet.h" #include "opt_inet6.h" #include /* * generally, I don't like #includes inside .h files, but it seems to * be the easiest way to handle the port. */ #include #include #include #include #include #include #include #include #include #include #ifdef KDTRACE_HOOKS dtrace_nfsclient_attrcache_flush_probe_func_t dtrace_nfscl_attrcache_flush_done_probe; uint32_t nfscl_attrcache_flush_done_id; dtrace_nfsclient_attrcache_get_hit_probe_func_t dtrace_nfscl_attrcache_get_hit_probe; uint32_t nfscl_attrcache_get_hit_id; dtrace_nfsclient_attrcache_get_miss_probe_func_t dtrace_nfscl_attrcache_get_miss_probe; uint32_t nfscl_attrcache_get_miss_id; dtrace_nfsclient_attrcache_load_probe_func_t dtrace_nfscl_attrcache_load_done_probe; uint32_t nfscl_attrcache_load_done_id; #endif /* !KDTRACE_HOOKS */ extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; extern struct vop_vector newnfs_vnodeops; extern struct vop_vector newnfs_fifoops; extern uma_zone_t newnfsnode_zone; extern struct buf_ops buf_ops_newnfs; extern uma_zone_t ncl_pbuf_zone; extern short nfsv4_cbport; extern int nfscl_enablecallb; extern int nfs_numnfscbd; extern int nfscl_inited; struct mtx ncl_iod_mutex; NFSDLOCKMUTEX; extern struct mtx nfsrv_dslock_mtx; extern void (*ncl_call_invalcaches)(struct vnode *); SYSCTL_DECL(_vfs_nfs); static int ncl_fileid_maxwarnings = 10; SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN, &ncl_fileid_maxwarnings, 0, "Limit fileid corruption warnings; 0 is off; -1 is unlimited"); static volatile int ncl_fileid_nwarnings; static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *, struct nfsvattr *); /* * Comparison function for vfs_hash functions. */ int newnfs_vncmpf(struct vnode *vp, void *arg) { struct nfsfh *nfhp = (struct nfsfh *)arg; struct nfsnode *np = VTONFS(vp); if (np->n_fhp->nfh_len != nfhp->nfh_len || NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len)) return (1); return (0); } /* * Look up a vnode/nfsnode by file handle. * Callers must check for mount points!! * In all cases, a pointer to a * nfsnode structure is returned. * This variant takes a "struct nfsfh *" as second argument and uses * that structure up, either by hanging off the nfsnode or FREEing it. */ int nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp, struct componentname *cnp, struct thread *td, struct nfsnode **npp, void *stuff, int lkflags) { struct nfsnode *np, *dnp; struct vnode *vp, *nvp; struct nfsv4node *newd, *oldd; int error; u_int hash; struct nfsmount *nmp; nmp = VFSTONFS(mntp); dnp = VTONFS(dvp); *npp = NULL; hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT); error = vfs_hash_get(mntp, hash, lkflags, td, &nvp, newnfs_vncmpf, nfhp); if (error == 0 && nvp != NULL) { /* * I believe there is a slight chance that vgonel() could * get called on this vnode between when NFSVOPLOCK() drops * the VI_LOCK() and vget() acquires it again, so that it * hasn't yet had v_usecount incremented. If this were to * happen, the VIRF_DOOMED flag would be set, so check for * that here. Since we now have the v_usecount incremented, * we should be ok until we vrele() it, if the VIRF_DOOMED * flag isn't set now. */ VI_LOCK(nvp); if (VN_IS_DOOMED(nvp)) { VI_UNLOCK(nvp); vrele(nvp); error = ENOENT; } else { VI_UNLOCK(nvp); } } if (error) { free(nfhp, M_NFSFH); return (error); } if (nvp != NULL) { np = VTONFS(nvp); /* * For NFSv4, check to see if it is the same name and * replace the name, if it is different. */ oldd = newd = NULL; if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL && nvp->v_type == VREG && (np->n_v4->n4_namelen != cnp->cn_namelen || NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), cnp->cn_namelen) || dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, dnp->n_fhp->nfh_len))) { newd = malloc( sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len + + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK); NFSLOCKNODE(np); if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG && (np->n_v4->n4_namelen != cnp->cn_namelen || NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), cnp->cn_namelen) || dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, dnp->n_fhp->nfh_len))) { oldd = np->n_v4; np->n_v4 = newd; newd = NULL; np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; np->n_v4->n4_namelen = cnp->cn_namelen; NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, dnp->n_fhp->nfh_len); NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), cnp->cn_namelen); } NFSUNLOCKNODE(np); } if (newd != NULL) free(newd, M_NFSV4NODE); if (oldd != NULL) free(oldd, M_NFSV4NODE); *npp = np; free(nfhp, M_NFSFH); return (0); } np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO); error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp); if (error) { uma_zfree(newnfsnode_zone, np); free(nfhp, M_NFSFH); return (error); } vp = nvp; KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0")); vp->v_bufobj.bo_ops = &buf_ops_newnfs; vp->v_data = np; np->n_vnode = vp; /* * Initialize the mutex even if the vnode is going to be a loser. * This simplifies the logic in reclaim, which can then unconditionally * destroy the mutex (in the case of the loser, or if hash_insert * happened to return an error no special casing is needed). */ mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK); lockinit(&np->n_excl, PVFS, "nfsupg", VLKTIMEOUT, LK_NOSHARE | LK_CANRECURSE); /* * Are we getting the root? If so, make sure the vnode flags * are correct */ if ((nfhp->nfh_len == nmp->nm_fhsize) && !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) { if (vp->v_type == VNON) vp->v_type = VDIR; vp->v_vflag |= VV_ROOT; } vp->v_vflag |= VV_VMSIZEVNLOCK; np->n_fhp = nfhp; /* * For NFSv4, we have to attach the directory file handle and * file name, so that Open Ops can be done later. */ if (nmp->nm_flag & NFSMNT_NFSV4) { np->n_v4 = malloc(sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK); np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; np->n_v4->n4_namelen = cnp->cn_namelen; NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, dnp->n_fhp->nfh_len); NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), cnp->cn_namelen); } else { np->n_v4 = NULL; } /* * NFS supports recursive and shared locking. */ lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL); VN_LOCK_AREC(vp); VN_LOCK_ASHARE(vp); error = insmntque(vp, mntp); if (error != 0) { *npp = NULL; mtx_destroy(&np->n_mtx); lockdestroy(&np->n_excl); free(nfhp, M_NFSFH); if (np->n_v4 != NULL) free(np->n_v4, M_NFSV4NODE); uma_zfree(newnfsnode_zone, np); return (error); } error = vfs_hash_insert(vp, hash, lkflags, td, &nvp, newnfs_vncmpf, nfhp); if (error) return (error); if (nvp != NULL) { *npp = VTONFS(nvp); /* vfs_hash_insert() vput()'s the losing vnode */ return (0); } *npp = np; return (0); } /* * Another variant of nfs_nget(). This one is only used by reopen. It * takes almost the same args as nfs_nget(), but only succeeds if an entry * exists in the cache. (Since files should already be "open" with a * vnode ref cnt on the node when reopen calls this, it should always * succeed.) * Also, don't get a vnode lock, since it may already be locked by some * other process that is handling it. This is ok, since all other threads * on the client are blocked by the nfsc_lock being exclusively held by the * caller of this function. */ int nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize, struct thread *td, struct nfsnode **npp) { struct vnode *nvp; u_int hash; struct nfsfh *nfhp; int error; *npp = NULL; /* For forced dismounts, just return error. */ if (NFSCL_FORCEDISM(mntp)) return (EINTR); nfhp = malloc(sizeof (struct nfsfh) + fhsize, M_NFSFH, M_WAITOK); bcopy(fhp, &nfhp->nfh_fh[0], fhsize); nfhp->nfh_len = fhsize; hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT); /* * First, try to get the vnode locked, but don't block for the lock. */ error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp, newnfs_vncmpf, nfhp); if (error == 0 && nvp != NULL) { NFSVOPUNLOCK(nvp); } else if (error == EBUSY) { /* * It is safe so long as a vflush() with * FORCECLOSE has not been done. Since the Renew thread is * stopped and the MNTK_UNMOUNTF flag is set before doing * a vflush() with FORCECLOSE, we should be ok here. */ if (NFSCL_FORCEDISM(mntp)) error = EINTR; else { vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp); if (nvp == NULL) { error = ENOENT; } else if (VN_IS_DOOMED(nvp)) { error = ENOENT; vrele(nvp); } else { error = 0; } } } free(nfhp, M_NFSFH); if (error) return (error); if (nvp != NULL) { *npp = VTONFS(nvp); return (0); } return (EINVAL); } static void nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap, struct nfsvattr *newnap) { int off; if (ncl_fileid_maxwarnings >= 0 && ncl_fileid_nwarnings >= ncl_fileid_maxwarnings) return; off = 0; if (ncl_fileid_maxwarnings >= 0) { if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings) off = 1; } printf("newnfs: server '%s' error: fileid changed. " "fsid %jx:%jx: expected fileid %#jx, got %#jx. " "(BROKEN NFS SERVER OR MIDDLEWARE)\n", nmp->nm_com.nmcom_hostname, (uintmax_t)nmp->nm_fsid[0], (uintmax_t)nmp->nm_fsid[1], (uintmax_t)oldnap->na_fileid, (uintmax_t)newnap->na_fileid); if (off) printf("newnfs: Logged %d times about fileid corruption; " "going quiet to avoid spamming logs excessively. (Limit " "is: %d).\n", ncl_fileid_nwarnings, ncl_fileid_maxwarnings); } /* * Load the attribute cache (that lives in the nfsnode entry) with * the attributes of the second argument and * Iff vaper not NULL * copy the attributes to *vaper * Similar to nfs_loadattrcache(), except the attributes are passed in * instead of being parsed out of the mbuf list. */ int nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper, void *stuff, int writeattr, int dontshrink) { struct vnode *vp = *vpp; struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper; struct nfsnode *np; struct nfsmount *nmp; struct timespec mtime_save; int error, force_fid_err; error = 0; /* * If v_type == VNON it is a new node, so fill in the v_type, * n_mtime fields. Check to see if it represents a special * device, and if so, check for a possible alias. Once the * correct vnode has been obtained, fill in the rest of the * information. */ np = VTONFS(vp); NFSLOCKNODE(np); if (vp->v_type != nvap->va_type) { vp->v_type = nvap->va_type; if (vp->v_type == VFIFO) vp->v_op = &newnfs_fifoops; np->n_mtime = nvap->va_mtime; } nmp = VFSTONFS(vp->v_mount); vap = &np->n_vattr.na_vattr; mtime_save = vap->va_mtime; if (writeattr) { np->n_vattr.na_filerev = nap->na_filerev; np->n_vattr.na_size = nap->na_size; np->n_vattr.na_mtime = nap->na_mtime; np->n_vattr.na_ctime = nap->na_ctime; np->n_vattr.na_fsid = nap->na_fsid; np->n_vattr.na_mode = nap->na_mode; } else { force_fid_err = 0; KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning, force_fid_err); /* * BROKEN NFS SERVER OR MIDDLEWARE * * Certain NFS servers (certain old proprietary filers ca. * 2006) or broken middleboxes (e.g. WAN accelerator products) * will respond to GETATTR requests with results for a * different fileid. * * The WAN accelerator we've observed not only serves stale * cache results for a given file, it also occasionally serves * results for wholly different files. This causes surprising * problems; for example the cached size attribute of a file * may truncate down and then back up, resulting in zero * regions in file contents read by applications. We observed * this reliably with Clang and .c files during parallel build. * A pcap revealed packet fragmentation and GETATTR RPC * responses with wholly wrong fileids. */ if ((np->n_vattr.na_fileid != 0 && np->n_vattr.na_fileid != nap->na_fileid) || force_fid_err) { nfscl_warn_fileid(nmp, &np->n_vattr, nap); error = EIDRM; goto out; } NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr, sizeof (struct nfsvattr)); } /* * For NFSv4, if the node's fsid is not equal to the mount point's * fsid, return the low order 32bits of the node's fsid. This * allows getcwd(3) to work. There is a chance that the fsid might * be the same as a local fs, but since this is in an NFS mount * point, I don't think that will cause any problems? */ if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) && (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] || nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) { /* * va_fsid needs to be set to some value derived from * np->n_vattr.na_filesid that is not equal * vp->v_mount->mnt_stat.f_fsid[0], so that it changes * from the value used for the top level server volume * in the mounted subtree. */ vn_fsid(vp, vap); if ((uint32_t)vap->va_fsid == np->n_vattr.na_filesid[0]) vap->va_fsid = hash32_buf( np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0); } else vn_fsid(vp, vap); np->n_attrstamp = time_second; if (vap->va_size != np->n_size) { if (vap->va_type == VREG) { if (dontshrink && vap->va_size < np->n_size) { /* * We've been told not to shrink the file; * zero np->n_attrstamp to indicate that * the attributes are stale. */ vap->va_size = np->n_size; np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); } else if (np->n_flag & NMODIFIED) { /* * We've modified the file: Use the larger * of our size, and the server's size. */ if (vap->va_size < np->n_size) { vap->va_size = np->n_size; } else { np->n_size = vap->va_size; np->n_flag |= NSIZECHANGED; } } else { np->n_size = vap->va_size; np->n_flag |= NSIZECHANGED; } } else { np->n_size = vap->va_size; } } /* * The following checks are added to prevent a race between (say) * a READDIR+ and a WRITE. * READDIR+, WRITE requests sent out. * READDIR+ resp, WRITE resp received on client. * However, the WRITE resp was handled before the READDIR+ resp * causing the post op attrs from the write to be loaded first * and the attrs from the READDIR+ to be loaded later. If this * happens, we have stale attrs loaded into the attrcache. * We detect this by for the mtime moving back. We invalidate the * attrcache when this happens. */ if (timespeccmp(&mtime_save, &vap->va_mtime, >)) { /* Size changed or mtime went backwards */ np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); } if (vaper != NULL) { NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap)); if (np->n_flag & NCHG) { if (np->n_flag & NACC) vaper->va_atime = np->n_atim; if (np->n_flag & NUPD) vaper->va_mtime = np->n_mtim; } } out: #ifdef KDTRACE_HOOKS if (np->n_attrstamp != 0) KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error); #endif (void)ncl_pager_setsize(vp, NULL); return (error); } /* * Call vnode_pager_setsize() if the size of the node changed, as * recorded in nfsnode vs. v_object, or delay the call if notifying * the pager is not possible at the moment. * * If nsizep is non-NULL, the call is delayed and the new node size is * provided. Caller should itself call vnode_pager_setsize() if * function returned true. If nsizep is NULL, function tries to call * vnode_pager_setsize() itself if needed and possible, and the nfs * node is unlocked unconditionally, the return value is not useful. */ bool ncl_pager_setsize(struct vnode *vp, u_quad_t *nsizep) { struct nfsnode *np; vm_object_t object; struct vattr *vap; u_quad_t nsize; bool setnsize; np = VTONFS(vp); NFSASSERTNODE(np); vap = &np->n_vattr.na_vattr; nsize = vap->va_size; object = vp->v_object; setnsize = false; if (object != NULL && nsize != object->un_pager.vnp.vnp_size) { if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && (curthread->td_pflags2 & TDP2_SBPAGES) == 0) setnsize = true; else np->n_flag |= NVNSETSZSKIP; } if (nsizep == NULL) { NFSUNLOCKNODE(np); if (setnsize) vnode_pager_setsize(vp, nsize); setnsize = false; } else { *nsizep = nsize; } return (setnsize); } /* * Fill in the client id name. For these bytes: * 1 - they must be unique * 2 - they should be persistent across client reboots * 1 is more critical than 2 * Use the mount point's unique id plus either the uuid or, if that * isn't set, random junk. */ void nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen) { int uuidlen; /* * First, put in the 64bit mount point identifier. */ if (idlen >= sizeof (u_int64_t)) { NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t)); cp += sizeof (u_int64_t); idlen -= sizeof (u_int64_t); } /* * If uuid is non-zero length, use it. */ uuidlen = strlen(uuid); if (uuidlen > 0 && idlen >= uuidlen) { NFSBCOPY(uuid, cp, uuidlen); cp += uuidlen; idlen -= uuidlen; } /* * This only normally happens if the uuid isn't set. */ while (idlen > 0) { *cp++ = (u_int8_t)(arc4random() % 256); idlen--; } } /* * Fill in a lock owner name. For now, pid + the process's creation time. */ void nfscl_filllockowner(void *id, u_int8_t *cp, int flags) { union { u_int32_t lval; u_int8_t cval[4]; } tl; struct proc *p; if (id == NULL) { /* Return the single open_owner of all 0 bytes. */ bzero(cp, NFSV4CL_LOCKNAMELEN); return; } if ((flags & F_POSIX) != 0) { p = (struct proc *)id; tl.lval = p->p_pid; *cp++ = tl.cval[0]; *cp++ = tl.cval[1]; *cp++ = tl.cval[2]; *cp++ = tl.cval[3]; tl.lval = p->p_stats->p_start.tv_sec; *cp++ = tl.cval[0]; *cp++ = tl.cval[1]; *cp++ = tl.cval[2]; *cp++ = tl.cval[3]; tl.lval = p->p_stats->p_start.tv_usec; *cp++ = tl.cval[0]; *cp++ = tl.cval[1]; *cp++ = tl.cval[2]; *cp = tl.cval[3]; } else if ((flags & F_FLOCK) != 0) { bcopy(&id, cp, sizeof(id)); bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id)); } else { printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n"); bzero(cp, NFSV4CL_LOCKNAMELEN); } } /* * Find the parent process for the thread passed in as an argument. * If none exists, return NULL, otherwise return a thread for the parent. * (Can be any of the threads, since it is only used for td->td_proc.) */ NFSPROC_T * nfscl_getparent(struct thread *td) { struct proc *p; struct thread *ptd; if (td == NULL) return (NULL); p = td->td_proc; if (p->p_pid == 0) return (NULL); p = p->p_pptr; if (p == NULL) return (NULL); ptd = TAILQ_FIRST(&p->p_threads); return (ptd); } /* * Start up the renew kernel thread. */ static void start_nfscl(void *arg) { struct nfsclclient *clp; struct thread *td; clp = (struct nfsclclient *)arg; td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads); nfscl_renewthread(clp, td); kproc_exit(0); } void nfscl_start_renewthread(struct nfsclclient *clp) { kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0, "nfscl"); } /* * Handle wcc_data. * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr * as the first Op after PutFH. * (For NFSv4, the postop attributes are after the Op, so they can't be * parsed here. A separate call to nfscl_postop_attr() is required.) */ int nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp, struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff) { u_int32_t *tl; struct nfsnode *np = VTONFS(vp); struct nfsvattr nfsva; int error = 0; if (wccflagp != NULL) *wccflagp = 0; if (nd->nd_flag & ND_NFSV3) { *flagp = 0; NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl == newnfs_true) { NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); if (wccflagp != NULL) { NFSLOCKNODE(np); *wccflagp = (np->n_mtime.tv_sec == fxdr_unsigned(u_int32_t, *(tl + 2)) && np->n_mtime.tv_nsec == fxdr_unsigned(u_int32_t, *(tl + 3))); NFSUNLOCKNODE(np); } } error = nfscl_postop_attr(nd, nap, flagp, stuff); if (wccflagp != NULL && *flagp == 0) *wccflagp = 0; } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == (ND_NFSV4 | ND_V4WCCATTR)) { error = nfsv4_loadattr(nd, NULL, &nfsva, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); if (error) return (error); /* * Get rid of Op# and status for next op. */ NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); if (*++tl) nd->nd_flag |= ND_NOMOREDATA; if (wccflagp != NULL && nfsva.na_vattr.va_mtime.tv_sec != 0) { NFSLOCKNODE(np); *wccflagp = (np->n_mtime.tv_sec == nfsva.na_vattr.va_mtime.tv_sec && np->n_mtime.tv_nsec == nfsva.na_vattr.va_mtime.tv_sec); NFSUNLOCKNODE(np); } } nfsmout: return (error); } /* * Get postop attributes. */ int nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp, void *stuff) { u_int32_t *tl; int error = 0; *retp = 0; if (nd->nd_flag & ND_NOMOREDATA) return (error); if (nd->nd_flag & ND_NFSV3) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); *retp = fxdr_unsigned(int, *tl); } else if (nd->nd_flag & ND_NFSV4) { /* * For NFSv4, the postop attr are at the end, so no point * in looking if nd_repstat != 0. */ if (!nd->nd_repstat) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); if (*(tl + 1)) /* should never happen since nd_repstat != 0 */ nd->nd_flag |= ND_NOMOREDATA; else *retp = 1; } } else if (!nd->nd_repstat) { /* For NFSv2, the attributes are here iff nd_repstat == 0 */ *retp = 1; } if (*retp) { error = nfsm_loadattr(nd, nap); if (error) *retp = 0; } nfsmout: return (error); } /* * nfscl_request() - mostly a wrapper for newnfs_request(). */ int nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p, struct ucred *cred, void *stuff) { int ret, vers; struct nfsmount *nmp; nmp = VFSTONFS(vp->v_mount); if (nd->nd_flag & ND_NFSV4) vers = NFS_VER4; else if (nd->nd_flag & ND_NFSV3) vers = NFS_VER3; else vers = NFS_VER2; ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred, NFS_PROG, vers, NULL, 1, NULL, NULL); return (ret); } /* * fill in this bsden's variant of statfs using nfsstatfs. */ void nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs) { struct statfs *sbp = (struct statfs *)statfs; if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) { sbp->f_bsize = NFS_FABLKSIZE; sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE; sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE; /* * Although sf_abytes is uint64_t and f_bavail is int64_t, * the value after dividing by NFS_FABLKSIZE is small * enough that it will fit in 63bits, so it is ok to * assign it to f_bavail without fear that it will become * negative. */ sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE; sbp->f_files = sfp->sf_tfiles; /* Since f_ffree is int64_t, clip it to 63bits. */ if (sfp->sf_ffiles > INT64_MAX) sbp->f_ffree = INT64_MAX; else sbp->f_ffree = sfp->sf_ffiles; } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) { /* * The type casts to (int32_t) ensure that this code is * compatible with the old NFS client, in that it will * propagate bit31 to the high order bits. This may or may * not be correct for NFSv2, but since it is a legacy * environment, I'd rather retain backwards compatibility. */ sbp->f_bsize = (int32_t)sfp->sf_bsize; sbp->f_blocks = (int32_t)sfp->sf_blocks; sbp->f_bfree = (int32_t)sfp->sf_bfree; sbp->f_bavail = (int32_t)sfp->sf_bavail; sbp->f_files = 0; sbp->f_ffree = 0; } } /* * Use the fsinfo stuff to update the mount point. */ void nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp) { if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) && fsp->fs_wtpref >= NFS_FABLKSIZE) nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) & ~(NFS_FABLKSIZE - 1); if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) { nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1); if (nmp->nm_wsize == 0) nmp->nm_wsize = fsp->fs_wtmax; } if (nmp->nm_wsize < NFS_FABLKSIZE) nmp->nm_wsize = NFS_FABLKSIZE; if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) && fsp->fs_rtpref >= NFS_FABLKSIZE) nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) & ~(NFS_FABLKSIZE - 1); if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) { nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1); if (nmp->nm_rsize == 0) nmp->nm_rsize = fsp->fs_rtmax; } if (nmp->nm_rsize < NFS_FABLKSIZE) nmp->nm_rsize = NFS_FABLKSIZE; if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize) && fsp->fs_dtpref >= NFS_DIRBLKSIZ) nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) & ~(NFS_DIRBLKSIZ - 1); if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) { nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1); if (nmp->nm_readdirsize == 0) nmp->nm_readdirsize = fsp->fs_rtmax; } if (nmp->nm_readdirsize < NFS_DIRBLKSIZ) nmp->nm_readdirsize = NFS_DIRBLKSIZ; if (fsp->fs_maxfilesize > 0 && fsp->fs_maxfilesize < nmp->nm_maxfilesize) nmp->nm_maxfilesize = fsp->fs_maxfilesize; nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp); nmp->nm_state |= NFSSTA_GOTFSINFO; } /* * Lookups source address which should be used to communicate with * @nmp and stores it inside @pdst. * * Returns 0 on success. */ u_int8_t * nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p) { #if defined(INET6) || defined(INET) int fibnum; fibnum = curthread->td_proc->p_fibnum; #endif #ifdef INET if (nmp->nm_nam->sa_family == AF_INET) { struct epoch_tracker et; struct nhop_object *nh; struct sockaddr_in *sin; struct in_addr addr = {}; sin = (struct sockaddr_in *)nmp->nm_nam; NET_EPOCH_ENTER(et); CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred)); nh = fib4_lookup(fibnum, sin->sin_addr, 0, NHR_NONE, 0); CURVNET_RESTORE(); if (nh != NULL) addr = IA_SIN(ifatoia(nh->nh_ifa))->sin_addr; NET_EPOCH_EXIT(et); if (nh == NULL) return (NULL); if (IN_LOOPBACK(ntohl(addr.s_addr))) { /* Ignore loopback addresses */ return (NULL); } *isinet6p = 0; *((struct in_addr *)paddr) = addr; return (u_int8_t *)paddr; } #endif #ifdef INET6 if (nmp->nm_nam->sa_family == AF_INET6) { struct sockaddr_in6 *sin6; int error; sin6 = (struct sockaddr_in6 *)nmp->nm_nam; CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred)); error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr, sin6->sin6_scope_id, NULL, paddr, NULL); CURVNET_RESTORE(); if (error != 0) return (NULL); if (IN6_IS_ADDR_LOOPBACK(paddr)) return (NULL); /* Scope is embedded in */ *isinet6p = 1; return (u_int8_t *)paddr; } #endif return (NULL); } /* * Copy NFS uid, gids from the cred structure. */ void newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr) { int i; KASSERT(cr->cr_ngroups >= 0, ("newnfs_copyincred: negative cr_ngroups")); nfscr->nfsc_uid = cr->cr_uid; nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1); for (i = 0; i < nfscr->nfsc_ngroups; i++) nfscr->nfsc_groups[i] = cr->cr_groups[i]; } /* * Do any client specific initialization. */ void nfscl_init(void) { static int inited = 0; if (inited) return; inited = 1; nfscl_inited = 1; ncl_pbuf_zone = pbuf_zsecond_create("nfspbuf", nswbuf / 2); } /* * Check each of the attributes to be set, to ensure they aren't already * the correct value. Disable setting ones already correct. */ int nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap) { if (vap->va_mode != (mode_t)VNOVAL) { if (vap->va_mode == nvap->na_mode) vap->va_mode = (mode_t)VNOVAL; } if (vap->va_uid != (uid_t)VNOVAL) { if (vap->va_uid == nvap->na_uid) vap->va_uid = (uid_t)VNOVAL; } if (vap->va_gid != (gid_t)VNOVAL) { if (vap->va_gid == nvap->na_gid) vap->va_gid = (gid_t)VNOVAL; } if (vap->va_size != VNOVAL) { if (vap->va_size == nvap->na_size) vap->va_size = VNOVAL; } /* * We are normally called with only a partially initialized * VAP. Since the NFSv3 spec says that server may use the * file attributes to store the verifier, the spec requires * us to do a SETATTR RPC. FreeBSD servers store the verifier * in atime, but we can't really assume that all servers will * so we ensure that our SETATTR sets both atime and mtime. * Set the VA_UTIMES_NULL flag for this case, so that * the server's time will be used. This is needed to * work around a bug in some Solaris servers, where * setting the time TOCLIENT causes the Setattr RPC * to return NFS_OK, but not set va_mode. */ if (vap->va_mtime.tv_sec == VNOVAL) { vfs_timestamp(&vap->va_mtime); vap->va_vaflags |= VA_UTIMES_NULL; } if (vap->va_atime.tv_sec == VNOVAL) vap->va_atime = vap->va_mtime; return (1); } /* * Map nfsv4 errors to errno.h errors. * The uid and gid arguments are only used for NFSERR_BADOWNER and that * error should only be returned for the Open, Create and Setattr Ops. * As such, most calls can just pass in 0 for those arguments. */ int nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid) { struct proc *p; if (error < 10000 || error >= NFSERR_STALEWRITEVERF) return (error); if (td != NULL) p = td->td_proc; else p = NULL; switch (error) { case NFSERR_BADOWNER: tprintf(p, LOG_INFO, "No name and/or group mapping for uid,gid:(%d,%d)\n", uid, gid); return (EPERM); case NFSERR_BADNAME: case NFSERR_BADCHAR: printf("nfsv4 char/name not handled by server\n"); return (ENOENT); case NFSERR_STALECLIENTID: case NFSERR_STALESTATEID: case NFSERR_EXPIRED: case NFSERR_BADSTATEID: case NFSERR_BADSESSION: printf("nfsv4 recover err returned %d\n", error); return (EIO); case NFSERR_BADHANDLE: case NFSERR_SERVERFAULT: case NFSERR_BADTYPE: case NFSERR_FHEXPIRED: case NFSERR_RESOURCE: case NFSERR_MOVED: case NFSERR_NOFILEHANDLE: case NFSERR_MINORVERMISMATCH: case NFSERR_OLDSTATEID: case NFSERR_BADSEQID: case NFSERR_LEASEMOVED: case NFSERR_RECLAIMBAD: case NFSERR_BADXDR: case NFSERR_OPILLEGAL: printf("nfsv4 client/server protocol prob err=%d\n", error); return (EIO); default: tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error); return (EIO); }; } /* * Check to see if the process for this owner exists. Return 1 if it doesn't * and 0 otherwise. */ int nfscl_procdoesntexist(u_int8_t *own) { union { u_int32_t lval; u_int8_t cval[4]; } tl; struct proc *p; pid_t pid; int i, ret = 0; /* For the single open_owner of all 0 bytes, just return 0. */ for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++) if (own[i] != 0) break; if (i == NFSV4CL_LOCKNAMELEN) return (0); tl.cval[0] = *own++; tl.cval[1] = *own++; tl.cval[2] = *own++; tl.cval[3] = *own++; pid = tl.lval; p = pfind_any_locked(pid); if (p == NULL) return (1); if (p->p_stats == NULL) { PROC_UNLOCK(p); return (0); } tl.cval[0] = *own++; tl.cval[1] = *own++; tl.cval[2] = *own++; tl.cval[3] = *own++; if (tl.lval != p->p_stats->p_start.tv_sec) { ret = 1; } else { tl.cval[0] = *own++; tl.cval[1] = *own++; tl.cval[2] = *own++; tl.cval[3] = *own; if (tl.lval != p->p_stats->p_start.tv_usec) ret = 1; } PROC_UNLOCK(p); return (ret); } /* * - nfs pseudo system call for the client */ /* * MPSAFE */ static int nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap) { struct file *fp; struct nfscbd_args nfscbdarg; struct nfsd_nfscbd_args nfscbdarg2; struct nameidata nd; struct nfscl_dumpmntopts dumpmntopts; cap_rights_t rights; char *buf; int error; struct mount *mp; struct nfsmount *nmp; if (uap->flag & NFSSVC_CBADDSOCK) { error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg)); if (error) return (error); /* * Since we don't know what rights might be required, * pretend that we need them all. It is better to be too * careful than too reckless. */ error = fget(td, nfscbdarg.sock, - cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp); + cap_rights_init_one(&rights, CAP_SOCK_CLIENT), &fp); if (error) return (error); if (fp->f_type != DTYPE_SOCKET) { fdrop(fp, td); return (EPERM); } error = nfscbd_addsock(fp); fdrop(fp, td); if (!error && nfscl_enablecallb == 0) { nfsv4_cbport = nfscbdarg.port; nfscl_enablecallb = 1; } } else if (uap->flag & NFSSVC_NFSCBD) { if (uap->argp == NULL) return (EINVAL); error = copyin(uap->argp, (caddr_t)&nfscbdarg2, sizeof(nfscbdarg2)); if (error) return (error); error = nfscbd_nfsd(td, &nfscbdarg2); } else if (uap->flag & NFSSVC_DUMPMNTOPTS) { error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts)); if (error == 0 && (dumpmntopts.ndmnt_blen < 256 || dumpmntopts.ndmnt_blen > 1024)) error = EINVAL; if (error == 0) error = nfsrv_lookupfilename(&nd, dumpmntopts.ndmnt_fname, td); if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { vput(nd.ni_vp); error = EINVAL; } if (error == 0) { buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK); nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf, dumpmntopts.ndmnt_blen); vput(nd.ni_vp); error = copyout(buf, dumpmntopts.ndmnt_buf, dumpmntopts.ndmnt_blen); free(buf, M_TEMP); } } else if (uap->flag & NFSSVC_FORCEDISM) { buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK); error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL); if (error == 0) { nmp = NULL; mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (strcmp(mp->mnt_stat.f_mntonname, buf) == 0 && strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 && mp->mnt_data != NULL) { nmp = VFSTONFS(mp); NFSDDSLOCK(); if (nfsv4_findmirror(nmp) != NULL) { NFSDDSUNLOCK(); error = ENXIO; nmp = NULL; break; } mtx_lock(&nmp->nm_mtx); if ((nmp->nm_privflag & NFSMNTP_FORCEDISM) == 0) { nmp->nm_privflag |= (NFSMNTP_FORCEDISM | NFSMNTP_CANCELRPCS); mtx_unlock(&nmp->nm_mtx); } else { mtx_unlock(&nmp->nm_mtx); nmp = NULL; } NFSDDSUNLOCK(); break; } } mtx_unlock(&mountlist_mtx); if (nmp != NULL) { /* * Call newnfs_nmcancelreqs() to cause * any RPCs in progress on the mount point to * fail. * This will cause any process waiting for an * RPC to complete while holding a vnode lock * on the mounted-on vnode (such as "df" or * a non-forced "umount") to fail. * This will unlock the mounted-on vnode so * a forced dismount can succeed. * Then clear NFSMNTP_CANCELRPCS and wakeup(), * so that nfs_unmount() can complete. */ newnfs_nmcancelreqs(nmp); mtx_lock(&nmp->nm_mtx); nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; wakeup(nmp); mtx_unlock(&nmp->nm_mtx); } else if (error == 0) error = EINVAL; } free(buf, M_TEMP); } else { error = EINVAL; } return (error); } extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *); /* * Called once to initialize data structures... */ static int nfscl_modevent(module_t mod, int type, void *data) { int error = 0; static int loaded = 0; switch (type) { case MOD_LOAD: if (loaded) return (0); newnfs_portinit(); mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF); nfscl_init(); NFSD_LOCK(); nfsrvd_cbinit(0); NFSD_UNLOCK(); ncl_call_invalcaches = ncl_invalcaches; nfsd_call_nfscl = nfssvc_nfscl; loaded = 1; break; case MOD_UNLOAD: if (nfs_numnfscbd != 0) { error = EBUSY; break; } /* * XXX: Unloading of nfscl module is unsupported. */ #if 0 ncl_call_invalcaches = NULL; nfsd_call_nfscl = NULL; uma_zdestroy(ncl_pbuf_zone); /* and get rid of the mutexes */ mtx_destroy(&ncl_iod_mutex); loaded = 0; break; #else /* FALLTHROUGH */ #endif default: error = EOPNOTSUPP; break; } return error; } static moduledata_t nfscl_mod = { "nfscl", nfscl_modevent, NULL, }; DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST); /* So that loader and kldload(2) can find us, wherever we are.. */ MODULE_VERSION(nfscl, 1); MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1); MODULE_DEPEND(nfscl, krpc, 1, 1, 1); MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1); diff --git a/sys/fs/nfsserver/nfs_nfsdport.c b/sys/fs/nfsserver/nfs_nfsdport.c index 8336a0b8fab2..b781503a6815 100644 --- a/sys/fs/nfsserver/nfs_nfsdport.c +++ b/sys/fs/nfsserver/nfs_nfsdport.c @@ -1,6723 +1,6723 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * 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. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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 /* * Functions that perform the vfs operations required by the routines in * nfsd_serv.c. It is hoped that this change will make the server more * portable. */ #include #include #include #include #include #include #include FEATURE(nfsd, "NFSv4 server"); extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; extern int nfsrv_useacl; extern int newnfs_numnfsd; extern struct mount nfsv4root_mnt; extern struct nfsrv_stablefirst nfsrv_stablefirst; extern void (*nfsd_call_servertimer)(void); extern SVCPOOL *nfsrvd_pool; extern struct nfsv4lock nfsd_suspend_lock; extern struct nfsclienthashhead *nfsclienthash; extern struct nfslockhashhead *nfslockhash; extern struct nfssessionhash *nfssessionhash; extern int nfsrv_sessionhashsize; extern struct nfsstatsv1 nfsstatsv1; extern struct nfslayouthash *nfslayouthash; extern int nfsrv_layouthashsize; extern struct mtx nfsrv_dslock_mtx; extern int nfs_pnfsiothreads; extern struct nfsdontlisthead nfsrv_dontlisthead; extern volatile int nfsrv_dontlistlen; extern volatile int nfsrv_devidcnt; extern int nfsrv_maxpnfsmirror; struct vfsoptlist nfsv4root_opt, nfsv4root_newopt; NFSDLOCKMUTEX; NFSSTATESPINLOCK; struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE]; struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE]; struct mtx nfsrc_udpmtx; struct mtx nfs_v4root_mutex; struct mtx nfsrv_dontlistlock_mtx; struct mtx nfsrv_recalllock_mtx; struct nfsrvfh nfs_rootfh, nfs_pubfh; int nfs_pubfhset = 0, nfs_rootfhset = 0; struct proc *nfsd_master_proc = NULL; int nfsd_debuglevel = 0; static pid_t nfsd_master_pid = (pid_t)-1; static char nfsd_master_comm[MAXCOMLEN + 1]; static struct timeval nfsd_master_start; static uint32_t nfsv4_sysid = 0; static fhandle_t zerofh; static int nfssvc_srvcall(struct thread *, struct nfssvc_args *, struct ucred *); int nfsrv_enable_crossmntpt = 1; static int nfs_commit_blks; static int nfs_commit_miss; extern int nfsrv_issuedelegs; extern int nfsrv_dolocallocks; extern int nfsd_enable_stringtouid; extern struct nfsdevicehead nfsrv_devidhead; static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **, struct iovec **); static int nfsrv_createiovec_extpgs(int, int, struct mbuf **, struct mbuf **, struct iovec **); static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **, int *); static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *, NFSPROC_T *); static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **, int *, char *, fhandle_t *); static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *, NFSPROC_T *); static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *, struct thread *, int, struct mbuf **, char *, struct mbuf **, struct nfsvattr *, struct acl *, off_t *, int, bool *); static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *); static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *, NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **); static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **, char *, int *); static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct acl *, int *); static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *); static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount *, struct nfsvattr *); static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *, NFSPROC_T *, struct nfsmount *); static int nfsrv_putfhname(fhandle_t *, char *); static int nfsrv_pnfslookupds(struct vnode *, struct vnode *, struct pnfsdsfile *, struct vnode **, NFSPROC_T *); static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *, struct vnode *, NFSPROC_T *); static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *); static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *, NFSPROC_T *); static int nfsrv_pnfsstatfs(struct statfs *, struct mount *); int nfs_pnfsio(task_fn_t *, void *); SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "NFS server"); SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW, &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points"); SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 0, ""); SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 0, ""); SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW, &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations"); SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW, &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files"); SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel, 0, "Debug level for NFS server"); SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW, &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names"); static int nfsrv_pnfsgetdsattr = 1; SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW, &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC"); /* * nfsrv_dsdirsize can only be increased and only when the nfsd threads are * not running. * The dsN subdirectories for the increased values must have been created * on all DS servers before this increase is done. */ u_int nfsrv_dsdirsize = 20; static int sysctl_dsdirsize(SYSCTL_HANDLER_ARGS) { int error, newdsdirsize; newdsdirsize = nfsrv_dsdirsize; error = sysctl_handle_int(oidp, &newdsdirsize, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 || newnfs_numnfsd != 0) return (EINVAL); nfsrv_dsdirsize = newdsdirsize; return (0); } SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize, CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize), sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers"); #define MAX_REORDERED_RPC 16 #define NUM_HEURISTIC 1031 #define NHUSE_INIT 64 #define NHUSE_INC 16 #define NHUSE_MAX 2048 static struct nfsheur { struct vnode *nh_vp; /* vp to match (unreferenced pointer) */ off_t nh_nextoff; /* next offset for sequential detection */ int nh_use; /* use count for selection */ int nh_seqcount; /* heuristic */ } nfsheur[NUM_HEURISTIC]; /* * Heuristic to detect sequential operation. */ static struct nfsheur * nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp) { struct nfsheur *nh; int hi, try; /* Locate best candidate. */ try = 32; hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC; nh = &nfsheur[hi]; while (try--) { if (nfsheur[hi].nh_vp == vp) { nh = &nfsheur[hi]; break; } if (nfsheur[hi].nh_use > 0) --nfsheur[hi].nh_use; hi = (hi + 1) % NUM_HEURISTIC; if (nfsheur[hi].nh_use < nh->nh_use) nh = &nfsheur[hi]; } /* Initialize hint if this is a new file. */ if (nh->nh_vp != vp) { nh->nh_vp = vp; nh->nh_nextoff = uio->uio_offset; nh->nh_use = NHUSE_INIT; if (uio->uio_offset == 0) nh->nh_seqcount = 4; else nh->nh_seqcount = 1; } /* Calculate heuristic. */ if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) || uio->uio_offset == nh->nh_nextoff) { /* See comments in vfs_vnops.c:sequential_heuristic(). */ nh->nh_seqcount += howmany(uio->uio_resid, 16384); if (nh->nh_seqcount > IO_SEQMAX) nh->nh_seqcount = IO_SEQMAX; } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC * imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) { /* Probably a reordered RPC, leave seqcount alone. */ } else if (nh->nh_seqcount > 1) { nh->nh_seqcount /= 2; } else { nh->nh_seqcount = 0; } nh->nh_use += NHUSE_INC; if (nh->nh_use > NHUSE_MAX) nh->nh_use = NHUSE_MAX; return (nh); } /* * Get attributes into nfsvattr structure. */ int nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct nfsrv_descript *nd, struct thread *p, int vpislocked, nfsattrbit_t *attrbitp) { int error, gotattr, lockedit = 0; struct nfsvattr na; if (vpislocked == 0) { /* * When vpislocked == 0, the vnode is either exclusively * locked by this thread or not locked by this thread. * As such, shared lock it, if not exclusively locked. */ if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { lockedit = 1; NFSVOPLOCK(vp, LK_SHARED | LK_RETRY); } } /* * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed * attributes, as required. * This needs to be done for regular files if: * - non-NFSv4 RPCs or * - when attrbitp == NULL or * - an NFSv4 RPC with any of the above attributes in attrbitp. * A return of 0 for nfsrv_proxyds() indicates that it has acquired * these attributes. nfsrv_proxyds() will return an error if the * server is not a pNFS one. */ gotattr = 0; if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL || (nd->nd_flag & ND_NFSV4) == 0 || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) { error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p, NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0, NULL); if (error == 0) gotattr = 1; } error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred); if (lockedit != 0) NFSVOPUNLOCK(vp); /* * If we got the Change, Size and Modify Time from the DS, * replace them. */ if (gotattr != 0) { nvap->na_atime = na.na_atime; nvap->na_mtime = na.na_mtime; nvap->na_filerev = na.na_filerev; nvap->na_size = na.na_size; nvap->na_bytes = na.na_bytes; } NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr, error, (uintmax_t)na.na_filerev); NFSEXITCODE(error); return (error); } /* * Get a file handle for a vnode. */ int nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p) { int error; NFSBZERO((caddr_t)fhp, sizeof(fhandle_t)); fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid; error = VOP_VPTOFH(vp, &fhp->fh_fid); NFSEXITCODE(error); return (error); } /* * Perform access checking for vnodes obtained from file handles that would * refer to files already opened by a Unix client. You cannot just use * vn_writechk() and VOP_ACCESSX() for two reasons. * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write * case. * 2 - The owner is to be given access irrespective of mode bits for some * operations, so that processes that chmod after opening a file don't * break. */ int nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred, struct nfsexstuff *exp, struct thread *p, int override, int vpislocked, u_int32_t *supportedtypep) { struct vattr vattr; int error = 0, getret = 0; if (vpislocked == 0) { if (NFSVOPLOCK(vp, LK_SHARED) != 0) { error = EPERM; goto out; } } if (accmode & VWRITE) { /* Just vn_writechk() changed to check rdonly */ /* * Disallow write attempts on read-only file systems; * unless the file is a socket or a block or character * device resident on the file system. */ if (NFSVNO_EXRDONLY(exp) || (vp->v_mount->mnt_flag & MNT_RDONLY)) { switch (vp->v_type) { case VREG: case VDIR: case VLNK: error = EROFS; default: break; } } /* * If there's shared text associated with * the inode, try to free it up once. If * we fail, we can't allow writing. */ if (VOP_IS_TEXT(vp) && error == 0) error = ETXTBSY; } if (error != 0) { if (vpislocked == 0) NFSVOPUNLOCK(vp); goto out; } /* * Should the override still be applied when ACLs are enabled? */ error = VOP_ACCESSX(vp, accmode, cred, p); if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) { /* * Try again with VEXPLICIT_DENY, to see if the test for * deletion is supported. */ error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p); if (error == 0) { if (vp->v_type == VDIR) { accmode &= ~(VDELETE | VDELETE_CHILD); accmode |= VWRITE; error = VOP_ACCESSX(vp, accmode, cred, p); } else if (supportedtypep != NULL) { *supportedtypep &= ~NFSACCESS_DELETE; } } } /* * Allow certain operations for the owner (reads and writes * on files that are already open). */ if (override != NFSACCCHK_NOOVERRIDE && (error == EPERM || error == EACCES)) { if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT)) error = 0; else if (override & NFSACCCHK_ALLOWOWNER) { getret = VOP_GETATTR(vp, &vattr, cred); if (getret == 0 && cred->cr_uid == vattr.va_uid) error = 0; } } if (vpislocked == 0) NFSVOPUNLOCK(vp); out: NFSEXITCODE(error); return (error); } /* * Set attribute(s) vnop. */ int nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { u_quad_t savsize = 0; int error, savedit; time_t savbtime; /* * If this is an exported file system and a pNFS service is running, * don't VOP_SETATTR() of size for the MDS file system. */ savedit = 0; error = 0; if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 && nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL && nvap->na_vattr.va_size > 0) { savsize = nvap->na_vattr.va_size; nvap->na_vattr.va_size = VNOVAL; if (nvap->na_vattr.va_uid != (uid_t)VNOVAL || nvap->na_vattr.va_gid != (gid_t)VNOVAL || nvap->na_vattr.va_mode != (mode_t)VNOVAL || nvap->na_vattr.va_atime.tv_sec != VNOVAL || nvap->na_vattr.va_mtime.tv_sec != VNOVAL) savedit = 1; else savedit = 2; } if (savedit != 2) error = VOP_SETATTR(vp, &nvap->na_vattr, cred); if (savedit != 0) nvap->na_vattr.va_size = savsize; if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL || nvap->na_vattr.va_gid != (gid_t)VNOVAL || nvap->na_vattr.va_size != VNOVAL || nvap->na_vattr.va_mode != (mode_t)VNOVAL || nvap->na_vattr.va_atime.tv_sec != VNOVAL || nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) { /* Never modify birthtime on a DS file. */ savbtime = nvap->na_vattr.va_birthtime.tv_sec; nvap->na_vattr.va_birthtime.tv_sec = VNOVAL; /* For a pNFS server, set the attributes on the DS file. */ error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR, NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL); nvap->na_vattr.va_birthtime.tv_sec = savbtime; if (error == ENOENT) error = 0; } NFSEXITCODE(error); return (error); } /* * Set up nameidata for a lookup() call and do it. */ int nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp, struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p, struct vnode **retdirp) { struct componentname *cnp = &ndp->ni_cnd; int i; struct iovec aiov; struct uio auio; int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen; int error = 0; char *cp; *retdirp = NULL; cnp->cn_nameptr = cnp->cn_pnbuf; ndp->ni_lcf = 0; /* * Extract and set starting directory. */ if (dp->v_type != VDIR) { if (islocked) vput(dp); else vrele(dp); nfsvno_relpathbuf(ndp); error = ENOTDIR; goto out1; } if (islocked) NFSVOPUNLOCK(dp); VREF(dp); *retdirp = dp; if (NFSVNO_EXRDONLY(exp)) cnp->cn_flags |= RDONLY; ndp->ni_segflg = UIO_SYSSPACE; if (nd->nd_flag & ND_PUBLOOKUP) { ndp->ni_loopcnt = 0; if (cnp->cn_pnbuf[0] == '/') { vrele(dp); /* * Check for degenerate pathnames here, since lookup() * panics on them. */ for (i = 1; i < ndp->ni_pathlen; i++) if (cnp->cn_pnbuf[i] != '/') break; if (i == ndp->ni_pathlen) { error = NFSERR_ACCES; goto out; } dp = rootvnode; VREF(dp); } } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) || (nd->nd_flag & ND_NFSV4) == 0) { /* * Only cross mount points for NFSv4 when doing a * mount while traversing the file system above * the mount point, unless nfsrv_enable_crossmntpt is set. */ cnp->cn_flags |= NOCROSSMOUNT; } /* * Initialize for scan, set ni_startdir and bump ref on dp again * because lookup() will dereference ni_startdir. */ cnp->cn_thread = p; ndp->ni_startdir = dp; ndp->ni_rootdir = rootvnode; ndp->ni_topdir = NULL; if (!lockleaf) cnp->cn_flags |= LOCKLEAF; for (;;) { cnp->cn_nameptr = cnp->cn_pnbuf; /* * Call lookup() to do the real work. If an error occurs, * ndp->ni_vp and ni_dvp are left uninitialized or NULL and * we do not have to dereference anything before returning. * In either case ni_startdir will be dereferenced and NULLed * out. */ error = lookup(ndp); if (error) break; /* * Check for encountering a symbolic link. Trivial * termination occurs if no symlink encountered. */ if ((cnp->cn_flags & ISSYMLINK) == 0) { if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) nfsvno_relpathbuf(ndp); if (ndp->ni_vp && !lockleaf) NFSVOPUNLOCK(ndp->ni_vp); break; } /* * Validate symlink */ if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1) NFSVOPUNLOCK(ndp->ni_dvp); if (!(nd->nd_flag & ND_PUBLOOKUP)) { error = EINVAL; goto badlink2; } if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { error = ELOOP; goto badlink2; } if (ndp->ni_pathlen > 1) cp = uma_zalloc(namei_zone, M_WAITOK); else cp = cnp->cn_pnbuf; aiov.iov_base = cp; aiov.iov_len = MAXPATHLEN; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = NULL; auio.uio_resid = MAXPATHLEN; error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); if (error) { badlink1: if (ndp->ni_pathlen > 1) uma_zfree(namei_zone, cp); badlink2: vrele(ndp->ni_dvp); vput(ndp->ni_vp); break; } linklen = MAXPATHLEN - auio.uio_resid; if (linklen == 0) { error = ENOENT; goto badlink1; } if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { error = ENAMETOOLONG; goto badlink1; } /* * Adjust or replace path */ if (ndp->ni_pathlen > 1) { NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen); uma_zfree(namei_zone, cnp->cn_pnbuf); cnp->cn_pnbuf = cp; } else cnp->cn_pnbuf[linklen] = '\0'; ndp->ni_pathlen += linklen; /* * Cleanup refs for next loop and check if root directory * should replace current directory. Normally ni_dvp * becomes the new base directory and is cleaned up when * we loop. Explicitly null pointers after invalidation * to clarify operation. */ vput(ndp->ni_vp); ndp->ni_vp = NULL; if (cnp->cn_pnbuf[0] == '/') { vrele(ndp->ni_dvp); ndp->ni_dvp = ndp->ni_rootdir; VREF(ndp->ni_dvp); } ndp->ni_startdir = ndp->ni_dvp; ndp->ni_dvp = NULL; } if (!lockleaf) cnp->cn_flags &= ~LOCKLEAF; out: if (error) { nfsvno_relpathbuf(ndp); ndp->ni_vp = NULL; ndp->ni_dvp = NULL; ndp->ni_startdir = NULL; } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) { ndp->ni_dvp = NULL; } out1: NFSEXITCODE2(error, nd); return (error); } /* * Set up a pathname buffer and return a pointer to it and, optionally * set a hash pointer. */ void nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp) { struct componentname *cnp = &ndp->ni_cnd; cnp->cn_flags |= (NOMACCHECK | HASBUF); cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); if (hashpp != NULL) *hashpp = NULL; *bufpp = cnp->cn_pnbuf; } /* * Release the above path buffer, if not released by nfsvno_namei(). */ void nfsvno_relpathbuf(struct nameidata *ndp) { if ((ndp->ni_cnd.cn_flags & HASBUF) == 0) panic("nfsrelpath"); uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); ndp->ni_cnd.cn_flags &= ~HASBUF; } /* * Readlink vnode op into an mbuf list. */ int nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz, struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp) { struct iovec *iv; struct uio io, *uiop = &io; struct mbuf *mp, *mp3; int len, tlen, error = 0; len = NFS_MAXPATHLEN; if (maxextsiz > 0) uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, &mp3, &mp, &iv); else uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv); uiop->uio_iov = iv; uiop->uio_offset = 0; uiop->uio_resid = len; uiop->uio_rw = UIO_READ; uiop->uio_segflg = UIO_SYSSPACE; uiop->uio_td = NULL; error = VOP_READLINK(vp, uiop, cred); free(iv, M_TEMP); if (error) { m_freem(mp3); *lenp = 0; goto out; } if (uiop->uio_resid > 0) { len -= uiop->uio_resid; tlen = NFSM_RNDUP(len); if (tlen == 0) { m_freem(mp3); mp3 = mp = NULL; } else if (tlen != NFS_MAXPATHLEN || tlen != len) mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len); } *lenp = len; *mpp = mp3; *mpendp = mp; out: NFSEXITCODE(error); return (error); } /* * Create an mbuf chain and an associated iovec that can be used to Read * or Getextattr of data. * Upon success, return pointers to the first and last mbufs in the chain * plus the malloc'd iovec and its iovlen. */ static int nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp, struct iovec **ivp) { struct mbuf *m, *m2 = NULL, *m3; struct iovec *iv; int i, left, siz; left = len; m3 = NULL; /* * Generate the mbuf list with the uio_iov ref. to it. */ i = 0; while (left > 0) { NFSMGET(m); MCLGET(m, M_WAITOK); m->m_len = 0; siz = min(M_TRAILINGSPACE(m), left); left -= siz; i++; if (m3) m2->m_next = m; else m3 = m; m2 = m; } *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); m = m3; left = len; i = 0; while (left > 0) { if (m == NULL) panic("nfsrv_createiovec iov"); siz = min(M_TRAILINGSPACE(m), left); if (siz > 0) { iv->iov_base = mtod(m, caddr_t) + m->m_len; iv->iov_len = siz; m->m_len += siz; left -= siz; iv++; i++; } m = m->m_next; } *mpp = m3; *mpendp = m2; return (i); } /* * Create an mbuf chain and an associated iovec that can be used to Read * or Getextattr of data. * Upon success, return pointers to the first and last mbufs in the chain * plus the malloc'd iovec and its iovlen. * Same as above, but creates ext_pgs mbuf(s). */ static int nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp, struct mbuf **mpendp, struct iovec **ivp) { struct mbuf *m, *m2 = NULL, *m3; struct iovec *iv; int i, left, pgno, siz; left = len; m3 = NULL; /* * Generate the mbuf list with the uio_iov ref. to it. */ i = 0; while (left > 0) { siz = min(left, maxextsiz); m = mb_alloc_ext_plus_pages(siz, M_WAITOK); left -= siz; i += m->m_epg_npgs; if (m3 != NULL) m2->m_next = m; else m3 = m; m2 = m; } *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); m = m3; left = len; i = 0; pgno = 0; while (left > 0) { if (m == NULL) panic("nfsvno_createiovec_extpgs iov"); siz = min(PAGE_SIZE, left); if (siz > 0) { iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]); iv->iov_len = siz; m->m_len += siz; if (pgno == m->m_epg_npgs - 1) m->m_epg_last_len = siz; left -= siz; iv++; i++; pgno++; } if (pgno == m->m_epg_npgs && left > 0) { m = m->m_next; if (m == NULL) panic("nfsvno_createiovec_extpgs iov"); pgno = 0; } } *mpp = m3; *mpendp = m2; return (i); } /* * Read vnode op call into mbuf list. */ int nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred, int maxextsiz, struct thread *p, struct mbuf **mpp, struct mbuf **mpendp) { struct mbuf *m; struct iovec *iv; int error = 0, len, tlen, ioflag = 0; struct mbuf *m3; struct uio io, *uiop = &io; struct nfsheur *nh; /* * Attempt to read from a DS file. A return of ENOENT implies * there is no DS file to read. */ error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp, NULL, mpendp, NULL, NULL, NULL, 0, NULL); if (error != ENOENT) return (error); len = NFSM_RNDUP(cnt); if (maxextsiz > 0) uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, &m3, &m, &iv); else uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv); uiop->uio_iov = iv; uiop->uio_offset = off; uiop->uio_resid = len; uiop->uio_rw = UIO_READ; uiop->uio_segflg = UIO_SYSSPACE; uiop->uio_td = NULL; nh = nfsrv_sequential_heuristic(uiop, vp); ioflag |= nh->nh_seqcount << IO_SEQSHIFT; /* XXX KDM make this more systematic? */ nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid; error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred); free(iv, M_TEMP); if (error) { m_freem(m3); *mpp = NULL; goto out; } nh->nh_nextoff = uiop->uio_offset; tlen = len - uiop->uio_resid; cnt = cnt < tlen ? cnt : tlen; tlen = NFSM_RNDUP(cnt); if (tlen == 0) { m_freem(m3); m3 = m = NULL; } else if (len != tlen || tlen != cnt) m = nfsrv_adj(m3, len - tlen, tlen - cnt); *mpp = m3; *mpendp = m; out: NFSEXITCODE(error); return (error); } /* * Create the iovec for the mbuf chain passed in as an argument. * The "cp" argument is where the data starts within the first mbuf in * the chain. It returns the iovec and the iovcnt. */ static int nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp, int *iovcntp) { struct mbuf *mp; struct iovec *ivp; int cnt, i, len; /* * Loop through the mbuf chain, counting how many mbufs are a * part of this write operation, so the iovec size is known. */ cnt = 0; len = retlen; mp = m; i = mtod(mp, caddr_t) + mp->m_len - cp; while (len > 0) { if (i > 0) { len -= i; cnt++; } mp = mp->m_next; if (!mp) { if (len > 0) return (EBADRPC); } else i = mp->m_len; } /* Now, create the iovec. */ mp = m; *ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK); *iovcntp = cnt; i = mtod(mp, caddr_t) + mp->m_len - cp; len = retlen; while (len > 0) { if (mp == NULL) panic("nfsrv_createiovecw"); if (i > 0) { i = min(i, len); ivp->iov_base = cp; ivp->iov_len = i; ivp++; len -= i; } mp = mp->m_next; if (mp) { i = mp->m_len; cp = mtod(mp, caddr_t); } } return (0); } /* * Write vnode op from an mbuf list. */ int nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable, struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p) { struct iovec *iv; int cnt, ioflags, error; struct uio io, *uiop = &io; struct nfsheur *nh; /* * Attempt to write to a DS file. A return of ENOENT implies * there is no DS file to write. */ error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS, &mp, cp, NULL, NULL, NULL, NULL, 0, NULL); if (error != ENOENT) { *stable = NFSWRITE_FILESYNC; return (error); } if (*stable == NFSWRITE_UNSTABLE) ioflags = IO_NODELOCKED; else ioflags = (IO_SYNC | IO_NODELOCKED); error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt); if (error != 0) return (error); uiop->uio_iov = iv; uiop->uio_iovcnt = cnt; uiop->uio_resid = retlen; uiop->uio_rw = UIO_WRITE; uiop->uio_segflg = UIO_SYSSPACE; NFSUIOPROC(uiop, p); uiop->uio_offset = off; nh = nfsrv_sequential_heuristic(uiop, vp); ioflags |= nh->nh_seqcount << IO_SEQSHIFT; /* XXX KDM make this more systematic? */ nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid; error = VOP_WRITE(vp, uiop, ioflags, cred); if (error == 0) nh->nh_nextoff = uiop->uio_offset; free(iv, M_TEMP); NFSEXITCODE(error); return (error); } /* * Common code for creating a regular file (plus special files for V2). */ int nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp, struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp, int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp) { u_quad_t tempsize; int error; struct thread *p = curthread; error = nd->nd_repstat; if (!error && ndp->ni_vp == NULL) { if (nvap->na_type == VREG || nvap->na_type == VSOCK) { vrele(ndp->ni_startdir); error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); /* For a pNFS server, create the data file on a DS. */ if (error == 0 && nvap->na_type == VREG) { /* * Create a data file on a DS for a pNFS server. * This function just returns if not * running a pNFS DS or the creation fails. */ nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, nd->nd_cred, p); } vput(ndp->ni_dvp); nfsvno_relpathbuf(ndp); if (!error) { if (*exclusive_flagp) { *exclusive_flagp = 0; NFSVNO_ATTRINIT(nvap); nvap->na_atime.tv_sec = cverf[0]; nvap->na_atime.tv_nsec = cverf[1]; error = VOP_SETATTR(ndp->ni_vp, &nvap->na_vattr, nd->nd_cred); if (error != 0) { vput(ndp->ni_vp); ndp->ni_vp = NULL; error = NFSERR_NOTSUPP; } } } /* * NFS V2 Only. nfsrvd_mknod() does this for V3. * (This implies, just get out on an error.) */ } else if (nvap->na_type == VCHR || nvap->na_type == VBLK || nvap->na_type == VFIFO) { if (nvap->na_type == VCHR && rdev == 0xffffffff) nvap->na_type = VFIFO; if (nvap->na_type != VFIFO && (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); goto out; } nvap->na_rdev = rdev; error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); vput(ndp->ni_dvp); nfsvno_relpathbuf(ndp); vrele(ndp->ni_startdir); if (error) goto out; } else { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); error = ENXIO; goto out; } *vpp = ndp->ni_vp; } else { /* * Handle cases where error is already set and/or * the file exists. * 1 - clean up the lookup * 2 - iff !error and na_size set, truncate it */ vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); *vpp = ndp->ni_vp; if (ndp->ni_dvp == *vpp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); if (!error && nvap->na_size != VNOVAL) { error = nfsvno_accchk(*vpp, VWRITE, nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, NFSACCCHK_VPISLOCKED, NULL); if (!error) { tempsize = nvap->na_size; NFSVNO_ATTRINIT(nvap); nvap->na_size = tempsize; error = VOP_SETATTR(*vpp, &nvap->na_vattr, nd->nd_cred); } } if (error) vput(*vpp); } out: NFSEXITCODE(error); return (error); } /* * Do a mknod vnode op. */ int nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred, struct thread *p) { int error = 0; enum vtype vtyp; vtyp = nvap->na_type; /* * Iff doesn't exist, create it. */ if (ndp->ni_vp) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); vrele(ndp->ni_vp); error = EEXIST; goto out; } if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); error = NFSERR_BADTYPE; goto out; } if (vtyp == VSOCK) { vrele(ndp->ni_startdir); error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); vput(ndp->ni_dvp); nfsvno_relpathbuf(ndp); } else { if (nvap->na_type != VFIFO && (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); goto out; } error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); vput(ndp->ni_dvp); nfsvno_relpathbuf(ndp); vrele(ndp->ni_startdir); /* * Since VOP_MKNOD returns the ni_vp, I can't * see any reason to do the lookup. */ } out: NFSEXITCODE(error); return (error); } /* * Mkdir vnode op. */ int nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { int error = 0; if (ndp->ni_vp != NULL) { if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); vrele(ndp->ni_vp); nfsvno_relpathbuf(ndp); error = EEXIST; goto out; } error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); vput(ndp->ni_dvp); nfsvno_relpathbuf(ndp); out: NFSEXITCODE(error); return (error); } /* * symlink vnode op. */ int nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp, int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { int error = 0; if (ndp->ni_vp) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); vrele(ndp->ni_vp); error = EEXIST; goto out; } error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr, pathcp); vput(ndp->ni_dvp); vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); /* * Although FreeBSD still had the lookup code in * it for 7/current, there doesn't seem to be any * point, since VOP_SYMLINK() returns the ni_vp. * Just vput it for v2. */ if (!not_v2 && !error) vput(ndp->ni_vp); out: NFSEXITCODE(error); return (error); } /* * Parse symbolic link arguments. * This function has an ugly side effect. It will malloc() an area for * the symlink and set iov_base to point to it, only if it succeeds. * So, if it returns with uiop->uio_iov->iov_base != NULL, that must * be FREE'd later. */ int nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap, struct thread *p, char **pathcpp, int *lenp) { u_int32_t *tl; char *pathcp = NULL; int error = 0, len; struct nfsv2_sattr *sp; *pathcpp = NULL; *lenp = 0; if ((nd->nd_flag & ND_NFSV3) && (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p))) goto nfsmout; NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); len = fxdr_unsigned(int, *tl); if (len > NFS_MAXPATHLEN || len <= 0) { error = EBADRPC; goto nfsmout; } pathcp = malloc(len + 1, M_TEMP, M_WAITOK); error = nfsrv_mtostr(nd, pathcp, len); if (error) goto nfsmout; if (nd->nd_flag & ND_NFSV2) { NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); } *pathcpp = pathcp; *lenp = len; NFSEXITCODE2(0, nd); return (0); nfsmout: if (pathcp) free(pathcp, M_TEMP); NFSEXITCODE2(error, nd); return (error); } /* * Remove a non-directory object. */ int nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS]; int error = 0, mirrorcnt; char fname[PNFS_FILENAME_LEN + 1]; fhandle_t fh; vp = ndp->ni_vp; dsdvp[0] = NULL; if (vp->v_type == VDIR) error = NFSERR_ISDIR; else if (is_v4) error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0), p); if (error == 0) nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh); if (!error) error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd); if (error == 0 && dsdvp[0] != NULL) nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); if (ndp->ni_dvp == vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); vput(vp); if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0) nfsvno_relpathbuf(ndp); NFSEXITCODE(error); return (error); } /* * Remove a directory. */ int nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { struct vnode *vp; int error = 0; vp = ndp->ni_vp; if (vp->v_type != VDIR) { error = ENOTDIR; goto out; } /* * No rmdir "." please. */ if (ndp->ni_dvp == vp) { error = EINVAL; goto out; } /* * The root of a mounted filesystem cannot be deleted. */ if (vp->v_vflag & VV_ROOT) error = EBUSY; out: if (!error) error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd); if (ndp->ni_dvp == vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); vput(vp); if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0) nfsvno_relpathbuf(ndp); NFSEXITCODE(error); return (error); } /* * Rename vnode op. */ int nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp, u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p) { struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS]; int error = 0, mirrorcnt; char fname[PNFS_FILENAME_LEN + 1]; fhandle_t fh; dsdvp[0] = NULL; fvp = fromndp->ni_vp; if (ndstat) { vrele(fromndp->ni_dvp); vrele(fvp); error = ndstat; goto out1; } tdvp = tondp->ni_dvp; tvp = tondp->ni_vp; if (tvp != NULL) { if (fvp->v_type == VDIR && tvp->v_type != VDIR) { error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST; goto out; } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST; goto out; } if (tvp->v_type == VDIR && tvp->v_mountedhere) { error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; goto out; } /* * A rename to '.' or '..' results in a prematurely * unlocked vnode on FreeBSD5, so I'm just going to fail that * here. */ if ((tondp->ni_cnd.cn_namelen == 1 && tondp->ni_cnd.cn_nameptr[0] == '.') || (tondp->ni_cnd.cn_namelen == 2 && tondp->ni_cnd.cn_nameptr[0] == '.' && tondp->ni_cnd.cn_nameptr[1] == '.')) { error = EINVAL; goto out; } } if (fvp->v_type == VDIR && fvp->v_mountedhere) { error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; goto out; } if (fvp->v_mount != tdvp->v_mount) { error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; goto out; } if (fvp == tdvp) { error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL; goto out; } if (fvp == tvp) { /* * If source and destination are the same, there is nothing to * do. Set error to -1 to indicate this. */ error = -1; goto out; } if (ndflag & ND_NFSV4) { if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) { error = nfsrv_checkremove(fvp, 0, NULL, (nfsquad_t)((u_quad_t)0), p); NFSVOPUNLOCK(fvp); } else error = EPERM; if (tvp && !error) error = nfsrv_checkremove(tvp, 1, NULL, (nfsquad_t)((u_quad_t)0), p); } else { /* * For NFSv2 and NFSv3, try to get rid of the delegation, so * that the NFSv4 client won't be confused by the rename. * Since nfsd_recalldelegation() can only be called on an * unlocked vnode at this point and fvp is the file that will * still exist after the rename, just do fvp. */ nfsd_recalldelegation(fvp, p); } if (error == 0 && tvp != NULL) { nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh); NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup" " dsdvp=%p\n", dsdvp[0]); } out: if (!error) { error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp, &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp, &tondp->ni_cnd); } else { if (tdvp == tvp) vrele(tdvp); else vput(tdvp); if (tvp) vput(tvp); vrele(fromndp->ni_dvp); vrele(fvp); if (error == -1) error = 0; } /* * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and * if the rename succeeded, the DS file for the tvp needs to be * removed. */ if (error == 0 && dsdvp[0] != NULL) { nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n"); } vrele(tondp->ni_startdir); nfsvno_relpathbuf(tondp); out1: vrele(fromndp->ni_startdir); nfsvno_relpathbuf(fromndp); NFSEXITCODE(error); return (error); } /* * Link vnode op. */ int nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { struct vnode *xp; int error = 0; xp = ndp->ni_vp; if (xp != NULL) { error = EEXIST; } else { xp = ndp->ni_dvp; if (vp->v_mount != xp->v_mount) error = EXDEV; } if (!error) { NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); if (!VN_IS_DOOMED(vp)) error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd); else error = EPERM; if (ndp->ni_dvp == vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); NFSVOPUNLOCK(vp); } else { if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); if (ndp->ni_vp) vrele(ndp->ni_vp); } nfsvno_relpathbuf(ndp); NFSEXITCODE(error); return (error); } /* * Do the fsync() appropriate for the commit. */ int nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred, struct thread *td) { int error = 0; /* * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of * file is done. At this time VOP_FSYNC does not accept offset and * byte count parameters so call VOP_FSYNC the whole file for now. * The same is true for NFSv4: RFC 3530 Sec. 14.2.3. * File systems that do not use the buffer cache (as indicated * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC(). */ if (cnt == 0 || cnt > MAX_COMMIT_COUNT || (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) { /* * Give up and do the whole thing */ if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC); VM_OBJECT_WUNLOCK(vp->v_object); } error = VOP_FSYNC(vp, MNT_WAIT, td); } else { /* * Locate and synchronously write any buffers that fall * into the requested range. Note: we are assuming that * f_iosize is a power of 2. */ int iosize = vp->v_mount->mnt_stat.f_iosize; int iomask = iosize - 1; struct bufobj *bo; daddr_t lblkno; /* * Align to iosize boundary, super-align to page boundary. */ if (off & iomask) { cnt += off & iomask; off &= ~(u_quad_t)iomask; } if (off & PAGE_MASK) { cnt += off & PAGE_MASK; off &= ~(u_quad_t)PAGE_MASK; } lblkno = off / iosize; if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, off, off + cnt, OBJPC_SYNC); VM_OBJECT_WUNLOCK(vp->v_object); } bo = &vp->v_bufobj; BO_LOCK(bo); while (cnt > 0) { struct buf *bp; /* * If we have a buffer and it is marked B_DELWRI we * have to lock and write it. Otherwise the prior * write is assumed to have already been committed. * * gbincore() can return invalid buffers now so we * have to check that bit as well (though B_DELWRI * should not be set if B_INVAL is set there could be * a race here since we haven't locked the buffer). */ if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) { if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { BO_LOCK(bo); continue; /* retry */ } if ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI) { bremfree(bp); bp->b_flags &= ~B_ASYNC; bwrite(bp); ++nfs_commit_miss; } else BUF_UNLOCK(bp); BO_LOCK(bo); } ++nfs_commit_blks; if (cnt < iosize) break; cnt -= iosize; ++lblkno; } BO_UNLOCK(bo); } NFSEXITCODE(error); return (error); } /* * Statfs vnode op. */ int nfsvno_statfs(struct vnode *vp, struct statfs *sf) { struct statfs *tsf; int error; tsf = NULL; if (nfsrv_devidcnt > 0) { /* For a pNFS service, get the DS numbers. */ tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO); error = nfsrv_pnfsstatfs(tsf, vp->v_mount); if (error != 0) { free(tsf, M_TEMP); tsf = NULL; } } error = VFS_STATFS(vp->v_mount, sf); if (error == 0) { if (tsf != NULL) { sf->f_blocks = tsf->f_blocks; sf->f_bavail = tsf->f_bavail; sf->f_bfree = tsf->f_bfree; sf->f_bsize = tsf->f_bsize; } /* * Since NFS handles these values as unsigned on the * wire, there is no way to represent negative values, * so set them to 0. Without this, they will appear * to be very large positive values for clients like * Solaris10. */ if (sf->f_bavail < 0) sf->f_bavail = 0; if (sf->f_ffree < 0) sf->f_ffree = 0; } free(tsf, M_TEMP); NFSEXITCODE(error); return (error); } /* * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but * must handle nfsrv_opencheck() calls after any other access checks. */ void nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp, nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp, int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create, NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct nfsexstuff *exp, struct vnode **vpp) { struct vnode *vp = NULL; u_quad_t tempsize; struct nfsexstuff nes; struct thread *p = curthread; if (ndp->ni_vp == NULL) nd->nd_repstat = nfsrv_opencheck(clientid, stateidp, stp, NULL, nd, p, nd->nd_repstat); if (!nd->nd_repstat) { if (ndp->ni_vp == NULL) { vrele(ndp->ni_startdir); nd->nd_repstat = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); /* For a pNFS server, create the data file on a DS. */ if (nd->nd_repstat == 0) { /* * Create a data file on a DS for a pNFS server. * This function just returns if not * running a pNFS DS or the creation fails. */ nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, cred, p); } vput(ndp->ni_dvp); nfsvno_relpathbuf(ndp); if (!nd->nd_repstat) { if (*exclusive_flagp) { *exclusive_flagp = 0; NFSVNO_ATTRINIT(nvap); nvap->na_atime.tv_sec = cverf[0]; nvap->na_atime.tv_nsec = cverf[1]; nd->nd_repstat = VOP_SETATTR(ndp->ni_vp, &nvap->na_vattr, cred); if (nd->nd_repstat != 0) { vput(ndp->ni_vp); ndp->ni_vp = NULL; nd->nd_repstat = NFSERR_NOTSUPP; } else NFSSETBIT_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS); } else { nfsrv_fixattr(nd, ndp->ni_vp, nvap, aclp, p, attrbitp, exp); } } vp = ndp->ni_vp; } else { if (ndp->ni_startdir) vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vp = ndp->ni_vp; if (create == NFSV4OPEN_CREATE) { if (ndp->ni_dvp == vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); } if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) { if (ndp->ni_cnd.cn_flags & RDONLY) NFSVNO_SETEXRDONLY(&nes); else NFSVNO_EXINIT(&nes); nd->nd_repstat = nfsvno_accchk(vp, VWRITE, cred, &nes, p, NFSACCCHK_NOOVERRIDE, NFSACCCHK_VPISLOCKED, NULL); nd->nd_repstat = nfsrv_opencheck(clientid, stateidp, stp, vp, nd, p, nd->nd_repstat); if (!nd->nd_repstat) { tempsize = nvap->na_size; NFSVNO_ATTRINIT(nvap); nvap->na_size = tempsize; nd->nd_repstat = VOP_SETATTR(vp, &nvap->na_vattr, cred); } } else if (vp->v_type == VREG) { nd->nd_repstat = nfsrv_opencheck(clientid, stateidp, stp, vp, nd, p, nd->nd_repstat); } } } else { if (ndp->ni_cnd.cn_flags & HASBUF) nfsvno_relpathbuf(ndp); if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) { vrele(ndp->ni_startdir); if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); if (ndp->ni_vp) vput(ndp->ni_vp); } } *vpp = vp; NFSEXITCODE2(0, nd); } /* * Updates the file rev and sets the mtime and ctime * to the current clock time, returning the va_filerev and va_Xtime * values. * Return ESTALE to indicate the vnode is VIRF_DOOMED. */ int nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap, struct nfsrv_descript *nd, struct thread *p) { struct vattr va; VATTR_NULL(&va); vfs_timestamp(&va.va_mtime); if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); if (VN_IS_DOOMED(vp)) return (ESTALE); } (void) VOP_SETATTR(vp, &va, nd->nd_cred); (void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL); return (0); } /* * Glue routine to nfsv4_fillattr(). */ int nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp, struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p, int isdgram, int reterr, int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno) { struct statfs *sf; int error; sf = NULL; if (nfsrv_devidcnt > 0 && (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) { sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO); error = nfsrv_pnfsstatfs(sf, mp); if (error != 0) { free(sf, M_TEMP); sf = NULL; } } error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror, attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root, mounted_on_fileno, sf); free(sf, M_TEMP); NFSEXITCODE2(0, nd); return (error); } /* Since the Readdir vnode ops vary, put the entire functions in here. */ /* * nfs readdir service * - mallocs what it thinks is enough to read * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR * - calls VOP_READDIR() * - loops around building the reply * if the output generated exceeds count break out of loop * The NFSM_CLGET macro is used here so that the reply will be packed * tightly in mbuf clusters. * - it trims out records with d_fileno == 0 * this doesn't matter for Unix clients, but they might confuse clients * for other os'. * - it trims out records with d_type == DT_WHT * these cannot be seen through NFS (unless we extend the protocol) * The alternate call nfsrvd_readdirplus() does lookups as well. * PS: The NFS protocol spec. does not clarify what the "count" byte * argument is a count of.. just name strings and file id's or the * entire reply rpc or ... * I tried just file name and id sizes and it confused the Sun client, * so I am using the full rpc size now. The "paranoia.." comment refers * to including the status longwords that are not a part of the dir. * "entry" structures, but are in the rpc. */ int nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram, struct vnode *vp, struct nfsexstuff *exp) { struct dirent *dp; u_int32_t *tl; int dirlen; char *cpos, *cend, *rbuf; struct nfsvattr at; int nlen, error = 0, getret = 1; int siz, cnt, fullsiz, eofflag, ncookies; u_int64_t off, toff, verf __unused; u_long *cookies = NULL, *cookiep; struct uio io; struct iovec iv; int is_ufs; struct thread *p = curthread; if (nd->nd_repstat) { nfsrv_postopattr(nd, getret, &at); goto out; } if (nd->nd_flag & ND_NFSV2) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); off = fxdr_unsigned(u_quad_t, *tl++); } else { NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); off = fxdr_hyper(tl); tl += 2; verf = fxdr_hyper(tl); tl += 2; } toff = off; cnt = fxdr_unsigned(int, *tl); if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) cnt = NFS_SRVMAXDATA(nd); siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); fullsiz = siz; if (nd->nd_flag & ND_NFSV3) { nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); #if 0 /* * va_filerev is not sufficient as a cookie verifier, * since it is not supposed to change when entries are * removed/added unless that offset cookies returned to * the client are no longer valid. */ if (!nd->nd_repstat && toff && verf != at.na_filerev) nd->nd_repstat = NFSERR_BAD_COOKIE; #endif } if (!nd->nd_repstat && vp->v_type != VDIR) nd->nd_repstat = NFSERR_NOTDIR; if (nd->nd_repstat == 0 && cnt == 0) { if (nd->nd_flag & ND_NFSV2) /* NFSv2 does not have NFSERR_TOOSMALL */ nd->nd_repstat = EPERM; else nd->nd_repstat = NFSERR_TOOSMALL; } if (!nd->nd_repstat) nd->nd_repstat = nfsvno_accchk(vp, VEXEC, nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, NFSACCCHK_VPISLOCKED, NULL); if (nd->nd_repstat) { vput(vp); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; rbuf = malloc(siz, M_TEMP, M_WAITOK); again: eofflag = 0; if (cookies) { free(cookies, M_TEMP); cookies = NULL; } iv.iov_base = rbuf; iv.iov_len = siz; io.uio_iov = &iv; io.uio_iovcnt = 1; io.uio_offset = (off_t)off; io.uio_resid = siz; io.uio_segflg = UIO_SYSSPACE; io.uio_rw = UIO_READ; io.uio_td = NULL; nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, &cookies); off = (u_int64_t)io.uio_offset; if (io.uio_resid) siz -= io.uio_resid; if (!cookies && !nd->nd_repstat) nd->nd_repstat = NFSERR_PERM; if (nd->nd_flag & ND_NFSV3) { getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); if (!nd->nd_repstat) nd->nd_repstat = getret; } /* * Handles the failed cases. nd->nd_repstat == 0 past here. */ if (nd->nd_repstat) { vput(vp); free(rbuf, M_TEMP); if (cookies) free(cookies, M_TEMP); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } /* * If nothing read, return eof * rpc reply */ if (siz == 0) { vput(vp); if (nd->nd_flag & ND_NFSV2) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); } else { nfsrv_postopattr(nd, getret, &at); NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); txdr_hyper(at.na_filerev, tl); tl += 2; } *tl++ = newnfs_false; *tl = newnfs_true; free(rbuf, M_TEMP); free(cookies, M_TEMP); goto out; } /* * Check for degenerate cases of nothing useful read. * If so go try again */ cpos = rbuf; cend = rbuf + siz; dp = (struct dirent *)cpos; cookiep = cookies; /* * For some reason FreeBSD's ufs_readdir() chooses to back the * directory offset up to a block boundary, so it is necessary to * skip over the records that precede the requested offset. This * requires the assumption that file offset cookies monotonically * increase. */ while (cpos < cend && ncookies > 0 && (dp->d_fileno == 0 || dp->d_type == DT_WHT || (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) { cpos += dp->d_reclen; dp = (struct dirent *)cpos; cookiep++; ncookies--; } if (cpos >= cend || ncookies == 0) { siz = fullsiz; toff = off; goto again; } vput(vp); /* * If cnt > MCLBYTES and the reply will not be saved, use * ext_pgs mbufs for TLS. * For NFSv4.0, we do not know for sure if the reply will * be saved, so do not use ext_pgs mbufs for NFSv4.0. */ if (cnt > MCLBYTES && siz > MCLBYTES && (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) nd->nd_flag |= ND_EXTPG; /* * dirlen is the size of the reply, including all XDR and must * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate * if the XDR should be included in "count", but to be safe, we do. * (Include the two booleans at the end of the reply in dirlen now.) */ if (nd->nd_flag & ND_NFSV3) { nfsrv_postopattr(nd, getret, &at); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); txdr_hyper(at.na_filerev, tl); dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; } else { dirlen = 2 * NFSX_UNSIGNED; } /* Loop through the records and build reply */ while (cpos < cend && ncookies > 0) { nlen = dp->d_namlen; if (dp->d_fileno != 0 && dp->d_type != DT_WHT && nlen <= NFS_MAXNAMLEN) { if (nd->nd_flag & ND_NFSV3) dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); else dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); if (dirlen > cnt) { eofflag = 0; break; } /* * Build the directory record xdr from * the dirent entry. */ if (nd->nd_flag & ND_NFSV3) { NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); *tl++ = newnfs_true; *tl++ = 0; } else { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = newnfs_true; } *tl = txdr_unsigned(dp->d_fileno); (void) nfsm_strtom(nd, dp->d_name, nlen); if (nd->nd_flag & ND_NFSV3) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = 0; } else NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(*cookiep); } cpos += dp->d_reclen; dp = (struct dirent *)cpos; cookiep++; ncookies--; } if (cpos < cend) eofflag = 0; NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = newnfs_false; if (eofflag) *tl = newnfs_true; else *tl = newnfs_false; free(rbuf, M_TEMP); free(cookies, M_TEMP); out: NFSEXITCODE2(0, nd); return (0); nfsmout: vput(vp); NFSEXITCODE2(error, nd); return (error); } /* * Readdirplus for V3 and Readdir for V4. */ int nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram, struct vnode *vp, struct nfsexstuff *exp) { struct dirent *dp; u_int32_t *tl; int dirlen; char *cpos, *cend, *rbuf; struct vnode *nvp; fhandle_t nfh; struct nfsvattr nva, at, *nvap = &nva; struct mbuf *mb0, *mb1; struct nfsreferral *refp; int nlen, r, error = 0, getret = 1, usevget = 1; int siz, cnt, fullsiz, eofflag, ncookies, entrycnt; caddr_t bpos0, bpos1; u_int64_t off, toff, verf; u_long *cookies = NULL, *cookiep; nfsattrbit_t attrbits, rderrbits, savbits; struct uio io; struct iovec iv; struct componentname cn; int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls; struct mount *mp, *new_mp; uint64_t mounted_on_fileno; struct thread *p = curthread; int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1; if (nd->nd_repstat) { nfsrv_postopattr(nd, getret, &at); goto out; } NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); off = fxdr_hyper(tl); toff = off; tl += 2; verf = fxdr_hyper(tl); tl += 2; siz = fxdr_unsigned(int, *tl++); cnt = fxdr_unsigned(int, *tl); /* * Use the server's maximum data transfer size as the upper bound * on reply datalen. */ if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) cnt = NFS_SRVMAXDATA(nd); /* * siz is a "hint" of how much directory information (name, fileid, * cookie) should be in the reply. At least one client "hints" 0, * so I set it to cnt for that case. I also round it up to the * next multiple of DIRBLKSIZ. * Since the size of a Readdirplus directory entry reply will always * be greater than a directory entry returned by VOP_READDIR(), it * does not make sense to read more than NFS_SRVMAXDATA() via * VOP_READDIR(). */ if (siz <= 0) siz = cnt; else if (siz > NFS_SRVMAXDATA(nd)) siz = NFS_SRVMAXDATA(nd); siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); if (nd->nd_flag & ND_NFSV4) { error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); if (error) goto nfsmout; NFSSET_ATTRBIT(&savbits, &attrbits); NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd); NFSZERO_ATTRBIT(&rderrbits); NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR); } else { NFSZERO_ATTRBIT(&attrbits); } fullsiz = siz; nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); #if 0 if (!nd->nd_repstat) { if (off && verf != at.na_filerev) { /* * va_filerev is not sufficient as a cookie verifier, * since it is not supposed to change when entries are * removed/added unless that offset cookies returned to * the client are no longer valid. */ if (nd->nd_flag & ND_NFSV4) { nd->nd_repstat = NFSERR_NOTSAME; } else { nd->nd_repstat = NFSERR_BAD_COOKIE; } } } #endif if (!nd->nd_repstat && vp->v_type != VDIR) nd->nd_repstat = NFSERR_NOTDIR; if (!nd->nd_repstat && cnt == 0) nd->nd_repstat = NFSERR_TOOSMALL; if (!nd->nd_repstat) nd->nd_repstat = nfsvno_accchk(vp, VEXEC, nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, NFSACCCHK_VPISLOCKED, NULL); if (nd->nd_repstat) { vput(vp); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0; rbuf = malloc(siz, M_TEMP, M_WAITOK); again: eofflag = 0; if (cookies) { free(cookies, M_TEMP); cookies = NULL; } iv.iov_base = rbuf; iv.iov_len = siz; io.uio_iov = &iv; io.uio_iovcnt = 1; io.uio_offset = (off_t)off; io.uio_resid = siz; io.uio_segflg = UIO_SYSSPACE; io.uio_rw = UIO_READ; io.uio_td = NULL; nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, &cookies); off = (u_int64_t)io.uio_offset; if (io.uio_resid) siz -= io.uio_resid; getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); if (!cookies && !nd->nd_repstat) nd->nd_repstat = NFSERR_PERM; if (!nd->nd_repstat) nd->nd_repstat = getret; if (nd->nd_repstat) { vput(vp); if (cookies) free(cookies, M_TEMP); free(rbuf, M_TEMP); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } /* * If nothing read, return eof * rpc reply */ if (siz == 0) { vput(vp); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); txdr_hyper(at.na_filerev, tl); tl += 2; *tl++ = newnfs_false; *tl = newnfs_true; free(cookies, M_TEMP); free(rbuf, M_TEMP); goto out; } /* * Check for degenerate cases of nothing useful read. * If so go try again */ cpos = rbuf; cend = rbuf + siz; dp = (struct dirent *)cpos; cookiep = cookies; /* * For some reason FreeBSD's ufs_readdir() chooses to back the * directory offset up to a block boundary, so it is necessary to * skip over the records that precede the requested offset. This * requires the assumption that file offset cookies monotonically * increase. */ while (cpos < cend && ncookies > 0 && (dp->d_fileno == 0 || dp->d_type == DT_WHT || (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) || ((nd->nd_flag & ND_NFSV4) && ((dp->d_namlen == 1 && dp->d_name[0] == '.') || (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) { cpos += dp->d_reclen; dp = (struct dirent *)cpos; cookiep++; ncookies--; } if (cpos >= cend || ncookies == 0) { siz = fullsiz; toff = off; goto again; } /* * Busy the file system so that the mount point won't go away * and, as such, VFS_VGET() can be used safely. */ mp = vp->v_mount; vfs_ref(mp); NFSVOPUNLOCK(vp); nd->nd_repstat = vfs_busy(mp, 0); vfs_rel(mp); if (nd->nd_repstat != 0) { vrele(vp); free(cookies, M_TEMP); free(rbuf, M_TEMP); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } /* * Check to see if entries in this directory can be safely acquired * via VFS_VGET() or if a switch to VOP_LOOKUP() is required. * ZFS snapshot directories need VOP_LOOKUP(), so that any * automount of the snapshot directory that is required will * be done. * This needs to be done here for NFSv4, since NFSv4 never does * a VFS_VGET() for "." or "..". */ if (is_zfs == 1) { r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp); if (r == EOPNOTSUPP) { usevget = 0; cn.cn_nameiop = LOOKUP; cn.cn_lkflags = LK_SHARED | LK_RETRY; cn.cn_cred = nd->nd_cred; cn.cn_thread = p; } else if (r == 0) vput(nvp); } /* * If the reply is likely to exceed MCLBYTES and the reply will * not be saved, use ext_pgs mbufs for TLS. * It is difficult to predict how large each entry will be and * how many entries have been read, so just assume the directory * entries grow by a factor of 4 when attributes are included. * For NFSv4.0, we do not know for sure if the reply will * be saved, so do not use ext_pgs mbufs for NFSv4.0. */ if (cnt > MCLBYTES && siz > MCLBYTES / 4 && (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) nd->nd_flag |= ND_EXTPG; /* * Save this position, in case there is an error before one entry * is created. */ mb0 = nd->nd_mb; bpos0 = nd->nd_bpos; bextpg0 = nd->nd_bextpg; bextpgsiz0 = nd->nd_bextpgsiz; /* * Fill in the first part of the reply. * dirlen is the reply length in bytes and cannot exceed cnt. * (Include the two booleans at the end of the reply in dirlen now, * so we recognize when we have exceeded cnt.) */ if (nd->nd_flag & ND_NFSV3) { dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; nfsrv_postopattr(nd, getret, &at); } else { dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED; } NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); txdr_hyper(at.na_filerev, tl); /* * Save this position, in case there is an empty reply needed. */ mb1 = nd->nd_mb; bpos1 = nd->nd_bpos; bextpg1 = nd->nd_bextpg; bextpgsiz1 = nd->nd_bextpgsiz; /* Loop through the records and build reply */ entrycnt = 0; while (cpos < cend && ncookies > 0 && dirlen < cnt) { nlen = dp->d_namlen; if (dp->d_fileno != 0 && dp->d_type != DT_WHT && nlen <= NFS_MAXNAMLEN && ((nd->nd_flag & ND_NFSV3) || nlen > 2 || (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.')) || (nlen == 1 && dp->d_name[0] != '.'))) { /* * Save the current position in the reply, in case * this entry exceeds cnt. */ mb1 = nd->nd_mb; bpos1 = nd->nd_bpos; bextpg1 = nd->nd_bextpg; bextpgsiz1 = nd->nd_bextpgsiz; /* * For readdir_and_lookup get the vnode using * the file number. */ nvp = NULL; refp = NULL; r = 0; at_root = 0; needs_unbusy = 0; new_mp = mp; mounted_on_fileno = (uint64_t)dp->d_fileno; if ((nd->nd_flag & ND_NFSV3) || NFSNONZERO_ATTRBIT(&savbits)) { if (nd->nd_flag & ND_NFSV4) refp = nfsv4root_getreferral(NULL, vp, dp->d_fileno); if (refp == NULL) { if (usevget) r = VFS_VGET(mp, dp->d_fileno, LK_SHARED, &nvp); else r = EOPNOTSUPP; if (r == EOPNOTSUPP) { if (usevget) { usevget = 0; cn.cn_nameiop = LOOKUP; cn.cn_lkflags = LK_SHARED | LK_RETRY; cn.cn_cred = nd->nd_cred; cn.cn_thread = p; } cn.cn_nameptr = dp->d_name; cn.cn_namelen = nlen; cn.cn_flags = ISLASTCN | NOFOLLOW | LOCKLEAF; if (nlen == 2 && dp->d_name[0] == '.' && dp->d_name[1] == '.') cn.cn_flags |= ISDOTDOT; if (NFSVOPLOCK(vp, LK_SHARED) != 0) { nd->nd_repstat = EPERM; break; } if ((vp->v_vflag & VV_ROOT) != 0 && (cn.cn_flags & ISDOTDOT) != 0) { vref(vp); nvp = vp; r = 0; } else { r = VOP_LOOKUP(vp, &nvp, &cn); if (vp != nvp) NFSVOPUNLOCK(vp); } } /* * For NFSv4, check to see if nvp is * a mount point and get the mount * point vnode, as required. */ if (r == 0 && nfsrv_enable_crossmntpt != 0 && (nd->nd_flag & ND_NFSV4) != 0 && nvp->v_type == VDIR && nvp->v_mountedhere != NULL) { new_mp = nvp->v_mountedhere; r = vfs_busy(new_mp, 0); vput(nvp); nvp = NULL; if (r == 0) { r = VFS_ROOT(new_mp, LK_SHARED, &nvp); needs_unbusy = 1; if (r == 0) at_root = 1; } } } /* * If we failed to look up the entry, then it * has become invalid, most likely removed. */ if (r != 0) { if (needs_unbusy) vfs_unbusy(new_mp); goto invalid; } KASSERT(refp != NULL || nvp != NULL, ("%s: undetected lookup error", __func__)); if (refp == NULL && ((nd->nd_flag & ND_NFSV3) || NFSNONZERO_ATTRBIT(&attrbits))) { r = nfsvno_getfh(nvp, &nfh, p); if (!r) r = nfsvno_getattr(nvp, nvap, nd, p, 1, &attrbits); if (r == 0 && is_zfs == 1 && nfsrv_enable_crossmntpt != 0 && (nd->nd_flag & ND_NFSV4) != 0 && nvp->v_type == VDIR && vp->v_mount != nvp->v_mount) { /* * For a ZFS snapshot, there is a * pseudo mount that does not set * v_mountedhere, so it needs to * be detected via a different * mount structure. */ at_root = 1; if (new_mp == mp) new_mp = nvp->v_mount; } } /* * If we failed to get attributes of the entry, * then just skip it for NFSv3 (the traditional * behavior in the old NFS server). * For NFSv4 the behavior is controlled by * RDATTRERROR: we either ignore the error or * fail the request. * Note that RDATTRERROR is never set for NFSv3. */ if (r != 0) { if (!NFSISSET_ATTRBIT(&attrbits, NFSATTRBIT_RDATTRERROR)) { vput(nvp); if (needs_unbusy != 0) vfs_unbusy(new_mp); if ((nd->nd_flag & ND_NFSV3)) goto invalid; nd->nd_repstat = r; break; } } } /* * Build the directory record xdr */ if (nd->nd_flag & ND_NFSV3) { NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); *tl++ = newnfs_true; *tl++ = 0; *tl = txdr_unsigned(dp->d_fileno); dirlen += nfsm_strtom(nd, dp->d_name, nlen); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = 0; *tl = txdr_unsigned(*cookiep); nfsrv_postopattr(nd, 0, nvap); dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1); dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR); if (nvp != NULL) vput(nvp); } else { NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); *tl++ = newnfs_true; *tl++ = 0; *tl = txdr_unsigned(*cookiep); dirlen += nfsm_strtom(nd, dp->d_name, nlen); if (nvp != NULL) { supports_nfsv4acls = nfs_supportsnfsv4acls(nvp); NFSVOPUNLOCK(nvp); } else supports_nfsv4acls = 0; if (refp != NULL) { dirlen += nfsrv_putreferralattr(nd, &savbits, refp, 0, &nd->nd_repstat); if (nd->nd_repstat) { if (nvp != NULL) vrele(nvp); if (needs_unbusy != 0) vfs_unbusy(new_mp); break; } } else if (r) { dirlen += nfsvno_fillattr(nd, new_mp, nvp, nvap, &nfh, r, &rderrbits, nd->nd_cred, p, isdgram, 0, supports_nfsv4acls, at_root, mounted_on_fileno); } else { dirlen += nfsvno_fillattr(nd, new_mp, nvp, nvap, &nfh, r, &attrbits, nd->nd_cred, p, isdgram, 0, supports_nfsv4acls, at_root, mounted_on_fileno); } if (nvp != NULL) vrele(nvp); dirlen += (3 * NFSX_UNSIGNED); } if (needs_unbusy != 0) vfs_unbusy(new_mp); if (dirlen <= cnt) entrycnt++; } invalid: cpos += dp->d_reclen; dp = (struct dirent *)cpos; cookiep++; ncookies--; } vrele(vp); vfs_unbusy(mp); /* * If dirlen > cnt, we must strip off the last entry. If that * results in an empty reply, report NFSERR_TOOSMALL. */ if (dirlen > cnt || nd->nd_repstat) { if (!nd->nd_repstat && entrycnt == 0) nd->nd_repstat = NFSERR_TOOSMALL; if (nd->nd_repstat) { nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); } else nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1); eofflag = 0; } else if (cpos < cend) eofflag = 0; if (!nd->nd_repstat) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = newnfs_false; if (eofflag) *tl = newnfs_true; else *tl = newnfs_false; } free(cookies, M_TEMP); free(rbuf, M_TEMP); out: NFSEXITCODE2(0, nd); return (0); nfsmout: vput(vp); NFSEXITCODE2(error, nd); return (error); } /* * Get the settable attributes out of the mbuf list. * (Return 0 or EBADRPC) */ int nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) { u_int32_t *tl; struct nfsv2_sattr *sp; int error = 0, toclient = 0; switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { case ND_NFSV2: NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); /* * Some old clients didn't fill in the high order 16bits. * --> check the low order 2 bytes for 0xffff */ if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) nvap->na_mode = nfstov_mode(sp->sa_mode); if (sp->sa_uid != newnfs_xdrneg1) nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid); if (sp->sa_gid != newnfs_xdrneg1) nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid); if (sp->sa_size != newnfs_xdrneg1) nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size); if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) { #ifdef notyet fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime); #else nvap->na_atime.tv_sec = fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); nvap->na_atime.tv_nsec = 0; #endif } if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1) fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime); break; case ND_NFSV3: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl == newnfs_true) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); nvap->na_mode = nfstov_mode(*tl); } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl == newnfs_true) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); nvap->na_uid = fxdr_unsigned(uid_t, *tl); } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl == newnfs_true) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); nvap->na_gid = fxdr_unsigned(gid_t, *tl); } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl == newnfs_true) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); nvap->na_size = fxdr_hyper(tl); } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); switch (fxdr_unsigned(int, *tl)) { case NFSV3SATTRTIME_TOCLIENT: NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); fxdr_nfsv3time(tl, &nvap->na_atime); toclient = 1; break; case NFSV3SATTRTIME_TOSERVER: vfs_timestamp(&nvap->na_atime); nvap->na_vaflags |= VA_UTIMES_NULL; break; } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); switch (fxdr_unsigned(int, *tl)) { case NFSV3SATTRTIME_TOCLIENT: NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); fxdr_nfsv3time(tl, &nvap->na_mtime); nvap->na_vaflags &= ~VA_UTIMES_NULL; break; case NFSV3SATTRTIME_TOSERVER: vfs_timestamp(&nvap->na_mtime); if (!toclient) nvap->na_vaflags |= VA_UTIMES_NULL; break; } break; case ND_NFSV4: error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p); } nfsmout: NFSEXITCODE2(error, nd); return (error); } /* * Handle the setable attributes for V4. * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise. */ int nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) { u_int32_t *tl; int attrsum = 0; int i, j; int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0; int moderet, toclient = 0; u_char *cp, namestr[NFSV4_SMALLSTR + 1]; uid_t uid; gid_t gid; u_short mode, mask; /* Same type as va_mode. */ struct vattr va; error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup); if (error) goto nfsmout; NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); attrsize = fxdr_unsigned(int, *tl); /* * Loop around getting the setable attributes. If an unsupported * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return. */ if (retnotsup) { nd->nd_repstat = NFSERR_ATTRNOTSUPP; bitpos = NFSATTRBIT_MAX; } else { bitpos = 0; } moderet = 0; for (; bitpos < NFSATTRBIT_MAX; bitpos++) { if (attrsum > attrsize) { error = NFSERR_BADXDR; goto nfsmout; } if (NFSISSET_ATTRBIT(attrbitp, bitpos)) switch (bitpos) { case NFSATTRBIT_SIZE: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (vp != NULL && vp->v_type != VREG) { error = (vp->v_type == VDIR) ? NFSERR_ISDIR : NFSERR_INVAL; goto nfsmout; } nvap->na_size = fxdr_hyper(tl); attrsum += NFSX_HYPER; break; case NFSATTRBIT_ACL: error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize, p); if (error) goto nfsmout; if (aceerr && !nd->nd_repstat) nd->nd_repstat = aceerr; attrsum += aclsize; break; case NFSATTRBIT_ARCHIVE: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_HIDDEN: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_MIMETYPE: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); i = fxdr_unsigned(int, *tl); error = nfsm_advance(nd, NFSM_RNDUP(i), -1); if (error) goto nfsmout; if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i)); break; case NFSATTRBIT_MODE: moderet = NFSERR_INVAL; /* Can't do MODESETMASKED. */ NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); nvap->na_mode = nfstov_mode(*tl); attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_OWNER: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); j = fxdr_unsigned(int, *tl); if (j < 0) { error = NFSERR_BADXDR; goto nfsmout; } if (j > NFSV4_SMALLSTR) cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); else cp = namestr; error = nfsrv_mtostr(nd, cp, j); if (error) { if (j > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); goto nfsmout; } if (!nd->nd_repstat) { nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid); if (!nd->nd_repstat) nvap->na_uid = uid; } if (j > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); break; case NFSATTRBIT_OWNERGROUP: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); j = fxdr_unsigned(int, *tl); if (j < 0) { error = NFSERR_BADXDR; goto nfsmout; } if (j > NFSV4_SMALLSTR) cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); else cp = namestr; error = nfsrv_mtostr(nd, cp, j); if (error) { if (j > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); goto nfsmout; } if (!nd->nd_repstat) { nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid); if (!nd->nd_repstat) nvap->na_gid = gid; } if (j > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); break; case NFSATTRBIT_SYSTEM: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_TIMEACCESSSET: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); attrsum += NFSX_UNSIGNED; if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); fxdr_nfsv4time(tl, &nvap->na_atime); toclient = 1; attrsum += NFSX_V4TIME; } else { vfs_timestamp(&nvap->na_atime); nvap->na_vaflags |= VA_UTIMES_NULL; } break; case NFSATTRBIT_TIMEBACKUP: NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += NFSX_V4TIME; break; case NFSATTRBIT_TIMECREATE: NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); fxdr_nfsv4time(tl, &nvap->na_btime); attrsum += NFSX_V4TIME; break; case NFSATTRBIT_TIMEMODIFYSET: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); attrsum += NFSX_UNSIGNED; if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); fxdr_nfsv4time(tl, &nvap->na_mtime); nvap->na_vaflags &= ~VA_UTIMES_NULL; attrsum += NFSX_V4TIME; } else { vfs_timestamp(&nvap->na_mtime); if (!toclient) nvap->na_vaflags |= VA_UTIMES_NULL; } break; case NFSATTRBIT_MODESETMASKED: NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); mode = fxdr_unsigned(u_short, *tl++); mask = fxdr_unsigned(u_short, *tl); /* * vp == NULL implies an Open/Create operation. * This attribute can only be used for Setattr and * only for NFSv4.1 or higher. * If moderet != 0, a mode attribute has also been * specified and this attribute cannot be done in the * same Setattr operation. */ if ((nd->nd_flag & ND_NFSV41) == 0) nd->nd_repstat = NFSERR_ATTRNOTSUPP; else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 || vp == NULL) nd->nd_repstat = NFSERR_INVAL; else if (moderet == 0) moderet = VOP_GETATTR(vp, &va, nd->nd_cred); if (moderet == 0) nvap->na_mode = (mode & mask) | (va.va_mode & ~mask); else nd->nd_repstat = moderet; attrsum += 2 * NFSX_UNSIGNED; break; default: nd->nd_repstat = NFSERR_ATTRNOTSUPP; /* * set bitpos so we drop out of the loop. */ bitpos = NFSATTRBIT_MAX; break; } } /* * some clients pad the attrlist, so we need to skip over the * padding. */ if (attrsum > attrsize) { error = NFSERR_BADXDR; } else { attrsize = NFSM_RNDUP(attrsize); if (attrsum < attrsize) error = nfsm_advance(nd, attrsize - attrsum, -1); } nfsmout: NFSEXITCODE2(error, nd); return (error); } /* * Check/setup export credentials. */ int nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp, struct ucred *credanon) { int error = 0; /* * Check/setup credentials. */ if (nd->nd_flag & ND_GSS) exp->nes_exflag &= ~MNT_EXPORTANON; /* * Check to see if the operation is allowed for this security flavor. * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS. * Also, allow Secinfo, so that it can acquire the correct flavor(s). */ if (nfsvno_testexp(nd, exp) && nd->nd_procnum != NFSV4OP_SECINFO && nd->nd_procnum != NFSPROC_FSINFO) { if (nd->nd_flag & ND_NFSV4) error = NFSERR_WRONGSEC; else error = (NFSERR_AUTHERR | AUTH_TOOWEAK); goto out; } /* * Check to see if the file system is exported V4 only. */ if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) { error = NFSERR_PROGNOTV4; goto out; } /* * Now, map the user credentials. * (Note that ND_AUTHNONE will only be set for an NFSv3 * Fsinfo RPC. If set for anything else, this code might need * to change.) */ if (NFSVNO_EXPORTED(exp)) { if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) || NFSVNO_EXPORTANON(exp) || (nd->nd_flag & ND_AUTHNONE) != 0) { nd->nd_cred->cr_uid = credanon->cr_uid; nd->nd_cred->cr_gid = credanon->cr_gid; crsetgroups(nd->nd_cred, credanon->cr_ngroups, credanon->cr_groups); } else if ((nd->nd_flag & ND_GSS) == 0) { /* * If using AUTH_SYS, call nfsrv_getgrpscred() to see * if there is a replacement credential with a group * list set up by "nfsuserd -manage-gids". * If there is no replacement, nfsrv_getgrpscred() * simply returns its argument. */ nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred); } } out: NFSEXITCODE2(error, nd); return (error); } /* * Check exports. */ int nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp, struct ucred **credp) { int error; error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, &exp->nes_numsecflavor, exp->nes_secflavors); if (error) { if (nfs_rootfhset) { exp->nes_exflag = 0; exp->nes_numsecflavor = 0; error = 0; } } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > MAXSECFLAVORS) { printf("nfsvno_checkexp: numsecflavors out of range\n"); exp->nes_numsecflavor = 0; error = EACCES; } NFSEXITCODE(error); return (error); } /* * Get a vnode for a file handle and export stuff. */ int nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam, int lktype, struct vnode **vpp, struct nfsexstuff *exp, struct ucred **credp) { int error; *credp = NULL; exp->nes_numsecflavor = 0; error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp); if (error != 0) /* Make sure the server replies ESTALE to the client. */ error = ESTALE; if (nam && !error) { error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, &exp->nes_numsecflavor, exp->nes_secflavors); if (error) { if (nfs_rootfhset) { exp->nes_exflag = 0; exp->nes_numsecflavor = 0; error = 0; } else { vput(*vpp); } } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > MAXSECFLAVORS) { printf("nfsvno_fhtovp: numsecflavors out of range\n"); exp->nes_numsecflavor = 0; error = EACCES; vput(*vpp); } } NFSEXITCODE(error); return (error); } /* * nfsd_fhtovp() - convert a fh to a vnode ptr * - look up fsid in mount list (if not found ret error) * - get vp and export rights by calling nfsvno_fhtovp() * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon * for AUTH_SYS * - if mpp != NULL, return the mount point so that it can * be used for vn_finished_write() by the caller */ void nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype, struct vnode **vpp, struct nfsexstuff *exp, struct mount **mpp, int startwrite) { struct mount *mp, *mpw; struct ucred *credanon; fhandle_t *fhp; int error; if (mpp != NULL) *mpp = NULL; *vpp = NULL; fhp = (fhandle_t *)nfp->nfsrvfh_data; mp = vfs_busyfs(&fhp->fh_fsid); if (mp == NULL) { nd->nd_repstat = ESTALE; goto out; } if (startwrite) { mpw = mp; error = vn_start_write(NULL, &mpw, V_WAIT); if (error != 0) { mpw = NULL; vfs_unbusy(mp); nd->nd_repstat = ESTALE; goto out; } if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp))) lktype = LK_EXCLUSIVE; } else mpw = NULL; nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp, &credanon); vfs_unbusy(mp); /* * For NFSv4 without a pseudo root fs, unexported file handles * can be returned, so that Lookup works everywhere. */ if (!nd->nd_repstat && exp->nes_exflag == 0 && !(nd->nd_flag & ND_NFSV4)) { vput(*vpp); *vpp = NULL; nd->nd_repstat = EACCES; } /* * If TLS is required by the export, check the flags in nd_flag. */ if (nd->nd_repstat == 0 && ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) || (NFSVNO_EXTLSCERT(exp) && (nd->nd_flag & ND_TLSCERT) == 0) || (NFSVNO_EXTLSCERTUSER(exp) && (nd->nd_flag & ND_TLSCERTUSER) == 0))) { vput(*vpp); nd->nd_repstat = NFSERR_ACCES; } /* * Personally, I've never seen any point in requiring a * reserved port#, since only in the rare case where the * clients are all boxes with secure system privileges, * does it provide any enhanced security, but... some people * believe it to be useful and keep putting this code back in. * (There is also some "security checker" out there that * complains if the nfs server doesn't enforce this.) * However, note the following: * RFC3530 (NFSv4) specifies that a reserved port# not be * required. * RFC2623 recommends that, if a reserved port# is checked for, * that there be a way to turn that off--> ifdef'd. */ #ifdef NFS_REQRSVPORT if (!nd->nd_repstat) { struct sockaddr_in *saddr; struct sockaddr_in6 *saddr6; saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *); if (!(nd->nd_flag & ND_NFSV4) && ((saddr->sin_family == AF_INET && ntohs(saddr->sin_port) >= IPPORT_RESERVED) || (saddr6->sin6_family == AF_INET6 && ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) { vput(*vpp); nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); } } #endif /* NFS_REQRSVPORT */ /* * Check/setup credentials. */ if (!nd->nd_repstat) { nd->nd_saveduid = nd->nd_cred->cr_uid; nd->nd_repstat = nfsd_excred(nd, exp, credanon); if (nd->nd_repstat) vput(*vpp); } if (credanon != NULL) crfree(credanon); if (nd->nd_repstat) { vn_finished_write(mpw); *vpp = NULL; } else if (mpp != NULL) { *mpp = mpw; } out: NFSEXITCODE2(0, nd); } /* * glue for fp. */ static int fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp) { struct filedesc *fdp; struct file *fp; int error = 0; fdp = p->td_proc->p_fd; if (fd < 0 || fd >= fdp->fd_nfiles || (fp = fdp->fd_ofiles[fd].fde_file) == NULL) { error = EBADF; goto out; } *fpp = fp; out: NFSEXITCODE(error); return (error); } /* * Called from nfssvc() to update the exports list. Just call * vfs_export(). This has to be done, since the v4 root fake fs isn't * in the mount list. */ int nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p) { struct nfsex_args *nfsexargp = (struct nfsex_args *)argp; int error = 0; struct nameidata nd; fhandle_t fh; error = vfs_export(&nfsv4root_mnt, &nfsexargp->export); if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0) nfs_rootfhset = 0; else if (error == 0) { if (nfsexargp->fspec == NULL) { error = EPERM; goto out; } /* * If fspec != NULL, this is the v4root path. */ NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec, p); if ((error = namei(&nd)) != 0) goto out; error = nfsvno_getfh(nd.ni_vp, &fh, p); vrele(nd.ni_vp); if (!error) { nfs_rootfh.nfsrvfh_len = NFSX_MYFH; NFSBCOPY((caddr_t)&fh, nfs_rootfh.nfsrvfh_data, sizeof (fhandle_t)); nfs_rootfhset = 1; } } out: NFSEXITCODE(error); return (error); } /* * This function needs to test to see if the system is near its limit * for memory allocation via malloc() or mget() and return True iff * either of these resources are near their limit. * XXX (For now, this is just a stub.) */ int nfsrv_testmalloclimit = 0; int nfsrv_mallocmget_limit(void) { static int printmesg = 0; static int testval = 1; if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) { if ((printmesg++ % 100) == 0) printf("nfsd: malloc/mget near limit\n"); return (1); } return (0); } /* * BSD specific initialization of a mount point. */ void nfsd_mntinit(void) { static int inited = 0; if (inited) return; inited = 1; nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED); TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist); TAILQ_INIT(&nfsv4root_mnt.mnt_lazyvnodelist); nfsv4root_mnt.mnt_export = NULL; TAILQ_INIT(&nfsv4root_opt); TAILQ_INIT(&nfsv4root_newopt); nfsv4root_mnt.mnt_opt = &nfsv4root_opt; nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt; nfsv4root_mnt.mnt_nvnodelistsize = 0; nfsv4root_mnt.mnt_lazyvnodelistsize = 0; } /* * Get a vnode for a file handle, without checking exports, etc. */ struct vnode * nfsvno_getvp(fhandle_t *fhp) { struct mount *mp; struct vnode *vp; int error; mp = vfs_busyfs(&fhp->fh_fsid); if (mp == NULL) return (NULL); error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp); vfs_unbusy(mp); if (error) return (NULL); return (vp); } /* * Do a local VOP_ADVLOCK(). */ int nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first, u_int64_t end, struct thread *td) { int error = 0; struct flock fl; u_int64_t tlen; if (nfsrv_dolocallocks == 0) goto out; ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked"); fl.l_whence = SEEK_SET; fl.l_type = ftype; fl.l_start = (off_t)first; if (end == NFS64BITSSET) { fl.l_len = 0; } else { tlen = end - first; fl.l_len = (off_t)tlen; } /* * For FreeBSD8, the l_pid and l_sysid must be set to the same * values for all calls, so that all locks will be held by the * nfsd server. (The nfsd server handles conflicts between the * various clients.) * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024 * bytes, so it can't be put in l_sysid. */ if (nfsv4_sysid == 0) nfsv4_sysid = nlm_acquire_next_sysid(); fl.l_pid = (pid_t)0; fl.l_sysid = (int)nfsv4_sysid; if (ftype == F_UNLCK) error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl, (F_POSIX | F_REMOTE)); else error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl, (F_POSIX | F_REMOTE)); out: NFSEXITCODE(error); return (error); } /* * Check the nfsv4 root exports. */ int nfsvno_v4rootexport(struct nfsrv_descript *nd) { struct ucred *credanon; int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i; uint64_t exflags; error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags, &credanon, &numsecflavor, secflavors); if (error) { error = NFSERR_PROGUNAVAIL; goto out; } if (credanon != NULL) crfree(credanon); for (i = 0; i < numsecflavor; i++) { if (secflavors[i] == AUTH_SYS) nd->nd_flag |= ND_EXAUTHSYS; else if (secflavors[i] == RPCSEC_GSS_KRB5) nd->nd_flag |= ND_EXGSS; else if (secflavors[i] == RPCSEC_GSS_KRB5I) nd->nd_flag |= ND_EXGSSINTEGRITY; else if (secflavors[i] == RPCSEC_GSS_KRB5P) nd->nd_flag |= ND_EXGSSPRIVACY; } /* And set ND_EXxx flags for TLS. */ if ((exflags & MNT_EXTLS) != 0) { nd->nd_flag |= ND_EXTLS; if ((exflags & MNT_EXTLSCERT) != 0) nd->nd_flag |= ND_EXTLSCERT; if ((exflags & MNT_EXTLSCERTUSER) != 0) nd->nd_flag |= ND_EXTLSCERTUSER; } out: NFSEXITCODE(error); return (error); } /* * Nfs server pseudo system call for the nfsd's */ /* * MPSAFE */ static int nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap) { struct file *fp; struct nfsd_addsock_args sockarg; struct nfsd_nfsd_args nfsdarg; struct nfsd_nfsd_oargs onfsdarg; struct nfsd_pnfsd_args pnfsdarg; struct vnode *vp, *nvp, *curdvp; struct pnfsdsfile *pf; struct nfsdevice *ds, *fds; cap_rights_t rights; int buflen, error, ret; char *buf, *cp, *cp2, *cp3; char fname[PNFS_FILENAME_LEN + 1]; if (uap->flag & NFSSVC_NFSDADDSOCK) { error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg)); if (error) goto out; /* * Since we don't know what rights might be required, * pretend that we need them all. It is better to be too * careful than too reckless. */ error = fget(td, sockarg.sock, - cap_rights_init(&rights, CAP_SOCK_SERVER), &fp); + cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp); if (error != 0) goto out; if (fp->f_type != DTYPE_SOCKET) { fdrop(fp, td); error = EPERM; goto out; } error = nfsrvd_addsock(fp); fdrop(fp, td); } else if (uap->flag & NFSSVC_NFSDNFSD) { if (uap->argp == NULL) { error = EINVAL; goto out; } if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) { error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg)); if (error == 0) { nfsdarg.principal = onfsdarg.principal; nfsdarg.minthreads = onfsdarg.minthreads; nfsdarg.maxthreads = onfsdarg.maxthreads; nfsdarg.version = 1; nfsdarg.addr = NULL; nfsdarg.addrlen = 0; nfsdarg.dnshost = NULL; nfsdarg.dnshostlen = 0; nfsdarg.dspath = NULL; nfsdarg.dspathlen = 0; nfsdarg.mdspath = NULL; nfsdarg.mdspathlen = 0; nfsdarg.mirrorcnt = 1; } } else error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg)); if (error) goto out; if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 && nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 && nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 && nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 && nfsdarg.mirrorcnt >= 1 && nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS && nfsdarg.addr != NULL && nfsdarg.dnshost != NULL && nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) { NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d" " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen, nfsdarg.dspathlen, nfsdarg.dnshostlen, nfsdarg.mdspathlen, nfsdarg.mirrorcnt); cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK); error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen); if (error != 0) { free(cp, M_TEMP); goto out; } cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */ nfsdarg.addr = cp; cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK); error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen); if (error != 0) { free(nfsdarg.addr, M_TEMP); free(cp, M_TEMP); goto out; } cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */ nfsdarg.dnshost = cp; cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK); error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen); if (error != 0) { free(nfsdarg.addr, M_TEMP); free(nfsdarg.dnshost, M_TEMP); free(cp, M_TEMP); goto out; } cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */ nfsdarg.dspath = cp; cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK); error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen); if (error != 0) { free(nfsdarg.addr, M_TEMP); free(nfsdarg.dnshost, M_TEMP); free(nfsdarg.dspath, M_TEMP); free(cp, M_TEMP); goto out; } cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */ nfsdarg.mdspath = cp; } else { nfsdarg.addr = NULL; nfsdarg.addrlen = 0; nfsdarg.dnshost = NULL; nfsdarg.dnshostlen = 0; nfsdarg.dspath = NULL; nfsdarg.dspathlen = 0; nfsdarg.mdspath = NULL; nfsdarg.mdspathlen = 0; nfsdarg.mirrorcnt = 1; } error = nfsrvd_nfsd(td, &nfsdarg); free(nfsdarg.addr, M_TEMP); free(nfsdarg.dnshost, M_TEMP); free(nfsdarg.dspath, M_TEMP); free(nfsdarg.mdspath, M_TEMP); } else if (uap->flag & NFSSVC_PNFSDS) { error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg)); if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER || pnfsdarg.op == PNFSDOP_FORCEDELDS)) { cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1, NULL); if (error == 0) error = nfsrv_deldsserver(pnfsdarg.op, cp, td); free(cp, M_TEMP); } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) { cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS; buf = malloc(buflen, M_TEMP, M_WAITOK); error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1, NULL); NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error); if (error == 0 && pnfsdarg.dspath != NULL) { cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); error = copyinstr(pnfsdarg.dspath, cp2, PATH_MAX + 1, NULL); NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n", error); } else cp2 = NULL; if (error == 0 && pnfsdarg.curdspath != NULL) { cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); error = copyinstr(pnfsdarg.curdspath, cp3, PATH_MAX + 1, NULL); NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n", error); } else cp3 = NULL; curdvp = NULL; fds = NULL; if (error == 0) error = nfsrv_mdscopymr(cp, cp2, cp3, buf, &buflen, fname, td, &vp, &nvp, &pf, &ds, &fds); NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error); if (error == 0) { if (pf->dsf_dir >= nfsrv_dsdirsize) { printf("copymr: dsdir out of range\n"); pf->dsf_dir = 0; } NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen); error = nfsrv_copymr(vp, nvp, ds->nfsdev_dsdir[pf->dsf_dir], ds, pf, (struct pnfsdsfile *)buf, buflen / sizeof(*pf), td->td_ucred, td); vput(vp); vput(nvp); if (fds != NULL && error == 0) { curdvp = fds->nfsdev_dsdir[pf->dsf_dir]; ret = vn_lock(curdvp, LK_EXCLUSIVE); if (ret == 0) { nfsrv_dsremove(curdvp, fname, td->td_ucred, td); NFSVOPUNLOCK(curdvp); } } NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error); } free(cp, M_TEMP); free(cp2, M_TEMP); free(cp3, M_TEMP); free(buf, M_TEMP); } } else { error = nfssvc_srvcall(td, uap, td->td_ucred); } out: NFSEXITCODE(error); return (error); } static int nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred) { struct nfsex_args export; struct nfsex_oldargs oexp; struct file *fp = NULL; int stablefd, i, len; struct nfsd_clid adminrevoke; struct nfsd_dumplist dumplist; struct nfsd_dumpclients *dumpclients; struct nfsd_dumplocklist dumplocklist; struct nfsd_dumplocks *dumplocks; struct nameidata nd; vnode_t vp; int error = EINVAL, igotlock; struct proc *procp; gid_t *grps; static int suspend_nfsd = 0; if (uap->flag & NFSSVC_PUBLICFH) { NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); error = copyin(uap->argp, &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); if (!error) nfs_pubfhset = 1; } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) { error = copyin(uap->argp,(caddr_t)&export, sizeof (struct nfsex_args)); if (!error) { grps = NULL; if (export.export.ex_ngroups > NGROUPS_MAX || export.export.ex_ngroups < 0) error = EINVAL; else if (export.export.ex_ngroups > 0) { grps = malloc(export.export.ex_ngroups * sizeof(gid_t), M_TEMP, M_WAITOK); error = copyin(export.export.ex_groups, grps, export.export.ex_ngroups * sizeof(gid_t)); export.export.ex_groups = grps; } else export.export.ex_groups = NULL; if (!error) error = nfsrv_v4rootexport(&export, cred, p); free(grps, M_TEMP); } } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == NFSSVC_V4ROOTEXPORT) { error = copyin(uap->argp,(caddr_t)&oexp, sizeof (struct nfsex_oldargs)); if (!error) { memset(&export.export, 0, sizeof(export.export)); export.export.ex_flags = (uint64_t)oexp.export.ex_flags; export.export.ex_root = oexp.export.ex_root; export.export.ex_uid = oexp.export.ex_anon.cr_uid; export.export.ex_ngroups = oexp.export.ex_anon.cr_ngroups; export.export.ex_groups = NULL; if (export.export.ex_ngroups > XU_NGROUPS || export.export.ex_ngroups < 0) error = EINVAL; else if (export.export.ex_ngroups > 0) { export.export.ex_groups = malloc( export.export.ex_ngroups * sizeof(gid_t), M_TEMP, M_WAITOK); for (i = 0; i < export.export.ex_ngroups; i++) export.export.ex_groups[i] = oexp.export.ex_anon.cr_groups[i]; } export.export.ex_addr = oexp.export.ex_addr; export.export.ex_addrlen = oexp.export.ex_addrlen; export.export.ex_mask = oexp.export.ex_mask; export.export.ex_masklen = oexp.export.ex_masklen; export.export.ex_indexfile = oexp.export.ex_indexfile; export.export.ex_numsecflavors = oexp.export.ex_numsecflavors; if (export.export.ex_numsecflavors >= MAXSECFLAVORS || export.export.ex_numsecflavors < 0) error = EINVAL; else { for (i = 0; i < export.export.ex_numsecflavors; i++) export.export.ex_secflavors[i] = oexp.export.ex_secflavors[i]; } export.fspec = oexp.fspec; if (error == 0) error = nfsrv_v4rootexport(&export, cred, p); free(export.export.ex_groups, M_TEMP); } } else if (uap->flag & NFSSVC_NOPUBLICFH) { nfs_pubfhset = 0; error = 0; } else if (uap->flag & NFSSVC_STABLERESTART) { error = copyin(uap->argp, (caddr_t)&stablefd, sizeof (int)); if (!error) error = fp_getfvp(p, stablefd, &fp, &vp); if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE)) error = EBADF; if (!error && newnfs_numnfsd != 0) error = EPERM; if (!error) { nfsrv_stablefirst.nsf_fp = fp; nfsrv_setupstable(p); } } else if (uap->flag & NFSSVC_ADMINREVOKE) { error = copyin(uap->argp, (caddr_t)&adminrevoke, sizeof (struct nfsd_clid)); if (!error) error = nfsrv_adminrevoke(&adminrevoke, p); } else if (uap->flag & NFSSVC_DUMPCLIENTS) { error = copyin(uap->argp, (caddr_t)&dumplist, sizeof (struct nfsd_dumplist)); if (!error && (dumplist.ndl_size < 1 || dumplist.ndl_size > NFSRV_MAXDUMPLIST)) error = EPERM; if (!error) { len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size; dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO); nfsrv_dumpclients(dumpclients, dumplist.ndl_size); error = copyout(dumpclients, dumplist.ndl_list, len); free(dumpclients, M_TEMP); } } else if (uap->flag & NFSSVC_DUMPLOCKS) { error = copyin(uap->argp, (caddr_t)&dumplocklist, sizeof (struct nfsd_dumplocklist)); if (!error && (dumplocklist.ndllck_size < 1 || dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST)) error = EPERM; if (!error) error = nfsrv_lookupfilename(&nd, dumplocklist.ndllck_fname, p); if (!error) { len = sizeof (struct nfsd_dumplocks) * dumplocklist.ndllck_size; dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO); nfsrv_dumplocks(nd.ni_vp, dumplocks, dumplocklist.ndllck_size, p); vput(nd.ni_vp); error = copyout(dumplocks, dumplocklist.ndllck_list, len); free(dumplocks, M_TEMP); } } else if (uap->flag & NFSSVC_BACKUPSTABLE) { procp = p->td_proc; PROC_LOCK(procp); nfsd_master_pid = procp->p_pid; bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1); nfsd_master_start = procp->p_stats->p_start; nfsd_master_proc = procp; PROC_UNLOCK(procp); } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) { NFSLOCKV4ROOTMUTEX(); if (suspend_nfsd == 0) { /* Lock out all nfsd threads */ do { igotlock = nfsv4_lock(&nfsd_suspend_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (igotlock == 0 && suspend_nfsd == 0); suspend_nfsd = 1; } NFSUNLOCKV4ROOTMUTEX(); error = 0; } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) { NFSLOCKV4ROOTMUTEX(); if (suspend_nfsd != 0) { nfsv4_unlock(&nfsd_suspend_lock, 0); suspend_nfsd = 0; } NFSUNLOCKV4ROOTMUTEX(); error = 0; } NFSEXITCODE(error); return (error); } /* * Check exports. * Returns 0 if ok, 1 otherwise. */ int nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp) { int i; /* * This seems odd, but allow the case where the security flavor * list is empty. This happens when NFSv4 is traversing non-exported * file systems. Exported file systems should always have a non-empty * security flavor list. */ if (exp->nes_numsecflavor == 0) return (0); for (i = 0; i < exp->nes_numsecflavor; i++) { /* * The tests for privacy and integrity must be first, * since ND_GSS is set for everything but AUTH_SYS. */ if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P && (nd->nd_flag & ND_GSSPRIVACY)) return (0); if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I && (nd->nd_flag & ND_GSSINTEGRITY)) return (0); if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 && (nd->nd_flag & ND_GSS)) return (0); if (exp->nes_secflavors[i] == AUTH_SYS && (nd->nd_flag & ND_GSS) == 0) return (0); } return (1); } /* * Calculate a hash value for the fid in a file handle. */ uint32_t nfsrv_hashfh(fhandle_t *fhp) { uint32_t hashval; hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0); return (hashval); } /* * Calculate a hash value for the sessionid. */ uint32_t nfsrv_hashsessionid(uint8_t *sessionid) { uint32_t hashval; hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0); return (hashval); } /* * Signal the userland master nfsd to backup the stable restart file. */ void nfsrv_backupstable(void) { struct proc *procp; if (nfsd_master_proc != NULL) { procp = pfind(nfsd_master_pid); /* Try to make sure it is the correct process. */ if (procp == nfsd_master_proc && procp->p_stats->p_start.tv_sec == nfsd_master_start.tv_sec && procp->p_stats->p_start.tv_usec == nfsd_master_start.tv_usec && strcmp(procp->p_comm, nfsd_master_comm) == 0) kern_psignal(procp, SIGUSR2); else nfsd_master_proc = NULL; if (procp != NULL) PROC_UNLOCK(procp); } } /* * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror. * The arguments are in a structure, so that they can be passed through * taskqueue for a kernel process to execute this function. */ struct nfsrvdscreate { int done; int inprog; struct task tsk; struct ucred *tcred; struct vnode *dvp; NFSPROC_T *p; struct pnfsdsfile *pf; int err; fhandle_t fh; struct vattr va; struct vattr createva; }; int nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap, fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa, char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp) { struct vnode *nvp; struct nameidata named; struct vattr va; char *bufp; u_long *hashp; struct nfsnode *np; struct nfsmount *nmp; int error; NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE, LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE); nfsvno_setpathbuf(&named, &bufp, &hashp); named.ni_cnd.cn_lkflags = LK_EXCLUSIVE; named.ni_cnd.cn_thread = p; named.ni_cnd.cn_nameptr = bufp; if (fnamep != NULL) { strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1); named.ni_cnd.cn_namelen = strlen(bufp); } else named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp); NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp); /* Create the date file in the DS mount. */ error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); if (error == 0) { error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap); NFSVOPUNLOCK(dvp); if (error == 0) { /* Set the ownership of the file. */ error = VOP_SETATTR(nvp, nvap, tcred); NFSD_DEBUG(4, "nfsrv_dscreate:" " setattr-uid=%d\n", error); if (error != 0) vput(nvp); } if (error != 0) printf("pNFS: pnfscreate failed=%d\n", error); } else printf("pNFS: pnfscreate vnlock=%d\n", error); if (error == 0) { np = VTONFS(nvp); nmp = VFSTONFS(nvp->v_mount); if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs") != 0 || nmp->nm_nam->sa_len > sizeof( struct sockaddr_in6) || np->n_fhp->nfh_len != NFSX_MYFH) { printf("Bad DS file: fstype=%s salen=%d" " fhlen=%d\n", nvp->v_mount->mnt_vfc->vfc_name, nmp->nm_nam->sa_len, np->n_fhp->nfh_len); error = ENOENT; } /* Set extattrs for the DS on the MDS file. */ if (error == 0) { if (dsa != NULL) { error = VOP_GETATTR(nvp, &va, tcred); if (error == 0) { dsa->dsa_filerev = va.va_filerev; dsa->dsa_size = va.va_size; dsa->dsa_atime = va.va_atime; dsa->dsa_mtime = va.va_mtime; dsa->dsa_bytes = va.va_bytes; } } if (error == 0) { NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH); NFSBCOPY(nmp->nm_nam, &pf->dsf_sin, nmp->nm_nam->sa_len); NFSBCOPY(named.ni_cnd.cn_nameptr, pf->dsf_filename, sizeof(pf->dsf_filename)); } } else printf("pNFS: pnfscreate can't get DS" " attr=%d\n", error); if (nvpp != NULL && error == 0) *nvpp = nvp; else vput(nvp); } nfsvno_relpathbuf(&named); return (error); } /* * Start up the thread that will execute nfsrv_dscreate(). */ static void start_dscreate(void *arg, int pending) { struct nfsrvdscreate *dsc; dsc = (struct nfsrvdscreate *)arg; dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh, dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL); dsc->done = 1; NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err); } /* * Create a pNFS data file on the Data Server(s). */ static void nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred, NFSPROC_T *p) { struct nfsrvdscreate *dsc, *tdsc = NULL; struct nfsdevice *ds, *tds, *fds; struct mount *mp; struct pnfsdsfile *pf, *tpf; struct pnfsdsattr dsattr; struct vattr va; struct vnode *dvp[NFSDEV_MAXMIRRORS]; struct nfsmount *nmp; fhandle_t fh; uid_t vauid; gid_t vagid; u_short vamode; struct ucred *tcred; int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret; int failpos, timo; /* Get a DS server directory in a round-robin order. */ mirrorcnt = 1; mp = vp->v_mount; ds = fds = NULL; NFSDDSLOCK(); /* * Search for the first entry that handles this MDS fs, but use the * first entry for all MDS fs's otherwise. */ TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { if (tds->nfsdev_nmp != NULL) { if (tds->nfsdev_mdsisset == 0 && ds == NULL) ds = tds; else if (tds->nfsdev_mdsisset != 0 && fsidcmp( &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) { ds = fds = tds; break; } } } if (ds == NULL) { NFSDDSUNLOCK(); NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n"); return; } i = dsdir[0] = ds->nfsdev_nextdir; ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize; dvp[0] = ds->nfsdev_dsdir[i]; tds = TAILQ_NEXT(ds, nfsdev_list); if (nfsrv_maxpnfsmirror > 1 && tds != NULL) { TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) { if (tds->nfsdev_nmp != NULL && ((tds->nfsdev_mdsisset == 0 && fds == NULL) || (tds->nfsdev_mdsisset != 0 && fds != NULL && fsidcmp(&mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0))) { dsdir[mirrorcnt] = i; dvp[mirrorcnt] = tds->nfsdev_dsdir[i]; mirrorcnt++; if (mirrorcnt >= nfsrv_maxpnfsmirror) break; } } } /* Put at end of list to implement round-robin usage. */ TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); NFSDDSUNLOCK(); dsc = NULL; if (mirrorcnt > 1) tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP, M_WAITOK | M_ZERO); tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK | M_ZERO); error = nfsvno_getfh(vp, &fh, p); if (error == 0) error = VOP_GETATTR(vp, &va, cred); if (error == 0) { /* Set the attributes for "vp" to Setattr the DS vp. */ vauid = va.va_uid; vagid = va.va_gid; vamode = va.va_mode; VATTR_NULL(&va); va.va_uid = vauid; va.va_gid = vagid; va.va_mode = vamode; va.va_size = 0; } else printf("pNFS: pnfscreate getfh+attr=%d\n", error); NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid, cred->cr_gid); /* Make data file name based on FH. */ tcred = newnfs_getcred(); /* * Create the file on each DS mirror, using kernel process(es) for the * additional mirrors. */ failpos = -1; for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) { tpf->dsf_dir = dsdir[i]; tdsc->tcred = tcred; tdsc->p = p; tdsc->pf = tpf; tdsc->createva = *vap; NFSBCOPY(&fh, &tdsc->fh, sizeof(fh)); tdsc->va = va; tdsc->dvp = dvp[i]; tdsc->done = 0; tdsc->inprog = 0; tdsc->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_dscreate, tdsc); NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL, NULL, tcred, p, NULL); if (ret != 0) { KASSERT(error == 0, ("nfsrv_dscreate err=%d", error)); if (failpos == -1 && nfsds_failerr(ret)) failpos = i; else error = ret; } } } if (error == 0) { tpf->dsf_dir = dsdir[mirrorcnt - 1]; error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf, &dsattr, NULL, tcred, p, NULL); if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) { failpos = mirrorcnt - 1; error = 0; } } timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; /* Wait for kernel task(s) to complete. */ for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) { while (tdsc->inprog != 0 && tdsc->done == 0) tsleep(&tdsc->tsk, PVFS, "srvdcr", timo); if (tdsc->err != 0) { if (failpos == -1 && nfsds_failerr(tdsc->err)) failpos = i; else if (error == 0) error = tdsc->err; } } /* * If failpos has been set, that mirror has failed, so it needs * to be disabled. */ if (failpos >= 0) { nmp = VFSTONFS(dvp[failpos]->v_mount); NFSLOCKMNT(nmp); if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | NFSMNTP_CANCELRPCS)) == 0) { nmp->nm_privflag |= NFSMNTP_CANCELRPCS; NFSUNLOCKMNT(nmp); ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos, ds); if (ds != NULL) nfsrv_killrpcs(nmp); NFSLOCKMNT(nmp); nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; wakeup(nmp); } NFSUNLOCKMNT(nmp); } NFSFREECRED(tcred); if (error == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp"); NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n", mirrorcnt, nfsrv_maxpnfsmirror); /* * For all mirrors that couldn't be created, fill in the * *pf structure, but with an IP address == 0.0.0.0. */ tpf = pf + mirrorcnt; for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) { *tpf = *pf; tpf->dsf_sin.sin_family = AF_INET; tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in); tpf->dsf_sin.sin_addr.s_addr = 0; tpf->dsf_sin.sin_port = 0; } error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p); if (error == 0) error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); if (error != 0) printf("pNFS: pnfscreate setextattr=%d\n", error); } else printf("pNFS: pnfscreate=%d\n", error); free(pf, M_TEMP); free(dsc, M_TEMP); } /* * Get the information needed to remove the pNFS Data Server file from the * Metadata file. Upon success, ddvp is set non-NULL to the locked * DS directory vnode. The caller must unlock *ddvp when done with it. */ static void nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp, int *mirrorcntp, char *fname, fhandle_t *fhp) { struct vattr va; struct ucred *tcred; char *buf; int buflen, error; dvpp[0] = NULL; /* If not an exported regular file or not a pNFS server, just return. */ if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || nfsrv_devidcnt == 0) return; /* Check to see if this is the last hard link. */ tcred = newnfs_getcred(); error = VOP_GETATTR(vp, &va, tcred); NFSFREECRED(tcred); if (error != 0) { printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error); return; } if (va.va_nlink > 1) return; error = nfsvno_getfh(vp, fhp, p); if (error != 0) { printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error); return; } buflen = 1024; buf = malloc(buflen, M_TEMP, M_WAITOK); /* Get the directory vnode for the DS mount and the file handle. */ error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp, NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL); free(buf, M_TEMP); if (error != 0) printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error); } /* * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror. * The arguments are in a structure, so that they can be passed through * taskqueue for a kernel process to execute this function. */ struct nfsrvdsremove { int done; int inprog; struct task tsk; struct ucred *tcred; struct vnode *dvp; NFSPROC_T *p; int err; char fname[PNFS_FILENAME_LEN + 1]; }; static int nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred, NFSPROC_T *p) { struct nameidata named; struct vnode *nvp; char *bufp; u_long *hashp; int error; error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); if (error != 0) return (error); named.ni_cnd.cn_nameiop = DELETE; named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY; named.ni_cnd.cn_cred = tcred; named.ni_cnd.cn_thread = p; named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME; nfsvno_setpathbuf(&named, &bufp, &hashp); named.ni_cnd.cn_nameptr = bufp; named.ni_cnd.cn_namelen = strlen(fname); strlcpy(bufp, fname, NAME_MAX); NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp); error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error); if (error == 0) { error = VOP_REMOVE(dvp, nvp, &named.ni_cnd); vput(nvp); } NFSVOPUNLOCK(dvp); nfsvno_relpathbuf(&named); if (error != 0) printf("pNFS: nfsrv_pnfsremove failed=%d\n", error); return (error); } /* * Start up the thread that will execute nfsrv_dsremove(). */ static void start_dsremove(void *arg, int pending) { struct nfsrvdsremove *dsrm; dsrm = (struct nfsrvdsremove *)arg; dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred, dsrm->p); dsrm->done = 1; NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err); } /* * Remove a pNFS data file from a Data Server. * nfsrv_pnfsremovesetup() must have been called before the MDS file was * removed to set up the dvp and fill in the FH. */ static void nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp, NFSPROC_T *p) { struct ucred *tcred; struct nfsrvdsremove *dsrm, *tdsrm; struct nfsdevice *ds; struct nfsmount *nmp; int failpos, i, ret, timo; tcred = newnfs_getcred(); dsrm = NULL; if (mirrorcnt > 1) dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK); /* * Remove the file on each DS mirror, using kernel process(es) for the * additional mirrors. */ failpos = -1; for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { tdsrm->tcred = tcred; tdsrm->p = p; tdsrm->dvp = dvp[i]; strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1); tdsrm->inprog = 0; tdsrm->done = 0; tdsrm->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_dsremove, tdsrm); NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_dsremove(dvp[i], fname, tcred, p); if (failpos == -1 && nfsds_failerr(ret)) failpos = i; } } ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p); if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret)) failpos = mirrorcnt - 1; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; /* Wait for kernel task(s) to complete. */ for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { while (tdsrm->inprog != 0 && tdsrm->done == 0) tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo); if (failpos == -1 && nfsds_failerr(tdsrm->err)) failpos = i; } /* * If failpos has been set, that mirror has failed, so it needs * to be disabled. */ if (failpos >= 0) { nmp = VFSTONFS(dvp[failpos]->v_mount); NFSLOCKMNT(nmp); if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | NFSMNTP_CANCELRPCS)) == 0) { nmp->nm_privflag |= NFSMNTP_CANCELRPCS; NFSUNLOCKMNT(nmp); ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos, ds); if (ds != NULL) nfsrv_killrpcs(nmp); NFSLOCKMNT(nmp); nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; wakeup(nmp); } NFSUNLOCKMNT(nmp); } /* Get rid all layouts for the file. */ nfsrv_freefilelayouts(fhp); NFSFREECRED(tcred); free(dsrm, M_TEMP); } /* * Generate a file name based on the file handle and put it in *bufp. * Return the number of bytes generated. */ static int nfsrv_putfhname(fhandle_t *fhp, char *bufp) { int i; uint8_t *cp; const uint8_t *hexdigits = "0123456789abcdef"; cp = (uint8_t *)fhp; for (i = 0; i < sizeof(*fhp); i++) { bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf]; bufp[2 * i + 1] = hexdigits[*cp++ & 0xf]; } bufp[2 * i] = '\0'; return (2 * i); } /* * Update the Metadata file's attributes from the DS file when a Read/Write * layout is returned. * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file. */ int nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) { struct ucred *tcred; int error; /* Do this as root so that it won't fail with EACCES. */ tcred = newnfs_getcred(); error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN, NULL, NULL, NULL, nap, NULL, NULL, 0, NULL); NFSFREECRED(tcred); return (error); } /* * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file. */ static int nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred, NFSPROC_T *p) { int error; error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL, NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL); return (error); } static int nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred, struct thread *p, int ioproc, struct mbuf **mpp, char *cp, struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp, off_t *offp, int content, bool *eofp) { struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp; fhandle_t fh[NFSDEV_MAXMIRRORS]; struct vnode *dvp[NFSDEV_MAXMIRRORS]; struct nfsdevice *ds; struct pnfsdsattr dsattr; struct opnfsdsattr odsattr; char *buf; int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt; NFSD_DEBUG(4, "in nfsrv_proxyds\n"); /* * If not a regular file, not exported or not a pNFS server, * just return ENOENT. */ if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || nfsrv_devidcnt == 0) return (ENOENT); buflen = 1024; buf = malloc(buflen, M_TEMP, M_WAITOK); error = 0; /* * For Getattr, get the Change attribute (va_filerev) and size (va_size) * from the MetaData file's extended attribute. */ if (ioproc == NFSPROC_GETATTR) { error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf, p); if (error == 0) { if (buflen == sizeof(odsattr)) { NFSBCOPY(buf, &odsattr, buflen); nap->na_filerev = odsattr.dsa_filerev; nap->na_size = odsattr.dsa_size; nap->na_atime = odsattr.dsa_atime; nap->na_mtime = odsattr.dsa_mtime; /* * Fake na_bytes by rounding up na_size. * Since we don't know the block size, just * use BLKDEV_IOSIZE. */ nap->na_bytes = (odsattr.dsa_size + BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1); } else if (buflen == sizeof(dsattr)) { NFSBCOPY(buf, &dsattr, buflen); nap->na_filerev = dsattr.dsa_filerev; nap->na_size = dsattr.dsa_size; nap->na_atime = dsattr.dsa_atime; nap->na_mtime = dsattr.dsa_mtime; nap->na_bytes = dsattr.dsa_bytes; } else error = ENXIO; } if (error == 0) { /* * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr() * returns 0, just return now. nfsrv_checkdsattr() * returns 0 if there is no Read/Write layout * plus either an Open/Write_access or Write * delegation issued to a client for the file. */ if (nfsrv_pnfsgetdsattr == 0 || nfsrv_checkdsattr(vp, p) == 0) { free(buf, M_TEMP); return (error); } } /* * Clear ENOATTR so the code below will attempt to do a * nfsrv_getattrdsrpc() to get the attributes and (re)create * the extended attribute. */ if (error == ENOATTR) error = 0; } origmircnt = -1; trycnt = 0; tryagain: if (error == 0) { buflen = 1024; if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) == LK_EXCLUSIVE) printf("nfsrv_proxyds: Readds vp exclusively locked\n"); error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen, &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL, NULL, NULL); if (error == 0) { for (i = 0; i < mirrorcnt; i++) nmp[i] = VFSTONFS(dvp[i]->v_mount); } else printf("pNFS: proxy getextattr sockaddr=%d\n", error); } else printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error); if (error == 0) { failpos = -1; if (origmircnt == -1) origmircnt = mirrorcnt; /* * If failpos is set to a mirror#, then that mirror has * failed and will be disabled. For Read, Getattr and Seek, the * function only tries one mirror, so if that mirror has * failed, it will need to be retried. As such, increment * tryitagain for these cases. * For Write, Setattr and Setacl, the function tries all * mirrors and will not return an error for the case where * one mirror has failed. For these cases, the functioning * mirror(s) will have been modified, so a retry isn't * necessary. These functions will set failpos for the * failed mirror#. */ if (ioproc == NFSPROC_READDS) { error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0], mpp, mpp2); if (nfsds_failerr(error) && mirrorcnt > 1) { /* * Setting failpos will cause the mirror * to be disabled and then a retry of this * read is required. */ failpos = 0; error = 0; trycnt++; } } else if (ioproc == NFSPROC_WRITEDS) error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp, &nmp[0], mirrorcnt, mpp, cp, &failpos); else if (ioproc == NFSPROC_SETATTR) error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0], mirrorcnt, nap, &failpos); else if (ioproc == NFSPROC_SETACL) error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0], mirrorcnt, aclp, &failpos); else if (ioproc == NFSPROC_SEEKDS) { error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred, p, nmp[0]); if (nfsds_failerr(error) && mirrorcnt > 1) { /* * Setting failpos will cause the mirror * to be disabled and then a retry of this * read is required. */ failpos = 0; error = 0; trycnt++; } } else if (ioproc == NFSPROC_ALLOCATE) error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp, &nmp[0], mirrorcnt, &failpos); else { error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p, vp, nmp[mirrorcnt - 1], nap); if (nfsds_failerr(error) && mirrorcnt > 1) { /* * Setting failpos will cause the mirror * to be disabled and then a retry of this * getattr is required. */ failpos = mirrorcnt - 1; error = 0; trycnt++; } } ds = NULL; if (failpos >= 0) { failnmp = nmp[failpos]; NFSLOCKMNT(failnmp); if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM | NFSMNTP_CANCELRPCS)) == 0) { failnmp->nm_privflag |= NFSMNTP_CANCELRPCS; NFSUNLOCKMNT(failnmp); ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, failnmp, p); NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n", failpos, ds); if (ds != NULL) nfsrv_killrpcs(failnmp); NFSLOCKMNT(failnmp); failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; wakeup(failnmp); } NFSUNLOCKMNT(failnmp); } for (i = 0; i < mirrorcnt; i++) NFSVOPUNLOCK(dvp[i]); NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error, trycnt); /* Try the Read/Getattr again if a mirror was deleted. */ if (ds != NULL && trycnt > 0 && trycnt < origmircnt) goto tryagain; } else { /* Return ENOENT for any Extended Attribute error. */ error = ENOENT; } free(buf, M_TEMP); NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error); return (error); } /* * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended * attribute. * newnmpp - If it points to a non-NULL nmp, that is the destination and needs * to be checked. If it points to a NULL nmp, then it returns * a suitable destination. * curnmp - If non-NULL, it is the source mount for the copy. */ int nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp, int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp, char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp, struct nfsmount *curnmp, int *ippos, int *dsdirp) { struct vnode *dvp, *nvp = NULL, **tdvpp; struct mount *mp; struct nfsmount *nmp, *newnmp; struct sockaddr *sad; struct sockaddr_in *sin; struct nfsdevice *ds, *tds, *fndds; struct pnfsdsfile *pf; uint32_t dsdir; int error, fhiszero, fnd, gotone, i, mirrorcnt; ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp"); *mirrorcntp = 1; tdvpp = dvpp; if (nvpp != NULL) *nvpp = NULL; if (dvpp != NULL) *dvpp = NULL; if (ippos != NULL) *ippos = -1; if (newnmpp != NULL) newnmp = *newnmpp; else newnmp = NULL; mp = vp->v_mount; error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", buflenp, buf, p); mirrorcnt = *buflenp / sizeof(*pf); if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS || *buflenp != sizeof(*pf) * mirrorcnt)) error = ENOATTR; pf = (struct pnfsdsfile *)buf; /* If curnmp != NULL, check for a match in the mirror list. */ if (curnmp != NULL && error == 0) { fnd = 0; for (i = 0; i < mirrorcnt; i++, pf++) { sad = (struct sockaddr *)&pf->dsf_sin; if (nfsaddr2_match(sad, curnmp->nm_nam)) { if (ippos != NULL) *ippos = i; fnd = 1; break; } } if (fnd == 0) error = ENXIO; } gotone = 0; pf = (struct pnfsdsfile *)buf; NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt, error); for (i = 0; i < mirrorcnt && error == 0; i++, pf++) { fhiszero = 0; sad = (struct sockaddr *)&pf->dsf_sin; sin = &pf->dsf_sin; dsdir = pf->dsf_dir; if (dsdir >= nfsrv_dsdirsize) { printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir); error = ENOATTR; } else if (nvpp != NULL && newnmp != NULL && nfsaddr2_match(sad, newnmp->nm_nam)) error = EEXIST; if (error == 0) { if (ippos != NULL && curnmp == NULL && sad->sa_family == AF_INET && sin->sin_addr.s_addr == 0) *ippos = i; if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0) fhiszero = 1; /* Use the socket address to find the mount point. */ fndds = NULL; NFSDDSLOCK(); /* Find a match for the IP address. */ TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds->nfsdev_nmp != NULL) { dvp = ds->nfsdev_dvp; nmp = VFSTONFS(dvp->v_mount); if (nmp != ds->nfsdev_nmp) printf("different2 nmp %p %p\n", nmp, ds->nfsdev_nmp); if (nfsaddr2_match(sad, nmp->nm_nam)) { fndds = ds; break; } } } if (fndds != NULL && newnmpp != NULL && newnmp == NULL) { /* Search for a place to make a mirror copy. */ TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { if (tds->nfsdev_nmp != NULL && fndds != tds && ((tds->nfsdev_mdsisset == 0 && fndds->nfsdev_mdsisset == 0) || (tds->nfsdev_mdsisset != 0 && fndds->nfsdev_mdsisset != 0 && fsidcmp(&tds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0))) { *newnmpp = tds->nfsdev_nmp; break; } } if (tds != NULL) { /* * Move this entry to the end of the * list, so it won't be selected as * easily the next time. */ TAILQ_REMOVE(&nfsrv_devidhead, tds, nfsdev_list); TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds, nfsdev_list); } } NFSDDSUNLOCK(); if (fndds != NULL) { dvp = fndds->nfsdev_dsdir[dsdir]; if (lktype != 0 || fhiszero != 0 || (nvpp != NULL && *nvpp == NULL)) { if (fhiszero != 0) error = vn_lock(dvp, LK_EXCLUSIVE); else if (lktype != 0) error = vn_lock(dvp, lktype); else error = vn_lock(dvp, LK_SHARED); /* * If the file handle is all 0's, try to * do a Lookup against the DS to acquire * it. * If dvpp == NULL or the Lookup fails, * unlock dvp after the call. */ if (error == 0 && (fhiszero != 0 || (nvpp != NULL && *nvpp == NULL))) { error = nfsrv_pnfslookupds(vp, dvp, pf, &nvp, p); if (error == 0) { if (fhiszero != 0) nfsrv_pnfssetfh( vp, pf, devid, fnamep, nvp, p); if (nvpp != NULL && *nvpp == NULL) { *nvpp = nvp; *dsdirp = dsdir; } else vput(nvp); } if (error != 0 || lktype == 0) NFSVOPUNLOCK(dvp); } } if (error == 0) { gotone++; NFSD_DEBUG(4, "gotone=%d\n", gotone); if (devid != NULL) { NFSBCOPY(fndds->nfsdev_deviceid, devid, NFSX_V4DEVICEID); devid += NFSX_V4DEVICEID; } if (dvpp != NULL) *tdvpp++ = dvp; if (fhp != NULL) NFSBCOPY(&pf->dsf_fh, fhp++, NFSX_MYFH); if (fnamep != NULL && gotone == 1) strlcpy(fnamep, pf->dsf_filename, sizeof(pf->dsf_filename)); } else NFSD_DEBUG(4, "nfsrv_dsgetsockmnt " "err=%d\n", error); } } } if (error == 0 && gotone == 0) error = ENOENT; NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone, error); if (error == 0) *mirrorcntp = gotone; else { if (gotone > 0 && dvpp != NULL) { /* * If the error didn't occur on the first one and * dvpp != NULL, the one(s) prior to the failure will * have locked dvp's that need to be unlocked. */ for (i = 0; i < gotone; i++) { NFSVOPUNLOCK(*dvpp); *dvpp++ = NULL; } } /* * If it found the vnode to be copied from before a failure, * it needs to be vput()'d. */ if (nvpp != NULL && *nvpp != NULL) { vput(*nvpp); *nvpp = NULL; } } return (error); } /* * Set the extended attribute for the Change attribute. */ static int nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) { struct pnfsdsattr dsattr; int error; ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp"); dsattr.dsa_filerev = nap->na_filerev; dsattr.dsa_size = nap->na_size; dsattr.dsa_atime = nap->na_atime; dsattr.dsa_mtime = nap->na_mtime; dsattr.dsa_bytes = nap->na_bytes; error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); if (error != 0) printf("pNFS: setextattr=%d\n", error); return (error); } static int nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp) { uint32_t *tl; struct nfsrv_descript *nd; nfsv4stateid_t st; struct mbuf *m, *m2; int error = 0, retlen, tlen, trimlen; NFSD_DEBUG(4, "in nfsrv_readdsrpc\n"); nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); *mpp = NULL; /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp), NULL, NULL, 0, 0); nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3); txdr_hyper(off, tl); *(tl + 2) = txdr_unsigned(len); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); NFSM_STRSIZ(retlen, len); if (retlen > 0) { /* Trim off the pre-data XDR from the mbuf chain. */ m = nd->nd_mrep; while (m != NULL && m != nd->nd_md) { if (m->m_next == nd->nd_md) { m->m_next = NULL; m_freem(nd->nd_mrep); nd->nd_mrep = m = nd->nd_md; } else m = m->m_next; } if (m == NULL) { printf("nfsrv_readdsrpc: busted mbuf list\n"); error = ENOENT; goto nfsmout; } /* * Now, adjust first mbuf so that any XDR before the * read data is skipped over. */ trimlen = nd->nd_dpos - mtod(m, char *); if (trimlen > 0) { m->m_len -= trimlen; NFSM_DATAP(m, trimlen); } /* * Truncate the mbuf chain at retlen bytes of data, * plus XDR padding that brings the length up to a * multiple of 4. */ tlen = NFSM_RNDUP(retlen); do { if (m->m_len >= tlen) { m->m_len = tlen; tlen = 0; m2 = m->m_next; m->m_next = NULL; m_freem(m2); break; } tlen -= m->m_len; m = m->m_next; } while (m != NULL); if (tlen > 0) { printf("nfsrv_readdsrpc: busted mbuf list\n"); error = ENOENT; goto nfsmout; } *mpp = nd->nd_mrep; *mpendp = m; nd->nd_mrep = NULL; } } else error = nd->nd_repstat; nfsmout: /* If nd->nd_mrep is already NULL, this is a no-op. */ m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error); return (error); } /* * Do a write RPC on a DS data file, using this structure for the arguments, * so that this function can be executed by a separate kernel process. */ struct nfsrvwritedsdorpc { int done; int inprog; struct task tsk; fhandle_t fh; off_t off; int len; struct nfsmount *nmp; struct ucred *cred; NFSPROC_T *p; struct mbuf *m; int err; }; static int nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len, struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p) { uint32_t *tl; struct nfsrv_descript *nd; nfsattrbit_t attrbits; nfsv4stateid_t st; int commit, error, retlen; nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp, sizeof(fhandle_t), NULL, NULL, 0, 0); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); txdr_hyper(off, tl); tl += 2; /* * Do all writes FileSync, since the server doesn't hold onto dirty * buffers. Since clients should be accessing the DS servers directly * using the pNFS layouts, this just needs to work correctly as a * fallback. */ *tl++ = txdr_unsigned(NFSWRITE_FILESYNC); *tl = txdr_unsigned(len); NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len); /* Put data in mbuf chain. */ nd->nd_mb->m_next = m; /* Set nd_mb and nd_bpos to end of data. */ while (m->m_next != NULL) m = m->m_next; nd->nd_mb = m; nfsm_set(nd, m->m_len); NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len); /* Do a Getattr for the attributes that change upon writing. */ NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat); /* Get rid of weak cache consistency data for now. */ if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == (ND_NFSV4 | ND_V4WCCATTR)) { error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error); if (error != 0) goto nfsmout; /* * Get rid of Op# and status for next op. */ NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); if (*++tl != 0) nd->nd_flag |= ND_NOMOREDATA; } if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); retlen = fxdr_unsigned(int, *tl++); commit = fxdr_unsigned(int, *tl); if (commit != NFSWRITE_FILESYNC) error = NFSERR_IO; NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n", retlen, commit, error); } else error = nd->nd_repstat; /* We have no use for the Write Verifier since we use FileSync. */ /* * Get the Change, Size, Access Time and Modify Time attributes and set * on the Metadata file, so its attributes will be what the file's * would be if it had been written. */ if (error == 0) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); } NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error); nfsmout: m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error); return (error); } /* * Start up the thread that will execute nfsrv_writedsdorpc(). */ static void start_writedsdorpc(void *arg, int pending) { struct nfsrvwritedsdorpc *drpc; drpc = (struct nfsrvwritedsdorpc *)arg; drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off, drpc->len, NULL, drpc->m, drpc->cred, drpc->p); drpc->done = 1; NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err); } static int nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct mbuf **mpp, char *cp, int *failposp) { struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL; struct nfsvattr na; struct mbuf *m; int error, i, offs, ret, timo; NFSD_DEBUG(4, "in nfsrv_writedsrpc\n"); KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain")); drpc = NULL; if (mirrorcnt > 1) tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, M_WAITOK); /* Calculate offset in mbuf chain that data starts. */ offs = cp - mtod(*mpp, char *); NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len); /* * Do the write RPC for every DS, using a separate kernel process * for every DS except the last one. */ error = 0; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { tdrpc->done = 0; NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); tdrpc->off = off; tdrpc->len = len; tdrpc->nmp = *nmpp; tdrpc->cred = cred; tdrpc->p = p; tdrpc->inprog = 0; tdrpc->err = 0; tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_writedsdorpc, tdrpc); NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL, tdrpc->m, cred, p); if (nfsds_failerr(ret) && *failposp == -1) *failposp = i; else if (error == 0 && ret != 0) error = ret; } nmpp++; fhp++; } m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p); if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) *failposp = mirrorcnt - 1; else if (error == 0 && ret != 0) error = ret; if (error == 0) error = nfsrv_setextattr(vp, &na, p); NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error); tdrpc = drpc; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { /* Wait for RPCs on separate threads to complete. */ while (tdrpc->inprog != 0 && tdrpc->done == 0) tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo); if (nfsds_failerr(tdrpc->err) && *failposp == -1) *failposp = i; else if (error == 0 && tdrpc->err != 0) error = tdrpc->err; } free(drpc, M_TEMP); return (error); } /* * Do a allocate RPC on a DS data file, using this structure for the arguments, * so that this function can be executed by a separate kernel process. */ struct nfsrvallocatedsdorpc { int done; int inprog; struct task tsk; fhandle_t fh; off_t off; off_t len; struct nfsmount *nmp; struct ucred *cred; NFSPROC_T *p; int err; }; static int nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) { uint32_t *tl; struct nfsrv_descript *nd; nfsattrbit_t attrbits; nfsv4stateid_t st; int error; nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp, sizeof(fhandle_t), NULL, NULL, 0, 0); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); txdr_hyper(off, tl); tl += 2; txdr_hyper(len, tl); tl += 2; NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len); *tl = txdr_unsigned(NFSV4OP_GETATTR); NFSGETATTR_ATTRBIT(&attrbits); nfsrv_putattrbit(nd, &attrbits); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n", nd->nd_repstat); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); } else error = nd->nd_repstat; NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error); nfsmout: m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error); return (error); } /* * Start up the thread that will execute nfsrv_allocatedsdorpc(). */ static void start_allocatedsdorpc(void *arg, int pending) { struct nfsrvallocatedsdorpc *drpc; drpc = (struct nfsrvallocatedsdorpc *)arg; drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, drpc->len, NULL, drpc->cred, drpc->p); drpc->done = 1; NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err); } static int nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, int *failposp) { struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL; struct nfsvattr na; int error, i, ret, timo; NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n"); drpc = NULL; if (mirrorcnt > 1) tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, M_WAITOK); /* * Do the allocate RPC for every DS, using a separate kernel process * for every DS except the last one. */ error = 0; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { tdrpc->done = 0; NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); tdrpc->off = off; tdrpc->len = len; tdrpc->nmp = *nmpp; tdrpc->cred = cred; tdrpc->p = p; tdrpc->inprog = 0; tdrpc->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc); NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL, cred, p); if (nfsds_failerr(ret) && *failposp == -1) *failposp = i; else if (error == 0 && ret != 0) error = ret; } nmpp++; fhp++; } ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) *failposp = mirrorcnt - 1; else if (error == 0 && ret != 0) error = ret; if (error == 0) error = nfsrv_setextattr(vp, &na, p); NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error); tdrpc = drpc; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { /* Wait for RPCs on separate threads to complete. */ while (tdrpc->inprog != 0 && tdrpc->done == 0) tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); if (nfsds_failerr(tdrpc->err) && *failposp == -1) *failposp = i; else if (error == 0 && tdrpc->err != 0) error = tdrpc->err; } free(drpc, M_TEMP); return (error); } static int nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap, struct nfsvattr *dsnap) { uint32_t *tl; struct nfsrv_descript *nd; nfsv4stateid_t st; nfsattrbit_t attrbits; int error; NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n"); nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp), NULL, NULL, 0, 0); nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0); /* Do a Getattr for the attributes that change due to writing. */ NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n", nd->nd_repstat); /* Get rid of weak cache consistency data for now. */ if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == (ND_NFSV4 | ND_V4WCCATTR)) { error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error); if (error != 0) goto nfsmout; /* * Get rid of Op# and status for next op. */ NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); if (*++tl != 0) nd->nd_flag |= ND_NOMOREDATA; } error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); if (error != 0) goto nfsmout; if (nd->nd_repstat != 0) error = nd->nd_repstat; /* * Get the Change, Size, Access Time and Modify Time attributes and set * on the Metadata file, so its attributes will be what the file's * would be if it had been written. */ if (error == 0) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); } NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error); nfsmout: m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error); return (error); } struct nfsrvsetattrdsdorpc { int done; int inprog; struct task tsk; fhandle_t fh; struct nfsmount *nmp; struct vnode *vp; struct ucred *cred; NFSPROC_T *p; struct nfsvattr na; struct nfsvattr dsna; int err; }; /* * Start up the thread that will execute nfsrv_setattrdsdorpc(). */ static void start_setattrdsdorpc(void *arg, int pending) { struct nfsrvsetattrdsdorpc *drpc; drpc = (struct nfsrvsetattrdsdorpc *)arg; drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p, drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna); drpc->done = 1; } static int nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct nfsvattr *nap, int *failposp) { struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL; struct nfsvattr na; int error, i, ret, timo; NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n"); drpc = NULL; if (mirrorcnt > 1) tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, M_WAITOK); /* * Do the setattr RPC for every DS, using a separate kernel process * for every DS except the last one. */ error = 0; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { tdrpc->done = 0; tdrpc->inprog = 0; NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); tdrpc->nmp = *nmpp; tdrpc->vp = vp; tdrpc->cred = cred; tdrpc->p = p; tdrpc->na = *nap; tdrpc->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc); NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na); if (nfsds_failerr(ret) && *failposp == -1) *failposp = i; else if (error == 0 && ret != 0) error = ret; } nmpp++; fhp++; } ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na); if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) *failposp = mirrorcnt - 1; else if (error == 0 && ret != 0) error = ret; if (error == 0) error = nfsrv_setextattr(vp, &na, p); NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error); tdrpc = drpc; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { /* Wait for RPCs on separate threads to complete. */ while (tdrpc->inprog != 0 && tdrpc->done == 0) tsleep(&tdrpc->tsk, PVFS, "srvsads", timo); if (nfsds_failerr(tdrpc->err) && *failposp == -1) *failposp = i; else if (error == 0 && tdrpc->err != 0) error = tdrpc->err; } free(drpc, M_TEMP); return (error); } /* * Do a Setattr of an NFSv4 ACL on the DS file. */ static int nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount *nmp, struct acl *aclp) { struct nfsrv_descript *nd; nfsv4stateid_t st; nfsattrbit_t attrbits; int error; NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n"); nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp), NULL, NULL, 0, 0); nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL); /* * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(), * so passing in the metadata "vp" will be ok, since it is of * the same type (VREG). */ nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL, NULL, 0, 0, 0, 0, 0, NULL); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n", nd->nd_repstat); error = nd->nd_repstat; m_freem(nd->nd_mrep); free(nd, M_TEMP); return (error); } struct nfsrvsetacldsdorpc { int done; int inprog; struct task tsk; fhandle_t fh; struct nfsmount *nmp; struct vnode *vp; struct ucred *cred; NFSPROC_T *p; struct acl *aclp; int err; }; /* * Start up the thread that will execute nfsrv_setacldsdorpc(). */ static void start_setacldsdorpc(void *arg, int pending) { struct nfsrvsetacldsdorpc *drpc; drpc = (struct nfsrvsetacldsdorpc *)arg; drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p, drpc->vp, drpc->nmp, drpc->aclp); drpc->done = 1; } static int nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp, int *failposp) { struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL; int error, i, ret, timo; NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n"); drpc = NULL; if (mirrorcnt > 1) tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, M_WAITOK); /* * Do the setattr RPC for every DS, using a separate kernel process * for every DS except the last one. */ error = 0; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { tdrpc->done = 0; tdrpc->inprog = 0; NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); tdrpc->nmp = *nmpp; tdrpc->vp = vp; tdrpc->cred = cred; tdrpc->p = p; tdrpc->aclp = aclp; tdrpc->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_setacldsdorpc, tdrpc); NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp); if (nfsds_failerr(ret) && *failposp == -1) *failposp = i; else if (error == 0 && ret != 0) error = ret; } nmpp++; fhp++; } ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp); if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) *failposp = mirrorcnt - 1; else if (error == 0 && ret != 0) error = ret; NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error); tdrpc = drpc; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { /* Wait for RPCs on separate threads to complete. */ while (tdrpc->inprog != 0 && tdrpc->done == 0) tsleep(&tdrpc->tsk, PVFS, "srvacds", timo); if (nfsds_failerr(tdrpc->err) && *failposp == -1) *failposp = i; else if (error == 0 && tdrpc->err != 0) error = tdrpc->err; } free(drpc, M_TEMP); return (error); } /* * Getattr call to the DS for the attributes that change due to writing. */ static int nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap) { struct nfsrv_descript *nd; int error; nfsattrbit_t attrbits; NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n"); nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp, sizeof(fhandle_t), NULL, NULL, 0, 0); NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); (void) nfsrv_putattrbit(nd, &attrbits); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n", nd->nd_repstat); if (nd->nd_repstat == 0) { error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); /* * We can only save the updated values in the extended * attribute if the vp is exclusively locked. * This should happen when any of the following operations * occur on the vnode: * Close, Delegreturn, LayoutCommit, LayoutReturn * As such, the updated extended attribute should get saved * before nfsrv_checkdsattr() returns 0 and allows the cached * attributes to be returned without calling this function. */ if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) { error = nfsrv_setextattr(vp, nap, p); NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n", error); } } else error = nd->nd_repstat; m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error); return (error); } /* * Seek call to a DS. */ static int nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp) { uint32_t *tl; struct nfsrv_descript *nd; nfsv4stateid_t st; int error; NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n"); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp, sizeof(fhandle_t), NULL, NULL, 0, 0); nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED); txdr_hyper(*offp, tl); tl += 2; *tl = txdr_unsigned(content); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER); if (*tl++ == newnfs_true) *eofp = true; else *eofp = false; *offp = fxdr_hyper(tl); } else error = nd->nd_repstat; nfsmout: m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error); return (error); } /* * Get the device id and file handle for a DS file. */ int nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp, fhandle_t *fhp, char *devid) { int buflen, error; char *buf; buflen = 1024; buf = malloc(buflen, M_TEMP, M_WAITOK); error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL, fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL); free(buf, M_TEMP); return (error); } /* * Do a Lookup against the DS for the filename. */ static int nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf, struct vnode **nvpp, NFSPROC_T *p) { struct nameidata named; struct ucred *tcred; char *bufp; u_long *hashp; struct vnode *nvp; int error; tcred = newnfs_getcred(); named.ni_cnd.cn_nameiop = LOOKUP; named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY; named.ni_cnd.cn_cred = tcred; named.ni_cnd.cn_thread = p; named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME; nfsvno_setpathbuf(&named, &bufp, &hashp); named.ni_cnd.cn_nameptr = bufp; named.ni_cnd.cn_namelen = strlen(pf->dsf_filename); strlcpy(bufp, pf->dsf_filename, NAME_MAX); NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp); error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error); NFSFREECRED(tcred); nfsvno_relpathbuf(&named); if (error == 0) *nvpp = nvp; NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error); return (error); } /* * Set the file handle to the correct one. */ static void nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid, char *fnamep, struct vnode *nvp, NFSPROC_T *p) { struct nfsnode *np; int ret = 0; np = VTONFS(nvp); NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH); /* * We can only do a vn_set_extattr() if the vnode is exclusively * locked and vn_start_write() has been done. If devid != NULL or * fnamep != NULL or the vnode is shared locked, vn_start_write() * may not have been done. * If not done now, it will be done on a future call. */ if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) == LK_EXCLUSIVE) ret = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf), (char *)pf, p); NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret); } /* * Cause RPCs waiting on "nmp" to fail. This is called for a DS mount point * when the DS has failed. */ void nfsrv_killrpcs(struct nfsmount *nmp) { /* * Call newnfs_nmcancelreqs() to cause * any RPCs in progress on the mount point to * fail. * This will cause any process waiting for an * RPC to complete while holding a vnode lock * on the mounted-on vnode (such as "df" or * a non-forced "umount") to fail. * This will unlock the mounted-on vnode so * a forced dismount can succeed. * The NFSMNTP_CANCELRPCS flag should be set when this function is * called. */ newnfs_nmcancelreqs(nmp); } /* * Sum up the statfs info for each of the DSs, so that the client will * receive the total for all DSs. */ static int nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp) { struct statfs *tsf; struct nfsdevice *ds; struct vnode **dvpp, **tdvpp, *dvp; uint64_t tot; int cnt, error = 0, i; if (nfsrv_devidcnt <= 0) return (ENXIO); dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); /* Get an array of the dvps for the DSs. */ tdvpp = dvpp; i = 0; NFSDDSLOCK(); /* First, search for matches for same file system. */ TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) { if (++i > nfsrv_devidcnt) break; *tdvpp++ = ds->nfsdev_dvp; } } /* * If no matches for same file system, total all servers not assigned * to a file system. */ if (i == 0) { TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset == 0) { if (++i > nfsrv_devidcnt) break; *tdvpp++ = ds->nfsdev_dvp; } } } NFSDDSUNLOCK(); cnt = i; /* Do a VFS_STATFS() for each of the DSs and sum them up. */ tdvpp = dvpp; for (i = 0; i < cnt && error == 0; i++) { dvp = *tdvpp++; error = VFS_STATFS(dvp->v_mount, tsf); if (error == 0) { if (sf->f_bsize == 0) { if (tsf->f_bsize > 0) sf->f_bsize = tsf->f_bsize; else sf->f_bsize = 8192; } if (tsf->f_blocks > 0) { if (sf->f_bsize != tsf->f_bsize) { tot = tsf->f_blocks * tsf->f_bsize; sf->f_blocks += (tot / sf->f_bsize); } else sf->f_blocks += tsf->f_blocks; } if (tsf->f_bfree > 0) { if (sf->f_bsize != tsf->f_bsize) { tot = tsf->f_bfree * tsf->f_bsize; sf->f_bfree += (tot / sf->f_bsize); } else sf->f_bfree += tsf->f_bfree; } if (tsf->f_bavail > 0) { if (sf->f_bsize != tsf->f_bsize) { tot = tsf->f_bavail * tsf->f_bsize; sf->f_bavail += (tot / sf->f_bsize); } else sf->f_bavail += tsf->f_bavail; } } } free(tsf, M_TEMP); free(dvpp, M_TEMP); return (error); } /* * Set an NFSv4 acl. */ int nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p) { int error; if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) { error = NFSERR_ATTRNOTSUPP; goto out; } /* * With NFSv4 ACLs, chmod(2) may need to add additional entries. * Make sure it has enough room for that - splitting every entry * into two and appending "canonical six" entries at the end. * Cribbed out of kern/vfs_acl.c - Rick M. */ if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) { error = NFSERR_ATTRNOTSUPP; goto out; } error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); if (error == 0) { error = nfsrv_dssetacl(vp, aclp, cred, p); if (error == ENOENT) error = 0; } out: NFSEXITCODE(error); return (error); } /* * Seek vnode op call (actually it is a VOP_IOCTL()). * This function is called with the vnode locked, but unlocks and vrele()s * the vp before returning. */ int nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd, off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p) { struct nfsvattr at; int error, ret; ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp"); /* * Attempt to seek on a DS file. A return of ENOENT implies * there is no DS file to seek on. */ error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL, NULL, NULL, NULL, NULL, offp, content, eofp); if (error != ENOENT) { vput(vp); return (error); } /* * Do the VOP_IOCTL() call. For the case where *offp == file_size, * VOP_IOCTL() will return ENXIO. However, the correct reply for * NFSv4.2 is *eofp == true and error == 0 for this case. */ NFSVOPUNLOCK(vp); error = VOP_IOCTL(vp, cmd, offp, 0, cred, p); *eofp = false; if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) { /* Handle the cases where we might be at EOF. */ ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL); if (ret == 0 && *offp == at.na_size) { *eofp = true; error = 0; } if (ret != 0 && error == 0) error = ret; } vrele(vp); NFSEXITCODE(error); return (error); } /* * Allocate vnode op call. */ int nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, NFSPROC_T *p) { int error, trycnt; ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp"); /* * Attempt to allocate on a DS file. A return of ENOENT implies * there is no DS file to allocate on. */ error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL, NULL, NULL, NULL, NULL, &len, 0, NULL); if (error != ENOENT) return (error); error = 0; /* * Do the actual VOP_ALLOCATE(), looping a reasonable number of * times to achieve completion. */ trycnt = 0; while (error == 0 && len > 0 && trycnt++ < 20) error = VOP_ALLOCATE(vp, &off, &len); if (error == 0 && len > 0) error = NFSERR_IO; NFSEXITCODE(error); return (error); } /* * Get Extended Atribute vnode op into an mbuf list. */ int nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp, struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp) { struct iovec *iv; struct uio io, *uiop = &io; struct mbuf *m, *m2; int alen, error, len, tlen; size_t siz; /* First, find out the size of the extended attribute. */ error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, &siz, cred, p); if (error != 0) return (NFSERR_NOXATTR); if (siz > maxresp - NFS_MAXXDR) return (NFSERR_XATTR2BIG); len = siz; tlen = NFSM_RNDUP(len); if (tlen > 0) { /* * If cnt > MCLBYTES and the reply will not be saved, use * ext_pgs mbufs for TLS. * For NFSv4.0, we do not know for sure if the reply will * be saved, so do not use ext_pgs mbufs for NFSv4.0. * Always use ext_pgs mbufs if ND_EXTPG is set. */ if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES && (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS && (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)) uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen, maxextsiz, &m, &m2, &iv); else uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2, &iv); uiop->uio_iov = iv; } else { uiop->uio_iovcnt = 0; uiop->uio_iov = iv = NULL; m = m2 = NULL; } uiop->uio_offset = 0; uiop->uio_resid = tlen; uiop->uio_rw = UIO_READ; uiop->uio_segflg = UIO_SYSSPACE; uiop->uio_td = p; #ifdef MAC error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER, name); if (error != 0) goto out; #endif if (tlen > 0) error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, NULL, cred, p); if (error != 0) goto out; if (uiop->uio_resid > 0) { alen = tlen; len = tlen - uiop->uio_resid; tlen = NFSM_RNDUP(len); if (alen != tlen) printf("nfsvno_getxattr: weird size read\n"); if (tlen == 0) { m_freem(m); m = m2 = NULL; } else if (alen != tlen || tlen != len) m2 = nfsrv_adj(m, alen - tlen, tlen - len); } *lenp = len; *mpp = m; *mpendp = m2; out: if (error != 0) { if (m != NULL) m_freem(m); *lenp = 0; } free(iv, M_TEMP); NFSEXITCODE(error); return (error); } /* * Set Extended attribute vnode op from an mbuf list. */ int nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m, char *cp, struct ucred *cred, struct thread *p) { struct iovec *iv; struct uio uio, *uiop = &uio; int cnt, error; error = 0; #ifdef MAC error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER, name); #endif if (error != 0) goto out; uiop->uio_rw = UIO_WRITE; uiop->uio_segflg = UIO_SYSSPACE; uiop->uio_td = p; uiop->uio_offset = 0; uiop->uio_resid = len; if (len > 0) { error = nfsrv_createiovecw(len, m, cp, &iv, &cnt); uiop->uio_iov = iv; uiop->uio_iovcnt = cnt; } else { uiop->uio_iov = iv = NULL; uiop->uio_iovcnt = 0; } if (error == 0) { error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, cred, p); free(iv, M_TEMP); } out: NFSEXITCODE(error); return (error); } /* * Remove Extended attribute vnode op. */ int nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name, struct ucred *cred, struct thread *p) { int error; /* * Get rid of any delegations. I am not sure why this is required, * but RFC-8276 says so. */ error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p); if (error != 0) goto out; #ifdef MAC error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER, name); if (error != 0) goto out; #endif error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p); if (error == EOPNOTSUPP) error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, cred, p); out: NFSEXITCODE(error); return (error); } /* * List Extended Atribute vnode op into an mbuf list. */ int nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred, struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp) { struct iovec iv; struct uio io; int error; size_t siz; *bufp = NULL; /* First, find out the size of the extended attribute. */ error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred, p); if (error != 0) return (NFSERR_NOXATTR); if (siz <= cookie) { *lenp = 0; *eofp = true; goto out; } if (siz > cookie + *lenp) { siz = cookie + *lenp; *eofp = false; } else *eofp = true; /* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */ if (siz > 10 * 1024 * 1024) { error = NFSERR_XATTR2BIG; goto out; } *bufp = malloc(siz, M_TEMP, M_WAITOK); iv.iov_base = *bufp; iv.iov_len = siz; io.uio_iovcnt = 1; io.uio_iov = &iv; io.uio_offset = 0; io.uio_resid = siz; io.uio_rw = UIO_READ; io.uio_segflg = UIO_SYSSPACE; io.uio_td = p; #ifdef MAC error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER); if (error != 0) goto out; #endif error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred, p); if (error != 0) goto out; if (io.uio_resid > 0) siz -= io.uio_resid; *lenp = siz; out: if (error != 0) { free(*bufp, M_TEMP); *bufp = NULL; } NFSEXITCODE(error); return (error); } /* * Trim trailing data off the mbuf list being built. */ void nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos, int bextpg, int bextpgsiz) { vm_page_t pg; int fullpgsiz, i; if (mb->m_next != NULL) { m_freem(mb->m_next); mb->m_next = NULL; } if ((mb->m_flags & M_EXTPG) != 0) { KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs, ("nfsm_trimtrailing: bextpg out of range")); KASSERT(bpos == (char *)(void *) PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz, ("nfsm_trimtrailing: bextpgsiz bad!")); /* First, get rid of any pages after this position. */ for (i = mb->m_epg_npgs - 1; i > bextpg; i--) { pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]); vm_page_unwire_noq(pg); vm_page_free(pg); } mb->m_epg_npgs = bextpg + 1; if (bextpg == 0) fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off; else fullpgsiz = PAGE_SIZE; mb->m_epg_last_len = fullpgsiz - bextpgsiz; mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off); for (i = 1; i < mb->m_epg_npgs; i++) mb->m_len += m_epg_pagelen(mb, i, 0); nd->nd_bextpgsiz = bextpgsiz; nd->nd_bextpg = bextpg; } else mb->m_len = bpos - mtod(mb, char *); nd->nd_mb = mb; nd->nd_bpos = bpos; } extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *); /* * Called once to initialize data structures... */ static int nfsd_modevent(module_t mod, int type, void *data) { int error = 0, i; static int loaded = 0; switch (type) { case MOD_LOAD: if (loaded) goto out; newnfs_portinit(); for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL, MTX_DEF); mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL, MTX_DEF); } mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF); mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF); mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF); mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF); mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF); lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0); nfsrvd_initcache(); nfsd_init(); NFSD_LOCK(); nfsrvd_init(0); NFSD_UNLOCK(); nfsd_mntinit(); #ifdef VV_DISABLEDELEG vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation; vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation; #endif nfsd_call_servertimer = nfsrv_servertimer; nfsd_call_nfsd = nfssvc_nfsd; loaded = 1; break; case MOD_UNLOAD: if (newnfs_numnfsd != 0) { error = EBUSY; break; } #ifdef VV_DISABLEDELEG vn_deleg_ops.vndeleg_recall = NULL; vn_deleg_ops.vndeleg_disable = NULL; #endif nfsd_call_servertimer = NULL; nfsd_call_nfsd = NULL; /* Clean out all NFSv4 state. */ nfsrv_throwawayallstate(curthread); /* Clean the NFS server reply cache */ nfsrvd_cleancache(); /* Free up the krpc server pool. */ if (nfsrvd_pool != NULL) svcpool_destroy(nfsrvd_pool); /* and get rid of the locks */ for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { mtx_destroy(&nfsrchash_table[i].mtx); mtx_destroy(&nfsrcahash_table[i].mtx); } mtx_destroy(&nfsrc_udpmtx); mtx_destroy(&nfs_v4root_mutex); mtx_destroy(&nfsv4root_mnt.mnt_mtx); mtx_destroy(&nfsrv_dontlistlock_mtx); mtx_destroy(&nfsrv_recalllock_mtx); for (i = 0; i < nfsrv_sessionhashsize; i++) mtx_destroy(&nfssessionhash[i].mtx); if (nfslayouthash != NULL) { for (i = 0; i < nfsrv_layouthashsize; i++) mtx_destroy(&nfslayouthash[i].mtx); free(nfslayouthash, M_NFSDSESSION); } lockdestroy(&nfsv4root_mnt.mnt_explock); free(nfsclienthash, M_NFSDCLIENT); free(nfslockhash, M_NFSDLOCKFILE); free(nfssessionhash, M_NFSDSESSION); loaded = 0; break; default: error = EOPNOTSUPP; break; } out: NFSEXITCODE(error); return (error); } static moduledata_t nfsd_mod = { "nfsd", nfsd_modevent, NULL, }; DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY); /* So that loader and kldload(2) can find us, wherever we are.. */ MODULE_VERSION(nfsd, 1); MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1); MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1); MODULE_DEPEND(nfsd, krpc, 1, 1, 1); MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1); diff --git a/sys/kern/kern_event.c b/sys/kern/kern_event.c index e22c987f8a08..5185723b8d10 100644 --- a/sys/kern/kern_event.c +++ b/sys/kern/kern_event.c @@ -1,2729 +1,2730 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 1999,2000,2001 Jonathan Lemon * Copyright 2004 John-Mark Gurney * Copyright (c) 2009 Apple, Inc. * 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_ktrace.h" #include "opt_kqueue.h" #ifdef COMPAT_FREEBSD11 #define _WANT_FREEBSD11_KEVENT #endif #include #include #include #include #include #include #include #include #include #include #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 KTRACE #include #endif #include #include static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); /* * This lock is used if multiple kq locks are required. This possibly * should be made into a per proc lock. */ static struct mtx kq_global; MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF); #define KQ_GLOBAL_LOCK(lck, haslck) do { \ if (!haslck) \ mtx_lock(lck); \ haslck = 1; \ } while (0) #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \ if (haslck) \ mtx_unlock(lck); \ haslck = 0; \ } while (0) TASKQUEUE_DEFINE_THREAD(kqueue_ctx); static int kevent_copyout(void *arg, struct kevent *kevp, int count); static int kevent_copyin(void *arg, struct kevent *kevp, int count); static int kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int mflag); static int kqueue_acquire(struct file *fp, struct kqueue **kqp); static void kqueue_release(struct kqueue *kq, int locked); static void kqueue_destroy(struct kqueue *kq); static void kqueue_drain(struct kqueue *kq, struct thread *td); static int kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident, int mflag); static void kqueue_task(void *arg, int pending); static int kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, const struct timespec *timeout, struct kevent *keva, struct thread *td); static void kqueue_wakeup(struct kqueue *kq); static struct filterops *kqueue_fo_find(int filt); static void kqueue_fo_release(int filt); struct g_kevent_args; static int kern_kevent_generic(struct thread *td, struct g_kevent_args *uap, struct kevent_copyops *k_ops, const char *struct_name); static fo_ioctl_t kqueue_ioctl; static fo_poll_t kqueue_poll; static fo_kqfilter_t kqueue_kqfilter; static fo_stat_t kqueue_stat; static fo_close_t kqueue_close; static fo_fill_kinfo_t kqueue_fill_kinfo; static struct fileops kqueueops = { .fo_read = invfo_rdwr, .fo_write = invfo_rdwr, .fo_truncate = invfo_truncate, .fo_ioctl = kqueue_ioctl, .fo_poll = kqueue_poll, .fo_kqfilter = kqueue_kqfilter, .fo_stat = kqueue_stat, .fo_close = kqueue_close, .fo_chmod = invfo_chmod, .fo_chown = invfo_chown, .fo_sendfile = invfo_sendfile, .fo_fill_kinfo = kqueue_fill_kinfo, }; static int knote_attach(struct knote *kn, struct kqueue *kq); static void knote_drop(struct knote *kn, struct thread *td); static void knote_drop_detached(struct knote *kn, struct thread *td); static void knote_enqueue(struct knote *kn); static void knote_dequeue(struct knote *kn); static void knote_init(void); static struct knote *knote_alloc(int mflag); static void knote_free(struct knote *kn); static void filt_kqdetach(struct knote *kn); static int filt_kqueue(struct knote *kn, long hint); static int filt_procattach(struct knote *kn); static void filt_procdetach(struct knote *kn); static int filt_proc(struct knote *kn, long hint); static int filt_fileattach(struct knote *kn); static void filt_timerexpire(void *knx); static int filt_timerattach(struct knote *kn); static void filt_timerdetach(struct knote *kn); static void filt_timerstart(struct knote *kn, sbintime_t to); static void filt_timertouch(struct knote *kn, struct kevent *kev, u_long type); static int filt_timervalidate(struct knote *kn, sbintime_t *to); static int filt_timer(struct knote *kn, long hint); static int filt_userattach(struct knote *kn); static void filt_userdetach(struct knote *kn); static int filt_user(struct knote *kn, long hint); static void filt_usertouch(struct knote *kn, struct kevent *kev, u_long type); static struct filterops file_filtops = { .f_isfd = 1, .f_attach = filt_fileattach, }; static struct filterops kqread_filtops = { .f_isfd = 1, .f_detach = filt_kqdetach, .f_event = filt_kqueue, }; /* XXX - move to kern_proc.c? */ static struct filterops proc_filtops = { .f_isfd = 0, .f_attach = filt_procattach, .f_detach = filt_procdetach, .f_event = filt_proc, }; static struct filterops timer_filtops = { .f_isfd = 0, .f_attach = filt_timerattach, .f_detach = filt_timerdetach, .f_event = filt_timer, .f_touch = filt_timertouch, }; static struct filterops user_filtops = { .f_attach = filt_userattach, .f_detach = filt_userdetach, .f_event = filt_user, .f_touch = filt_usertouch, }; static uma_zone_t knote_zone; static unsigned int kq_ncallouts = 0; static unsigned int kq_calloutmax = 4 * 1024; SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); /* XXX - ensure not influx ? */ #define KNOTE_ACTIVATE(kn, islock) do { \ if ((islock)) \ mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \ else \ KQ_LOCK((kn)->kn_kq); \ (kn)->kn_status |= KN_ACTIVE; \ if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ knote_enqueue((kn)); \ if (!(islock)) \ KQ_UNLOCK((kn)->kn_kq); \ } while(0) #define KQ_LOCK(kq) do { \ mtx_lock(&(kq)->kq_lock); \ } while (0) #define KQ_FLUX_WAKEUP(kq) do { \ if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \ (kq)->kq_state &= ~KQ_FLUXWAIT; \ wakeup((kq)); \ } \ } while (0) #define KQ_UNLOCK_FLUX(kq) do { \ KQ_FLUX_WAKEUP(kq); \ mtx_unlock(&(kq)->kq_lock); \ } while (0) #define KQ_UNLOCK(kq) do { \ mtx_unlock(&(kq)->kq_lock); \ } while (0) #define KQ_OWNED(kq) do { \ mtx_assert(&(kq)->kq_lock, MA_OWNED); \ } while (0) #define KQ_NOTOWNED(kq) do { \ mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \ } while (0) static struct knlist * kn_list_lock(struct knote *kn) { struct knlist *knl; knl = kn->kn_knlist; if (knl != NULL) knl->kl_lock(knl->kl_lockarg); return (knl); } static void kn_list_unlock(struct knlist *knl) { bool do_free; if (knl == NULL) return; do_free = knl->kl_autodestroy && knlist_empty(knl); knl->kl_unlock(knl->kl_lockarg); if (do_free) { knlist_destroy(knl); free(knl, M_KQUEUE); } } static bool kn_in_flux(struct knote *kn) { return (kn->kn_influx > 0); } static void kn_enter_flux(struct knote *kn) { KQ_OWNED(kn->kn_kq); MPASS(kn->kn_influx < INT_MAX); kn->kn_influx++; } static bool kn_leave_flux(struct knote *kn) { KQ_OWNED(kn->kn_kq); MPASS(kn->kn_influx > 0); kn->kn_influx--; return (kn->kn_influx == 0); } #define KNL_ASSERT_LOCK(knl, islocked) do { \ if (islocked) \ KNL_ASSERT_LOCKED(knl); \ else \ KNL_ASSERT_UNLOCKED(knl); \ } while (0) #ifdef INVARIANTS #define KNL_ASSERT_LOCKED(knl) do { \ knl->kl_assert_lock((knl)->kl_lockarg, LA_LOCKED); \ } while (0) #define KNL_ASSERT_UNLOCKED(knl) do { \ knl->kl_assert_lock((knl)->kl_lockarg, LA_UNLOCKED); \ } while (0) #else /* !INVARIANTS */ #define KNL_ASSERT_LOCKED(knl) do {} while(0) #define KNL_ASSERT_UNLOCKED(knl) do {} while (0) #endif /* INVARIANTS */ #ifndef KN_HASHSIZE #define KN_HASHSIZE 64 /* XXX should be tunable */ #endif #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) static int filt_nullattach(struct knote *kn) { return (ENXIO); }; struct filterops null_filtops = { .f_isfd = 0, .f_attach = filt_nullattach, }; /* XXX - make SYSINIT to add these, and move into respective modules. */ extern struct filterops sig_filtops; extern struct filterops fs_filtops; /* * Table for for all system-defined filters. */ static struct mtx filterops_lock; MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", MTX_DEF); static struct { struct filterops *for_fop; int for_nolock; int for_refcnt; } sysfilt_ops[EVFILT_SYSCOUNT] = { { &file_filtops, 1 }, /* EVFILT_READ */ { &file_filtops, 1 }, /* EVFILT_WRITE */ { &null_filtops }, /* EVFILT_AIO */ { &file_filtops, 1 }, /* EVFILT_VNODE */ { &proc_filtops, 1 }, /* EVFILT_PROC */ { &sig_filtops, 1 }, /* EVFILT_SIGNAL */ { &timer_filtops, 1 }, /* EVFILT_TIMER */ { &file_filtops, 1 }, /* EVFILT_PROCDESC */ { &fs_filtops, 1 }, /* EVFILT_FS */ { &null_filtops }, /* EVFILT_LIO */ { &user_filtops, 1 }, /* EVFILT_USER */ { &null_filtops }, /* EVFILT_SENDFILE */ { &file_filtops, 1 }, /* EVFILT_EMPTY */ }; /* * Simple redirection for all cdevsw style objects to call their fo_kqfilter * method. */ static int filt_fileattach(struct knote *kn) { return (fo_kqfilter(kn->kn_fp, kn)); } /*ARGSUSED*/ static int kqueue_kqfilter(struct file *fp, struct knote *kn) { struct kqueue *kq = kn->kn_fp->f_data; if (kn->kn_filter != EVFILT_READ) return (EINVAL); kn->kn_status |= KN_KQUEUE; kn->kn_fop = &kqread_filtops; knlist_add(&kq->kq_sel.si_note, kn, 0); return (0); } static void filt_kqdetach(struct knote *kn) { struct kqueue *kq = kn->kn_fp->f_data; knlist_remove(&kq->kq_sel.si_note, kn, 0); } /*ARGSUSED*/ static int filt_kqueue(struct knote *kn, long hint) { struct kqueue *kq = kn->kn_fp->f_data; kn->kn_data = kq->kq_count; return (kn->kn_data > 0); } /* XXX - move to kern_proc.c? */ static int filt_procattach(struct knote *kn) { struct proc *p; int error; bool exiting, immediate; exiting = immediate = false; if (kn->kn_sfflags & NOTE_EXIT) p = pfind_any(kn->kn_id); else p = pfind(kn->kn_id); if (p == NULL) return (ESRCH); if (p->p_flag & P_WEXIT) exiting = true; if ((error = p_cansee(curthread, p))) { PROC_UNLOCK(p); return (error); } kn->kn_ptr.p_proc = p; kn->kn_flags |= EV_CLEAR; /* automatically set */ /* * Internal flag indicating registration done by kernel for the * purposes of getting a NOTE_CHILD notification. */ if (kn->kn_flags & EV_FLAG2) { kn->kn_flags &= ~EV_FLAG2; kn->kn_data = kn->kn_sdata; /* ppid */ kn->kn_fflags = NOTE_CHILD; kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK); immediate = true; /* Force immediate activation of child note. */ } /* * Internal flag indicating registration done by kernel (for other than * NOTE_CHILD). */ if (kn->kn_flags & EV_FLAG1) { kn->kn_flags &= ~EV_FLAG1; } knlist_add(p->p_klist, kn, 1); /* * Immediately activate any child notes or, in the case of a zombie * target process, exit notes. The latter is necessary to handle the * case where the target process, e.g. a child, dies before the kevent * is registered. */ if (immediate || (exiting && filt_proc(kn, NOTE_EXIT))) KNOTE_ACTIVATE(kn, 0); PROC_UNLOCK(p); return (0); } /* * The knote may be attached to a different process, which may exit, * leaving nothing for the knote to be attached to. So when the process * exits, the knote is marked as DETACHED and also flagged as ONESHOT so * it will be deleted when read out. However, as part of the knote deletion, * this routine is called, so a check is needed to avoid actually performing * a detach, because the original process does not exist any more. */ /* XXX - move to kern_proc.c? */ static void filt_procdetach(struct knote *kn) { knlist_remove(kn->kn_knlist, kn, 0); kn->kn_ptr.p_proc = NULL; } /* XXX - move to kern_proc.c? */ static int filt_proc(struct knote *kn, long hint) { struct proc *p; u_int event; p = kn->kn_ptr.p_proc; if (p == NULL) /* already activated, from attach filter */ return (0); /* Mask off extra data. */ event = (u_int)hint & NOTE_PCTRLMASK; /* If the user is interested in this event, record it. */ if (kn->kn_sfflags & event) kn->kn_fflags |= event; /* Process is gone, so flag the event as finished. */ if (event == NOTE_EXIT) { kn->kn_flags |= EV_EOF | EV_ONESHOT; kn->kn_ptr.p_proc = NULL; if (kn->kn_fflags & NOTE_EXIT) kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig); if (kn->kn_fflags == 0) kn->kn_flags |= EV_DROP; return (1); } return (kn->kn_fflags != 0); } /* * Called when the process forked. It mostly does the same as the * knote(), activating all knotes registered to be activated when the * process forked. Additionally, for each knote attached to the * parent, check whether user wants to track the new process. If so * attach a new knote to it, and immediately report an event with the * child's pid. */ void knote_fork(struct knlist *list, int pid) { struct kqueue *kq; struct knote *kn; struct kevent kev; int error; MPASS(list != NULL); KNL_ASSERT_LOCKED(list); if (SLIST_EMPTY(&list->kl_list)) return; memset(&kev, 0, sizeof(kev)); SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { kq = kn->kn_kq; KQ_LOCK(kq); if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) { KQ_UNLOCK(kq); continue; } /* * The same as knote(), activate the event. */ if ((kn->kn_sfflags & NOTE_TRACK) == 0) { if (kn->kn_fop->f_event(kn, NOTE_FORK)) KNOTE_ACTIVATE(kn, 1); KQ_UNLOCK(kq); continue; } /* * The NOTE_TRACK case. In addition to the activation * of the event, we need to register new events to * track the child. Drop the locks in preparation for * the call to kqueue_register(). */ kn_enter_flux(kn); KQ_UNLOCK(kq); list->kl_unlock(list->kl_lockarg); /* * Activate existing knote and register tracking knotes with * new process. * * First register a knote to get just the child notice. This * must be a separate note from a potential NOTE_EXIT * notification since both NOTE_CHILD and NOTE_EXIT are defined * to use the data field (in conflicting ways). */ kev.ident = pid; kev.filter = kn->kn_filter; kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT | EV_FLAG2; kev.fflags = kn->kn_sfflags; kev.data = kn->kn_id; /* parent */ kev.udata = kn->kn_kevent.udata;/* preserve udata */ error = kqueue_register(kq, &kev, NULL, M_NOWAIT); if (error) kn->kn_fflags |= NOTE_TRACKERR; /* * Then register another knote to track other potential events * from the new process. */ kev.ident = pid; kev.filter = kn->kn_filter; kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; kev.fflags = kn->kn_sfflags; kev.data = kn->kn_id; /* parent */ kev.udata = kn->kn_kevent.udata;/* preserve udata */ error = kqueue_register(kq, &kev, NULL, M_NOWAIT); if (error) kn->kn_fflags |= NOTE_TRACKERR; if (kn->kn_fop->f_event(kn, NOTE_FORK)) KNOTE_ACTIVATE(kn, 0); list->kl_lock(list->kl_lockarg); KQ_LOCK(kq); kn_leave_flux(kn); KQ_UNLOCK_FLUX(kq); } } /* * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the * interval timer support code. */ #define NOTE_TIMER_PRECMASK \ (NOTE_SECONDS | NOTE_MSECONDS | NOTE_USECONDS | NOTE_NSECONDS) static sbintime_t timer2sbintime(int64_t data, int flags) { int64_t secs; /* * Macros for converting to the fractional second portion of an * sbintime_t using 64bit multiplication to improve precision. */ #define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32) #define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32) #define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32) switch (flags & NOTE_TIMER_PRECMASK) { case NOTE_SECONDS: #ifdef __LP64__ if (data > (SBT_MAX / SBT_1S)) return (SBT_MAX); #endif return ((sbintime_t)data << 32); case NOTE_MSECONDS: /* FALLTHROUGH */ case 0: if (data >= 1000) { secs = data / 1000; #ifdef __LP64__ if (secs > (SBT_MAX / SBT_1S)) return (SBT_MAX); #endif return (secs << 32 | MS_TO_SBT(data % 1000)); } return (MS_TO_SBT(data)); case NOTE_USECONDS: if (data >= 1000000) { secs = data / 1000000; #ifdef __LP64__ if (secs > (SBT_MAX / SBT_1S)) return (SBT_MAX); #endif return (secs << 32 | US_TO_SBT(data % 1000000)); } return (US_TO_SBT(data)); case NOTE_NSECONDS: if (data >= 1000000000) { secs = data / 1000000000; #ifdef __LP64__ if (secs > (SBT_MAX / SBT_1S)) return (SBT_MAX); #endif return (secs << 32 | NS_TO_SBT(data % 1000000000)); } return (NS_TO_SBT(data)); default: break; } return (-1); } struct kq_timer_cb_data { struct callout c; sbintime_t next; /* next timer event fires at */ sbintime_t to; /* precalculated timer period, 0 for abs */ }; static void filt_timerexpire(void *knx) { struct knote *kn; struct kq_timer_cb_data *kc; kn = knx; kn->kn_data++; KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */ if ((kn->kn_flags & EV_ONESHOT) != 0) return; kc = kn->kn_ptr.p_v; if (kc->to == 0) return; kc->next += kc->to; callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kn, PCPU_GET(cpuid), C_ABSOLUTE); } /* * data contains amount of time to sleep */ static int filt_timervalidate(struct knote *kn, sbintime_t *to) { struct bintime bt; sbintime_t sbt; if (kn->kn_sdata < 0) return (EINVAL); if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0) kn->kn_sdata = 1; /* * The only fflags values supported are the timer unit * (precision) and the absolute time indicator. */ if ((kn->kn_sfflags & ~(NOTE_TIMER_PRECMASK | NOTE_ABSTIME)) != 0) return (EINVAL); *to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags); if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) { getboottimebin(&bt); sbt = bttosbt(bt); *to -= sbt; } if (*to < 0) return (EINVAL); return (0); } static int filt_timerattach(struct knote *kn) { struct kq_timer_cb_data *kc; sbintime_t to; unsigned int ncallouts; int error; error = filt_timervalidate(kn, &to); if (error != 0) return (error); do { ncallouts = kq_ncallouts; if (ncallouts >= kq_calloutmax) return (ENOMEM); } while (!atomic_cmpset_int(&kq_ncallouts, ncallouts, ncallouts + 1)); if ((kn->kn_sfflags & NOTE_ABSTIME) == 0) kn->kn_flags |= EV_CLEAR; /* automatically set */ kn->kn_status &= ~KN_DETACHED; /* knlist_add clears it */ kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK); callout_init(&kc->c, 1); filt_timerstart(kn, to); return (0); } static void filt_timerstart(struct knote *kn, sbintime_t to) { struct kq_timer_cb_data *kc; kc = kn->kn_ptr.p_v; if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) { kc->next = to; kc->to = 0; } else { kc->next = to + sbinuptime(); kc->to = to; } callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kn, PCPU_GET(cpuid), C_ABSOLUTE); } static void filt_timerdetach(struct knote *kn) { struct kq_timer_cb_data *kc; unsigned int old __unused; kc = kn->kn_ptr.p_v; callout_drain(&kc->c); free(kc, M_KQUEUE); old = atomic_fetchadd_int(&kq_ncallouts, -1); KASSERT(old > 0, ("Number of callouts cannot become negative")); kn->kn_status |= KN_DETACHED; /* knlist_remove sets it */ } static void filt_timertouch(struct knote *kn, struct kevent *kev, u_long type) { struct kq_timer_cb_data *kc; struct kqueue *kq; sbintime_t to; int error; switch (type) { case EVENT_REGISTER: /* Handle re-added timers that update data/fflags */ if (kev->flags & EV_ADD) { kc = kn->kn_ptr.p_v; /* Drain any existing callout. */ callout_drain(&kc->c); /* Throw away any existing undelivered record * of the timer expiration. This is done under * the presumption that if a process is * re-adding this timer with new parameters, * it is no longer interested in what may have * happened under the old parameters. If it is * interested, it can wait for the expiration, * delete the old timer definition, and then * add the new one. * * This has to be done while the kq is locked: * - if enqueued, dequeue * - make it no longer active * - clear the count of expiration events */ kq = kn->kn_kq; KQ_LOCK(kq); if (kn->kn_status & KN_QUEUED) knote_dequeue(kn); kn->kn_status &= ~KN_ACTIVE; kn->kn_data = 0; KQ_UNLOCK(kq); /* Reschedule timer based on new data/fflags */ kn->kn_sfflags = kev->fflags; kn->kn_sdata = kev->data; error = filt_timervalidate(kn, &to); if (error != 0) { kn->kn_flags |= EV_ERROR; kn->kn_data = error; } else filt_timerstart(kn, to); } break; case EVENT_PROCESS: *kev = kn->kn_kevent; if (kn->kn_flags & EV_CLEAR) { kn->kn_data = 0; kn->kn_fflags = 0; } break; default: panic("filt_timertouch() - invalid type (%ld)", type); break; } } static int filt_timer(struct knote *kn, long hint) { return (kn->kn_data != 0); } static int filt_userattach(struct knote *kn) { /* * EVFILT_USER knotes are not attached to anything in the kernel. */ kn->kn_hook = NULL; if (kn->kn_fflags & NOTE_TRIGGER) kn->kn_hookid = 1; else kn->kn_hookid = 0; return (0); } static void filt_userdetach(__unused struct knote *kn) { /* * EVFILT_USER knotes are not attached to anything in the kernel. */ } static int filt_user(struct knote *kn, __unused long hint) { return (kn->kn_hookid); } static void filt_usertouch(struct knote *kn, struct kevent *kev, u_long type) { u_int ffctrl; switch (type) { case EVENT_REGISTER: if (kev->fflags & NOTE_TRIGGER) kn->kn_hookid = 1; ffctrl = kev->fflags & NOTE_FFCTRLMASK; kev->fflags &= NOTE_FFLAGSMASK; switch (ffctrl) { case NOTE_FFNOP: break; case NOTE_FFAND: kn->kn_sfflags &= kev->fflags; break; case NOTE_FFOR: kn->kn_sfflags |= kev->fflags; break; case NOTE_FFCOPY: kn->kn_sfflags = kev->fflags; break; default: /* XXX Return error? */ break; } kn->kn_sdata = kev->data; if (kev->flags & EV_CLEAR) { kn->kn_hookid = 0; kn->kn_data = 0; kn->kn_fflags = 0; } break; case EVENT_PROCESS: *kev = kn->kn_kevent; kev->fflags = kn->kn_sfflags; kev->data = kn->kn_sdata; if (kn->kn_flags & EV_CLEAR) { kn->kn_hookid = 0; kn->kn_data = 0; kn->kn_fflags = 0; } break; default: panic("filt_usertouch() - invalid type (%ld)", type); break; } } int sys_kqueue(struct thread *td, struct kqueue_args *uap) { return (kern_kqueue(td, 0, NULL)); } static void kqueue_init(struct kqueue *kq) { mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK); TAILQ_INIT(&kq->kq_head); knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock); TASK_INIT(&kq->kq_task, 0, kqueue_task, kq); } int kern_kqueue(struct thread *td, int flags, struct filecaps *fcaps) { struct filedesc *fdp; struct kqueue *kq; struct file *fp; struct ucred *cred; int fd, error; fdp = td->td_proc->p_fd; cred = td->td_ucred; if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES))) return (ENOMEM); error = falloc_caps(td, &fp, &fd, flags, fcaps); if (error != 0) { chgkqcnt(cred->cr_ruidinfo, -1, 0); return (error); } /* An extra reference on `fp' has been held for us by falloc(). */ kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO); kqueue_init(kq); kq->kq_fdp = fdp; kq->kq_cred = crhold(cred); FILEDESC_XLOCK(fdp); TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list); FILEDESC_XUNLOCK(fdp); finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops); fdrop(fp, td); td->td_retval[0] = fd; return (0); } struct g_kevent_args { int fd; void *changelist; int nchanges; void *eventlist; int nevents; const struct timespec *timeout; }; int sys_kevent(struct thread *td, struct kevent_args *uap) { struct kevent_copyops k_ops = { .arg = uap, .k_copyout = kevent_copyout, .k_copyin = kevent_copyin, .kevent_size = sizeof(struct kevent), }; struct g_kevent_args gk_args = { .fd = uap->fd, .changelist = uap->changelist, .nchanges = uap->nchanges, .eventlist = uap->eventlist, .nevents = uap->nevents, .timeout = uap->timeout, }; return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent")); } static int kern_kevent_generic(struct thread *td, struct g_kevent_args *uap, struct kevent_copyops *k_ops, const char *struct_name) { struct timespec ts, *tsp; #ifdef KTRACE struct kevent *eventlist = uap->eventlist; #endif int error; if (uap->timeout != NULL) { error = copyin(uap->timeout, &ts, sizeof(ts)); if (error) return (error); tsp = &ts; } else tsp = NULL; #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT_ARRAY)) ktrstructarray(struct_name, UIO_USERSPACE, uap->changelist, uap->nchanges, k_ops->kevent_size); #endif error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents, k_ops, tsp); #ifdef KTRACE if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY)) ktrstructarray(struct_name, UIO_USERSPACE, eventlist, td->td_retval[0], k_ops->kevent_size); #endif return (error); } /* * Copy 'count' items into the destination list pointed to by uap->eventlist. */ static int kevent_copyout(void *arg, struct kevent *kevp, int count) { struct kevent_args *uap; int error; KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); uap = (struct kevent_args *)arg; error = copyout(kevp, uap->eventlist, count * sizeof *kevp); if (error == 0) uap->eventlist += count; return (error); } /* * Copy 'count' items from the list pointed to by uap->changelist. */ static int kevent_copyin(void *arg, struct kevent *kevp, int count) { struct kevent_args *uap; int error; KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); uap = (struct kevent_args *)arg; error = copyin(uap->changelist, kevp, count * sizeof *kevp); if (error == 0) uap->changelist += count; return (error); } #ifdef COMPAT_FREEBSD11 static int kevent11_copyout(void *arg, struct kevent *kevp, int count) { struct freebsd11_kevent_args *uap; struct kevent_freebsd11 kev11; int error, i; KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); uap = (struct freebsd11_kevent_args *)arg; for (i = 0; i < count; i++) { kev11.ident = kevp->ident; kev11.filter = kevp->filter; kev11.flags = kevp->flags; kev11.fflags = kevp->fflags; kev11.data = kevp->data; kev11.udata = kevp->udata; error = copyout(&kev11, uap->eventlist, sizeof(kev11)); if (error != 0) break; uap->eventlist++; kevp++; } return (error); } /* * Copy 'count' items from the list pointed to by uap->changelist. */ static int kevent11_copyin(void *arg, struct kevent *kevp, int count) { struct freebsd11_kevent_args *uap; struct kevent_freebsd11 kev11; int error, i; KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); uap = (struct freebsd11_kevent_args *)arg; for (i = 0; i < count; i++) { error = copyin(uap->changelist, &kev11, sizeof(kev11)); if (error != 0) break; kevp->ident = kev11.ident; kevp->filter = kev11.filter; kevp->flags = kev11.flags; kevp->fflags = kev11.fflags; kevp->data = (uintptr_t)kev11.data; kevp->udata = kev11.udata; bzero(&kevp->ext, sizeof(kevp->ext)); uap->changelist++; kevp++; } return (error); } int freebsd11_kevent(struct thread *td, struct freebsd11_kevent_args *uap) { struct kevent_copyops k_ops = { .arg = uap, .k_copyout = kevent11_copyout, .k_copyin = kevent11_copyin, .kevent_size = sizeof(struct kevent_freebsd11), }; struct g_kevent_args gk_args = { .fd = uap->fd, .changelist = uap->changelist, .nchanges = uap->nchanges, .eventlist = uap->eventlist, .nevents = uap->nevents, .timeout = uap->timeout, }; return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent_freebsd11")); } #endif int kern_kevent(struct thread *td, int fd, int nchanges, int nevents, struct kevent_copyops *k_ops, const struct timespec *timeout) { cap_rights_t rights; struct file *fp; int error; cap_rights_init_zero(&rights); if (nchanges > 0) cap_rights_set_one(&rights, CAP_KQUEUE_CHANGE); if (nevents > 0) cap_rights_set_one(&rights, CAP_KQUEUE_EVENT); error = fget(td, fd, &rights, &fp); if (error != 0) return (error); error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout); fdrop(fp, td); return (error); } static int kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents, struct kevent_copyops *k_ops, const struct timespec *timeout) { struct kevent keva[KQ_NEVENTS]; struct kevent *kevp, *changes; int i, n, nerrors, error; nerrors = 0; while (nchanges > 0) { n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges; error = k_ops->k_copyin(k_ops->arg, keva, n); if (error) return (error); changes = keva; for (i = 0; i < n; i++) { kevp = &changes[i]; if (!kevp->filter) continue; kevp->flags &= ~EV_SYSFLAGS; error = kqueue_register(kq, kevp, td, M_WAITOK); if (error || (kevp->flags & EV_RECEIPT)) { if (nevents == 0) return (error); kevp->flags = EV_ERROR; kevp->data = error; (void)k_ops->k_copyout(k_ops->arg, kevp, 1); nevents--; nerrors++; } } nchanges -= n; } if (nerrors) { td->td_retval[0] = nerrors; return (0); } return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td)); } int kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents, struct kevent_copyops *k_ops, const struct timespec *timeout) { struct kqueue *kq; int error; error = kqueue_acquire(fp, &kq); if (error != 0) return (error); error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout); kqueue_release(kq, 0); return (error); } /* * Performs a kevent() call on a temporarily created kqueue. This can be * used to perform one-shot polling, similar to poll() and select(). */ int kern_kevent_anonymous(struct thread *td, int nevents, struct kevent_copyops *k_ops) { struct kqueue kq = {}; int error; kqueue_init(&kq); kq.kq_refcnt = 1; error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL); kqueue_drain(&kq, td); kqueue_destroy(&kq); return (error); } int kqueue_add_filteropts(int filt, struct filterops *filtops) { int error; error = 0; if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) { printf( "trying to add a filterop that is out of range: %d is beyond %d\n", ~filt, EVFILT_SYSCOUNT); return EINVAL; } mtx_lock(&filterops_lock); if (sysfilt_ops[~filt].for_fop != &null_filtops && sysfilt_ops[~filt].for_fop != NULL) error = EEXIST; else { sysfilt_ops[~filt].for_fop = filtops; sysfilt_ops[~filt].for_refcnt = 0; } mtx_unlock(&filterops_lock); return (error); } int kqueue_del_filteropts(int filt) { int error; error = 0; if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) return EINVAL; mtx_lock(&filterops_lock); if (sysfilt_ops[~filt].for_fop == &null_filtops || sysfilt_ops[~filt].for_fop == NULL) error = EINVAL; else if (sysfilt_ops[~filt].for_refcnt != 0) error = EBUSY; else { sysfilt_ops[~filt].for_fop = &null_filtops; sysfilt_ops[~filt].for_refcnt = 0; } mtx_unlock(&filterops_lock); return error; } static struct filterops * kqueue_fo_find(int filt) { if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) return NULL; if (sysfilt_ops[~filt].for_nolock) return sysfilt_ops[~filt].for_fop; mtx_lock(&filterops_lock); sysfilt_ops[~filt].for_refcnt++; if (sysfilt_ops[~filt].for_fop == NULL) sysfilt_ops[~filt].for_fop = &null_filtops; mtx_unlock(&filterops_lock); return sysfilt_ops[~filt].for_fop; } static void kqueue_fo_release(int filt) { if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) return; if (sysfilt_ops[~filt].for_nolock) return; mtx_lock(&filterops_lock); KASSERT(sysfilt_ops[~filt].for_refcnt > 0, ("filter object refcount not valid on release")); sysfilt_ops[~filt].for_refcnt--; mtx_unlock(&filterops_lock); } /* * A ref to kq (obtained via kqueue_acquire) must be held. */ static int kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int mflag) { struct filterops *fops; struct file *fp; struct knote *kn, *tkn; struct knlist *knl; int error, filt, event; int haskqglobal, filedesc_unlock; if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE)) return (EINVAL); fp = NULL; kn = NULL; knl = NULL; error = 0; haskqglobal = 0; filedesc_unlock = 0; filt = kev->filter; fops = kqueue_fo_find(filt); if (fops == NULL) return EINVAL; if (kev->flags & EV_ADD) { /* * Prevent waiting with locks. Non-sleepable * allocation failures are handled in the loop, only * if the spare knote appears to be actually required. */ tkn = knote_alloc(mflag); } else { tkn = NULL; } findkn: if (fops->f_isfd) { KASSERT(td != NULL, ("td is NULL")); if (kev->ident > INT_MAX) error = EBADF; else error = fget(td, kev->ident, &cap_event_rights, &fp); if (error) goto done; if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops, kev->ident, M_NOWAIT) != 0) { /* try again */ fdrop(fp, td); fp = NULL; error = kqueue_expand(kq, fops, kev->ident, mflag); if (error) goto done; goto findkn; } if (fp->f_type == DTYPE_KQUEUE) { /* * If we add some intelligence about what we are doing, * we should be able to support events on ourselves. * We need to know when we are doing this to prevent * getting both the knlist lock and the kq lock since * they are the same thing. */ if (fp->f_data == kq) { error = EINVAL; goto done; } /* * Pre-lock the filedesc before the global * lock mutex, see the comment in * kqueue_close(). */ FILEDESC_XLOCK(td->td_proc->p_fd); filedesc_unlock = 1; KQ_GLOBAL_LOCK(&kq_global, haskqglobal); } KQ_LOCK(kq); if (kev->ident < kq->kq_knlistsize) { SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link) if (kev->filter == kn->kn_filter) break; } } else { if ((kev->flags & EV_ADD) == EV_ADD) { error = kqueue_expand(kq, fops, kev->ident, mflag); if (error != 0) goto done; } KQ_LOCK(kq); /* * If possible, find an existing knote to use for this kevent. */ if (kev->filter == EVFILT_PROC && (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) { /* This is an internal creation of a process tracking * note. Don't attempt to coalesce this with an * existing note. */ ; } else if (kq->kq_knhashmask != 0) { struct klist *list; list = &kq->kq_knhash[ KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; SLIST_FOREACH(kn, list, kn_link) if (kev->ident == kn->kn_id && kev->filter == kn->kn_filter) break; } } /* knote is in the process of changing, wait for it to stabilize. */ if (kn != NULL && kn_in_flux(kn)) { KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); if (filedesc_unlock) { FILEDESC_XUNLOCK(td->td_proc->p_fd); filedesc_unlock = 0; } kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0); if (fp != NULL) { fdrop(fp, td); fp = NULL; } goto findkn; } /* * kn now contains the matching knote, or NULL if no match */ if (kn == NULL) { if (kev->flags & EV_ADD) { kn = tkn; tkn = NULL; if (kn == NULL) { KQ_UNLOCK(kq); error = ENOMEM; goto done; } kn->kn_fp = fp; kn->kn_kq = kq; kn->kn_fop = fops; /* * apply reference counts to knote structure, and * do not release it at the end of this routine. */ fops = NULL; fp = NULL; kn->kn_sfflags = kev->fflags; kn->kn_sdata = kev->data; kev->fflags = 0; kev->data = 0; kn->kn_kevent = *kev; kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE | EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT); kn->kn_status = KN_DETACHED; if ((kev->flags & EV_DISABLE) != 0) kn->kn_status |= KN_DISABLED; kn_enter_flux(kn); error = knote_attach(kn, kq); KQ_UNLOCK(kq); if (error != 0) { tkn = kn; goto done; } if ((error = kn->kn_fop->f_attach(kn)) != 0) { knote_drop_detached(kn, td); goto done; } knl = kn_list_lock(kn); goto done_ev_add; } else { /* No matching knote and the EV_ADD flag is not set. */ KQ_UNLOCK(kq); error = ENOENT; goto done; } } if (kev->flags & EV_DELETE) { kn_enter_flux(kn); KQ_UNLOCK(kq); knote_drop(kn, td); goto done; } if (kev->flags & EV_FORCEONESHOT) { kn->kn_flags |= EV_ONESHOT; KNOTE_ACTIVATE(kn, 1); } if ((kev->flags & EV_ENABLE) != 0) kn->kn_status &= ~KN_DISABLED; else if ((kev->flags & EV_DISABLE) != 0) kn->kn_status |= KN_DISABLED; /* * The user may change some filter values after the initial EV_ADD, * but doing so will not reset any filter which has already been * triggered. */ kn->kn_status |= KN_SCAN; kn_enter_flux(kn); KQ_UNLOCK(kq); knl = kn_list_lock(kn); kn->kn_kevent.udata = kev->udata; if (!fops->f_isfd && fops->f_touch != NULL) { fops->f_touch(kn, kev, EVENT_REGISTER); } else { kn->kn_sfflags = kev->fflags; kn->kn_sdata = kev->data; } done_ev_add: /* * We can get here with kn->kn_knlist == NULL. This can happen when * the initial attach event decides that the event is "completed" * already, e.g., filt_procattach() is called on a zombie process. It * will call filt_proc() which will remove it from the list, and NULL * kn_knlist. * * KN_DISABLED will be stable while the knote is in flux, so the * unlocked read will not race with an update. */ if ((kn->kn_status & KN_DISABLED) == 0) event = kn->kn_fop->f_event(kn, 0); else event = 0; KQ_LOCK(kq); if (event) kn->kn_status |= KN_ACTIVE; if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) == KN_ACTIVE) knote_enqueue(kn); kn->kn_status &= ~KN_SCAN; kn_leave_flux(kn); kn_list_unlock(knl); KQ_UNLOCK_FLUX(kq); done: KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); if (filedesc_unlock) FILEDESC_XUNLOCK(td->td_proc->p_fd); if (fp != NULL) fdrop(fp, td); knote_free(tkn); if (fops != NULL) kqueue_fo_release(filt); return (error); } static int kqueue_acquire(struct file *fp, struct kqueue **kqp) { int error; struct kqueue *kq; error = 0; kq = fp->f_data; if (fp->f_type != DTYPE_KQUEUE || kq == NULL) return (EBADF); *kqp = kq; KQ_LOCK(kq); if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) { KQ_UNLOCK(kq); return (EBADF); } kq->kq_refcnt++; KQ_UNLOCK(kq); return error; } static void kqueue_release(struct kqueue *kq, int locked) { if (locked) KQ_OWNED(kq); else KQ_LOCK(kq); kq->kq_refcnt--; if (kq->kq_refcnt == 1) wakeup(&kq->kq_refcnt); if (!locked) KQ_UNLOCK(kq); } static void kqueue_schedtask(struct kqueue *kq) { KQ_OWNED(kq); KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN), ("scheduling kqueue task while draining")); if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) { taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task); kq->kq_state |= KQ_TASKSCHED; } } /* * Expand the kq to make sure we have storage for fops/ident pair. * * Return 0 on success (or no work necessary), return errno on failure. */ static int kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident, int mflag) { struct klist *list, *tmp_knhash, *to_free; u_long tmp_knhashmask; int error, fd, size; KQ_NOTOWNED(kq); error = 0; to_free = NULL; if (fops->f_isfd) { fd = ident; if (kq->kq_knlistsize <= fd) { size = kq->kq_knlistsize; while (size <= fd) size += KQEXTENT; list = malloc(size * sizeof(*list), M_KQUEUE, mflag); if (list == NULL) return ENOMEM; KQ_LOCK(kq); if ((kq->kq_state & KQ_CLOSING) != 0) { to_free = list; error = EBADF; } else if (kq->kq_knlistsize > fd) { to_free = list; } else { if (kq->kq_knlist != NULL) { bcopy(kq->kq_knlist, list, kq->kq_knlistsize * sizeof(*list)); to_free = kq->kq_knlist; kq->kq_knlist = NULL; } bzero((caddr_t)list + kq->kq_knlistsize * sizeof(*list), (size - kq->kq_knlistsize) * sizeof(*list)); kq->kq_knlistsize = size; kq->kq_knlist = list; } KQ_UNLOCK(kq); } } else { if (kq->kq_knhashmask == 0) { tmp_knhash = hashinit_flags(KN_HASHSIZE, M_KQUEUE, &tmp_knhashmask, (mflag & M_WAITOK) != 0 ? HASH_WAITOK : HASH_NOWAIT); if (tmp_knhash == NULL) return (ENOMEM); KQ_LOCK(kq); if ((kq->kq_state & KQ_CLOSING) != 0) { to_free = tmp_knhash; error = EBADF; } else if (kq->kq_knhashmask == 0) { kq->kq_knhash = tmp_knhash; kq->kq_knhashmask = tmp_knhashmask; } else { to_free = tmp_knhash; } KQ_UNLOCK(kq); } } free(to_free, M_KQUEUE); KQ_NOTOWNED(kq); return (error); } static void kqueue_task(void *arg, int pending) { struct kqueue *kq; int haskqglobal; haskqglobal = 0; kq = arg; KQ_GLOBAL_LOCK(&kq_global, haskqglobal); KQ_LOCK(kq); KNOTE_LOCKED(&kq->kq_sel.si_note, 0); kq->kq_state &= ~KQ_TASKSCHED; if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) { wakeup(&kq->kq_state); } KQ_UNLOCK(kq); KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); } /* * Scan, update kn_data (if not ONESHOT), and copyout triggered events. * We treat KN_MARKER knotes as if they are in flux. */ static int kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, const struct timespec *tsp, struct kevent *keva, struct thread *td) { struct kevent *kevp; struct knote *kn, *marker; struct knlist *knl; sbintime_t asbt, rsbt; int count, error, haskqglobal, influx, nkev, touch; count = maxevents; nkev = 0; error = 0; haskqglobal = 0; if (maxevents == 0) goto done_nl; rsbt = 0; if (tsp != NULL) { if (tsp->tv_sec < 0 || tsp->tv_nsec < 0 || tsp->tv_nsec >= 1000000000) { error = EINVAL; goto done_nl; } if (timespecisset(tsp)) { if (tsp->tv_sec <= INT32_MAX) { rsbt = tstosbt(*tsp); if (TIMESEL(&asbt, rsbt)) asbt += tc_tick_sbt; if (asbt <= SBT_MAX - rsbt) asbt += rsbt; else asbt = 0; rsbt >>= tc_precexp; } else asbt = 0; } else asbt = -1; } else asbt = 0; marker = knote_alloc(M_WAITOK); marker->kn_status = KN_MARKER; KQ_LOCK(kq); retry: kevp = keva; if (kq->kq_count == 0) { if (asbt == -1) { error = EWOULDBLOCK; } else { kq->kq_state |= KQ_SLEEP; error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH, "kqread", asbt, rsbt, C_ABSOLUTE); } if (error == 0) goto retry; /* don't restart after signals... */ if (error == ERESTART) error = EINTR; else if (error == EWOULDBLOCK) error = 0; goto done; } TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); influx = 0; while (count) { KQ_OWNED(kq); kn = TAILQ_FIRST(&kq->kq_head); if ((kn->kn_status == KN_MARKER && kn != marker) || kn_in_flux(kn)) { if (influx) { influx = 0; KQ_FLUX_WAKEUP(kq); } kq->kq_state |= KQ_FLUXWAIT; error = msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); continue; } TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) { kn->kn_status &= ~KN_QUEUED; kq->kq_count--; continue; } if (kn == marker) { KQ_FLUX_WAKEUP(kq); if (count == maxevents) goto retry; goto done; } KASSERT(!kn_in_flux(kn), ("knote %p is unexpectedly in flux", kn)); if ((kn->kn_flags & EV_DROP) == EV_DROP) { kn->kn_status &= ~KN_QUEUED; kn_enter_flux(kn); kq->kq_count--; KQ_UNLOCK(kq); /* * We don't need to lock the list since we've * marked it as in flux. */ knote_drop(kn, td); KQ_LOCK(kq); continue; } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) { kn->kn_status &= ~KN_QUEUED; kn_enter_flux(kn); kq->kq_count--; KQ_UNLOCK(kq); /* * We don't need to lock the list since we've * marked the knote as being in flux. */ *kevp = kn->kn_kevent; knote_drop(kn, td); KQ_LOCK(kq); kn = NULL; } else { kn->kn_status |= KN_SCAN; kn_enter_flux(kn); KQ_UNLOCK(kq); if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE) KQ_GLOBAL_LOCK(&kq_global, haskqglobal); knl = kn_list_lock(kn); if (kn->kn_fop->f_event(kn, 0) == 0) { KQ_LOCK(kq); KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE | KN_SCAN); kn_leave_flux(kn); kq->kq_count--; kn_list_unlock(knl); influx = 1; continue; } touch = (!kn->kn_fop->f_isfd && kn->kn_fop->f_touch != NULL); if (touch) kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS); else *kevp = kn->kn_kevent; KQ_LOCK(kq); KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) { /* * Manually clear knotes who weren't * 'touch'ed. */ if (touch == 0 && kn->kn_flags & EV_CLEAR) { kn->kn_data = 0; kn->kn_fflags = 0; } if (kn->kn_flags & EV_DISPATCH) kn->kn_status |= KN_DISABLED; kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); kq->kq_count--; } else TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); kn->kn_status &= ~KN_SCAN; kn_leave_flux(kn); kn_list_unlock(knl); influx = 1; } /* we are returning a copy to the user */ kevp++; nkev++; count--; if (nkev == KQ_NEVENTS) { influx = 0; KQ_UNLOCK_FLUX(kq); error = k_ops->k_copyout(k_ops->arg, keva, nkev); nkev = 0; kevp = keva; KQ_LOCK(kq); if (error) break; } } TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); done: KQ_OWNED(kq); KQ_UNLOCK_FLUX(kq); knote_free(marker); done_nl: KQ_NOTOWNED(kq); if (nkev != 0) error = k_ops->k_copyout(k_ops->arg, keva, nkev); td->td_retval[0] = maxevents - count; return (error); } /*ARGSUSED*/ static int kqueue_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred, struct thread *td) { /* * Enabling sigio causes two major problems: * 1) infinite recursion: * Synopsys: kevent is being used to track signals and have FIOASYNC * set. On receipt of a signal this will cause a kqueue to recurse * into itself over and over. Sending the sigio causes the kqueue * to become ready, which in turn posts sigio again, forever. * Solution: this can be solved by setting a flag in the kqueue that * we have a SIGIO in progress. * 2) locking problems: * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts * us above the proc and pgrp locks. * Solution: Post a signal using an async mechanism, being sure to * record a generation count in the delivery so that we do not deliver * a signal to the wrong process. * * Note, these two mechanisms are somewhat mutually exclusive! */ #if 0 struct kqueue *kq; kq = fp->f_data; switch (cmd) { case FIOASYNC: if (*(int *)data) { kq->kq_state |= KQ_ASYNC; } else { kq->kq_state &= ~KQ_ASYNC; } return (0); case FIOSETOWN: return (fsetown(*(int *)data, &kq->kq_sigio)); case FIOGETOWN: *(int *)data = fgetown(&kq->kq_sigio); return (0); } #endif return (ENOTTY); } /*ARGSUSED*/ static int kqueue_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { struct kqueue *kq; int revents = 0; int error; if ((error = kqueue_acquire(fp, &kq))) return POLLERR; KQ_LOCK(kq); if (events & (POLLIN | POLLRDNORM)) { if (kq->kq_count) { revents |= events & (POLLIN | POLLRDNORM); } else { selrecord(td, &kq->kq_sel); if (SEL_WAITING(&kq->kq_sel)) kq->kq_state |= KQ_SEL; } } kqueue_release(kq, 1); KQ_UNLOCK(kq); return (revents); } /*ARGSUSED*/ static int kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred, struct thread *td) { bzero((void *)st, sizeof *st); /* * We no longer return kq_count because the unlocked value is useless. * If you spent all this time getting the count, why not spend your * syscall better by calling kevent? * * XXX - This is needed for libc_r. */ st->st_mode = S_IFIFO; return (0); } static void kqueue_drain(struct kqueue *kq, struct thread *td) { struct knote *kn; int i; KQ_LOCK(kq); KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING, ("kqueue already closing")); kq->kq_state |= KQ_CLOSING; if (kq->kq_refcnt > 1) msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0); KASSERT(kq->kq_refcnt == 1, ("other refs are out there!")); KASSERT(knlist_empty(&kq->kq_sel.si_note), ("kqueue's knlist not empty")); for (i = 0; i < kq->kq_knlistsize; i++) { while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) { if (kn_in_flux(kn)) { kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0); continue; } kn_enter_flux(kn); KQ_UNLOCK(kq); knote_drop(kn, td); KQ_LOCK(kq); } } if (kq->kq_knhashmask != 0) { for (i = 0; i <= kq->kq_knhashmask; i++) { while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) { if (kn_in_flux(kn)) { kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK, "kqclo2", 0); continue; } kn_enter_flux(kn); KQ_UNLOCK(kq); knote_drop(kn, td); KQ_LOCK(kq); } } } if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) { kq->kq_state |= KQ_TASKDRAIN; msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0); } if ((kq->kq_state & KQ_SEL) == KQ_SEL) { selwakeuppri(&kq->kq_sel, PSOCK); if (!SEL_WAITING(&kq->kq_sel)) kq->kq_state &= ~KQ_SEL; } KQ_UNLOCK(kq); } static void kqueue_destroy(struct kqueue *kq) { KASSERT(kq->kq_fdp == NULL, ("kqueue still attached to a file descriptor")); seldrain(&kq->kq_sel); knlist_destroy(&kq->kq_sel.si_note); mtx_destroy(&kq->kq_lock); if (kq->kq_knhash != NULL) free(kq->kq_knhash, M_KQUEUE); if (kq->kq_knlist != NULL) free(kq->kq_knlist, M_KQUEUE); funsetown(&kq->kq_sigio); } /*ARGSUSED*/ static int kqueue_close(struct file *fp, struct thread *td) { struct kqueue *kq = fp->f_data; struct filedesc *fdp; int error; int filedesc_unlock; if ((error = kqueue_acquire(fp, &kq))) return error; kqueue_drain(kq, td); /* * We could be called due to the knote_drop() doing fdrop(), * called from kqueue_register(). In this case the global * lock is owned, and filedesc sx is locked before, to not * take the sleepable lock after non-sleepable. */ fdp = kq->kq_fdp; kq->kq_fdp = NULL; if (!sx_xlocked(FILEDESC_LOCK(fdp))) { FILEDESC_XLOCK(fdp); filedesc_unlock = 1; } else filedesc_unlock = 0; TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list); if (filedesc_unlock) FILEDESC_XUNLOCK(fdp); kqueue_destroy(kq); chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0); crfree(kq->kq_cred); free(kq, M_KQUEUE); fp->f_data = NULL; return (0); } static int kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { kif->kf_type = KF_TYPE_KQUEUE; return (0); } static void kqueue_wakeup(struct kqueue *kq) { KQ_OWNED(kq); if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) { kq->kq_state &= ~KQ_SLEEP; wakeup(kq); } if ((kq->kq_state & KQ_SEL) == KQ_SEL) { selwakeuppri(&kq->kq_sel, PSOCK); if (!SEL_WAITING(&kq->kq_sel)) kq->kq_state &= ~KQ_SEL; } if (!knlist_empty(&kq->kq_sel.si_note)) kqueue_schedtask(kq); if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) { pgsigio(&kq->kq_sigio, SIGIO, 0); } } /* * Walk down a list of knotes, activating them if their event has triggered. * * There is a possibility to optimize in the case of one kq watching another. * Instead of scheduling a task to wake it up, you could pass enough state * down the chain to make up the parent kqueue. Make this code functional * first. */ void knote(struct knlist *list, long hint, int lockflags) { struct kqueue *kq; struct knote *kn, *tkn; int error; if (list == NULL) return; KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED); if ((lockflags & KNF_LISTLOCKED) == 0) list->kl_lock(list->kl_lockarg); /* * If we unlock the list lock (and enter influx), we can * eliminate the kqueue scheduling, but this will introduce * four lock/unlock's for each knote to test. Also, marker * would be needed to keep iteration position, since filters * or other threads could remove events. */ SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) { kq = kn->kn_kq; KQ_LOCK(kq); if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) { /* * Do not process the influx notes, except for * the influx coming from the kq unlock in the * kqueue_scan(). In the later case, we do * not interfere with the scan, since the code * fragment in kqueue_scan() locks the knlist, * and cannot proceed until we finished. */ KQ_UNLOCK(kq); } else if ((lockflags & KNF_NOKQLOCK) != 0) { kn_enter_flux(kn); KQ_UNLOCK(kq); error = kn->kn_fop->f_event(kn, hint); KQ_LOCK(kq); kn_leave_flux(kn); if (error) KNOTE_ACTIVATE(kn, 1); KQ_UNLOCK_FLUX(kq); } else { if (kn->kn_fop->f_event(kn, hint)) KNOTE_ACTIVATE(kn, 1); KQ_UNLOCK(kq); } } if ((lockflags & KNF_LISTLOCKED) == 0) list->kl_unlock(list->kl_lockarg); } /* * add a knote to a knlist */ void knlist_add(struct knlist *knl, struct knote *kn, int islocked) { KNL_ASSERT_LOCK(knl, islocked); KQ_NOTOWNED(kn->kn_kq); KASSERT(kn_in_flux(kn), ("knote %p not in flux", kn)); KASSERT((kn->kn_status & KN_DETACHED) != 0, ("knote %p was not detached", kn)); if (!islocked) knl->kl_lock(knl->kl_lockarg); SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext); if (!islocked) knl->kl_unlock(knl->kl_lockarg); KQ_LOCK(kn->kn_kq); kn->kn_knlist = knl; kn->kn_status &= ~KN_DETACHED; KQ_UNLOCK(kn->kn_kq); } static void knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked) { KASSERT(!kqislocked || knlislocked, ("kq locked w/o knl locked")); KNL_ASSERT_LOCK(knl, knlislocked); mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED); KASSERT(kqislocked || kn_in_flux(kn), ("knote %p not in flux", kn)); KASSERT((kn->kn_status & KN_DETACHED) == 0, ("knote %p was already detached", kn)); if (!knlislocked) knl->kl_lock(knl->kl_lockarg); SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext); kn->kn_knlist = NULL; if (!knlislocked) kn_list_unlock(knl); if (!kqislocked) KQ_LOCK(kn->kn_kq); kn->kn_status |= KN_DETACHED; if (!kqislocked) KQ_UNLOCK(kn->kn_kq); } /* * remove knote from the specified knlist */ void knlist_remove(struct knlist *knl, struct knote *kn, int islocked) { knlist_remove_kq(knl, kn, islocked, 0); } int knlist_empty(struct knlist *knl) { KNL_ASSERT_LOCKED(knl); return (SLIST_EMPTY(&knl->kl_list)); } static struct mtx knlist_lock; MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects", MTX_DEF); static void knlist_mtx_lock(void *arg); static void knlist_mtx_unlock(void *arg); static void knlist_mtx_lock(void *arg) { mtx_lock((struct mtx *)arg); } static void knlist_mtx_unlock(void *arg) { mtx_unlock((struct mtx *)arg); } static void knlist_mtx_assert_lock(void *arg, int what) { if (what == LA_LOCKED) mtx_assert((struct mtx *)arg, MA_OWNED); else mtx_assert((struct mtx *)arg, MA_NOTOWNED); } static void knlist_rw_rlock(void *arg) { rw_rlock((struct rwlock *)arg); } static void knlist_rw_runlock(void *arg) { rw_runlock((struct rwlock *)arg); } static void knlist_rw_assert_lock(void *arg, int what) { if (what == LA_LOCKED) rw_assert((struct rwlock *)arg, RA_LOCKED); else rw_assert((struct rwlock *)arg, RA_UNLOCKED); } void knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *), void (*kl_unlock)(void *), void (*kl_assert_lock)(void *, int)) { if (lock == NULL) knl->kl_lockarg = &knlist_lock; else knl->kl_lockarg = lock; if (kl_lock == NULL) knl->kl_lock = knlist_mtx_lock; else knl->kl_lock = kl_lock; if (kl_unlock == NULL) knl->kl_unlock = knlist_mtx_unlock; else knl->kl_unlock = kl_unlock; if (kl_assert_lock == NULL) knl->kl_assert_lock = knlist_mtx_assert_lock; else knl->kl_assert_lock = kl_assert_lock; knl->kl_autodestroy = 0; SLIST_INIT(&knl->kl_list); } void knlist_init_mtx(struct knlist *knl, struct mtx *lock) { knlist_init(knl, lock, NULL, NULL, NULL); } struct knlist * knlist_alloc(struct mtx *lock) { struct knlist *knl; knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK); knlist_init_mtx(knl, lock); return (knl); } void knlist_init_rw_reader(struct knlist *knl, struct rwlock *lock) { knlist_init(knl, lock, knlist_rw_rlock, knlist_rw_runlock, knlist_rw_assert_lock); } void knlist_destroy(struct knlist *knl) { KASSERT(KNLIST_EMPTY(knl), ("destroying knlist %p with knotes on it", knl)); } void knlist_detach(struct knlist *knl) { KNL_ASSERT_LOCKED(knl); knl->kl_autodestroy = 1; if (knlist_empty(knl)) { knlist_destroy(knl); free(knl, M_KQUEUE); } } /* * Even if we are locked, we may need to drop the lock to allow any influx * knotes time to "settle". */ void knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn) { struct knote *kn, *kn2; struct kqueue *kq; KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl)); if (islocked) KNL_ASSERT_LOCKED(knl); else { KNL_ASSERT_UNLOCKED(knl); again: /* need to reacquire lock since we have dropped it */ knl->kl_lock(knl->kl_lockarg); } SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) { kq = kn->kn_kq; KQ_LOCK(kq); if (kn_in_flux(kn)) { KQ_UNLOCK(kq); continue; } knlist_remove_kq(knl, kn, 1, 1); if (killkn) { kn_enter_flux(kn); KQ_UNLOCK(kq); knote_drop_detached(kn, td); } else { /* Make sure cleared knotes disappear soon */ kn->kn_flags |= EV_EOF | EV_ONESHOT; KQ_UNLOCK(kq); } kq = NULL; } if (!SLIST_EMPTY(&knl->kl_list)) { /* there are still in flux knotes remaining */ kn = SLIST_FIRST(&knl->kl_list); kq = kn->kn_kq; KQ_LOCK(kq); KASSERT(kn_in_flux(kn), ("knote removed w/o list lock")); knl->kl_unlock(knl->kl_lockarg); kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0); kq = NULL; goto again; } if (islocked) KNL_ASSERT_LOCKED(knl); else { knl->kl_unlock(knl->kl_lockarg); KNL_ASSERT_UNLOCKED(knl); } } /* * Remove all knotes referencing a specified fd must be called with FILEDESC * lock. This prevents a race where a new fd comes along and occupies the * entry and we attach a knote to the fd. */ void knote_fdclose(struct thread *td, int fd) { struct filedesc *fdp = td->td_proc->p_fd; struct kqueue *kq; struct knote *kn; int influx; FILEDESC_XLOCK_ASSERT(fdp); /* * We shouldn't have to worry about new kevents appearing on fd * since filedesc is locked. */ TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) { KQ_LOCK(kq); again: influx = 0; while (kq->kq_knlistsize > fd && (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) { if (kn_in_flux(kn)) { /* someone else might be waiting on our knote */ if (influx) wakeup(kq); kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); goto again; } kn_enter_flux(kn); KQ_UNLOCK(kq); influx = 1; knote_drop(kn, td); KQ_LOCK(kq); } KQ_UNLOCK_FLUX(kq); } } static int knote_attach(struct knote *kn, struct kqueue *kq) { struct klist *list; KASSERT(kn_in_flux(kn), ("knote %p not marked influx", kn)); KQ_OWNED(kq); if ((kq->kq_state & KQ_CLOSING) != 0) return (EBADF); if (kn->kn_fop->f_isfd) { if (kn->kn_id >= kq->kq_knlistsize) return (ENOMEM); list = &kq->kq_knlist[kn->kn_id]; } else { if (kq->kq_knhash == NULL) return (ENOMEM); list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; } SLIST_INSERT_HEAD(list, kn, kn_link); return (0); } static void knote_drop(struct knote *kn, struct thread *td) { if ((kn->kn_status & KN_DETACHED) == 0) kn->kn_fop->f_detach(kn); knote_drop_detached(kn, td); } static void knote_drop_detached(struct knote *kn, struct thread *td) { struct kqueue *kq; struct klist *list; kq = kn->kn_kq; KASSERT((kn->kn_status & KN_DETACHED) != 0, ("knote %p still attached", kn)); KQ_NOTOWNED(kq); KQ_LOCK(kq); KASSERT(kn->kn_influx == 1, ("knote_drop called on %p with influx %d", kn, kn->kn_influx)); if (kn->kn_fop->f_isfd) list = &kq->kq_knlist[kn->kn_id]; else list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; if (!SLIST_EMPTY(list)) SLIST_REMOVE(list, kn, knote, kn_link); if (kn->kn_status & KN_QUEUED) knote_dequeue(kn); KQ_UNLOCK_FLUX(kq); if (kn->kn_fop->f_isfd) { fdrop(kn->kn_fp, td); kn->kn_fp = NULL; } kqueue_fo_release(kn->kn_kevent.filter); kn->kn_fop = NULL; knote_free(kn); } static void knote_enqueue(struct knote *kn) { struct kqueue *kq = kn->kn_kq; KQ_OWNED(kn->kn_kq); KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); kn->kn_status |= KN_QUEUED; kq->kq_count++; kqueue_wakeup(kq); } static void knote_dequeue(struct knote *kn) { struct kqueue *kq = kn->kn_kq; KQ_OWNED(kn->kn_kq); KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); kn->kn_status &= ~KN_QUEUED; kq->kq_count--; } static void knote_init(void) { knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); } SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL); static struct knote * knote_alloc(int mflag) { return (uma_zalloc(knote_zone, mflag | M_ZERO)); } static void knote_free(struct knote *kn) { uma_zfree(knote_zone, kn); } /* * Register the kev w/ the kq specified by fd. */ int kqfd_register(int fd, struct kevent *kev, struct thread *td, int mflag) { struct kqueue *kq; struct file *fp; cap_rights_t rights; int error; - error = fget(td, fd, cap_rights_init(&rights, CAP_KQUEUE_CHANGE), &fp); + error = fget(td, fd, cap_rights_init_one(&rights, CAP_KQUEUE_CHANGE), + &fp); if (error != 0) return (error); if ((error = kqueue_acquire(fp, &kq)) != 0) goto noacquire; error = kqueue_register(kq, kev, td, mflag); kqueue_release(kq, 0); noacquire: fdrop(fp, td); return (error); } diff --git a/sys/kern/tty.c b/sys/kern/tty.c index 4b1f7ca52abe..00b4df675311 100644 --- a/sys/kern/tty.c +++ b/sys/kern/tty.c @@ -1,2454 +1,2454 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2008 Ed Schouten * All rights reserved. * * Portions of this software were developed under sponsorship from Snow * B.V., the Netherlands. * * 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_capsicum.h" #include "opt_printf.h" #include #include #include #include #include #include #include #include #ifdef COMPAT_43TTY #include #endif /* COMPAT_43TTY */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define TTYDEFCHARS #include #undef TTYDEFCHARS #include #include #include #include static MALLOC_DEFINE(M_TTY, "tty", "tty device"); static void tty_rel_free(struct tty *tp); static TAILQ_HEAD(, tty) tty_list = TAILQ_HEAD_INITIALIZER(tty_list); static struct sx tty_list_sx; SX_SYSINIT(tty_list, &tty_list_sx, "tty list"); static unsigned int tty_list_count = 0; /* Character device of /dev/console. */ static struct cdev *dev_console; static const char *dev_console_filename; /* * Flags that are supported and stored by this implementation. */ #define TTYSUP_IFLAG (IGNBRK|BRKINT|IGNPAR|PARMRK|INPCK|ISTRIP|\ INLCR|IGNCR|ICRNL|IXON|IXOFF|IXANY|IMAXBEL) #define TTYSUP_OFLAG (OPOST|ONLCR|TAB3|ONOEOT|OCRNL|ONOCR|ONLRET) #define TTYSUP_LFLAG (ECHOKE|ECHOE|ECHOK|ECHO|ECHONL|ECHOPRT|\ ECHOCTL|ISIG|ICANON|ALTWERASE|IEXTEN|TOSTOP|\ FLUSHO|NOKERNINFO|NOFLSH) #define TTYSUP_CFLAG (CIGNORE|CSIZE|CSTOPB|CREAD|PARENB|PARODD|\ HUPCL|CLOCAL|CCTS_OFLOW|CRTS_IFLOW|CDTR_IFLOW|\ CDSR_OFLOW|CCAR_OFLOW|CNO_RTSDTR) #define TTY_CALLOUT(tp,d) (dev2unit(d) & TTYUNIT_CALLOUT) static int tty_drainwait = 5 * 60; SYSCTL_INT(_kern, OID_AUTO, tty_drainwait, CTLFLAG_RWTUN, &tty_drainwait, 0, "Default output drain timeout in seconds"); /* * Set TTY buffer sizes. */ #define TTYBUF_MAX 65536 #ifdef PRINTF_BUFR_SIZE #define TTY_PRBUF_SIZE PRINTF_BUFR_SIZE #else #define TTY_PRBUF_SIZE 256 #endif /* * Allocate buffer space if necessary, and set low watermarks, based on speed. * Note that the ttyxxxq_setsize() functions may drop and then reacquire the tty * lock during memory allocation. They will return ENXIO if the tty disappears * while unlocked. */ static int tty_watermarks(struct tty *tp) { size_t bs = 0; int error; /* Provide an input buffer for 2 seconds of data. */ if (tp->t_termios.c_cflag & CREAD) bs = MIN(tp->t_termios.c_ispeed / 5, TTYBUF_MAX); error = ttyinq_setsize(&tp->t_inq, tp, bs); if (error != 0) return (error); /* Set low watermark at 10% (when 90% is available). */ tp->t_inlow = (ttyinq_getallocatedsize(&tp->t_inq) * 9) / 10; /* Provide an output buffer for 2 seconds of data. */ bs = MIN(tp->t_termios.c_ospeed / 5, TTYBUF_MAX); error = ttyoutq_setsize(&tp->t_outq, tp, bs); if (error != 0) return (error); /* Set low watermark at 10% (when 90% is available). */ tp->t_outlow = (ttyoutq_getallocatedsize(&tp->t_outq) * 9) / 10; return (0); } static int tty_drain(struct tty *tp, int leaving) { sbintime_t timeout_at; size_t bytes; int error; if (ttyhook_hashook(tp, getc_inject)) /* buffer is inaccessible */ return (0); /* * For close(), use the recent historic timeout of "1 second without * making progress". For tcdrain(), use t_drainwait as the timeout, * with zero meaning "no timeout" which gives POSIX behavior. */ if (leaving) timeout_at = getsbinuptime() + SBT_1S; else if (tp->t_drainwait != 0) timeout_at = getsbinuptime() + SBT_1S * tp->t_drainwait; else timeout_at = 0; /* * Poll the output buffer and the hardware for completion, at 10 Hz. * Polling is required for devices which are not able to signal an * interrupt when the transmitter becomes idle (most USB serial devs). * The unusual structure of this loop ensures we check for busy one more * time after tty_timedwait() returns EWOULDBLOCK, so that success has * higher priority than timeout if the IO completed in the last 100mS. */ error = 0; bytes = ttyoutq_bytesused(&tp->t_outq); for (;;) { if (ttyoutq_bytesused(&tp->t_outq) == 0 && !ttydevsw_busy(tp)) return (0); if (error != 0) return (error); ttydevsw_outwakeup(tp); error = tty_timedwait(tp, &tp->t_outwait, hz / 10); if (error != 0 && error != EWOULDBLOCK) return (error); else if (timeout_at == 0 || getsbinuptime() < timeout_at) error = 0; else if (leaving && ttyoutq_bytesused(&tp->t_outq) < bytes) { /* In close, making progress, grant an extra second. */ error = 0; timeout_at += SBT_1S; bytes = ttyoutq_bytesused(&tp->t_outq); } } } /* * Though ttydev_enter() and ttydev_leave() seem to be related, they * don't have to be used together. ttydev_enter() is used by the cdev * operations to prevent an actual operation from being processed when * the TTY has been abandoned. ttydev_leave() is used by ttydev_open() * and ttydev_close() to determine whether per-TTY data should be * deallocated. */ static __inline int ttydev_enter(struct tty *tp) { tty_lock(tp); if (tty_gone(tp) || !tty_opened(tp)) { /* Device is already gone. */ tty_unlock(tp); return (ENXIO); } return (0); } static void ttydev_leave(struct tty *tp) { tty_assert_locked(tp); if (tty_opened(tp) || tp->t_flags & TF_OPENCLOSE) { /* Device is still opened somewhere. */ tty_unlock(tp); return; } tp->t_flags |= TF_OPENCLOSE; /* Remove console TTY. */ if (constty == tp) constty_clear(); /* Drain any output. */ if (!tty_gone(tp)) tty_drain(tp, 1); ttydisc_close(tp); /* Free i/o queues now since they might be large. */ ttyinq_free(&tp->t_inq); tp->t_inlow = 0; ttyoutq_free(&tp->t_outq); tp->t_outlow = 0; knlist_clear(&tp->t_inpoll.si_note, 1); knlist_clear(&tp->t_outpoll.si_note, 1); if (!tty_gone(tp)) ttydevsw_close(tp); tp->t_flags &= ~TF_OPENCLOSE; cv_broadcast(&tp->t_dcdwait); tty_rel_free(tp); } /* * Operations that are exposed through the character device in /dev. */ static int ttydev_open(struct cdev *dev, int oflags, int devtype __unused, struct thread *td) { struct tty *tp; int error; tp = dev->si_drv1; error = 0; tty_lock(tp); if (tty_gone(tp)) { /* Device is already gone. */ tty_unlock(tp); return (ENXIO); } /* * Block when other processes are currently opening or closing * the TTY. */ while (tp->t_flags & TF_OPENCLOSE) { error = tty_wait(tp, &tp->t_dcdwait); if (error != 0) { tty_unlock(tp); return (error); } } tp->t_flags |= TF_OPENCLOSE; /* * Make sure the "tty" and "cua" device cannot be opened at the * same time. The console is a "tty" device. */ if (TTY_CALLOUT(tp, dev)) { if (tp->t_flags & (TF_OPENED_CONS | TF_OPENED_IN)) { error = EBUSY; goto done; } } else { if (tp->t_flags & TF_OPENED_OUT) { error = EBUSY; goto done; } } if (tp->t_flags & TF_EXCLUDE && priv_check(td, PRIV_TTY_EXCLUSIVE)) { error = EBUSY; goto done; } if (!tty_opened(tp)) { /* Set proper termios flags. */ if (TTY_CALLOUT(tp, dev)) tp->t_termios = tp->t_termios_init_out; else tp->t_termios = tp->t_termios_init_in; ttydevsw_param(tp, &tp->t_termios); /* Prevent modem control on callout devices and /dev/console. */ if (TTY_CALLOUT(tp, dev) || dev == dev_console) tp->t_termios.c_cflag |= CLOCAL; if ((tp->t_termios.c_cflag & CNO_RTSDTR) == 0) ttydevsw_modem(tp, SER_DTR|SER_RTS, 0); error = ttydevsw_open(tp); if (error != 0) goto done; ttydisc_open(tp); error = tty_watermarks(tp); if (error != 0) goto done; } /* Wait for Carrier Detect. */ if ((oflags & O_NONBLOCK) == 0 && (tp->t_termios.c_cflag & CLOCAL) == 0) { while ((ttydevsw_modem(tp, 0, 0) & SER_DCD) == 0) { error = tty_wait(tp, &tp->t_dcdwait); if (error != 0) goto done; } } if (dev == dev_console) tp->t_flags |= TF_OPENED_CONS; else if (TTY_CALLOUT(tp, dev)) tp->t_flags |= TF_OPENED_OUT; else tp->t_flags |= TF_OPENED_IN; MPASS((tp->t_flags & (TF_OPENED_CONS | TF_OPENED_IN)) == 0 || (tp->t_flags & TF_OPENED_OUT) == 0); done: tp->t_flags &= ~TF_OPENCLOSE; cv_broadcast(&tp->t_dcdwait); ttydev_leave(tp); return (error); } static int ttydev_close(struct cdev *dev, int fflag, int devtype __unused, struct thread *td __unused) { struct tty *tp = dev->si_drv1; tty_lock(tp); /* * Don't actually close the device if it is being used as the * console. */ MPASS((tp->t_flags & (TF_OPENED_CONS | TF_OPENED_IN)) == 0 || (tp->t_flags & TF_OPENED_OUT) == 0); if (dev == dev_console) tp->t_flags &= ~TF_OPENED_CONS; else tp->t_flags &= ~(TF_OPENED_IN|TF_OPENED_OUT); if (tp->t_flags & TF_OPENED) { tty_unlock(tp); return (0); } /* If revoking, flush output now to avoid draining it later. */ if (fflag & FREVOKE) tty_flush(tp, FWRITE); tp->t_flags &= ~TF_EXCLUDE; /* Properly wake up threads that are stuck - revoke(). */ tp->t_revokecnt++; tty_wakeup(tp, FREAD|FWRITE); cv_broadcast(&tp->t_bgwait); cv_broadcast(&tp->t_dcdwait); ttydev_leave(tp); return (0); } static __inline int tty_is_ctty(struct tty *tp, struct proc *p) { tty_assert_locked(tp); return (p->p_session == tp->t_session && p->p_flag & P_CONTROLT); } int tty_wait_background(struct tty *tp, struct thread *td, int sig) { struct proc *p; struct pgrp *pg; ksiginfo_t ksi; int error; MPASS(sig == SIGTTIN || sig == SIGTTOU); tty_assert_locked(tp); p = td->td_proc; for (;;) { pg = p->p_pgrp; PGRP_LOCK(pg); PROC_LOCK(p); /* * pg may no longer be our process group. * Re-check after locking. */ if (p->p_pgrp != pg) { PROC_UNLOCK(p); PGRP_UNLOCK(pg); continue; } /* * The process should only sleep, when: * - This terminal is the controlling terminal * - Its process group is not the foreground process * group * - The parent process isn't waiting for the child to * exit * - the signal to send to the process isn't masked */ if (!tty_is_ctty(tp, p) || p->p_pgrp == tp->t_pgrp) { /* Allow the action to happen. */ PROC_UNLOCK(p); PGRP_UNLOCK(pg); return (0); } if (SIGISMEMBER(p->p_sigacts->ps_sigignore, sig) || SIGISMEMBER(td->td_sigmask, sig)) { /* Only allow them in write()/ioctl(). */ PROC_UNLOCK(p); PGRP_UNLOCK(pg); return (sig == SIGTTOU ? 0 : EIO); } if ((p->p_flag & P_PPWAIT) != 0 || (pg->pg_flags & PGRP_ORPHANED) != 0) { /* Don't allow the action to happen. */ PROC_UNLOCK(p); PGRP_UNLOCK(pg); return (EIO); } PROC_UNLOCK(p); /* * Send the signal and sleep until we're the new * foreground process group. */ if (sig != 0) { ksiginfo_init(&ksi); ksi.ksi_code = SI_KERNEL; ksi.ksi_signo = sig; sig = 0; } pgsignal(pg, ksi.ksi_signo, 1, &ksi); PGRP_UNLOCK(pg); error = tty_wait(tp, &tp->t_bgwait); if (error) return (error); } } static int ttydev_read(struct cdev *dev, struct uio *uio, int ioflag) { struct tty *tp = dev->si_drv1; int error; error = ttydev_enter(tp); if (error) goto done; error = ttydisc_read(tp, uio, ioflag); tty_unlock(tp); /* * The read() call should not throw an error when the device is * being destroyed. Silently convert it to an EOF. */ done: if (error == ENXIO) error = 0; return (error); } static int ttydev_write(struct cdev *dev, struct uio *uio, int ioflag) { struct tty *tp = dev->si_drv1; int error; error = ttydev_enter(tp); if (error) return (error); if (tp->t_termios.c_lflag & TOSTOP) { error = tty_wait_background(tp, curthread, SIGTTOU); if (error) goto done; } if (ioflag & IO_NDELAY && tp->t_flags & TF_BUSY_OUT) { /* Allow non-blocking writes to bypass serialization. */ error = ttydisc_write(tp, uio, ioflag); } else { /* Serialize write() calls. */ while (tp->t_flags & TF_BUSY_OUT) { error = tty_wait(tp, &tp->t_outserwait); if (error) goto done; } tp->t_flags |= TF_BUSY_OUT; error = ttydisc_write(tp, uio, ioflag); tp->t_flags &= ~TF_BUSY_OUT; cv_signal(&tp->t_outserwait); } done: tty_unlock(tp); return (error); } static int ttydev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) { struct tty *tp = dev->si_drv1; int error; error = ttydev_enter(tp); if (error) return (error); switch (cmd) { case TIOCCBRK: case TIOCCONS: case TIOCDRAIN: case TIOCEXCL: case TIOCFLUSH: case TIOCNXCL: case TIOCSBRK: case TIOCSCTTY: case TIOCSETA: case TIOCSETAF: case TIOCSETAW: case TIOCSPGRP: case TIOCSTART: case TIOCSTAT: case TIOCSTI: case TIOCSTOP: case TIOCSWINSZ: #if 0 case TIOCSDRAINWAIT: case TIOCSETD: #endif #ifdef COMPAT_43TTY case TIOCLBIC: case TIOCLBIS: case TIOCLSET: case TIOCSETC: case OTIOCSETD: case TIOCSETN: case TIOCSETP: case TIOCSLTC: #endif /* COMPAT_43TTY */ /* * If the ioctl() causes the TTY to be modified, let it * wait in the background. */ error = tty_wait_background(tp, curthread, SIGTTOU); if (error) goto done; } if (cmd == TIOCSETA || cmd == TIOCSETAW || cmd == TIOCSETAF) { struct termios *old = &tp->t_termios; struct termios *new = (struct termios *)data; struct termios *lock = TTY_CALLOUT(tp, dev) ? &tp->t_termios_lock_out : &tp->t_termios_lock_in; int cc; /* * Lock state devices. Just overwrite the values of the * commands that are currently in use. */ new->c_iflag = (old->c_iflag & lock->c_iflag) | (new->c_iflag & ~lock->c_iflag); new->c_oflag = (old->c_oflag & lock->c_oflag) | (new->c_oflag & ~lock->c_oflag); new->c_cflag = (old->c_cflag & lock->c_cflag) | (new->c_cflag & ~lock->c_cflag); new->c_lflag = (old->c_lflag & lock->c_lflag) | (new->c_lflag & ~lock->c_lflag); for (cc = 0; cc < NCCS; ++cc) if (lock->c_cc[cc]) new->c_cc[cc] = old->c_cc[cc]; if (lock->c_ispeed) new->c_ispeed = old->c_ispeed; if (lock->c_ospeed) new->c_ospeed = old->c_ospeed; } error = tty_ioctl(tp, cmd, data, fflag, td); done: tty_unlock(tp); return (error); } static int ttydev_poll(struct cdev *dev, int events, struct thread *td) { struct tty *tp = dev->si_drv1; int error, revents = 0; error = ttydev_enter(tp); if (error) return ((events & (POLLIN|POLLRDNORM)) | POLLHUP); if (events & (POLLIN|POLLRDNORM)) { /* See if we can read something. */ if (ttydisc_read_poll(tp) > 0) revents |= events & (POLLIN|POLLRDNORM); } if (tp->t_flags & TF_ZOMBIE) { /* Hangup flag on zombie state. */ revents |= POLLHUP; } else if (events & (POLLOUT|POLLWRNORM)) { /* See if we can write something. */ if (ttydisc_write_poll(tp) > 0) revents |= events & (POLLOUT|POLLWRNORM); } if (revents == 0) { if (events & (POLLIN|POLLRDNORM)) selrecord(td, &tp->t_inpoll); if (events & (POLLOUT|POLLWRNORM)) selrecord(td, &tp->t_outpoll); } tty_unlock(tp); return (revents); } static int ttydev_mmap(struct cdev *dev, vm_ooffset_t offset, vm_paddr_t *paddr, int nprot, vm_memattr_t *memattr) { struct tty *tp = dev->si_drv1; int error; /* Handle mmap() through the driver. */ error = ttydev_enter(tp); if (error) return (-1); error = ttydevsw_mmap(tp, offset, paddr, nprot, memattr); tty_unlock(tp); return (error); } /* * kqueue support. */ static void tty_kqops_read_detach(struct knote *kn) { struct tty *tp = kn->kn_hook; knlist_remove(&tp->t_inpoll.si_note, kn, 0); } static int tty_kqops_read_event(struct knote *kn, long hint __unused) { struct tty *tp = kn->kn_hook; tty_assert_locked(tp); if (tty_gone(tp) || tp->t_flags & TF_ZOMBIE) { kn->kn_flags |= EV_EOF; return (1); } else { kn->kn_data = ttydisc_read_poll(tp); return (kn->kn_data > 0); } } static void tty_kqops_write_detach(struct knote *kn) { struct tty *tp = kn->kn_hook; knlist_remove(&tp->t_outpoll.si_note, kn, 0); } static int tty_kqops_write_event(struct knote *kn, long hint __unused) { struct tty *tp = kn->kn_hook; tty_assert_locked(tp); if (tty_gone(tp)) { kn->kn_flags |= EV_EOF; return (1); } else { kn->kn_data = ttydisc_write_poll(tp); return (kn->kn_data > 0); } } static struct filterops tty_kqops_read = { .f_isfd = 1, .f_detach = tty_kqops_read_detach, .f_event = tty_kqops_read_event, }; static struct filterops tty_kqops_write = { .f_isfd = 1, .f_detach = tty_kqops_write_detach, .f_event = tty_kqops_write_event, }; static int ttydev_kqfilter(struct cdev *dev, struct knote *kn) { struct tty *tp = dev->si_drv1; int error; error = ttydev_enter(tp); if (error) return (error); switch (kn->kn_filter) { case EVFILT_READ: kn->kn_hook = tp; kn->kn_fop = &tty_kqops_read; knlist_add(&tp->t_inpoll.si_note, kn, 1); break; case EVFILT_WRITE: kn->kn_hook = tp; kn->kn_fop = &tty_kqops_write; knlist_add(&tp->t_outpoll.si_note, kn, 1); break; default: error = EINVAL; break; } tty_unlock(tp); return (error); } static struct cdevsw ttydev_cdevsw = { .d_version = D_VERSION, .d_open = ttydev_open, .d_close = ttydev_close, .d_read = ttydev_read, .d_write = ttydev_write, .d_ioctl = ttydev_ioctl, .d_kqfilter = ttydev_kqfilter, .d_poll = ttydev_poll, .d_mmap = ttydev_mmap, .d_name = "ttydev", .d_flags = D_TTY, }; /* * Init/lock-state devices */ static int ttyil_open(struct cdev *dev, int oflags __unused, int devtype __unused, struct thread *td) { struct tty *tp; int error; tp = dev->si_drv1; error = 0; tty_lock(tp); if (tty_gone(tp)) error = ENODEV; tty_unlock(tp); return (error); } static int ttyil_close(struct cdev *dev __unused, int flag __unused, int mode __unused, struct thread *td __unused) { return (0); } static int ttyil_rdwr(struct cdev *dev __unused, struct uio *uio __unused, int ioflag __unused) { return (ENODEV); } static int ttyil_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) { struct tty *tp = dev->si_drv1; int error; tty_lock(tp); if (tty_gone(tp)) { error = ENODEV; goto done; } error = ttydevsw_cioctl(tp, dev2unit(dev), cmd, data, td); if (error != ENOIOCTL) goto done; error = 0; switch (cmd) { case TIOCGETA: /* Obtain terminal flags through tcgetattr(). */ *(struct termios*)data = *(struct termios*)dev->si_drv2; break; case TIOCSETA: /* Set terminal flags through tcsetattr(). */ error = priv_check(td, PRIV_TTY_SETA); if (error) break; *(struct termios*)dev->si_drv2 = *(struct termios*)data; break; case TIOCGETD: *(int *)data = TTYDISC; break; case TIOCGWINSZ: bzero(data, sizeof(struct winsize)); break; default: error = ENOTTY; } done: tty_unlock(tp); return (error); } static struct cdevsw ttyil_cdevsw = { .d_version = D_VERSION, .d_open = ttyil_open, .d_close = ttyil_close, .d_read = ttyil_rdwr, .d_write = ttyil_rdwr, .d_ioctl = ttyil_ioctl, .d_name = "ttyil", .d_flags = D_TTY, }; static void tty_init_termios(struct tty *tp) { struct termios *t = &tp->t_termios_init_in; t->c_cflag = TTYDEF_CFLAG; t->c_iflag = TTYDEF_IFLAG; t->c_lflag = TTYDEF_LFLAG; t->c_oflag = TTYDEF_OFLAG; t->c_ispeed = TTYDEF_SPEED; t->c_ospeed = TTYDEF_SPEED; memcpy(&t->c_cc, ttydefchars, sizeof ttydefchars); tp->t_termios_init_out = *t; } void tty_init_console(struct tty *tp, speed_t s) { struct termios *ti = &tp->t_termios_init_in; struct termios *to = &tp->t_termios_init_out; if (s != 0) { ti->c_ispeed = ti->c_ospeed = s; to->c_ispeed = to->c_ospeed = s; } ti->c_cflag |= CLOCAL; to->c_cflag |= CLOCAL; } /* * Standard device routine implementations, mostly meant for * pseudo-terminal device drivers. When a driver creates a new terminal * device class, missing routines are patched. */ static int ttydevsw_defopen(struct tty *tp __unused) { return (0); } static void ttydevsw_defclose(struct tty *tp __unused) { } static void ttydevsw_defoutwakeup(struct tty *tp __unused) { panic("Terminal device has output, while not implemented"); } static void ttydevsw_definwakeup(struct tty *tp __unused) { } static int ttydevsw_defioctl(struct tty *tp __unused, u_long cmd __unused, caddr_t data __unused, struct thread *td __unused) { return (ENOIOCTL); } static int ttydevsw_defcioctl(struct tty *tp __unused, int unit __unused, u_long cmd __unused, caddr_t data __unused, struct thread *td __unused) { return (ENOIOCTL); } static int ttydevsw_defparam(struct tty *tp __unused, struct termios *t) { /* * Allow the baud rate to be adjusted for pseudo-devices, but at * least restrict it to 115200 to prevent excessive buffer * usage. Also disallow 0, to prevent foot shooting. */ if (t->c_ispeed < B50) t->c_ispeed = B50; else if (t->c_ispeed > B115200) t->c_ispeed = B115200; if (t->c_ospeed < B50) t->c_ospeed = B50; else if (t->c_ospeed > B115200) t->c_ospeed = B115200; t->c_cflag |= CREAD; return (0); } static int ttydevsw_defmodem(struct tty *tp __unused, int sigon __unused, int sigoff __unused) { /* Simulate a carrier to make the TTY layer happy. */ return (SER_DCD); } static int ttydevsw_defmmap(struct tty *tp __unused, vm_ooffset_t offset __unused, vm_paddr_t *paddr __unused, int nprot __unused, vm_memattr_t *memattr __unused) { return (-1); } static void ttydevsw_defpktnotify(struct tty *tp __unused, char event __unused) { } static void ttydevsw_deffree(void *softc __unused) { panic("Terminal device freed without a free-handler"); } static bool ttydevsw_defbusy(struct tty *tp __unused) { return (FALSE); } /* * TTY allocation and deallocation. TTY devices can be deallocated when * the driver doesn't use it anymore, when the TTY isn't a session's * controlling TTY and when the device node isn't opened through devfs. */ struct tty * tty_alloc(struct ttydevsw *tsw, void *sc) { return (tty_alloc_mutex(tsw, sc, NULL)); } struct tty * tty_alloc_mutex(struct ttydevsw *tsw, void *sc, struct mtx *mutex) { struct tty *tp; /* Make sure the driver defines all routines. */ #define PATCH_FUNC(x) do { \ if (tsw->tsw_ ## x == NULL) \ tsw->tsw_ ## x = ttydevsw_def ## x; \ } while (0) PATCH_FUNC(open); PATCH_FUNC(close); PATCH_FUNC(outwakeup); PATCH_FUNC(inwakeup); PATCH_FUNC(ioctl); PATCH_FUNC(cioctl); PATCH_FUNC(param); PATCH_FUNC(modem); PATCH_FUNC(mmap); PATCH_FUNC(pktnotify); PATCH_FUNC(free); PATCH_FUNC(busy); #undef PATCH_FUNC tp = malloc(sizeof(struct tty) + TTY_PRBUF_SIZE, M_TTY, M_WAITOK | M_ZERO); tp->t_prbufsz = TTY_PRBUF_SIZE; tp->t_devsw = tsw; tp->t_devswsoftc = sc; tp->t_flags = tsw->tsw_flags; tp->t_drainwait = tty_drainwait; tty_init_termios(tp); cv_init(&tp->t_inwait, "ttyin"); cv_init(&tp->t_outwait, "ttyout"); cv_init(&tp->t_outserwait, "ttyosr"); cv_init(&tp->t_bgwait, "ttybg"); cv_init(&tp->t_dcdwait, "ttydcd"); /* Allow drivers to use a custom mutex to lock the TTY. */ if (mutex != NULL) { tp->t_mtx = mutex; } else { tp->t_mtx = &tp->t_mtxobj; mtx_init(&tp->t_mtxobj, "ttymtx", NULL, MTX_DEF); } knlist_init_mtx(&tp->t_inpoll.si_note, tp->t_mtx); knlist_init_mtx(&tp->t_outpoll.si_note, tp->t_mtx); return (tp); } static void tty_dealloc(void *arg) { struct tty *tp = arg; /* * ttyydev_leave() usually frees the i/o queues earlier, but it is * not always called between queue allocation and here. The queues * may be allocated by ioctls on a pty control device without the * corresponding pty slave device ever being open, or after it is * closed. */ ttyinq_free(&tp->t_inq); ttyoutq_free(&tp->t_outq); seldrain(&tp->t_inpoll); seldrain(&tp->t_outpoll); knlist_destroy(&tp->t_inpoll.si_note); knlist_destroy(&tp->t_outpoll.si_note); cv_destroy(&tp->t_inwait); cv_destroy(&tp->t_outwait); cv_destroy(&tp->t_bgwait); cv_destroy(&tp->t_dcdwait); cv_destroy(&tp->t_outserwait); if (tp->t_mtx == &tp->t_mtxobj) mtx_destroy(&tp->t_mtxobj); ttydevsw_free(tp); free(tp, M_TTY); } static void tty_rel_free(struct tty *tp) { struct cdev *dev; tty_assert_locked(tp); #define TF_ACTIVITY (TF_GONE|TF_OPENED|TF_HOOK|TF_OPENCLOSE) if (tp->t_sessioncnt != 0 || (tp->t_flags & TF_ACTIVITY) != TF_GONE) { /* TTY is still in use. */ tty_unlock(tp); return; } /* Stop asynchronous I/O. */ funsetown(&tp->t_sigio); /* TTY can be deallocated. */ dev = tp->t_dev; tp->t_dev = NULL; tty_unlock(tp); if (dev != NULL) { sx_xlock(&tty_list_sx); TAILQ_REMOVE(&tty_list, tp, t_list); tty_list_count--; sx_xunlock(&tty_list_sx); destroy_dev_sched_cb(dev, tty_dealloc, tp); } } void tty_rel_pgrp(struct tty *tp, struct pgrp *pg) { MPASS(tp->t_sessioncnt > 0); tty_assert_locked(tp); if (tp->t_pgrp == pg) tp->t_pgrp = NULL; tty_unlock(tp); } void tty_rel_sess(struct tty *tp, struct session *sess) { MPASS(tp->t_sessioncnt > 0); /* Current session has left. */ if (tp->t_session == sess) { tp->t_session = NULL; MPASS(tp->t_pgrp == NULL); } tp->t_sessioncnt--; tty_rel_free(tp); } void tty_rel_gone(struct tty *tp) { tty_assert_locked(tp); MPASS(!tty_gone(tp)); /* Simulate carrier removal. */ ttydisc_modem(tp, 0); /* Wake up all blocked threads. */ tty_wakeup(tp, FREAD|FWRITE); cv_broadcast(&tp->t_bgwait); cv_broadcast(&tp->t_dcdwait); tp->t_flags |= TF_GONE; tty_rel_free(tp); } static int tty_drop_ctty(struct tty *tp, struct proc *p) { struct session *session; struct vnode *vp; /* * This looks terrible, but it's generally safe as long as the tty * hasn't gone away while we had the lock dropped. All of our sanity * checking that this operation is OK happens after we've picked it back * up, so other state changes are generally not fatal and the potential * for this particular operation to happen out-of-order in a * multithreaded scenario is likely a non-issue. */ tty_unlock(tp); sx_xlock(&proctree_lock); tty_lock(tp); if (tty_gone(tp)) { sx_xunlock(&proctree_lock); return (ENODEV); } /* * If the session doesn't have a controlling TTY, or if we weren't * invoked on the controlling TTY, we'll return ENOIOCTL as we've * historically done. */ session = p->p_session; if (session->s_ttyp == NULL || session->s_ttyp != tp) { sx_xunlock(&proctree_lock); return (ENOTTY); } if (!SESS_LEADER(p)) { sx_xunlock(&proctree_lock); return (EPERM); } PROC_LOCK(p); SESS_LOCK(session); vp = session->s_ttyvp; session->s_ttyp = NULL; session->s_ttyvp = NULL; session->s_ttydp = NULL; SESS_UNLOCK(session); tp->t_sessioncnt--; p->p_flag &= ~P_CONTROLT; PROC_UNLOCK(p); sx_xunlock(&proctree_lock); /* * If we did have a vnode, release our reference. Ordinarily we manage * these at the devfs layer, but we can't necessarily know that we were * invoked on the vnode referenced in the session (i.e. the vnode we * hold a reference to). We explicitly don't check VBAD/VIRF_DOOMED here * to avoid a vnode leak -- in circumstances elsewhere where we'd hit a * VIRF_DOOMED vnode, release has been deferred until the controlling TTY * is either changed or released. */ if (vp != NULL) devfs_ctty_unref(vp); return (0); } /* * Exposing information about current TTY's through sysctl */ static void tty_to_xtty(struct tty *tp, struct xtty *xt) { tty_assert_locked(tp); xt->xt_size = sizeof(struct xtty); xt->xt_insize = ttyinq_getsize(&tp->t_inq); xt->xt_incc = ttyinq_bytescanonicalized(&tp->t_inq); xt->xt_inlc = ttyinq_bytesline(&tp->t_inq); xt->xt_inlow = tp->t_inlow; xt->xt_outsize = ttyoutq_getsize(&tp->t_outq); xt->xt_outcc = ttyoutq_bytesused(&tp->t_outq); xt->xt_outlow = tp->t_outlow; xt->xt_column = tp->t_column; xt->xt_pgid = tp->t_pgrp ? tp->t_pgrp->pg_id : 0; xt->xt_sid = tp->t_session ? tp->t_session->s_sid : 0; xt->xt_flags = tp->t_flags; xt->xt_dev = tp->t_dev ? dev2udev(tp->t_dev) : (uint32_t)NODEV; } static int sysctl_kern_ttys(SYSCTL_HANDLER_ARGS) { unsigned long lsize; struct xtty *xtlist, *xt; struct tty *tp; int error; sx_slock(&tty_list_sx); lsize = tty_list_count * sizeof(struct xtty); if (lsize == 0) { sx_sunlock(&tty_list_sx); return (0); } xtlist = xt = malloc(lsize, M_TTY, M_WAITOK); TAILQ_FOREACH(tp, &tty_list, t_list) { tty_lock(tp); tty_to_xtty(tp, xt); tty_unlock(tp); xt++; } sx_sunlock(&tty_list_sx); error = SYSCTL_OUT(req, xtlist, lsize); free(xtlist, M_TTY); return (error); } SYSCTL_PROC(_kern, OID_AUTO, ttys, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 0, 0, sysctl_kern_ttys, "S,xtty", "List of TTYs"); /* * Device node creation. Device has been set up, now we can expose it to * the user. */ int tty_makedevf(struct tty *tp, struct ucred *cred, int flags, const char *fmt, ...) { va_list ap; struct make_dev_args args; struct cdev *dev, *init, *lock, *cua, *cinit, *clock; const char *prefix = "tty"; char name[SPECNAMELEN - 3]; /* for "tty" and "cua". */ uid_t uid; gid_t gid; mode_t mode; int error; /* Remove "tty" prefix from devices like PTY's. */ if (tp->t_flags & TF_NOPREFIX) prefix = ""; va_start(ap, fmt); vsnrprintf(name, sizeof name, 32, fmt, ap); va_end(ap); if (cred == NULL) { /* System device. */ uid = UID_ROOT; gid = GID_WHEEL; mode = S_IRUSR|S_IWUSR; } else { /* User device. */ uid = cred->cr_ruid; gid = GID_TTY; mode = S_IRUSR|S_IWUSR|S_IWGRP; } flags = flags & TTYMK_CLONING ? MAKEDEV_REF : 0; flags |= MAKEDEV_CHECKNAME; /* Master call-in device. */ make_dev_args_init(&args); args.mda_flags = flags; args.mda_devsw = &ttydev_cdevsw; args.mda_cr = cred; args.mda_uid = uid; args.mda_gid = gid; args.mda_mode = mode; args.mda_si_drv1 = tp; error = make_dev_s(&args, &dev, "%s%s", prefix, name); if (error != 0) return (error); tp->t_dev = dev; init = lock = cua = cinit = clock = NULL; /* Slave call-in devices. */ if (tp->t_flags & TF_INITLOCK) { args.mda_devsw = &ttyil_cdevsw; args.mda_unit = TTYUNIT_INIT; args.mda_si_drv1 = tp; args.mda_si_drv2 = &tp->t_termios_init_in; error = make_dev_s(&args, &init, "%s%s.init", prefix, name); if (error != 0) goto fail; dev_depends(dev, init); args.mda_unit = TTYUNIT_LOCK; args.mda_si_drv2 = &tp->t_termios_lock_in; error = make_dev_s(&args, &lock, "%s%s.lock", prefix, name); if (error != 0) goto fail; dev_depends(dev, lock); } /* Call-out devices. */ if (tp->t_flags & TF_CALLOUT) { make_dev_args_init(&args); args.mda_flags = flags; args.mda_devsw = &ttydev_cdevsw; args.mda_cr = cred; args.mda_uid = UID_UUCP; args.mda_gid = GID_DIALER; args.mda_mode = 0660; args.mda_unit = TTYUNIT_CALLOUT; args.mda_si_drv1 = tp; error = make_dev_s(&args, &cua, "cua%s", name); if (error != 0) goto fail; dev_depends(dev, cua); /* Slave call-out devices. */ if (tp->t_flags & TF_INITLOCK) { args.mda_devsw = &ttyil_cdevsw; args.mda_unit = TTYUNIT_CALLOUT | TTYUNIT_INIT; args.mda_si_drv2 = &tp->t_termios_init_out; error = make_dev_s(&args, &cinit, "cua%s.init", name); if (error != 0) goto fail; dev_depends(dev, cinit); args.mda_unit = TTYUNIT_CALLOUT | TTYUNIT_LOCK; args.mda_si_drv2 = &tp->t_termios_lock_out; error = make_dev_s(&args, &clock, "cua%s.lock", name); if (error != 0) goto fail; dev_depends(dev, clock); } } sx_xlock(&tty_list_sx); TAILQ_INSERT_TAIL(&tty_list, tp, t_list); tty_list_count++; sx_xunlock(&tty_list_sx); return (0); fail: destroy_dev(dev); if (init) destroy_dev(init); if (lock) destroy_dev(lock); if (cinit) destroy_dev(cinit); if (clock) destroy_dev(clock); return (error); } /* * Signalling processes. */ void tty_signal_sessleader(struct tty *tp, int sig) { struct proc *p; struct session *s; tty_assert_locked(tp); MPASS(sig >= 1 && sig < NSIG); /* Make signals start output again. */ tp->t_flags &= ~TF_STOPPED; tp->t_termios.c_lflag &= ~FLUSHO; /* * Load s_leader exactly once to avoid race where s_leader is * set to NULL by a concurrent invocation of killjobc() by the * session leader. Note that we are not holding t_session's * lock for the read. */ if ((s = tp->t_session) != NULL && (p = atomic_load_ptr(&s->s_leader)) != NULL) { PROC_LOCK(p); kern_psignal(p, sig); PROC_UNLOCK(p); } } void tty_signal_pgrp(struct tty *tp, int sig) { ksiginfo_t ksi; tty_assert_locked(tp); MPASS(sig >= 1 && sig < NSIG); /* Make signals start output again. */ tp->t_flags &= ~TF_STOPPED; tp->t_termios.c_lflag &= ~FLUSHO; if (sig == SIGINFO && !(tp->t_termios.c_lflag & NOKERNINFO)) tty_info(tp); if (tp->t_pgrp != NULL) { ksiginfo_init(&ksi); ksi.ksi_signo = sig; ksi.ksi_code = SI_KERNEL; PGRP_LOCK(tp->t_pgrp); pgsignal(tp->t_pgrp, sig, 1, &ksi); PGRP_UNLOCK(tp->t_pgrp); } } void tty_wakeup(struct tty *tp, int flags) { if (tp->t_flags & TF_ASYNC && tp->t_sigio != NULL) pgsigio(&tp->t_sigio, SIGIO, (tp->t_session != NULL)); if (flags & FWRITE) { cv_broadcast(&tp->t_outwait); selwakeup(&tp->t_outpoll); KNOTE_LOCKED(&tp->t_outpoll.si_note, 0); } if (flags & FREAD) { cv_broadcast(&tp->t_inwait); selwakeup(&tp->t_inpoll); KNOTE_LOCKED(&tp->t_inpoll.si_note, 0); } } int tty_wait(struct tty *tp, struct cv *cv) { int error; int revokecnt = tp->t_revokecnt; tty_lock_assert(tp, MA_OWNED|MA_NOTRECURSED); MPASS(!tty_gone(tp)); error = cv_wait_sig(cv, tp->t_mtx); /* Bail out when the device slipped away. */ if (tty_gone(tp)) return (ENXIO); /* Restart the system call when we may have been revoked. */ if (tp->t_revokecnt != revokecnt) return (ERESTART); return (error); } int tty_timedwait(struct tty *tp, struct cv *cv, int hz) { int error; int revokecnt = tp->t_revokecnt; tty_lock_assert(tp, MA_OWNED|MA_NOTRECURSED); MPASS(!tty_gone(tp)); error = cv_timedwait_sig(cv, tp->t_mtx, hz); /* Bail out when the device slipped away. */ if (tty_gone(tp)) return (ENXIO); /* Restart the system call when we may have been revoked. */ if (tp->t_revokecnt != revokecnt) return (ERESTART); return (error); } void tty_flush(struct tty *tp, int flags) { if (flags & FWRITE) { tp->t_flags &= ~TF_HIWAT_OUT; ttyoutq_flush(&tp->t_outq); tty_wakeup(tp, FWRITE); if (!tty_gone(tp)) { ttydevsw_outwakeup(tp); ttydevsw_pktnotify(tp, TIOCPKT_FLUSHWRITE); } } if (flags & FREAD) { tty_hiwat_in_unblock(tp); ttyinq_flush(&tp->t_inq); tty_wakeup(tp, FREAD); if (!tty_gone(tp)) { ttydevsw_inwakeup(tp); ttydevsw_pktnotify(tp, TIOCPKT_FLUSHREAD); } } } void tty_set_winsize(struct tty *tp, const struct winsize *wsz) { if (memcmp(&tp->t_winsize, wsz, sizeof(*wsz)) == 0) return; tp->t_winsize = *wsz; tty_signal_pgrp(tp, SIGWINCH); } static int tty_generic_ioctl(struct tty *tp, u_long cmd, void *data, int fflag, struct thread *td) { int error; switch (cmd) { /* * Modem commands. * The SER_* and TIOCM_* flags are the same, but one bit * shifted. I don't know why. */ case TIOCSDTR: ttydevsw_modem(tp, SER_DTR, 0); return (0); case TIOCCDTR: ttydevsw_modem(tp, 0, SER_DTR); return (0); case TIOCMSET: { int bits = *(int *)data; ttydevsw_modem(tp, (bits & (TIOCM_DTR | TIOCM_RTS)) >> 1, ((~bits) & (TIOCM_DTR | TIOCM_RTS)) >> 1); return (0); } case TIOCMBIS: { int bits = *(int *)data; ttydevsw_modem(tp, (bits & (TIOCM_DTR | TIOCM_RTS)) >> 1, 0); return (0); } case TIOCMBIC: { int bits = *(int *)data; ttydevsw_modem(tp, 0, (bits & (TIOCM_DTR | TIOCM_RTS)) >> 1); return (0); } case TIOCMGET: *(int *)data = TIOCM_LE + (ttydevsw_modem(tp, 0, 0) << 1); return (0); case FIOASYNC: if (*(int *)data) tp->t_flags |= TF_ASYNC; else tp->t_flags &= ~TF_ASYNC; return (0); case FIONBIO: /* This device supports non-blocking operation. */ return (0); case FIONREAD: *(int *)data = ttyinq_bytescanonicalized(&tp->t_inq); return (0); case FIONWRITE: case TIOCOUTQ: *(int *)data = ttyoutq_bytesused(&tp->t_outq); return (0); case FIOSETOWN: if (tp->t_session != NULL && !tty_is_ctty(tp, td->td_proc)) /* Not allowed to set ownership. */ return (ENOTTY); /* Temporarily unlock the TTY to set ownership. */ tty_unlock(tp); error = fsetown(*(int *)data, &tp->t_sigio); tty_lock(tp); return (error); case FIOGETOWN: if (tp->t_session != NULL && !tty_is_ctty(tp, td->td_proc)) /* Not allowed to set ownership. */ return (ENOTTY); /* Get ownership. */ *(int *)data = fgetown(&tp->t_sigio); return (0); case TIOCGETA: /* Obtain terminal flags through tcgetattr(). */ *(struct termios*)data = tp->t_termios; return (0); case TIOCSETA: case TIOCSETAW: case TIOCSETAF: { struct termios *t = data; /* * Who makes up these funny rules? According to POSIX, * input baud rate is set equal to the output baud rate * when zero. */ if (t->c_ispeed == 0) t->c_ispeed = t->c_ospeed; /* Discard any unsupported bits. */ t->c_iflag &= TTYSUP_IFLAG; t->c_oflag &= TTYSUP_OFLAG; t->c_lflag &= TTYSUP_LFLAG; t->c_cflag &= TTYSUP_CFLAG; /* Set terminal flags through tcsetattr(). */ if (cmd == TIOCSETAW || cmd == TIOCSETAF) { error = tty_drain(tp, 0); if (error) return (error); if (cmd == TIOCSETAF) tty_flush(tp, FREAD); } /* * Only call param() when the flags really change. */ if ((t->c_cflag & CIGNORE) == 0 && (tp->t_termios.c_cflag != t->c_cflag || ((tp->t_termios.c_iflag ^ t->c_iflag) & (IXON|IXOFF|IXANY)) || tp->t_termios.c_ispeed != t->c_ispeed || tp->t_termios.c_ospeed != t->c_ospeed)) { error = ttydevsw_param(tp, t); if (error) return (error); /* XXX: CLOCAL? */ tp->t_termios.c_cflag = t->c_cflag & ~CIGNORE; tp->t_termios.c_ispeed = t->c_ispeed; tp->t_termios.c_ospeed = t->c_ospeed; /* Baud rate has changed - update watermarks. */ error = tty_watermarks(tp); if (error) return (error); } /* Copy new non-device driver parameters. */ tp->t_termios.c_iflag = t->c_iflag; tp->t_termios.c_oflag = t->c_oflag; tp->t_termios.c_lflag = t->c_lflag; memcpy(&tp->t_termios.c_cc, t->c_cc, sizeof t->c_cc); ttydisc_optimize(tp); if ((t->c_lflag & ICANON) == 0) { /* * When in non-canonical mode, wake up all * readers. Canonicalize any partial input. VMIN * and VTIME could also be adjusted. */ ttyinq_canonicalize(&tp->t_inq); tty_wakeup(tp, FREAD); } /* * For packet mode: notify the PTY consumer that VSTOP * and VSTART may have been changed. */ if (tp->t_termios.c_iflag & IXON && tp->t_termios.c_cc[VSTOP] == CTRL('S') && tp->t_termios.c_cc[VSTART] == CTRL('Q')) ttydevsw_pktnotify(tp, TIOCPKT_DOSTOP); else ttydevsw_pktnotify(tp, TIOCPKT_NOSTOP); return (0); } case TIOCGETD: /* For compatibility - we only support TTYDISC. */ *(int *)data = TTYDISC; return (0); case TIOCGPGRP: if (!tty_is_ctty(tp, td->td_proc)) return (ENOTTY); if (tp->t_pgrp != NULL) *(int *)data = tp->t_pgrp->pg_id; else *(int *)data = NO_PID; return (0); case TIOCGSID: if (!tty_is_ctty(tp, td->td_proc)) return (ENOTTY); MPASS(tp->t_session); *(int *)data = tp->t_session->s_sid; return (0); case TIOCNOTTY: return (tty_drop_ctty(tp, td->td_proc)); case TIOCSCTTY: { struct proc *p = td->td_proc; /* XXX: This looks awful. */ tty_unlock(tp); sx_xlock(&proctree_lock); tty_lock(tp); if (!SESS_LEADER(p)) { /* Only the session leader may do this. */ sx_xunlock(&proctree_lock); return (EPERM); } if (tp->t_session != NULL && tp->t_session == p->p_session) { /* This is already our controlling TTY. */ sx_xunlock(&proctree_lock); return (0); } if (p->p_session->s_ttyp != NULL || (tp->t_session != NULL && tp->t_session->s_ttyvp != NULL && tp->t_session->s_ttyvp->v_type != VBAD)) { /* * There is already a relation between a TTY and * a session, or the caller is not the session * leader. * * Allow the TTY to be stolen when the vnode is * invalid, but the reference to the TTY is * still active. This allows immediate reuse of * TTYs of which the session leader has been * killed or the TTY revoked. */ sx_xunlock(&proctree_lock); return (EPERM); } /* Connect the session to the TTY. */ tp->t_session = p->p_session; tp->t_session->s_ttyp = tp; tp->t_sessioncnt++; /* Assign foreground process group. */ tp->t_pgrp = p->p_pgrp; PROC_LOCK(p); p->p_flag |= P_CONTROLT; PROC_UNLOCK(p); sx_xunlock(&proctree_lock); return (0); } case TIOCSPGRP: { struct pgrp *pg; /* * XXX: Temporarily unlock the TTY to locate the process * group. This code would be lot nicer if we would ever * decompose proctree_lock. */ tty_unlock(tp); sx_slock(&proctree_lock); pg = pgfind(*(int *)data); if (pg != NULL) PGRP_UNLOCK(pg); if (pg == NULL || pg->pg_session != td->td_proc->p_session) { sx_sunlock(&proctree_lock); tty_lock(tp); return (EPERM); } tty_lock(tp); /* * Determine if this TTY is the controlling TTY after * relocking the TTY. */ if (!tty_is_ctty(tp, td->td_proc)) { sx_sunlock(&proctree_lock); return (ENOTTY); } tp->t_pgrp = pg; sx_sunlock(&proctree_lock); /* Wake up the background process groups. */ cv_broadcast(&tp->t_bgwait); return (0); } case TIOCFLUSH: { int flags = *(int *)data; if (flags == 0) flags = (FREAD|FWRITE); else flags &= (FREAD|FWRITE); tty_flush(tp, flags); return (0); } case TIOCDRAIN: /* Drain TTY output. */ return tty_drain(tp, 0); case TIOCGDRAINWAIT: *(int *)data = tp->t_drainwait; return (0); case TIOCSDRAINWAIT: error = priv_check(td, PRIV_TTY_DRAINWAIT); if (error == 0) tp->t_drainwait = *(int *)data; return (error); case TIOCCONS: /* Set terminal as console TTY. */ if (*(int *)data) { error = priv_check(td, PRIV_TTY_CONSOLE); if (error) return (error); /* * XXX: constty should really need to be locked! * XXX: allow disconnected constty's to be stolen! */ if (constty == tp) return (0); if (constty != NULL) return (EBUSY); tty_unlock(tp); constty_set(tp); tty_lock(tp); } else if (constty == tp) { constty_clear(); } return (0); case TIOCGWINSZ: /* Obtain window size. */ *(struct winsize*)data = tp->t_winsize; return (0); case TIOCSWINSZ: /* Set window size. */ tty_set_winsize(tp, data); return (0); case TIOCEXCL: tp->t_flags |= TF_EXCLUDE; return (0); case TIOCNXCL: tp->t_flags &= ~TF_EXCLUDE; return (0); case TIOCSTOP: tp->t_flags |= TF_STOPPED; ttydevsw_pktnotify(tp, TIOCPKT_STOP); return (0); case TIOCSTART: tp->t_flags &= ~TF_STOPPED; tp->t_termios.c_lflag &= ~FLUSHO; ttydevsw_outwakeup(tp); ttydevsw_pktnotify(tp, TIOCPKT_START); return (0); case TIOCSTAT: tty_info(tp); return (0); case TIOCSTI: if ((fflag & FREAD) == 0 && priv_check(td, PRIV_TTY_STI)) return (EPERM); if (!tty_is_ctty(tp, td->td_proc) && priv_check(td, PRIV_TTY_STI)) return (EACCES); ttydisc_rint(tp, *(char *)data, 0); ttydisc_rint_done(tp); return (0); } #ifdef COMPAT_43TTY return tty_ioctl_compat(tp, cmd, data, fflag, td); #else /* !COMPAT_43TTY */ return (ENOIOCTL); #endif /* COMPAT_43TTY */ } int tty_ioctl(struct tty *tp, u_long cmd, void *data, int fflag, struct thread *td) { int error; tty_assert_locked(tp); if (tty_gone(tp)) return (ENXIO); error = ttydevsw_ioctl(tp, cmd, data, td); if (error == ENOIOCTL) error = tty_generic_ioctl(tp, cmd, data, fflag, td); return (error); } dev_t tty_udev(struct tty *tp) { if (tp->t_dev) return (dev2udev(tp->t_dev)); else return (NODEV); } int tty_checkoutq(struct tty *tp) { /* 256 bytes should be enough to print a log message. */ return (ttyoutq_bytesleft(&tp->t_outq) >= 256); } void tty_hiwat_in_block(struct tty *tp) { if ((tp->t_flags & TF_HIWAT_IN) == 0 && tp->t_termios.c_iflag & IXOFF && tp->t_termios.c_cc[VSTOP] != _POSIX_VDISABLE) { /* * Input flow control. Only enter the high watermark when we * can successfully store the VSTOP character. */ if (ttyoutq_write_nofrag(&tp->t_outq, &tp->t_termios.c_cc[VSTOP], 1) == 0) tp->t_flags |= TF_HIWAT_IN; } else { /* No input flow control. */ tp->t_flags |= TF_HIWAT_IN; } } void tty_hiwat_in_unblock(struct tty *tp) { if (tp->t_flags & TF_HIWAT_IN && tp->t_termios.c_iflag & IXOFF && tp->t_termios.c_cc[VSTART] != _POSIX_VDISABLE) { /* * Input flow control. Only leave the high watermark when we * can successfully store the VSTART character. */ if (ttyoutq_write_nofrag(&tp->t_outq, &tp->t_termios.c_cc[VSTART], 1) == 0) tp->t_flags &= ~TF_HIWAT_IN; } else { /* No input flow control. */ tp->t_flags &= ~TF_HIWAT_IN; } if (!tty_gone(tp)) ttydevsw_inwakeup(tp); } /* * TTY hooks interface. */ static int ttyhook_defrint(struct tty *tp, char c, int flags) { if (ttyhook_rint_bypass(tp, &c, 1) != 1) return (-1); return (0); } int ttyhook_register(struct tty **rtp, struct proc *p, int fd, struct ttyhook *th, void *softc) { struct tty *tp; struct file *fp; struct cdev *dev; struct cdevsw *cdp; struct filedesc *fdp; cap_rights_t rights; int error, ref; /* Validate the file descriptor. */ fdp = p->p_fd; - error = fget_unlocked(fdp, fd, cap_rights_init(&rights, CAP_TTYHOOK), + error = fget_unlocked(fdp, fd, cap_rights_init_one(&rights, CAP_TTYHOOK), &fp); if (error != 0) return (error); if (fp->f_ops == &badfileops) { error = EBADF; goto done1; } /* * Make sure the vnode is bound to a character device. * Unlocked check for the vnode type is ok there, because we * only shall prevent calling devvn_refthread on the file that * never has been opened over a character device. */ if (fp->f_type != DTYPE_VNODE || fp->f_vnode->v_type != VCHR) { error = EINVAL; goto done1; } /* Make sure it is a TTY. */ cdp = devvn_refthread(fp->f_vnode, &dev, &ref); if (cdp == NULL) { error = ENXIO; goto done1; } if (dev != fp->f_data) { error = ENXIO; goto done2; } if (cdp != &ttydev_cdevsw) { error = ENOTTY; goto done2; } tp = dev->si_drv1; /* Try to attach the hook to the TTY. */ error = EBUSY; tty_lock(tp); MPASS((tp->t_hook == NULL) == ((tp->t_flags & TF_HOOK) == 0)); if (tp->t_flags & TF_HOOK) goto done3; tp->t_flags |= TF_HOOK; tp->t_hook = th; tp->t_hooksoftc = softc; *rtp = tp; error = 0; /* Maybe we can switch into bypass mode now. */ ttydisc_optimize(tp); /* Silently convert rint() calls to rint_bypass() when possible. */ if (!ttyhook_hashook(tp, rint) && ttyhook_hashook(tp, rint_bypass)) th->th_rint = ttyhook_defrint; done3: tty_unlock(tp); done2: dev_relthread(dev, ref); done1: fdrop(fp, curthread); return (error); } void ttyhook_unregister(struct tty *tp) { tty_assert_locked(tp); MPASS(tp->t_flags & TF_HOOK); /* Disconnect the hook. */ tp->t_flags &= ~TF_HOOK; tp->t_hook = NULL; /* Maybe we need to leave bypass mode. */ ttydisc_optimize(tp); /* Maybe deallocate the TTY as well. */ tty_rel_free(tp); } /* * /dev/console handling. */ static int ttyconsdev_open(struct cdev *dev, int oflags, int devtype, struct thread *td) { struct tty *tp; /* System has no console device. */ if (dev_console_filename == NULL) return (ENXIO); /* Look up corresponding TTY by device name. */ sx_slock(&tty_list_sx); TAILQ_FOREACH(tp, &tty_list, t_list) { if (strcmp(dev_console_filename, tty_devname(tp)) == 0) { dev_console->si_drv1 = tp; break; } } sx_sunlock(&tty_list_sx); /* System console has no TTY associated. */ if (dev_console->si_drv1 == NULL) return (ENXIO); return (ttydev_open(dev, oflags, devtype, td)); } static int ttyconsdev_write(struct cdev *dev, struct uio *uio, int ioflag) { log_console(uio); return (ttydev_write(dev, uio, ioflag)); } /* * /dev/console is a little different than normal TTY's. When opened, * it determines which TTY to use. When data gets written to it, it * will be logged in the kernel message buffer. */ static struct cdevsw ttyconsdev_cdevsw = { .d_version = D_VERSION, .d_open = ttyconsdev_open, .d_close = ttydev_close, .d_read = ttydev_read, .d_write = ttyconsdev_write, .d_ioctl = ttydev_ioctl, .d_kqfilter = ttydev_kqfilter, .d_poll = ttydev_poll, .d_mmap = ttydev_mmap, .d_name = "ttyconsdev", .d_flags = D_TTY, }; static void ttyconsdev_init(void *unused __unused) { dev_console = make_dev_credf(MAKEDEV_ETERNAL, &ttyconsdev_cdevsw, 0, NULL, UID_ROOT, GID_WHEEL, 0600, "console"); } SYSINIT(tty, SI_SUB_DRIVERS, SI_ORDER_FIRST, ttyconsdev_init, NULL); void ttyconsdev_select(const char *name) { dev_console_filename = name; } /* * Debugging routines. */ #include "opt_ddb.h" #ifdef DDB #include #include static const struct { int flag; char val; } ttystates[] = { #if 0 { TF_NOPREFIX, 'N' }, #endif { TF_INITLOCK, 'I' }, { TF_CALLOUT, 'C' }, /* Keep these together -> 'Oi' and 'Oo'. */ { TF_OPENED, 'O' }, { TF_OPENED_IN, 'i' }, { TF_OPENED_OUT, 'o' }, { TF_OPENED_CONS, 'c' }, { TF_GONE, 'G' }, { TF_OPENCLOSE, 'B' }, { TF_ASYNC, 'Y' }, { TF_LITERAL, 'L' }, /* Keep these together -> 'Hi' and 'Ho'. */ { TF_HIWAT, 'H' }, { TF_HIWAT_IN, 'i' }, { TF_HIWAT_OUT, 'o' }, { TF_STOPPED, 'S' }, { TF_EXCLUDE, 'X' }, { TF_BYPASS, 'l' }, { TF_ZOMBIE, 'Z' }, { TF_HOOK, 's' }, /* Keep these together -> 'bi' and 'bo'. */ { TF_BUSY, 'b' }, { TF_BUSY_IN, 'i' }, { TF_BUSY_OUT, 'o' }, { 0, '\0'}, }; #define TTY_FLAG_BITS \ "\20\1NOPREFIX\2INITLOCK\3CALLOUT\4OPENED_IN" \ "\5OPENED_OUT\6OPENED_CONS\7GONE\10OPENCLOSE" \ "\11ASYNC\12LITERAL\13HIWAT_IN\14HIWAT_OUT" \ "\15STOPPED\16EXCLUDE\17BYPASS\20ZOMBIE" \ "\21HOOK\22BUSY_IN\23BUSY_OUT" #define DB_PRINTSYM(name, addr) \ db_printf("%s " #name ": ", sep); \ db_printsym((db_addr_t) addr, DB_STGY_ANY); \ db_printf("\n"); static void _db_show_devsw(const char *sep, const struct ttydevsw *tsw) { db_printf("%sdevsw: ", sep); db_printsym((db_addr_t)tsw, DB_STGY_ANY); db_printf(" (%p)\n", tsw); DB_PRINTSYM(open, tsw->tsw_open); DB_PRINTSYM(close, tsw->tsw_close); DB_PRINTSYM(outwakeup, tsw->tsw_outwakeup); DB_PRINTSYM(inwakeup, tsw->tsw_inwakeup); DB_PRINTSYM(ioctl, tsw->tsw_ioctl); DB_PRINTSYM(param, tsw->tsw_param); DB_PRINTSYM(modem, tsw->tsw_modem); DB_PRINTSYM(mmap, tsw->tsw_mmap); DB_PRINTSYM(pktnotify, tsw->tsw_pktnotify); DB_PRINTSYM(free, tsw->tsw_free); } static void _db_show_hooks(const char *sep, const struct ttyhook *th) { db_printf("%shook: ", sep); db_printsym((db_addr_t)th, DB_STGY_ANY); db_printf(" (%p)\n", th); if (th == NULL) return; DB_PRINTSYM(rint, th->th_rint); DB_PRINTSYM(rint_bypass, th->th_rint_bypass); DB_PRINTSYM(rint_done, th->th_rint_done); DB_PRINTSYM(rint_poll, th->th_rint_poll); DB_PRINTSYM(getc_inject, th->th_getc_inject); DB_PRINTSYM(getc_capture, th->th_getc_capture); DB_PRINTSYM(getc_poll, th->th_getc_poll); DB_PRINTSYM(close, th->th_close); } static void _db_show_termios(const char *name, const struct termios *t) { db_printf("%s: iflag 0x%x oflag 0x%x cflag 0x%x " "lflag 0x%x ispeed %u ospeed %u\n", name, t->c_iflag, t->c_oflag, t->c_cflag, t->c_lflag, t->c_ispeed, t->c_ospeed); } /* DDB command to show TTY statistics. */ DB_SHOW_COMMAND(tty, db_show_tty) { struct tty *tp; if (!have_addr) { db_printf("usage: show tty \n"); return; } tp = (struct tty *)addr; db_printf("%p: %s\n", tp, tty_devname(tp)); db_printf("\tmtx: %p\n", tp->t_mtx); db_printf("\tflags: 0x%b\n", tp->t_flags, TTY_FLAG_BITS); db_printf("\trevokecnt: %u\n", tp->t_revokecnt); /* Buffering mechanisms. */ db_printf("\tinq: %p begin %u linestart %u reprint %u end %u " "nblocks %u quota %u\n", &tp->t_inq, tp->t_inq.ti_begin, tp->t_inq.ti_linestart, tp->t_inq.ti_reprint, tp->t_inq.ti_end, tp->t_inq.ti_nblocks, tp->t_inq.ti_quota); db_printf("\toutq: %p begin %u end %u nblocks %u quota %u\n", &tp->t_outq, tp->t_outq.to_begin, tp->t_outq.to_end, tp->t_outq.to_nblocks, tp->t_outq.to_quota); db_printf("\tinlow: %zu\n", tp->t_inlow); db_printf("\toutlow: %zu\n", tp->t_outlow); _db_show_termios("\ttermios", &tp->t_termios); db_printf("\twinsize: row %u col %u xpixel %u ypixel %u\n", tp->t_winsize.ws_row, tp->t_winsize.ws_col, tp->t_winsize.ws_xpixel, tp->t_winsize.ws_ypixel); db_printf("\tcolumn: %u\n", tp->t_column); db_printf("\twritepos: %u\n", tp->t_writepos); db_printf("\tcompatflags: 0x%x\n", tp->t_compatflags); /* Init/lock-state devices. */ _db_show_termios("\ttermios_init_in", &tp->t_termios_init_in); _db_show_termios("\ttermios_init_out", &tp->t_termios_init_out); _db_show_termios("\ttermios_lock_in", &tp->t_termios_lock_in); _db_show_termios("\ttermios_lock_out", &tp->t_termios_lock_out); /* Hooks */ _db_show_devsw("\t", tp->t_devsw); _db_show_hooks("\t", tp->t_hook); /* Process info. */ db_printf("\tpgrp: %p gid %d\n", tp->t_pgrp, tp->t_pgrp ? tp->t_pgrp->pg_id : 0); db_printf("\tsession: %p", tp->t_session); if (tp->t_session != NULL) db_printf(" count %u leader %p tty %p sid %d login %s", tp->t_session->s_count, tp->t_session->s_leader, tp->t_session->s_ttyp, tp->t_session->s_sid, tp->t_session->s_login); db_printf("\n"); db_printf("\tsessioncnt: %u\n", tp->t_sessioncnt); db_printf("\tdevswsoftc: %p\n", tp->t_devswsoftc); db_printf("\thooksoftc: %p\n", tp->t_hooksoftc); db_printf("\tdev: %p\n", tp->t_dev); } /* DDB command to list TTYs. */ DB_SHOW_ALL_COMMAND(ttys, db_show_all_ttys) { struct tty *tp; size_t isiz, osiz; int i, j; /* Make the output look like `pstat -t'. */ db_printf("PTR "); #if defined(__LP64__) db_printf(" "); #endif db_printf(" LINE INQ CAN LIN LOW OUTQ USE LOW " "COL SESS PGID STATE\n"); TAILQ_FOREACH(tp, &tty_list, t_list) { isiz = tp->t_inq.ti_nblocks * TTYINQ_DATASIZE; osiz = tp->t_outq.to_nblocks * TTYOUTQ_DATASIZE; db_printf("%p %10s %5zu %4u %4u %4zu %5zu %4u %4zu %5u %5d " "%5d ", tp, tty_devname(tp), isiz, tp->t_inq.ti_linestart - tp->t_inq.ti_begin, tp->t_inq.ti_end - tp->t_inq.ti_linestart, isiz - tp->t_inlow, osiz, tp->t_outq.to_end - tp->t_outq.to_begin, osiz - tp->t_outlow, MIN(tp->t_column, 99999), tp->t_session ? tp->t_session->s_sid : 0, tp->t_pgrp ? tp->t_pgrp->pg_id : 0); /* Flag bits. */ for (i = j = 0; ttystates[i].flag; i++) if (tp->t_flags & ttystates[i].flag) { db_printf("%c", ttystates[i].val); j++; } if (j == 0) db_printf("-"); db_printf("\n"); } } #endif /* DDB */ diff --git a/sys/kern/uipc_sem.c b/sys/kern/uipc_sem.c index bb47fe9a3905..db1d84696df0 100644 --- a/sys/kern/uipc_sem.c +++ b/sys/kern/uipc_sem.c @@ -1,1110 +1,1111 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2002 Alfred Perlstein * Copyright (c) 2003-2005 SPARTA, Inc. * Copyright (c) 2005, 2016-2017 Robert N. M. Watson * All rights reserved. * * This software was developed for the FreeBSD Project in part by Network * Associates Laboratories, the Security Research Division of Network * Associates, Inc. under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), * as part of the DARPA CHATS research program. * * Portions of this software were developed by BAE Systems, the University of * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent * Computing (TC) research program. * * 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_posix.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 #include #include #include #include #include #include FEATURE(p1003_1b_semaphores, "POSIX P1003.1B semaphores support"); /* * TODO * * - Resource limits? * - Replace global sem_lock with mtx_pool locks? * - Add a MAC check_create() hook for creating new named semaphores. */ #ifndef SEM_MAX #define SEM_MAX 30 #endif #ifdef SEM_DEBUG #define DP(x) printf x #else #define DP(x) #endif struct ksem_mapping { char *km_path; Fnv32_t km_fnv; struct ksem *km_ksem; LIST_ENTRY(ksem_mapping) km_link; }; static MALLOC_DEFINE(M_KSEM, "ksem", "semaphore file descriptor"); static LIST_HEAD(, ksem_mapping) *ksem_dictionary; static struct sx ksem_dict_lock; static struct mtx ksem_count_lock; static struct mtx sem_lock; static u_long ksem_hash; static int ksem_dead; #define KSEM_HASH(fnv) (&ksem_dictionary[(fnv) & ksem_hash]) static int nsems = 0; SYSCTL_DECL(_p1003_1b); SYSCTL_INT(_p1003_1b, OID_AUTO, nsems, CTLFLAG_RD, &nsems, 0, "Number of active kernel POSIX semaphores"); static int kern_sem_wait(struct thread *td, semid_t id, int tryflag, struct timespec *abstime); static int ksem_access(struct ksem *ks, struct ucred *ucred); static struct ksem *ksem_alloc(struct ucred *ucred, mode_t mode, unsigned int value); static int ksem_create(struct thread *td, const char *path, semid_t *semidp, mode_t mode, unsigned int value, int flags, int compat32); static void ksem_drop(struct ksem *ks); static int ksem_get(struct thread *td, semid_t id, cap_rights_t *rightsp, struct file **fpp); static struct ksem *ksem_hold(struct ksem *ks); static void ksem_insert(char *path, Fnv32_t fnv, struct ksem *ks); static struct ksem *ksem_lookup(char *path, Fnv32_t fnv); static void ksem_module_destroy(void); static int ksem_module_init(void); static int ksem_remove(char *path, Fnv32_t fnv, struct ucred *ucred); static int sem_modload(struct module *module, int cmd, void *arg); static fo_stat_t ksem_stat; static fo_close_t ksem_closef; static fo_chmod_t ksem_chmod; static fo_chown_t ksem_chown; static fo_fill_kinfo_t ksem_fill_kinfo; /* File descriptor operations. */ static struct fileops ksem_ops = { .fo_read = invfo_rdwr, .fo_write = invfo_rdwr, .fo_truncate = invfo_truncate, .fo_ioctl = invfo_ioctl, .fo_poll = invfo_poll, .fo_kqfilter = invfo_kqfilter, .fo_stat = ksem_stat, .fo_close = ksem_closef, .fo_chmod = ksem_chmod, .fo_chown = ksem_chown, .fo_sendfile = invfo_sendfile, .fo_fill_kinfo = ksem_fill_kinfo, .fo_flags = DFLAG_PASSABLE }; FEATURE(posix_sem, "POSIX semaphores"); static int ksem_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td) { struct ksem *ks; #ifdef MAC int error; #endif ks = fp->f_data; #ifdef MAC error = mac_posixsem_check_stat(active_cred, fp->f_cred, ks); if (error) return (error); #endif /* * Attempt to return sanish values for fstat() on a semaphore * file descriptor. */ bzero(sb, sizeof(*sb)); mtx_lock(&sem_lock); sb->st_atim = ks->ks_atime; sb->st_ctim = ks->ks_ctime; sb->st_mtim = ks->ks_mtime; sb->st_birthtim = ks->ks_birthtime; sb->st_uid = ks->ks_uid; sb->st_gid = ks->ks_gid; sb->st_mode = S_IFREG | ks->ks_mode; /* XXX */ mtx_unlock(&sem_lock); return (0); } static int ksem_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { struct ksem *ks; int error; error = 0; ks = fp->f_data; mtx_lock(&sem_lock); #ifdef MAC error = mac_posixsem_check_setmode(active_cred, ks, mode); if (error != 0) goto out; #endif error = vaccess(VREG, ks->ks_mode, ks->ks_uid, ks->ks_gid, VADMIN, active_cred); if (error != 0) goto out; ks->ks_mode = mode & ACCESSPERMS; out: mtx_unlock(&sem_lock); return (error); } static int ksem_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { struct ksem *ks; int error; error = 0; ks = fp->f_data; mtx_lock(&sem_lock); #ifdef MAC error = mac_posixsem_check_setowner(active_cred, ks, uid, gid); if (error != 0) goto out; #endif if (uid == (uid_t)-1) uid = ks->ks_uid; if (gid == (gid_t)-1) gid = ks->ks_gid; if (((uid != ks->ks_uid && uid != active_cred->cr_uid) || (gid != ks->ks_gid && !groupmember(gid, active_cred))) && (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN))) goto out; ks->ks_uid = uid; ks->ks_gid = gid; out: mtx_unlock(&sem_lock); return (error); } static int ksem_closef(struct file *fp, struct thread *td) { struct ksem *ks; ks = fp->f_data; fp->f_data = NULL; ksem_drop(ks); return (0); } static int ksem_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { const char *path, *pr_path; struct ksem *ks; size_t pr_pathlen; kif->kf_type = KF_TYPE_SEM; ks = fp->f_data; mtx_lock(&sem_lock); kif->kf_un.kf_sem.kf_sem_value = ks->ks_value; kif->kf_un.kf_sem.kf_sem_mode = S_IFREG | ks->ks_mode; /* XXX */ mtx_unlock(&sem_lock); if (ks->ks_path != NULL) { sx_slock(&ksem_dict_lock); if (ks->ks_path != NULL) { path = ks->ks_path; pr_path = curthread->td_ucred->cr_prison->pr_path; if (strcmp(pr_path, "/") != 0) { /* Return the jail-rooted pathname. */ pr_pathlen = strlen(pr_path); if (strncmp(path, pr_path, pr_pathlen) == 0 && path[pr_pathlen] == '/') path += pr_pathlen; } strlcpy(kif->kf_path, path, sizeof(kif->kf_path)); } sx_sunlock(&ksem_dict_lock); } return (0); } /* * ksem object management including creation and reference counting * routines. */ static struct ksem * ksem_alloc(struct ucred *ucred, mode_t mode, unsigned int value) { struct ksem *ks; mtx_lock(&ksem_count_lock); if (nsems == p31b_getcfg(CTL_P1003_1B_SEM_NSEMS_MAX) || ksem_dead) { mtx_unlock(&ksem_count_lock); return (NULL); } nsems++; mtx_unlock(&ksem_count_lock); ks = malloc(sizeof(*ks), M_KSEM, M_WAITOK | M_ZERO); ks->ks_uid = ucred->cr_uid; ks->ks_gid = ucred->cr_gid; ks->ks_mode = mode; ks->ks_value = value; cv_init(&ks->ks_cv, "ksem"); vfs_timestamp(&ks->ks_birthtime); ks->ks_atime = ks->ks_mtime = ks->ks_ctime = ks->ks_birthtime; refcount_init(&ks->ks_ref, 1); #ifdef MAC mac_posixsem_init(ks); mac_posixsem_create(ucred, ks); #endif return (ks); } static struct ksem * ksem_hold(struct ksem *ks) { refcount_acquire(&ks->ks_ref); return (ks); } static void ksem_drop(struct ksem *ks) { if (refcount_release(&ks->ks_ref)) { #ifdef MAC mac_posixsem_destroy(ks); #endif cv_destroy(&ks->ks_cv); free(ks, M_KSEM); mtx_lock(&ksem_count_lock); nsems--; mtx_unlock(&ksem_count_lock); } } /* * Determine if the credentials have sufficient permissions for read * and write access. */ static int ksem_access(struct ksem *ks, struct ucred *ucred) { int error; error = vaccess(VREG, ks->ks_mode, ks->ks_uid, ks->ks_gid, VREAD | VWRITE, ucred); if (error) error = priv_check_cred(ucred, PRIV_SEM_WRITE); return (error); } /* * Dictionary management. We maintain an in-kernel dictionary to map * paths to semaphore objects. We use the FNV hash on the path to * store the mappings in a hash table. */ static struct ksem * ksem_lookup(char *path, Fnv32_t fnv) { struct ksem_mapping *map; LIST_FOREACH(map, KSEM_HASH(fnv), km_link) { if (map->km_fnv != fnv) continue; if (strcmp(map->km_path, path) == 0) return (map->km_ksem); } return (NULL); } static void ksem_insert(char *path, Fnv32_t fnv, struct ksem *ks) { struct ksem_mapping *map; map = malloc(sizeof(struct ksem_mapping), M_KSEM, M_WAITOK); map->km_path = path; map->km_fnv = fnv; map->km_ksem = ksem_hold(ks); ks->ks_path = path; LIST_INSERT_HEAD(KSEM_HASH(fnv), map, km_link); } static int ksem_remove(char *path, Fnv32_t fnv, struct ucred *ucred) { struct ksem_mapping *map; int error; LIST_FOREACH(map, KSEM_HASH(fnv), km_link) { if (map->km_fnv != fnv) continue; if (strcmp(map->km_path, path) == 0) { #ifdef MAC error = mac_posixsem_check_unlink(ucred, map->km_ksem); if (error) return (error); #endif error = ksem_access(map->km_ksem, ucred); if (error) return (error); map->km_ksem->ks_path = NULL; LIST_REMOVE(map, km_link); ksem_drop(map->km_ksem); free(map->km_path, M_KSEM); free(map, M_KSEM); return (0); } } return (ENOENT); } static int ksem_create_copyout_semid(struct thread *td, semid_t *semidp, int fd, int compat32) { semid_t semid; #ifdef COMPAT_FREEBSD32 int32_t semid32; #endif void *ptr; size_t ptrs; #ifdef COMPAT_FREEBSD32 if (compat32) { semid32 = fd; ptr = &semid32; ptrs = sizeof(semid32); } else { #endif semid = fd; ptr = &semid; ptrs = sizeof(semid); compat32 = 0; /* silence gcc */ #ifdef COMPAT_FREEBSD32 } #endif return (copyout(ptr, semidp, ptrs)); } /* Other helper routines. */ static int ksem_create(struct thread *td, const char *name, semid_t *semidp, mode_t mode, unsigned int value, int flags, int compat32) { struct pwddesc *pdp; struct ksem *ks; struct file *fp; char *path; const char *pr_path; size_t pr_pathlen; Fnv32_t fnv; int error, fd; AUDIT_ARG_FFLAGS(flags); AUDIT_ARG_MODE(mode); AUDIT_ARG_VALUE(value); if (value > SEM_VALUE_MAX) return (EINVAL); pdp = td->td_proc->p_pd; mode = (mode & ~pdp->pd_cmask) & ACCESSPERMS; error = falloc(td, &fp, &fd, O_CLOEXEC); if (error) { if (name == NULL) error = ENOSPC; return (error); } /* * Go ahead and copyout the file descriptor now. This is a bit * premature, but it is a lot easier to handle errors as opposed * to later when we've possibly created a new semaphore, etc. */ error = ksem_create_copyout_semid(td, semidp, fd, compat32); if (error) { fdclose(td, fp, fd); fdrop(fp, td); return (error); } if (name == NULL) { /* Create an anonymous semaphore. */ ks = ksem_alloc(td->td_ucred, mode, value); if (ks == NULL) error = ENOSPC; else ks->ks_flags |= KS_ANONYMOUS; } else { path = malloc(MAXPATHLEN, M_KSEM, M_WAITOK); pr_path = td->td_ucred->cr_prison->pr_path; /* Construct a full pathname for jailed callers. */ pr_pathlen = strcmp(pr_path, "/") == 0 ? 0 : strlcpy(path, pr_path, MAXPATHLEN); error = copyinstr(name, path + pr_pathlen, MAXPATHLEN - pr_pathlen, NULL); /* Require paths to start with a '/' character. */ if (error == 0 && path[pr_pathlen] != '/') error = EINVAL; if (error) { fdclose(td, fp, fd); fdrop(fp, td); free(path, M_KSEM); return (error); } AUDIT_ARG_UPATH1_CANON(path); fnv = fnv_32_str(path, FNV1_32_INIT); sx_xlock(&ksem_dict_lock); ks = ksem_lookup(path, fnv); if (ks == NULL) { /* Object does not exist, create it if requested. */ if (flags & O_CREAT) { ks = ksem_alloc(td->td_ucred, mode, value); if (ks == NULL) error = ENFILE; else { ksem_insert(path, fnv, ks); path = NULL; } } else error = ENOENT; } else { /* * Object already exists, obtain a new * reference if requested and permitted. */ if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) error = EEXIST; else { #ifdef MAC error = mac_posixsem_check_open(td->td_ucred, ks); if (error == 0) #endif error = ksem_access(ks, td->td_ucred); } if (error == 0) ksem_hold(ks); #ifdef INVARIANTS else ks = NULL; #endif } sx_xunlock(&ksem_dict_lock); if (path) free(path, M_KSEM); } if (error) { KASSERT(ks == NULL, ("ksem_create error with a ksem")); fdclose(td, fp, fd); fdrop(fp, td); return (error); } KASSERT(ks != NULL, ("ksem_create w/o a ksem")); finit(fp, FREAD | FWRITE, DTYPE_SEM, ks, &ksem_ops); fdrop(fp, td); return (0); } static int ksem_get(struct thread *td, semid_t id, cap_rights_t *rightsp, struct file **fpp) { struct ksem *ks; struct file *fp; int error; error = fget(td, id, rightsp, &fp); if (error) return (EINVAL); if (fp->f_type != DTYPE_SEM) { fdrop(fp, td); return (EINVAL); } ks = fp->f_data; if (ks->ks_flags & KS_DEAD) { fdrop(fp, td); return (EINVAL); } *fpp = fp; return (0); } /* System calls. */ #ifndef _SYS_SYSPROTO_H_ struct ksem_init_args { unsigned int value; semid_t *idp; }; #endif int sys_ksem_init(struct thread *td, struct ksem_init_args *uap) { return (ksem_create(td, NULL, uap->idp, S_IRWXU | S_IRWXG, uap->value, 0, 0)); } #ifndef _SYS_SYSPROTO_H_ struct ksem_open_args { char *name; int oflag; mode_t mode; unsigned int value; semid_t *idp; }; #endif int sys_ksem_open(struct thread *td, struct ksem_open_args *uap) { DP((">>> ksem_open start, pid=%d\n", (int)td->td_proc->p_pid)); if ((uap->oflag & ~(O_CREAT | O_EXCL)) != 0) return (EINVAL); return (ksem_create(td, uap->name, uap->idp, uap->mode, uap->value, uap->oflag, 0)); } #ifndef _SYS_SYSPROTO_H_ struct ksem_unlink_args { char *name; }; #endif int sys_ksem_unlink(struct thread *td, struct ksem_unlink_args *uap) { char *path; const char *pr_path; size_t pr_pathlen; Fnv32_t fnv; int error; path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); pr_path = td->td_ucred->cr_prison->pr_path; pr_pathlen = strcmp(pr_path, "/") == 0 ? 0 : strlcpy(path, pr_path, MAXPATHLEN); error = copyinstr(uap->name, path + pr_pathlen, MAXPATHLEN - pr_pathlen, NULL); if (error) { free(path, M_TEMP); return (error); } AUDIT_ARG_UPATH1_CANON(path); fnv = fnv_32_str(path, FNV1_32_INIT); sx_xlock(&ksem_dict_lock); error = ksem_remove(path, fnv, td->td_ucred); sx_xunlock(&ksem_dict_lock); free(path, M_TEMP); return (error); } #ifndef _SYS_SYSPROTO_H_ struct ksem_close_args { semid_t id; }; #endif int sys_ksem_close(struct thread *td, struct ksem_close_args *uap) { struct ksem *ks; struct file *fp; int error; /* No capability rights required to close a semaphore. */ AUDIT_ARG_FD(uap->id); error = ksem_get(td, uap->id, &cap_no_rights, &fp); if (error) return (error); ks = fp->f_data; if (ks->ks_flags & KS_ANONYMOUS) { fdrop(fp, td); return (EINVAL); } error = kern_close(td, uap->id); fdrop(fp, td); return (error); } #ifndef _SYS_SYSPROTO_H_ struct ksem_post_args { semid_t id; }; #endif int sys_ksem_post(struct thread *td, struct ksem_post_args *uap) { cap_rights_t rights; struct file *fp; struct ksem *ks; int error; AUDIT_ARG_FD(uap->id); error = ksem_get(td, uap->id, - cap_rights_init(&rights, CAP_SEM_POST), &fp); + cap_rights_init_one(&rights, CAP_SEM_POST), &fp); if (error) return (error); ks = fp->f_data; mtx_lock(&sem_lock); #ifdef MAC error = mac_posixsem_check_post(td->td_ucred, fp->f_cred, ks); if (error) goto err; #endif if (ks->ks_value == SEM_VALUE_MAX) { error = EOVERFLOW; goto err; } ++ks->ks_value; if (ks->ks_waiters > 0) cv_signal(&ks->ks_cv); error = 0; vfs_timestamp(&ks->ks_ctime); err: mtx_unlock(&sem_lock); fdrop(fp, td); return (error); } #ifndef _SYS_SYSPROTO_H_ struct ksem_wait_args { semid_t id; }; #endif int sys_ksem_wait(struct thread *td, struct ksem_wait_args *uap) { return (kern_sem_wait(td, uap->id, 0, NULL)); } #ifndef _SYS_SYSPROTO_H_ struct ksem_timedwait_args { semid_t id; const struct timespec *abstime; }; #endif int sys_ksem_timedwait(struct thread *td, struct ksem_timedwait_args *uap) { struct timespec abstime; struct timespec *ts; int error; /* * We allow a null timespec (wait forever). */ if (uap->abstime == NULL) ts = NULL; else { error = copyin(uap->abstime, &abstime, sizeof(abstime)); if (error != 0) return (error); if (abstime.tv_nsec >= 1000000000 || abstime.tv_nsec < 0) return (EINVAL); ts = &abstime; } return (kern_sem_wait(td, uap->id, 0, ts)); } #ifndef _SYS_SYSPROTO_H_ struct ksem_trywait_args { semid_t id; }; #endif int sys_ksem_trywait(struct thread *td, struct ksem_trywait_args *uap) { return (kern_sem_wait(td, uap->id, 1, NULL)); } static int kern_sem_wait(struct thread *td, semid_t id, int tryflag, struct timespec *abstime) { struct timespec ts1, ts2; struct timeval tv; cap_rights_t rights; struct file *fp; struct ksem *ks; int error; DP((">>> kern_sem_wait entered! pid=%d\n", (int)td->td_proc->p_pid)); AUDIT_ARG_FD(id); - error = ksem_get(td, id, cap_rights_init(&rights, CAP_SEM_WAIT), &fp); + error = ksem_get(td, id, cap_rights_init_one(&rights, CAP_SEM_WAIT), + &fp); if (error) return (error); ks = fp->f_data; mtx_lock(&sem_lock); DP((">>> kern_sem_wait critical section entered! pid=%d\n", (int)td->td_proc->p_pid)); #ifdef MAC error = mac_posixsem_check_wait(td->td_ucred, fp->f_cred, ks); if (error) { DP(("kern_sem_wait mac failed\n")); goto err; } #endif DP(("kern_sem_wait value = %d, tryflag %d\n", ks->ks_value, tryflag)); vfs_timestamp(&ks->ks_atime); while (ks->ks_value == 0) { ks->ks_waiters++; if (tryflag != 0) error = EAGAIN; else if (abstime == NULL) error = cv_wait_sig(&ks->ks_cv, &sem_lock); else { for (;;) { ts1 = *abstime; getnanotime(&ts2); timespecsub(&ts1, &ts2, &ts1); TIMESPEC_TO_TIMEVAL(&tv, &ts1); if (tv.tv_sec < 0) { error = ETIMEDOUT; break; } error = cv_timedwait_sig(&ks->ks_cv, &sem_lock, tvtohz(&tv)); if (error != EWOULDBLOCK) break; } } ks->ks_waiters--; if (error) goto err; } ks->ks_value--; DP(("kern_sem_wait value post-decrement = %d\n", ks->ks_value)); error = 0; err: mtx_unlock(&sem_lock); fdrop(fp, td); DP(("<<< kern_sem_wait leaving, pid=%d, error = %d\n", (int)td->td_proc->p_pid, error)); return (error); } #ifndef _SYS_SYSPROTO_H_ struct ksem_getvalue_args { semid_t id; int *val; }; #endif int sys_ksem_getvalue(struct thread *td, struct ksem_getvalue_args *uap) { cap_rights_t rights; struct file *fp; struct ksem *ks; int error, val; AUDIT_ARG_FD(uap->id); error = ksem_get(td, uap->id, - cap_rights_init(&rights, CAP_SEM_GETVALUE), &fp); + cap_rights_init_one(&rights, CAP_SEM_GETVALUE), &fp); if (error) return (error); ks = fp->f_data; mtx_lock(&sem_lock); #ifdef MAC error = mac_posixsem_check_getvalue(td->td_ucred, fp->f_cred, ks); if (error) { mtx_unlock(&sem_lock); fdrop(fp, td); return (error); } #endif val = ks->ks_value; vfs_timestamp(&ks->ks_atime); mtx_unlock(&sem_lock); fdrop(fp, td); error = copyout(&val, uap->val, sizeof(val)); return (error); } #ifndef _SYS_SYSPROTO_H_ struct ksem_destroy_args { semid_t id; }; #endif int sys_ksem_destroy(struct thread *td, struct ksem_destroy_args *uap) { struct file *fp; struct ksem *ks; int error; /* No capability rights required to close a semaphore. */ AUDIT_ARG_FD(uap->id); error = ksem_get(td, uap->id, &cap_no_rights, &fp); if (error) return (error); ks = fp->f_data; if (!(ks->ks_flags & KS_ANONYMOUS)) { fdrop(fp, td); return (EINVAL); } mtx_lock(&sem_lock); if (ks->ks_waiters != 0) { mtx_unlock(&sem_lock); error = EBUSY; goto err; } ks->ks_flags |= KS_DEAD; mtx_unlock(&sem_lock); error = kern_close(td, uap->id); err: fdrop(fp, td); return (error); } static struct syscall_helper_data ksem_syscalls[] = { SYSCALL_INIT_HELPER(ksem_init), SYSCALL_INIT_HELPER(ksem_open), SYSCALL_INIT_HELPER(ksem_unlink), SYSCALL_INIT_HELPER(ksem_close), SYSCALL_INIT_HELPER(ksem_post), SYSCALL_INIT_HELPER(ksem_wait), SYSCALL_INIT_HELPER(ksem_timedwait), SYSCALL_INIT_HELPER(ksem_trywait), SYSCALL_INIT_HELPER(ksem_getvalue), SYSCALL_INIT_HELPER(ksem_destroy), SYSCALL_INIT_LAST }; #ifdef COMPAT_FREEBSD32 #include #include #include #include #include int freebsd32_ksem_init(struct thread *td, struct freebsd32_ksem_init_args *uap) { return (ksem_create(td, NULL, uap->idp, S_IRWXU | S_IRWXG, uap->value, 0, 1)); } int freebsd32_ksem_open(struct thread *td, struct freebsd32_ksem_open_args *uap) { if ((uap->oflag & ~(O_CREAT | O_EXCL)) != 0) return (EINVAL); return (ksem_create(td, uap->name, uap->idp, uap->mode, uap->value, uap->oflag, 1)); } int freebsd32_ksem_timedwait(struct thread *td, struct freebsd32_ksem_timedwait_args *uap) { struct timespec32 abstime32; struct timespec *ts, abstime; int error; /* * We allow a null timespec (wait forever). */ if (uap->abstime == NULL) ts = NULL; else { error = copyin(uap->abstime, &abstime32, sizeof(abstime32)); if (error != 0) return (error); CP(abstime32, abstime, tv_sec); CP(abstime32, abstime, tv_nsec); if (abstime.tv_nsec >= 1000000000 || abstime.tv_nsec < 0) return (EINVAL); ts = &abstime; } return (kern_sem_wait(td, uap->id, 0, ts)); } static struct syscall_helper_data ksem32_syscalls[] = { SYSCALL32_INIT_HELPER(freebsd32_ksem_init), SYSCALL32_INIT_HELPER(freebsd32_ksem_open), SYSCALL32_INIT_HELPER_COMPAT(ksem_unlink), SYSCALL32_INIT_HELPER_COMPAT(ksem_close), SYSCALL32_INIT_HELPER_COMPAT(ksem_post), SYSCALL32_INIT_HELPER_COMPAT(ksem_wait), SYSCALL32_INIT_HELPER(freebsd32_ksem_timedwait), SYSCALL32_INIT_HELPER_COMPAT(ksem_trywait), SYSCALL32_INIT_HELPER_COMPAT(ksem_getvalue), SYSCALL32_INIT_HELPER_COMPAT(ksem_destroy), SYSCALL_INIT_LAST }; #endif static int ksem_module_init(void) { int error; mtx_init(&sem_lock, "sem", NULL, MTX_DEF); mtx_init(&ksem_count_lock, "ksem count", NULL, MTX_DEF); sx_init(&ksem_dict_lock, "ksem dictionary"); ksem_dictionary = hashinit(1024, M_KSEM, &ksem_hash); p31b_setcfg(CTL_P1003_1B_SEMAPHORES, 200112L); p31b_setcfg(CTL_P1003_1B_SEM_NSEMS_MAX, SEM_MAX); p31b_setcfg(CTL_P1003_1B_SEM_VALUE_MAX, SEM_VALUE_MAX); error = syscall_helper_register(ksem_syscalls, SY_THR_STATIC_KLD); if (error) return (error); #ifdef COMPAT_FREEBSD32 error = syscall32_helper_register(ksem32_syscalls, SY_THR_STATIC_KLD); if (error) return (error); #endif return (0); } static void ksem_module_destroy(void) { #ifdef COMPAT_FREEBSD32 syscall32_helper_unregister(ksem32_syscalls); #endif syscall_helper_unregister(ksem_syscalls); p31b_setcfg(CTL_P1003_1B_SEMAPHORES, 0); hashdestroy(ksem_dictionary, M_KSEM, ksem_hash); sx_destroy(&ksem_dict_lock); mtx_destroy(&ksem_count_lock); mtx_destroy(&sem_lock); p31b_unsetcfg(CTL_P1003_1B_SEM_VALUE_MAX); p31b_unsetcfg(CTL_P1003_1B_SEM_NSEMS_MAX); } static int sem_modload(struct module *module, int cmd, void *arg) { int error = 0; switch (cmd) { case MOD_LOAD: error = ksem_module_init(); if (error) ksem_module_destroy(); break; case MOD_UNLOAD: mtx_lock(&ksem_count_lock); if (nsems != 0) { error = EOPNOTSUPP; mtx_unlock(&ksem_count_lock); break; } ksem_dead = 1; mtx_unlock(&ksem_count_lock); ksem_module_destroy(); break; case MOD_SHUTDOWN: break; default: error = EINVAL; break; } return (error); } static moduledata_t sem_mod = { "sem", &sem_modload, NULL }; DECLARE_MODULE(sem, sem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST); MODULE_VERSION(sem, 1); diff --git a/sys/kern/uipc_usrreq.c b/sys/kern/uipc_usrreq.c index afea72fd2700..ad369fc7b23e 100644 --- a/sys/kern/uipc_usrreq.c +++ b/sys/kern/uipc_usrreq.c @@ -1,2992 +1,2994 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1989, 1991, 1993 * The Regents of the University of California. All Rights Reserved. * Copyright (c) 2004-2009 Robert N. M. Watson All Rights Reserved. * Copyright (c) 2018 Matthew Macy * * 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. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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. * * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 */ /* * UNIX Domain (Local) Sockets * * This is an implementation of UNIX (local) domain sockets. Each socket has * an associated struct unpcb (UNIX protocol control block). Stream sockets * may be connected to 0 or 1 other socket. Datagram sockets may be * connected to 0, 1, or many other sockets. Sockets may be created and * connected in pairs (socketpair(2)), or bound/connected to using the file * system name space. For most purposes, only the receive socket buffer is * used, as sending on one socket delivers directly to the receive socket * buffer of a second socket. * * The implementation is substantially complicated by the fact that * "ancillary data", such as file descriptors or credentials, may be passed * across UNIX domain sockets. The potential for passing UNIX domain sockets * over other UNIX domain sockets requires the implementation of a simple * garbage collector to find and tear down cycles of disconnected sockets. * * TODO: * RDM * rethink name space problems * need a proper out-of-band */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.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 #include #include #include #include #include #include #include #ifdef DDB #include #endif #include #include MALLOC_DECLARE(M_FILECAPS); /* * See unpcb.h for the locking key. */ static uma_zone_t unp_zone; static unp_gen_t unp_gencnt; /* (l) */ static u_int unp_count; /* (l) Count of local sockets. */ static ino_t unp_ino; /* Prototype for fake inode numbers. */ static int unp_rights; /* (g) File descriptors in flight. */ static struct unp_head unp_shead; /* (l) List of stream sockets. */ static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ struct unp_defer { SLIST_ENTRY(unp_defer) ud_link; struct file *ud_fp; }; static SLIST_HEAD(, unp_defer) unp_defers; static int unp_defers_count; static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; /* * Garbage collection of cyclic file descriptor/socket references occurs * asynchronously in a taskqueue context in order to avoid recursion and * reentrance in the UNIX domain socket, file descriptor, and socket layer * code. See unp_gc() for a full description. */ static struct timeout_task unp_gc_task; /* * The close of unix domain sockets attached as SCM_RIGHTS is * postponed to the taskqueue, to avoid arbitrary recursion depth. * The attached sockets might have another sockets attached. */ static struct task unp_defer_task; /* * Both send and receive buffers are allocated PIPSIZ bytes of buffering for * stream sockets, although the total for sender and receiver is actually * only PIPSIZ. * * Datagram sockets really use the sendspace as the maximum datagram size, * and don't really want to reserve the sendspace. Their recvspace should be * large enough for at least one max-size datagram plus address. */ #ifndef PIPSIZ #define PIPSIZ 8192 #endif static u_long unpst_sendspace = PIPSIZ; static u_long unpst_recvspace = PIPSIZ; static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ static u_long unpdg_recvspace = 4*1024; static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ static u_long unpsp_recvspace = PIPSIZ; static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Local domain"); static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "SOCK_STREAM"); static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "SOCK_DGRAM"); static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "SOCK_SEQPACKET"); SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, &unpst_sendspace, 0, "Default stream send space."); SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, &unpst_recvspace, 0, "Default stream receive space."); SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, &unpdg_sendspace, 0, "Default datagram send space."); SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, &unpdg_recvspace, 0, "Default datagram receive space."); SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, &unpsp_sendspace, 0, "Default seqpacket send space."); SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, &unpsp_recvspace, 0, "Default seqpacket receive space."); SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "File descriptors in flight."); SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, &unp_defers_count, 0, "File descriptors deferred to taskqueue for close."); /* * Locking and synchronization: * * Several types of locks exist in the local domain socket implementation: * - a global linkage lock * - a global connection list lock * - the mtxpool lock * - per-unpcb mutexes * * The linkage lock protects the global socket lists, the generation number * counter and garbage collector state. * * The connection list lock protects the list of referring sockets in a datagram * socket PCB. This lock is also overloaded to protect a global list of * sockets whose buffers contain socket references in the form of SCM_RIGHTS * messages. To avoid recursion, such references are released by a dedicated * thread. * * The mtxpool lock protects the vnode from being modified while referenced. * Lock ordering rules require that it be acquired before any PCB locks. * * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the * unpcb. This includes the unp_conn field, which either links two connected * PCBs together (for connected socket types) or points at the destination * socket (for connectionless socket types). The operations of creating or * destroying a connection therefore involve locking multiple PCBs. To avoid * lock order reversals, in some cases this involves dropping a PCB lock and * using a reference counter to maintain liveness. * * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, * allocated in pru_attach() and freed in pru_detach(). The validity of that * pointer is an invariant, so no lock is required to dereference the so_pcb * pointer if a valid socket reference is held by the caller. In practice, * this is always true during operations performed on a socket. Each unpcb * has a back-pointer to its socket, unp_socket, which will be stable under * the same circumstances. * * This pointer may only be safely dereferenced as long as a valid reference * to the unpcb is held. Typically, this reference will be from the socket, * or from another unpcb when the referring unpcb's lock is held (in order * that the reference not be invalidated during use). For example, to follow * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee * that detach is not run clearing unp_socket. * * Blocking with UNIX domain sockets is a tricky issue: unlike most network * protocols, bind() is a non-atomic operation, and connect() requires * potential sleeping in the protocol, due to potentially waiting on local or * distributed file systems. We try to separate "lookup" operations, which * may sleep, and the IPC operations themselves, which typically can occur * with relative atomicity as locks can be held over the entire operation. * * Another tricky issue is simultaneous multi-threaded or multi-process * access to a single UNIX domain socket. These are handled by the flags * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or * binding, both of which involve dropping UNIX domain socket locks in order * to perform namei() and other file system operations. */ static struct rwlock unp_link_rwlock; static struct mtx unp_defers_lock; #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ "unp_link_rwlock") #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ RA_LOCKED) #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ RA_UNLOCKED) #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ RA_WLOCKED) #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock) #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ "unp_defer", NULL, MTX_DEF) #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK(); #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK(); #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ "unp", "unp", \ MTX_DUPOK|MTX_DEF) #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) #define UNP_PCB_LOCKPTR(unp) (&(unp)->unp_mtx) #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx) #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx) #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED) static int uipc_connect2(struct socket *, struct socket *); static int uipc_ctloutput(struct socket *, struct sockopt *); static int unp_connect(struct socket *, struct sockaddr *, struct thread *); static int unp_connectat(int, struct socket *, struct sockaddr *, struct thread *); static int unp_connect2(struct socket *so, struct socket *so2, int); static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); static void unp_dispose(struct socket *so); static void unp_dispose_mbuf(struct mbuf *); static void unp_shutdown(struct unpcb *); static void unp_drop(struct unpcb *); static void unp_gc(__unused void *, int); static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int)); static void unp_discard(struct file *); static void unp_freerights(struct filedescent **, int); static void unp_init(void); static int unp_internalize(struct mbuf **, struct thread *); static void unp_internalize_fp(struct file *); static int unp_externalize(struct mbuf *, struct mbuf **, int); static int unp_externalize_fp(struct file *); static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *, int); static void unp_process_defers(void * __unused, int); static void unp_pcb_hold(struct unpcb *unp) { u_int old __unused; old = refcount_acquire(&unp->unp_refcount); KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp)); } static __result_use_check bool unp_pcb_rele(struct unpcb *unp) { bool ret; UNP_PCB_LOCK_ASSERT(unp); if ((ret = refcount_release(&unp->unp_refcount))) { UNP_PCB_UNLOCK(unp); UNP_PCB_LOCK_DESTROY(unp); uma_zfree(unp_zone, unp); } return (ret); } static void unp_pcb_rele_notlast(struct unpcb *unp) { bool ret __unused; ret = refcount_release(&unp->unp_refcount); KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp)); } static void unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2) { UNP_PCB_UNLOCK_ASSERT(unp); UNP_PCB_UNLOCK_ASSERT(unp2); if (unp == unp2) { UNP_PCB_LOCK(unp); } else if ((uintptr_t)unp2 > (uintptr_t)unp) { UNP_PCB_LOCK(unp); UNP_PCB_LOCK(unp2); } else { UNP_PCB_LOCK(unp2); UNP_PCB_LOCK(unp); } } static void unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2) { UNP_PCB_UNLOCK(unp); if (unp != unp2) UNP_PCB_UNLOCK(unp2); } /* * Try to lock the connected peer of an already locked socket. In some cases * this requires that we unlock the current socket. The pairbusy counter is * used to block concurrent connection attempts while the lock is dropped. The * caller must be careful to revalidate PCB state. */ static struct unpcb * unp_pcb_lock_peer(struct unpcb *unp) { struct unpcb *unp2; UNP_PCB_LOCK_ASSERT(unp); unp2 = unp->unp_conn; if (unp2 == NULL) return (NULL); if (__predict_false(unp == unp2)) return (unp); UNP_PCB_UNLOCK_ASSERT(unp2); if (__predict_true(UNP_PCB_TRYLOCK(unp2))) return (unp2); if ((uintptr_t)unp2 > (uintptr_t)unp) { UNP_PCB_LOCK(unp2); return (unp2); } unp->unp_pairbusy++; unp_pcb_hold(unp2); UNP_PCB_UNLOCK(unp); UNP_PCB_LOCK(unp2); UNP_PCB_LOCK(unp); KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL, ("%s: socket %p was reconnected", __func__, unp)); if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) { unp->unp_flags &= ~UNP_WAITING; wakeup(unp); } if (unp_pcb_rele(unp2)) { /* unp2 is unlocked. */ return (NULL); } if (unp->unp_conn == NULL) { UNP_PCB_UNLOCK(unp2); return (NULL); } return (unp2); } /* * Definitions of protocols supported in the LOCAL domain. */ static struct domain localdomain; static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream; static struct pr_usrreqs uipc_usrreqs_seqpacket; static struct protosw localsw[] = { { .pr_type = SOCK_STREAM, .pr_domain = &localdomain, .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS, .pr_ctloutput = &uipc_ctloutput, .pr_usrreqs = &uipc_usrreqs_stream }, { .pr_type = SOCK_DGRAM, .pr_domain = &localdomain, .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS, .pr_ctloutput = &uipc_ctloutput, .pr_usrreqs = &uipc_usrreqs_dgram }, { .pr_type = SOCK_SEQPACKET, .pr_domain = &localdomain, /* * XXXRW: For now, PR_ADDR because soreceive will bump into them * due to our use of sbappendaddr. A new sbappend variants is needed * that supports both atomic record writes and control data. */ .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD| PR_RIGHTS, .pr_ctloutput = &uipc_ctloutput, .pr_usrreqs = &uipc_usrreqs_seqpacket, }, }; static struct domain localdomain = { .dom_family = AF_LOCAL, .dom_name = "local", .dom_init = unp_init, .dom_externalize = unp_externalize, .dom_dispose = unp_dispose, .dom_protosw = localsw, .dom_protoswNPROTOSW = &localsw[nitems(localsw)] }; DOMAIN_SET(local); static void uipc_abort(struct socket *so) { struct unpcb *unp, *unp2; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); UNP_PCB_UNLOCK_ASSERT(unp); UNP_PCB_LOCK(unp); unp2 = unp->unp_conn; if (unp2 != NULL) { unp_pcb_hold(unp2); UNP_PCB_UNLOCK(unp); unp_drop(unp2); } else UNP_PCB_UNLOCK(unp); } static int uipc_accept(struct socket *so, struct sockaddr **nam) { struct unpcb *unp, *unp2; const struct sockaddr *sa; /* * Pass back name of connected socket, if it was bound and we are * still connected (our peer may have closed already!). */ unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); UNP_PCB_LOCK(unp); unp2 = unp_pcb_lock_peer(unp); if (unp2 != NULL && unp2->unp_addr != NULL) sa = (struct sockaddr *)unp2->unp_addr; else sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); if (unp2 != NULL) unp_pcb_unlock_pair(unp, unp2); else UNP_PCB_UNLOCK(unp); return (0); } static int uipc_attach(struct socket *so, int proto, struct thread *td) { u_long sendspace, recvspace; struct unpcb *unp; int error; bool locked; KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { switch (so->so_type) { case SOCK_STREAM: sendspace = unpst_sendspace; recvspace = unpst_recvspace; break; case SOCK_DGRAM: sendspace = unpdg_sendspace; recvspace = unpdg_recvspace; break; case SOCK_SEQPACKET: sendspace = unpsp_sendspace; recvspace = unpsp_recvspace; break; default: panic("uipc_attach"); } error = soreserve(so, sendspace, recvspace); if (error) return (error); } unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); if (unp == NULL) return (ENOBUFS); LIST_INIT(&unp->unp_refs); UNP_PCB_LOCK_INIT(unp); unp->unp_socket = so; so->so_pcb = unp; refcount_init(&unp->unp_refcount, 1); if ((locked = UNP_LINK_WOWNED()) == false) UNP_LINK_WLOCK(); unp->unp_gencnt = ++unp_gencnt; unp->unp_ino = ++unp_ino; unp_count++; switch (so->so_type) { case SOCK_STREAM: LIST_INSERT_HEAD(&unp_shead, unp, unp_link); break; case SOCK_DGRAM: LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); break; case SOCK_SEQPACKET: LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); break; default: panic("uipc_attach"); } if (locked == false) UNP_LINK_WUNLOCK(); return (0); } static int uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { struct sockaddr_un *soun = (struct sockaddr_un *)nam; struct vattr vattr; int error, namelen; struct nameidata nd; struct unpcb *unp; struct vnode *vp; struct mount *mp; cap_rights_t rights; char *buf; if (nam->sa_family != AF_UNIX) return (EAFNOSUPPORT); unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); if (soun->sun_len > sizeof(struct sockaddr_un)) return (EINVAL); namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); if (namelen <= 0) return (EINVAL); /* * We don't allow simultaneous bind() calls on a single UNIX domain * socket, so flag in-progress operations, and return an error if an * operation is already in progress. * * Historically, we have not allowed a socket to be rebound, so this * also returns an error. Not allowing re-binding simplifies the * implementation and avoids a great many possible failure modes. */ UNP_PCB_LOCK(unp); if (unp->unp_vnode != NULL) { UNP_PCB_UNLOCK(unp); return (EINVAL); } if (unp->unp_flags & UNP_BINDING) { UNP_PCB_UNLOCK(unp); return (EALREADY); } unp->unp_flags |= UNP_BINDING; UNP_PCB_UNLOCK(unp); buf = malloc(namelen + 1, M_TEMP, M_WAITOK); bcopy(soun->sun_path, buf, namelen); buf[namelen] = 0; restart: NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE, - UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td); + UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT), + td); /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ error = namei(&nd); if (error) goto error; vp = nd.ni_vp; if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_dvp == vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); if (vp != NULL) { vrele(vp); error = EADDRINUSE; goto error; } error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); if (error) goto error; goto restart; } VATTR_NULL(&vattr); vattr.va_type = VSOCK; vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_pd->pd_cmask); #ifdef MAC error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); #endif if (error == 0) error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if (error) { vn_finished_write(mp); if (error == ERELOOKUP) goto restart; goto error; } vp = nd.ni_vp; ASSERT_VOP_ELOCKED(vp, "uipc_bind"); soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); UNP_PCB_LOCK(unp); VOP_UNP_BIND(vp, unp); unp->unp_vnode = vp; unp->unp_addr = soun; unp->unp_flags &= ~UNP_BINDING; UNP_PCB_UNLOCK(unp); VOP_UNLOCK(vp); vn_finished_write(mp); free(buf, M_TEMP); return (0); error: UNP_PCB_LOCK(unp); unp->unp_flags &= ~UNP_BINDING; UNP_PCB_UNLOCK(unp); free(buf, M_TEMP); return (error); } static int uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) { return (uipc_bindat(AT_FDCWD, so, nam, td)); } static int uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) { int error; KASSERT(td == curthread, ("uipc_connect: td != curthread")); error = unp_connect(so, nam, td); return (error); } static int uipc_connectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { int error; KASSERT(td == curthread, ("uipc_connectat: td != curthread")); error = unp_connectat(fd, so, nam, td); return (error); } static void uipc_close(struct socket *so) { struct unpcb *unp, *unp2; struct vnode *vp = NULL; struct mtx *vplock; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_close: unp == NULL")); vplock = NULL; if ((vp = unp->unp_vnode) != NULL) { vplock = mtx_pool_find(mtxpool_sleep, vp); mtx_lock(vplock); } UNP_PCB_LOCK(unp); if (vp && unp->unp_vnode == NULL) { mtx_unlock(vplock); vp = NULL; } if (vp != NULL) { VOP_UNP_DETACH(vp); unp->unp_vnode = NULL; } if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) unp_disconnect(unp, unp2); else UNP_PCB_UNLOCK(unp); if (vp) { mtx_unlock(vplock); vrele(vp); } } static int uipc_connect2(struct socket *so1, struct socket *so2) { struct unpcb *unp, *unp2; int error; unp = so1->so_pcb; KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); unp2 = so2->so_pcb; KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); unp_pcb_lock_pair(unp, unp2); error = unp_connect2(so1, so2, PRU_CONNECT2); unp_pcb_unlock_pair(unp, unp2); return (error); } static void uipc_detach(struct socket *so) { struct unpcb *unp, *unp2; struct mtx *vplock; struct vnode *vp; int local_unp_rights; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); vp = NULL; vplock = NULL; SOCK_LOCK(so); if (!SOLISTENING(so)) { /* * Once the socket is removed from the global lists, * uipc_ready() will not be able to locate its socket buffer, so * clear the buffer now. At this point internalized rights have * already been disposed of. */ sbrelease(&so->so_rcv, so); } SOCK_UNLOCK(so); UNP_LINK_WLOCK(); LIST_REMOVE(unp, unp_link); if (unp->unp_gcflag & UNPGC_DEAD) LIST_REMOVE(unp, unp_dead); unp->unp_gencnt = ++unp_gencnt; --unp_count; UNP_LINK_WUNLOCK(); UNP_PCB_UNLOCK_ASSERT(unp); restart: if ((vp = unp->unp_vnode) != NULL) { vplock = mtx_pool_find(mtxpool_sleep, vp); mtx_lock(vplock); } UNP_PCB_LOCK(unp); if (unp->unp_vnode != vp && unp->unp_vnode != NULL) { if (vplock) mtx_unlock(vplock); UNP_PCB_UNLOCK(unp); goto restart; } if ((vp = unp->unp_vnode) != NULL) { VOP_UNP_DETACH(vp); unp->unp_vnode = NULL; } if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) unp_disconnect(unp, unp2); else UNP_PCB_UNLOCK(unp); UNP_REF_LIST_LOCK(); while (!LIST_EMPTY(&unp->unp_refs)) { struct unpcb *ref = LIST_FIRST(&unp->unp_refs); unp_pcb_hold(ref); UNP_REF_LIST_UNLOCK(); MPASS(ref != unp); UNP_PCB_UNLOCK_ASSERT(ref); unp_drop(ref); UNP_REF_LIST_LOCK(); } UNP_REF_LIST_UNLOCK(); UNP_PCB_LOCK(unp); local_unp_rights = unp_rights; unp->unp_socket->so_pcb = NULL; unp->unp_socket = NULL; free(unp->unp_addr, M_SONAME); unp->unp_addr = NULL; if (!unp_pcb_rele(unp)) UNP_PCB_UNLOCK(unp); if (vp) { mtx_unlock(vplock); vrele(vp); } if (local_unp_rights) taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); } static int uipc_disconnect(struct socket *so) { struct unpcb *unp, *unp2; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); UNP_PCB_LOCK(unp); if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) unp_disconnect(unp, unp2); else UNP_PCB_UNLOCK(unp); return (0); } static int uipc_listen(struct socket *so, int backlog, struct thread *td) { struct unpcb *unp; int error; if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET) return (EOPNOTSUPP); unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); UNP_PCB_LOCK(unp); if (unp->unp_vnode == NULL) { /* Already connected or not bound to an address. */ error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ; UNP_PCB_UNLOCK(unp); return (error); } SOCK_LOCK(so); error = solisten_proto_check(so); if (error == 0) { cru2xt(td, &unp->unp_peercred); solisten_proto(so, backlog); } SOCK_UNLOCK(so); UNP_PCB_UNLOCK(unp); return (error); } static int uipc_peeraddr(struct socket *so, struct sockaddr **nam) { struct unpcb *unp, *unp2; const struct sockaddr *sa; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); UNP_LINK_RLOCK(); /* * XXX: It seems that this test always fails even when connection is * established. So, this else clause is added as workaround to * return PF_LOCAL sockaddr. */ unp2 = unp->unp_conn; if (unp2 != NULL) { UNP_PCB_LOCK(unp2); if (unp2->unp_addr != NULL) sa = (struct sockaddr *) unp2->unp_addr; else sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); UNP_PCB_UNLOCK(unp2); } else { sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); } UNP_LINK_RUNLOCK(); return (0); } static int uipc_rcvd(struct socket *so, int flags) { struct unpcb *unp, *unp2; struct socket *so2; u_int mbcnt, sbcc; unp = sotounpcb(so); KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, ("%s: socktype %d", __func__, so->so_type)); /* * Adjust backpressure on sender and wakeup any waiting to write. * * The unp lock is acquired to maintain the validity of the unp_conn * pointer; no lock on unp2 is required as unp2->unp_socket will be * static as long as we don't permit unp2 to disconnect from unp, * which is prevented by the lock on unp. We cache values from * so_rcv to avoid holding the so_rcv lock over the entire * transaction on the remote so_snd. */ SOCKBUF_LOCK(&so->so_rcv); mbcnt = so->so_rcv.sb_mbcnt; sbcc = sbavail(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_rcv); /* * There is a benign race condition at this point. If we're planning to * clear SB_STOP, but uipc_send is called on the connected socket at * this instant, it might add data to the sockbuf and set SB_STOP. Then * we would erroneously clear SB_STOP below, even though the sockbuf is * full. The race is benign because the only ill effect is to allow the * sockbuf to exceed its size limit, and the size limits are not * strictly guaranteed anyway. */ UNP_PCB_LOCK(unp); unp2 = unp->unp_conn; if (unp2 == NULL) { UNP_PCB_UNLOCK(unp); return (0); } so2 = unp2->unp_socket; SOCKBUF_LOCK(&so2->so_snd); if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax) so2->so_snd.sb_flags &= ~SB_STOP; sowwakeup_locked(so2); UNP_PCB_UNLOCK(unp); return (0); } static int uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, struct mbuf *control, struct thread *td) { struct unpcb *unp, *unp2; struct socket *so2; u_int mbcnt, sbcc; int freed, error; unp = sotounpcb(so); KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM || so->so_type == SOCK_SEQPACKET, ("%s: socktype %d", __func__, so->so_type)); freed = error = 0; if (flags & PRUS_OOB) { error = EOPNOTSUPP; goto release; } if (control != NULL && (error = unp_internalize(&control, td))) goto release; unp2 = NULL; switch (so->so_type) { case SOCK_DGRAM: { const struct sockaddr *from; if (nam != NULL) { error = unp_connect(so, nam, td); if (error != 0) break; } UNP_PCB_LOCK(unp); /* * Because connect() and send() are non-atomic in a sendto() * with a target address, it's possible that the socket will * have disconnected before the send() can run. In that case * return the slightly counter-intuitive but otherwise * correct error that the socket is not connected. */ unp2 = unp_pcb_lock_peer(unp); if (unp2 == NULL) { UNP_PCB_UNLOCK(unp); error = ENOTCONN; break; } if (unp2->unp_flags & UNP_WANTCRED_MASK) control = unp_addsockcred(td, control, unp2->unp_flags); if (unp->unp_addr != NULL) from = (struct sockaddr *)unp->unp_addr; else from = &sun_noname; so2 = unp2->unp_socket; SOCKBUF_LOCK(&so2->so_rcv); if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { sorwakeup_locked(so2); m = NULL; control = NULL; } else { SOCKBUF_UNLOCK(&so2->so_rcv); error = ENOBUFS; } if (nam != NULL) unp_disconnect(unp, unp2); else unp_pcb_unlock_pair(unp, unp2); break; } case SOCK_SEQPACKET: case SOCK_STREAM: if ((so->so_state & SS_ISCONNECTED) == 0) { if (nam != NULL) { error = unp_connect(so, nam, td); if (error != 0) break; } else { error = ENOTCONN; break; } } UNP_PCB_LOCK(unp); if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) { UNP_PCB_UNLOCK(unp); error = ENOTCONN; break; } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) { unp_pcb_unlock_pair(unp, unp2); error = EPIPE; break; } UNP_PCB_UNLOCK(unp); if ((so2 = unp2->unp_socket) == NULL) { UNP_PCB_UNLOCK(unp2); error = ENOTCONN; break; } SOCKBUF_LOCK(&so2->so_rcv); if (unp2->unp_flags & UNP_WANTCRED_MASK) { /* * Credentials are passed only once on SOCK_STREAM and * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS). */ control = unp_addsockcred(td, control, unp2->unp_flags); unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT; } /* * Send to paired receive port and wake up readers. Don't * check for space available in the receive buffer if we're * attaching ancillary data; Unix domain sockets only check * for space in the sending sockbuf, and that check is * performed one level up the stack. At that level we cannot * precisely account for the amount of buffer space used * (e.g., because control messages are not yet internalized). */ switch (so->so_type) { case SOCK_STREAM: if (control != NULL) { sbappendcontrol_locked(&so2->so_rcv, m, control, flags); control = NULL; } else sbappend_locked(&so2->so_rcv, m, flags); break; case SOCK_SEQPACKET: if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, &sun_noname, m, control)) control = NULL; break; } mbcnt = so2->so_rcv.sb_mbcnt; sbcc = sbavail(&so2->so_rcv); if (sbcc) sorwakeup_locked(so2); else SOCKBUF_UNLOCK(&so2->so_rcv); /* * The PCB lock on unp2 protects the SB_STOP flag. Without it, * it would be possible for uipc_rcvd to be called at this * point, drain the receiving sockbuf, clear SB_STOP, and then * we would set SB_STOP below. That could lead to an empty * sockbuf having SB_STOP set */ SOCKBUF_LOCK(&so->so_snd); if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax) so->so_snd.sb_flags |= SB_STOP; SOCKBUF_UNLOCK(&so->so_snd); UNP_PCB_UNLOCK(unp2); m = NULL; break; } /* * PRUS_EOF is equivalent to pru_send followed by pru_shutdown. */ if (flags & PRUS_EOF) { UNP_PCB_LOCK(unp); socantsendmore(so); unp_shutdown(unp); UNP_PCB_UNLOCK(unp); } if (control != NULL && error != 0) unp_dispose_mbuf(control); release: if (control != NULL) m_freem(control); /* * In case of PRUS_NOTREADY, uipc_ready() is responsible * for freeing memory. */ if (m != NULL && (flags & PRUS_NOTREADY) == 0) m_freem(m); return (error); } static bool uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp) { struct mbuf *mb, *n; struct sockbuf *sb; SOCK_LOCK(so); if (SOLISTENING(so)) { SOCK_UNLOCK(so); return (false); } mb = NULL; sb = &so->so_rcv; SOCKBUF_LOCK(sb); if (sb->sb_fnrdy != NULL) { for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) { if (mb == m) { *errorp = sbready(sb, m, count); break; } mb = mb->m_next; if (mb == NULL) { mb = n; if (mb != NULL) n = mb->m_nextpkt; } } } SOCKBUF_UNLOCK(sb); SOCK_UNLOCK(so); return (mb != NULL); } static int uipc_ready(struct socket *so, struct mbuf *m, int count) { struct unpcb *unp, *unp2; struct socket *so2; int error, i; unp = sotounpcb(so); KASSERT(so->so_type == SOCK_STREAM, ("%s: unexpected socket type for %p", __func__, so)); UNP_PCB_LOCK(unp); if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { UNP_PCB_UNLOCK(unp); so2 = unp2->unp_socket; SOCKBUF_LOCK(&so2->so_rcv); if ((error = sbready(&so2->so_rcv, m, count)) == 0) sorwakeup_locked(so2); else SOCKBUF_UNLOCK(&so2->so_rcv); UNP_PCB_UNLOCK(unp2); return (error); } UNP_PCB_UNLOCK(unp); /* * The receiving socket has been disconnected, but may still be valid. * In this case, the now-ready mbufs are still present in its socket * buffer, so perform an exhaustive search before giving up and freeing * the mbufs. */ UNP_LINK_RLOCK(); LIST_FOREACH(unp, &unp_shead, unp_link) { if (uipc_ready_scan(unp->unp_socket, m, count, &error)) break; } UNP_LINK_RUNLOCK(); if (unp == NULL) { for (i = 0; i < count; i++) m = m_free(m); error = ECONNRESET; } return (error); } static int uipc_sense(struct socket *so, struct stat *sb) { struct unpcb *unp; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); sb->st_blksize = so->so_snd.sb_hiwat; sb->st_dev = NODEV; sb->st_ino = unp->unp_ino; return (0); } static int uipc_shutdown(struct socket *so) { struct unpcb *unp; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); UNP_PCB_LOCK(unp); socantsendmore(so); unp_shutdown(unp); UNP_PCB_UNLOCK(unp); return (0); } static int uipc_sockaddr(struct socket *so, struct sockaddr **nam) { struct unpcb *unp; const struct sockaddr *sa; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); UNP_PCB_LOCK(unp); if (unp->unp_addr != NULL) sa = (struct sockaddr *) unp->unp_addr; else sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); UNP_PCB_UNLOCK(unp); return (0); } static struct pr_usrreqs uipc_usrreqs_dgram = { .pru_abort = uipc_abort, .pru_accept = uipc_accept, .pru_attach = uipc_attach, .pru_bind = uipc_bind, .pru_bindat = uipc_bindat, .pru_connect = uipc_connect, .pru_connectat = uipc_connectat, .pru_connect2 = uipc_connect2, .pru_detach = uipc_detach, .pru_disconnect = uipc_disconnect, .pru_listen = uipc_listen, .pru_peeraddr = uipc_peeraddr, .pru_rcvd = uipc_rcvd, .pru_send = uipc_send, .pru_sense = uipc_sense, .pru_shutdown = uipc_shutdown, .pru_sockaddr = uipc_sockaddr, .pru_soreceive = soreceive_dgram, .pru_close = uipc_close, }; static struct pr_usrreqs uipc_usrreqs_seqpacket = { .pru_abort = uipc_abort, .pru_accept = uipc_accept, .pru_attach = uipc_attach, .pru_bind = uipc_bind, .pru_bindat = uipc_bindat, .pru_connect = uipc_connect, .pru_connectat = uipc_connectat, .pru_connect2 = uipc_connect2, .pru_detach = uipc_detach, .pru_disconnect = uipc_disconnect, .pru_listen = uipc_listen, .pru_peeraddr = uipc_peeraddr, .pru_rcvd = uipc_rcvd, .pru_send = uipc_send, .pru_sense = uipc_sense, .pru_shutdown = uipc_shutdown, .pru_sockaddr = uipc_sockaddr, .pru_soreceive = soreceive_generic, /* XXX: or...? */ .pru_close = uipc_close, }; static struct pr_usrreqs uipc_usrreqs_stream = { .pru_abort = uipc_abort, .pru_accept = uipc_accept, .pru_attach = uipc_attach, .pru_bind = uipc_bind, .pru_bindat = uipc_bindat, .pru_connect = uipc_connect, .pru_connectat = uipc_connectat, .pru_connect2 = uipc_connect2, .pru_detach = uipc_detach, .pru_disconnect = uipc_disconnect, .pru_listen = uipc_listen, .pru_peeraddr = uipc_peeraddr, .pru_rcvd = uipc_rcvd, .pru_send = uipc_send, .pru_ready = uipc_ready, .pru_sense = uipc_sense, .pru_shutdown = uipc_shutdown, .pru_sockaddr = uipc_sockaddr, .pru_soreceive = soreceive_generic, .pru_close = uipc_close, }; static int uipc_ctloutput(struct socket *so, struct sockopt *sopt) { struct unpcb *unp; struct xucred xu; int error, optval; if (sopt->sopt_level != SOL_LOCAL) return (EINVAL); unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); error = 0; switch (sopt->sopt_dir) { case SOPT_GET: switch (sopt->sopt_name) { case LOCAL_PEERCRED: UNP_PCB_LOCK(unp); if (unp->unp_flags & UNP_HAVEPC) xu = unp->unp_peercred; else { if (so->so_type == SOCK_STREAM) error = ENOTCONN; else error = EINVAL; } UNP_PCB_UNLOCK(unp); if (error == 0) error = sooptcopyout(sopt, &xu, sizeof(xu)); break; case LOCAL_CREDS: /* Unlocked read. */ optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0; error = sooptcopyout(sopt, &optval, sizeof(optval)); break; case LOCAL_CREDS_PERSISTENT: /* Unlocked read. */ optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0; error = sooptcopyout(sopt, &optval, sizeof(optval)); break; case LOCAL_CONNWAIT: /* Unlocked read. */ optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; error = sooptcopyout(sopt, &optval, sizeof(optval)); break; default: error = EOPNOTSUPP; break; } break; case SOPT_SET: switch (sopt->sopt_name) { case LOCAL_CREDS: case LOCAL_CREDS_PERSISTENT: case LOCAL_CONNWAIT: error = sooptcopyin(sopt, &optval, sizeof(optval), sizeof(optval)); if (error) break; #define OPTSET(bit, exclusive) do { \ UNP_PCB_LOCK(unp); \ if (optval) { \ if ((unp->unp_flags & (exclusive)) != 0) { \ UNP_PCB_UNLOCK(unp); \ error = EINVAL; \ break; \ } \ unp->unp_flags |= (bit); \ } else \ unp->unp_flags &= ~(bit); \ UNP_PCB_UNLOCK(unp); \ } while (0) switch (sopt->sopt_name) { case LOCAL_CREDS: OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS); break; case LOCAL_CREDS_PERSISTENT: OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT); break; case LOCAL_CONNWAIT: OPTSET(UNP_CONNWAIT, 0); break; default: break; } break; #undef OPTSET default: error = ENOPROTOOPT; break; } break; default: error = EOPNOTSUPP; break; } return (error); } static int unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) { return (unp_connectat(AT_FDCWD, so, nam, td)); } static int unp_connectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { struct mtx *vplock; struct sockaddr_un *soun; struct vnode *vp; struct socket *so2; struct unpcb *unp, *unp2, *unp3; struct nameidata nd; char buf[SOCK_MAXADDRLEN]; struct sockaddr *sa; cap_rights_t rights; int error, len; bool connreq; if (nam->sa_family != AF_UNIX) return (EAFNOSUPPORT); if (nam->sa_len > sizeof(struct sockaddr_un)) return (EINVAL); len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); if (len <= 0) return (EINVAL); soun = (struct sockaddr_un *)nam; bcopy(soun->sun_path, buf, len); buf[len] = 0; unp = sotounpcb(so); UNP_PCB_LOCK(unp); for (;;) { /* * Wait for connection state to stabilize. If a connection * already exists, give up. For datagram sockets, which permit * multiple consecutive connect(2) calls, upper layers are * responsible for disconnecting in advance of a subsequent * connect(2), but this is not synchronized with PCB connection * state. * * Also make sure that no threads are currently attempting to * lock the peer socket, to ensure that unp_conn cannot * transition between two valid sockets while locks are dropped. */ if (unp->unp_conn != NULL) { UNP_PCB_UNLOCK(unp); return (EISCONN); } if ((unp->unp_flags & UNP_CONNECTING) != 0) { UNP_PCB_UNLOCK(unp); return (EALREADY); } if (unp->unp_pairbusy > 0) { unp->unp_flags |= UNP_WAITING; mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0); continue; } break; } unp->unp_flags |= UNP_CONNECTING; UNP_PCB_UNLOCK(unp); connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0; if (connreq) sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); else sa = NULL; NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, - UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td); + UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT), + td); error = namei(&nd); if (error) vp = NULL; else vp = nd.ni_vp; ASSERT_VOP_LOCKED(vp, "unp_connect"); NDFREE(&nd, NDF_ONLY_PNBUF); if (error) goto bad; if (vp->v_type != VSOCK) { error = ENOTSOCK; goto bad; } #ifdef MAC error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); if (error) goto bad; #endif error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); if (error) goto bad; unp = sotounpcb(so); KASSERT(unp != NULL, ("unp_connect: unp == NULL")); vplock = mtx_pool_find(mtxpool_sleep, vp); mtx_lock(vplock); VOP_UNP_CONNECT(vp, &unp2); if (unp2 == NULL) { error = ECONNREFUSED; goto bad2; } so2 = unp2->unp_socket; if (so->so_type != so2->so_type) { error = EPROTOTYPE; goto bad2; } if (connreq) { if (so2->so_options & SO_ACCEPTCONN) { CURVNET_SET(so2->so_vnet); so2 = sonewconn(so2, 0); CURVNET_RESTORE(); } else so2 = NULL; if (so2 == NULL) { error = ECONNREFUSED; goto bad2; } unp3 = sotounpcb(so2); unp_pcb_lock_pair(unp2, unp3); if (unp2->unp_addr != NULL) { bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); unp3->unp_addr = (struct sockaddr_un *) sa; sa = NULL; } unp_copy_peercred(td, unp3, unp, unp2); UNP_PCB_UNLOCK(unp2); unp2 = unp3; /* * It is safe to block on the PCB lock here since unp2 is * nascent and cannot be connected to any other sockets. */ UNP_PCB_LOCK(unp); #ifdef MAC mac_socketpeer_set_from_socket(so, so2); mac_socketpeer_set_from_socket(so2, so); #endif } else { unp_pcb_lock_pair(unp, unp2); } KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 && sotounpcb(so2) == unp2, ("%s: unp2 %p so2 %p", __func__, unp2, so2)); error = unp_connect2(so, so2, PRU_CONNECT); unp_pcb_unlock_pair(unp, unp2); bad2: mtx_unlock(vplock); bad: if (vp != NULL) { vput(vp); } free(sa, M_SONAME); UNP_PCB_LOCK(unp); KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); unp->unp_flags &= ~UNP_CONNECTING; UNP_PCB_UNLOCK(unp); return (error); } /* * Set socket peer credentials at connection time. * * The client's PCB credentials are copied from its process structure. The * server's PCB credentials are copied from the socket on which it called * listen(2). uipc_listen cached that process's credentials at the time. */ void unp_copy_peercred(struct thread *td, struct unpcb *client_unp, struct unpcb *server_unp, struct unpcb *listen_unp) { cru2xt(td, &client_unp->unp_peercred); client_unp->unp_flags |= UNP_HAVEPC; memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred, sizeof(server_unp->unp_peercred)); server_unp->unp_flags |= UNP_HAVEPC; client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK); } static int unp_connect2(struct socket *so, struct socket *so2, int req) { struct unpcb *unp; struct unpcb *unp2; unp = sotounpcb(so); KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); unp2 = sotounpcb(so2); KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); UNP_PCB_LOCK_ASSERT(unp); UNP_PCB_LOCK_ASSERT(unp2); KASSERT(unp->unp_conn == NULL, ("%s: socket %p is already connected", __func__, unp)); if (so2->so_type != so->so_type) return (EPROTOTYPE); unp->unp_conn = unp2; unp_pcb_hold(unp2); unp_pcb_hold(unp); switch (so->so_type) { case SOCK_DGRAM: UNP_REF_LIST_LOCK(); LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); UNP_REF_LIST_UNLOCK(); soisconnected(so); break; case SOCK_STREAM: case SOCK_SEQPACKET: KASSERT(unp2->unp_conn == NULL, ("%s: socket %p is already connected", __func__, unp2)); unp2->unp_conn = unp; if (req == PRU_CONNECT && ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) soisconnecting(so); else soisconnected(so); soisconnected(so2); break; default: panic("unp_connect2"); } return (0); } static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2) { struct socket *so, *so2; #ifdef INVARIANTS struct unpcb *unptmp; #endif UNP_PCB_LOCK_ASSERT(unp); UNP_PCB_LOCK_ASSERT(unp2); KASSERT(unp->unp_conn == unp2, ("%s: unpcb %p is not connected to %p", __func__, unp, unp2)); unp->unp_conn = NULL; so = unp->unp_socket; so2 = unp2->unp_socket; switch (unp->unp_socket->so_type) { case SOCK_DGRAM: UNP_REF_LIST_LOCK(); #ifdef INVARIANTS LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) { if (unptmp == unp) break; } KASSERT(unptmp != NULL, ("%s: %p not found in reflist of %p", __func__, unp, unp2)); #endif LIST_REMOVE(unp, unp_reflink); UNP_REF_LIST_UNLOCK(); if (so) { SOCK_LOCK(so); so->so_state &= ~SS_ISCONNECTED; SOCK_UNLOCK(so); } break; case SOCK_STREAM: case SOCK_SEQPACKET: if (so) soisdisconnected(so); MPASS(unp2->unp_conn == unp); unp2->unp_conn = NULL; if (so2) soisdisconnected(so2); break; } if (unp == unp2) { unp_pcb_rele_notlast(unp); if (!unp_pcb_rele(unp)) UNP_PCB_UNLOCK(unp); } else { if (!unp_pcb_rele(unp)) UNP_PCB_UNLOCK(unp); if (!unp_pcb_rele(unp2)) UNP_PCB_UNLOCK(unp2); } } /* * unp_pcblist() walks the global list of struct unpcb's to generate a * pointer list, bumping the refcount on each unpcb. It then copies them out * sequentially, validating the generation number on each to see if it has * been detached. All of this is necessary because copyout() may sleep on * disk I/O. */ static int unp_pcblist(SYSCTL_HANDLER_ARGS) { struct unpcb *unp, **unp_list; unp_gen_t gencnt; struct xunpgen *xug; struct unp_head *head; struct xunpcb *xu; u_int i; int error, n; switch ((intptr_t)arg1) { case SOCK_STREAM: head = &unp_shead; break; case SOCK_DGRAM: head = &unp_dhead; break; case SOCK_SEQPACKET: head = &unp_sphead; break; default: panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); } /* * The process of preparing the PCB list is too time-consuming and * resource-intensive to repeat twice on every request. */ if (req->oldptr == NULL) { n = unp_count; req->oldidx = 2 * (sizeof *xug) + (n + n/8) * sizeof(struct xunpcb); return (0); } if (req->newptr != NULL) return (EPERM); /* * OK, now we're committed to doing something. */ xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO); UNP_LINK_RLOCK(); gencnt = unp_gencnt; n = unp_count; UNP_LINK_RUNLOCK(); xug->xug_len = sizeof *xug; xug->xug_count = n; xug->xug_gen = gencnt; xug->xug_sogen = so_gencnt; error = SYSCTL_OUT(req, xug, sizeof *xug); if (error) { free(xug, M_TEMP); return (error); } unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); UNP_LINK_RLOCK(); for (unp = LIST_FIRST(head), i = 0; unp && i < n; unp = LIST_NEXT(unp, unp_link)) { UNP_PCB_LOCK(unp); if (unp->unp_gencnt <= gencnt) { if (cr_cansee(req->td->td_ucred, unp->unp_socket->so_cred)) { UNP_PCB_UNLOCK(unp); continue; } unp_list[i++] = unp; unp_pcb_hold(unp); } UNP_PCB_UNLOCK(unp); } UNP_LINK_RUNLOCK(); n = i; /* In case we lost some during malloc. */ error = 0; xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); for (i = 0; i < n; i++) { unp = unp_list[i]; UNP_PCB_LOCK(unp); if (unp_pcb_rele(unp)) continue; if (unp->unp_gencnt <= gencnt) { xu->xu_len = sizeof *xu; xu->xu_unpp = (uintptr_t)unp; /* * XXX - need more locking here to protect against * connect/disconnect races for SMP. */ if (unp->unp_addr != NULL) bcopy(unp->unp_addr, &xu->xu_addr, unp->unp_addr->sun_len); else bzero(&xu->xu_addr, sizeof(xu->xu_addr)); if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) bcopy(unp->unp_conn->unp_addr, &xu->xu_caddr, unp->unp_conn->unp_addr->sun_len); else bzero(&xu->xu_caddr, sizeof(xu->xu_caddr)); xu->unp_vnode = (uintptr_t)unp->unp_vnode; xu->unp_conn = (uintptr_t)unp->unp_conn; xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs); xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink); xu->unp_gencnt = unp->unp_gencnt; sotoxsocket(unp->unp_socket, &xu->xu_socket); UNP_PCB_UNLOCK(unp); error = SYSCTL_OUT(req, xu, sizeof *xu); } else { UNP_PCB_UNLOCK(unp); } } free(xu, M_TEMP); if (!error) { /* * Give the user an updated idea of our state. If the * generation differs from what we told her before, she knows * that something happened while we were processing this * request, and it might be necessary to retry. */ xug->xug_gen = unp_gencnt; xug->xug_sogen = so_gencnt; xug->xug_count = unp_count; error = SYSCTL_OUT(req, xug, sizeof *xug); } free(unp_list, M_TEMP); free(xug, M_TEMP); return (error); } SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", "List of active local datagram sockets"); SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", "List of active local stream sockets"); SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", "List of active local seqpacket sockets"); static void unp_shutdown(struct unpcb *unp) { struct unpcb *unp2; struct socket *so; UNP_PCB_LOCK_ASSERT(unp); unp2 = unp->unp_conn; if ((unp->unp_socket->so_type == SOCK_STREAM || (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { so = unp2->unp_socket; if (so != NULL) socantrcvmore(so); } } static void unp_drop(struct unpcb *unp) { struct socket *so = unp->unp_socket; struct unpcb *unp2; /* * Regardless of whether the socket's peer dropped the connection * with this socket by aborting or disconnecting, POSIX requires * that ECONNRESET is returned. */ UNP_PCB_LOCK(unp); if (so) so->so_error = ECONNRESET; if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { /* Last reference dropped in unp_disconnect(). */ unp_pcb_rele_notlast(unp); unp_disconnect(unp, unp2); } else if (!unp_pcb_rele(unp)) { UNP_PCB_UNLOCK(unp); } } static void unp_freerights(struct filedescent **fdep, int fdcount) { struct file *fp; int i; KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); for (i = 0; i < fdcount; i++) { fp = fdep[i]->fde_file; filecaps_free(&fdep[i]->fde_caps); unp_discard(fp); } free(fdep[0], M_FILECAPS); } static int unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) { struct thread *td = curthread; /* XXX */ struct cmsghdr *cm = mtod(control, struct cmsghdr *); int i; int *fdp; struct filedesc *fdesc = td->td_proc->p_fd; struct filedescent **fdep; void *data; socklen_t clen = control->m_len, datalen; int error, newfds; u_int newlen; UNP_LINK_UNLOCK_ASSERT(); error = 0; if (controlp != NULL) /* controlp == NULL => free control messages */ *controlp = NULL; while (cm != NULL) { if (sizeof(*cm) > clen || cm->cmsg_len > clen) { error = EINVAL; break; } data = CMSG_DATA(cm); datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; if (cm->cmsg_level == SOL_SOCKET && cm->cmsg_type == SCM_RIGHTS) { newfds = datalen / sizeof(*fdep); if (newfds == 0) goto next; fdep = data; /* If we're not outputting the descriptors free them. */ if (error || controlp == NULL) { unp_freerights(fdep, newfds); goto next; } FILEDESC_XLOCK(fdesc); /* * Now change each pointer to an fd in the global * table to an integer that is the index to the local * fd table entry that we set up to point to the * global one we are transferring. */ newlen = newfds * sizeof(int); *controlp = sbcreatecontrol(NULL, newlen, SCM_RIGHTS, SOL_SOCKET); if (*controlp == NULL) { FILEDESC_XUNLOCK(fdesc); error = E2BIG; unp_freerights(fdep, newfds); goto next; } fdp = (int *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); if (fdallocn(td, 0, fdp, newfds) != 0) { FILEDESC_XUNLOCK(fdesc); error = EMSGSIZE; unp_freerights(fdep, newfds); m_freem(*controlp); *controlp = NULL; goto next; } for (i = 0; i < newfds; i++, fdp++) { _finstall(fdesc, fdep[i]->fde_file, *fdp, (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0, &fdep[i]->fde_caps); unp_externalize_fp(fdep[i]->fde_file); } /* * The new type indicates that the mbuf data refers to * kernel resources that may need to be released before * the mbuf is freed. */ m_chtype(*controlp, MT_EXTCONTROL); FILEDESC_XUNLOCK(fdesc); free(fdep[0], M_FILECAPS); } else { /* We can just copy anything else across. */ if (error || controlp == NULL) goto next; *controlp = sbcreatecontrol(NULL, datalen, cm->cmsg_type, cm->cmsg_level); if (*controlp == NULL) { error = ENOBUFS; goto next; } bcopy(data, CMSG_DATA(mtod(*controlp, struct cmsghdr *)), datalen); } controlp = &(*controlp)->m_next; next: if (CMSG_SPACE(datalen) < clen) { clen -= CMSG_SPACE(datalen); cm = (struct cmsghdr *) ((caddr_t)cm + CMSG_SPACE(datalen)); } else { clen = 0; cm = NULL; } } m_freem(control); return (error); } static void unp_zone_change(void *tag) { uma_zone_set_max(unp_zone, maxsockets); } #ifdef INVARIANTS static void unp_zdtor(void *mem, int size __unused, void *arg __unused) { struct unpcb *unp; unp = mem; KASSERT(LIST_EMPTY(&unp->unp_refs), ("%s: unpcb %p has lingering refs", __func__, unp)); KASSERT(unp->unp_socket == NULL, ("%s: unpcb %p has socket backpointer", __func__, unp)); KASSERT(unp->unp_vnode == NULL, ("%s: unpcb %p has vnode references", __func__, unp)); KASSERT(unp->unp_conn == NULL, ("%s: unpcb %p is still connected", __func__, unp)); KASSERT(unp->unp_addr == NULL, ("%s: unpcb %p has leaked addr", __func__, unp)); } #endif static void unp_init(void) { uma_dtor dtor; #ifdef VIMAGE if (!IS_DEFAULT_VNET(curvnet)) return; #endif #ifdef INVARIANTS dtor = unp_zdtor; #else dtor = NULL; #endif unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor, NULL, NULL, UMA_ALIGN_CACHE, 0); uma_zone_set_max(unp_zone, maxsockets); uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, NULL, EVENTHANDLER_PRI_ANY); LIST_INIT(&unp_dhead); LIST_INIT(&unp_shead); LIST_INIT(&unp_sphead); SLIST_INIT(&unp_defers); TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); UNP_LINK_LOCK_INIT(); UNP_DEFERRED_LOCK_INIT(); } static void unp_internalize_cleanup_rights(struct mbuf *control) { struct cmsghdr *cp; struct mbuf *m; void *data; socklen_t datalen; for (m = control; m != NULL; m = m->m_next) { cp = mtod(m, struct cmsghdr *); if (cp->cmsg_level != SOL_SOCKET || cp->cmsg_type != SCM_RIGHTS) continue; data = CMSG_DATA(cp); datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data; unp_freerights(data, datalen / sizeof(struct filedesc *)); } } static int unp_internalize(struct mbuf **controlp, struct thread *td) { struct mbuf *control, **initial_controlp; struct proc *p; struct filedesc *fdesc; struct bintime *bt; struct cmsghdr *cm; struct cmsgcred *cmcred; struct filedescent *fde, **fdep, *fdev; struct file *fp; struct timeval *tv; struct timespec *ts; void *data; socklen_t clen, datalen; int i, j, error, *fdp, oldfds; u_int newlen; UNP_LINK_UNLOCK_ASSERT(); p = td->td_proc; fdesc = p->p_fd; error = 0; control = *controlp; clen = control->m_len; *controlp = NULL; initial_controlp = controlp; for (cm = mtod(control, struct cmsghdr *); cm != NULL;) { if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) { error = EINVAL; goto out; } data = CMSG_DATA(cm); datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; switch (cm->cmsg_type) { /* * Fill in credential information. */ case SCM_CREDS: *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), SCM_CREDS, SOL_SOCKET); if (*controlp == NULL) { error = ENOBUFS; goto out; } cmcred = (struct cmsgcred *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); cmcred->cmcred_pid = p->p_pid; cmcred->cmcred_uid = td->td_ucred->cr_ruid; cmcred->cmcred_gid = td->td_ucred->cr_rgid; cmcred->cmcred_euid = td->td_ucred->cr_uid; cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); for (i = 0; i < cmcred->cmcred_ngroups; i++) cmcred->cmcred_groups[i] = td->td_ucred->cr_groups[i]; break; case SCM_RIGHTS: oldfds = datalen / sizeof (int); if (oldfds == 0) break; /* * Check that all the FDs passed in refer to legal * files. If not, reject the entire operation. */ fdp = data; FILEDESC_SLOCK(fdesc); for (i = 0; i < oldfds; i++, fdp++) { fp = fget_locked(fdesc, *fdp); if (fp == NULL) { FILEDESC_SUNLOCK(fdesc); error = EBADF; goto out; } if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { FILEDESC_SUNLOCK(fdesc); error = EOPNOTSUPP; goto out; } } /* * Now replace the integer FDs with pointers to the * file structure and capability rights. */ newlen = oldfds * sizeof(fdep[0]); *controlp = sbcreatecontrol(NULL, newlen, SCM_RIGHTS, SOL_SOCKET); if (*controlp == NULL) { FILEDESC_SUNLOCK(fdesc); error = E2BIG; goto out; } fdp = data; for (i = 0; i < oldfds; i++, fdp++) { if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) { fdp = data; for (j = 0; j < i; j++, fdp++) { fdrop(fdesc->fd_ofiles[*fdp]. fde_file, td); } FILEDESC_SUNLOCK(fdesc); error = EBADF; goto out; } } fdp = data; fdep = (struct filedescent **) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, M_WAITOK); for (i = 0; i < oldfds; i++, fdev++, fdp++) { fde = &fdesc->fd_ofiles[*fdp]; fdep[i] = fdev; fdep[i]->fde_file = fde->fde_file; filecaps_copy(&fde->fde_caps, &fdep[i]->fde_caps, true); unp_internalize_fp(fdep[i]->fde_file); } FILEDESC_SUNLOCK(fdesc); break; case SCM_TIMESTAMP: *controlp = sbcreatecontrol(NULL, sizeof(*tv), SCM_TIMESTAMP, SOL_SOCKET); if (*controlp == NULL) { error = ENOBUFS; goto out; } tv = (struct timeval *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); microtime(tv); break; case SCM_BINTIME: *controlp = sbcreatecontrol(NULL, sizeof(*bt), SCM_BINTIME, SOL_SOCKET); if (*controlp == NULL) { error = ENOBUFS; goto out; } bt = (struct bintime *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); bintime(bt); break; case SCM_REALTIME: *controlp = sbcreatecontrol(NULL, sizeof(*ts), SCM_REALTIME, SOL_SOCKET); if (*controlp == NULL) { error = ENOBUFS; goto out; } ts = (struct timespec *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); nanotime(ts); break; case SCM_MONOTONIC: *controlp = sbcreatecontrol(NULL, sizeof(*ts), SCM_MONOTONIC, SOL_SOCKET); if (*controlp == NULL) { error = ENOBUFS; goto out; } ts = (struct timespec *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); nanouptime(ts); break; default: error = EINVAL; goto out; } if (*controlp != NULL) controlp = &(*controlp)->m_next; if (CMSG_SPACE(datalen) < clen) { clen -= CMSG_SPACE(datalen); cm = (struct cmsghdr *) ((caddr_t)cm + CMSG_SPACE(datalen)); } else { clen = 0; cm = NULL; } } out: if (error != 0 && initial_controlp != NULL) unp_internalize_cleanup_rights(*initial_controlp); m_freem(control); return (error); } static struct mbuf * unp_addsockcred(struct thread *td, struct mbuf *control, int mode) { struct mbuf *m, *n, *n_prev; const struct cmsghdr *cm; int ngroups, i, cmsgtype; size_t ctrlsz; ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); if (mode & UNP_WANTCRED_ALWAYS) { ctrlsz = SOCKCRED2SIZE(ngroups); cmsgtype = SCM_CREDS2; } else { ctrlsz = SOCKCREDSIZE(ngroups); cmsgtype = SCM_CREDS; } m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET); if (m == NULL) return (control); if (mode & UNP_WANTCRED_ALWAYS) { struct sockcred2 *sc; sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); sc->sc_version = 0; sc->sc_pid = td->td_proc->p_pid; sc->sc_uid = td->td_ucred->cr_ruid; sc->sc_euid = td->td_ucred->cr_uid; sc->sc_gid = td->td_ucred->cr_rgid; sc->sc_egid = td->td_ucred->cr_gid; sc->sc_ngroups = ngroups; for (i = 0; i < sc->sc_ngroups; i++) sc->sc_groups[i] = td->td_ucred->cr_groups[i]; } else { struct sockcred *sc; sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); sc->sc_uid = td->td_ucred->cr_ruid; sc->sc_euid = td->td_ucred->cr_uid; sc->sc_gid = td->td_ucred->cr_rgid; sc->sc_egid = td->td_ucred->cr_gid; sc->sc_ngroups = ngroups; for (i = 0; i < sc->sc_ngroups; i++) sc->sc_groups[i] = td->td_ucred->cr_groups[i]; } /* * Unlink SCM_CREDS control messages (struct cmsgcred), since just * created SCM_CREDS control message (struct sockcred) has another * format. */ if (control != NULL && cmsgtype == SCM_CREDS) for (n = control, n_prev = NULL; n != NULL;) { cm = mtod(n, struct cmsghdr *); if (cm->cmsg_level == SOL_SOCKET && cm->cmsg_type == SCM_CREDS) { if (n_prev == NULL) control = n->m_next; else n_prev->m_next = n->m_next; n = m_free(n); } else { n_prev = n; n = n->m_next; } } /* Prepend it to the head. */ m->m_next = control; return (m); } static struct unpcb * fptounp(struct file *fp) { struct socket *so; if (fp->f_type != DTYPE_SOCKET) return (NULL); if ((so = fp->f_data) == NULL) return (NULL); if (so->so_proto->pr_domain != &localdomain) return (NULL); return sotounpcb(so); } static void unp_discard(struct file *fp) { struct unp_defer *dr; if (unp_externalize_fp(fp)) { dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); dr->ud_fp = fp; UNP_DEFERRED_LOCK(); SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); UNP_DEFERRED_UNLOCK(); atomic_add_int(&unp_defers_count, 1); taskqueue_enqueue(taskqueue_thread, &unp_defer_task); } else (void) closef(fp, (struct thread *)NULL); } static void unp_process_defers(void *arg __unused, int pending) { struct unp_defer *dr; SLIST_HEAD(, unp_defer) drl; int count; SLIST_INIT(&drl); for (;;) { UNP_DEFERRED_LOCK(); if (SLIST_FIRST(&unp_defers) == NULL) { UNP_DEFERRED_UNLOCK(); break; } SLIST_SWAP(&unp_defers, &drl, unp_defer); UNP_DEFERRED_UNLOCK(); count = 0; while ((dr = SLIST_FIRST(&drl)) != NULL) { SLIST_REMOVE_HEAD(&drl, ud_link); closef(dr->ud_fp, NULL); free(dr, M_TEMP); count++; } atomic_add_int(&unp_defers_count, -count); } } static void unp_internalize_fp(struct file *fp) { struct unpcb *unp; UNP_LINK_WLOCK(); if ((unp = fptounp(fp)) != NULL) { unp->unp_file = fp; unp->unp_msgcount++; } unp_rights++; UNP_LINK_WUNLOCK(); } static int unp_externalize_fp(struct file *fp) { struct unpcb *unp; int ret; UNP_LINK_WLOCK(); if ((unp = fptounp(fp)) != NULL) { unp->unp_msgcount--; ret = 1; } else ret = 0; unp_rights--; UNP_LINK_WUNLOCK(); return (ret); } /* * unp_defer indicates whether additional work has been defered for a future * pass through unp_gc(). It is thread local and does not require explicit * synchronization. */ static int unp_marked; static void unp_remove_dead_ref(struct filedescent **fdep, int fdcount) { struct unpcb *unp; struct file *fp; int i; /* * This function can only be called from the gc task. */ KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, ("%s: not on gc callout", __func__)); UNP_LINK_LOCK_ASSERT(); for (i = 0; i < fdcount; i++) { fp = fdep[i]->fde_file; if ((unp = fptounp(fp)) == NULL) continue; if ((unp->unp_gcflag & UNPGC_DEAD) == 0) continue; unp->unp_gcrefs--; } } static void unp_restore_undead_ref(struct filedescent **fdep, int fdcount) { struct unpcb *unp; struct file *fp; int i; /* * This function can only be called from the gc task. */ KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, ("%s: not on gc callout", __func__)); UNP_LINK_LOCK_ASSERT(); for (i = 0; i < fdcount; i++) { fp = fdep[i]->fde_file; if ((unp = fptounp(fp)) == NULL) continue; if ((unp->unp_gcflag & UNPGC_DEAD) == 0) continue; unp->unp_gcrefs++; unp_marked++; } } static void unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int)) { struct socket *so, *soa; so = unp->unp_socket; SOCK_LOCK(so); if (SOLISTENING(so)) { /* * Mark all sockets in our accept queue. */ TAILQ_FOREACH(soa, &so->sol_comp, so_list) { if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS) continue; SOCKBUF_LOCK(&soa->so_rcv); unp_scan(soa->so_rcv.sb_mb, op); SOCKBUF_UNLOCK(&soa->so_rcv); } } else { /* * Mark all sockets we reference with RIGHTS. */ if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) { SOCKBUF_LOCK(&so->so_rcv); unp_scan(so->so_rcv.sb_mb, op); SOCKBUF_UNLOCK(&so->so_rcv); } } SOCK_UNLOCK(so); } static int unp_recycled; SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, "Number of unreachable sockets claimed by the garbage collector."); static int unp_taskcount; SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, "Number of times the garbage collector has run."); SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0, "Number of active local sockets."); static void unp_gc(__unused void *arg, int pending) { struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, NULL }; struct unp_head **head; struct unp_head unp_deadhead; /* List of potentially-dead sockets. */ struct file *f, **unref; struct unpcb *unp, *unptmp; int i, total, unp_unreachable; LIST_INIT(&unp_deadhead); unp_taskcount++; UNP_LINK_RLOCK(); /* * First determine which sockets may be in cycles. */ unp_unreachable = 0; for (head = heads; *head != NULL; head++) LIST_FOREACH(unp, *head, unp_link) { KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0, ("%s: unp %p has unexpected gc flags 0x%x", __func__, unp, (unsigned int)unp->unp_gcflag)); f = unp->unp_file; /* * Check for an unreachable socket potentially in a * cycle. It must be in a queue as indicated by * msgcount, and this must equal the file reference * count. Note that when msgcount is 0 the file is * NULL. */ if (f != NULL && unp->unp_msgcount != 0 && refcount_load(&f->f_count) == unp->unp_msgcount) { LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead); unp->unp_gcflag |= UNPGC_DEAD; unp->unp_gcrefs = unp->unp_msgcount; unp_unreachable++; } } /* * Scan all sockets previously marked as potentially being in a cycle * and remove the references each socket holds on any UNPGC_DEAD * sockets in its queue. After this step, all remaining references on * sockets marked UNPGC_DEAD should not be part of any cycle. */ LIST_FOREACH(unp, &unp_deadhead, unp_dead) unp_gc_scan(unp, unp_remove_dead_ref); /* * If a socket still has a non-negative refcount, it cannot be in a * cycle. In this case increment refcount of all children iteratively. * Stop the scan once we do a complete loop without discovering * a new reachable socket. */ do { unp_marked = 0; LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp) if (unp->unp_gcrefs > 0) { unp->unp_gcflag &= ~UNPGC_DEAD; LIST_REMOVE(unp, unp_dead); KASSERT(unp_unreachable > 0, ("%s: unp_unreachable underflow.", __func__)); unp_unreachable--; unp_gc_scan(unp, unp_restore_undead_ref); } } while (unp_marked); UNP_LINK_RUNLOCK(); if (unp_unreachable == 0) return; /* * Allocate space for a local array of dead unpcbs. * TODO: can this path be simplified by instead using the local * dead list at unp_deadhead, after taking out references * on the file object and/or unpcb and dropping the link lock? */ unref = malloc(unp_unreachable * sizeof(struct file *), M_TEMP, M_WAITOK); /* * Iterate looking for sockets which have been specifically marked * as unreachable and store them locally. */ UNP_LINK_RLOCK(); total = 0; LIST_FOREACH(unp, &unp_deadhead, unp_dead) { KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0, ("%s: unp %p not marked UNPGC_DEAD", __func__, unp)); unp->unp_gcflag &= ~UNPGC_DEAD; f = unp->unp_file; if (unp->unp_msgcount == 0 || f == NULL || refcount_load(&f->f_count) != unp->unp_msgcount || !fhold(f)) continue; unref[total++] = f; KASSERT(total <= unp_unreachable, ("%s: incorrect unreachable count.", __func__)); } UNP_LINK_RUNLOCK(); /* * Now flush all sockets, free'ing rights. This will free the * struct files associated with these sockets but leave each socket * with one remaining ref. */ for (i = 0; i < total; i++) { struct socket *so; so = unref[i]->f_data; CURVNET_SET(so->so_vnet); sorflush(so); CURVNET_RESTORE(); } /* * And finally release the sockets so they can be reclaimed. */ for (i = 0; i < total; i++) fdrop(unref[i], NULL); unp_recycled += total; free(unref, M_TEMP); } static void unp_dispose_mbuf(struct mbuf *m) { if (m) unp_scan(m, unp_freerights); } /* * Synchronize against unp_gc, which can trip over data as we are freeing it. */ static void unp_dispose(struct socket *so) { struct unpcb *unp; unp = sotounpcb(so); UNP_LINK_WLOCK(); unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; UNP_LINK_WUNLOCK(); if (!SOLISTENING(so)) unp_dispose_mbuf(so->so_rcv.sb_mb); } static void unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) { struct mbuf *m; struct cmsghdr *cm; void *data; socklen_t clen, datalen; while (m0 != NULL) { for (m = m0; m; m = m->m_next) { if (m->m_type != MT_CONTROL) continue; cm = mtod(m, struct cmsghdr *); clen = m->m_len; while (cm != NULL) { if (sizeof(*cm) > clen || cm->cmsg_len > clen) break; data = CMSG_DATA(cm); datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; if (cm->cmsg_level == SOL_SOCKET && cm->cmsg_type == SCM_RIGHTS) { (*op)(data, datalen / sizeof(struct filedescent *)); } if (CMSG_SPACE(datalen) < clen) { clen -= CMSG_SPACE(datalen); cm = (struct cmsghdr *) ((caddr_t)cm + CMSG_SPACE(datalen)); } else { clen = 0; cm = NULL; } } } m0 = m0->m_nextpkt; } } /* * A helper function called by VFS before socket-type vnode reclamation. * For an active vnode it clears unp_vnode pointer and decrements unp_vnode * use count. */ void vfs_unp_reclaim(struct vnode *vp) { struct unpcb *unp; int active; struct mtx *vplock; ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); KASSERT(vp->v_type == VSOCK, ("vfs_unp_reclaim: vp->v_type != VSOCK")); active = 0; vplock = mtx_pool_find(mtxpool_sleep, vp); mtx_lock(vplock); VOP_UNP_CONNECT(vp, &unp); if (unp == NULL) goto done; UNP_PCB_LOCK(unp); if (unp->unp_vnode == vp) { VOP_UNP_DETACH(vp); unp->unp_vnode = NULL; active = 1; } UNP_PCB_UNLOCK(unp); done: mtx_unlock(vplock); if (active) vunref(vp); } #ifdef DDB static void db_print_indent(int indent) { int i; for (i = 0; i < indent; i++) db_printf(" "); } static void db_print_unpflags(int unp_flags) { int comma; comma = 0; if (unp_flags & UNP_HAVEPC) { db_printf("%sUNP_HAVEPC", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_WANTCRED_ALWAYS) { db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_WANTCRED_ONESHOT) { db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_CONNWAIT) { db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_CONNECTING) { db_printf("%sUNP_CONNECTING", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_BINDING) { db_printf("%sUNP_BINDING", comma ? ", " : ""); comma = 1; } } static void db_print_xucred(int indent, struct xucred *xu) { int comma, i; db_print_indent(indent); db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n", xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups); db_print_indent(indent); db_printf("cr_groups: "); comma = 0; for (i = 0; i < xu->cr_ngroups; i++) { db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); comma = 1; } db_printf("\n"); } static void db_print_unprefs(int indent, struct unp_head *uh) { struct unpcb *unp; int counter; counter = 0; LIST_FOREACH(unp, uh, unp_reflink) { if (counter % 4 == 0) db_print_indent(indent); db_printf("%p ", unp); if (counter % 4 == 3) db_printf("\n"); counter++; } if (counter != 0 && counter % 4 != 0) db_printf("\n"); } DB_SHOW_COMMAND(unpcb, db_show_unpcb) { struct unpcb *unp; if (!have_addr) { db_printf("usage: show unpcb \n"); return; } unp = (struct unpcb *)addr; db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, unp->unp_vnode); db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, unp->unp_conn); db_printf("unp_refs:\n"); db_print_unprefs(2, &unp->unp_refs); /* XXXRW: Would be nice to print the full address, if any. */ db_printf("unp_addr: %p\n", unp->unp_addr); db_printf("unp_gencnt: %llu\n", (unsigned long long)unp->unp_gencnt); db_printf("unp_flags: %x (", unp->unp_flags); db_print_unpflags(unp->unp_flags); db_printf(")\n"); db_printf("unp_peercred:\n"); db_print_xucred(2, &unp->unp_peercred); db_printf("unp_refcount: %u\n", unp->unp_refcount); } #endif diff --git a/sys/netinet/sctp_syscalls.c b/sys/netinet/sctp_syscalls.c index 0dad76b0f8bc..2697d139300c 100644 --- a/sys/netinet/sctp_syscalls.c +++ b/sys/netinet/sctp_syscalls.c @@ -1,589 +1,589 @@ /*- * Copyright (c) 1982, 1986, 1989, 1990, 1993 * The Regents of the University of California. 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. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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_capsicum.h" #include "opt_sctp.h" #include "opt_ktrace.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 #include #include #include #include #ifdef KTRACE #include #endif #ifdef COMPAT_FREEBSD32 #include #include #include #endif #include #include #include #include #include #include static struct syscall_helper_data sctp_syscalls[] = { SYSCALL_INIT_HELPER_F(sctp_peeloff, SYF_CAPENABLED), SYSCALL_INIT_HELPER_F(sctp_generic_sendmsg, SYF_CAPENABLED), SYSCALL_INIT_HELPER_F(sctp_generic_sendmsg_iov, SYF_CAPENABLED), SYSCALL_INIT_HELPER_F(sctp_generic_recvmsg, SYF_CAPENABLED), SYSCALL_INIT_LAST }; #ifdef COMPAT_FREEBSD32 static struct syscall_helper_data sctp32_syscalls[] = { SYSCALL32_INIT_HELPER_COMPAT(sctp_peeloff), SYSCALL32_INIT_HELPER_COMPAT(sctp_generic_sendmsg), SYSCALL32_INIT_HELPER_COMPAT(sctp_generic_sendmsg_iov), SYSCALL32_INIT_HELPER_COMPAT(sctp_generic_recvmsg), SYSCALL_INIT_LAST }; #endif int sctp_syscalls_init(void) { int error; error = syscall_helper_register(sctp_syscalls, SY_THR_STATIC_KLD); if (error != 0) return (error); #ifdef COMPAT_FREEBSD32 error = syscall32_helper_register(sctp32_syscalls, SY_THR_STATIC_KLD); if (error != 0) return (error); #endif return (0); } #ifdef SCTP SYSINIT(sctp_syscalls, SI_SUB_SYSCALLS, SI_ORDER_ANY, sctp_syscalls_init, NULL); #endif int sctp_syscalls_uninit(void) { int error; #ifdef COMPAT_FREEBSD32 error = syscall32_helper_unregister(sctp32_syscalls); if (error != 0) return (error); #endif error = syscall_helper_unregister(sctp_syscalls); if (error != 0) return (error); return (0); } /* * SCTP syscalls. */ int sys_sctp_peeloff(td, uap) struct thread *td; struct sctp_peeloff_args /* { int sd; caddr_t name; } */ *uap; { struct file *headfp, *nfp = NULL; struct socket *head, *so; cap_rights_t rights; u_int fflag; int error, fd; AUDIT_ARG_FD(uap->sd); - error = getsock_cap(td, uap->sd, cap_rights_init(&rights, CAP_PEELOFF), + error = getsock_cap(td, uap->sd, cap_rights_init_one(&rights, CAP_PEELOFF), &headfp, &fflag, NULL); if (error != 0) goto done2; head = headfp->f_data; if (head->so_proto->pr_protocol != IPPROTO_SCTP) { error = EOPNOTSUPP; goto done; } error = sctp_can_peel_off(head, (sctp_assoc_t)uap->name); if (error != 0) goto done; /* * At this point we know we do have a assoc to pull * we proceed to get the fd setup. This may block * but that is ok. */ error = falloc(td, &nfp, &fd, 0); if (error != 0) goto done; td->td_retval[0] = fd; CURVNET_SET(head->so_vnet); so = sopeeloff(head); if (so == NULL) { error = ENOMEM; goto noconnection; } finit(nfp, fflag, DTYPE_SOCKET, so, &socketops); error = sctp_do_peeloff(head, so, (sctp_assoc_t)uap->name); if (error != 0) goto noconnection; if (head->so_sigio != NULL) fsetown(fgetown(&head->so_sigio), &so->so_sigio); noconnection: /* * close the new descriptor, assuming someone hasn't ripped it * out from under us. */ if (error != 0) fdclose(td, nfp, fd); /* * Release explicitly held references before returning. */ CURVNET_RESTORE(); done: if (nfp != NULL) fdrop(nfp, td); fdrop(headfp, td); done2: return (error); } int sys_sctp_generic_sendmsg (td, uap) struct thread *td; struct sctp_generic_sendmsg_args /* { int sd, caddr_t msg, int mlen, caddr_t to, __socklen_t tolen, struct sctp_sndrcvinfo *sinfo, int flags } */ *uap; { struct sctp_sndrcvinfo sinfo, *u_sinfo = NULL; struct socket *so; struct file *fp = NULL; struct sockaddr *to = NULL; #ifdef KTRACE struct uio *ktruio = NULL; #endif struct uio auio; struct iovec iov[1]; cap_rights_t rights; int error = 0, len; if (uap->sinfo != NULL) { error = copyin(uap->sinfo, &sinfo, sizeof (sinfo)); if (error != 0) return (error); u_sinfo = &sinfo; } cap_rights_init_one(&rights, CAP_SEND); if (uap->tolen != 0) { error = getsockaddr(&to, uap->to, uap->tolen); if (error != 0) { to = NULL; goto sctp_bad2; } cap_rights_set_one(&rights, CAP_CONNECT); } AUDIT_ARG_FD(uap->sd); error = getsock_cap(td, uap->sd, &rights, &fp, NULL, NULL); if (error != 0) goto sctp_bad; #ifdef KTRACE if (to && (KTRPOINT(td, KTR_STRUCT))) ktrsockaddr(to); #endif iov[0].iov_base = uap->msg; iov[0].iov_len = uap->mlen; so = (struct socket *)fp->f_data; if (so->so_proto->pr_protocol != IPPROTO_SCTP) { error = EOPNOTSUPP; goto sctp_bad; } #ifdef MAC error = mac_socket_check_send(td->td_ucred, so); if (error != 0) goto sctp_bad; #endif /* MAC */ auio.uio_iov = iov; auio.uio_iovcnt = 1; auio.uio_segflg = UIO_USERSPACE; auio.uio_rw = UIO_WRITE; auio.uio_td = td; auio.uio_offset = 0; /* XXX */ auio.uio_resid = 0; #ifdef KTRACE if (KTRPOINT(td, KTR_GENIO)) ktruio = cloneuio(&auio); #endif /* KTRACE */ len = auio.uio_resid = uap->mlen; CURVNET_SET(so->so_vnet); error = sctp_lower_sosend(so, to, &auio, (struct mbuf *)NULL, (struct mbuf *)NULL, uap->flags, u_sinfo, td); CURVNET_RESTORE(); if (error != 0) { if (auio.uio_resid != len && (error == ERESTART || error == EINTR || error == EWOULDBLOCK)) error = 0; /* Generation of SIGPIPE can be controlled per socket. */ if (error == EPIPE && !(so->so_options & SO_NOSIGPIPE) && !(uap->flags & MSG_NOSIGNAL)) { PROC_LOCK(td->td_proc); tdsignal(td, SIGPIPE); PROC_UNLOCK(td->td_proc); } } if (error == 0) td->td_retval[0] = len - auio.uio_resid; #ifdef KTRACE if (ktruio != NULL) { ktruio->uio_resid = td->td_retval[0]; ktrgenio(uap->sd, UIO_WRITE, ktruio, error); } #endif /* KTRACE */ sctp_bad: if (fp != NULL) fdrop(fp, td); sctp_bad2: free(to, M_SONAME); return (error); } int sys_sctp_generic_sendmsg_iov(td, uap) struct thread *td; struct sctp_generic_sendmsg_iov_args /* { int sd, struct iovec *iov, int iovlen, caddr_t to, __socklen_t tolen, struct sctp_sndrcvinfo *sinfo, int flags } */ *uap; { struct sctp_sndrcvinfo sinfo, *u_sinfo = NULL; struct socket *so; struct file *fp = NULL; struct sockaddr *to = NULL; #ifdef KTRACE struct uio *ktruio = NULL; #endif struct uio auio; struct iovec *iov, *tiov; cap_rights_t rights; ssize_t len; int error, i; if (uap->sinfo != NULL) { error = copyin(uap->sinfo, &sinfo, sizeof (sinfo)); if (error != 0) return (error); u_sinfo = &sinfo; } cap_rights_init_one(&rights, CAP_SEND); if (uap->tolen != 0) { error = getsockaddr(&to, uap->to, uap->tolen); if (error != 0) { to = NULL; goto sctp_bad2; } cap_rights_set_one(&rights, CAP_CONNECT); } AUDIT_ARG_FD(uap->sd); error = getsock_cap(td, uap->sd, &rights, &fp, NULL, NULL); if (error != 0) goto sctp_bad1; #ifdef COMPAT_FREEBSD32 if (SV_CURPROC_FLAG(SV_ILP32)) error = freebsd32_copyiniov((struct iovec32 *)uap->iov, uap->iovlen, &iov, EMSGSIZE); else #endif error = copyiniov(uap->iov, uap->iovlen, &iov, EMSGSIZE); if (error != 0) goto sctp_bad1; #ifdef KTRACE if (to && (KTRPOINT(td, KTR_STRUCT))) ktrsockaddr(to); #endif so = (struct socket *)fp->f_data; if (so->so_proto->pr_protocol != IPPROTO_SCTP) { error = EOPNOTSUPP; goto sctp_bad; } #ifdef MAC error = mac_socket_check_send(td->td_ucred, so); if (error != 0) goto sctp_bad; #endif /* MAC */ auio.uio_iov = iov; auio.uio_iovcnt = uap->iovlen; auio.uio_segflg = UIO_USERSPACE; auio.uio_rw = UIO_WRITE; auio.uio_td = td; auio.uio_offset = 0; /* XXX */ auio.uio_resid = 0; tiov = iov; for (i = 0; i iovlen; i++, tiov++) { if ((auio.uio_resid += tiov->iov_len) < 0) { error = EINVAL; goto sctp_bad; } } #ifdef KTRACE if (KTRPOINT(td, KTR_GENIO)) ktruio = cloneuio(&auio); #endif /* KTRACE */ len = auio.uio_resid; CURVNET_SET(so->so_vnet); error = sctp_lower_sosend(so, to, &auio, (struct mbuf *)NULL, (struct mbuf *)NULL, uap->flags, u_sinfo, td); CURVNET_RESTORE(); if (error != 0) { if (auio.uio_resid != len && (error == ERESTART || error == EINTR || error == EWOULDBLOCK)) error = 0; /* Generation of SIGPIPE can be controlled per socket */ if (error == EPIPE && !(so->so_options & SO_NOSIGPIPE) && !(uap->flags & MSG_NOSIGNAL)) { PROC_LOCK(td->td_proc); tdsignal(td, SIGPIPE); PROC_UNLOCK(td->td_proc); } } if (error == 0) td->td_retval[0] = len - auio.uio_resid; #ifdef KTRACE if (ktruio != NULL) { ktruio->uio_resid = td->td_retval[0]; ktrgenio(uap->sd, UIO_WRITE, ktruio, error); } #endif /* KTRACE */ sctp_bad: free(iov, M_IOV); sctp_bad1: if (fp != NULL) fdrop(fp, td); sctp_bad2: free(to, M_SONAME); return (error); } int sys_sctp_generic_recvmsg(td, uap) struct thread *td; struct sctp_generic_recvmsg_args /* { int sd, struct iovec *iov, int iovlen, struct sockaddr *from, __socklen_t *fromlenaddr, struct sctp_sndrcvinfo *sinfo, int *msg_flags } */ *uap; { uint8_t sockbufstore[256]; struct uio auio; struct iovec *iov, *tiov; struct sctp_sndrcvinfo sinfo; struct socket *so; struct file *fp = NULL; struct sockaddr *fromsa; cap_rights_t rights; #ifdef KTRACE struct uio *ktruio = NULL; #endif ssize_t len; int error, fromlen, i, msg_flags; AUDIT_ARG_FD(uap->sd); - error = getsock_cap(td, uap->sd, cap_rights_init(&rights, CAP_RECV), + error = getsock_cap(td, uap->sd, cap_rights_init_one(&rights, CAP_RECV), &fp, NULL, NULL); if (error != 0) return (error); #ifdef COMPAT_FREEBSD32 if (SV_CURPROC_FLAG(SV_ILP32)) error = freebsd32_copyiniov((struct iovec32 *)uap->iov, uap->iovlen, &iov, EMSGSIZE); else #endif error = copyiniov(uap->iov, uap->iovlen, &iov, EMSGSIZE); if (error != 0) goto out1; so = fp->f_data; if (so->so_proto->pr_protocol != IPPROTO_SCTP) { error = EOPNOTSUPP; goto out; } #ifdef MAC error = mac_socket_check_receive(td->td_ucred, so); if (error != 0) goto out; #endif /* MAC */ if (uap->fromlenaddr != NULL) { error = copyin(uap->fromlenaddr, &fromlen, sizeof (fromlen)); if (error != 0) goto out; } else { fromlen = 0; } if (uap->msg_flags) { error = copyin(uap->msg_flags, &msg_flags, sizeof (int)); if (error != 0) goto out; } else { msg_flags = 0; } auio.uio_iov = iov; auio.uio_iovcnt = uap->iovlen; auio.uio_segflg = UIO_USERSPACE; auio.uio_rw = UIO_READ; auio.uio_td = td; auio.uio_offset = 0; /* XXX */ auio.uio_resid = 0; tiov = iov; for (i = 0; i iovlen; i++, tiov++) { if ((auio.uio_resid += tiov->iov_len) < 0) { error = EINVAL; goto out; } } len = auio.uio_resid; fromsa = (struct sockaddr *)sockbufstore; #ifdef KTRACE if (KTRPOINT(td, KTR_GENIO)) ktruio = cloneuio(&auio); #endif /* KTRACE */ memset(&sinfo, 0, sizeof(struct sctp_sndrcvinfo)); CURVNET_SET(so->so_vnet); error = sctp_sorecvmsg(so, &auio, (struct mbuf **)NULL, fromsa, fromlen, &msg_flags, (struct sctp_sndrcvinfo *)&sinfo, 1); CURVNET_RESTORE(); if (error != 0) { if (auio.uio_resid != len && (error == ERESTART || error == EINTR || error == EWOULDBLOCK)) error = 0; } else { if (uap->sinfo) error = copyout(&sinfo, uap->sinfo, sizeof (sinfo)); } #ifdef KTRACE if (ktruio != NULL) { ktruio->uio_resid = len - auio.uio_resid; ktrgenio(uap->sd, UIO_READ, ktruio, error); } #endif /* KTRACE */ if (error != 0) goto out; td->td_retval[0] = len - auio.uio_resid; if (fromlen && uap->from) { len = fromlen; if (len <= 0 || fromsa == NULL) len = 0; else { len = MIN(len, fromsa->sa_len); error = copyout(fromsa, uap->from, (size_t)len); if (error != 0) goto out; } error = copyout(&len, uap->fromlenaddr, sizeof (socklen_t)); if (error != 0) goto out; } #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT)) ktrsockaddr(fromsa); #endif if (uap->msg_flags) { error = copyout(&msg_flags, uap->msg_flags, sizeof (int)); if (error != 0) goto out; } out: free(iov, M_IOV); out1: if (fp != NULL) fdrop(fp, td); return (error); } diff --git a/sys/security/mac/mac_syscalls.c b/sys/security/mac/mac_syscalls.c index c28a5e99c656..e7d71b2e22da 100644 --- a/sys/security/mac/mac_syscalls.c +++ b/sys/security/mac/mac_syscalls.c @@ -1,661 +1,663 @@ /*- * Copyright (c) 1999-2002, 2006, 2009 Robert N. M. Watson * Copyright (c) 2001 Ilmar S. Habibulin * Copyright (c) 2001-2005 Networks Associates Technology, Inc. * Copyright (c) 2005-2006 SPARTA, Inc. * Copyright (c) 2008 Apple Inc. * All rights reserved. * * This software was developed by Robert Watson and Ilmar Habibulin for the * TrustedBSD Project. * * This software was developed for the FreeBSD Project in part by Network * Associates Laboratories, the Security Research Division of Network * Associates, Inc. under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), * as part of the DARPA CHATS research program. * * This software was enhanced by SPARTA ISSO under SPAWAR contract * N66001-04-C-6019 ("SEFOS"). * * This software was developed at the University of Cambridge Computer * Laboratory with support from a grant from Google, Inc. * * 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_mac.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef MAC FEATURE(security_mac, "Mandatory Access Control Framework support"); static int kern___mac_get_path(struct thread *td, const char *path_p, struct mac *mac_p, int follow); static int kern___mac_set_path(struct thread *td, const char *path_p, struct mac *mac_p, int follow); int sys___mac_get_pid(struct thread *td, struct __mac_get_pid_args *uap) { char *elements, *buffer; struct mac mac; struct proc *tproc; struct ucred *tcred; int error; error = copyin(uap->mac_p, &mac, sizeof(mac)); if (error) return (error); error = mac_check_structmac_consistent(&mac); if (error) return (error); tproc = pfind(uap->pid); if (tproc == NULL) return (ESRCH); tcred = NULL; /* Satisfy gcc. */ error = p_cansee(td, tproc); if (error == 0) tcred = crhold(tproc->p_ucred); PROC_UNLOCK(tproc); if (error) return (error); elements = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK); error = copyinstr(mac.m_string, elements, mac.m_buflen, NULL); if (error) { free(elements, M_MACTEMP); crfree(tcred); return (error); } buffer = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK | M_ZERO); error = mac_cred_externalize_label(tcred->cr_label, elements, buffer, mac.m_buflen); if (error == 0) error = copyout(buffer, mac.m_string, strlen(buffer)+1); free(buffer, M_MACTEMP); free(elements, M_MACTEMP); crfree(tcred); return (error); } int sys___mac_get_proc(struct thread *td, struct __mac_get_proc_args *uap) { char *elements, *buffer; struct mac mac; int error; error = copyin(uap->mac_p, &mac, sizeof(mac)); if (error) return (error); error = mac_check_structmac_consistent(&mac); if (error) return (error); elements = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK); error = copyinstr(mac.m_string, elements, mac.m_buflen, NULL); if (error) { free(elements, M_MACTEMP); return (error); } buffer = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK | M_ZERO); error = mac_cred_externalize_label(td->td_ucred->cr_label, elements, buffer, mac.m_buflen); if (error == 0) error = copyout(buffer, mac.m_string, strlen(buffer)+1); free(buffer, M_MACTEMP); free(elements, M_MACTEMP); return (error); } int sys___mac_set_proc(struct thread *td, struct __mac_set_proc_args *uap) { struct ucred *newcred, *oldcred; struct label *intlabel; struct proc *p; struct mac mac; char *buffer; int error; if (!(mac_labeled & MPC_OBJECT_CRED)) return (EINVAL); error = copyin(uap->mac_p, &mac, sizeof(mac)); if (error) return (error); error = mac_check_structmac_consistent(&mac); if (error) return (error); buffer = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK); error = copyinstr(mac.m_string, buffer, mac.m_buflen, NULL); if (error) { free(buffer, M_MACTEMP); return (error); } intlabel = mac_cred_label_alloc(); error = mac_cred_internalize_label(intlabel, buffer); free(buffer, M_MACTEMP); if (error) goto out; newcred = crget(); p = td->td_proc; PROC_LOCK(p); oldcred = p->p_ucred; error = mac_cred_check_relabel(oldcred, intlabel); if (error) { PROC_UNLOCK(p); crfree(newcred); goto out; } setsugid(p); crcopy(newcred, oldcred); mac_cred_relabel(newcred, intlabel); proc_set_cred(p, newcred); PROC_UNLOCK(p); crfree(oldcred); mac_proc_vm_revoke(td); out: mac_cred_label_free(intlabel); return (error); } int sys___mac_get_fd(struct thread *td, struct __mac_get_fd_args *uap) { char *elements, *buffer; struct label *intlabel; struct file *fp; struct mac mac; struct vnode *vp; struct pipe *pipe; struct socket *so; cap_rights_t rights; int error; error = copyin(uap->mac_p, &mac, sizeof(mac)); if (error) return (error); error = mac_check_structmac_consistent(&mac); if (error) return (error); elements = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK); error = copyinstr(mac.m_string, elements, mac.m_buflen, NULL); if (error) { free(elements, M_MACTEMP); return (error); } buffer = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK | M_ZERO); - error = fget(td, uap->fd, cap_rights_init(&rights, CAP_MAC_GET), &fp); + error = fget(td, uap->fd, cap_rights_init_one(&rights, CAP_MAC_GET), + &fp); if (error) goto out; switch (fp->f_type) { case DTYPE_FIFO: case DTYPE_VNODE: if (!(mac_labeled & MPC_OBJECT_VNODE)) { error = EINVAL; goto out_fdrop; } vp = fp->f_vnode; intlabel = mac_vnode_label_alloc(); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); mac_vnode_copy_label(vp->v_label, intlabel); VOP_UNLOCK(vp); error = mac_vnode_externalize_label(intlabel, elements, buffer, mac.m_buflen); mac_vnode_label_free(intlabel); break; case DTYPE_PIPE: if (!(mac_labeled & MPC_OBJECT_PIPE)) { error = EINVAL; goto out_fdrop; } pipe = fp->f_data; intlabel = mac_pipe_label_alloc(); PIPE_LOCK(pipe); mac_pipe_copy_label(pipe->pipe_pair->pp_label, intlabel); PIPE_UNLOCK(pipe); error = mac_pipe_externalize_label(intlabel, elements, buffer, mac.m_buflen); mac_pipe_label_free(intlabel); break; case DTYPE_SOCKET: if (!(mac_labeled & MPC_OBJECT_SOCKET)) { error = EINVAL; goto out_fdrop; } so = fp->f_data; intlabel = mac_socket_label_alloc(M_WAITOK); SOCK_LOCK(so); mac_socket_copy_label(so->so_label, intlabel); SOCK_UNLOCK(so); error = mac_socket_externalize_label(intlabel, elements, buffer, mac.m_buflen); mac_socket_label_free(intlabel); break; default: error = EINVAL; } if (error == 0) error = copyout(buffer, mac.m_string, strlen(buffer)+1); out_fdrop: fdrop(fp, td); out: free(buffer, M_MACTEMP); free(elements, M_MACTEMP); return (error); } int sys___mac_get_file(struct thread *td, struct __mac_get_file_args *uap) { return (kern___mac_get_path(td, uap->path_p, uap->mac_p, FOLLOW)); } int sys___mac_get_link(struct thread *td, struct __mac_get_link_args *uap) { return (kern___mac_get_path(td, uap->path_p, uap->mac_p, NOFOLLOW)); } static int kern___mac_get_path(struct thread *td, const char *path_p, struct mac *mac_p, int follow) { char *elements, *buffer; struct nameidata nd; struct label *intlabel; struct mac mac; int error; if (!(mac_labeled & MPC_OBJECT_VNODE)) return (EINVAL); error = copyin(mac_p, &mac, sizeof(mac)); if (error) return (error); error = mac_check_structmac_consistent(&mac); if (error) return (error); elements = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK); error = copyinstr(mac.m_string, elements, mac.m_buflen, NULL); if (error) { free(elements, M_MACTEMP); return (error); } buffer = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK | M_ZERO); NDINIT(&nd, LOOKUP, LOCKLEAF | follow, UIO_USERSPACE, path_p, td); error = namei(&nd); if (error) goto out; intlabel = mac_vnode_label_alloc(); mac_vnode_copy_label(nd.ni_vp->v_label, intlabel); error = mac_vnode_externalize_label(intlabel, elements, buffer, mac.m_buflen); NDFREE(&nd, 0); mac_vnode_label_free(intlabel); if (error == 0) error = copyout(buffer, mac.m_string, strlen(buffer)+1); out: free(buffer, M_MACTEMP); free(elements, M_MACTEMP); return (error); } int sys___mac_set_fd(struct thread *td, struct __mac_set_fd_args *uap) { struct label *intlabel; struct pipe *pipe; struct socket *so; struct file *fp; struct mount *mp; struct vnode *vp; struct mac mac; cap_rights_t rights; char *buffer; int error; error = copyin(uap->mac_p, &mac, sizeof(mac)); if (error) return (error); error = mac_check_structmac_consistent(&mac); if (error) return (error); buffer = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK); error = copyinstr(mac.m_string, buffer, mac.m_buflen, NULL); if (error) { free(buffer, M_MACTEMP); return (error); } - error = fget(td, uap->fd, cap_rights_init(&rights, CAP_MAC_SET), &fp); + error = fget(td, uap->fd, cap_rights_init_one(&rights, CAP_MAC_SET), + &fp); if (error) goto out; switch (fp->f_type) { case DTYPE_FIFO: case DTYPE_VNODE: if (!(mac_labeled & MPC_OBJECT_VNODE)) { error = EINVAL; goto out_fdrop; } intlabel = mac_vnode_label_alloc(); error = mac_vnode_internalize_label(intlabel, buffer); if (error) { mac_vnode_label_free(intlabel); break; } vp = fp->f_vnode; error = vn_start_write(vp, &mp, V_WAIT | PCATCH); if (error != 0) { mac_vnode_label_free(intlabel); break; } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); error = vn_setlabel(vp, intlabel, td->td_ucred); VOP_UNLOCK(vp); vn_finished_write(mp); mac_vnode_label_free(intlabel); break; case DTYPE_PIPE: if (!(mac_labeled & MPC_OBJECT_PIPE)) { error = EINVAL; goto out_fdrop; } intlabel = mac_pipe_label_alloc(); error = mac_pipe_internalize_label(intlabel, buffer); if (error == 0) { pipe = fp->f_data; PIPE_LOCK(pipe); error = mac_pipe_label_set(td->td_ucred, pipe->pipe_pair, intlabel); PIPE_UNLOCK(pipe); } mac_pipe_label_free(intlabel); break; case DTYPE_SOCKET: if (!(mac_labeled & MPC_OBJECT_SOCKET)) { error = EINVAL; goto out_fdrop; } intlabel = mac_socket_label_alloc(M_WAITOK); error = mac_socket_internalize_label(intlabel, buffer); if (error == 0) { so = fp->f_data; error = mac_socket_label_set(td->td_ucred, so, intlabel); } mac_socket_label_free(intlabel); break; default: error = EINVAL; } out_fdrop: fdrop(fp, td); out: free(buffer, M_MACTEMP); return (error); } int sys___mac_set_file(struct thread *td, struct __mac_set_file_args *uap) { return (kern___mac_set_path(td, uap->path_p, uap->mac_p, FOLLOW)); } int sys___mac_set_link(struct thread *td, struct __mac_set_link_args *uap) { return (kern___mac_set_path(td, uap->path_p, uap->mac_p, NOFOLLOW)); } static int kern___mac_set_path(struct thread *td, const char *path_p, struct mac *mac_p, int follow) { struct label *intlabel; struct nameidata nd; struct mount *mp; struct mac mac; char *buffer; int error; if (!(mac_labeled & MPC_OBJECT_VNODE)) return (EINVAL); error = copyin(mac_p, &mac, sizeof(mac)); if (error) return (error); error = mac_check_structmac_consistent(&mac); if (error) return (error); buffer = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK); error = copyinstr(mac.m_string, buffer, mac.m_buflen, NULL); if (error) { free(buffer, M_MACTEMP); return (error); } intlabel = mac_vnode_label_alloc(); error = mac_vnode_internalize_label(intlabel, buffer); free(buffer, M_MACTEMP); if (error) goto out; NDINIT(&nd, LOOKUP, LOCKLEAF | follow, UIO_USERSPACE, path_p, td); error = namei(&nd); if (error == 0) { error = vn_start_write(nd.ni_vp, &mp, V_WAIT | PCATCH); if (error == 0) { error = vn_setlabel(nd.ni_vp, intlabel, td->td_ucred); vn_finished_write(mp); } } NDFREE(&nd, 0); out: mac_vnode_label_free(intlabel); return (error); } int sys_mac_syscall(struct thread *td, struct mac_syscall_args *uap) { struct mac_policy_conf *mpc; char target[MAC_MAX_POLICY_NAME]; int error; error = copyinstr(uap->policy, target, sizeof(target), NULL); if (error) return (error); error = ENOSYS; LIST_FOREACH(mpc, &mac_static_policy_list, mpc_list) { if (strcmp(mpc->mpc_name, target) == 0 && mpc->mpc_ops->mpo_syscall != NULL) { error = mpc->mpc_ops->mpo_syscall(td, uap->call, uap->arg); goto out; } } if (!LIST_EMPTY(&mac_policy_list)) { mac_policy_slock_sleep(); LIST_FOREACH(mpc, &mac_policy_list, mpc_list) { if (strcmp(mpc->mpc_name, target) == 0 && mpc->mpc_ops->mpo_syscall != NULL) { error = mpc->mpc_ops->mpo_syscall(td, uap->call, uap->arg); break; } } mac_policy_sunlock_sleep(); } out: return (error); } #else /* !MAC */ int sys___mac_get_pid(struct thread *td, struct __mac_get_pid_args *uap) { return (ENOSYS); } int sys___mac_get_proc(struct thread *td, struct __mac_get_proc_args *uap) { return (ENOSYS); } int sys___mac_set_proc(struct thread *td, struct __mac_set_proc_args *uap) { return (ENOSYS); } int sys___mac_get_fd(struct thread *td, struct __mac_get_fd_args *uap) { return (ENOSYS); } int sys___mac_get_file(struct thread *td, struct __mac_get_file_args *uap) { return (ENOSYS); } int sys___mac_get_link(struct thread *td, struct __mac_get_link_args *uap) { return (ENOSYS); } int sys___mac_set_fd(struct thread *td, struct __mac_set_fd_args *uap) { return (ENOSYS); } int sys___mac_set_file(struct thread *td, struct __mac_set_file_args *uap) { return (ENOSYS); } int sys___mac_set_link(struct thread *td, struct __mac_set_link_args *uap) { return (ENOSYS); } int sys_mac_syscall(struct thread *td, struct mac_syscall_args *uap) { return (ENOSYS); } #endif /* !MAC */ diff --git a/sys/security/mac_veriexec/mac_veriexec.c b/sys/security/mac_veriexec/mac_veriexec.c index 8d43d59c9a3c..ecaa8cc35e09 100644 --- a/sys/security/mac_veriexec/mac_veriexec.c +++ b/sys/security/mac_veriexec/mac_veriexec.c @@ -1,882 +1,882 @@ /* * $FreeBSD$ * * Copyright (c) 2011, 2012, 2013, 2015, 2016, 2019 Juniper Networks, Inc. * 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 ``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 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 #include "opt_capsicum.h" #include "opt_mac.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "mac_veriexec.h" #include "mac_veriexec_internal.h" #define SLOT(l) \ mac_label_get((l), mac_veriexec_slot) #define SLOT_SET(l, v) \ mac_label_set((l), mac_veriexec_slot, (v)) #ifdef MAC_DEBUG #define MAC_VERIEXEC_DBG(_lvl, _fmt, ...) \ do { \ VERIEXEC_DEBUG((_lvl), (MAC_VERIEXEC_FULLNAME ": " _fmt \ "\n", ##__VA_ARGS__)); \ } while(0) #else #define MAC_VERIEXEC_DBG(_lvl, _fmt, ...) #endif static int sysctl_mac_veriexec_state(SYSCTL_HANDLER_ARGS); static int sysctl_mac_veriexec_db(SYSCTL_HANDLER_ARGS); SYSCTL_DECL(_security_mac); SYSCTL_NODE(_security_mac, OID_AUTO, veriexec, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "MAC/veriexec policy controls"); int mac_veriexec_debug; SYSCTL_INT(_security_mac_veriexec, OID_AUTO, debug, CTLFLAG_RW, &mac_veriexec_debug, 0, "Debug level"); static int mac_veriexec_state; SYSCTL_PROC(_security_mac_veriexec, OID_AUTO, state, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 0, 0, sysctl_mac_veriexec_state, "A", "Verified execution subsystem state"); SYSCTL_PROC(_security_mac_veriexec, OID_AUTO, db, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, 0, 0, sysctl_mac_veriexec_db, "A", "Verified execution fingerprint database"); static int mac_veriexec_slot; MALLOC_DEFINE(M_VERIEXEC, "veriexec", "Verified execution data"); /** * @internal * @brief Handler for security.mac.veriexec.db sysctl * * Display a human-readable form of the current fingerprint database. */ static int sysctl_mac_veriexec_db(SYSCTL_HANDLER_ARGS) { struct sbuf sb; int error; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sbuf_new_for_sysctl(&sb, NULL, 1024, req); mac_veriexec_metadata_print_db(&sb); error = sbuf_finish(&sb); sbuf_delete(&sb); return (error); } /** * @internal * @brief Generate human-readable output about the current verified execution * state. * * @param sbp sbuf to write output to */ static void mac_veriexec_print_state(struct sbuf *sbp) { if (mac_veriexec_state & VERIEXEC_STATE_INACTIVE) sbuf_printf(sbp, "inactive "); if (mac_veriexec_state & VERIEXEC_STATE_LOADED) sbuf_printf(sbp, "loaded "); if (mac_veriexec_state & VERIEXEC_STATE_ACTIVE) sbuf_printf(sbp, "active "); if (mac_veriexec_state & VERIEXEC_STATE_ENFORCE) sbuf_printf(sbp, "enforce "); if (mac_veriexec_state & VERIEXEC_STATE_LOCKED) sbuf_printf(sbp, "locked "); if (mac_veriexec_state != 0) sbuf_trim(sbp); } /** * @internal * @brief Handler for security.mac.veriexec.state sysctl * * Display a human-readable form of the current verified execution subsystem * state. */ static int sysctl_mac_veriexec_state(SYSCTL_HANDLER_ARGS) { struct sbuf sb; int error; sbuf_new(&sb, NULL, 128, SBUF_AUTOEXTEND); mac_veriexec_print_state(&sb); sbuf_finish(&sb); error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb)); sbuf_delete(&sb); return (error); } /** * @internal * @brief Event handler called when a virtual file system is mounted. * * We need to record the file system identifier in the MAC per-policy slot * assigned to veriexec, so we have a key to use in order to reference the * mount point in the meta-data store. * * @param arg unused argument * @param mp mount point that is being mounted * @param fsrootvp vnode of the file system root * @param td calling thread */ static void mac_veriexec_vfs_mounted(void *arg __unused, struct mount *mp, struct vnode *fsrootvp, struct thread *td) { struct vattr va; int error; error = VOP_GETATTR(fsrootvp, &va, td->td_ucred); if (error) return; SLOT_SET(mp->mnt_label, va.va_fsid); #ifdef MAC_DEBUG MAC_VERIEXEC_DBG(3, "set fsid to %ju for mount %p", (uintmax_t)va.va_fsid, mp); #endif } /** * @internal * @brief Event handler called when a virtual file system is unmounted. * * If we recorded a file system identifier in the MAC per-policy slot assigned * to veriexec, then we need to tell the meta-data store to clean up. * * @param arg unused argument * @param mp mount point that is being unmounted * @param td calling thread */ static void mac_veriexec_vfs_unmounted(void *arg __unused, struct mount *mp, struct thread *td) { dev_t fsid; fsid = SLOT(mp->mnt_label); if (fsid) { MAC_VERIEXEC_DBG(3, "fsid %ju, cleaning up mount", (uintmax_t)fsid); mac_veriexec_metadata_unmounted(fsid, td); } } /** * @internal * @brief The mount point is being initialized, set the value in the MAC * per-policy slot for veriexec to zero. * * @note A value of zero in this slot indicates no file system identifier * is assigned. * * @param label the label that is being initialized */ static void mac_veriexec_mount_init_label(struct label *label) { SLOT_SET(label, 0); } /** * @internal * @brief The mount-point is being destroyed, reset the value in the MAC * per-policy slot for veriexec back to zero. * * @note A value of zero in this slot indicates no file system identifier * is assigned. * * @param label the label that is being destroyed */ static void mac_veriexec_mount_destroy_label(struct label *label) { SLOT_SET(label, 0); } /** * @internal * @brief The vnode label is being initialized, set the value in the MAC * per-policy slot for veriexec to @c FINGERPRINT_INVALID * * @note @c FINGERPRINT_INVALID indicates the fingerprint is invalid. * * @param label the label that is being initialized */ static void mac_veriexec_vnode_init_label(struct label *label) { SLOT_SET(label, FINGERPRINT_INVALID); } /** * @internal * @brief The vnode label is being destroyed, reset the value in the MAC * per-policy slot for veriexec back to @c FINGERPRINT_INVALID * * @note @c FINGERPRINT_INVALID indicates the fingerprint is invalid. * * @param label the label that is being destroyed */ static void mac_veriexec_vnode_destroy_label(struct label *label) { SLOT_SET(label, FINGERPRINT_INVALID); } /** * @internal * @brief Copy the value in the MAC per-policy slot assigned to veriexec from * the @p src label to the @p dest label */ static void mac_veriexec_copy_label(struct label *src, struct label *dest) { SLOT_SET(dest, SLOT(src)); } /** * @internal * @brief Check if the requested process can be debugged * * @param cred credentials to use * @param p process to debug * * @return 0 if debugging is allowed, otherwise an error code. */ static int mac_veriexec_proc_check_debug(struct ucred *cred, struct proc *p) { int error, flags; /* If we are not enforcing veriexec, nothing for us to check */ if ((mac_veriexec_state & VERIEXEC_STATE_ENFORCE) == 0) return (0); error = mac_veriexec_metadata_get_executable_flags(cred, p, &flags, 0); if (error != 0) return (0); return ((flags & VERIEXEC_NOTRACE) ? EACCES : 0); } /** * @internal * @brief A KLD load has been requested and needs to be validated. * * @param cred credentials to use * @param vp vnode of the KLD that has been requested * @param vlabel vnode label assigned to the vnode * * @return 0 if the KLD load is allowed, otherwise an error code. */ static int mac_veriexec_kld_check_load(struct ucred *cred, struct vnode *vp, struct label *vlabel) { struct vattr va; struct thread *td = curthread; fingerprint_status_t status; int error; /* * If we are not actively enforcing, allow it */ if ((mac_veriexec_state & VERIEXEC_STATE_ENFORCE) == 0) return (0); /* Get vnode attributes */ error = VOP_GETATTR(vp, &va, cred); if (error) return (error); /* * Fetch the fingerprint status for the vnode * (starting with files first) */ error = mac_veriexec_metadata_fetch_fingerprint_status(vp, &va, td, VERIEXEC_FILES_FIRST); if (error && error != EAUTH) return (error); /* * By now we should have status... */ status = mac_veriexec_get_fingerprint_status(vp); switch (status) { case FINGERPRINT_FILE: case FINGERPRINT_VALID: case FINGERPRINT_INDIRECT: if (error) return (error); break; default: /* * kldload should fail unless there is a valid fingerprint * registered. */ MAC_VERIEXEC_DBG(2, "fingerprint status is %d for dev %ju, " "file %ju.%ju\n", status, (uintmax_t)va.va_fsid, (uintmax_t)va.va_fileid, (uintmax_t)va.va_gen); return (EAUTH); } /* Everything is good, allow the KLD to be loaded */ return (0); } /** * @internal * @brief Check privileges that veriexec needs to be concerned about. * * The following privileges are checked by this function: * - PRIV_KMEM_WRITE\n * Check if writes to /dev/mem and /dev/kmem are allowed\n * (Only trusted processes are allowed) * * @param cred credentials to use * @param priv privilege to check * * @return 0 if the privilege is allowed, error code otherwise. */ static int mac_veriexec_priv_check(struct ucred *cred, int priv) { /* If we are not enforcing veriexec, nothing for us to check */ if ((mac_veriexec_state & VERIEXEC_STATE_ENFORCE) == 0) return (0); switch (priv) { case PRIV_KMEM_WRITE: if (!mac_veriexec_proc_is_trusted(cred, curproc)) return (EPERM); break; default: break; } return (0); } static int mac_veriexec_sysctl_check(struct ucred *cred, struct sysctl_oid *oidp, void *arg1, int arg2, struct sysctl_req *req) { struct sysctl_oid *oid; /* If we are not enforcing veriexec, nothing for us to check */ if ((mac_veriexec_state & VERIEXEC_STATE_ENFORCE) == 0) return (0); oid = oidp; if (oid->oid_kind & CTLFLAG_SECURE) { return (EPERM); /* XXX call mac_veriexec_priv_check? */ } return 0; } /** * @internal * @brief A program is being executed and needs to be validated. * * @param cred credentials to use * @param vp vnode of the program that is being executed * @param label vnode label assigned to the vnode * @param imgp parameters for the image to be executed * @param execlabel optional exec label * * @return 0 if the program should be allowed to execute, otherwise an error * code. */ static int mac_veriexec_vnode_check_exec(struct ucred *cred __unused, struct vnode *vp __unused, struct label *label __unused, struct image_params *imgp, struct label *execlabel __unused) { struct thread *td = curthread; int error; error = mac_veriexec_fingerprint_check_image(imgp, 0, td); return (error); } /** * @brief Check fingerprint for the specified vnode and validate it * * @param cred credentials to use * @param vp vnode of the file * @param accmode access mode to check (read, write, append, create, * verify, etc.) * * @return 0 if the file validated, otherwise an error code. */ static int mac_veriexec_check_vp(struct ucred *cred, struct vnode *vp, accmode_t accmode) { struct vattr va; struct thread *td = curthread; fingerprint_status_t status; int error; /* Get vnode attributes */ error = VOP_GETATTR(vp, &va, cred); if (error) return (error); /* Get the fingerprint status for the file */ error = mac_veriexec_metadata_fetch_fingerprint_status(vp, &va, td, VERIEXEC_FILES_FIRST); if (error && error != EAUTH) return (error); /* * By now we should have status... */ status = mac_veriexec_get_fingerprint_status(vp); if (accmode & VWRITE) { /* * If file has a fingerprint then deny the write request, * otherwise invalidate the status so we don't keep checking * for the file having a fingerprint. */ switch (status) { case FINGERPRINT_FILE: case FINGERPRINT_VALID: case FINGERPRINT_INDIRECT: MAC_VERIEXEC_DBG(2, "attempted write to fingerprinted file for dev " "%ju, file %ju.%ju\n", (uintmax_t)va.va_fsid, (uintmax_t)va.va_fileid, (uintmax_t)va.va_gen); return (EPERM); default: break; } } if (accmode & VVERIFY) { switch (status) { case FINGERPRINT_FILE: case FINGERPRINT_VALID: case FINGERPRINT_INDIRECT: if (error) return (error); break; default: /* * Caller wants open to fail unless there is a valid * fingerprint registered. */ MAC_VERIEXEC_DBG(2, "fingerprint status is %d for dev " "%ju, file %ju.%ju\n", status, (uintmax_t)va.va_fsid, (uintmax_t)va.va_fileid, (uintmax_t)va.va_gen); return (EAUTH); } } return (0); } /** * @brief Opening a file has been requested and may need to be validated. * * @param cred credentials to use * @param vp vnode of the file to open * @param label vnode label assigned to the vnode * @param accmode access mode to use for opening the file (read, write, * append, create, verify, etc.) * * @return 0 if opening the file should be allowed, otherwise an error code. */ static int mac_veriexec_vnode_check_open(struct ucred *cred, struct vnode *vp, struct label *label __unused, accmode_t accmode) { int error; /* * Look for the file on the fingerprint lists iff it has not been seen * before. */ if ((mac_veriexec_state & VERIEXEC_STATE_ENFORCE) == 0) return (0); error = mac_veriexec_check_vp(cred, vp, accmode); return (error); } /** * @brief Check mode changes on file to ensure they should be allowed. * * We cannot allow chmod of SUID or SGID on verified files. * * @param cred credentials to use * @param vp vnode of the file to open * @param label vnode label assigned to the vnode * @param mode mode flags to set * * @return 0 if the mode change should be allowed, EAUTH otherwise. */ static int mac_veriexec_vnode_check_setmode(struct ucred *cred, struct vnode *vp, struct label *label __unused, mode_t mode) { int error; if ((mac_veriexec_state & VERIEXEC_STATE_ENFORCE) == 0) return (0); /* * Do not allow chmod (set-[gu]id) of verified file */ error = mac_veriexec_check_vp(cred, vp, VVERIFY); if (error == EAUTH) /* it isn't verified */ return (0); if (error == 0 && (mode & (S_ISUID|S_ISGID)) != 0) return (EAUTH); return (0); } /** * @internal * @brief Initialize the mac_veriexec MAC policy * * @param mpc MAC policy configuration */ static void mac_veriexec_init(struct mac_policy_conf *mpc __unused) { /* Initialize state */ mac_veriexec_state = VERIEXEC_STATE_INACTIVE; /* Initialize meta-data storage */ mac_veriexec_metadata_init(); /* Initialize fingerprint ops */ mac_veriexec_fingerprint_init(); /* Register event handlers */ EVENTHANDLER_REGISTER(vfs_mounted, mac_veriexec_vfs_mounted, NULL, EVENTHANDLER_PRI_FIRST); EVENTHANDLER_REGISTER(vfs_unmounted, mac_veriexec_vfs_unmounted, NULL, EVENTHANDLER_PRI_LAST); } /** * @internal * @brief MAC policy-specific syscall for mac_veriexec * * The following syscalls are implemented: * - @c MAC_VERIEXEC_CHECK_SYSCALL * Check if the file referenced by a file descriptor has a fingerprint * registered in the meta-data store. * * @param td calling thread * @param call system call number * @param arg arugments to the syscall * * @return 0 on success, otherwise an error code. */ static int mac_veriexec_syscall(struct thread *td, int call, void *arg) { struct image_params img; struct nameidata nd; cap_rights_t rights; struct vattr va; struct file *fp; int error; switch (call) { case MAC_VERIEXEC_CHECK_FD_SYSCALL: /* Get the vnode associated with the file descriptor passed */ - error = getvnode(td, (uintptr_t) arg, cap_rights_init(&rights, - CAP_READ), &fp); + error = getvnode(td, (uintptr_t) arg, + cap_rights_init_one(&rights, CAP_READ), &fp); if (error) return (error); if (fp->f_type != DTYPE_VNODE) { MAC_VERIEXEC_DBG(3, "MAC_VERIEXEC_CHECK_SYSCALL: " "file is not vnode type (type=0x%x)", fp->f_type); error = EINVAL; goto cleanup_file; } /* * setup the bits of image_params that are used by * mac_veriexec_check_fingerprint(). */ bzero(&img, sizeof(img)); img.proc = td->td_proc; img.vp = fp->f_vnode; img.attr = &va; /* * Get vnode attributes * (need to obtain a lock on the vnode first) */ vn_lock(img.vp, LK_EXCLUSIVE | LK_RETRY); error = VOP_GETATTR(fp->f_vnode, &va, td->td_ucred); if (error) goto check_done; MAC_VERIEXEC_DBG(2, "mac_veriexec_fingerprint_check_image: " "va_mode=%o, check_files=%d\n", va.va_mode, ((va.va_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)); error = mac_veriexec_fingerprint_check_image(&img, ((va.va_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0), td); check_done: /* Release the lock we obtained earlier */ VOP_UNLOCK(img.vp); cleanup_file: fdrop(fp, td); break; case MAC_VERIEXEC_CHECK_PATH_SYSCALL: /* Look up the path to get the vnode */ NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | LOCKSHARED | AUDITVNODE1, UIO_USERSPACE, arg, td); error = namei(&nd); if (error != 0) break; NDFREE(&nd, NDF_ONLY_PNBUF); /* Check the fingerprint status of the vnode */ error = mac_veriexec_check_vp(td->td_ucred, nd.ni_vp, VVERIFY); vput(nd.ni_vp); break; default: error = EOPNOTSUPP; } return (error); } static struct mac_policy_ops mac_veriexec_ops = { .mpo_init = mac_veriexec_init, .mpo_kld_check_load = mac_veriexec_kld_check_load, .mpo_mount_destroy_label = mac_veriexec_mount_destroy_label, .mpo_mount_init_label = mac_veriexec_mount_init_label, .mpo_priv_check = mac_veriexec_priv_check, .mpo_proc_check_debug = mac_veriexec_proc_check_debug, .mpo_syscall = mac_veriexec_syscall, .mpo_system_check_sysctl = mac_veriexec_sysctl_check, .mpo_vnode_check_exec = mac_veriexec_vnode_check_exec, .mpo_vnode_check_open = mac_veriexec_vnode_check_open, .mpo_vnode_check_setmode = mac_veriexec_vnode_check_setmode, .mpo_vnode_copy_label = mac_veriexec_copy_label, .mpo_vnode_destroy_label = mac_veriexec_vnode_destroy_label, .mpo_vnode_init_label = mac_veriexec_vnode_init_label, }; MAC_POLICY_SET(&mac_veriexec_ops, mac_veriexec, MAC_VERIEXEC_FULLNAME, MPC_LOADTIME_FLAG_NOTLATE, &mac_veriexec_slot); MODULE_VERSION(mac_veriexec, 1); static struct vnode * mac_veriexec_bottom_vnode(struct vnode *vp) { struct vnode *ldvp = NULL; /* * XXX This code is bogus. nullfs is not the only stacking * filesystem. Less bogus code would add a VOP to reach bottom * vnode and would not make assumptions how to get there. */ if (vp->v_mount != NULL && strcmp(vp->v_mount->mnt_vfc->vfc_name, "nullfs") == 0) ldvp = NULLVPTOLOWERVP(vp); return (ldvp); } /** * @brief Get the fingerprint status set on a vnode. * * @param vp vnode to obtain fingerprint status from * * @return Fingerprint status assigned to the vnode. */ fingerprint_status_t mac_veriexec_get_fingerprint_status(struct vnode *vp) { fingerprint_status_t fps; struct vnode *ldvp; fps = SLOT(vp->v_label); switch (fps) { case FINGERPRINT_VALID: case FINGERPRINT_INDIRECT: case FINGERPRINT_FILE: break; default: /* we may need to recurse */ ldvp = mac_veriexec_bottom_vnode(vp); if (ldvp != NULL) return mac_veriexec_get_fingerprint_status(ldvp); break; } return fps; } /** * @brief Get the current verified execution subsystem state. * * @return Current set of verified execution subsystem state flags. */ int mac_veriexec_get_state(void) { return (mac_veriexec_state); } /** * @brief Determine if the verified execution subsystem state has specific * flags set. * * @param state mask of flags to check * * @return State flags set within the masked bits */ int mac_veriexec_in_state(int state) { return (mac_veriexec_state & state); } /** * @brief Set the fingerprint status for a vnode * * Fingerprint status is stored in the MAC per-policy slot assigned to * mac_veriexec. * * @param vp vnode to store the fingerprint status on * @param fp_status fingerprint status to store */ void mac_veriexec_set_fingerprint_status(struct vnode *vp, fingerprint_status_t fp_status) { struct vnode *ldvp; /* recurse until we find the real storage */ ldvp = mac_veriexec_bottom_vnode(vp); if (ldvp != NULL) { mac_veriexec_set_fingerprint_status(ldvp, fp_status); return; } SLOT_SET(vp->v_label, fp_status); } /** * @brief Set verified execution subsystem state flags * * @note Flags can only be added to the current state, not removed. * * @param state state flags to add to the current state */ void mac_veriexec_set_state(int state) { mac_veriexec_state |= state; } /** * @brief Determine if the process is trusted * * @param cred credentials to use * @param p the process in question * * @return 1 if the process is trusted, otherwise 0. */ int mac_veriexec_proc_is_trusted(struct ucred *cred, struct proc *p) { int already_locked, error, flags; /* Make sure we lock the process if we do not already have the lock */ already_locked = PROC_LOCKED(p); if (!already_locked) PROC_LOCK(p); error = mac_veriexec_metadata_get_executable_flags(cred, p, &flags, 0); /* Unlock the process if we locked it previously */ if (!already_locked) PROC_UNLOCK(p); /* Any errors, deny access */ if (error != 0) return (0); /* Check that the trusted flag is set */ return ((flags & VERIEXEC_TRUSTED) == VERIEXEC_TRUSTED); } diff --git a/sys/ufs/ffs/ffs_alloc.c b/sys/ufs/ffs/ffs_alloc.c index be93dd16cd0a..3b652ea14303 100644 --- a/sys/ufs/ffs/ffs_alloc.c +++ b/sys/ufs/ffs/ffs_alloc.c @@ -1,3517 +1,3517 @@ /*- * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND BSD-3-Clause) * * Copyright (c) 2002 Networks Associates Technology, Inc. * All rights reserved. * * This software was developed for the FreeBSD Project by Marshall * Kirk McKusick and Network Associates Laboratories, the Security * Research Division of Network Associates, Inc. under DARPA/SPAWAR * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS * research program * * 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. * * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. 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. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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. * * @(#)ffs_alloc.c 8.18 (Berkeley) 5/26/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_quota.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 #include #include #include #include #include #include #include typedef ufs2_daddr_t allocfcn_t(struct inode *ip, u_int cg, ufs2_daddr_t bpref, int size, int rsize); static ufs2_daddr_t ffs_alloccg(struct inode *, u_int, ufs2_daddr_t, int, int); static ufs2_daddr_t ffs_alloccgblk(struct inode *, struct buf *, ufs2_daddr_t, int); static void ffs_blkfree_cg(struct ufsmount *, struct fs *, struct vnode *, ufs2_daddr_t, long, ino_t, struct workhead *); #ifdef INVARIANTS static int ffs_checkblk(struct inode *, ufs2_daddr_t, long); #endif static ufs2_daddr_t ffs_clusteralloc(struct inode *, u_int, ufs2_daddr_t, int); static ino_t ffs_dirpref(struct inode *); static ufs2_daddr_t ffs_fragextend(struct inode *, u_int, ufs2_daddr_t, int, int); static ufs2_daddr_t ffs_hashalloc (struct inode *, u_int, ufs2_daddr_t, int, int, allocfcn_t *); static ufs2_daddr_t ffs_nodealloccg(struct inode *, u_int, ufs2_daddr_t, int, int); static ufs1_daddr_t ffs_mapsearch(struct fs *, struct cg *, ufs2_daddr_t, int); static int ffs_reallocblks_ufs1(struct vop_reallocblks_args *); static int ffs_reallocblks_ufs2(struct vop_reallocblks_args *); static void ffs_ckhash_cg(struct buf *); /* * Allocate a block in the filesystem. * * The size of the requested block is given, which must be some * multiple of fs_fsize and <= fs_bsize. * A preference may be optionally specified. If a preference is given * the following hierarchy is used to allocate a block: * 1) allocate the requested block. * 2) allocate a rotationally optimal block in the same cylinder. * 3) allocate a block in the same cylinder group. * 4) quadradically rehash into other cylinder groups, until an * available block is located. * If no block preference is given the following hierarchy is used * to allocate a block: * 1) allocate a block in the cylinder group that contains the * inode for the file. * 2) quadradically rehash into other cylinder groups, until an * available block is located. */ int ffs_alloc(ip, lbn, bpref, size, flags, cred, bnp) struct inode *ip; ufs2_daddr_t lbn, bpref; int size, flags; struct ucred *cred; ufs2_daddr_t *bnp; { struct fs *fs; struct ufsmount *ump; ufs2_daddr_t bno; u_int cg, reclaimed; int64_t delta; #ifdef QUOTA int error; #endif *bnp = 0; ump = ITOUMP(ip); fs = ump->um_fs; mtx_assert(UFS_MTX(ump), MA_OWNED); #ifdef INVARIANTS if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { printf("dev = %s, bsize = %ld, size = %d, fs = %s\n", devtoname(ump->um_dev), (long)fs->fs_bsize, size, fs->fs_fsmnt); panic("ffs_alloc: bad size"); } if (cred == NOCRED) panic("ffs_alloc: missing credential"); #endif /* INVARIANTS */ reclaimed = 0; retry: #ifdef QUOTA UFS_UNLOCK(ump); error = chkdq(ip, btodb(size), cred, 0); if (error) return (error); UFS_LOCK(ump); #endif if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0) goto nospace; if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE) && freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0) goto nospace; if (bpref >= fs->fs_size) bpref = 0; if (bpref == 0) cg = ino_to_cg(fs, ip->i_number); else cg = dtog(fs, bpref); bno = ffs_hashalloc(ip, cg, bpref, size, size, ffs_alloccg); if (bno > 0) { delta = btodb(size); DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta); if (flags & IO_EXT) UFS_INODE_SET_FLAG(ip, IN_CHANGE); else UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); *bnp = bno; return (0); } nospace: #ifdef QUOTA UFS_UNLOCK(ump); /* * Restore user's disk quota because allocation failed. */ (void) chkdq(ip, -btodb(size), cred, FORCE); UFS_LOCK(ump); #endif if (reclaimed == 0 && (flags & IO_BUFLOCKED) == 0) { reclaimed = 1; softdep_request_cleanup(fs, ITOV(ip), cred, FLUSH_BLOCKS_WAIT); goto retry; } if (ffs_fsfail_cleanup_locked(ump, 0)) { UFS_UNLOCK(ump); return (ENXIO); } if (reclaimed > 0 && ppsratecheck(&ump->um_last_fullmsg, &ump->um_secs_fullmsg, 1)) { UFS_UNLOCK(ump); ffs_fserr(fs, ip->i_number, "filesystem full"); uprintf("\n%s: write failed, filesystem is full\n", fs->fs_fsmnt); } else { UFS_UNLOCK(ump); } return (ENOSPC); } /* * Reallocate a fragment to a bigger size * * The number and size of the old block is given, and a preference * and new size is also specified. The allocator attempts to extend * the original block. Failing that, the regular block allocator is * invoked to get an appropriate block. */ int ffs_realloccg(ip, lbprev, bprev, bpref, osize, nsize, flags, cred, bpp) struct inode *ip; ufs2_daddr_t lbprev; ufs2_daddr_t bprev; ufs2_daddr_t bpref; int osize, nsize, flags; struct ucred *cred; struct buf **bpp; { struct vnode *vp; struct fs *fs; struct buf *bp; struct ufsmount *ump; u_int cg, request, reclaimed; int error, gbflags; ufs2_daddr_t bno; int64_t delta; vp = ITOV(ip); ump = ITOUMP(ip); fs = ump->um_fs; bp = NULL; gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0; mtx_assert(UFS_MTX(ump), MA_OWNED); #ifdef INVARIANTS if (vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) panic("ffs_realloccg: allocation on suspended filesystem"); if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 || (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) { printf( "dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n", devtoname(ump->um_dev), (long)fs->fs_bsize, osize, nsize, fs->fs_fsmnt); panic("ffs_realloccg: bad size"); } if (cred == NOCRED) panic("ffs_realloccg: missing credential"); #endif /* INVARIANTS */ reclaimed = 0; retry: if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE) && freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0) { goto nospace; } if (bprev == 0) { printf("dev = %s, bsize = %ld, bprev = %jd, fs = %s\n", devtoname(ump->um_dev), (long)fs->fs_bsize, (intmax_t)bprev, fs->fs_fsmnt); panic("ffs_realloccg: bad bprev"); } UFS_UNLOCK(ump); /* * Allocate the extra space in the buffer. */ error = bread_gb(vp, lbprev, osize, NOCRED, gbflags, &bp); if (error) { return (error); } if (bp->b_blkno == bp->b_lblkno) { if (lbprev >= UFS_NDADDR) panic("ffs_realloccg: lbprev out of range"); bp->b_blkno = fsbtodb(fs, bprev); } #ifdef QUOTA error = chkdq(ip, btodb(nsize - osize), cred, 0); if (error) { brelse(bp); return (error); } #endif /* * Check for extension in the existing location. */ *bpp = NULL; cg = dtog(fs, bprev); UFS_LOCK(ump); bno = ffs_fragextend(ip, cg, bprev, osize, nsize); if (bno) { if (bp->b_blkno != fsbtodb(fs, bno)) panic("ffs_realloccg: bad blockno"); delta = btodb(nsize - osize); DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta); if (flags & IO_EXT) UFS_INODE_SET_FLAG(ip, IN_CHANGE); else UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); allocbuf(bp, nsize); bp->b_flags |= B_DONE; vfs_bio_bzero_buf(bp, osize, nsize - osize); if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO) vfs_bio_set_valid(bp, osize, nsize - osize); *bpp = bp; return (0); } /* * Allocate a new disk location. */ if (bpref >= fs->fs_size) bpref = 0; switch ((int)fs->fs_optim) { case FS_OPTSPACE: /* * Allocate an exact sized fragment. Although this makes * best use of space, we will waste time relocating it if * the file continues to grow. If the fragmentation is * less than half of the minimum free reserve, we choose * to begin optimizing for time. */ request = nsize; if (fs->fs_minfree <= 5 || fs->fs_cstotal.cs_nffree > (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100)) break; log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n", fs->fs_fsmnt); fs->fs_optim = FS_OPTTIME; break; case FS_OPTTIME: /* * At this point we have discovered a file that is trying to * grow a small fragment to a larger fragment. To save time, * we allocate a full sized block, then free the unused portion. * If the file continues to grow, the `ffs_fragextend' call * above will be able to grow it in place without further * copying. If aberrant programs cause disk fragmentation to * grow within 2% of the free reserve, we choose to begin * optimizing for space. */ request = fs->fs_bsize; if (fs->fs_cstotal.cs_nffree < (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100) break; log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n", fs->fs_fsmnt); fs->fs_optim = FS_OPTSPACE; break; default: printf("dev = %s, optim = %ld, fs = %s\n", devtoname(ump->um_dev), (long)fs->fs_optim, fs->fs_fsmnt); panic("ffs_realloccg: bad optim"); /* NOTREACHED */ } bno = ffs_hashalloc(ip, cg, bpref, request, nsize, ffs_alloccg); if (bno > 0) { bp->b_blkno = fsbtodb(fs, bno); if (!DOINGSOFTDEP(vp)) /* * The usual case is that a smaller fragment that * was just allocated has been replaced with a bigger * fragment or a full-size block. If it is marked as * B_DELWRI, the current contents have not been written * to disk. It is possible that the block was written * earlier, but very uncommon. If the block has never * been written, there is no need to send a BIO_DELETE * for it when it is freed. The gain from avoiding the * TRIMs for the common case of unwritten blocks far * exceeds the cost of the write amplification for the * uncommon case of failing to send a TRIM for a block * that had been written. */ ffs_blkfree(ump, fs, ump->um_devvp, bprev, (long)osize, ip->i_number, vp->v_type, NULL, (bp->b_flags & B_DELWRI) != 0 ? NOTRIM_KEY : SINGLETON_KEY); delta = btodb(nsize - osize); DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta); if (flags & IO_EXT) UFS_INODE_SET_FLAG(ip, IN_CHANGE); else UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); allocbuf(bp, nsize); bp->b_flags |= B_DONE; vfs_bio_bzero_buf(bp, osize, nsize - osize); if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO) vfs_bio_set_valid(bp, osize, nsize - osize); *bpp = bp; return (0); } #ifdef QUOTA UFS_UNLOCK(ump); /* * Restore user's disk quota because allocation failed. */ (void) chkdq(ip, -btodb(nsize - osize), cred, FORCE); UFS_LOCK(ump); #endif nospace: /* * no space available */ if (reclaimed == 0 && (flags & IO_BUFLOCKED) == 0) { reclaimed = 1; UFS_UNLOCK(ump); if (bp) { brelse(bp); bp = NULL; } UFS_LOCK(ump); softdep_request_cleanup(fs, vp, cred, FLUSH_BLOCKS_WAIT); goto retry; } if (bp) brelse(bp); if (ffs_fsfail_cleanup_locked(ump, 0)) { UFS_UNLOCK(ump); return (ENXIO); } if (reclaimed > 0 && ppsratecheck(&ump->um_last_fullmsg, &ump->um_secs_fullmsg, 1)) { UFS_UNLOCK(ump); ffs_fserr(fs, ip->i_number, "filesystem full"); uprintf("\n%s: write failed, filesystem is full\n", fs->fs_fsmnt); } else { UFS_UNLOCK(ump); } return (ENOSPC); } /* * Reallocate a sequence of blocks into a contiguous sequence of blocks. * * The vnode and an array of buffer pointers for a range of sequential * logical blocks to be made contiguous is given. The allocator attempts * to find a range of sequential blocks starting as close as possible * from the end of the allocation for the logical block immediately * preceding the current range. If successful, the physical block numbers * in the buffer pointers and in the inode are changed to reflect the new * allocation. If unsuccessful, the allocation is left unchanged. The * success in doing the reallocation is returned. Note that the error * return is not reflected back to the user. Rather the previous block * allocation will be used. */ SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "FFS filesystem"); static int doasyncfree = 1; SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "do not force synchronous writes when blocks are reallocated"); static int doreallocblks = 1; SYSCTL_INT(_vfs_ffs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "enable block reallocation"); static int dotrimcons = 1; SYSCTL_INT(_vfs_ffs, OID_AUTO, dotrimcons, CTLFLAG_RWTUN, &dotrimcons, 0, "enable BIO_DELETE / TRIM consolidation"); static int maxclustersearch = 10; SYSCTL_INT(_vfs_ffs, OID_AUTO, maxclustersearch, CTLFLAG_RW, &maxclustersearch, 0, "max number of cylinder group to search for contigous blocks"); #ifdef DIAGNOSTIC static int prtrealloc = 0; SYSCTL_INT(_debug, OID_AUTO, ffs_prtrealloc, CTLFLAG_RW, &prtrealloc, 0, "print out FFS filesystem block reallocation operations"); #endif int ffs_reallocblks(ap) struct vop_reallocblks_args /* { struct vnode *a_vp; struct cluster_save *a_buflist; } */ *ap; { struct ufsmount *ump; /* * We used to skip reallocating the blocks of a file into a * contiguous sequence if the underlying flash device requested * BIO_DELETE notifications, because devices that benefit from * BIO_DELETE also benefit from not moving the data. However, * the destination for the data is usually moved before the data * is written to the initially allocated location, so we rarely * suffer the penalty of extra writes. With the addition of the * consolidation of contiguous blocks into single BIO_DELETE * operations, having fewer but larger contiguous blocks reduces * the number of (slow and expensive) BIO_DELETE operations. So * when doing BIO_DELETE consolidation, we do block reallocation. * * Skip if reallocblks has been disabled globally. */ ump = ap->a_vp->v_mount->mnt_data; if ((((ump->um_flags) & UM_CANDELETE) != 0 && dotrimcons == 0) || doreallocblks == 0) return (ENOSPC); /* * We can't wait in softdep prealloc as it may fsync and recurse * here. Instead we simply fail to reallocate blocks if this * rare condition arises. */ if (DOINGSOFTDEP(ap->a_vp)) if (softdep_prealloc(ap->a_vp, MNT_NOWAIT) != 0) return (ENOSPC); if (ump->um_fstype == UFS1) return (ffs_reallocblks_ufs1(ap)); return (ffs_reallocblks_ufs2(ap)); } static int ffs_reallocblks_ufs1(ap) struct vop_reallocblks_args /* { struct vnode *a_vp; struct cluster_save *a_buflist; } */ *ap; { struct fs *fs; struct inode *ip; struct vnode *vp; struct buf *sbp, *ebp, *bp; ufs1_daddr_t *bap, *sbap, *ebap; struct cluster_save *buflist; struct ufsmount *ump; ufs_lbn_t start_lbn, end_lbn; ufs1_daddr_t soff, newblk, blkno; ufs2_daddr_t pref; struct indir start_ap[UFS_NIADDR + 1], end_ap[UFS_NIADDR + 1], *idp; int i, cg, len, start_lvl, end_lvl, ssize; vp = ap->a_vp; ip = VTOI(vp); ump = ITOUMP(ip); fs = ump->um_fs; /* * If we are not tracking block clusters or if we have less than 4% * free blocks left, then do not attempt to cluster. Running with * less than 5% free block reserve is not recommended and those that * choose to do so do not expect to have good file layout. */ if (fs->fs_contigsumsize <= 0 || freespace(fs, 4) < 0) return (ENOSPC); buflist = ap->a_buflist; len = buflist->bs_nchildren; start_lbn = buflist->bs_children[0]->b_lblkno; end_lbn = start_lbn + len - 1; #ifdef INVARIANTS for (i = 0; i < len; i++) if (!ffs_checkblk(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 1"); for (i = 1; i < len; i++) if (buflist->bs_children[i]->b_lblkno != start_lbn + i) panic("ffs_reallocblks: non-logical cluster"); blkno = buflist->bs_children[0]->b_blkno; ssize = fsbtodb(fs, fs->fs_frag); for (i = 1; i < len - 1; i++) if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize)) panic("ffs_reallocblks: non-physical cluster %d", i); #endif /* * If the cluster crosses the boundary for the first indirect * block, leave space for the indirect block. Indirect blocks * are initially laid out in a position after the last direct * block. Block reallocation would usually destroy locality by * moving the indirect block out of the way to make room for * data blocks if we didn't compensate here. We should also do * this for other indirect block boundaries, but it is only * important for the first one. */ if (start_lbn < UFS_NDADDR && end_lbn >= UFS_NDADDR) return (ENOSPC); /* * If the latest allocation is in a new cylinder group, assume that * the filesystem has decided to move and do not force it back to * the previous cylinder group. */ if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) return (ENOSPC); if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) || ufs_getlbns(vp, end_lbn, end_ap, &end_lvl)) return (ENOSPC); /* * Get the starting offset and block map for the first block. */ if (start_lvl == 0) { sbap = &ip->i_din1->di_db[0]; soff = start_lbn; } else { idp = &start_ap[start_lvl - 1]; if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) { brelse(sbp); return (ENOSPC); } sbap = (ufs1_daddr_t *)sbp->b_data; soff = idp->in_off; } /* * If the block range spans two block maps, get the second map. */ ebap = NULL; if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { ssize = len; } else { #ifdef INVARIANTS if (start_lvl > 0 && start_ap[start_lvl - 1].in_lbn == idp->in_lbn) panic("ffs_reallocblk: start == end"); #endif ssize = len - (idp->in_off + 1); if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp)) goto fail; ebap = (ufs1_daddr_t *)ebp->b_data; } /* * Find the preferred location for the cluster. If we have not * previously failed at this endeavor, then follow our standard * preference calculation. If we have failed at it, then pick up * where we last ended our search. */ UFS_LOCK(ump); if (ip->i_nextclustercg == -1) pref = ffs_blkpref_ufs1(ip, start_lbn, soff, sbap); else pref = cgdata(fs, ip->i_nextclustercg); /* * Search the block map looking for an allocation of the desired size. * To avoid wasting too much time, we limit the number of cylinder * groups that we will search. */ cg = dtog(fs, pref); for (i = min(maxclustersearch, fs->fs_ncg); i > 0; i--) { if ((newblk = ffs_clusteralloc(ip, cg, pref, len)) != 0) break; cg += 1; if (cg >= fs->fs_ncg) cg = 0; } /* * If we have failed in our search, record where we gave up for * next time. Otherwise, fall back to our usual search citerion. */ if (newblk == 0) { ip->i_nextclustercg = cg; UFS_UNLOCK(ump); goto fail; } ip->i_nextclustercg = -1; /* * We have found a new contiguous block. * * First we have to replace the old block pointers with the new * block pointers in the inode and indirect blocks associated * with the file. */ #ifdef DIAGNOSTIC if (prtrealloc) printf("realloc: ino %ju, lbns %jd-%jd\n\told:", (uintmax_t)ip->i_number, (intmax_t)start_lbn, (intmax_t)end_lbn); #endif blkno = newblk; for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) { if (i == ssize) { bap = ebap; soff = -i; } #ifdef INVARIANTS if (!ffs_checkblk(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 2"); if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap) panic("ffs_reallocblks: alloc mismatch"); #endif #ifdef DIAGNOSTIC if (prtrealloc) printf(" %d,", *bap); #endif if (DOINGSOFTDEP(vp)) { if (sbap == &ip->i_din1->di_db[0] && i < ssize) softdep_setup_allocdirect(ip, start_lbn + i, blkno, *bap, fs->fs_bsize, fs->fs_bsize, buflist->bs_children[i]); else softdep_setup_allocindir_page(ip, start_lbn + i, i < ssize ? sbp : ebp, soff + i, blkno, *bap, buflist->bs_children[i]); } *bap++ = blkno; } /* * Next we must write out the modified inode and indirect blocks. * For strict correctness, the writes should be synchronous since * the old block values may have been written to disk. In practise * they are almost never written, but if we are concerned about * strict correctness, the `doasyncfree' flag should be set to zero. * * The test on `doasyncfree' should be changed to test a flag * that shows whether the associated buffers and inodes have * been written. The flag should be set when the cluster is * started and cleared whenever the buffer or inode is flushed. * We can then check below to see if it is set, and do the * synchronous write only when it has been cleared. */ if (sbap != &ip->i_din1->di_db[0]) { if (doasyncfree) bdwrite(sbp); else bwrite(sbp); } else { UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); if (!doasyncfree) ffs_update(vp, 1); } if (ssize < len) { if (doasyncfree) bdwrite(ebp); else bwrite(ebp); } /* * Last, free the old blocks and assign the new blocks to the buffers. */ #ifdef DIAGNOSTIC if (prtrealloc) printf("\n\tnew:"); #endif for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) { bp = buflist->bs_children[i]; if (!DOINGSOFTDEP(vp)) /* * The usual case is that a set of N-contiguous blocks * that was just allocated has been replaced with a * set of N+1-contiguous blocks. If they are marked as * B_DELWRI, the current contents have not been written * to disk. It is possible that the blocks were written * earlier, but very uncommon. If the blocks have never * been written, there is no need to send a BIO_DELETE * for them when they are freed. The gain from avoiding * the TRIMs for the common case of unwritten blocks * far exceeds the cost of the write amplification for * the uncommon case of failing to send a TRIM for the * blocks that had been written. */ ffs_blkfree(ump, fs, ump->um_devvp, dbtofsb(fs, bp->b_blkno), fs->fs_bsize, ip->i_number, vp->v_type, NULL, (bp->b_flags & B_DELWRI) != 0 ? NOTRIM_KEY : SINGLETON_KEY); bp->b_blkno = fsbtodb(fs, blkno); #ifdef INVARIANTS if (!ffs_checkblk(ip, dbtofsb(fs, bp->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 3"); #endif #ifdef DIAGNOSTIC if (prtrealloc) printf(" %d,", blkno); #endif } #ifdef DIAGNOSTIC if (prtrealloc) { prtrealloc--; printf("\n"); } #endif return (0); fail: if (ssize < len) brelse(ebp); if (sbap != &ip->i_din1->di_db[0]) brelse(sbp); return (ENOSPC); } static int ffs_reallocblks_ufs2(ap) struct vop_reallocblks_args /* { struct vnode *a_vp; struct cluster_save *a_buflist; } */ *ap; { struct fs *fs; struct inode *ip; struct vnode *vp; struct buf *sbp, *ebp, *bp; ufs2_daddr_t *bap, *sbap, *ebap; struct cluster_save *buflist; struct ufsmount *ump; ufs_lbn_t start_lbn, end_lbn; ufs2_daddr_t soff, newblk, blkno, pref; struct indir start_ap[UFS_NIADDR + 1], end_ap[UFS_NIADDR + 1], *idp; int i, cg, len, start_lvl, end_lvl, ssize; vp = ap->a_vp; ip = VTOI(vp); ump = ITOUMP(ip); fs = ump->um_fs; /* * If we are not tracking block clusters or if we have less than 4% * free blocks left, then do not attempt to cluster. Running with * less than 5% free block reserve is not recommended and those that * choose to do so do not expect to have good file layout. */ if (fs->fs_contigsumsize <= 0 || freespace(fs, 4) < 0) return (ENOSPC); buflist = ap->a_buflist; len = buflist->bs_nchildren; start_lbn = buflist->bs_children[0]->b_lblkno; end_lbn = start_lbn + len - 1; #ifdef INVARIANTS for (i = 0; i < len; i++) if (!ffs_checkblk(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 1"); for (i = 1; i < len; i++) if (buflist->bs_children[i]->b_lblkno != start_lbn + i) panic("ffs_reallocblks: non-logical cluster"); blkno = buflist->bs_children[0]->b_blkno; ssize = fsbtodb(fs, fs->fs_frag); for (i = 1; i < len - 1; i++) if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize)) panic("ffs_reallocblks: non-physical cluster %d", i); #endif /* * If the cluster crosses the boundary for the first indirect * block, do not move anything in it. Indirect blocks are * usually initially laid out in a position between the data * blocks. Block reallocation would usually destroy locality by * moving the indirect block out of the way to make room for * data blocks if we didn't compensate here. We should also do * this for other indirect block boundaries, but it is only * important for the first one. */ if (start_lbn < UFS_NDADDR && end_lbn >= UFS_NDADDR) return (ENOSPC); /* * If the latest allocation is in a new cylinder group, assume that * the filesystem has decided to move and do not force it back to * the previous cylinder group. */ if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) return (ENOSPC); if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) || ufs_getlbns(vp, end_lbn, end_ap, &end_lvl)) return (ENOSPC); /* * Get the starting offset and block map for the first block. */ if (start_lvl == 0) { sbap = &ip->i_din2->di_db[0]; soff = start_lbn; } else { idp = &start_ap[start_lvl - 1]; if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) { brelse(sbp); return (ENOSPC); } sbap = (ufs2_daddr_t *)sbp->b_data; soff = idp->in_off; } /* * If the block range spans two block maps, get the second map. */ ebap = NULL; if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { ssize = len; } else { #ifdef INVARIANTS if (start_lvl > 0 && start_ap[start_lvl - 1].in_lbn == idp->in_lbn) panic("ffs_reallocblk: start == end"); #endif ssize = len - (idp->in_off + 1); if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp)) goto fail; ebap = (ufs2_daddr_t *)ebp->b_data; } /* * Find the preferred location for the cluster. If we have not * previously failed at this endeavor, then follow our standard * preference calculation. If we have failed at it, then pick up * where we last ended our search. */ UFS_LOCK(ump); if (ip->i_nextclustercg == -1) pref = ffs_blkpref_ufs2(ip, start_lbn, soff, sbap); else pref = cgdata(fs, ip->i_nextclustercg); /* * Search the block map looking for an allocation of the desired size. * To avoid wasting too much time, we limit the number of cylinder * groups that we will search. */ cg = dtog(fs, pref); for (i = min(maxclustersearch, fs->fs_ncg); i > 0; i--) { if ((newblk = ffs_clusteralloc(ip, cg, pref, len)) != 0) break; cg += 1; if (cg >= fs->fs_ncg) cg = 0; } /* * If we have failed in our search, record where we gave up for * next time. Otherwise, fall back to our usual search citerion. */ if (newblk == 0) { ip->i_nextclustercg = cg; UFS_UNLOCK(ump); goto fail; } ip->i_nextclustercg = -1; /* * We have found a new contiguous block. * * First we have to replace the old block pointers with the new * block pointers in the inode and indirect blocks associated * with the file. */ #ifdef DIAGNOSTIC if (prtrealloc) printf("realloc: ino %ju, lbns %jd-%jd\n\told:", (uintmax_t)ip->i_number, (intmax_t)start_lbn, (intmax_t)end_lbn); #endif blkno = newblk; for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) { if (i == ssize) { bap = ebap; soff = -i; } #ifdef INVARIANTS if (!ffs_checkblk(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 2"); if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap) panic("ffs_reallocblks: alloc mismatch"); #endif #ifdef DIAGNOSTIC if (prtrealloc) printf(" %jd,", (intmax_t)*bap); #endif if (DOINGSOFTDEP(vp)) { if (sbap == &ip->i_din2->di_db[0] && i < ssize) softdep_setup_allocdirect(ip, start_lbn + i, blkno, *bap, fs->fs_bsize, fs->fs_bsize, buflist->bs_children[i]); else softdep_setup_allocindir_page(ip, start_lbn + i, i < ssize ? sbp : ebp, soff + i, blkno, *bap, buflist->bs_children[i]); } *bap++ = blkno; } /* * Next we must write out the modified inode and indirect blocks. * For strict correctness, the writes should be synchronous since * the old block values may have been written to disk. In practise * they are almost never written, but if we are concerned about * strict correctness, the `doasyncfree' flag should be set to zero. * * The test on `doasyncfree' should be changed to test a flag * that shows whether the associated buffers and inodes have * been written. The flag should be set when the cluster is * started and cleared whenever the buffer or inode is flushed. * We can then check below to see if it is set, and do the * synchronous write only when it has been cleared. */ if (sbap != &ip->i_din2->di_db[0]) { if (doasyncfree) bdwrite(sbp); else bwrite(sbp); } else { UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); if (!doasyncfree) ffs_update(vp, 1); } if (ssize < len) { if (doasyncfree) bdwrite(ebp); else bwrite(ebp); } /* * Last, free the old blocks and assign the new blocks to the buffers. */ #ifdef DIAGNOSTIC if (prtrealloc) printf("\n\tnew:"); #endif for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) { bp = buflist->bs_children[i]; if (!DOINGSOFTDEP(vp)) /* * The usual case is that a set of N-contiguous blocks * that was just allocated has been replaced with a * set of N+1-contiguous blocks. If they are marked as * B_DELWRI, the current contents have not been written * to disk. It is possible that the blocks were written * earlier, but very uncommon. If the blocks have never * been written, there is no need to send a BIO_DELETE * for them when they are freed. The gain from avoiding * the TRIMs for the common case of unwritten blocks * far exceeds the cost of the write amplification for * the uncommon case of failing to send a TRIM for the * blocks that had been written. */ ffs_blkfree(ump, fs, ump->um_devvp, dbtofsb(fs, bp->b_blkno), fs->fs_bsize, ip->i_number, vp->v_type, NULL, (bp->b_flags & B_DELWRI) != 0 ? NOTRIM_KEY : SINGLETON_KEY); bp->b_blkno = fsbtodb(fs, blkno); #ifdef INVARIANTS if (!ffs_checkblk(ip, dbtofsb(fs, bp->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 3"); #endif #ifdef DIAGNOSTIC if (prtrealloc) printf(" %jd,", (intmax_t)blkno); #endif } #ifdef DIAGNOSTIC if (prtrealloc) { prtrealloc--; printf("\n"); } #endif return (0); fail: if (ssize < len) brelse(ebp); if (sbap != &ip->i_din2->di_db[0]) brelse(sbp); return (ENOSPC); } /* * Allocate an inode in the filesystem. * * If allocating a directory, use ffs_dirpref to select the inode. * If allocating in a directory, the following hierarchy is followed: * 1) allocate the preferred inode. * 2) allocate an inode in the same cylinder group. * 3) quadradically rehash into other cylinder groups, until an * available inode is located. * If no inode preference is given the following hierarchy is used * to allocate an inode: * 1) allocate an inode in cylinder group 0. * 2) quadradically rehash into other cylinder groups, until an * available inode is located. */ int ffs_valloc(pvp, mode, cred, vpp) struct vnode *pvp; int mode; struct ucred *cred; struct vnode **vpp; { struct inode *pip; struct fs *fs; struct inode *ip; struct timespec ts; struct ufsmount *ump; ino_t ino, ipref; u_int cg; int error, reclaimed; *vpp = NULL; pip = VTOI(pvp); ump = ITOUMP(pip); fs = ump->um_fs; UFS_LOCK(ump); reclaimed = 0; retry: if (fs->fs_cstotal.cs_nifree == 0) goto noinodes; if ((mode & IFMT) == IFDIR) ipref = ffs_dirpref(pip); else ipref = pip->i_number; if (ipref >= fs->fs_ncg * fs->fs_ipg) ipref = 0; cg = ino_to_cg(fs, ipref); /* * Track number of dirs created one after another * in a same cg without intervening by files. */ if ((mode & IFMT) == IFDIR) { if (fs->fs_contigdirs[cg] < 255) fs->fs_contigdirs[cg]++; } else { if (fs->fs_contigdirs[cg] > 0) fs->fs_contigdirs[cg]--; } ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode, 0, (allocfcn_t *)ffs_nodealloccg); if (ino == 0) goto noinodes; /* * Get rid of the cached old vnode, force allocation of a new vnode * for this inode. If this fails, release the allocated ino and * return the error. */ if ((error = ffs_vgetf(pvp->v_mount, ino, LK_EXCLUSIVE, vpp, FFSV_FORCEINSMQ | FFSV_REPLACE)) != 0) { ffs_vfree(pvp, ino, mode); return (error); } /* * We got an inode, so check mode and panic if it is already allocated. */ ip = VTOI(*vpp); if (ip->i_mode) { printf("mode = 0%o, inum = %ju, fs = %s\n", ip->i_mode, (uintmax_t)ip->i_number, fs->fs_fsmnt); panic("ffs_valloc: dup alloc"); } if (DIP(ip, i_blocks) && (fs->fs_flags & FS_UNCLEAN) == 0) { /* XXX */ printf("free inode %s/%lu had %ld blocks\n", fs->fs_fsmnt, (u_long)ino, (long)DIP(ip, i_blocks)); DIP_SET(ip, i_blocks, 0); } ip->i_flags = 0; DIP_SET(ip, i_flags, 0); /* * Set up a new generation number for this inode. */ while (ip->i_gen == 0 || ++ip->i_gen == 0) ip->i_gen = arc4random(); DIP_SET(ip, i_gen, ip->i_gen); if (fs->fs_magic == FS_UFS2_MAGIC) { vfs_timestamp(&ts); ip->i_din2->di_birthtime = ts.tv_sec; ip->i_din2->di_birthnsec = ts.tv_nsec; } ip->i_flag = 0; (*vpp)->v_vflag = 0; (*vpp)->v_type = VNON; if (fs->fs_magic == FS_UFS2_MAGIC) { (*vpp)->v_op = &ffs_vnodeops2; UFS_INODE_SET_FLAG(ip, IN_UFS2); } else { (*vpp)->v_op = &ffs_vnodeops1; } return (0); noinodes: if (reclaimed == 0) { reclaimed = 1; softdep_request_cleanup(fs, pvp, cred, FLUSH_INODES_WAIT); goto retry; } if (ffs_fsfail_cleanup_locked(ump, 0)) { UFS_UNLOCK(ump); return (ENXIO); } if (ppsratecheck(&ump->um_last_fullmsg, &ump->um_secs_fullmsg, 1)) { UFS_UNLOCK(ump); ffs_fserr(fs, pip->i_number, "out of inodes"); uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt); } else { UFS_UNLOCK(ump); } return (ENOSPC); } /* * Find a cylinder group to place a directory. * * The policy implemented by this algorithm is to allocate a * directory inode in the same cylinder group as its parent * directory, but also to reserve space for its files inodes * and data. Restrict the number of directories which may be * allocated one after another in the same cylinder group * without intervening allocation of files. * * If we allocate a first level directory then force allocation * in another cylinder group. */ static ino_t ffs_dirpref(pip) struct inode *pip; { struct fs *fs; int cg, prefcg, dirsize, cgsize; u_int avgifree, avgbfree, avgndir, curdirsize; u_int minifree, minbfree, maxndir; u_int mincg, minndir; u_int maxcontigdirs; mtx_assert(UFS_MTX(ITOUMP(pip)), MA_OWNED); fs = ITOFS(pip); avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg; avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg; /* * Force allocation in another cg if creating a first level dir. */ ASSERT_VOP_LOCKED(ITOV(pip), "ffs_dirpref"); if (ITOV(pip)->v_vflag & VV_ROOT) { prefcg = arc4random() % fs->fs_ncg; mincg = prefcg; minndir = fs->fs_ipg; for (cg = prefcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_ndir < minndir && fs->fs_cs(fs, cg).cs_nifree >= avgifree && fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { mincg = cg; minndir = fs->fs_cs(fs, cg).cs_ndir; } for (cg = 0; cg < prefcg; cg++) if (fs->fs_cs(fs, cg).cs_ndir < minndir && fs->fs_cs(fs, cg).cs_nifree >= avgifree && fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { mincg = cg; minndir = fs->fs_cs(fs, cg).cs_ndir; } return ((ino_t)(fs->fs_ipg * mincg)); } /* * Count various limits which used for * optimal allocation of a directory inode. */ maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg); minifree = avgifree - avgifree / 4; if (minifree < 1) minifree = 1; minbfree = avgbfree - avgbfree / 4; if (minbfree < 1) minbfree = 1; cgsize = fs->fs_fsize * fs->fs_fpg; dirsize = fs->fs_avgfilesize * fs->fs_avgfpdir; curdirsize = avgndir ? (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0; if (dirsize < curdirsize) dirsize = curdirsize; if (dirsize <= 0) maxcontigdirs = 0; /* dirsize overflowed */ else maxcontigdirs = min((avgbfree * fs->fs_bsize) / dirsize, 255); if (fs->fs_avgfpdir > 0) maxcontigdirs = min(maxcontigdirs, fs->fs_ipg / fs->fs_avgfpdir); if (maxcontigdirs == 0) maxcontigdirs = 1; /* * Limit number of dirs in one cg and reserve space for * regular files, but only if we have no deficit in * inodes or space. * * We are trying to find a suitable cylinder group nearby * our preferred cylinder group to place a new directory. * We scan from our preferred cylinder group forward looking * for a cylinder group that meets our criterion. If we get * to the final cylinder group and do not find anything, * we start scanning forwards from the beginning of the * filesystem. While it might seem sensible to start scanning * backwards or even to alternate looking forward and backward, * this approach fails badly when the filesystem is nearly full. * Specifically, we first search all the areas that have no space * and finally try the one preceding that. We repeat this on * every request and in the case of the final block end up * searching the entire filesystem. By jumping to the front * of the filesystem, our future forward searches always look * in new cylinder groups so finds every possible block after * one pass over the filesystem. */ prefcg = ino_to_cg(fs, pip->i_number); for (cg = prefcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_ndir < maxndir && fs->fs_cs(fs, cg).cs_nifree >= minifree && fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { if (fs->fs_contigdirs[cg] < maxcontigdirs) return ((ino_t)(fs->fs_ipg * cg)); } for (cg = 0; cg < prefcg; cg++) if (fs->fs_cs(fs, cg).cs_ndir < maxndir && fs->fs_cs(fs, cg).cs_nifree >= minifree && fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { if (fs->fs_contigdirs[cg] < maxcontigdirs) return ((ino_t)(fs->fs_ipg * cg)); } /* * This is a backstop when we have deficit in space. */ for (cg = prefcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) return ((ino_t)(fs->fs_ipg * cg)); for (cg = 0; cg < prefcg; cg++) if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) break; return ((ino_t)(fs->fs_ipg * cg)); } /* * Select the desired position for the next block in a file. The file is * logically divided into sections. The first section is composed of the * direct blocks and the next fs_maxbpg blocks. Each additional section * contains fs_maxbpg blocks. * * If no blocks have been allocated in the first section, the policy is to * request a block in the same cylinder group as the inode that describes * the file. The first indirect is allocated immediately following the last * direct block and the data blocks for the first indirect immediately * follow it. * * If no blocks have been allocated in any other section, the indirect * block(s) are allocated in the same cylinder group as its inode in an * area reserved immediately following the inode blocks. The policy for * the data blocks is to place them in a cylinder group with a greater than * average number of free blocks. An appropriate cylinder group is found * by using a rotor that sweeps the cylinder groups. When a new group of * blocks is needed, the sweep begins in the cylinder group following the * cylinder group from which the previous allocation was made. The sweep * continues until a cylinder group with greater than the average number * of free blocks is found. If the allocation is for the first block in an * indirect block or the previous block is a hole, then the information on * the previous allocation is unavailable; here a best guess is made based * on the logical block number being allocated. * * If a section is already partially allocated, the policy is to * allocate blocks contiguously within the section if possible. */ ufs2_daddr_t ffs_blkpref_ufs1(ip, lbn, indx, bap) struct inode *ip; ufs_lbn_t lbn; int indx; ufs1_daddr_t *bap; { struct fs *fs; u_int cg, inocg; u_int avgbfree, startcg; ufs2_daddr_t pref, prevbn; KASSERT(indx <= 0 || bap != NULL, ("need non-NULL bap")); mtx_assert(UFS_MTX(ITOUMP(ip)), MA_OWNED); fs = ITOFS(ip); /* * Allocation of indirect blocks is indicated by passing negative * values in indx: -1 for single indirect, -2 for double indirect, * -3 for triple indirect. As noted below, we attempt to allocate * the first indirect inline with the file data. For all later * indirect blocks, the data is often allocated in other cylinder * groups. However to speed random file access and to speed up * fsck, the filesystem reserves the first fs_metaspace blocks * (typically half of fs_minfree) of the data area of each cylinder * group to hold these later indirect blocks. */ inocg = ino_to_cg(fs, ip->i_number); if (indx < 0) { /* * Our preference for indirect blocks is the zone at the * beginning of the inode's cylinder group data area that * we try to reserve for indirect blocks. */ pref = cgmeta(fs, inocg); /* * If we are allocating the first indirect block, try to * place it immediately following the last direct block. */ if (indx == -1 && lbn < UFS_NDADDR + NINDIR(fs) && ip->i_din1->di_db[UFS_NDADDR - 1] != 0) pref = ip->i_din1->di_db[UFS_NDADDR - 1] + fs->fs_frag; return (pref); } /* * If we are allocating the first data block in the first indirect * block and the indirect has been allocated in the data block area, * try to place it immediately following the indirect block. */ if (lbn == UFS_NDADDR) { pref = ip->i_din1->di_ib[0]; if (pref != 0 && pref >= cgdata(fs, inocg) && pref < cgbase(fs, inocg + 1)) return (pref + fs->fs_frag); } /* * If we are at the beginning of a file, or we have already allocated * the maximum number of blocks per cylinder group, or we do not * have a block allocated immediately preceding us, then we need * to decide where to start allocating new blocks. */ if (indx == 0) { prevbn = 0; } else { prevbn = bap[indx - 1]; if (UFS_CHECK_BLKNO(ITOVFS(ip), ip->i_number, prevbn, fs->fs_bsize) != 0) prevbn = 0; } if (indx % fs->fs_maxbpg == 0 || prevbn == 0) { /* * If we are allocating a directory data block, we want * to place it in the metadata area. */ if ((ip->i_mode & IFMT) == IFDIR) return (cgmeta(fs, inocg)); /* * Until we fill all the direct and all the first indirect's * blocks, we try to allocate in the data area of the inode's * cylinder group. */ if (lbn < UFS_NDADDR + NINDIR(fs)) return (cgdata(fs, inocg)); /* * Find a cylinder with greater than average number of * unused data blocks. */ if (indx == 0 || prevbn == 0) startcg = inocg + lbn / fs->fs_maxbpg; else startcg = dtog(fs, prevbn) + 1; startcg %= fs->fs_ncg; avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; for (cg = startcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { fs->fs_cgrotor = cg; return (cgdata(fs, cg)); } for (cg = 0; cg <= startcg; cg++) if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { fs->fs_cgrotor = cg; return (cgdata(fs, cg)); } return (0); } /* * Otherwise, we just always try to lay things out contiguously. */ return (prevbn + fs->fs_frag); } /* * Same as above, but for UFS2 */ ufs2_daddr_t ffs_blkpref_ufs2(ip, lbn, indx, bap) struct inode *ip; ufs_lbn_t lbn; int indx; ufs2_daddr_t *bap; { struct fs *fs; u_int cg, inocg; u_int avgbfree, startcg; ufs2_daddr_t pref, prevbn; KASSERT(indx <= 0 || bap != NULL, ("need non-NULL bap")); mtx_assert(UFS_MTX(ITOUMP(ip)), MA_OWNED); fs = ITOFS(ip); /* * Allocation of indirect blocks is indicated by passing negative * values in indx: -1 for single indirect, -2 for double indirect, * -3 for triple indirect. As noted below, we attempt to allocate * the first indirect inline with the file data. For all later * indirect blocks, the data is often allocated in other cylinder * groups. However to speed random file access and to speed up * fsck, the filesystem reserves the first fs_metaspace blocks * (typically half of fs_minfree) of the data area of each cylinder * group to hold these later indirect blocks. */ inocg = ino_to_cg(fs, ip->i_number); if (indx < 0) { /* * Our preference for indirect blocks is the zone at the * beginning of the inode's cylinder group data area that * we try to reserve for indirect blocks. */ pref = cgmeta(fs, inocg); /* * If we are allocating the first indirect block, try to * place it immediately following the last direct block. */ if (indx == -1 && lbn < UFS_NDADDR + NINDIR(fs) && ip->i_din2->di_db[UFS_NDADDR - 1] != 0) pref = ip->i_din2->di_db[UFS_NDADDR - 1] + fs->fs_frag; return (pref); } /* * If we are allocating the first data block in the first indirect * block and the indirect has been allocated in the data block area, * try to place it immediately following the indirect block. */ if (lbn == UFS_NDADDR) { pref = ip->i_din2->di_ib[0]; if (pref != 0 && pref >= cgdata(fs, inocg) && pref < cgbase(fs, inocg + 1)) return (pref + fs->fs_frag); } /* * If we are at the beginning of a file, or we have already allocated * the maximum number of blocks per cylinder group, or we do not * have a block allocated immediately preceding us, then we need * to decide where to start allocating new blocks. */ if (indx == 0) { prevbn = 0; } else { prevbn = bap[indx - 1]; if (UFS_CHECK_BLKNO(ITOVFS(ip), ip->i_number, prevbn, fs->fs_bsize) != 0) prevbn = 0; } if (indx % fs->fs_maxbpg == 0 || prevbn == 0) { /* * If we are allocating a directory data block, we want * to place it in the metadata area. */ if ((ip->i_mode & IFMT) == IFDIR) return (cgmeta(fs, inocg)); /* * Until we fill all the direct and all the first indirect's * blocks, we try to allocate in the data area of the inode's * cylinder group. */ if (lbn < UFS_NDADDR + NINDIR(fs)) return (cgdata(fs, inocg)); /* * Find a cylinder with greater than average number of * unused data blocks. */ if (indx == 0 || prevbn == 0) startcg = inocg + lbn / fs->fs_maxbpg; else startcg = dtog(fs, prevbn) + 1; startcg %= fs->fs_ncg; avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; for (cg = startcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { fs->fs_cgrotor = cg; return (cgdata(fs, cg)); } for (cg = 0; cg <= startcg; cg++) if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { fs->fs_cgrotor = cg; return (cgdata(fs, cg)); } return (0); } /* * Otherwise, we just always try to lay things out contiguously. */ return (prevbn + fs->fs_frag); } /* * Implement the cylinder overflow algorithm. * * The policy implemented by this algorithm is: * 1) allocate the block in its requested cylinder group. * 2) quadradically rehash on the cylinder group number. * 3) brute force search for a free block. * * Must be called with the UFS lock held. Will release the lock on success * and return with it held on failure. */ /*VARARGS5*/ static ufs2_daddr_t ffs_hashalloc(ip, cg, pref, size, rsize, allocator) struct inode *ip; u_int cg; ufs2_daddr_t pref; int size; /* Search size for data blocks, mode for inodes */ int rsize; /* Real allocated size. */ allocfcn_t *allocator; { struct fs *fs; ufs2_daddr_t result; u_int i, icg = cg; mtx_assert(UFS_MTX(ITOUMP(ip)), MA_OWNED); #ifdef INVARIANTS if (ITOV(ip)->v_mount->mnt_kern_flag & MNTK_SUSPENDED) panic("ffs_hashalloc: allocation on suspended filesystem"); #endif fs = ITOFS(ip); /* * 1: preferred cylinder group */ result = (*allocator)(ip, cg, pref, size, rsize); if (result) return (result); /* * 2: quadratic rehash */ for (i = 1; i < fs->fs_ncg; i *= 2) { cg += i; if (cg >= fs->fs_ncg) cg -= fs->fs_ncg; result = (*allocator)(ip, cg, 0, size, rsize); if (result) return (result); } /* * 3: brute force search * Note that we start at i == 2, since 0 was checked initially, * and 1 is always checked in the quadratic rehash. */ cg = (icg + 2) % fs->fs_ncg; for (i = 2; i < fs->fs_ncg; i++) { result = (*allocator)(ip, cg, 0, size, rsize); if (result) return (result); cg++; if (cg == fs->fs_ncg) cg = 0; } return (0); } /* * Determine whether a fragment can be extended. * * Check to see if the necessary fragments are available, and * if they are, allocate them. */ static ufs2_daddr_t ffs_fragextend(ip, cg, bprev, osize, nsize) struct inode *ip; u_int cg; ufs2_daddr_t bprev; int osize, nsize; { struct fs *fs; struct cg *cgp; struct buf *bp; struct ufsmount *ump; int nffree; long bno; int frags, bbase; int i, error; u_int8_t *blksfree; ump = ITOUMP(ip); fs = ump->um_fs; if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize)) return (0); frags = numfrags(fs, nsize); bbase = fragnum(fs, bprev); if (bbase > fragnum(fs, (bprev + frags - 1))) { /* cannot extend across a block boundary */ return (0); } UFS_UNLOCK(ump); if ((error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp)) != 0) goto fail; bno = dtogd(fs, bprev); blksfree = cg_blksfree(cgp); for (i = numfrags(fs, osize); i < frags; i++) if (isclr(blksfree, bno + i)) goto fail; /* * the current fragment can be extended * deduct the count on fragment being extended into * increase the count on the remaining fragment (if any) * allocate the extended piece */ for (i = frags; i < fs->fs_frag - bbase; i++) if (isclr(blksfree, bno + i)) break; cgp->cg_frsum[i - numfrags(fs, osize)]--; if (i != frags) cgp->cg_frsum[i - frags]++; for (i = numfrags(fs, osize), nffree = 0; i < frags; i++) { clrbit(blksfree, bno + i); cgp->cg_cs.cs_nffree--; nffree++; } UFS_LOCK(ump); fs->fs_cstotal.cs_nffree -= nffree; fs->fs_cs(fs, cg).cs_nffree -= nffree; fs->fs_fmod = 1; ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); if (DOINGSOFTDEP(ITOV(ip))) softdep_setup_blkmapdep(bp, UFSTOVFS(ump), bprev, frags, numfrags(fs, osize)); bdwrite(bp); return (bprev); fail: brelse(bp); UFS_LOCK(ump); return (0); } /* * Determine whether a block can be allocated. * * Check to see if a block of the appropriate size is available, * and if it is, allocate it. */ static ufs2_daddr_t ffs_alloccg(ip, cg, bpref, size, rsize) struct inode *ip; u_int cg; ufs2_daddr_t bpref; int size; int rsize; { struct fs *fs; struct cg *cgp; struct buf *bp; struct ufsmount *ump; ufs1_daddr_t bno; ufs2_daddr_t blkno; int i, allocsiz, error, frags; u_int8_t *blksfree; ump = ITOUMP(ip); fs = ump->um_fs; if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize) return (0); UFS_UNLOCK(ump); if ((error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp)) != 0 || (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) goto fail; if (size == fs->fs_bsize) { UFS_LOCK(ump); blkno = ffs_alloccgblk(ip, bp, bpref, rsize); ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); bdwrite(bp); return (blkno); } /* * check to see if any fragments are already available * allocsiz is the size which will be allocated, hacking * it down to a smaller size if necessary */ blksfree = cg_blksfree(cgp); frags = numfrags(fs, size); for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++) if (cgp->cg_frsum[allocsiz] != 0) break; if (allocsiz == fs->fs_frag) { /* * no fragments were available, so a block will be * allocated, and hacked up */ if (cgp->cg_cs.cs_nbfree == 0) goto fail; UFS_LOCK(ump); blkno = ffs_alloccgblk(ip, bp, bpref, rsize); ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); bdwrite(bp); return (blkno); } KASSERT(size == rsize, ("ffs_alloccg: size(%d) != rsize(%d)", size, rsize)); bno = ffs_mapsearch(fs, cgp, bpref, allocsiz); if (bno < 0) goto fail; for (i = 0; i < frags; i++) clrbit(blksfree, bno + i); cgp->cg_cs.cs_nffree -= frags; cgp->cg_frsum[allocsiz]--; if (frags != allocsiz) cgp->cg_frsum[allocsiz - frags]++; UFS_LOCK(ump); fs->fs_cstotal.cs_nffree -= frags; fs->fs_cs(fs, cg).cs_nffree -= frags; fs->fs_fmod = 1; blkno = cgbase(fs, cg) + bno; ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); if (DOINGSOFTDEP(ITOV(ip))) softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno, frags, 0); bdwrite(bp); return (blkno); fail: brelse(bp); UFS_LOCK(ump); return (0); } /* * Allocate a block in a cylinder group. * * This algorithm implements the following policy: * 1) allocate the requested block. * 2) allocate a rotationally optimal block in the same cylinder. * 3) allocate the next available block on the block rotor for the * specified cylinder group. * Note that this routine only allocates fs_bsize blocks; these * blocks may be fragmented by the routine that allocates them. */ static ufs2_daddr_t ffs_alloccgblk(ip, bp, bpref, size) struct inode *ip; struct buf *bp; ufs2_daddr_t bpref; int size; { struct fs *fs; struct cg *cgp; struct ufsmount *ump; ufs1_daddr_t bno; ufs2_daddr_t blkno; u_int8_t *blksfree; int i, cgbpref; ump = ITOUMP(ip); fs = ump->um_fs; mtx_assert(UFS_MTX(ump), MA_OWNED); cgp = (struct cg *)bp->b_data; blksfree = cg_blksfree(cgp); if (bpref == 0) { bpref = cgbase(fs, cgp->cg_cgx) + cgp->cg_rotor + fs->fs_frag; } else if ((cgbpref = dtog(fs, bpref)) != cgp->cg_cgx) { /* map bpref to correct zone in this cg */ if (bpref < cgdata(fs, cgbpref)) bpref = cgmeta(fs, cgp->cg_cgx); else bpref = cgdata(fs, cgp->cg_cgx); } /* * if the requested block is available, use it */ bno = dtogd(fs, blknum(fs, bpref)); if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno))) goto gotit; /* * Take the next available block in this cylinder group. */ bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag); if (bno < 0) return (0); /* Update cg_rotor only if allocated from the data zone */ if (bno >= dtogd(fs, cgdata(fs, cgp->cg_cgx))) cgp->cg_rotor = bno; gotit: blkno = fragstoblks(fs, bno); ffs_clrblock(fs, blksfree, (long)blkno); ffs_clusteracct(fs, cgp, blkno, -1); cgp->cg_cs.cs_nbfree--; fs->fs_cstotal.cs_nbfree--; fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--; fs->fs_fmod = 1; blkno = cgbase(fs, cgp->cg_cgx) + bno; /* * If the caller didn't want the whole block free the frags here. */ size = numfrags(fs, size); if (size != fs->fs_frag) { bno = dtogd(fs, blkno); for (i = size; i < fs->fs_frag; i++) setbit(blksfree, bno + i); i = fs->fs_frag - size; cgp->cg_cs.cs_nffree += i; fs->fs_cstotal.cs_nffree += i; fs->fs_cs(fs, cgp->cg_cgx).cs_nffree += i; fs->fs_fmod = 1; cgp->cg_frsum[i]++; } /* XXX Fixme. */ UFS_UNLOCK(ump); if (DOINGSOFTDEP(ITOV(ip))) softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno, size, 0); UFS_LOCK(ump); return (blkno); } /* * Determine whether a cluster can be allocated. * * We do not currently check for optimal rotational layout if there * are multiple choices in the same cylinder group. Instead we just * take the first one that we find following bpref. */ static ufs2_daddr_t ffs_clusteralloc(ip, cg, bpref, len) struct inode *ip; u_int cg; ufs2_daddr_t bpref; int len; { struct fs *fs; struct cg *cgp; struct buf *bp; struct ufsmount *ump; int i, run, bit, map, got, error; ufs2_daddr_t bno; u_char *mapp; int32_t *lp; u_int8_t *blksfree; ump = ITOUMP(ip); fs = ump->um_fs; if (fs->fs_maxcluster[cg] < len) return (0); UFS_UNLOCK(ump); if ((error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp)) != 0) { UFS_LOCK(ump); return (0); } /* * Check to see if a cluster of the needed size (or bigger) is * available in this cylinder group. */ lp = &cg_clustersum(cgp)[len]; for (i = len; i <= fs->fs_contigsumsize; i++) if (*lp++ > 0) break; if (i > fs->fs_contigsumsize) { /* * This is the first time looking for a cluster in this * cylinder group. Update the cluster summary information * to reflect the true maximum sized cluster so that * future cluster allocation requests can avoid reading * the cylinder group map only to find no clusters. */ lp = &cg_clustersum(cgp)[len - 1]; for (i = len - 1; i > 0; i--) if (*lp-- > 0) break; UFS_LOCK(ump); fs->fs_maxcluster[cg] = i; brelse(bp); return (0); } /* * Search the cluster map to find a big enough cluster. * We take the first one that we find, even if it is larger * than we need as we prefer to get one close to the previous * block allocation. We do not search before the current * preference point as we do not want to allocate a block * that is allocated before the previous one (as we will * then have to wait for another pass of the elevator * algorithm before it will be read). We prefer to fail and * be recalled to try an allocation in the next cylinder group. */ if (dtog(fs, bpref) != cg) bpref = cgdata(fs, cg); else bpref = blknum(fs, bpref); bpref = fragstoblks(fs, dtogd(fs, bpref)); mapp = &cg_clustersfree(cgp)[bpref / NBBY]; map = *mapp++; bit = 1 << (bpref % NBBY); for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) { if ((map & bit) == 0) { run = 0; } else { run++; if (run == len) break; } if ((got & (NBBY - 1)) != (NBBY - 1)) { bit <<= 1; } else { map = *mapp++; bit = 1; } } if (got >= cgp->cg_nclusterblks) { UFS_LOCK(ump); brelse(bp); return (0); } /* * Allocate the cluster that we have found. */ blksfree = cg_blksfree(cgp); for (i = 1; i <= len; i++) if (!ffs_isblock(fs, blksfree, got - run + i)) panic("ffs_clusteralloc: map mismatch"); bno = cgbase(fs, cg) + blkstofrags(fs, got - run + 1); if (dtog(fs, bno) != cg) panic("ffs_clusteralloc: allocated out of group"); len = blkstofrags(fs, len); UFS_LOCK(ump); for (i = 0; i < len; i += fs->fs_frag) if (ffs_alloccgblk(ip, bp, bno + i, fs->fs_bsize) != bno + i) panic("ffs_clusteralloc: lost block"); ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); bdwrite(bp); return (bno); } static inline struct buf * getinobuf(struct inode *ip, u_int cg, u_int32_t cginoblk, int gbflags) { struct fs *fs; fs = ITOFS(ip); return (getblk(ITODEVVP(ip), fsbtodb(fs, ino_to_fsba(fs, cg * fs->fs_ipg + cginoblk)), (int)fs->fs_bsize, 0, 0, gbflags)); } /* * Synchronous inode initialization is needed only when barrier writes do not * work as advertised, and will impose a heavy cost on file creation in a newly * created filesystem. */ static int doasyncinodeinit = 1; SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncinodeinit, CTLFLAG_RWTUN, &doasyncinodeinit, 0, "Perform inode block initialization using asynchronous writes"); /* * Determine whether an inode can be allocated. * * Check to see if an inode is available, and if it is, * allocate it using the following policy: * 1) allocate the requested inode. * 2) allocate the next available inode after the requested * inode in the specified cylinder group. */ static ufs2_daddr_t ffs_nodealloccg(ip, cg, ipref, mode, unused) struct inode *ip; u_int cg; ufs2_daddr_t ipref; int mode; int unused; { struct fs *fs; struct cg *cgp; struct buf *bp, *ibp; struct ufsmount *ump; u_int8_t *inosused, *loc; struct ufs2_dinode *dp2; int error, start, len, i; u_int32_t old_initediblk; ump = ITOUMP(ip); fs = ump->um_fs; check_nifree: if (fs->fs_cs(fs, cg).cs_nifree == 0) return (0); UFS_UNLOCK(ump); if ((error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp)) != 0) { UFS_LOCK(ump); return (0); } restart: if (cgp->cg_cs.cs_nifree == 0) { brelse(bp); UFS_LOCK(ump); return (0); } inosused = cg_inosused(cgp); if (ipref) { ipref %= fs->fs_ipg; if (isclr(inosused, ipref)) goto gotit; } start = cgp->cg_irotor / NBBY; len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY); loc = memcchr(&inosused[start], 0xff, len); if (loc == NULL) { len = start + 1; start = 0; loc = memcchr(&inosused[start], 0xff, len); if (loc == NULL) { printf("cg = %d, irotor = %ld, fs = %s\n", cg, (long)cgp->cg_irotor, fs->fs_fsmnt); panic("ffs_nodealloccg: map corrupted"); /* NOTREACHED */ } } ipref = (loc - inosused) * NBBY + ffs(~*loc) - 1; gotit: /* * Check to see if we need to initialize more inodes. */ if (fs->fs_magic == FS_UFS2_MAGIC && ipref + INOPB(fs) > cgp->cg_initediblk && cgp->cg_initediblk < cgp->cg_niblk) { old_initediblk = cgp->cg_initediblk; /* * Free the cylinder group lock before writing the * initialized inode block. Entering the * babarrierwrite() with the cylinder group lock * causes lock order violation between the lock and * snaplk. * * Another thread can decide to initialize the same * inode block, but whichever thread first gets the * cylinder group lock after writing the newly * allocated inode block will update it and the other * will realize that it has lost and leave the * cylinder group unchanged. */ ibp = getinobuf(ip, cg, old_initediblk, GB_LOCK_NOWAIT); brelse(bp); if (ibp == NULL) { /* * The inode block buffer is already owned by * another thread, which must initialize it. * Wait on the buffer to allow another thread * to finish the updates, with dropped cg * buffer lock, then retry. */ ibp = getinobuf(ip, cg, old_initediblk, 0); brelse(ibp); UFS_LOCK(ump); goto check_nifree; } bzero(ibp->b_data, (int)fs->fs_bsize); dp2 = (struct ufs2_dinode *)(ibp->b_data); for (i = 0; i < INOPB(fs); i++) { while (dp2->di_gen == 0) dp2->di_gen = arc4random(); dp2++; } /* * Rather than adding a soft updates dependency to ensure * that the new inode block is written before it is claimed * by the cylinder group map, we just do a barrier write * here. The barrier write will ensure that the inode block * gets written before the updated cylinder group map can be * written. The barrier write should only slow down bulk * loading of newly created filesystems. */ if (doasyncinodeinit) babarrierwrite(ibp); else bwrite(ibp); /* * After the inode block is written, try to update the * cg initediblk pointer. If another thread beat us * to it, then leave it unchanged as the other thread * has already set it correctly. */ error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp); UFS_LOCK(ump); ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); if (error != 0) return (error); if (cgp->cg_initediblk == old_initediblk) cgp->cg_initediblk += INOPB(fs); goto restart; } cgp->cg_irotor = ipref; UFS_LOCK(ump); ACTIVECLEAR(fs, cg); setbit(inosused, ipref); cgp->cg_cs.cs_nifree--; fs->fs_cstotal.cs_nifree--; fs->fs_cs(fs, cg).cs_nifree--; fs->fs_fmod = 1; if ((mode & IFMT) == IFDIR) { cgp->cg_cs.cs_ndir++; fs->fs_cstotal.cs_ndir++; fs->fs_cs(fs, cg).cs_ndir++; } UFS_UNLOCK(ump); if (DOINGSOFTDEP(ITOV(ip))) softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref, mode); bdwrite(bp); return ((ino_t)(cg * fs->fs_ipg + ipref)); } /* * Free a block or fragment. * * The specified block or fragment is placed back in the * free map. If a fragment is deallocated, a possible * block reassembly is checked. */ static void ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd) struct ufsmount *ump; struct fs *fs; struct vnode *devvp; ufs2_daddr_t bno; long size; ino_t inum; struct workhead *dephd; { struct mount *mp; struct cg *cgp; struct buf *bp; daddr_t dbn; ufs1_daddr_t fragno, cgbno; int i, blk, frags, bbase, error; u_int cg; u_int8_t *blksfree; struct cdev *dev; cg = dtog(fs, bno); if (devvp->v_type == VREG) { /* devvp is a snapshot */ MPASS(devvp->v_mount->mnt_data == ump); dev = ump->um_devvp->v_rdev; } else if (devvp->v_type == VCHR) { /* devvp is a normal disk device */ dev = devvp->v_rdev; ASSERT_VOP_LOCKED(devvp, "ffs_blkfree_cg"); } else return; #ifdef INVARIANTS if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 || fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) { printf("dev=%s, bno = %jd, bsize = %ld, size = %ld, fs = %s\n", devtoname(dev), (intmax_t)bno, (long)fs->fs_bsize, size, fs->fs_fsmnt); panic("ffs_blkfree_cg: bad size"); } #endif if ((u_int)bno >= fs->fs_size) { printf("bad block %jd, ino %lu\n", (intmax_t)bno, (u_long)inum); ffs_fserr(fs, inum, "bad block"); return; } if ((error = ffs_getcg(fs, devvp, cg, GB_CVTENXIO, &bp, &cgp)) != 0) { if (!ffs_fsfail_cleanup(ump, error) || !MOUNTEDSOFTDEP(UFSTOVFS(ump)) || devvp->v_type != VCHR) return; if (devvp->v_type == VREG) dbn = fragstoblks(fs, cgtod(fs, cg)); else dbn = fsbtodb(fs, cgtod(fs, cg)); error = getblkx(devvp, dbn, dbn, fs->fs_cgsize, 0, 0, 0, &bp); KASSERT(error == 0, ("getblkx failed")); softdep_setup_blkfree(UFSTOVFS(ump), bp, bno, numfrags(fs, size), dephd); bp->b_flags |= B_RELBUF | B_NOCACHE; bp->b_flags &= ~B_CACHE; bawrite(bp); return; } cgbno = dtogd(fs, bno); blksfree = cg_blksfree(cgp); UFS_LOCK(ump); if (size == fs->fs_bsize) { fragno = fragstoblks(fs, cgbno); if (!ffs_isfreeblock(fs, blksfree, fragno)) { if (devvp->v_type == VREG) { UFS_UNLOCK(ump); /* devvp is a snapshot */ brelse(bp); return; } printf("dev = %s, block = %jd, fs = %s\n", devtoname(dev), (intmax_t)bno, fs->fs_fsmnt); panic("ffs_blkfree_cg: freeing free block"); } ffs_setblock(fs, blksfree, fragno); ffs_clusteracct(fs, cgp, fragno, 1); cgp->cg_cs.cs_nbfree++; fs->fs_cstotal.cs_nbfree++; fs->fs_cs(fs, cg).cs_nbfree++; } else { bbase = cgbno - fragnum(fs, cgbno); /* * decrement the counts associated with the old frags */ blk = blkmap(fs, blksfree, bbase); ffs_fragacct(fs, blk, cgp->cg_frsum, -1); /* * deallocate the fragment */ frags = numfrags(fs, size); for (i = 0; i < frags; i++) { if (isset(blksfree, cgbno + i)) { printf("dev = %s, block = %jd, fs = %s\n", devtoname(dev), (intmax_t)(bno + i), fs->fs_fsmnt); panic("ffs_blkfree_cg: freeing free frag"); } setbit(blksfree, cgbno + i); } cgp->cg_cs.cs_nffree += i; fs->fs_cstotal.cs_nffree += i; fs->fs_cs(fs, cg).cs_nffree += i; /* * add back in counts associated with the new frags */ blk = blkmap(fs, blksfree, bbase); ffs_fragacct(fs, blk, cgp->cg_frsum, 1); /* * if a complete block has been reassembled, account for it */ fragno = fragstoblks(fs, bbase); if (ffs_isblock(fs, blksfree, fragno)) { cgp->cg_cs.cs_nffree -= fs->fs_frag; fs->fs_cstotal.cs_nffree -= fs->fs_frag; fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag; ffs_clusteracct(fs, cgp, fragno, 1); cgp->cg_cs.cs_nbfree++; fs->fs_cstotal.cs_nbfree++; fs->fs_cs(fs, cg).cs_nbfree++; } } fs->fs_fmod = 1; ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); mp = UFSTOVFS(ump); if (MOUNTEDSOFTDEP(mp) && devvp->v_type == VCHR) softdep_setup_blkfree(UFSTOVFS(ump), bp, bno, numfrags(fs, size), dephd); bdwrite(bp); } /* * Structures and routines associated with trim management. * * The following requests are passed to trim_lookup to indicate * the actions that should be taken. */ #define NEW 1 /* if found, error else allocate and hash it */ #define OLD 2 /* if not found, error, else return it */ #define REPLACE 3 /* if not found, error else unhash and reallocate it */ #define DONE 4 /* if not found, error else unhash and return it */ #define SINGLE 5 /* don't look up, just allocate it and don't hash it */ MALLOC_DEFINE(M_TRIM, "ufs_trim", "UFS trim structures"); #define TRIMLIST_HASH(ump, key) \ (&(ump)->um_trimhash[(key) & (ump)->um_trimlisthashsize]) /* * These structures describe each of the block free requests aggregated * together to make up a trim request. */ struct trim_blkreq { TAILQ_ENTRY(trim_blkreq) blkreqlist; ufs2_daddr_t bno; long size; struct workhead *pdephd; struct workhead dephd; }; /* * Description of a trim request. */ struct ffs_blkfree_trim_params { TAILQ_HEAD(, trim_blkreq) blklist; LIST_ENTRY(ffs_blkfree_trim_params) hashlist; struct task task; struct ufsmount *ump; struct vnode *devvp; ino_t inum; ufs2_daddr_t bno; long size; long key; }; static void ffs_blkfree_trim_completed(struct buf *); static void ffs_blkfree_trim_task(void *ctx, int pending __unused); static struct ffs_blkfree_trim_params *trim_lookup(struct ufsmount *, struct vnode *, ufs2_daddr_t, long, ino_t, u_long, int); static void ffs_blkfree_sendtrim(struct ffs_blkfree_trim_params *); /* * Called on trim completion to start a task to free the associated block(s). */ static void ffs_blkfree_trim_completed(bp) struct buf *bp; { struct ffs_blkfree_trim_params *tp; tp = bp->b_fsprivate1; free(bp, M_TRIM); TASK_INIT(&tp->task, 0, ffs_blkfree_trim_task, tp); taskqueue_enqueue(tp->ump->um_trim_tq, &tp->task); } /* * Trim completion task that free associated block(s). */ static void ffs_blkfree_trim_task(ctx, pending) void *ctx; int pending; { struct ffs_blkfree_trim_params *tp; struct trim_blkreq *blkelm; struct ufsmount *ump; tp = ctx; ump = tp->ump; while ((blkelm = TAILQ_FIRST(&tp->blklist)) != NULL) { ffs_blkfree_cg(ump, ump->um_fs, tp->devvp, blkelm->bno, blkelm->size, tp->inum, blkelm->pdephd); TAILQ_REMOVE(&tp->blklist, blkelm, blkreqlist); free(blkelm, M_TRIM); } vn_finished_secondary_write(UFSTOVFS(ump)); UFS_LOCK(ump); ump->um_trim_inflight -= 1; ump->um_trim_inflight_blks -= numfrags(ump->um_fs, tp->size); UFS_UNLOCK(ump); free(tp, M_TRIM); } /* * Lookup a trim request by inode number. * Allocate if requested (NEW, REPLACE, SINGLE). */ static struct ffs_blkfree_trim_params * trim_lookup(ump, devvp, bno, size, inum, key, alloctype) struct ufsmount *ump; struct vnode *devvp; ufs2_daddr_t bno; long size; ino_t inum; u_long key; int alloctype; { struct trimlist_hashhead *tphashhead; struct ffs_blkfree_trim_params *tp, *ntp; ntp = malloc(sizeof(struct ffs_blkfree_trim_params), M_TRIM, M_WAITOK); if (alloctype != SINGLE) { KASSERT(key >= FIRST_VALID_KEY, ("trim_lookup: invalid key")); UFS_LOCK(ump); tphashhead = TRIMLIST_HASH(ump, key); LIST_FOREACH(tp, tphashhead, hashlist) if (key == tp->key) break; } switch (alloctype) { case NEW: KASSERT(tp == NULL, ("trim_lookup: found trim")); break; case OLD: KASSERT(tp != NULL, ("trim_lookup: missing call to ffs_blkrelease_start()")); UFS_UNLOCK(ump); free(ntp, M_TRIM); return (tp); case REPLACE: KASSERT(tp != NULL, ("trim_lookup: missing REPLACE trim")); LIST_REMOVE(tp, hashlist); /* tp will be freed by caller */ break; case DONE: KASSERT(tp != NULL, ("trim_lookup: missing DONE trim")); LIST_REMOVE(tp, hashlist); UFS_UNLOCK(ump); free(ntp, M_TRIM); return (tp); } TAILQ_INIT(&ntp->blklist); ntp->ump = ump; ntp->devvp = devvp; ntp->bno = bno; ntp->size = size; ntp->inum = inum; ntp->key = key; if (alloctype != SINGLE) { LIST_INSERT_HEAD(tphashhead, ntp, hashlist); UFS_UNLOCK(ump); } return (ntp); } /* * Dispatch a trim request. */ static void ffs_blkfree_sendtrim(tp) struct ffs_blkfree_trim_params *tp; { struct ufsmount *ump; struct mount *mp; struct buf *bp; /* * Postpone the set of the free bit in the cg bitmap until the * BIO_DELETE is completed. Otherwise, due to disk queue * reordering, TRIM might be issued after we reuse the block * and write some new data into it. */ ump = tp->ump; bp = malloc(sizeof(*bp), M_TRIM, M_WAITOK | M_ZERO); bp->b_iocmd = BIO_DELETE; bp->b_iooffset = dbtob(fsbtodb(ump->um_fs, tp->bno)); bp->b_iodone = ffs_blkfree_trim_completed; bp->b_bcount = tp->size; bp->b_fsprivate1 = tp; UFS_LOCK(ump); ump->um_trim_total += 1; ump->um_trim_inflight += 1; ump->um_trim_inflight_blks += numfrags(ump->um_fs, tp->size); ump->um_trim_total_blks += numfrags(ump->um_fs, tp->size); UFS_UNLOCK(ump); mp = UFSTOVFS(ump); vn_start_secondary_write(NULL, &mp, 0); g_vfs_strategy(ump->um_bo, bp); } /* * Allocate a new key to use to identify a range of blocks. */ u_long ffs_blkrelease_start(ump, devvp, inum) struct ufsmount *ump; struct vnode *devvp; ino_t inum; { static u_long masterkey; u_long key; if (((ump->um_flags & UM_CANDELETE) == 0) || dotrimcons == 0) return (SINGLETON_KEY); do { key = atomic_fetchadd_long(&masterkey, 1); } while (key < FIRST_VALID_KEY); (void) trim_lookup(ump, devvp, 0, 0, inum, key, NEW); return (key); } /* * Deallocate a key that has been used to identify a range of blocks. */ void ffs_blkrelease_finish(ump, key) struct ufsmount *ump; u_long key; { struct ffs_blkfree_trim_params *tp; if (((ump->um_flags & UM_CANDELETE) == 0) || dotrimcons == 0) return; /* * If the vfs.ffs.dotrimcons sysctl option is enabled while * a file deletion is active, specifically after a call * to ffs_blkrelease_start() but before the call to * ffs_blkrelease_finish(), ffs_blkrelease_start() will * have handed out SINGLETON_KEY rather than starting a * collection sequence. Thus if we get a SINGLETON_KEY * passed to ffs_blkrelease_finish(), we just return rather * than trying to finish the nonexistent sequence. */ if (key == SINGLETON_KEY) { #ifdef INVARIANTS printf("%s: vfs.ffs.dotrimcons enabled on active filesystem\n", ump->um_mountp->mnt_stat.f_mntonname); #endif return; } /* * We are done with sending blocks using this key. Look up the key * using the DONE alloctype (in tp) to request that it be unhashed * as we will not be adding to it. If the key has never been used, * tp->size will be zero, so we can just free tp. Otherwise the call * to ffs_blkfree_sendtrim(tp) causes the block range described by * tp to be issued (and then tp to be freed). */ tp = trim_lookup(ump, NULL, 0, 0, 0, key, DONE); if (tp->size == 0) free(tp, M_TRIM); else ffs_blkfree_sendtrim(tp); } /* * Setup to free a block or fragment. * * Check for snapshots that might want to claim the block. * If trims are requested, prepare a trim request. Attempt to * aggregate consecutive blocks into a single trim request. */ void ffs_blkfree(ump, fs, devvp, bno, size, inum, vtype, dephd, key) struct ufsmount *ump; struct fs *fs; struct vnode *devvp; ufs2_daddr_t bno; long size; ino_t inum; enum vtype vtype; struct workhead *dephd; u_long key; { struct ffs_blkfree_trim_params *tp, *ntp; struct trim_blkreq *blkelm; /* * Check to see if a snapshot wants to claim the block. * Check that devvp is a normal disk device, not a snapshot, * it has a snapshot(s) associated with it, and one of the * snapshots wants to claim the block. */ if (devvp->v_type == VCHR && (devvp->v_vflag & VV_COPYONWRITE) && ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, dephd)) { return; } /* * Nothing to delay if TRIM is not required for this block or TRIM * is disabled or the operation is performed on a snapshot. */ if (key == NOTRIM_KEY || ((ump->um_flags & UM_CANDELETE) == 0) || devvp->v_type == VREG) { ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd); return; } blkelm = malloc(sizeof(struct trim_blkreq), M_TRIM, M_WAITOK); blkelm->bno = bno; blkelm->size = size; if (dephd == NULL) { blkelm->pdephd = NULL; } else { LIST_INIT(&blkelm->dephd); LIST_SWAP(dephd, &blkelm->dephd, worklist, wk_list); blkelm->pdephd = &blkelm->dephd; } if (key == SINGLETON_KEY) { /* * Just a single non-contiguous piece. Use the SINGLE * alloctype to return a trim request that will not be * hashed for future lookup. */ tp = trim_lookup(ump, devvp, bno, size, inum, key, SINGLE); TAILQ_INSERT_HEAD(&tp->blklist, blkelm, blkreqlist); ffs_blkfree_sendtrim(tp); return; } /* * The callers of this function are not tracking whether or not * the blocks are contiguous. They are just saying that they * are freeing a set of blocks. It is this code that determines * the pieces of that range that are actually contiguous. * * Calling ffs_blkrelease_start() will have created an entry * that we will use. */ tp = trim_lookup(ump, devvp, bno, size, inum, key, OLD); if (tp->size == 0) { /* * First block of a potential range, set block and size * for the trim block. */ tp->bno = bno; tp->size = size; TAILQ_INSERT_HEAD(&tp->blklist, blkelm, blkreqlist); return; } /* * If this block is a continuation of the range (either * follows at the end or preceeds in the front) then we * add it to the front or back of the list and return. * * If it is not a continuation of the trim that we were * building, using the REPLACE alloctype, we request that * the old trim request (still in tp) be unhashed and a * new range started (in ntp). The ffs_blkfree_sendtrim(tp) * call causes the block range described by tp to be issued * (and then tp to be freed). */ if (bno + numfrags(fs, size) == tp->bno) { TAILQ_INSERT_HEAD(&tp->blklist, blkelm, blkreqlist); tp->bno = bno; tp->size += size; return; } else if (bno == tp->bno + numfrags(fs, tp->size)) { TAILQ_INSERT_TAIL(&tp->blklist, blkelm, blkreqlist); tp->size += size; return; } ntp = trim_lookup(ump, devvp, bno, size, inum, key, REPLACE); TAILQ_INSERT_HEAD(&ntp->blklist, blkelm, blkreqlist); ffs_blkfree_sendtrim(tp); } #ifdef INVARIANTS /* * Verify allocation of a block or fragment. Returns true if block or * fragment is allocated, false if it is free. */ static int ffs_checkblk(ip, bno, size) struct inode *ip; ufs2_daddr_t bno; long size; { struct fs *fs; struct cg *cgp; struct buf *bp; ufs1_daddr_t cgbno; int i, error, frags, free; u_int8_t *blksfree; fs = ITOFS(ip); if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { printf("bsize = %ld, size = %ld, fs = %s\n", (long)fs->fs_bsize, size, fs->fs_fsmnt); panic("ffs_checkblk: bad size"); } if ((u_int)bno >= fs->fs_size) panic("ffs_checkblk: bad block %jd", (intmax_t)bno); error = ffs_getcg(fs, ITODEVVP(ip), dtog(fs, bno), 0, &bp, &cgp); if (error) panic("ffs_checkblk: cylinder group read failed"); blksfree = cg_blksfree(cgp); cgbno = dtogd(fs, bno); if (size == fs->fs_bsize) { free = ffs_isblock(fs, blksfree, fragstoblks(fs, cgbno)); } else { frags = numfrags(fs, size); for (free = 0, i = 0; i < frags; i++) if (isset(blksfree, cgbno + i)) free++; if (free != 0 && free != frags) panic("ffs_checkblk: partially free fragment"); } brelse(bp); return (!free); } #endif /* INVARIANTS */ /* * Free an inode. */ int ffs_vfree(pvp, ino, mode) struct vnode *pvp; ino_t ino; int mode; { struct ufsmount *ump; if (DOINGSOFTDEP(pvp)) { softdep_freefile(pvp, ino, mode); return (0); } ump = VFSTOUFS(pvp->v_mount); return (ffs_freefile(ump, ump->um_fs, ump->um_devvp, ino, mode, NULL)); } /* * Do the actual free operation. * The specified inode is placed back in the free map. */ int ffs_freefile(ump, fs, devvp, ino, mode, wkhd) struct ufsmount *ump; struct fs *fs; struct vnode *devvp; ino_t ino; int mode; struct workhead *wkhd; { struct cg *cgp; struct buf *bp; daddr_t dbn; int error; u_int cg; u_int8_t *inosused; struct cdev *dev; ino_t cgino; cg = ino_to_cg(fs, ino); if (devvp->v_type == VREG) { /* devvp is a snapshot */ MPASS(devvp->v_mount->mnt_data == ump); dev = ump->um_devvp->v_rdev; } else if (devvp->v_type == VCHR) { /* devvp is a normal disk device */ dev = devvp->v_rdev; } else { bp = NULL; return (0); } if (ino >= fs->fs_ipg * fs->fs_ncg) panic("ffs_freefile: range: dev = %s, ino = %ju, fs = %s", devtoname(dev), (uintmax_t)ino, fs->fs_fsmnt); if ((error = ffs_getcg(fs, devvp, cg, GB_CVTENXIO, &bp, &cgp)) != 0) { if (!ffs_fsfail_cleanup(ump, error) || !MOUNTEDSOFTDEP(UFSTOVFS(ump)) || devvp->v_type != VCHR) return (error); if (devvp->v_type == VREG) dbn = fragstoblks(fs, cgtod(fs, cg)); else dbn = fsbtodb(fs, cgtod(fs, cg)); error = getblkx(devvp, dbn, dbn, fs->fs_cgsize, 0, 0, 0, &bp); KASSERT(error == 0, ("getblkx failed")); softdep_setup_inofree(UFSTOVFS(ump), bp, ino, wkhd); bp->b_flags |= B_RELBUF | B_NOCACHE; bp->b_flags &= ~B_CACHE; bawrite(bp); return (error); } inosused = cg_inosused(cgp); cgino = ino % fs->fs_ipg; if (isclr(inosused, cgino)) { printf("dev = %s, ino = %ju, fs = %s\n", devtoname(dev), (uintmax_t)ino, fs->fs_fsmnt); if (fs->fs_ronly == 0) panic("ffs_freefile: freeing free inode"); } clrbit(inosused, cgino); if (cgino < cgp->cg_irotor) cgp->cg_irotor = cgino; cgp->cg_cs.cs_nifree++; UFS_LOCK(ump); fs->fs_cstotal.cs_nifree++; fs->fs_cs(fs, cg).cs_nifree++; if ((mode & IFMT) == IFDIR) { cgp->cg_cs.cs_ndir--; fs->fs_cstotal.cs_ndir--; fs->fs_cs(fs, cg).cs_ndir--; } fs->fs_fmod = 1; ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); if (MOUNTEDSOFTDEP(UFSTOVFS(ump)) && devvp->v_type == VCHR) softdep_setup_inofree(UFSTOVFS(ump), bp, ino, wkhd); bdwrite(bp); return (0); } /* * Check to see if a file is free. * Used to check for allocated files in snapshots. */ int ffs_checkfreefile(fs, devvp, ino) struct fs *fs; struct vnode *devvp; ino_t ino; { struct cg *cgp; struct buf *bp; int ret, error; u_int cg; u_int8_t *inosused; cg = ino_to_cg(fs, ino); if ((devvp->v_type != VREG) && (devvp->v_type != VCHR)) return (1); if (ino >= fs->fs_ipg * fs->fs_ncg) return (1); if ((error = ffs_getcg(fs, devvp, cg, 0, &bp, &cgp)) != 0) return (1); inosused = cg_inosused(cgp); ino %= fs->fs_ipg; ret = isclr(inosused, ino); brelse(bp); return (ret); } /* * Find a block of the specified size in the specified cylinder group. * * It is a panic if a request is made to find a block if none are * available. */ static ufs1_daddr_t ffs_mapsearch(fs, cgp, bpref, allocsiz) struct fs *fs; struct cg *cgp; ufs2_daddr_t bpref; int allocsiz; { ufs1_daddr_t bno; int start, len, loc, i; int blk, field, subfield, pos; u_int8_t *blksfree; /* * find the fragment by searching through the free block * map for an appropriate bit pattern */ if (bpref) start = dtogd(fs, bpref) / NBBY; else start = cgp->cg_frotor / NBBY; blksfree = cg_blksfree(cgp); len = howmany(fs->fs_fpg, NBBY) - start; loc = scanc((u_int)len, (u_char *)&blksfree[start], fragtbl[fs->fs_frag], (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); if (loc == 0) { len = start + 1; start = 0; loc = scanc((u_int)len, (u_char *)&blksfree[0], fragtbl[fs->fs_frag], (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); if (loc == 0) { printf("start = %d, len = %d, fs = %s\n", start, len, fs->fs_fsmnt); panic("ffs_alloccg: map corrupted"); /* NOTREACHED */ } } bno = (start + len - loc) * NBBY; cgp->cg_frotor = bno; /* * found the byte in the map * sift through the bits to find the selected frag */ for (i = bno + NBBY; bno < i; bno += fs->fs_frag) { blk = blkmap(fs, blksfree, bno); blk <<= 1; field = around[allocsiz]; subfield = inside[allocsiz]; for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) { if ((blk & field) == subfield) return (bno + pos); field <<= 1; subfield <<= 1; } } printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt); panic("ffs_alloccg: block not in map"); return (-1); } static const struct statfs * ffs_getmntstat(struct vnode *devvp) { if (devvp->v_type == VCHR) return (&devvp->v_rdev->si_mountpt->mnt_stat); return (ffs_getmntstat(VFSTOUFS(devvp->v_mount)->um_devvp)); } /* * Fetch and verify a cylinder group. */ int ffs_getcg(fs, devvp, cg, flags, bpp, cgpp) struct fs *fs; struct vnode *devvp; u_int cg; int flags; struct buf **bpp; struct cg **cgpp; { struct buf *bp; struct cg *cgp; const struct statfs *sfs; daddr_t blkno; int error; *bpp = NULL; *cgpp = NULL; if ((fs->fs_metackhash & CK_CYLGRP) != 0) flags |= GB_CKHASH; if (devvp->v_type == VREG) blkno = fragstoblks(fs, cgtod(fs, cg)); else blkno = fsbtodb(fs, cgtod(fs, cg)); error = breadn_flags(devvp, blkno, blkno, (int)fs->fs_cgsize, NULL, NULL, 0, NOCRED, flags, ffs_ckhash_cg, &bp); if (error != 0) return (error); cgp = (struct cg *)bp->b_data; if ((fs->fs_metackhash & CK_CYLGRP) != 0 && (bp->b_flags & B_CKHASH) != 0 && cgp->cg_ckhash != bp->b_ckhash) { sfs = ffs_getmntstat(devvp); printf("UFS %s%s (%s) cylinder checksum failed: cg %u, cgp: " "0x%x != bp: 0x%jx\n", devvp->v_type == VCHR ? "" : "snapshot of ", sfs->f_mntfromname, sfs->f_mntonname, cg, cgp->cg_ckhash, (uintmax_t)bp->b_ckhash); bp->b_flags &= ~B_CKHASH; bp->b_flags |= B_INVAL | B_NOCACHE; brelse(bp); return (EIO); } if (!cg_chkmagic(cgp) || cgp->cg_cgx != cg) { sfs = ffs_getmntstat(devvp); printf("UFS %s%s (%s)", devvp->v_type == VCHR ? "" : "snapshot of ", sfs->f_mntfromname, sfs->f_mntonname); if (!cg_chkmagic(cgp)) printf(" cg %u: bad magic number 0x%x should be 0x%x\n", cg, cgp->cg_magic, CG_MAGIC); else printf(": wrong cylinder group cg %u != cgx %u\n", cg, cgp->cg_cgx); bp->b_flags &= ~B_CKHASH; bp->b_flags |= B_INVAL | B_NOCACHE; brelse(bp); return (EIO); } bp->b_flags &= ~B_CKHASH; bp->b_xflags |= BX_BKGRDWRITE; /* * If we are using check hashes on the cylinder group then we want * to limit changing the cylinder group time to when we are actually * going to write it to disk so that its check hash remains correct * in memory. If the CK_CYLGRP flag is set the time is updated in * ffs_bufwrite() as the buffer is queued for writing. Otherwise we * update the time here as we have done historically. */ if ((fs->fs_metackhash & CK_CYLGRP) != 0) bp->b_xflags |= BX_CYLGRP; else cgp->cg_old_time = cgp->cg_time = time_second; *bpp = bp; *cgpp = cgp; return (0); } static void ffs_ckhash_cg(bp) struct buf *bp; { uint32_t ckhash; struct cg *cgp; cgp = (struct cg *)bp->b_data; ckhash = cgp->cg_ckhash; cgp->cg_ckhash = 0; bp->b_ckhash = calculate_crc32c(~0L, bp->b_data, bp->b_bcount); cgp->cg_ckhash = ckhash; } /* * Fserr prints the name of a filesystem with an error diagnostic. * * The form of the error message is: * fs: error message */ void ffs_fserr(fs, inum, cp) struct fs *fs; ino_t inum; char *cp; { struct thread *td = curthread; /* XXX */ struct proc *p = td->td_proc; log(LOG_ERR, "pid %d (%s), uid %d inumber %ju on %s: %s\n", p->p_pid, p->p_comm, td->td_ucred->cr_uid, (uintmax_t)inum, fs->fs_fsmnt, cp); } /* * This function provides the capability for the fsck program to * update an active filesystem. Fourteen operations are provided: * * adjrefcnt(inode, amt) - adjusts the reference count on the * specified inode by the specified amount. Under normal * operation the count should always go down. Decrementing * the count to zero will cause the inode to be freed. * adjblkcnt(inode, amt) - adjust the number of blocks used by the * inode by the specified amount. * setsize(inode, size) - set the size of the inode to the * specified size. * adjndir, adjbfree, adjifree, adjffree, adjnumclusters(amt) - * adjust the superblock summary. * freedirs(inode, count) - directory inodes [inode..inode + count - 1] * are marked as free. Inodes should never have to be marked * as in use. * freefiles(inode, count) - file inodes [inode..inode + count - 1] * are marked as free. Inodes should never have to be marked * as in use. * freeblks(blockno, size) - blocks [blockno..blockno + size - 1] * are marked as free. Blocks should never have to be marked * as in use. * setflags(flags, set/clear) - the fs_flags field has the specified * flags set (second parameter +1) or cleared (second parameter -1). * setcwd(dirinode) - set the current directory to dirinode in the * filesystem associated with the snapshot. * setdotdot(oldvalue, newvalue) - Verify that the inode number for ".." * in the current directory is oldvalue then change it to newvalue. * unlink(nameptr, oldvalue) - Verify that the inode number associated * with nameptr in the current directory is oldvalue then unlink it. */ static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vfs_ffs, FFS_ADJ_REFCNT, adjrefcnt, CTLFLAG_WR | CTLTYPE_STRUCT | CTLFLAG_NEEDGIANT, 0, 0, sysctl_ffs_fsck, "S,fsck", "Adjust Inode Reference Count"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_BLKCNT, adjblkcnt, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Adjust Inode Used Blocks Count"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_SIZE, setsize, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Set the inode size"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NDIR, adjndir, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Adjust number of directories"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NBFREE, adjnbfree, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Adjust number of free blocks"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NIFREE, adjnifree, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Adjust number of free inodes"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NFFREE, adjnffree, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Adjust number of free frags"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NUMCLUSTERS, adjnumclusters, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Adjust number of free clusters"); static SYSCTL_NODE(_vfs_ffs, FFS_DIR_FREE, freedirs, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Free Range of Directory Inodes"); static SYSCTL_NODE(_vfs_ffs, FFS_FILE_FREE, freefiles, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Free Range of File Inodes"); static SYSCTL_NODE(_vfs_ffs, FFS_BLK_FREE, freeblks, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Free Range of Blocks"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_FLAGS, setflags, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Change Filesystem Flags"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_CWD, setcwd, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Set Current Working Directory"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_DOTDOT, setdotdot, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Change Value of .. Entry"); static SYSCTL_NODE(_vfs_ffs, FFS_UNLINK, unlink, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Unlink a Duplicate Name"); #ifdef DIAGNOSTIC static int fsckcmds = 0; SYSCTL_INT(_debug, OID_AUTO, ffs_fsckcmds, CTLFLAG_RW, &fsckcmds, 0, "print out fsck_ffs-based filesystem update commands"); #endif /* DIAGNOSTIC */ static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS) { struct thread *td = curthread; struct fsck_cmd cmd; struct ufsmount *ump; struct vnode *vp, *dvp, *fdvp; struct inode *ip, *dp; struct mount *mp; struct fs *fs; struct pwd *pwd; ufs2_daddr_t blkno; long blkcnt, blksize; u_long key; struct file *fp; cap_rights_t rights; int filetype, error; if (req->newlen > sizeof cmd) return (EBADRPC); if ((error = SYSCTL_IN(req, &cmd, sizeof cmd)) != 0) return (error); if (cmd.version != FFS_CMD_VERSION) return (ERPCMISMATCH); if ((error = getvnode(td, cmd.handle, - cap_rights_init(&rights, CAP_FSCK), &fp)) != 0) + cap_rights_init_one(&rights, CAP_FSCK), &fp)) != 0) return (error); vp = fp->f_vnode; if (vp->v_type != VREG && vp->v_type != VDIR) { fdrop(fp, td); return (EINVAL); } vn_start_write(vp, &mp, V_WAIT); if (mp == NULL || strncmp(mp->mnt_stat.f_fstypename, "ufs", MFSNAMELEN)) { vn_finished_write(mp); fdrop(fp, td); return (EINVAL); } ump = VFSTOUFS(mp); if ((mp->mnt_flag & MNT_RDONLY) && ump->um_fsckpid != td->td_proc->p_pid) { vn_finished_write(mp); fdrop(fp, td); return (EROFS); } fs = ump->um_fs; filetype = IFREG; switch (oidp->oid_number) { case FFS_SET_FLAGS: #ifdef DIAGNOSTIC if (fsckcmds) printf("%s: %s flags\n", mp->mnt_stat.f_mntonname, cmd.size > 0 ? "set" : "clear"); #endif /* DIAGNOSTIC */ if (cmd.size > 0) fs->fs_flags |= (long)cmd.value; else fs->fs_flags &= ~(long)cmd.value; break; case FFS_ADJ_REFCNT: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: adjust inode %jd link count by %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, (intmax_t)cmd.size); } #endif /* DIAGNOSTIC */ if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) break; ip = VTOI(vp); ip->i_nlink += cmd.size; DIP_SET(ip, i_nlink, ip->i_nlink); ip->i_effnlink += cmd.size; UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_MODIFIED); error = ffs_update(vp, 1); if (DOINGSOFTDEP(vp)) softdep_change_linkcnt(ip); vput(vp); break; case FFS_ADJ_BLKCNT: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: adjust inode %jd block count by %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, (intmax_t)cmd.size); } #endif /* DIAGNOSTIC */ if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) break; ip = VTOI(vp); DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + cmd.size); UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_MODIFIED); error = ffs_update(vp, 1); vput(vp); break; case FFS_SET_SIZE: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: set inode %jd size to %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, (intmax_t)cmd.size); } #endif /* DIAGNOSTIC */ if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) break; ip = VTOI(vp); DIP_SET(ip, i_size, cmd.size); UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_MODIFIED); error = ffs_update(vp, 1); vput(vp); break; case FFS_DIR_FREE: filetype = IFDIR; /* fall through */ case FFS_FILE_FREE: #ifdef DIAGNOSTIC if (fsckcmds) { if (cmd.size == 1) printf("%s: free %s inode %ju\n", mp->mnt_stat.f_mntonname, filetype == IFDIR ? "directory" : "file", (uintmax_t)cmd.value); else printf("%s: free %s inodes %ju-%ju\n", mp->mnt_stat.f_mntonname, filetype == IFDIR ? "directory" : "file", (uintmax_t)cmd.value, (uintmax_t)(cmd.value + cmd.size - 1)); } #endif /* DIAGNOSTIC */ while (cmd.size > 0) { if ((error = ffs_freefile(ump, fs, ump->um_devvp, cmd.value, filetype, NULL))) break; cmd.size -= 1; cmd.value += 1; } break; case FFS_BLK_FREE: #ifdef DIAGNOSTIC if (fsckcmds) { if (cmd.size == 1) printf("%s: free block %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); else printf("%s: free blocks %jd-%jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, (intmax_t)cmd.value + cmd.size - 1); } #endif /* DIAGNOSTIC */ blkno = cmd.value; blkcnt = cmd.size; blksize = fs->fs_frag - (blkno % fs->fs_frag); key = ffs_blkrelease_start(ump, ump->um_devvp, UFS_ROOTINO); while (blkcnt > 0) { if (blkcnt < blksize) blksize = blkcnt; ffs_blkfree(ump, fs, ump->um_devvp, blkno, blksize * fs->fs_fsize, UFS_ROOTINO, VDIR, NULL, key); blkno += blksize; blkcnt -= blksize; blksize = fs->fs_frag; } ffs_blkrelease_finish(ump, key); break; /* * Adjust superblock summaries. fsck(8) is expected to * submit deltas when necessary. */ case FFS_ADJ_NDIR: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: adjust number of directories by %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DIAGNOSTIC */ fs->fs_cstotal.cs_ndir += cmd.value; break; case FFS_ADJ_NBFREE: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: adjust number of free blocks by %+jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DIAGNOSTIC */ fs->fs_cstotal.cs_nbfree += cmd.value; break; case FFS_ADJ_NIFREE: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: adjust number of free inodes by %+jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DIAGNOSTIC */ fs->fs_cstotal.cs_nifree += cmd.value; break; case FFS_ADJ_NFFREE: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: adjust number of free frags by %+jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DIAGNOSTIC */ fs->fs_cstotal.cs_nffree += cmd.value; break; case FFS_ADJ_NUMCLUSTERS: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: adjust number of free clusters by %+jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DIAGNOSTIC */ fs->fs_cstotal.cs_numclusters += cmd.value; break; case FFS_SET_CWD: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: set current directory to inode %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DIAGNOSTIC */ if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_SHARED, &vp))) break; AUDIT_ARG_VNODE1(vp); if ((error = change_dir(vp, td)) != 0) { vput(vp); break; } VOP_UNLOCK(vp); pwd_chdir(td, vp); break; case FFS_SET_DOTDOT: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: change .. in cwd from %jd to %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, (intmax_t)cmd.size); } #endif /* DIAGNOSTIC */ /* * First we have to get and lock the parent directory * to which ".." points. */ error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &fdvp); if (error) break; /* * Now we get and lock the child directory containing "..". */ pwd = pwd_hold(td); dvp = pwd->pwd_cdir; if ((error = vget(dvp, LK_EXCLUSIVE)) != 0) { vput(fdvp); pwd_drop(pwd); break; } dp = VTOI(dvp); SET_I_OFFSET(dp, 12); /* XXX mastertemplate.dot_reclen */ error = ufs_dirrewrite(dp, VTOI(fdvp), (ino_t)cmd.size, DT_DIR, 0); cache_purge(fdvp); cache_purge(dvp); vput(dvp); vput(fdvp); pwd_drop(pwd); break; case FFS_UNLINK: #ifdef DIAGNOSTIC if (fsckcmds) { char buf[32]; if (copyinstr((char *)(intptr_t)cmd.value, buf,32,NULL)) strncpy(buf, "Name_too_long", 32); printf("%s: unlink %s (inode %jd)\n", mp->mnt_stat.f_mntonname, buf, (intmax_t)cmd.size); } #endif /* DIAGNOSTIC */ /* * kern_funlinkat will do its own start/finish writes and * they do not nest, so drop ours here. Setting mp == NULL * indicates that vn_finished_write is not needed down below. */ vn_finished_write(mp); mp = NULL; error = kern_funlinkat(td, AT_FDCWD, (char *)(intptr_t)cmd.value, FD_NONE, UIO_USERSPACE, 0, (ino_t)cmd.size); break; default: #ifdef DIAGNOSTIC if (fsckcmds) { printf("Invalid request %d from fsck\n", oidp->oid_number); } #endif /* DIAGNOSTIC */ error = EINVAL; break; } fdrop(fp, td); vn_finished_write(mp); return (error); }