diff --git a/sys/amd64/amd64/exec_machdep.c b/sys/amd64/amd64/exec_machdep.c index 492aab177b9b..5c5132278cd0 100644 --- a/sys/amd64/amd64/exec_machdep.c +++ b/sys/amd64/amd64/exec_machdep.c @@ -1,978 +1,977 @@ /*- * SPDX-License-Identifier: BSD-4-Clause * * Copyright (c) 2003 Peter Wemm. * Copyright (c) 1992 Terrence R. Lambert. * Copyright (c) 1982, 1987, 1990 The Regents of the University of California. * All rights reserved. * * This code is derived from software contributed to Berkeley by * William Jolitz. * * 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. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. 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: @(#)machdep.c 7.4 (Berkeley) 6/3/91 */ #include __FBSDID("$FreeBSD$"); #include "opt_cpu.h" #include "opt_ddb.h" #include "opt_kstack_pages.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef SMP #include #endif #include #include #include #include #include #include #include #include #include #include #ifdef DDB #ifndef KDB #error KDB must be enabled in order for DDB to work! #endif #include #include #endif #include #include #include #include #include #include #include #include #include static void get_fpcontext(struct thread *td, mcontext_t *mcp, char **xfpusave, size_t *xfpusave_len); static int set_fpcontext(struct thread *td, mcontext_t *mcp, char *xfpustate, size_t xfpustate_len); /* * Send an interrupt to process. * * Stack is set up to allow sigcode stored at top to call routine, * followed by call to sigreturn routine below. After sigreturn * resets the signal mask, the stack, and the frame pointer, it * returns to the user specified pc, psl. */ void sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct sigframe sf, *sfp; struct pcb *pcb; struct proc *p; struct thread *td; struct sigacts *psp; char *sp; struct trapframe *regs; char *xfpusave; size_t xfpusave_len; int sig; int oonstack; td = curthread; pcb = td->td_pcb; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); sig = ksi->ksi_signo; psp = p->p_sigacts; mtx_assert(&psp->ps_mtx, MA_OWNED); regs = td->td_frame; oonstack = sigonstack(regs->tf_rsp); /* Save user context. */ bzero(&sf, sizeof(sf)); sf.sf_uc.uc_sigmask = *mask; sf.sf_uc.uc_stack = td->td_sigstk; sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE; sf.sf_uc.uc_mcontext.mc_onstack = (oonstack) ? 1 : 0; bcopy(regs, &sf.sf_uc.uc_mcontext.mc_rdi, sizeof(*regs)); sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext); /* magic */ get_fpcontext(td, &sf.sf_uc.uc_mcontext, &xfpusave, &xfpusave_len); fpstate_drop(td); update_pcb_bases(pcb); sf.sf_uc.uc_mcontext.mc_fsbase = pcb->pcb_fsbase; sf.sf_uc.uc_mcontext.mc_gsbase = pcb->pcb_gsbase; bzero(sf.sf_uc.uc_mcontext.mc_spare, sizeof(sf.sf_uc.uc_mcontext.mc_spare)); /* Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { sp = (char *)td->td_sigstk.ss_sp + td->td_sigstk.ss_size; #if defined(COMPAT_43) td->td_sigstk.ss_flags |= SS_ONSTACK; #endif } else sp = (char *)regs->tf_rsp - 128; if (xfpusave != NULL) { sp -= xfpusave_len; sp = (char *)((unsigned long)sp & ~0x3Ful); sf.sf_uc.uc_mcontext.mc_xfpustate = (register_t)sp; } sp -= sizeof(struct sigframe); /* Align to 16 bytes. */ sfp = (struct sigframe *)((unsigned long)sp & ~0xFul); /* Build the argument list for the signal handler. */ regs->tf_rdi = sig; /* arg 1 in %rdi */ regs->tf_rdx = (register_t)&sfp->sf_uc; /* arg 3 in %rdx */ bzero(&sf.sf_si, sizeof(sf.sf_si)); if (SIGISMEMBER(psp->ps_siginfo, sig)) { /* Signal handler installed with SA_SIGINFO. */ regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher; /* Fill in POSIX parts */ sf.sf_si = ksi->ksi_info; sf.sf_si.si_signo = sig; /* maybe a translated signal */ regs->tf_rcx = (register_t)ksi->ksi_addr; /* arg 4 in %rcx */ } else { /* Old FreeBSD-style arguments. */ regs->tf_rsi = ksi->ksi_code; /* arg 2 in %rsi */ regs->tf_rcx = (register_t)ksi->ksi_addr; /* arg 4 in %rcx */ sf.sf_ahu.sf_handler = catcher; } mtx_unlock(&psp->ps_mtx); PROC_UNLOCK(p); /* * Copy the sigframe out to the user's stack. */ if (copyout(&sf, sfp, sizeof(*sfp)) != 0 || (xfpusave != NULL && copyout(xfpusave, (void *)sf.sf_uc.uc_mcontext.mc_xfpustate, xfpusave_len) != 0)) { -#ifdef DEBUG - printf("process %ld has trashed its stack\n", (long)p->p_pid); -#endif + uprintf("pid %d comm %s has trashed its stack, killing\n", + p->p_pid, p->p_comm); PROC_LOCK(p); sigexit(td, SIGILL); } regs->tf_rsp = (long)sfp; regs->tf_rip = p->p_sysent->sv_sigcode_base; regs->tf_rflags &= ~(PSL_T | PSL_D); regs->tf_cs = _ucodesel; regs->tf_ds = _udatasel; regs->tf_ss = _udatasel; regs->tf_es = _udatasel; regs->tf_fs = _ufssel; regs->tf_gs = _ugssel; regs->tf_flags = TF_HASSEGS; PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } /* * System call to cleanup state after a signal * has been taken. Reset signal mask and * stack state from context left by sendsig (above). * Return to previous pc and psl as specified by * context left by sendsig. Check carefully to * make sure that the user has not modified the * state to gain improper privileges. */ int sys_sigreturn(struct thread *td, struct sigreturn_args *uap) { ucontext_t uc; struct pcb *pcb; struct proc *p; struct trapframe *regs; ucontext_t *ucp; char *xfpustate; size_t xfpustate_len; long rflags; int cs, error, ret; ksiginfo_t ksi; pcb = td->td_pcb; p = td->td_proc; error = copyin(uap->sigcntxp, &uc, sizeof(uc)); if (error != 0) { uprintf("pid %d (%s): sigreturn copyin failed\n", p->p_pid, td->td_name); return (error); } ucp = &uc; if ((ucp->uc_mcontext.mc_flags & ~_MC_FLAG_MASK) != 0) { uprintf("pid %d (%s): sigreturn mc_flags %x\n", p->p_pid, td->td_name, ucp->uc_mcontext.mc_flags); return (EINVAL); } regs = td->td_frame; rflags = ucp->uc_mcontext.mc_rflags; /* * Don't allow users to change privileged or reserved flags. */ if (!EFL_SECURE(rflags, regs->tf_rflags)) { uprintf("pid %d (%s): sigreturn rflags = 0x%lx\n", p->p_pid, td->td_name, rflags); return (EINVAL); } /* * Don't allow users to load a valid privileged %cs. Let the * hardware check for invalid selectors, excess privilege in * other selectors, invalid %eip's and invalid %esp's. */ cs = ucp->uc_mcontext.mc_cs; if (!CS_SECURE(cs)) { uprintf("pid %d (%s): sigreturn cs = 0x%x\n", p->p_pid, td->td_name, cs); ksiginfo_init_trap(&ksi); ksi.ksi_signo = SIGBUS; ksi.ksi_code = BUS_OBJERR; ksi.ksi_trapno = T_PROTFLT; ksi.ksi_addr = (void *)regs->tf_rip; trapsignal(td, &ksi); return (EINVAL); } if ((uc.uc_mcontext.mc_flags & _MC_HASFPXSTATE) != 0) { xfpustate_len = uc.uc_mcontext.mc_xfpustate_len; if (xfpustate_len > cpu_max_ext_state_size - sizeof(struct savefpu)) { uprintf("pid %d (%s): sigreturn xfpusave_len = 0x%zx\n", p->p_pid, td->td_name, xfpustate_len); return (EINVAL); } xfpustate = (char *)fpu_save_area_alloc(); error = copyin((const void *)uc.uc_mcontext.mc_xfpustate, xfpustate, xfpustate_len); if (error != 0) { fpu_save_area_free((struct savefpu *)xfpustate); uprintf( "pid %d (%s): sigreturn copying xfpustate failed\n", p->p_pid, td->td_name); return (error); } } else { xfpustate = NULL; xfpustate_len = 0; } ret = set_fpcontext(td, &ucp->uc_mcontext, xfpustate, xfpustate_len); fpu_save_area_free((struct savefpu *)xfpustate); if (ret != 0) { uprintf("pid %d (%s): sigreturn set_fpcontext err %d\n", p->p_pid, td->td_name, ret); return (ret); } bcopy(&ucp->uc_mcontext.mc_rdi, regs, sizeof(*regs)); update_pcb_bases(pcb); pcb->pcb_fsbase = ucp->uc_mcontext.mc_fsbase; pcb->pcb_gsbase = ucp->uc_mcontext.mc_gsbase; #if defined(COMPAT_43) if (ucp->uc_mcontext.mc_onstack & 1) td->td_sigstk.ss_flags |= SS_ONSTACK; else td->td_sigstk.ss_flags &= ~SS_ONSTACK; #endif kern_sigprocmask(td, SIG_SETMASK, &ucp->uc_sigmask, NULL, 0); return (EJUSTRETURN); } #ifdef COMPAT_FREEBSD4 int freebsd4_sigreturn(struct thread *td, struct freebsd4_sigreturn_args *uap) { return sys_sigreturn(td, (struct sigreturn_args *)uap); } #endif /* * Reset the hardware debug registers if they were in use. * They won't have any meaning for the newly exec'd process. */ void x86_clear_dbregs(struct pcb *pcb) { if ((pcb->pcb_flags & PCB_DBREGS) == 0) return; pcb->pcb_dr0 = 0; pcb->pcb_dr1 = 0; pcb->pcb_dr2 = 0; pcb->pcb_dr3 = 0; pcb->pcb_dr6 = 0; pcb->pcb_dr7 = 0; if (pcb == curpcb) { /* * Clear the debug registers on the running CPU, * otherwise they will end up affecting the next * process we switch to. */ reset_dbregs(); } clear_pcb_flags(pcb, PCB_DBREGS); } /* * Reset registers to default values on exec. */ void exec_setregs(struct thread *td, struct image_params *imgp, uintptr_t stack) { struct trapframe *regs; struct pcb *pcb; register_t saved_rflags; regs = td->td_frame; pcb = td->td_pcb; if (td->td_proc->p_md.md_ldt != NULL) user_ldt_free(td); update_pcb_bases(pcb); pcb->pcb_fsbase = 0; pcb->pcb_gsbase = 0; clear_pcb_flags(pcb, PCB_32BIT); pcb->pcb_initial_fpucw = __INITIAL_FPUCW__; saved_rflags = regs->tf_rflags & PSL_T; bzero((char *)regs, sizeof(struct trapframe)); regs->tf_rip = imgp->entry_addr; regs->tf_rsp = ((stack - 8) & ~0xFul) + 8; regs->tf_rdi = stack; /* argv */ regs->tf_rflags = PSL_USER | saved_rflags; regs->tf_ss = _udatasel; regs->tf_cs = _ucodesel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; regs->tf_fs = _ufssel; regs->tf_gs = _ugssel; regs->tf_flags = TF_HASSEGS; x86_clear_dbregs(pcb); /* * Drop the FP state if we hold it, so that the process gets a * clean FP state if it uses the FPU again. */ fpstate_drop(td); } int fill_regs(struct thread *td, struct reg *regs) { struct trapframe *tp; tp = td->td_frame; return (fill_frame_regs(tp, regs)); } int fill_frame_regs(struct trapframe *tp, struct reg *regs) { regs->r_r15 = tp->tf_r15; regs->r_r14 = tp->tf_r14; regs->r_r13 = tp->tf_r13; regs->r_r12 = tp->tf_r12; regs->r_r11 = tp->tf_r11; regs->r_r10 = tp->tf_r10; regs->r_r9 = tp->tf_r9; regs->r_r8 = tp->tf_r8; regs->r_rdi = tp->tf_rdi; regs->r_rsi = tp->tf_rsi; regs->r_rbp = tp->tf_rbp; regs->r_rbx = tp->tf_rbx; regs->r_rdx = tp->tf_rdx; regs->r_rcx = tp->tf_rcx; regs->r_rax = tp->tf_rax; regs->r_rip = tp->tf_rip; regs->r_cs = tp->tf_cs; regs->r_rflags = tp->tf_rflags; regs->r_rsp = tp->tf_rsp; regs->r_ss = tp->tf_ss; if (tp->tf_flags & TF_HASSEGS) { regs->r_ds = tp->tf_ds; regs->r_es = tp->tf_es; regs->r_fs = tp->tf_fs; regs->r_gs = tp->tf_gs; } else { regs->r_ds = 0; regs->r_es = 0; regs->r_fs = 0; regs->r_gs = 0; } regs->r_err = 0; regs->r_trapno = 0; return (0); } int set_regs(struct thread *td, struct reg *regs) { struct trapframe *tp; register_t rflags; tp = td->td_frame; rflags = regs->r_rflags & 0xffffffff; if (!EFL_SECURE(rflags, tp->tf_rflags) || !CS_SECURE(regs->r_cs)) return (EINVAL); tp->tf_r15 = regs->r_r15; tp->tf_r14 = regs->r_r14; tp->tf_r13 = regs->r_r13; tp->tf_r12 = regs->r_r12; tp->tf_r11 = regs->r_r11; tp->tf_r10 = regs->r_r10; tp->tf_r9 = regs->r_r9; tp->tf_r8 = regs->r_r8; tp->tf_rdi = regs->r_rdi; tp->tf_rsi = regs->r_rsi; tp->tf_rbp = regs->r_rbp; tp->tf_rbx = regs->r_rbx; tp->tf_rdx = regs->r_rdx; tp->tf_rcx = regs->r_rcx; tp->tf_rax = regs->r_rax; tp->tf_rip = regs->r_rip; tp->tf_cs = regs->r_cs; tp->tf_rflags = rflags; tp->tf_rsp = regs->r_rsp; tp->tf_ss = regs->r_ss; if (0) { /* XXXKIB */ tp->tf_ds = regs->r_ds; tp->tf_es = regs->r_es; tp->tf_fs = regs->r_fs; tp->tf_gs = regs->r_gs; tp->tf_flags = TF_HASSEGS; } set_pcb_flags(td->td_pcb, PCB_FULL_IRET); return (0); } /* XXX check all this stuff! */ /* externalize from sv_xmm */ static void fill_fpregs_xmm(struct savefpu *sv_xmm, struct fpreg *fpregs) { struct envxmm *penv_fpreg = (struct envxmm *)&fpregs->fpr_env; struct envxmm *penv_xmm = &sv_xmm->sv_env; int i; /* pcb -> fpregs */ bzero(fpregs, sizeof(*fpregs)); /* FPU control/status */ penv_fpreg->en_cw = penv_xmm->en_cw; penv_fpreg->en_sw = penv_xmm->en_sw; penv_fpreg->en_tw = penv_xmm->en_tw; penv_fpreg->en_opcode = penv_xmm->en_opcode; penv_fpreg->en_rip = penv_xmm->en_rip; penv_fpreg->en_rdp = penv_xmm->en_rdp; penv_fpreg->en_mxcsr = penv_xmm->en_mxcsr; penv_fpreg->en_mxcsr_mask = penv_xmm->en_mxcsr_mask; /* FPU registers */ for (i = 0; i < 8; ++i) bcopy(sv_xmm->sv_fp[i].fp_acc.fp_bytes, fpregs->fpr_acc[i], 10); /* SSE registers */ for (i = 0; i < 16; ++i) bcopy(sv_xmm->sv_xmm[i].xmm_bytes, fpregs->fpr_xacc[i], 16); } /* internalize from fpregs into sv_xmm */ static void set_fpregs_xmm(struct fpreg *fpregs, struct savefpu *sv_xmm) { struct envxmm *penv_xmm = &sv_xmm->sv_env; struct envxmm *penv_fpreg = (struct envxmm *)&fpregs->fpr_env; int i; /* fpregs -> pcb */ /* FPU control/status */ penv_xmm->en_cw = penv_fpreg->en_cw; penv_xmm->en_sw = penv_fpreg->en_sw; penv_xmm->en_tw = penv_fpreg->en_tw; penv_xmm->en_opcode = penv_fpreg->en_opcode; penv_xmm->en_rip = penv_fpreg->en_rip; penv_xmm->en_rdp = penv_fpreg->en_rdp; penv_xmm->en_mxcsr = penv_fpreg->en_mxcsr; penv_xmm->en_mxcsr_mask = penv_fpreg->en_mxcsr_mask & cpu_mxcsr_mask; /* FPU registers */ for (i = 0; i < 8; ++i) bcopy(fpregs->fpr_acc[i], sv_xmm->sv_fp[i].fp_acc.fp_bytes, 10); /* SSE registers */ for (i = 0; i < 16; ++i) bcopy(fpregs->fpr_xacc[i], sv_xmm->sv_xmm[i].xmm_bytes, 16); } /* externalize from td->pcb */ int fill_fpregs(struct thread *td, struct fpreg *fpregs) { KASSERT(td == curthread || TD_IS_SUSPENDED(td) || P_SHOULDSTOP(td->td_proc), ("not suspended thread %p", td)); fpugetregs(td); fill_fpregs_xmm(get_pcb_user_save_td(td), fpregs); return (0); } /* internalize to td->pcb */ int set_fpregs(struct thread *td, struct fpreg *fpregs) { critical_enter(); set_fpregs_xmm(fpregs, get_pcb_user_save_td(td)); fpuuserinited(td); critical_exit(); return (0); } /* * Get machine context. */ int get_mcontext(struct thread *td, mcontext_t *mcp, int flags) { struct pcb *pcb; struct trapframe *tp; pcb = td->td_pcb; tp = td->td_frame; PROC_LOCK(curthread->td_proc); mcp->mc_onstack = sigonstack(tp->tf_rsp); PROC_UNLOCK(curthread->td_proc); mcp->mc_r15 = tp->tf_r15; mcp->mc_r14 = tp->tf_r14; mcp->mc_r13 = tp->tf_r13; mcp->mc_r12 = tp->tf_r12; mcp->mc_r11 = tp->tf_r11; mcp->mc_r10 = tp->tf_r10; mcp->mc_r9 = tp->tf_r9; mcp->mc_r8 = tp->tf_r8; mcp->mc_rdi = tp->tf_rdi; mcp->mc_rsi = tp->tf_rsi; mcp->mc_rbp = tp->tf_rbp; mcp->mc_rbx = tp->tf_rbx; mcp->mc_rcx = tp->tf_rcx; mcp->mc_rflags = tp->tf_rflags; if (flags & GET_MC_CLEAR_RET) { mcp->mc_rax = 0; mcp->mc_rdx = 0; mcp->mc_rflags &= ~PSL_C; } else { mcp->mc_rax = tp->tf_rax; mcp->mc_rdx = tp->tf_rdx; } mcp->mc_rip = tp->tf_rip; mcp->mc_cs = tp->tf_cs; mcp->mc_rsp = tp->tf_rsp; mcp->mc_ss = tp->tf_ss; mcp->mc_ds = tp->tf_ds; mcp->mc_es = tp->tf_es; mcp->mc_fs = tp->tf_fs; mcp->mc_gs = tp->tf_gs; mcp->mc_flags = tp->tf_flags; mcp->mc_len = sizeof(*mcp); get_fpcontext(td, mcp, NULL, 0); update_pcb_bases(pcb); mcp->mc_fsbase = pcb->pcb_fsbase; mcp->mc_gsbase = pcb->pcb_gsbase; mcp->mc_xfpustate = 0; mcp->mc_xfpustate_len = 0; bzero(mcp->mc_spare, sizeof(mcp->mc_spare)); return (0); } /* * Set machine context. * * However, we don't set any but the user modifiable flags, and we won't * touch the cs selector. */ int set_mcontext(struct thread *td, mcontext_t *mcp) { struct pcb *pcb; struct trapframe *tp; char *xfpustate; long rflags; int ret; pcb = td->td_pcb; tp = td->td_frame; if (mcp->mc_len != sizeof(*mcp) || (mcp->mc_flags & ~_MC_FLAG_MASK) != 0) return (EINVAL); rflags = (mcp->mc_rflags & PSL_USERCHANGE) | (tp->tf_rflags & ~PSL_USERCHANGE); if (mcp->mc_flags & _MC_HASFPXSTATE) { if (mcp->mc_xfpustate_len > cpu_max_ext_state_size - sizeof(struct savefpu)) return (EINVAL); xfpustate = (char *)fpu_save_area_alloc(); ret = copyin((void *)mcp->mc_xfpustate, xfpustate, mcp->mc_xfpustate_len); if (ret != 0) { fpu_save_area_free((struct savefpu *)xfpustate); return (ret); } } else xfpustate = NULL; ret = set_fpcontext(td, mcp, xfpustate, mcp->mc_xfpustate_len); fpu_save_area_free((struct savefpu *)xfpustate); if (ret != 0) return (ret); tp->tf_r15 = mcp->mc_r15; tp->tf_r14 = mcp->mc_r14; tp->tf_r13 = mcp->mc_r13; tp->tf_r12 = mcp->mc_r12; tp->tf_r11 = mcp->mc_r11; tp->tf_r10 = mcp->mc_r10; tp->tf_r9 = mcp->mc_r9; tp->tf_r8 = mcp->mc_r8; tp->tf_rdi = mcp->mc_rdi; tp->tf_rsi = mcp->mc_rsi; tp->tf_rbp = mcp->mc_rbp; tp->tf_rbx = mcp->mc_rbx; tp->tf_rdx = mcp->mc_rdx; tp->tf_rcx = mcp->mc_rcx; tp->tf_rax = mcp->mc_rax; tp->tf_rip = mcp->mc_rip; tp->tf_rflags = rflags; tp->tf_rsp = mcp->mc_rsp; tp->tf_ss = mcp->mc_ss; tp->tf_flags = mcp->mc_flags; if (tp->tf_flags & TF_HASSEGS) { tp->tf_ds = mcp->mc_ds; tp->tf_es = mcp->mc_es; tp->tf_fs = mcp->mc_fs; tp->tf_gs = mcp->mc_gs; } set_pcb_flags(pcb, PCB_FULL_IRET); if (mcp->mc_flags & _MC_HASBASES) { pcb->pcb_fsbase = mcp->mc_fsbase; pcb->pcb_gsbase = mcp->mc_gsbase; } return (0); } static void get_fpcontext(struct thread *td, mcontext_t *mcp, char **xfpusave, size_t *xfpusave_len) { mcp->mc_ownedfp = fpugetregs(td); bcopy(get_pcb_user_save_td(td), &mcp->mc_fpstate[0], sizeof(mcp->mc_fpstate)); mcp->mc_fpformat = fpuformat(); if (xfpusave == NULL) return; if (!use_xsave || cpu_max_ext_state_size <= sizeof(struct savefpu)) { *xfpusave_len = 0; *xfpusave = NULL; } else { mcp->mc_flags |= _MC_HASFPXSTATE; *xfpusave_len = mcp->mc_xfpustate_len = cpu_max_ext_state_size - sizeof(struct savefpu); *xfpusave = (char *)(get_pcb_user_save_td(td) + 1); } } static int set_fpcontext(struct thread *td, mcontext_t *mcp, char *xfpustate, size_t xfpustate_len) { int error; if (mcp->mc_fpformat == _MC_FPFMT_NODEV) return (0); else if (mcp->mc_fpformat != _MC_FPFMT_XMM) return (EINVAL); else if (mcp->mc_ownedfp == _MC_FPOWNED_NONE) { /* We don't care what state is left in the FPU or PCB. */ fpstate_drop(td); error = 0; } else if (mcp->mc_ownedfp == _MC_FPOWNED_FPU || mcp->mc_ownedfp == _MC_FPOWNED_PCB) { error = fpusetregs(td, (struct savefpu *)&mcp->mc_fpstate, xfpustate, xfpustate_len); } else return (EINVAL); return (error); } void fpstate_drop(struct thread *td) { KASSERT(PCB_USER_FPU(td->td_pcb), ("fpstate_drop: kernel-owned fpu")); critical_enter(); if (PCPU_GET(fpcurthread) == td) fpudrop(); /* * XXX force a full drop of the fpu. The above only drops it if we * owned it. * * XXX I don't much like fpugetuserregs()'s semantics of doing a full * drop. Dropping only to the pcb matches fnsave's behaviour. * We only need to drop to !PCB_INITDONE in sendsig(). But * sendsig() is the only caller of fpugetuserregs()... perhaps we just * have too many layers. */ clear_pcb_flags(curthread->td_pcb, PCB_FPUINITDONE | PCB_USERFPUINITDONE); critical_exit(); } int fill_dbregs(struct thread *td, struct dbreg *dbregs) { struct pcb *pcb; if (td == NULL) { dbregs->dr[0] = rdr0(); dbregs->dr[1] = rdr1(); dbregs->dr[2] = rdr2(); dbregs->dr[3] = rdr3(); dbregs->dr[6] = rdr6(); dbregs->dr[7] = rdr7(); } else { pcb = td->td_pcb; dbregs->dr[0] = pcb->pcb_dr0; dbregs->dr[1] = pcb->pcb_dr1; dbregs->dr[2] = pcb->pcb_dr2; dbregs->dr[3] = pcb->pcb_dr3; dbregs->dr[6] = pcb->pcb_dr6; dbregs->dr[7] = pcb->pcb_dr7; } dbregs->dr[4] = 0; dbregs->dr[5] = 0; dbregs->dr[8] = 0; dbregs->dr[9] = 0; dbregs->dr[10] = 0; dbregs->dr[11] = 0; dbregs->dr[12] = 0; dbregs->dr[13] = 0; dbregs->dr[14] = 0; dbregs->dr[15] = 0; return (0); } int set_dbregs(struct thread *td, struct dbreg *dbregs) { struct pcb *pcb; int i; if (td == NULL) { load_dr0(dbregs->dr[0]); load_dr1(dbregs->dr[1]); load_dr2(dbregs->dr[2]); load_dr3(dbregs->dr[3]); load_dr6(dbregs->dr[6]); load_dr7(dbregs->dr[7]); } else { /* * Don't let an illegal value for dr7 get set. Specifically, * check for undefined settings. Setting these bit patterns * result in undefined behaviour and can lead to an unexpected * TRCTRAP or a general protection fault right here. * Upper bits of dr6 and dr7 must not be set */ for (i = 0; i < 4; i++) { if (DBREG_DR7_ACCESS(dbregs->dr[7], i) == 0x02) return (EINVAL); if (td->td_frame->tf_cs == _ucode32sel && DBREG_DR7_LEN(dbregs->dr[7], i) == DBREG_DR7_LEN_8) return (EINVAL); } if ((dbregs->dr[6] & 0xffffffff00000000ul) != 0 || (dbregs->dr[7] & 0xffffffff00000000ul) != 0) return (EINVAL); pcb = td->td_pcb; /* * Don't let a process set a breakpoint that is not within the * process's address space. If a process could do this, it * could halt the system by setting a breakpoint in the kernel * (if ddb was enabled). Thus, we need to check to make sure * that no breakpoints are being enabled for addresses outside * process's address space. * * XXX - what about when the watched area of the user's * address space is written into from within the kernel * ... wouldn't that still cause a breakpoint to be generated * from within kernel mode? */ if (DBREG_DR7_ENABLED(dbregs->dr[7], 0)) { /* dr0 is enabled */ if (dbregs->dr[0] >= VM_MAXUSER_ADDRESS) return (EINVAL); } if (DBREG_DR7_ENABLED(dbregs->dr[7], 1)) { /* dr1 is enabled */ if (dbregs->dr[1] >= VM_MAXUSER_ADDRESS) return (EINVAL); } if (DBREG_DR7_ENABLED(dbregs->dr[7], 2)) { /* dr2 is enabled */ if (dbregs->dr[2] >= VM_MAXUSER_ADDRESS) return (EINVAL); } if (DBREG_DR7_ENABLED(dbregs->dr[7], 3)) { /* dr3 is enabled */ if (dbregs->dr[3] >= VM_MAXUSER_ADDRESS) return (EINVAL); } pcb->pcb_dr0 = dbregs->dr[0]; pcb->pcb_dr1 = dbregs->dr[1]; pcb->pcb_dr2 = dbregs->dr[2]; pcb->pcb_dr3 = dbregs->dr[3]; pcb->pcb_dr6 = dbregs->dr[6]; pcb->pcb_dr7 = dbregs->dr[7]; set_pcb_flags(pcb, PCB_DBREGS); } return (0); } void reset_dbregs(void) { load_dr7(0); /* Turn off the control bits first */ load_dr0(0); load_dr1(0); load_dr2(0); load_dr3(0); load_dr6(0); } /* * Return > 0 if a hardware breakpoint has been hit, and the * breakpoint was in user space. Return 0, otherwise. */ int user_dbreg_trap(register_t dr6) { u_int64_t dr7; u_int64_t bp; /* breakpoint bits extracted from dr6 */ int nbp; /* number of breakpoints that triggered */ caddr_t addr[4]; /* breakpoint addresses */ int i; bp = dr6 & DBREG_DR6_BMASK; if (bp == 0) { /* * None of the breakpoint bits are set meaning this * trap was not caused by any of the debug registers */ return (0); } dr7 = rdr7(); if ((dr7 & 0x000000ff) == 0) { /* * all GE and LE bits in the dr7 register are zero, * thus the trap couldn't have been caused by the * hardware debug registers */ return (0); } nbp = 0; /* * at least one of the breakpoints were hit, check to see * which ones and if any of them are user space addresses */ if (bp & 0x01) { addr[nbp++] = (caddr_t)rdr0(); } if (bp & 0x02) { addr[nbp++] = (caddr_t)rdr1(); } if (bp & 0x04) { addr[nbp++] = (caddr_t)rdr2(); } if (bp & 0x08) { addr[nbp++] = (caddr_t)rdr3(); } for (i = 0; i < nbp; i++) { if (addr[i] < (caddr_t)VM_MAXUSER_ADDRESS) { /* * addr[i] is in user space */ return (nbp); } } /* * None of the breakpoints are in user space. */ return (0); } diff --git a/sys/amd64/linux/linux_sysvec.c b/sys/amd64/linux/linux_sysvec.c index 28fac2ea7452..e02221c324ec 100644 --- a/sys/amd64/linux/linux_sysvec.c +++ b/sys/amd64/linux/linux_sysvec.c @@ -1,941 +1,943 @@ /*- * Copyright (c) 2013 Dmitry Chagin * Copyright (c) 2004 Tim J. Robbins * Copyright (c) 2003 Peter Wemm * Copyright (c) 2002 Doug Rabson * Copyright (c) 1998-1999 Andrew Gallatin * Copyright (c) 1994-1996 Søren Schmidt * 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 * in this position and unchanged. * 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. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission * * 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 __FBSDID("$FreeBSD$"); #define __ELF_WORD_SIZE 64 #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 MODULE_VERSION(linux64, 1); const char *linux_kplatform; static int linux_szsigcode; static vm_object_t linux_shared_page_obj; static char *linux_shared_page_mapping; extern char _binary_linux_locore_o_start; extern char _binary_linux_locore_o_end; extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); static int linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base); static int linux_fixup_elf(uintptr_t *stack_base, struct image_params *iparams); static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel); static void linux_vdso_install(void *param); static void linux_vdso_deinstall(void *param); static void linux_set_syscall_retval(struct thread *td, int error); static int linux_fetch_syscall_args(struct thread *td); static void linux_exec_setregs(struct thread *td, struct image_params *imgp, uintptr_t stack); static int linux_vsyscall(struct thread *td); #define LINUX_T_UNKNOWN 255 static int _bsd_to_linux_trapcode[] = { LINUX_T_UNKNOWN, /* 0 */ 6, /* 1 T_PRIVINFLT */ LINUX_T_UNKNOWN, /* 2 */ 3, /* 3 T_BPTFLT */ LINUX_T_UNKNOWN, /* 4 */ LINUX_T_UNKNOWN, /* 5 */ 16, /* 6 T_ARITHTRAP */ 254, /* 7 T_ASTFLT */ LINUX_T_UNKNOWN, /* 8 */ 13, /* 9 T_PROTFLT */ 1, /* 10 T_TRCTRAP */ LINUX_T_UNKNOWN, /* 11 */ 14, /* 12 T_PAGEFLT */ LINUX_T_UNKNOWN, /* 13 */ 17, /* 14 T_ALIGNFLT */ LINUX_T_UNKNOWN, /* 15 */ LINUX_T_UNKNOWN, /* 16 */ LINUX_T_UNKNOWN, /* 17 */ 0, /* 18 T_DIVIDE */ 2, /* 19 T_NMI */ 4, /* 20 T_OFLOW */ 5, /* 21 T_BOUND */ 7, /* 22 T_DNA */ 8, /* 23 T_DOUBLEFLT */ 9, /* 24 T_FPOPFLT */ 10, /* 25 T_TSSFLT */ 11, /* 26 T_SEGNPFLT */ 12, /* 27 T_STKFLT */ 18, /* 28 T_MCHK */ 19, /* 29 T_XMMFLT */ 15 /* 30 T_RESERVED */ }; #define bsd_to_linux_trapcode(code) \ ((code)td_proc; frame = td->td_frame; sa = &td->td_sa; sa->args[0] = frame->tf_rdi; sa->args[1] = frame->tf_rsi; sa->args[2] = frame->tf_rdx; sa->args[3] = frame->tf_rcx; sa->args[4] = frame->tf_r8; sa->args[5] = frame->tf_r9; sa->code = frame->tf_rax; if (sa->code >= p->p_sysent->sv_size) /* nosys */ sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; else sa->callp = &p->p_sysent->sv_table[sa->code]; td->td_retval[0] = 0; return (0); } static void linux_set_syscall_retval(struct thread *td, int error) { struct trapframe *frame; frame = td->td_frame; switch (error) { case 0: frame->tf_rax = td->td_retval[0]; frame->tf_r10 = frame->tf_rcx; break; case ERESTART: /* * Reconstruct pc, we know that 'syscall' is 2 bytes, * lcall $X,y is 7 bytes, int 0x80 is 2 bytes. * We saved this in tf_err. * */ frame->tf_rip -= frame->tf_err; frame->tf_r10 = frame->tf_rcx; break; case EJUSTRETURN: break; default: frame->tf_rax = bsd_to_linux_errno(error); frame->tf_r10 = frame->tf_rcx; break; } /* * Differently from FreeBSD native ABI, on Linux only %rcx * and %r11 values are not preserved across the syscall. * Require full context restore to get all registers except * those two restored at return to usermode. * * XXX: Would be great to be able to avoid PCB_FULL_IRET * for the error == 0 case. */ set_pcb_flags(td->td_pcb, PCB_FULL_IRET); } static int linux_copyout_auxargs(struct image_params *imgp, uintptr_t base) { Elf_Auxargs *args; Elf_Auxinfo *argarray, *pos; struct proc *p; int error, issetugid; p = imgp->proc; args = (Elf64_Auxargs *)imgp->auxargs; argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, M_WAITOK | M_ZERO); issetugid = p->p_flag & P_SUGID ? 1 : 0; AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, imgp->proc->p_sysent->sv_shared_page_base); AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); AUXARGS_ENTRY(pos, AT_PHENT, args->phent); AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); AUXARGS_ENTRY(pos, AT_BASE, args->base); AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary); if (imgp->execpathp != 0) AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp); if (args->execfd != -1) AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); AUXARGS_ENTRY(pos, AT_NULL, 0); free(imgp->auxargs, M_TEMP); imgp->auxargs = NULL; KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); error = copyout(argarray, (void *)base, sizeof(*argarray) * LINUX_AT_COUNT); free(argarray, M_TEMP); return (error); } static int linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp) { Elf_Addr *base; base = (Elf64_Addr *)*stack_base; base--; if (suword(base, (uint64_t)imgp->args->argc) == -1) return (EFAULT); *stack_base = (uintptr_t)base; return (0); } /* * Copy strings out to the new process address space, constructing new arg * and env vector tables. Return a pointer to the base so that it can be used * as the initial stack pointer. */ static int linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) { int argc, envc, error; char **vectp; char *stringp; uintptr_t destp, ustringp; struct ps_strings *arginfo; char canary[LINUX_AT_RANDOM_LEN]; size_t execpath_len; struct proc *p; /* Calculate string base and vector table pointers. */ if (imgp->execpath != NULL && imgp->auxargs != NULL) execpath_len = strlen(imgp->execpath) + 1; else execpath_len = 0; p = imgp->proc; arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; destp = (uintptr_t)arginfo; if (execpath_len != 0) { destp -= execpath_len; destp = rounddown2(destp, sizeof(void *)); 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 -= roundup(sizeof(canary), sizeof(void *)); imgp->canary = (void *)destp; error = copyout(canary, imgp->canary, sizeof(canary)); if (error != 0) return (error); /* Allocate room for the argument and environment strings. */ destp -= ARG_MAX - imgp->args->stringspace; destp = rounddown2(destp, sizeof(void *)); ustringp = destp; if (imgp->auxargs) { /* * Allocate room on the stack for the ELF auxargs * array. It has LINUX_AT_COUNT entries. */ destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo); destp = rounddown2(destp, sizeof(void *)); } vectp = (char **)destp; /* * Allocate room for the argv[] and env vectors including the * terminating NULL pointers. */ vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; /* * Starting with 2.24, glibc depends on a 16-byte stack alignment. * One "long argc" will be prepended later. */ vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8); /* 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. */ if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || suword(&arginfo->ps_nargvstr, argc) != 0) return (EFAULT); /* Fill in argument portion of vector table. */ for (; argc > 0; --argc) { if (suword(vectp++, ustringp) != 0) return (EFAULT); while (*stringp++ != 0) ustringp++; ustringp++; } /* A null vector table pointer separates the argp's from the envp's. */ if (suword(vectp++, 0) != 0) return (EFAULT); if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || suword(&arginfo->ps_nenvstr, envc) != 0) return (EFAULT); /* Fill in environment portion of vector table. */ for (; envc > 0; --envc) { if (suword(vectp++, ustringp) != 0) return (EFAULT); while (*stringp++ != 0) ustringp++; ustringp++; } /* The end of the vector table is a null pointer. */ if (suword(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); } /* * Reset registers to default values on exec. */ static void linux_exec_setregs(struct thread *td, struct image_params *imgp, uintptr_t stack) { struct trapframe *regs; struct pcb *pcb; register_t saved_rflags; regs = td->td_frame; pcb = td->td_pcb; if (td->td_proc->p_md.md_ldt != NULL) user_ldt_free(td); pcb->pcb_fsbase = 0; pcb->pcb_gsbase = 0; clear_pcb_flags(pcb, PCB_32BIT); pcb->pcb_initial_fpucw = __LINUX_NPXCW__; set_pcb_flags(pcb, PCB_FULL_IRET); saved_rflags = regs->tf_rflags & PSL_T; bzero((char *)regs, sizeof(struct trapframe)); regs->tf_rip = imgp->entry_addr; regs->tf_rsp = stack; regs->tf_rflags = PSL_USER | saved_rflags; regs->tf_ss = _udatasel; regs->tf_cs = _ucodesel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; regs->tf_fs = _ufssel; regs->tf_gs = _ugssel; regs->tf_flags = TF_HASSEGS; x86_clear_dbregs(pcb); /* * Drop the FP state if we hold it, so that the process gets a * clean FP state if it uses the FPU again. */ fpstate_drop(td); } /* * Copied from amd64/amd64/machdep.c * * XXX fpu state need? don't think so */ int linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) { struct proc *p; struct l_ucontext uc; struct l_sigcontext *context; struct trapframe *regs; unsigned long rflags; int error; ksiginfo_t ksi; regs = td->td_frame; error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc)); if (error != 0) return (error); p = td->td_proc; context = &uc.uc_mcontext; rflags = context->sc_rflags; /* * Don't allow users to change privileged or reserved flags. */ /* * XXX do allow users to change the privileged flag PSL_RF. * The cpu sets PSL_RF in tf_rflags for faults. Debuggers * should sometimes set it there too. tf_rflags is kept in * the signal context during signal handling and there is no * other place to remember it, so the PSL_RF bit may be * corrupted by the signal handler without us knowing. * Corruption of the PSL_RF bit at worst causes one more or * one less debugger trap, so allowing it is fairly harmless. */ if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { - printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags); + uprintf("pid %d comm %s linux mangled rflags %#lx\n", + p->p_pid, p->p_comm, rflags); return (EINVAL); } /* * Don't allow users to load a valid privileged %cs. Let the * hardware check for invalid selectors, excess privilege in * other selectors, invalid %eip's and invalid %esp's. */ if (!CS_SECURE(context->sc_cs)) { - printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs); + uprintf("pid %d comm %s linux mangled cs %#x\n", + p->p_pid, p->p_comm, context->sc_cs); ksiginfo_init_trap(&ksi); ksi.ksi_signo = SIGBUS; ksi.ksi_code = BUS_OBJERR; ksi.ksi_trapno = T_PROTFLT; ksi.ksi_addr = (void *)regs->tf_rip; trapsignal(td, &ksi); return (EINVAL); } PROC_LOCK(p); linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask); SIG_CANTMASK(td->td_sigmask); signotify(td); PROC_UNLOCK(p); regs->tf_rdi = context->sc_rdi; regs->tf_rsi = context->sc_rsi; regs->tf_rdx = context->sc_rdx; regs->tf_rbp = context->sc_rbp; regs->tf_rbx = context->sc_rbx; regs->tf_rcx = context->sc_rcx; regs->tf_rax = context->sc_rax; regs->tf_rip = context->sc_rip; regs->tf_rsp = context->sc_rsp; regs->tf_r8 = context->sc_r8; regs->tf_r9 = context->sc_r9; regs->tf_r10 = context->sc_r10; regs->tf_r11 = context->sc_r11; regs->tf_r12 = context->sc_r12; regs->tf_r13 = context->sc_r13; regs->tf_r14 = context->sc_r14; regs->tf_r15 = context->sc_r15; regs->tf_cs = context->sc_cs; regs->tf_err = context->sc_err; regs->tf_rflags = rflags; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); return (EJUSTRETURN); } /* * copied from amd64/amd64/machdep.c * * Send an interrupt to process. */ static void linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct l_rt_sigframe sf, *sfp; struct proc *p; struct thread *td; struct sigacts *psp; caddr_t sp; struct trapframe *regs; int sig, code; int oonstack; td = curthread; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); sig = ksi->ksi_signo; psp = p->p_sigacts; code = ksi->ksi_code; mtx_assert(&psp->ps_mtx, MA_OWNED); regs = td->td_frame; oonstack = sigonstack(regs->tf_rsp); LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u", catcher, sig, mask, code); /* Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe); } else sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128; /* Align to 16 bytes. */ sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); mtx_unlock(&psp->ps_mtx); /* Translate the signal. */ sig = bsd_to_linux_signal(sig); /* Save user context. */ bzero(&sf, sizeof(sf)); bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask); bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask); sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; PROC_UNLOCK(p); sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi; sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi; sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx; sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp; sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx; sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx; sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax; sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip; sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp; sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8; sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9; sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10; sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11; sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12; sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13; sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14; sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15; sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags; sf.sf_sc.uc_mcontext.sc_err = regs->tf_err; sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; /* Build the argument list for the signal handler. */ regs->tf_rdi = sig; /* arg 1 in %rdi */ regs->tf_rax = 0; regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */ sf.sf_handler = catcher; /* Fill in POSIX parts. */ ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig); /* Copy the sigframe out to the user's stack. */ if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { PROC_LOCK(p); sigexit(td, SIGILL); } regs->tf_rsp = (long)sfp; regs->tf_rip = linux_rt_sigcode; regs->tf_rflags &= ~(PSL_T | PSL_D); regs->tf_cs = _ucodesel; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } #define LINUX_VSYSCALL_START (-10UL << 20) #define LINUX_VSYSCALL_SZ 1024 const unsigned long linux_vsyscall_vector[] = { LINUX_SYS_gettimeofday, LINUX_SYS_linux_time, LINUX_SYS_linux_getcpu, }; static int linux_vsyscall(struct thread *td) { struct trapframe *frame; uint64_t retqaddr; int code, traced; int error; frame = td->td_frame; /* Check %rip for vsyscall area. */ if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START)) return (EINVAL); if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0) return (EINVAL); code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ; if (code >= nitems(linux_vsyscall_vector)) return (EINVAL); /* * vsyscall called as callq *(%rax), so we must * use return address from %rsp and also fixup %rsp. */ error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr)); if (error) return (error); frame->tf_rip = retqaddr; frame->tf_rax = linux_vsyscall_vector[code]; frame->tf_rsp += 8; traced = (frame->tf_flags & PSL_T); amd64_syscall(td, traced); return (0); } struct sysentvec elf_linux_sysvec = { .sv_size = LINUX_SYS_MAXSYSCALL, .sv_table = linux_sysent, .sv_transtrap = linux_translate_traps, .sv_fixup = linux_fixup_elf, .sv_sendsig = linux_rt_sendsig, .sv_sigcode = &_binary_linux_locore_o_start, .sv_szsigcode = &linux_szsigcode, .sv_name = "Linux ELF64", .sv_coredump = elf64_coredump, .sv_imgact_try = linux_exec_imgact_try, .sv_minsigstksz = LINUX_MINSIGSTKSZ, .sv_minuser = VM_MIN_ADDRESS, .sv_maxuser = VM_MAXUSER_ADDRESS_LA48, .sv_usrstack = USRSTACK_LA48, .sv_psstrings = PS_STRINGS_LA48, .sv_stackprot = VM_PROT_ALL, .sv_copyout_auxargs = linux_copyout_auxargs, .sv_copyout_strings = linux_copyout_strings, .sv_setregs = linux_exec_setregs, .sv_fixlimit = NULL, .sv_maxssiz = NULL, .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN | SV_SIG_WAITNDQ, .sv_set_syscall_retval = linux_set_syscall_retval, .sv_fetch_syscall_args = linux_fetch_syscall_args, .sv_syscallnames = NULL, .sv_shared_page_base = SHAREDPAGE_LA48, .sv_shared_page_len = PAGE_SIZE, .sv_schedtail = linux_schedtail, .sv_thread_detach = linux_thread_detach, .sv_trap = linux_vsyscall, .sv_onexec = linux_on_exec, .sv_onexit = linux_on_exit, .sv_ontdexit = linux_thread_dtor, .sv_setid_allowed = &linux_setid_allowed_query, }; static void linux_vdso_install(void *param) { amd64_lower_shared_page(&elf_linux_sysvec); linux_szsigcode = (&_binary_linux_locore_o_end - &_binary_linux_locore_o_start); if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) panic("Linux invalid vdso size\n"); __elfN(linux_vdso_fixup)(&elf_linux_sysvec); linux_shared_page_obj = __elfN(linux_shared_page_init) (&linux_shared_page_mapping); __elfN(linux_vdso_reloc)(&elf_linux_sysvec); bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, linux_szsigcode); elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; linux_kplatform = linux_shared_page_mapping + (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base); } SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, linux_vdso_install, NULL); static void linux_vdso_deinstall(void *param) { __elfN(linux_shared_page_fini)(linux_shared_page_obj, linux_shared_page_mapping); } SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, linux_vdso_deinstall, NULL); static char GNULINUX_ABI_VENDOR[] = "GNU"; static int GNULINUX_ABI_DESC = 0; static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel) { const Elf32_Word *desc; uintptr_t p; p = (uintptr_t)(note + 1); p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); desc = (const Elf32_Word *)p; if (desc[0] != GNULINUX_ABI_DESC) return (false); /* * For Linux we encode osrel using the Linux convention of * (version << 16) | (major << 8) | (minor) * See macro in linux_mib.h */ *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); return (true); } static Elf_Brandnote linux64_brandnote = { .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR), .hdr.n_descsz = 16, .hdr.n_type = 1, .vendor = GNULINUX_ABI_VENDOR, .flags = BN_TRANSLATE_OSREL, .trans_osrel = linux_trans_osrel }; static Elf64_Brandinfo linux_glibc2brand = { .brand = ELFOSABI_LINUX, .machine = EM_X86_64, .compat_3_brand = "Linux", .emul_path = linux_emul_path, .interp_path = "/lib64/ld-linux-x86-64.so.2", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux64_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; static Elf64_Brandinfo linux_glibc2brandshort = { .brand = ELFOSABI_LINUX, .machine = EM_X86_64, .compat_3_brand = "Linux", .emul_path = linux_emul_path, .interp_path = "/lib64/ld-linux.so.2", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux64_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; static Elf64_Brandinfo linux_muslbrand = { .brand = ELFOSABI_LINUX, .machine = EM_X86_64, .compat_3_brand = "Linux", .emul_path = linux_emul_path, .interp_path = "/lib/ld-musl-x86_64.so.1", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux64_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; Elf64_Brandinfo *linux_brandlist[] = { &linux_glibc2brand, &linux_glibc2brandshort, &linux_muslbrand, NULL }; static int linux64_elf_modevent(module_t mod, int type, void *data) { Elf64_Brandinfo **brandinfo; int error; struct linux_ioctl_handler **lihp; error = 0; switch(type) { case MOD_LOAD: for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; ++brandinfo) if (elf64_insert_brand_entry(*brandinfo) < 0) error = EINVAL; if (error == 0) { SET_FOREACH(lihp, linux_ioctl_handler_set) linux_ioctl_register_handler(*lihp); stclohz = (stathz ? stathz : hz); if (bootverbose) printf("Linux x86-64 ELF exec handler installed\n"); } else printf("cannot insert Linux x86-64 ELF brand handler\n"); break; case MOD_UNLOAD: for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; ++brandinfo) if (elf64_brand_inuse(*brandinfo)) error = EBUSY; if (error == 0) { for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; ++brandinfo) if (elf64_remove_brand_entry(*brandinfo) < 0) error = EINVAL; } if (error == 0) { SET_FOREACH(lihp, linux_ioctl_handler_set) linux_ioctl_unregister_handler(*lihp); if (bootverbose) printf("Linux ELF exec handler removed\n"); } else printf("Could not deinstall ELF interpreter entry\n"); break; default: return (EOPNOTSUPP); } return (error); } static moduledata_t linux64_elf_mod = { "linux64elf", linux64_elf_modevent, 0 }; DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1); FEATURE(linux64, "Linux 64bit support");