diff --git a/sys/amd64/amd64/exec_machdep.c b/sys/amd64/amd64/exec_machdep.c index 8361f9fcfcab..46e4f0c9e07b 100644 --- a/sys/amd64/amd64/exec_machdep.c +++ b/sys/amd64/amd64/exec_machdep.c @@ -1,973 +1,975 @@ /*- * 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. */ #include #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 #include #ifdef SMP #include #endif #include #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 _Static_assert(sizeof(mcontext_t) == 800, "mcontext_t size incorrect"); _Static_assert(sizeof(ucontext_t) == 880, "ucontext_t size incorrect"); _Static_assert(sizeof(siginfo_t) == 80, "siginfo_t size incorrect"); /* * 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); 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)) { uprintf("pid %d comm %s has trashed its stack, killing\n", p->p_pid, p->p_comm); PROC_LOCK(p); sigexit(td, SIGILL); } fpstate_drop(td); regs->tf_rsp = (long)sfp; regs->tf_rip = PROC_SIGCODE(p); 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; + if ((pcb->pcb_flags & PCB_TLSBASE) != 0) + pcb->pcb_fsbase = pcb->pcb_tlsbase; 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_fsbase = pcb->pcb_tlsbase = 0; pcb->pcb_gsbase = 0; - clear_pcb_flags(pcb, PCB_32BIT); + clear_pcb_flags(pcb, PCB_32BIT | PCB_TLSBASE); 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, NULL); 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); } 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); } } 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/amd64/sys_machdep.c b/sys/amd64/amd64/sys_machdep.c index 70a369ec64a3..6deebba5babc 100644 --- a/sys/amd64/amd64/sys_machdep.c +++ b/sys/amd64/amd64/sys_machdep.c @@ -1,808 +1,823 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2003 Peter Wemm. * Copyright (c) 1990 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 #include "opt_capsicum.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* for kernel_map */ #include #include #include #include #include #include #include #include #include #include static void user_ldt_deref(struct proc_ldt *pldt); static void user_ldt_derefl(struct proc_ldt *pldt); #define MAX_LD 8192 int max_ldt_segment = 512; SYSCTL_INT(_machdep, OID_AUTO, max_ldt_segment, CTLFLAG_RDTUN, &max_ldt_segment, 0, "Maximum number of allowed LDT segments in the single address space"); static void max_ldt_segment_init(void *arg __unused) { if (max_ldt_segment <= 0) max_ldt_segment = 1; if (max_ldt_segment > MAX_LD) max_ldt_segment = MAX_LD; } SYSINIT(maxldt, SI_SUB_VM_CONF, SI_ORDER_ANY, max_ldt_segment_init, NULL); #ifndef _SYS_SYSPROTO_H_ struct sysarch_args { int op; char *parms; }; #endif int sysarch_ldt(struct thread *td, struct sysarch_args *uap, int uap_space) { struct i386_ldt_args *largs, la; struct user_segment_descriptor *lp; int error = 0; /* * XXXKIB check that the BSM generation code knows to encode * the op argument. */ AUDIT_ARG_CMD(uap->op); if (uap_space == UIO_USERSPACE) { error = copyin(uap->parms, &la, sizeof(struct i386_ldt_args)); if (error != 0) return (error); largs = &la; } else largs = (struct i386_ldt_args *)uap->parms; switch (uap->op) { case I386_GET_LDT: error = amd64_get_ldt(td, largs); break; case I386_SET_LDT: if (largs->descs != NULL && largs->num > max_ldt_segment) return (EINVAL); set_pcb_flags(td->td_pcb, PCB_FULL_IRET); if (largs->descs != NULL) { lp = malloc(largs->num * sizeof(struct user_segment_descriptor), M_TEMP, M_WAITOK); error = copyin(largs->descs, lp, largs->num * sizeof(struct user_segment_descriptor)); if (error == 0) error = amd64_set_ldt(td, largs, lp); free(lp, M_TEMP); } else { error = amd64_set_ldt(td, largs, NULL); } break; } return (error); } void update_gdt_gsbase(struct thread *td, uint32_t base) { struct user_segment_descriptor *sd; if (td != curthread) return; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); critical_enter(); sd = PCPU_GET(gs32p); sd->sd_lobase = base & 0xffffff; sd->sd_hibase = (base >> 24) & 0xff; critical_exit(); } void update_gdt_fsbase(struct thread *td, uint32_t base) { struct user_segment_descriptor *sd; if (td != curthread) return; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); critical_enter(); sd = PCPU_GET(fs32p); sd->sd_lobase = base & 0xffffff; sd->sd_hibase = (base >> 24) & 0xff; critical_exit(); } int sysarch(struct thread *td, struct sysarch_args *uap) { struct pcb *pcb; struct vm_map *map; uint32_t i386base; uint64_t a64base; struct i386_ioperm_args iargs; struct i386_get_xfpustate i386xfpu; struct i386_set_pkru i386pkru; struct amd64_get_xfpustate a64xfpu; struct amd64_set_pkru a64pkru; int error; #ifdef CAPABILITY_MODE /* * When adding new operations, add a new case statement here to * explicitly indicate whether or not the operation is safe to * perform in capability mode. */ switch (uap->op) { case I386_GET_LDT: case I386_SET_LDT: case I386_GET_IOPERM: case I386_GET_FSBASE: case I386_SET_FSBASE: case I386_GET_GSBASE: case I386_SET_GSBASE: case I386_GET_XFPUSTATE: case I386_SET_PKRU: case I386_CLEAR_PKRU: case AMD64_GET_FSBASE: case AMD64_SET_FSBASE: case AMD64_GET_GSBASE: case AMD64_SET_GSBASE: case AMD64_GET_XFPUSTATE: case AMD64_SET_PKRU: case AMD64_CLEAR_PKRU: + case AMD64_GET_TLSBASE: + case AMD64_SET_TLSBASE: break; case I386_SET_IOPERM: default: if (CAP_TRACING(td)) ktrcapfail(CAPFAIL_SYSCALL, &uap->op); if (IN_CAPABILITY_MODE(td)) return (ECAPMODE); break; } #endif if (uap->op == I386_GET_LDT || uap->op == I386_SET_LDT) return (sysarch_ldt(td, uap, UIO_USERSPACE)); error = 0; pcb = td->td_pcb; /* * XXXKIB check that the BSM generation code knows to encode * the op argument. */ AUDIT_ARG_CMD(uap->op); switch (uap->op) { case I386_GET_IOPERM: case I386_SET_IOPERM: if ((error = copyin(uap->parms, &iargs, sizeof(struct i386_ioperm_args))) != 0) return (error); break; case I386_GET_XFPUSTATE: if ((error = copyin(uap->parms, &i386xfpu, sizeof(struct i386_get_xfpustate))) != 0) return (error); a64xfpu.addr = (void *)(uintptr_t)i386xfpu.addr; a64xfpu.len = i386xfpu.len; break; case I386_SET_PKRU: case I386_CLEAR_PKRU: if ((error = copyin(uap->parms, &i386pkru, sizeof(struct i386_set_pkru))) != 0) return (error); a64pkru.addr = (void *)(uintptr_t)i386pkru.addr; a64pkru.len = i386pkru.len; a64pkru.keyidx = i386pkru.keyidx; a64pkru.flags = i386pkru.flags; break; case AMD64_GET_XFPUSTATE: if ((error = copyin(uap->parms, &a64xfpu, sizeof(struct amd64_get_xfpustate))) != 0) return (error); break; case AMD64_SET_PKRU: case AMD64_CLEAR_PKRU: if ((error = copyin(uap->parms, &a64pkru, sizeof(struct amd64_set_pkru))) != 0) return (error); break; default: break; } switch (uap->op) { case I386_GET_IOPERM: error = amd64_get_ioperm(td, &iargs); if (error == 0) error = copyout(&iargs, uap->parms, sizeof(struct i386_ioperm_args)); break; case I386_SET_IOPERM: error = amd64_set_ioperm(td, &iargs); break; case I386_GET_FSBASE: update_pcb_bases(pcb); i386base = pcb->pcb_fsbase; error = copyout(&i386base, uap->parms, sizeof(i386base)); break; case I386_SET_FSBASE: error = copyin(uap->parms, &i386base, sizeof(i386base)); if (error == 0) { set_pcb_flags(pcb, PCB_FULL_IRET); pcb->pcb_fsbase = i386base; td->td_frame->tf_fs = _ufssel; update_gdt_fsbase(td, i386base); } break; case I386_GET_GSBASE: update_pcb_bases(pcb); i386base = pcb->pcb_gsbase; error = copyout(&i386base, uap->parms, sizeof(i386base)); break; case I386_SET_GSBASE: error = copyin(uap->parms, &i386base, sizeof(i386base)); if (error == 0) { set_pcb_flags(pcb, PCB_FULL_IRET); pcb->pcb_gsbase = i386base; td->td_frame->tf_gs = _ugssel; update_gdt_gsbase(td, i386base); } break; case AMD64_GET_FSBASE: update_pcb_bases(pcb); error = copyout(&pcb->pcb_fsbase, uap->parms, sizeof(pcb->pcb_fsbase)); break; + case AMD64_GET_TLSBASE: + if ((pcb->pcb_flags & PCB_TLSBASE) == 0) { + error = ESRCH; + } else { + error = copyout(&pcb->pcb_tlsbase, uap->parms, + sizeof(pcb->pcb_tlsbase)); + } + break; case AMD64_SET_FSBASE: + case AMD64_SET_TLSBASE: error = copyin(uap->parms, &a64base, sizeof(a64base)); if (error == 0) { if (a64base < curproc->p_sysent->sv_maxuser) { set_pcb_flags(pcb, PCB_FULL_IRET); pcb->pcb_fsbase = a64base; td->td_frame->tf_fs = _ufssel; + if (uap->op == AMD64_SET_TLSBASE) { + pcb->pcb_tlsbase = a64base; + set_pcb_flags(pcb, PCB_TLSBASE); + } } else error = EINVAL; } break; case AMD64_GET_GSBASE: update_pcb_bases(pcb); error = copyout(&pcb->pcb_gsbase, uap->parms, sizeof(pcb->pcb_gsbase)); break; case AMD64_SET_GSBASE: error = copyin(uap->parms, &a64base, sizeof(a64base)); if (error == 0) { if (a64base < curproc->p_sysent->sv_maxuser) { set_pcb_flags(pcb, PCB_FULL_IRET); pcb->pcb_gsbase = a64base; td->td_frame->tf_gs = _ugssel; } else error = EINVAL; } break; case I386_GET_XFPUSTATE: case AMD64_GET_XFPUSTATE: if (a64xfpu.len > cpu_max_ext_state_size - sizeof(struct savefpu)) return (EINVAL); fpugetregs(td); error = copyout((char *)(get_pcb_user_save_td(td) + 1), a64xfpu.addr, a64xfpu.len); break; case I386_SET_PKRU: case AMD64_SET_PKRU: /* * Read-lock the map to synchronize with parallel * pmap_vmspace_copy() on fork. */ map = &td->td_proc->p_vmspace->vm_map; vm_map_lock_read(map); error = pmap_pkru_set(PCPU_GET(curpmap), (vm_offset_t)a64pkru.addr, (vm_offset_t)a64pkru.addr + a64pkru.len, a64pkru.keyidx, a64pkru.flags); vm_map_unlock_read(map); break; case I386_CLEAR_PKRU: case AMD64_CLEAR_PKRU: if (a64pkru.flags != 0 || a64pkru.keyidx != 0) { error = EINVAL; break; } map = &td->td_proc->p_vmspace->vm_map; vm_map_lock_read(map); error = pmap_pkru_clear(PCPU_GET(curpmap), (vm_offset_t)a64pkru.addr, (vm_offset_t)a64pkru.addr + a64pkru.len); vm_map_unlock_read(map); break; default: error = EINVAL; break; } return (error); } int amd64_set_ioperm(struct thread *td, struct i386_ioperm_args *uap) { char *iomap; struct amd64tss *tssp; struct system_segment_descriptor *tss_sd; struct pcb *pcb; u_int i; int error; if ((error = priv_check(td, PRIV_IO)) != 0) return (error); if ((error = securelevel_gt(td->td_ucred, 0)) != 0) return (error); if (uap->start > uap->start + uap->length || uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY) return (EINVAL); /* * XXX * While this is restricted to root, we should probably figure out * whether any other driver is using this i/o address, as so not to * cause confusion. This probably requires a global 'usage registry'. */ pcb = td->td_pcb; if (pcb->pcb_tssp == NULL) { tssp = kmem_malloc(ctob(IOPAGES + 1), M_WAITOK); pmap_pti_add_kva((vm_offset_t)tssp, (vm_offset_t)tssp + ctob(IOPAGES + 1), false); iomap = (char *)&tssp[1]; memset(iomap, 0xff, IOPERM_BITMAP_SIZE); critical_enter(); /* Takes care of tss_rsp0. */ memcpy(tssp, PCPU_PTR(common_tss), sizeof(struct amd64tss)); tssp->tss_iobase = sizeof(*tssp); pcb->pcb_tssp = tssp; tss_sd = PCPU_GET(tss); tss_sd->sd_lobase = (u_long)tssp & 0xffffff; tss_sd->sd_hibase = ((u_long)tssp >> 24) & 0xfffffffffful; tss_sd->sd_type = SDT_SYSTSS; ltr(GSEL(GPROC0_SEL, SEL_KPL)); PCPU_SET(tssp, tssp); critical_exit(); } else iomap = (char *)&pcb->pcb_tssp[1]; for (i = uap->start; i < uap->start + uap->length; i++) { if (uap->enable) iomap[i >> 3] &= ~(1 << (i & 7)); else iomap[i >> 3] |= (1 << (i & 7)); } return (error); } int amd64_get_ioperm(struct thread *td, struct i386_ioperm_args *uap) { int i, state; char *iomap; if (uap->start >= IOPAGES * PAGE_SIZE * NBBY) return (EINVAL); if (td->td_pcb->pcb_tssp == NULL) { uap->length = 0; goto done; } iomap = (char *)&td->td_pcb->pcb_tssp[1]; i = uap->start; state = (iomap[i >> 3] >> (i & 7)) & 1; uap->enable = !state; uap->length = 1; for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) { if (state != ((iomap[i >> 3] >> (i & 7)) & 1)) break; uap->length++; } done: return (0); } /* * Update the GDT entry pointing to the LDT to point to the LDT of the * current process. */ static void set_user_ldt(struct mdproc *mdp) { *PCPU_GET(ldt) = mdp->md_ldt_sd; lldt(GSEL(GUSERLDT_SEL, SEL_KPL)); } static void set_user_ldt_rv(void *arg) { struct proc *orig, *target; struct proc_ldt *ldt; orig = arg; target = curthread->td_proc; ldt = (void *)atomic_load_acq_ptr((uintptr_t *)&orig->p_md.md_ldt); if (target->p_md.md_ldt != ldt) return; set_user_ldt(&target->p_md); } struct proc_ldt * user_ldt_alloc(struct proc *p, int force) { struct proc_ldt *pldt, *new_ldt; struct mdproc *mdp; struct soft_segment_descriptor sldt; vm_offset_t sva; vm_size_t sz; mtx_assert(&dt_lock, MA_OWNED); mdp = &p->p_md; if (!force && mdp->md_ldt != NULL) return (mdp->md_ldt); mtx_unlock(&dt_lock); new_ldt = malloc(sizeof(struct proc_ldt), M_SUBPROC, M_WAITOK); sz = max_ldt_segment * sizeof(struct user_segment_descriptor); new_ldt->ldt_base = kmem_malloc(sz, M_WAITOK | M_ZERO); sva = (uintptr_t)new_ldt->ldt_base; pmap_pti_add_kva(sva, sva + sz, false); new_ldt->ldt_refcnt = 1; sldt.ssd_base = sva; sldt.ssd_limit = sz - 1; sldt.ssd_type = SDT_SYSLDT; sldt.ssd_dpl = SEL_KPL; sldt.ssd_p = 1; sldt.ssd_long = 0; sldt.ssd_def32 = 0; sldt.ssd_gran = 0; mtx_lock(&dt_lock); pldt = mdp->md_ldt; if (pldt != NULL && !force) { pmap_pti_remove_kva(sva, sva + sz); kmem_free(new_ldt->ldt_base, sz); free(new_ldt, M_SUBPROC); return (pldt); } if (pldt != NULL) { bcopy(pldt->ldt_base, new_ldt->ldt_base, max_ldt_segment * sizeof(struct user_segment_descriptor)); user_ldt_derefl(pldt); } critical_enter(); ssdtosyssd(&sldt, &p->p_md.md_ldt_sd); atomic_thread_fence_rel(); mdp->md_ldt = new_ldt; critical_exit(); smp_rendezvous(NULL, set_user_ldt_rv, NULL, p); return (mdp->md_ldt); } void user_ldt_free(struct thread *td) { struct proc *p = td->td_proc; struct mdproc *mdp = &p->p_md; struct proc_ldt *pldt; mtx_lock(&dt_lock); if ((pldt = mdp->md_ldt) == NULL) { mtx_unlock(&dt_lock); return; } critical_enter(); mdp->md_ldt = NULL; atomic_thread_fence_rel(); bzero(&mdp->md_ldt_sd, sizeof(mdp->md_ldt_sd)); if (td == curthread) lldt(GSEL(GNULL_SEL, SEL_KPL)); critical_exit(); user_ldt_deref(pldt); } static void user_ldt_derefl(struct proc_ldt *pldt) { vm_offset_t sva; vm_size_t sz; if (--pldt->ldt_refcnt == 0) { sva = (vm_offset_t)pldt->ldt_base; sz = max_ldt_segment * sizeof(struct user_segment_descriptor); pmap_pti_remove_kva(sva, sva + sz); kmem_free(pldt->ldt_base, sz); free(pldt, M_SUBPROC); } } static void user_ldt_deref(struct proc_ldt *pldt) { mtx_assert(&dt_lock, MA_OWNED); user_ldt_derefl(pldt); mtx_unlock(&dt_lock); } /* * Note for the authors of compat layers (linux, etc): copyout() in * the function below is not a problem since it presents data in * arch-specific format (i.e. i386-specific in this case), not in * the OS-specific one. */ int amd64_get_ldt(struct thread *td, struct i386_ldt_args *uap) { struct proc_ldt *pldt; struct user_segment_descriptor *lp; uint64_t *data; u_int i, num; int error; #ifdef DEBUG printf("amd64_get_ldt: start=%u num=%u descs=%p\n", uap->start, uap->num, (void *)uap->descs); #endif pldt = td->td_proc->p_md.md_ldt; if (pldt == NULL || uap->start >= max_ldt_segment || uap->num == 0) { td->td_retval[0] = 0; return (0); } num = min(uap->num, max_ldt_segment - uap->start); lp = &((struct user_segment_descriptor *)(pldt->ldt_base))[uap->start]; data = malloc(num * sizeof(struct user_segment_descriptor), M_TEMP, M_WAITOK); mtx_lock(&dt_lock); for (i = 0; i < num; i++) data[i] = ((volatile uint64_t *)lp)[i]; mtx_unlock(&dt_lock); error = copyout(data, uap->descs, num * sizeof(struct user_segment_descriptor)); free(data, M_TEMP); if (error == 0) td->td_retval[0] = num; return (error); } int amd64_set_ldt(struct thread *td, struct i386_ldt_args *uap, struct user_segment_descriptor *descs) { struct mdproc *mdp; struct proc_ldt *pldt; struct user_segment_descriptor *dp; struct proc *p; u_int largest_ld, i; int error; #ifdef DEBUG printf("amd64_set_ldt: start=%u num=%u descs=%p\n", uap->start, uap->num, (void *)uap->descs); #endif mdp = &td->td_proc->p_md; error = 0; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); p = td->td_proc; if (descs == NULL) { /* Free descriptors */ if (uap->start == 0 && uap->num == 0) uap->num = max_ldt_segment; if (uap->num == 0) return (EINVAL); if ((pldt = mdp->md_ldt) == NULL || uap->start >= max_ldt_segment) return (0); largest_ld = uap->start + uap->num; if (largest_ld > max_ldt_segment) largest_ld = max_ldt_segment; if (largest_ld < uap->start) return (EINVAL); mtx_lock(&dt_lock); for (i = uap->start; i < largest_ld; i++) ((volatile uint64_t *)(pldt->ldt_base))[i] = 0; mtx_unlock(&dt_lock); return (0); } if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) { /* verify range of descriptors to modify */ largest_ld = uap->start + uap->num; if (uap->start >= max_ldt_segment || largest_ld > max_ldt_segment || largest_ld < uap->start) return (EINVAL); } /* Check descriptors for access violations */ for (i = 0; i < uap->num; i++) { dp = &descs[i]; switch (dp->sd_type) { case SDT_SYSNULL: /* system null */ dp->sd_p = 0; break; case SDT_SYS286TSS: case SDT_SYSLDT: case SDT_SYS286BSY: case SDT_SYS286CGT: case SDT_SYSTASKGT: case SDT_SYS286IGT: case SDT_SYS286TGT: case SDT_SYSNULL2: case SDT_SYSTSS: case SDT_SYSNULL3: case SDT_SYSBSY: case SDT_SYSCGT: case SDT_SYSNULL4: case SDT_SYSIGT: case SDT_SYSTGT: return (EACCES); /* memory segment types */ case SDT_MEMEC: /* memory execute only conforming */ case SDT_MEMEAC: /* memory execute only accessed conforming */ case SDT_MEMERC: /* memory execute read conforming */ case SDT_MEMERAC: /* memory execute read accessed conforming */ /* Must be "present" if executable and conforming. */ if (dp->sd_p == 0) return (EACCES); break; case SDT_MEMRO: /* memory read only */ case SDT_MEMROA: /* memory read only accessed */ case SDT_MEMRW: /* memory read write */ case SDT_MEMRWA: /* memory read write accessed */ case SDT_MEMROD: /* memory read only expand dwn limit */ case SDT_MEMRODA: /* memory read only expand dwn lim accessed */ case SDT_MEMRWD: /* memory read write expand dwn limit */ case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */ case SDT_MEME: /* memory execute only */ case SDT_MEMEA: /* memory execute only accessed */ case SDT_MEMER: /* memory execute read */ case SDT_MEMERA: /* memory execute read accessed */ break; default: return(EINVAL); } /* Only user (ring-3) descriptors may be present. */ if ((dp->sd_p != 0) && (dp->sd_dpl != SEL_UPL)) return (EACCES); } if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) { /* Allocate a free slot */ mtx_lock(&dt_lock); pldt = user_ldt_alloc(p, 0); if (pldt == NULL) { mtx_unlock(&dt_lock); return (ENOMEM); } /* * start scanning a bit up to leave room for NVidia and * Wine, which still user the "Blat" method of allocation. */ i = 16; dp = &((struct user_segment_descriptor *)(pldt->ldt_base))[i]; for (; i < max_ldt_segment; ++i, ++dp) { if (dp->sd_type == SDT_SYSNULL) break; } if (i >= max_ldt_segment) { mtx_unlock(&dt_lock); return (ENOSPC); } uap->start = i; error = amd64_set_ldt_data(td, i, 1, descs); mtx_unlock(&dt_lock); } else { largest_ld = uap->start + uap->num; if (largest_ld > max_ldt_segment) return (EINVAL); mtx_lock(&dt_lock); if (user_ldt_alloc(p, 0) != NULL) { error = amd64_set_ldt_data(td, uap->start, uap->num, descs); } mtx_unlock(&dt_lock); } if (error == 0) td->td_retval[0] = uap->start; return (error); } int amd64_set_ldt_data(struct thread *td, int start, int num, struct user_segment_descriptor *descs) { struct mdproc *mdp; struct proc_ldt *pldt; volatile uint64_t *dst, *src; int i; mtx_assert(&dt_lock, MA_OWNED); mdp = &td->td_proc->p_md; pldt = mdp->md_ldt; dst = (volatile uint64_t *)(pldt->ldt_base); src = (volatile uint64_t *)descs; for (i = 0; i < num; i++) dst[start + i] = src[i]; return (0); } diff --git a/sys/amd64/amd64/vm_machdep.c b/sys/amd64/amd64/vm_machdep.c index 2b280e960b2e..4001f40554af 100644 --- a/sys/amd64/amd64/vm_machdep.c +++ b/sys/amd64/amd64/vm_machdep.c @@ -1,682 +1,682 @@ /*- * SPDX-License-Identifier: BSD-4-Clause * * Copyright (c) 1982, 1986 The Regents of the University of California. * Copyright (c) 1989, 1990 William Jolitz * Copyright (c) 1994 John Dyson * All rights reserved. * * This code is derived from software contributed to Berkeley by * the Systems Programming Group of the University of Utah Computer * Science Department, and 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. * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ */ #include #include "opt_isa.h" #include "opt_cpu.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 #include _Static_assert(OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf), "OFFSETOF_MONITORBUF does not correspond with offset of pc_monitorbuf."); void set_top_of_stack_td(struct thread *td) { td->td_md.md_stack_base = td->td_kstack + td->td_kstack_pages * PAGE_SIZE; } struct savefpu * get_pcb_user_save_td(struct thread *td) { KASSERT(((vm_offset_t)td->td_md.md_usr_fpu_save % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area ptr %p td %p", td->td_md.md_usr_fpu_save, td)); return (td->td_md.md_usr_fpu_save); } struct pcb * get_pcb_td(struct thread *td) { return (&td->td_md.md_pcb); } struct savefpu * get_pcb_user_save_pcb(struct pcb *pcb) { struct thread *td; td = __containerof(pcb, struct thread, td_md.md_pcb); return (get_pcb_user_save_td(td)); } void * alloc_fpusave(int flags) { void *res; struct savefpu_ymm *sf; res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags); if (use_xsave) { sf = (struct savefpu_ymm *)res; bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd)); sf->sv_xstate.sx_hd.xstate_bv = xsave_mask; } return (res); } /* * Common code shared between cpu_fork() and cpu_copy_thread() for * initializing a thread. */ static void copy_thread(struct thread *td1, struct thread *td2) { struct pcb *pcb2; pcb2 = td2->td_pcb; /* Ensure that td1's pcb is up to date for user threads. */ if ((td2->td_pflags & TDP_KTHREAD) == 0) { MPASS(td1 == curthread); fpuexit(td1); update_pcb_bases(td1->td_pcb); } /* Copy td1's pcb */ bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); /* Properly initialize pcb_save */ pcb2->pcb_save = get_pcb_user_save_pcb(pcb2); /* Kernel threads start with clean FPU and segment bases. */ if ((td2->td_pflags & TDP_KTHREAD) != 0) { - pcb2->pcb_fsbase = 0; + pcb2->pcb_fsbase = pcb2->pcb_tlsbase = 0; pcb2->pcb_gsbase = 0; clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE | PCB_KERNFPU | PCB_KERNFPU_THR); } else { MPASS((pcb2->pcb_flags & (PCB_KERNFPU | PCB_KERNFPU_THR)) == 0); bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2), cpu_max_ext_state_size); } td2->td_frame = (struct trapframe *)td2->td_md.md_stack_base - 1; /* * Set registers for trampoline to user mode. Leave space for the * return address on stack. These are the kernel mode register values. */ pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */ pcb2->pcb_rbp = 0; pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *); pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */ pcb2->pcb_rip = (register_t)fork_trampoline; /*- * pcb2->pcb_dr*: cloned above. * pcb2->pcb_savefpu: cloned above. * pcb2->pcb_flags: cloned above. * pcb2->pcb_onfault: cloned above (always NULL here?). - * pcb2->pcb_[fg]sbase: cloned above + * pcb2->pcb_[f,g,tls]sbase: cloned above */ pcb2->pcb_tssp = NULL; /* Setup to release spin count in fork_exit(). */ td2->td_md.md_spinlock_count = 1; td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I; pmap_thread_init_invl_gen(td2); /* * Copy the trap frame for the return to user mode as if from a syscall. * This copies most of the user mode register values. Some of these * registers are rewritten by cpu_set_upcall() and linux_set_upcall(). */ if ((td1->td_proc->p_flag & P_KPROC) == 0) { bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); /* * If the current thread has the trap bit set (i.e. a debugger * had single stepped the process to the system call), we need * to clear the trap flag from the new frame. Otherwise, the new * thread will receive a (likely unexpected) SIGTRAP when it * executes the first instruction after returning to userland. */ td2->td_frame->tf_rflags &= ~PSL_T; } } /* * Finish a fork operation, with process p2 nearly set up. * Copy and update the pcb, set up the stack so that the child * ready to run and return to user mode. */ void cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags) { struct proc *p1; struct pcb *pcb2; struct mdproc *mdp1, *mdp2; struct proc_ldt *pldt; p1 = td1->td_proc; if ((flags & RFPROC) == 0) { if ((flags & RFMEM) == 0) { /* unshare user LDT */ mdp1 = &p1->p_md; mtx_lock(&dt_lock); if ((pldt = mdp1->md_ldt) != NULL && pldt->ldt_refcnt > 1 && user_ldt_alloc(p1, 1) == NULL) panic("could not copy LDT"); mtx_unlock(&dt_lock); } return; } /* Point the stack and pcb to the actual location */ set_top_of_stack_td(td2); td2->td_pcb = pcb2 = get_pcb_td(td2); copy_thread(td1, td2); /* Reset debug registers in the new process */ x86_clear_dbregs(pcb2); /* Point mdproc and then copy over p1's contents */ mdp2 = &p2->p_md; bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); /* Set child return values. */ p2->p_sysent->sv_set_fork_retval(td2); /* As on i386, do not copy io permission bitmap. */ pcb2->pcb_tssp = NULL; /* New segment registers. */ set_pcb_flags_raw(pcb2, PCB_FULL_IRET); /* Copy the LDT, if necessary. */ mdp1 = &td1->td_proc->p_md; mdp2 = &p2->p_md; if (mdp1->md_ldt == NULL) { mdp2->md_ldt = NULL; return; } mtx_lock(&dt_lock); if (mdp1->md_ldt != NULL) { if (flags & RFMEM) { mdp1->md_ldt->ldt_refcnt++; mdp2->md_ldt = mdp1->md_ldt; bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct system_segment_descriptor)); } else { mdp2->md_ldt = NULL; mdp2->md_ldt = user_ldt_alloc(p2, 0); if (mdp2->md_ldt == NULL) panic("could not copy LDT"); amd64_set_ldt_data(td2, 0, max_ldt_segment, (struct user_segment_descriptor *) mdp1->md_ldt->ldt_base); } } else mdp2->md_ldt = NULL; mtx_unlock(&dt_lock); /* * Now, cpu_switch() can schedule the new process. * pcb_rsp is loaded pointing to the cpu_switch() stack frame * containing the return address when exiting cpu_switch. * This will normally be to fork_trampoline(), which will have * %rbx loaded with the new proc's pointer. fork_trampoline() * will set up a stack to call fork_return(p, frame); to complete * the return to user-mode. */ } void x86_set_fork_retval(struct thread *td) { struct trapframe *frame = td->td_frame; frame->tf_rax = 0; /* Child returns zero */ frame->tf_rflags &= ~PSL_C; /* success */ frame->tf_rdx = 1; /* System V emulation */ } /* * Intercept the return address from a freshly forked process that has NOT * been scheduled yet. * * This is needed to make kernel threads stay in kernel mode. */ void cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg) { /* * Note that the trap frame follows the args, so the function * is really called like this: func(arg, frame); */ td->td_pcb->pcb_r12 = (long) func; /* function */ td->td_pcb->pcb_rbx = (long) arg; /* first arg */ } void cpu_exit(struct thread *td) { /* * If this process has a custom LDT, release it. */ if (td->td_proc->p_md.md_ldt != NULL) user_ldt_free(td); } void cpu_thread_exit(struct thread *td) { struct pcb *pcb; critical_enter(); if (td == PCPU_GET(fpcurthread)) fpudrop(); critical_exit(); pcb = td->td_pcb; /* Disable any hardware breakpoints. */ if (pcb->pcb_flags & PCB_DBREGS) { reset_dbregs(); clear_pcb_flags(pcb, PCB_DBREGS); } } void cpu_thread_clean(struct thread *td) { struct pcb *pcb; pcb = td->td_pcb; /* * Clean TSS/iomap */ if (pcb->pcb_tssp != NULL) { pmap_pti_remove_kva((vm_offset_t)pcb->pcb_tssp, (vm_offset_t)pcb->pcb_tssp + ctob(IOPAGES + 1)); kmem_free(pcb->pcb_tssp, ctob(IOPAGES + 1)); pcb->pcb_tssp = NULL; } } void cpu_thread_alloc(struct thread *td) { struct pcb *pcb; struct xstate_hdr *xhdr; set_top_of_stack_td(td); td->td_pcb = pcb = get_pcb_td(td); td->td_frame = (struct trapframe *)td->td_md.md_stack_base - 1; td->td_md.md_usr_fpu_save = fpu_save_area_alloc(); pcb->pcb_save = get_pcb_user_save_pcb(pcb); if (use_xsave) { xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1); bzero(xhdr, sizeof(*xhdr)); xhdr->xstate_bv = xsave_mask; } } void cpu_thread_free(struct thread *td) { cpu_thread_clean(td); fpu_save_area_free(td->td_md.md_usr_fpu_save); td->td_md.md_usr_fpu_save = NULL; } bool cpu_exec_vmspace_reuse(struct proc *p, vm_map_t map) { return (((curproc->p_md.md_flags & P_MD_KPTI) != 0) == (vm_map_pmap(map)->pm_ucr3 != PMAP_NO_CR3)); } static void cpu_procctl_kpti_ctl(struct proc *p, int val) { if (pti && val == PROC_KPTI_CTL_ENABLE_ON_EXEC) p->p_md.md_flags |= P_MD_KPTI; if (val == PROC_KPTI_CTL_DISABLE_ON_EXEC) p->p_md.md_flags &= ~P_MD_KPTI; } static void cpu_procctl_kpti_status(struct proc *p, int *val) { *val = (p->p_md.md_flags & P_MD_KPTI) != 0 ? PROC_KPTI_CTL_ENABLE_ON_EXEC: PROC_KPTI_CTL_DISABLE_ON_EXEC; if (vmspace_pmap(p->p_vmspace)->pm_ucr3 != PMAP_NO_CR3) *val |= PROC_KPTI_STATUS_ACTIVE; } static int cpu_procctl_la_ctl(struct proc *p, int val) { int error; error = 0; switch (val) { case PROC_LA_CTL_LA48_ON_EXEC: p->p_md.md_flags |= P_MD_LA48; p->p_md.md_flags &= ~P_MD_LA57; break; case PROC_LA_CTL_LA57_ON_EXEC: if (la57) { p->p_md.md_flags &= ~P_MD_LA48; p->p_md.md_flags |= P_MD_LA57; } else { error = ENOTSUP; } break; case PROC_LA_CTL_DEFAULT_ON_EXEC: p->p_md.md_flags &= ~(P_MD_LA48 | P_MD_LA57); break; } return (error); } static void cpu_procctl_la_status(struct proc *p, int *val) { int res; if ((p->p_md.md_flags & P_MD_LA48) != 0) res = PROC_LA_CTL_LA48_ON_EXEC; else if ((p->p_md.md_flags & P_MD_LA57) != 0) res = PROC_LA_CTL_LA57_ON_EXEC; else res = PROC_LA_CTL_DEFAULT_ON_EXEC; if (p->p_sysent->sv_maxuser == VM_MAXUSER_ADDRESS_LA48) res |= PROC_LA_STATUS_LA48; else res |= PROC_LA_STATUS_LA57; *val = res; } int cpu_procctl(struct thread *td, int idtype, id_t id, int com, void *data) { struct proc *p; int error, val; switch (com) { case PROC_KPTI_CTL: case PROC_KPTI_STATUS: case PROC_LA_CTL: case PROC_LA_STATUS: if (idtype != P_PID) { error = EINVAL; break; } if (com == PROC_KPTI_CTL) { /* sad but true and not a joke */ error = priv_check(td, PRIV_IO); if (error != 0) break; } if (com == PROC_KPTI_CTL || com == PROC_LA_CTL) { error = copyin(data, &val, sizeof(val)); if (error != 0) break; } if (com == PROC_KPTI_CTL && val != PROC_KPTI_CTL_ENABLE_ON_EXEC && val != PROC_KPTI_CTL_DISABLE_ON_EXEC) { error = EINVAL; break; } if (com == PROC_LA_CTL && val != PROC_LA_CTL_LA48_ON_EXEC && val != PROC_LA_CTL_LA57_ON_EXEC && val != PROC_LA_CTL_DEFAULT_ON_EXEC) { error = EINVAL; break; } error = pget(id, PGET_CANSEE | PGET_NOTWEXIT | PGET_NOTID, &p); if (error != 0) break; switch (com) { case PROC_KPTI_CTL: cpu_procctl_kpti_ctl(p, val); break; case PROC_KPTI_STATUS: cpu_procctl_kpti_status(p, &val); break; case PROC_LA_CTL: error = cpu_procctl_la_ctl(p, val); break; case PROC_LA_STATUS: cpu_procctl_la_status(p, &val); break; } PROC_UNLOCK(p); if (com == PROC_KPTI_STATUS || com == PROC_LA_STATUS) error = copyout(&val, data, sizeof(val)); break; default: error = EINVAL; break; } return (error); } void cpu_set_syscall_retval(struct thread *td, int error) { struct trapframe *frame; frame = td->td_frame; if (__predict_true(error == 0)) { frame->tf_rax = td->td_retval[0]; frame->tf_rdx = td->td_retval[1]; frame->tf_rflags &= ~PSL_C; return; } switch (error) { 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. * %r10 (which was holding the value of %rcx) is restored * for the next iteration. * %r10 restore is only required for freebsd/amd64 processes, * but shall be innocent for any ia32 ABI. * * Require full context restore to get the arguments * in the registers reloaded at return to usermode. */ frame->tf_rip -= frame->tf_err; frame->tf_r10 = frame->tf_rcx; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); break; case EJUSTRETURN: break; default: frame->tf_rax = error; frame->tf_rflags |= PSL_C; break; } } /* * Initialize machine state, mostly pcb and trap frame for a new * thread, about to return to userspace. Put enough state in the new * thread's PCB to get it to go back to the fork_return(), which * finalizes the thread state and handles peculiarities of the first * return to userspace for the new thread. */ void cpu_copy_thread(struct thread *td, struct thread *td0) { copy_thread(td0, td); set_pcb_flags_raw(td->td_pcb, PCB_FULL_IRET); } /* * Set that machine state for performing an upcall that starts * the entry function with the given argument. */ int cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg, stack_t *stack) { #ifdef COMPAT_FREEBSD32 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { /* * Set the trap frame to point at the beginning of the entry * function. */ td->td_frame->tf_rbp = 0; td->td_frame->tf_rsp = (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4; td->td_frame->tf_rip = (uintptr_t)entry; /* Return address sentinel value to stop stack unwinding. */ if (suword32((void *)td->td_frame->tf_rsp, 0) != 0) return (EFAULT); /* Pass the argument to the entry point. */ if (suword32( (void *)(td->td_frame->tf_rsp + sizeof(int32_t)), (uint32_t)(uintptr_t)arg) != 0) return (EFAULT); return (0); } #endif /* * Set the trap frame to point at the beginning of the uts * function. */ td->td_frame->tf_rbp = 0; td->td_frame->tf_rsp = ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f; td->td_frame->tf_rsp -= 8; td->td_frame->tf_rip = (register_t)entry; td->td_frame->tf_ds = _udatasel; td->td_frame->tf_es = _udatasel; td->td_frame->tf_fs = _ufssel; td->td_frame->tf_gs = _ugssel; td->td_frame->tf_flags = TF_HASSEGS; /* Return address sentinel value to stop stack unwinding. */ if (suword((void *)td->td_frame->tf_rsp, 0) != 0) return (EFAULT); /* Pass the argument to the entry point. */ td->td_frame->tf_rdi = (register_t)arg; return (0); } int cpu_set_user_tls(struct thread *td, void *tls_base) { struct pcb *pcb; if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS) return (EINVAL); pcb = td->td_pcb; - set_pcb_flags(pcb, PCB_FULL_IRET); + set_pcb_flags(pcb, PCB_FULL_IRET | PCB_TLSBASE); #ifdef COMPAT_FREEBSD32 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { pcb->pcb_gsbase = (register_t)tls_base; return (0); } #endif - pcb->pcb_fsbase = (register_t)tls_base; + pcb->pcb_fsbase = pcb->pcb_tlsbase = (register_t)tls_base; return (0); } void cpu_update_pcb(struct thread *td) { MPASS(td == curthread); update_pcb_bases(td->td_pcb); } diff --git a/sys/amd64/ia32/ia32_signal.c b/sys/amd64/ia32/ia32_signal.c index dea12c5346a3..54e170450dba 100644 --- a/sys/amd64/ia32/ia32_signal.c +++ b/sys/amd64/ia32/ia32_signal.c @@ -1,961 +1,962 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2003 Peter Wemm * 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. 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 #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 "vdso_ia32_offsets.h" extern const char _binary_elf_vdso32_so_1_start[]; extern const char _binary_elf_vdso32_so_1_end[]; extern char _binary_elf_vdso32_so_1_size; #ifdef COMPAT_FREEBSD4 static void freebsd4_ia32_sendsig(sig_t, ksiginfo_t *, sigset_t *); #endif static void ia32_get_fpcontext(struct thread *td, struct ia32_mcontext *mcp, char **xfpusave, size_t *xfpusave_len) { /* * XXX Format of 64bit and 32bit FXSAVE areas differs. FXSAVE * in 32bit mode saves %cs and %ds, while on 64bit it saves * 64bit instruction and data pointers. Ignore the difference * for now, it should be irrelevant for most applications. */ 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_IA32_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 ia32_set_fpcontext(struct thread *td, struct ia32_mcontext *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); } /* * Get machine context. */ static int ia32_get_mcontext(struct thread *td, struct ia32_mcontext *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); /* Entry into kernel always sets TF_HASSEGS */ mcp->mc_gs = tp->tf_gs; mcp->mc_fs = tp->tf_fs; mcp->mc_es = tp->tf_es; mcp->mc_ds = tp->tf_ds; mcp->mc_edi = tp->tf_rdi; mcp->mc_esi = tp->tf_rsi; mcp->mc_ebp = tp->tf_rbp; mcp->mc_isp = tp->tf_rsp; mcp->mc_eflags = tp->tf_rflags; if (flags & GET_MC_CLEAR_RET) { mcp->mc_eax = 0; mcp->mc_edx = 0; mcp->mc_eflags &= ~PSL_C; } else { mcp->mc_eax = tp->tf_rax; mcp->mc_edx = tp->tf_rdx; } mcp->mc_ebx = tp->tf_rbx; mcp->mc_ecx = tp->tf_rcx; mcp->mc_eip = tp->tf_rip; mcp->mc_cs = tp->tf_cs; mcp->mc_esp = tp->tf_rsp; mcp->mc_ss = tp->tf_ss; mcp->mc_len = sizeof(*mcp); mcp->mc_flags = tp->tf_flags; ia32_get_fpcontext(td, mcp, NULL, 0); mcp->mc_fsbase = pcb->pcb_fsbase; mcp->mc_gsbase = pcb->pcb_gsbase; mcp->mc_xfpustate = 0; mcp->mc_xfpustate_len = 0; bzero(mcp->mc_spare2, sizeof(mcp->mc_spare2)); return (0); } /* * Set machine context. * * However, we don't set any but the user modifiable flags, and we won't * touch the cs selector. */ static int ia32_set_mcontext(struct thread *td, struct ia32_mcontext *mcp) { struct trapframe *tp; char *xfpustate; long rflags; int ret; tp = td->td_frame; if (mcp->mc_len != sizeof(*mcp)) return (EINVAL); rflags = (mcp->mc_eflags & PSL_USERCHANGE) | (tp->tf_rflags & ~PSL_USERCHANGE); if (mcp->mc_flags & _MC_IA32_HASFPXSTATE) { if (mcp->mc_xfpustate_len > cpu_max_ext_state_size - sizeof(struct savefpu)) return (EINVAL); xfpustate = (char *)fpu_save_area_alloc(); ret = copyin(PTRIN(mcp->mc_xfpustate), xfpustate, mcp->mc_xfpustate_len); if (ret != 0) { fpu_save_area_free((struct savefpu *)xfpustate); return (ret); } } else xfpustate = NULL; ret = ia32_set_fpcontext(td, mcp, xfpustate, mcp->mc_xfpustate_len); fpu_save_area_free((struct savefpu *)xfpustate); if (ret != 0) return (ret); tp->tf_gs = mcp->mc_gs; tp->tf_fs = mcp->mc_fs; tp->tf_es = mcp->mc_es; tp->tf_ds = mcp->mc_ds; tp->tf_flags = TF_HASSEGS; tp->tf_rdi = mcp->mc_edi; tp->tf_rsi = mcp->mc_esi; tp->tf_rbp = mcp->mc_ebp; tp->tf_rbx = mcp->mc_ebx; tp->tf_rdx = mcp->mc_edx; tp->tf_rcx = mcp->mc_ecx; tp->tf_rax = mcp->mc_eax; /* trapno, err */ tp->tf_rip = mcp->mc_eip; tp->tf_rflags = rflags; tp->tf_rsp = mcp->mc_esp; tp->tf_ss = mcp->mc_ss; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); return (0); } /* * The first two fields of a ucontext_t are the signal mask and * the machine context. The next field is uc_link; we want to * avoid destroying the link when copying out contexts. */ #define UC_COPY_SIZE offsetof(struct ia32_ucontext, uc_link) int freebsd32_getcontext(struct thread *td, struct freebsd32_getcontext_args *uap) { struct ia32_ucontext uc; int ret; if (uap->ucp == NULL) ret = EINVAL; else { bzero(&uc, sizeof(uc)); ia32_get_mcontext(td, &uc.uc_mcontext, GET_MC_CLEAR_RET); PROC_LOCK(td->td_proc); uc.uc_sigmask = td->td_sigmask; PROC_UNLOCK(td->td_proc); ret = copyout(&uc, uap->ucp, UC_COPY_SIZE); } return (ret); } int freebsd32_setcontext(struct thread *td, struct freebsd32_setcontext_args *uap) { struct ia32_ucontext uc; int ret; if (uap->ucp == NULL) ret = EINVAL; else { ret = copyin(uap->ucp, &uc, UC_COPY_SIZE); if (ret == 0) { ret = ia32_set_mcontext(td, &uc.uc_mcontext); if (ret == 0) { kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0); } } } return (ret == 0 ? EJUSTRETURN : ret); } int freebsd32_swapcontext(struct thread *td, struct freebsd32_swapcontext_args *uap) { struct ia32_ucontext uc; int ret; if (uap->oucp == NULL || uap->ucp == NULL) ret = EINVAL; else { bzero(&uc, sizeof(uc)); ia32_get_mcontext(td, &uc.uc_mcontext, GET_MC_CLEAR_RET); PROC_LOCK(td->td_proc); uc.uc_sigmask = td->td_sigmask; PROC_UNLOCK(td->td_proc); ret = copyout(&uc, uap->oucp, UC_COPY_SIZE); if (ret == 0) { ret = copyin(uap->ucp, &uc, UC_COPY_SIZE); if (ret == 0) { ret = ia32_set_mcontext(td, &uc.uc_mcontext); if (ret == 0) { kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0); } } } } return (ret == 0 ? EJUSTRETURN : ret); } /* * Send an interrupt to process. * * Stack is set up to allow sigcode stored * at top to call routine, followed by kcall * to sigreturn routine below. After sigreturn * resets the signal mask, the stack, and the * frame pointer, it returns to the user * specified pc, psl. */ #ifdef COMPAT_43 static void ia32_osendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct ia32_osigframe sf, *fp; struct proc *p; struct thread *td; struct sigacts *psp; struct trapframe *regs; int sig; int oonstack; td = curthread; 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); /* Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { fp = (struct ia32_osigframe *)((uintptr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - sizeof(sf)); td->td_sigstk.ss_flags |= SS_ONSTACK; } else fp = (struct ia32_osigframe *)regs->tf_rsp - 1; /* Build the argument list for the signal handler. */ sf.sf_signum = sig; sf.sf_scp = (register_t)&fp->sf_siginfo.si_sc; bzero(&sf.sf_siginfo, sizeof(sf.sf_siginfo)); if (SIGISMEMBER(psp->ps_siginfo, sig)) { /* Signal handler installed with SA_SIGINFO. */ sf.sf_arg2 = (register_t)&fp->sf_siginfo; sf.sf_siginfo.si_signo = sig; sf.sf_siginfo.si_code = ksi->ksi_code; sf.sf_ah = (uintptr_t)catcher; sf.sf_addr = 0; } else { /* Old FreeBSD-style arguments. */ sf.sf_arg2 = ksi->ksi_code; sf.sf_addr = (register_t)ksi->ksi_addr; sf.sf_ah = (uintptr_t)catcher; } mtx_unlock(&psp->ps_mtx); PROC_UNLOCK(p); /* Save most if not all of trap frame. */ sf.sf_siginfo.si_sc.sc_eax = regs->tf_rax; sf.sf_siginfo.si_sc.sc_ebx = regs->tf_rbx; sf.sf_siginfo.si_sc.sc_ecx = regs->tf_rcx; sf.sf_siginfo.si_sc.sc_edx = regs->tf_rdx; sf.sf_siginfo.si_sc.sc_esi = regs->tf_rsi; sf.sf_siginfo.si_sc.sc_edi = regs->tf_rdi; sf.sf_siginfo.si_sc.sc_cs = regs->tf_cs; sf.sf_siginfo.si_sc.sc_ds = regs->tf_ds; sf.sf_siginfo.si_sc.sc_ss = regs->tf_ss; sf.sf_siginfo.si_sc.sc_es = regs->tf_es; sf.sf_siginfo.si_sc.sc_fs = regs->tf_fs; sf.sf_siginfo.si_sc.sc_gs = regs->tf_gs; sf.sf_siginfo.si_sc.sc_isp = regs->tf_rsp; /* Build the signal context to be used by osigreturn(). */ sf.sf_siginfo.si_sc.sc_onstack = (oonstack) ? 1 : 0; SIG2OSIG(*mask, sf.sf_siginfo.si_sc.sc_mask); sf.sf_siginfo.si_sc.sc_esp = regs->tf_rsp; sf.sf_siginfo.si_sc.sc_ebp = regs->tf_rbp; sf.sf_siginfo.si_sc.sc_eip = regs->tf_rip; sf.sf_siginfo.si_sc.sc_eflags = regs->tf_rflags; sf.sf_siginfo.si_sc.sc_trapno = regs->tf_trapno; sf.sf_siginfo.si_sc.sc_err = regs->tf_err; /* * Copy the sigframe out to the user's stack. */ if (copyout(&sf, fp, sizeof(*fp)) != 0) { #ifdef DEBUG printf("process %ld has trashed its stack\n", (long)p->p_pid); #endif PROC_LOCK(p); sigexit(td, SIGILL); } regs->tf_rsp = (uintptr_t)fp; regs->tf_rip = PROC_PS_STRINGS(p) - (_binary_elf_vdso32_so_1_end - _binary_elf_vdso32_so_1_start) + VDSO_IA32_OSIGCODE_OFFSET; regs->tf_rflags &= ~(PSL_T | PSL_D); regs->tf_cs = _ucode32sel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; regs->tf_fs = _udatasel; regs->tf_ss = _udatasel; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } #endif #ifdef COMPAT_FREEBSD4 static void freebsd4_ia32_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct ia32_freebsd4_sigframe sf, *sfp; struct __siginfo32 siginfo; struct proc *p; struct thread *td; struct sigacts *psp; struct trapframe *regs; int oonstack; int sig; td = curthread; p = td->td_proc; siginfo_to_siginfo32(&ksi->ksi_info, &siginfo); PROC_LOCK_ASSERT(p, MA_OWNED); sig = siginfo.si_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.ss_sp = (uintptr_t)td->td_sigstk.ss_sp; sf.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size; 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; sf.sf_uc.uc_mcontext.mc_edi = regs->tf_rdi; sf.sf_uc.uc_mcontext.mc_esi = regs->tf_rsi; sf.sf_uc.uc_mcontext.mc_ebp = regs->tf_rbp; sf.sf_uc.uc_mcontext.mc_isp = regs->tf_rsp; /* XXX */ sf.sf_uc.uc_mcontext.mc_ebx = regs->tf_rbx; sf.sf_uc.uc_mcontext.mc_edx = regs->tf_rdx; sf.sf_uc.uc_mcontext.mc_ecx = regs->tf_rcx; sf.sf_uc.uc_mcontext.mc_eax = regs->tf_rax; sf.sf_uc.uc_mcontext.mc_trapno = regs->tf_trapno; sf.sf_uc.uc_mcontext.mc_err = regs->tf_err; sf.sf_uc.uc_mcontext.mc_eip = regs->tf_rip; sf.sf_uc.uc_mcontext.mc_cs = regs->tf_cs; sf.sf_uc.uc_mcontext.mc_eflags = regs->tf_rflags; sf.sf_uc.uc_mcontext.mc_esp = regs->tf_rsp; sf.sf_uc.uc_mcontext.mc_ss = regs->tf_ss; sf.sf_uc.uc_mcontext.mc_ds = regs->tf_ds; sf.sf_uc.uc_mcontext.mc_es = regs->tf_es; sf.sf_uc.uc_mcontext.mc_fs = regs->tf_fs; sf.sf_uc.uc_mcontext.mc_gs = regs->tf_gs; bzero(sf.sf_uc.uc_mcontext.mc_fpregs, sizeof(sf.sf_uc.uc_mcontext.mc_fpregs)); bzero(sf.sf_uc.uc_mcontext.__spare__, sizeof(sf.sf_uc.uc_mcontext.__spare__)); bzero(sf.sf_uc.__spare__, sizeof(sf.sf_uc.__spare__)); /* Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { sfp = (struct ia32_freebsd4_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - sizeof(sf)); } else sfp = (struct ia32_freebsd4_sigframe *)regs->tf_rsp - 1; PROC_UNLOCK(p); /* Build the argument list for the signal handler. */ sf.sf_signum = sig; sf.sf_ucontext = (register_t)&sfp->sf_uc; bzero(&sf.sf_si, sizeof(sf.sf_si)); if (SIGISMEMBER(psp->ps_siginfo, sig)) { /* Signal handler installed with SA_SIGINFO. */ sf.sf_siginfo = (u_int32_t)(uintptr_t)&sfp->sf_si; sf.sf_ah = (u_int32_t)(uintptr_t)catcher; /* Fill in POSIX parts */ sf.sf_si = siginfo; sf.sf_si.si_signo = sig; } else { /* Old FreeBSD-style arguments. */ sf.sf_siginfo = siginfo.si_code; sf.sf_addr = (u_int32_t)siginfo.si_addr; sf.sf_ah = (u_int32_t)(uintptr_t)catcher; } mtx_unlock(&psp->ps_mtx); /* * Copy the sigframe out to the user's stack. */ if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { #ifdef DEBUG printf("process %ld has trashed its stack\n", (long)p->p_pid); #endif PROC_LOCK(p); sigexit(td, SIGILL); } regs->tf_rsp = (uintptr_t)sfp; regs->tf_rip = PROC_SIGCODE(p) + VDSO_FREEBSD4_IA32_SIGCODE_OFFSET - VDSO_IA32_SIGCODE_OFFSET; regs->tf_rflags &= ~(PSL_T | PSL_D); regs->tf_cs = _ucode32sel; regs->tf_ss = _udatasel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); /* leave user %fs and %gs untouched */ PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } #endif /* COMPAT_FREEBSD4 */ void ia32_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct ia32_sigframe sf, *sfp; struct __siginfo32 siginfo; struct proc *p; struct thread *td; struct sigacts *psp; char *sp; struct trapframe *regs; char *xfpusave; size_t xfpusave_len; int oonstack; int sig; siginfo_to_siginfo32(&ksi->ksi_info, &siginfo); td = curthread; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); sig = siginfo.si_signo; psp = p->p_sigacts; #ifdef COMPAT_FREEBSD4 if (SIGISMEMBER(psp->ps_freebsd4, sig)) { freebsd4_ia32_sendsig(catcher, ksi, mask); return; } #endif #ifdef COMPAT_43 if (SIGISMEMBER(psp->ps_osigset, sig)) { ia32_osendsig(catcher, ksi, mask); return; } #endif 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.ss_sp = (uintptr_t)td->td_sigstk.ss_sp; sf.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size; 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; sf.sf_uc.uc_mcontext.mc_edi = regs->tf_rdi; sf.sf_uc.uc_mcontext.mc_esi = regs->tf_rsi; sf.sf_uc.uc_mcontext.mc_ebp = regs->tf_rbp; sf.sf_uc.uc_mcontext.mc_isp = regs->tf_rsp; /* XXX */ sf.sf_uc.uc_mcontext.mc_ebx = regs->tf_rbx; sf.sf_uc.uc_mcontext.mc_edx = regs->tf_rdx; sf.sf_uc.uc_mcontext.mc_ecx = regs->tf_rcx; sf.sf_uc.uc_mcontext.mc_eax = regs->tf_rax; sf.sf_uc.uc_mcontext.mc_trapno = regs->tf_trapno; sf.sf_uc.uc_mcontext.mc_err = regs->tf_err; sf.sf_uc.uc_mcontext.mc_eip = regs->tf_rip; sf.sf_uc.uc_mcontext.mc_cs = regs->tf_cs; sf.sf_uc.uc_mcontext.mc_eflags = regs->tf_rflags; sf.sf_uc.uc_mcontext.mc_esp = regs->tf_rsp; sf.sf_uc.uc_mcontext.mc_ss = regs->tf_ss; sf.sf_uc.uc_mcontext.mc_ds = regs->tf_ds; sf.sf_uc.uc_mcontext.mc_es = regs->tf_es; sf.sf_uc.uc_mcontext.mc_fs = regs->tf_fs; sf.sf_uc.uc_mcontext.mc_gs = regs->tf_gs; sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext); /* magic */ ia32_get_fpcontext(td, &sf.sf_uc.uc_mcontext, &xfpusave, &xfpusave_len); sf.sf_uc.uc_mcontext.mc_fsbase = td->td_pcb->pcb_fsbase; sf.sf_uc.uc_mcontext.mc_gsbase = td->td_pcb->pcb_gsbase; /* 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; else sp = (char *)regs->tf_rsp; if (xfpusave != NULL) { sp -= xfpusave_len; sp = (char *)((unsigned long)sp & ~0x3Ful); sf.sf_uc.uc_mcontext.mc_xfpustate = (register_t)sp; } sp -= sizeof(sf); /* Align to 16 bytes. */ sfp = (struct ia32_sigframe *)((uintptr_t)sp & ~0xF); PROC_UNLOCK(p); /* Build the argument list for the signal handler. */ sf.sf_signum = sig; sf.sf_ucontext = (register_t)&sfp->sf_uc; bzero(&sf.sf_si, sizeof(sf.sf_si)); if (SIGISMEMBER(psp->ps_siginfo, sig)) { /* Signal handler installed with SA_SIGINFO. */ sf.sf_siginfo = (u_int32_t)(uintptr_t)&sfp->sf_si; sf.sf_ah = (u_int32_t)(uintptr_t)catcher; /* Fill in POSIX parts */ sf.sf_si = siginfo; sf.sf_si.si_signo = sig; } else { /* Old FreeBSD-style arguments. */ sf.sf_siginfo = siginfo.si_code; sf.sf_addr = (u_int32_t)siginfo.si_addr; sf.sf_ah = (u_int32_t)(uintptr_t)catcher; } mtx_unlock(&psp->ps_mtx); /* * Copy the sigframe out to the user's stack. */ if (copyout(&sf, sfp, sizeof(*sfp)) != 0 || (xfpusave != NULL && copyout(xfpusave, PTRIN(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 PROC_LOCK(p); sigexit(td, SIGILL); } fpstate_drop(td); regs->tf_rsp = (uintptr_t)sfp; regs->tf_rip = PROC_SIGCODE(p); regs->tf_rflags &= ~(PSL_T | PSL_D); regs->tf_cs = _ucode32sel; regs->tf_ss = _udatasel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); /* XXXKIB leave user %fs and %gs untouched */ 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. */ #ifdef COMPAT_43 int ofreebsd32_sigreturn(struct thread *td, struct ofreebsd32_sigreturn_args *uap) { struct ia32_osigcontext sc, *scp; struct trapframe *regs; int eflags, error; ksiginfo_t ksi; regs = td->td_frame; error = copyin(uap->sigcntxp, &sc, sizeof(sc)); if (error != 0) return (error); scp = ≻ eflags = scp->sc_eflags; if (!EFL_SECURE(eflags, regs->tf_rflags)) { return (EINVAL); } if (!CS_SECURE(scp->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); } regs->tf_ds = scp->sc_ds; regs->tf_es = scp->sc_es; regs->tf_fs = scp->sc_fs; regs->tf_gs = scp->sc_gs; regs->tf_rax = scp->sc_eax; regs->tf_rbx = scp->sc_ebx; regs->tf_rcx = scp->sc_ecx; regs->tf_rdx = scp->sc_edx; regs->tf_rsi = scp->sc_esi; regs->tf_rdi = scp->sc_edi; regs->tf_cs = scp->sc_cs; regs->tf_ss = scp->sc_ss; regs->tf_rbp = scp->sc_ebp; regs->tf_rsp = scp->sc_esp; regs->tf_rip = scp->sc_eip; regs->tf_rflags = eflags; if (scp->sc_onstack & 1) td->td_sigstk.ss_flags |= SS_ONSTACK; else td->td_sigstk.ss_flags &= ~SS_ONSTACK; kern_sigprocmask(td, SIG_SETMASK, (sigset_t *)&scp->sc_mask, NULL, SIGPROCMASK_OLD); set_pcb_flags(td->td_pcb, PCB_FULL_IRET); return (EJUSTRETURN); } #endif #ifdef COMPAT_FREEBSD4 int freebsd4_freebsd32_sigreturn(struct thread *td, struct freebsd4_freebsd32_sigreturn_args *uap) { struct ia32_freebsd4_ucontext uc; struct trapframe *regs; struct ia32_freebsd4_ucontext *ucp; int cs, eflags, error; ksiginfo_t ksi; error = copyin(uap->sigcntxp, &uc, sizeof(uc)); if (error != 0) return (error); ucp = &uc; regs = td->td_frame; eflags = ucp->uc_mcontext.mc_eflags; /* * Don't allow users to change privileged or reserved flags. */ if (!EFL_SECURE(eflags, regs->tf_rflags)) { uprintf("pid %d (%s): freebsd4_freebsd32_sigreturn eflags = 0x%x\n", td->td_proc->p_pid, td->td_name, eflags); 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): freebsd4_sigreturn cs = 0x%x\n", td->td_proc->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); } regs->tf_rdi = ucp->uc_mcontext.mc_edi; regs->tf_rsi = ucp->uc_mcontext.mc_esi; regs->tf_rbp = ucp->uc_mcontext.mc_ebp; regs->tf_rbx = ucp->uc_mcontext.mc_ebx; regs->tf_rdx = ucp->uc_mcontext.mc_edx; regs->tf_rcx = ucp->uc_mcontext.mc_ecx; regs->tf_rax = ucp->uc_mcontext.mc_eax; regs->tf_trapno = ucp->uc_mcontext.mc_trapno; regs->tf_err = ucp->uc_mcontext.mc_err; regs->tf_rip = ucp->uc_mcontext.mc_eip; regs->tf_cs = cs; regs->tf_rflags = ucp->uc_mcontext.mc_eflags; regs->tf_rsp = ucp->uc_mcontext.mc_esp; regs->tf_ss = ucp->uc_mcontext.mc_ss; regs->tf_ds = ucp->uc_mcontext.mc_ds; regs->tf_es = ucp->uc_mcontext.mc_es; regs->tf_fs = ucp->uc_mcontext.mc_fs; regs->tf_gs = ucp->uc_mcontext.mc_gs; kern_sigprocmask(td, SIG_SETMASK, &ucp->uc_sigmask, NULL, 0); set_pcb_flags(td->td_pcb, PCB_FULL_IRET); return (EJUSTRETURN); } #endif /* COMPAT_FREEBSD4 */ int freebsd32_sigreturn(struct thread *td, struct freebsd32_sigreturn_args *uap) { struct ia32_ucontext uc; struct trapframe *regs; struct ia32_ucontext *ucp; char *xfpustate; size_t xfpustate_len; int cs, eflags, error, ret; ksiginfo_t ksi; error = copyin(uap->sigcntxp, &uc, sizeof(uc)); if (error != 0) return (error); ucp = &uc; regs = td->td_frame; eflags = ucp->uc_mcontext.mc_eflags; /* * Don't allow users to change privileged or reserved flags. */ if (!EFL_SECURE(eflags, regs->tf_rflags)) { uprintf("pid %d (%s): freebsd32_sigreturn eflags = 0x%x\n", td->td_proc->p_pid, td->td_name, eflags); 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", td->td_proc->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 ((ucp->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", td->td_proc->p_pid, td->td_name, xfpustate_len); return (EINVAL); } xfpustate = (char *)fpu_save_area_alloc(); error = copyin(PTRIN(ucp->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", td->td_proc->p_pid, td->td_name); return (error); } } else { xfpustate = NULL; xfpustate_len = 0; } ret = ia32_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", td->td_proc->p_pid, td->td_name, ret); return (ret); } regs->tf_rdi = ucp->uc_mcontext.mc_edi; regs->tf_rsi = ucp->uc_mcontext.mc_esi; regs->tf_rbp = ucp->uc_mcontext.mc_ebp; regs->tf_rbx = ucp->uc_mcontext.mc_ebx; regs->tf_rdx = ucp->uc_mcontext.mc_edx; regs->tf_rcx = ucp->uc_mcontext.mc_ecx; regs->tf_rax = ucp->uc_mcontext.mc_eax; regs->tf_trapno = ucp->uc_mcontext.mc_trapno; regs->tf_err = ucp->uc_mcontext.mc_err; regs->tf_rip = ucp->uc_mcontext.mc_eip; regs->tf_cs = cs; regs->tf_rflags = ucp->uc_mcontext.mc_eflags; regs->tf_rsp = ucp->uc_mcontext.mc_esp; regs->tf_ss = ucp->uc_mcontext.mc_ss; regs->tf_ds = ucp->uc_mcontext.mc_ds; regs->tf_es = ucp->uc_mcontext.mc_es; regs->tf_fs = ucp->uc_mcontext.mc_fs; regs->tf_gs = ucp->uc_mcontext.mc_gs; regs->tf_flags = TF_HASSEGS; kern_sigprocmask(td, SIG_SETMASK, &ucp->uc_sigmask, NULL, 0); set_pcb_flags(td->td_pcb, PCB_FULL_IRET); return (EJUSTRETURN); } /* * Clear registers on exec */ void ia32_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); #ifdef COMPAT_43 setup_lcall_gate(); #endif pcb->pcb_fsbase = 0; pcb->pcb_gsbase = 0; pcb->pcb_initial_fpucw = __INITIAL_FPUCW_I386__; 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 = _ucode32sel; regs->tf_rbx = (register_t)imgp->ps_strings; 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); fpstate_drop(td); /* Return via doreti so that we can change to a different %cs */ set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET); + clear_pcb_flags(pcb, PCB_TLSBASE); } diff --git a/sys/amd64/include/pcb.h b/sys/amd64/include/pcb.h index 1caa33ac0019..27e1dce08ee1 100644 --- a/sys/amd64/include/pcb.h +++ b/sys/amd64/include/pcb.h @@ -1,137 +1,139 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2003 Peter Wemm. * Copyright (c) 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. 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. */ #ifndef _AMD64_PCB_H_ #define _AMD64_PCB_H_ /* * AMD64 process control block */ #include #include #ifdef __amd64__ /* * NB: The fields marked with (*) are used by kernel debuggers. Their * ABI should be preserved. */ struct pcb { register_t pcb_r15; /* (*) */ register_t pcb_r14; /* (*) */ register_t pcb_r13; /* (*) */ register_t pcb_r12; /* (*) */ register_t pcb_rbp; /* (*) */ register_t pcb_rsp; /* (*) */ register_t pcb_rbx; /* (*) */ register_t pcb_rip; /* (*) */ register_t pcb_fsbase; register_t pcb_gsbase; register_t pcb_kgsbase; register_t pcb_cr0; register_t pcb_cr2; register_t pcb_cr3; register_t pcb_cr4; register_t pcb_dr0; register_t pcb_dr1; register_t pcb_dr2; register_t pcb_dr3; register_t pcb_dr6; register_t pcb_dr7; struct region_descriptor pcb_gdt; struct region_descriptor pcb_idt; struct region_descriptor pcb_ldt; uint16_t pcb_tr; u_int pcb_flags; #define PCB_FULL_IRET 0x0001 /* full iret is required */ #define PCB_DBREGS 0x0002 /* process using debug registers */ #define PCB_KERNFPU 0x0004 /* kernel uses fpu */ #define PCB_FPUINITDONE 0x0008 /* fpu state is initialized */ #define PCB_USERFPUINITDONE 0x0010 /* fpu user state is initialized */ #define PCB_KERNFPU_THR 0x0020 /* fpu_kern_thread() */ #define PCB_32BIT 0x0040 /* process has 32 bit context (segs etc) */ #define PCB_FPUNOSAVE 0x0080 /* no save area for current FPU ctx */ +#define PCB_TLSBASE 0x0100 /* tlsbase was set */ uint16_t pcb_initial_fpucw; /* copyin/out fault recovery */ caddr_t pcb_onfault; uint64_t pcb_saved_ucr3; /* local tss, with i/o bitmap; NULL for common */ struct amd64tss *pcb_tssp; /* model specific registers */ register_t pcb_efer; register_t pcb_star; register_t pcb_lstar; register_t pcb_cstar; register_t pcb_sfmask; struct savefpu *pcb_save; - uint64_t pcb_pad[5]; + register_t pcb_tlsbase; /* not same as pcb_fsbase */ + uint64_t pcb_pad[4]; }; /* Per-CPU state saved during suspend and resume. */ struct susppcb { struct pcb sp_pcb; /* fpu context for suspend/resume */ void *sp_fpususpend; }; #else /* 32bit */ struct pcb { uint64_t pcb_dummy[40]; }; #endif #ifdef _KERNEL struct trapframe; void clear_pcb_flags(struct pcb *pcb, const u_int flags); void makectx(struct trapframe *, struct pcb *); void set_pcb_flags(struct pcb *pcb, const u_int flags); void set_pcb_flags_raw(struct pcb *pcb, const u_int flags); int savectx(struct pcb *) __returns_twice; void resumectx(struct pcb *); /* Ensure that pcb_gsbase and pcb_fsbase are up to date */ #define update_pcb_bases(pcb) set_pcb_flags((pcb), PCB_FULL_IRET) #endif #endif /* _AMD64_PCB_H_ */ diff --git a/sys/amd64/linux/linux_sysvec.c b/sys/amd64/linux/linux_sysvec.c index 9c3d7e6405c1..3760a5455a73 100644 --- a/sys/amd64/linux/linux_sysvec.c +++ b/sys/amd64/linux/linux_sysvec.c @@ -1,965 +1,965 @@ /*- * 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. * Copyright (c) 2013, 2021 Dmitry Chagin * * 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. */ #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 _Static_assert(sizeof(struct l_fpstate) == sizeof(__typeof(((mcontext_t *)0)->mc_fpstate)), "fxsave area size incorrect"); MODULE_VERSION(linux64, 1); #define LINUX_VDSOPAGE_SIZE PAGE_SIZE * 2 #define LINUX_VDSOPAGE_LA48 (VM_MAXUSER_ADDRESS_LA48 - \ LINUX_VDSOPAGE_SIZE) #define LINUX_SHAREDPAGE_LA48 (LINUX_VDSOPAGE_LA48 - PAGE_SIZE) /* * PAGE_SIZE - the size * of the native SHAREDPAGE */ #define LINUX_USRSTACK_LA48 LINUX_SHAREDPAGE_LA48 #define LINUX_PS_STRINGS_LA48 (LINUX_USRSTACK_LA48 - \ sizeof(struct ps_strings)) static int linux_szsigcode; static vm_object_t linux_vdso_obj; static char *linux_vdso_mapping; extern char _binary_linux_vdso_so_o_start; extern char _binary_linux_vdso_so_o_end; static vm_offset_t linux_vdso_base; extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; extern const char *linux_syscallnames[]; SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); static void linux_vdso_install(const void *param); static void linux_vdso_deinstall(const void *param); static void linux_vdso_reloc(char *mapping, Elf_Addr offset); 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 void linux_exec_sysvec_init(void *param); static int linux_on_exec_vmspace(struct proc *p, struct image_params *imgp); static void linux_set_fork_retval(struct thread *td); static int linux_vsyscall(struct thread *td); LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode); LINUX_VDSO_SYM_CHAR(linux_platform); LINUX_VDSO_SYM_INTPTR(kern_timekeep_base); LINUX_VDSO_SYM_INTPTR(kern_tsc_selector); LINUX_VDSO_SYM_INTPTR(kern_cpu_selector); /* * According to the Intel x86 ISA 64-bit syscall * saves %rip to %rcx and rflags to %r11. Registers on syscall entry: * %rax system call number * %rcx return address * %r11 saved rflags * %rdi arg1 * %rsi arg2 * %rdx arg3 * %r10 arg4 * %r8 arg5 * %r9 arg6 * * Then FreeBSD fast_syscall() move registers: * %rcx -> trapframe.tf_rip * %r10 -> trapframe.tf_rcx */ static int linux_fetch_syscall_args(struct thread *td) { struct proc *p; struct trapframe *frame; struct syscall_args *sa; p = td->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; sa->original_code = sa->code; if (sa->code >= p->p_sysent->sv_size) /* nosys */ sa->callp = &nosys_sysent; else sa->callp = &p->p_sysent->sv_table[sa->code]; /* Restore r10 earlier to avoid doing this multiply times. */ frame->tf_r10 = frame->tf_rcx; /* Restore %rcx for machine context. */ frame->tf_rcx = frame->tf_rip; 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]; 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; break; case EJUSTRETURN: break; default: frame->tf_rax = bsd_to_linux_errno(error); 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. */ set_pcb_flags(td->td_pcb, PCB_FULL_IRET); } static void linux_set_fork_retval(struct thread *td) { struct trapframe *frame = td->td_frame; frame->tf_rax = 0; } void linux64_arch_copyout_auxargs(struct image_params *imgp, Elf_Auxinfo **pos) { AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO_EHDR, linux_vdso_base); AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP, cpu_feature); AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP2, linux_x86_elf_hwcap2()); AUXARGS_ENTRY((*pos), LINUX_AT_PLATFORM, PTROUT(linux_platform)); } /* * 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); + clear_pcb_flags(pcb, PCB_32BIT | PCB_TLSBASE); 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); } static int linux_fxrstor(struct thread *td, mcontext_t *mcp, struct l_sigcontext *sc) { struct savefpu *fp = (struct savefpu *)&mcp->mc_fpstate[0]; int error; error = copyin(PTRIN(sc->sc_fpstate), fp, sizeof(mcp->mc_fpstate)); if (error != 0) return (error); bzero(&fp->sv_pad[0], sizeof(fp->sv_pad)); return (set_fpcontext(td, mcp, NULL, 0)); } static int linux_xrstor(struct thread *td, mcontext_t *mcp, struct l_sigcontext *sc) { struct savefpu *fp = (struct savefpu *)&mcp->mc_fpstate[0]; char *xfpustate; struct proc *p; uint32_t magic2; int error; p = td->td_proc; mcp->mc_xfpustate_len = cpu_max_ext_state_size - sizeof(struct savefpu); /* Legacy region of an xsave area. */ error = copyin(PTRIN(sc->sc_fpstate), fp, sizeof(mcp->mc_fpstate)); if (error != 0) return (error); bzero(&fp->sv_pad[0], sizeof(fp->sv_pad)); /* Extended region of an xsave area. */ sc->sc_fpstate += sizeof(mcp->mc_fpstate); xfpustate = (char *)fpu_save_area_alloc(); error = copyin(PTRIN(sc->sc_fpstate), xfpustate, mcp->mc_xfpustate_len); if (error != 0) { fpu_save_area_free((struct savefpu *)xfpustate); uprintf("pid %d (%s): linux xrstor failed\n", p->p_pid, td->td_name); return (error); } /* Linux specific end of xsave area marker. */ sc->sc_fpstate += mcp->mc_xfpustate_len; error = copyin(PTRIN(sc->sc_fpstate), &magic2, LINUX_FP_XSTATE_MAGIC2_SIZE); if (error != 0 || magic2 != LINUX_FP_XSTATE_MAGIC2) { fpu_save_area_free((struct savefpu *)xfpustate); uprintf("pid %d (%s): sigreturn magic2 0x%x error %d\n", p->p_pid, td->td_name, magic2, error); return (error); } error = set_fpcontext(td, mcp, xfpustate, mcp->mc_xfpustate_len); fpu_save_area_free((struct savefpu *)xfpustate); if (error != 0) { uprintf("pid %d (%s): sigreturn set_fpcontext error %d\n", p->p_pid, td->td_name, error); } return (error); } static int linux_copyin_fpstate(struct thread *td, struct l_ucontext *uc) { mcontext_t mc; bzero(&mc, sizeof(mc)); mc.mc_ownedfp = _MC_FPOWNED_FPU; mc.mc_fpformat = _MC_FPFMT_XMM; if ((uc->uc_flags & LINUX_UC_FP_XSTATE) != 0) return (linux_xrstor(td, &mc, &uc->uc_mcontext)); else return (linux_fxrstor(td, &mc, &uc->uc_mcontext)); } /* * Copied from amd64/amd64/machdep.c */ int linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) { struct proc *p; struct l_rt_sigframe sf; struct l_sigcontext *context; struct trapframe *regs; unsigned long rflags; sigset_t bmask; int error; ksiginfo_t ksi; regs = td->td_frame; error = copyin((void *)regs->tf_rbx, &sf, sizeof(sf)); if (error != 0) return (error); p = td->td_proc; context = &sf.sf_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)) { 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)) { 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); } linux_to_bsd_sigset(&sf.sf_uc.uc_sigmask, &bmask); kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 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; error = linux_copyin_fpstate(td, &sf.sf_uc); if (error != 0) { uprintf("pid %d comm %s linux can't restore fpu state %d\n", p->p_pid, p->p_comm, error); return (error); } set_pcb_flags(td->td_pcb, PCB_FULL_IRET); return (EJUSTRETURN); } static int linux_fxsave(mcontext_t *mcp, void *ufp) { struct l_fpstate *fx = (struct l_fpstate *)&mcp->mc_fpstate[0]; bzero(&fx->reserved2[0], sizeof(fx->reserved2)); return (copyout(fx, ufp, sizeof(*fx))); } static int linux_xsave(mcontext_t *mcp, char *xfpusave, char *ufp) { struct l_fpstate *fx = (struct l_fpstate *)&mcp->mc_fpstate[0]; uint32_t magic2; int error; /* Legacy region of an xsave area. */ fx->sw_reserved.magic1 = LINUX_FP_XSTATE_MAGIC1; fx->sw_reserved.xstate_size = mcp->mc_xfpustate_len + sizeof(*fx); fx->sw_reserved.extended_size = fx->sw_reserved.xstate_size + LINUX_FP_XSTATE_MAGIC2_SIZE; fx->sw_reserved.xfeatures = xsave_mask; error = copyout(fx, ufp, sizeof(*fx)); if (error != 0) return (error); ufp += sizeof(*fx); /* Extended region of an xsave area. */ error = copyout(xfpusave, ufp, mcp->mc_xfpustate_len); if (error != 0) return (error); /* Linux specific end of xsave area marker. */ ufp += mcp->mc_xfpustate_len; magic2 = LINUX_FP_XSTATE_MAGIC2; return (copyout(&magic2, ufp, LINUX_FP_XSTATE_MAGIC2_SIZE)); } static int linux_copyout_fpstate(struct thread *td, struct l_ucontext *uc, char **sp) { size_t xfpusave_len; char *xfpusave; mcontext_t mc; char *ufp = *sp; get_fpcontext(td, &mc, &xfpusave, &xfpusave_len); KASSERT(mc.mc_fpformat != _MC_FPFMT_NODEV, ("fpu not present")); /* Room for fxsave area. */ ufp -= sizeof(struct l_fpstate); if (xfpusave != NULL) { /* Room for xsave area. */ ufp -= (xfpusave_len + LINUX_FP_XSTATE_MAGIC2_SIZE); uc->uc_flags |= LINUX_UC_FP_XSTATE; } *sp = ufp = (char *)((unsigned long)ufp & ~0x3Ful); if (xfpusave != NULL) return (linux_xsave(&mc, xfpusave, ufp)); else return (linux_fxsave(&mc, ufp)); } /* * 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; char *sp; struct trapframe *regs; int sig, code; int oonstack, issiginfo; td = curthread; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno); psp = p->p_sigacts; issiginfo = SIGISMEMBER(psp->ps_siginfo, sig); 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); bzero(&sf, sizeof(sf)); sf.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); sf.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size; sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; /* 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; } else sp = (char *)regs->tf_rsp - 128; mtx_unlock(&psp->ps_mtx); PROC_UNLOCK(p); if (linux_copyout_fpstate(td, &sf.sf_uc, &sp) != 0) { uprintf("pid %d comm %s linux can't save fpu state, killing\n", p->p_pid, p->p_comm); PROC_LOCK(p); sigexit(td, SIGILL); } sf.sf_uc.uc_mcontext.sc_fpstate = (register_t)sp; /* Make room, keeping the stack aligned. */ sp -= sizeof(struct l_rt_sigframe); sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); /* Save user context. */ bsd_to_linux_sigset(mask, &sf.sf_uc.uc_sigmask); sf.sf_uc.uc_mcontext.sc_mask = sf.sf_uc.uc_sigmask; sf.sf_uc.uc_mcontext.sc_rdi = regs->tf_rdi; sf.sf_uc.uc_mcontext.sc_rsi = regs->tf_rsi; sf.sf_uc.uc_mcontext.sc_rdx = regs->tf_rdx; sf.sf_uc.uc_mcontext.sc_rbp = regs->tf_rbp; sf.sf_uc.uc_mcontext.sc_rbx = regs->tf_rbx; sf.sf_uc.uc_mcontext.sc_rcx = regs->tf_rcx; sf.sf_uc.uc_mcontext.sc_rax = regs->tf_rax; sf.sf_uc.uc_mcontext.sc_rip = regs->tf_rip; sf.sf_uc.uc_mcontext.sc_rsp = regs->tf_rsp; sf.sf_uc.uc_mcontext.sc_r8 = regs->tf_r8; sf.sf_uc.uc_mcontext.sc_r9 = regs->tf_r9; sf.sf_uc.uc_mcontext.sc_r10 = regs->tf_r10; sf.sf_uc.uc_mcontext.sc_r11 = regs->tf_r11; sf.sf_uc.uc_mcontext.sc_r12 = regs->tf_r12; sf.sf_uc.uc_mcontext.sc_r13 = regs->tf_r13; sf.sf_uc.uc_mcontext.sc_r14 = regs->tf_r14; sf.sf_uc.uc_mcontext.sc_r15 = regs->tf_r15; sf.sf_uc.uc_mcontext.sc_cs = regs->tf_cs; sf.sf_uc.uc_mcontext.sc_rflags = regs->tf_rflags; sf.sf_uc.uc_mcontext.sc_err = regs->tf_err; sf.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); sf.sf_uc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; /* Translate the signal. */ sig = bsd_to_linux_signal(sig); /* Fill in POSIX parts. */ siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig); /* Copy the sigframe out to the user's stack. */ if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { uprintf("pid %d comm %s has trashed its stack, killing\n", p->p_pid, p->p_comm); PROC_LOCK(p); sigexit(td, SIGILL); } fpstate_drop(td); /* Build the argument list for the signal handler. */ regs->tf_rdi = sig; /* arg 1 in %rdi */ regs->tf_rax = 0; if (issiginfo) { regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ regs->tf_rdx = (register_t)&sfp->sf_uc; /* arg 3 in %rdx */ } else { regs->tf_rsi = 0; regs->tf_rdx = 0; } regs->tf_rcx = (register_t)catcher; 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_fixup = __elfN(freebsd_fixup), .sv_sendsig = linux_rt_sendsig, .sv_sigcode = &_binary_linux_vdso_so_o_start, .sv_szsigcode = &linux_szsigcode, .sv_name = "Linux ELF64", .sv_coredump = elf64_coredump, .sv_elf_core_osabi = ELFOSABI_NONE, .sv_elf_core_abi_vendor = LINUX_ABI_VENDOR, .sv_elf_core_prepare_notes = linux64_prepare_notes, .sv_minsigstksz = LINUX_MINSIGSTKSZ, .sv_minuser = VM_MIN_ADDRESS, .sv_maxuser = VM_MAXUSER_ADDRESS_LA48, .sv_usrstack = LINUX_USRSTACK_LA48, .sv_psstrings = LINUX_PS_STRINGS_LA48, .sv_psstringssz = sizeof(struct ps_strings), .sv_stackprot = VM_PROT_ALL, .sv_copyout_auxargs = __linuxN(copyout_auxargs), .sv_copyout_strings = __linuxN(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_TIMEKEEP, .sv_set_syscall_retval = linux_set_syscall_retval, .sv_fetch_syscall_args = linux_fetch_syscall_args, .sv_syscallnames = linux_syscallnames, .sv_shared_page_base = LINUX_SHAREDPAGE_LA48, .sv_shared_page_len = PAGE_SIZE, .sv_schedtail = linux_schedtail, .sv_thread_detach = linux_thread_detach, .sv_trap = linux_vsyscall, .sv_hwcap = NULL, .sv_hwcap2 = NULL, .sv_onexec = linux_on_exec_vmspace, .sv_onexit = linux_on_exit, .sv_ontdexit = linux_thread_dtor, .sv_setid_allowed = &linux_setid_allowed_query, .sv_set_fork_retval = linux_set_fork_retval, }; static int linux_on_exec_vmspace(struct proc *p, struct image_params *imgp) { int error; error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base, LINUX_VDSOPAGE_SIZE, imgp); if (error == 0) error = linux_on_exec(p, imgp); return (error); } /* * linux_vdso_install() and linux_exec_sysvec_init() must be called * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY). */ static void linux_exec_sysvec_init(void *param) { l_uintptr_t *ktimekeep_base, *ktsc_selector; struct sysentvec *sv; ptrdiff_t tkoff; sv = param; amd64_lower_shared_page(sv); /* Fill timekeep_base */ exec_sysvec_init(sv); tkoff = kern_timekeep_base - linux_vdso_base; ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff); *ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset; tkoff = kern_tsc_selector - linux_vdso_base; ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); *ktsc_selector = linux_vdso_tsc_selector_idx(); if (bootverbose) printf("Linux x86-64 vDSO tsc_selector: %lu\n", *ktsc_selector); tkoff = kern_cpu_selector - linux_vdso_base; ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); *ktsc_selector = linux_vdso_cpu_selector_idx(); if (bootverbose) printf("Linux x86-64 vDSO cpu_selector: %lu\n", *ktsc_selector); } SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY, linux_exec_sysvec_init, &elf_linux_sysvec); static void linux_vdso_install(const void *param) { char *vdso_start = &_binary_linux_vdso_so_o_start; char *vdso_end = &_binary_linux_vdso_so_o_end; linux_szsigcode = vdso_end - vdso_start; MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE); linux_vdso_base = LINUX_VDSOPAGE_LA48; if (hw_lower_amd64_sharedpage != 0) linux_vdso_base -= PAGE_SIZE; __elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base); linux_vdso_obj = __elfN(linux_shared_page_init) (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode); linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base); } SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST, linux_vdso_install, NULL); static void linux_vdso_deinstall(const void *param) { __elfN(linux_shared_page_fini)(linux_vdso_obj, linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); } SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, linux_vdso_deinstall, NULL); static void linux_vdso_reloc(char *mapping, Elf_Addr offset) { const Elf_Ehdr *ehdr; const Elf_Shdr *shdr; Elf64_Addr *where, val; Elf_Size rtype, symidx; const Elf_Rela *rela; Elf_Addr addr, addend; int relacnt; int i, j; MPASS(offset != 0); relacnt = 0; ehdr = (const Elf_Ehdr *)mapping; shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff); for (i = 0; i < ehdr->e_shnum; i++) { switch (shdr[i].sh_type) { case SHT_REL: printf("Linux x86_64 vDSO: unexpected Rel section\n"); break; case SHT_RELA: rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset); relacnt = shdr[i].sh_size / sizeof(*rela); } } for (j = 0; j < relacnt; j++, rela++) { where = (Elf_Addr *)(mapping + rela->r_offset); addend = rela->r_addend; rtype = ELF_R_TYPE(rela->r_info); symidx = ELF_R_SYM(rela->r_info); switch (rtype) { case R_X86_64_NONE: /* none */ break; case R_X86_64_RELATIVE: /* B + A */ addr = (Elf_Addr)(offset + addend); val = addr; if (*where != val) *where = val; break; case R_X86_64_IRELATIVE: printf("Linux x86_64 vDSO: unexpected ifunc relocation, " "symbol index %ld\n", symidx); break; default: printf("Linux x86_64 vDSO: unexpected relocation type %ld, " "symbol index %ld\n", rtype, symidx); } } } static Elf_Brandnote linux64_brandnote = { .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), .hdr.n_descsz = 16, .hdr.n_type = 1, .vendor = GNU_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", .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", .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", .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 | LINUX_BI_FUTEX_REQUEUE }; static 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 x86_64 ELF exec handler removed\n"); } else printf("Could not deinstall Linux x86_64 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"); diff --git a/sys/amd64/linux32/linux32_sysvec.c b/sys/amd64/linux32/linux32_sysvec.c index 19b4af7661f0..a8cc03ea8c71 100644 --- a/sys/amd64/linux32/linux32_sysvec.c +++ b/sys/amd64/linux32/linux32_sysvec.c @@ -1,1061 +1,1062 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * 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. */ #ifndef COMPAT_FREEBSD32 #error "Unable to compile Linux-emulator due to missing COMPAT_FREEBSD32 option!" #endif #define __ELF_WORD_SIZE 32 #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(linux, 1); #define LINUX32_MAXUSER ((1ul << 32) - PAGE_SIZE) #define LINUX32_VDSOPAGE_SIZE PAGE_SIZE * 2 #define LINUX32_VDSOPAGE (LINUX32_MAXUSER - LINUX32_VDSOPAGE_SIZE) #define LINUX32_SHAREDPAGE (LINUX32_VDSOPAGE - PAGE_SIZE) /* * PAGE_SIZE - the size * of the native SHAREDPAGE */ #define LINUX32_USRSTACK LINUX32_SHAREDPAGE static int linux_szsigcode; static vm_object_t linux_vdso_obj; static char *linux_vdso_mapping; extern char _binary_linux32_vdso_so_o_start; extern char _binary_linux32_vdso_so_o_end; static vm_offset_t linux_vdso_base; extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL]; extern const char *linux32_syscallnames[]; SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); static int linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base); static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask); static void linux_exec_setregs(struct thread *td, struct image_params *imgp, uintptr_t stack); static void linux_exec_sysvec_init(void *param); static int linux_on_exec_vmspace(struct proc *p, struct image_params *imgp); static void linux32_fixlimit(struct rlimit *rl, int which); static void linux_vdso_install(const void *param); static void linux_vdso_deinstall(const void *param); static void linux_vdso_reloc(char *mapping, Elf_Addr offset); static void linux32_set_fork_retval(struct thread *td); static void linux32_set_syscall_retval(struct thread *td, int error); struct linux32_ps_strings { u_int32_t ps_argvstr; /* first of 0 or more argument strings */ u_int ps_nargvstr; /* the number of argument strings */ u_int32_t ps_envstr; /* first of 0 or more environment strings */ u_int ps_nenvstr; /* the number of environment strings */ }; #define LINUX32_PS_STRINGS (LINUX32_USRSTACK - \ sizeof(struct linux32_ps_strings)) LINUX_VDSO_SYM_INTPTR(__kernel_vsyscall); LINUX_VDSO_SYM_INTPTR(linux32_vdso_sigcode); LINUX_VDSO_SYM_INTPTR(linux32_vdso_rt_sigcode); LINUX_VDSO_SYM_INTPTR(kern_timekeep_base); LINUX_VDSO_SYM_INTPTR(kern_tsc_selector); LINUX_VDSO_SYM_INTPTR(kern_cpu_selector); LINUX_VDSO_SYM_CHAR(linux_platform); void linux32_arch_copyout_auxargs(struct image_params *imgp, Elf_Auxinfo **pos) { AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO, __kernel_vsyscall); AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO_EHDR, linux_vdso_base); AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP, cpu_feature); AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP2, linux_x86_elf_hwcap2()); AUXARGS_ENTRY((*pos), LINUX_AT_PLATFORM, PTROUT(linux_platform)); } static void linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct thread *td = curthread; struct proc *p = td->td_proc; struct sigacts *psp; struct trapframe *regs; struct l_rt_sigframe *fp, frame; int oonstack; int sig; int code; sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno); code = ksi->ksi_code; PROC_LOCK_ASSERT(p, MA_OWNED); psp = p->p_sigacts; mtx_assert(&psp->ps_mtx, MA_OWNED); regs = td->td_frame; oonstack = sigonstack(regs->tf_rsp); /* Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe)); } else fp = (struct l_rt_sigframe *)regs->tf_rsp - 1; mtx_unlock(&psp->ps_mtx); /* Build the argument list for the signal handler. */ sig = bsd_to_linux_signal(sig); bzero(&frame, sizeof(frame)); frame.sf_sig = sig; frame.sf_siginfo = PTROUT(&fp->sf_si); frame.sf_ucontext = PTROUT(&fp->sf_uc); /* Fill in POSIX parts. */ siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig); /* * Build the signal context to be used by sigreturn and libgcc unwind. */ frame.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); frame.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size; frame.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; PROC_UNLOCK(p); bsd_to_linux_sigset(mask, &frame.sf_uc.uc_sigmask); frame.sf_uc.uc_mcontext.sc_mask = frame.sf_uc.uc_sigmask.__mask; frame.sf_uc.uc_mcontext.sc_edi = regs->tf_rdi; frame.sf_uc.uc_mcontext.sc_esi = regs->tf_rsi; frame.sf_uc.uc_mcontext.sc_ebp = regs->tf_rbp; frame.sf_uc.uc_mcontext.sc_ebx = regs->tf_rbx; frame.sf_uc.uc_mcontext.sc_esp = regs->tf_rsp; frame.sf_uc.uc_mcontext.sc_edx = regs->tf_rdx; frame.sf_uc.uc_mcontext.sc_ecx = regs->tf_rcx; frame.sf_uc.uc_mcontext.sc_eax = regs->tf_rax; frame.sf_uc.uc_mcontext.sc_eip = regs->tf_rip; frame.sf_uc.uc_mcontext.sc_cs = regs->tf_cs; frame.sf_uc.uc_mcontext.sc_gs = regs->tf_gs; frame.sf_uc.uc_mcontext.sc_fs = regs->tf_fs; frame.sf_uc.uc_mcontext.sc_es = regs->tf_es; frame.sf_uc.uc_mcontext.sc_ds = regs->tf_ds; frame.sf_uc.uc_mcontext.sc_eflags = regs->tf_rflags; frame.sf_uc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp; frame.sf_uc.uc_mcontext.sc_ss = regs->tf_ss; frame.sf_uc.uc_mcontext.sc_err = regs->tf_err; frame.sf_uc.uc_mcontext.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; frame.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); if (copyout(&frame, fp, sizeof(frame)) != 0) { /* * Process has trashed its stack; give it an illegal * instruction to halt it in its tracks. */ PROC_LOCK(p); sigexit(td, SIGILL); } /* Build context to run handler in. */ regs->tf_rsp = PTROUT(fp); regs->tf_rip = linux32_vdso_rt_sigcode; regs->tf_rdi = PTROUT(catcher); regs->tf_rflags &= ~(PSL_T | PSL_D); regs->tf_cs = _ucode32sel; regs->tf_ss = _udatasel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; regs->tf_fs = _ufssel; regs->tf_gs = _ugssel; regs->tf_flags = TF_HASSEGS; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } /* * Send an interrupt to process. * * Stack is set up to allow sigcode stored * in u. to call routine, followed by kcall * to sigreturn routine below. After sigreturn * resets the signal mask, the stack, and the * frame pointer, it returns to the user * specified pc, psl. */ static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct thread *td = curthread; struct proc *p = td->td_proc; struct sigacts *psp; struct trapframe *regs; struct l_sigframe *fp, frame; l_sigset_t lmask; int oonstack; int sig, code; sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno); code = ksi->ksi_code; PROC_LOCK_ASSERT(p, MA_OWNED); psp = p->p_sigacts; mtx_assert(&psp->ps_mtx, MA_OWNED); if (SIGISMEMBER(psp->ps_siginfo, sig)) { /* Signal handler installed with SA_SIGINFO. */ linux_rt_sendsig(catcher, ksi, mask); return; } regs = td->td_frame; oonstack = sigonstack(regs->tf_rsp); /* Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - sizeof(struct l_sigframe)); } else fp = (struct l_sigframe *)regs->tf_rsp - 1; mtx_unlock(&psp->ps_mtx); PROC_UNLOCK(p); /* Build the argument list for the signal handler. */ sig = bsd_to_linux_signal(sig); bzero(&frame, sizeof(frame)); frame.sf_sig = sig; frame.sf_sigmask = *mask; bsd_to_linux_sigset(mask, &lmask); /* Build the signal context to be used by sigreturn. */ frame.sf_sc.sc_mask = lmask.__mask; frame.sf_sc.sc_gs = regs->tf_gs; frame.sf_sc.sc_fs = regs->tf_fs; frame.sf_sc.sc_es = regs->tf_es; frame.sf_sc.sc_ds = regs->tf_ds; frame.sf_sc.sc_edi = regs->tf_rdi; frame.sf_sc.sc_esi = regs->tf_rsi; frame.sf_sc.sc_ebp = regs->tf_rbp; frame.sf_sc.sc_ebx = regs->tf_rbx; frame.sf_sc.sc_esp = regs->tf_rsp; frame.sf_sc.sc_edx = regs->tf_rdx; frame.sf_sc.sc_ecx = regs->tf_rcx; frame.sf_sc.sc_eax = regs->tf_rax; frame.sf_sc.sc_eip = regs->tf_rip; frame.sf_sc.sc_cs = regs->tf_cs; frame.sf_sc.sc_eflags = regs->tf_rflags; frame.sf_sc.sc_esp_at_signal = regs->tf_rsp; frame.sf_sc.sc_ss = regs->tf_ss; frame.sf_sc.sc_err = regs->tf_err; frame.sf_sc.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code); if (copyout(&frame, fp, sizeof(frame)) != 0) { /* * Process has trashed its stack; give it an illegal * instruction to halt it in its tracks. */ PROC_LOCK(p); sigexit(td, SIGILL); } /* Build context to run handler in. */ regs->tf_rsp = PTROUT(fp); regs->tf_rip = linux32_vdso_sigcode; regs->tf_rdi = PTROUT(catcher); regs->tf_rflags &= ~(PSL_T | PSL_D); regs->tf_cs = _ucode32sel; regs->tf_ss = _udatasel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; regs->tf_fs = _ufssel; regs->tf_gs = _ugssel; regs->tf_flags = TF_HASSEGS; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 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 * psl to gain improper privileges or to cause * a machine fault. */ int linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args) { struct l_sigframe frame; struct trapframe *regs; int eflags; ksiginfo_t ksi; regs = td->td_frame; /* * The trampoline code hands us the sigframe. * It is unsafe to keep track of it ourselves, in the event that a * program jumps out of a signal handler. */ if (copyin(args->sfp, &frame, sizeof(frame)) != 0) return (EFAULT); /* Check for security violations. */ eflags = frame.sf_sc.sc_eflags; if (!EFL_SECURE(eflags, regs->tf_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(frame.sf_sc.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); } kern_sigprocmask(td, SIG_SETMASK, &frame.sf_sigmask, NULL, 0); /* Restore signal context. */ regs->tf_rdi = frame.sf_sc.sc_edi; regs->tf_rsi = frame.sf_sc.sc_esi; regs->tf_rbp = frame.sf_sc.sc_ebp; regs->tf_rbx = frame.sf_sc.sc_ebx; regs->tf_rdx = frame.sf_sc.sc_edx; regs->tf_rcx = frame.sf_sc.sc_ecx; regs->tf_rax = frame.sf_sc.sc_eax; regs->tf_rip = frame.sf_sc.sc_eip; regs->tf_cs = frame.sf_sc.sc_cs; regs->tf_ds = frame.sf_sc.sc_ds; regs->tf_es = frame.sf_sc.sc_es; regs->tf_fs = frame.sf_sc.sc_fs; regs->tf_gs = frame.sf_sc.sc_gs; regs->tf_rflags = eflags; regs->tf_rsp = frame.sf_sc.sc_esp_at_signal; regs->tf_ss = frame.sf_sc.sc_ss; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); return (EJUSTRETURN); } /* * System call to cleanup state after a signal * has been taken. Reset signal mask and * stack state from context left by rt_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 * psl to gain improper privileges or to cause * a machine fault. */ int linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) { struct l_ucontext uc; struct l_sigcontext *context; sigset_t bmask; l_stack_t *lss; stack_t ss; struct trapframe *regs; int eflags; ksiginfo_t ksi; regs = td->td_frame; /* * The trampoline code hands us the ucontext. * It is unsafe to keep track of it ourselves, in the event that a * program jumps out of a signal handler. */ if (copyin(args->ucp, &uc, sizeof(uc)) != 0) return (EFAULT); context = &uc.uc_mcontext; /* Check for security violations. */ eflags = context->sc_eflags; if (!EFL_SECURE(eflags, regs->tf_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)) { 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); } linux_to_bsd_sigset(&uc.uc_sigmask, &bmask); kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); /* * Restore signal context */ regs->tf_gs = context->sc_gs; regs->tf_fs = context->sc_fs; regs->tf_es = context->sc_es; regs->tf_ds = context->sc_ds; regs->tf_rdi = context->sc_edi; regs->tf_rsi = context->sc_esi; regs->tf_rbp = context->sc_ebp; regs->tf_rbx = context->sc_ebx; regs->tf_rdx = context->sc_edx; regs->tf_rcx = context->sc_ecx; regs->tf_rax = context->sc_eax; regs->tf_rip = context->sc_eip; regs->tf_cs = context->sc_cs; regs->tf_rflags = eflags; regs->tf_rsp = context->sc_esp_at_signal; regs->tf_ss = context->sc_ss; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); /* * call sigaltstack & ignore results.. */ lss = &uc.uc_stack; ss.ss_sp = PTRIN(lss->ss_sp); ss.ss_size = lss->ss_size; ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags); (void)kern_sigaltstack(td, &ss, NULL); return (EJUSTRETURN); } static int linux32_fetch_syscall_args(struct thread *td) { struct proc *p; struct trapframe *frame; struct syscall_args *sa; p = td->td_proc; frame = td->td_frame; sa = &td->td_sa; sa->args[0] = frame->tf_rbx; sa->args[1] = frame->tf_rcx; sa->args[2] = frame->tf_rdx; sa->args[3] = frame->tf_rsi; sa->args[4] = frame->tf_rdi; sa->args[5] = frame->tf_rbp; sa->code = frame->tf_rax; sa->original_code = sa->code; if (sa->code >= p->p_sysent->sv_size) /* nosys */ sa->callp = &nosys_sysent; else sa->callp = &p->p_sysent->sv_table[sa->code]; td->td_retval[0] = 0; td->td_retval[1] = frame->tf_rdx; return (0); } static void linux32_set_syscall_retval(struct thread *td, int error) { struct trapframe *frame = td->td_frame; cpu_set_syscall_retval(td, error); if (__predict_false(error != 0)) { if (error != ERESTART && error != EJUSTRETURN) frame->tf_rax = bsd_to_linux_errno(error); } } static void linux32_set_fork_retval(struct thread *td) { struct trapframe *frame = td->td_frame; frame->tf_rax = 0; } /* * Clear registers on exec * XXX copied from ia32_signal.c. */ static void linux_exec_setregs(struct thread *td, struct image_params *imgp, uintptr_t stack) { struct trapframe *regs = td->td_frame; struct pcb *pcb = td->td_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); + /* Do full restore on return so that we can change to a different %cs */ + set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET); + clear_pcb_flags(pcb, PCB_TLSBASE); + critical_enter(); wrmsr(MSR_FSBASE, 0); wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */ pcb->pcb_fsbase = 0; pcb->pcb_gsbase = 0; critical_exit(); pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 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_gs = _ugssel; regs->tf_fs = _ufssel; regs->tf_es = _udatasel; regs->tf_ds = _udatasel; regs->tf_ss = _udatasel; regs->tf_flags = TF_HASSEGS; regs->tf_cs = _ucode32sel; regs->tf_rbx = (register_t)imgp->ps_strings; x86_clear_dbregs(pcb); fpstate_drop(td); - - /* Do full restore on return so that we can change to a different %cs */ - set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET); } /* * XXX copied from ia32_sysvec.c. */ static int linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) { int argc, envc, error; u_int32_t *vectp; char *stringp; uintptr_t destp, ustringp; struct linux32_ps_strings *arginfo; char canary[LINUX_AT_RANDOM_LEN]; size_t execpath_len; arginfo = (struct linux32_ps_strings *)PROC_PS_STRINGS(imgp->proc); destp = (uintptr_t)arginfo; if (imgp->execpath != NULL && imgp->auxargs != NULL) { execpath_len = strlen(imgp->execpath) + 1; destp -= execpath_len; destp = rounddown2(destp, sizeof(uint32_t)); 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(uint32_t)); 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(uint32_t)); 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(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. */ if (suword32(&arginfo->ps_argvstr, (uint32_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); if (suword32(&arginfo->ps_envstr, (uint32_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++; } /* The end of the 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); } static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "32-bit Linux emulation"); static u_long linux32_maxdsiz = LINUX32_MAXDSIZ; SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW, &linux32_maxdsiz, 0, ""); static u_long linux32_maxssiz = LINUX32_MAXSSIZ; SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW, &linux32_maxssiz, 0, ""); static u_long linux32_maxvmem = LINUX32_MAXVMEM; SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW, &linux32_maxvmem, 0, ""); bool linux32_emulate_i386 = false; SYSCTL_BOOL(_compat_linux32, OID_AUTO, emulate_i386, CTLFLAG_RWTUN, &linux32_emulate_i386, 0, "Emulate the real i386"); static void linux32_fixlimit(struct rlimit *rl, int which) { switch (which) { case RLIMIT_DATA: if (linux32_maxdsiz != 0) { if (rl->rlim_cur > linux32_maxdsiz) rl->rlim_cur = linux32_maxdsiz; if (rl->rlim_max > linux32_maxdsiz) rl->rlim_max = linux32_maxdsiz; } break; case RLIMIT_STACK: if (linux32_maxssiz != 0) { if (rl->rlim_cur > linux32_maxssiz) rl->rlim_cur = linux32_maxssiz; if (rl->rlim_max > linux32_maxssiz) rl->rlim_max = linux32_maxssiz; } break; case RLIMIT_VMEM: if (linux32_maxvmem != 0) { if (rl->rlim_cur > linux32_maxvmem) rl->rlim_cur = linux32_maxvmem; if (rl->rlim_max > linux32_maxvmem) rl->rlim_max = linux32_maxvmem; } break; } } struct sysentvec elf_linux_sysvec = { .sv_size = LINUX32_SYS_MAXSYSCALL, .sv_table = linux32_sysent, .sv_fixup = elf32_freebsd_fixup, .sv_sendsig = linux_sendsig, .sv_sigcode = &_binary_linux32_vdso_so_o_start, .sv_szsigcode = &linux_szsigcode, .sv_name = "Linux ELF32", .sv_coredump = elf32_coredump, .sv_elf_core_osabi = ELFOSABI_NONE, .sv_elf_core_abi_vendor = LINUX_ABI_VENDOR, .sv_elf_core_prepare_notes = linux32_prepare_notes, .sv_minsigstksz = LINUX_MINSIGSTKSZ, .sv_minuser = VM_MIN_ADDRESS, .sv_maxuser = LINUX32_MAXUSER, .sv_usrstack = LINUX32_USRSTACK, .sv_psstrings = LINUX32_PS_STRINGS, .sv_psstringssz = sizeof(struct linux32_ps_strings), .sv_stackprot = VM_PROT_ALL, .sv_copyout_auxargs = __linuxN(copyout_auxargs), .sv_copyout_strings = linux_copyout_strings, .sv_setregs = linux_exec_setregs, .sv_fixlimit = linux32_fixlimit, .sv_maxssiz = &linux32_maxssiz, .sv_flags = SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP | SV_SIG_DISCIGN | SV_SIG_WAITNDQ | SV_TIMEKEEP, .sv_set_syscall_retval = linux32_set_syscall_retval, .sv_fetch_syscall_args = linux32_fetch_syscall_args, .sv_syscallnames = linux32_syscallnames, .sv_shared_page_base = LINUX32_SHAREDPAGE, .sv_shared_page_len = PAGE_SIZE, .sv_schedtail = linux_schedtail, .sv_thread_detach = linux_thread_detach, .sv_trap = NULL, .sv_hwcap = NULL, .sv_hwcap2 = NULL, .sv_onexec = linux_on_exec_vmspace, .sv_onexit = linux_on_exit, .sv_ontdexit = linux_thread_dtor, .sv_setid_allowed = &linux_setid_allowed_query, .sv_set_fork_retval = linux32_set_fork_retval, }; static int linux_on_exec_vmspace(struct proc *p, struct image_params *imgp) { int error; error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base, LINUX32_VDSOPAGE_SIZE, imgp); if (error == 0) error = linux_on_exec(p, imgp); return (error); } /* * linux_vdso_install() and linux_exec_sysvec_init() must be called * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY). */ static void linux_exec_sysvec_init(void *param) { l_uintptr_t *ktimekeep_base, *ktsc_selector; struct sysentvec *sv; ptrdiff_t tkoff; sv = param; /* Fill timekeep_base */ exec_sysvec_init(sv); tkoff = kern_timekeep_base - linux_vdso_base; ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff); *ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset; tkoff = kern_tsc_selector - linux_vdso_base; ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); *ktsc_selector = linux_vdso_tsc_selector_idx(); if (bootverbose) printf("Linux i386 vDSO tsc_selector: %u\n", *ktsc_selector); tkoff = kern_cpu_selector - linux_vdso_base; ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); *ktsc_selector = linux_vdso_cpu_selector_idx(); if (bootverbose) printf("Linux i386 vDSO cpu_selector: %u\n", *ktsc_selector); } SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY, linux_exec_sysvec_init, &elf_linux_sysvec); static void linux_vdso_install(const void *param) { char *vdso_start = &_binary_linux32_vdso_so_o_start; char *vdso_end = &_binary_linux32_vdso_so_o_end; linux_szsigcode = vdso_end - vdso_start; MPASS(linux_szsigcode <= LINUX32_VDSOPAGE_SIZE); linux_vdso_base = LINUX32_VDSOPAGE; __elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base); linux_vdso_obj = __elfN(linux_shared_page_init) (&linux_vdso_mapping, LINUX32_VDSOPAGE_SIZE); bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode); linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base); } SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST, linux_vdso_install, NULL); static void linux_vdso_deinstall(const void *param) { __elfN(linux_shared_page_fini)(linux_vdso_obj, linux_vdso_mapping, LINUX32_VDSOPAGE_SIZE); } SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, linux_vdso_deinstall, NULL); static void linux_vdso_reloc(char *mapping, Elf_Addr offset) { const Elf_Shdr *shdr; const Elf_Rel *rel; const Elf_Ehdr *ehdr; Elf32_Addr *where; Elf_Size rtype, symidx; Elf32_Addr addr, addend; int i, relcnt; MPASS(offset != 0); relcnt = 0; ehdr = (const Elf_Ehdr *)mapping; shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff); for (i = 0; i < ehdr->e_shnum; i++) { switch (shdr[i].sh_type) { case SHT_REL: rel = (const Elf_Rel *)(mapping + shdr[i].sh_offset); relcnt = shdr[i].sh_size / sizeof(*rel); break; case SHT_RELA: printf("Linux i386 vDSO: unexpected Rela section\n"); break; } } for (i = 0; i < relcnt; i++, rel++) { where = (Elf32_Addr *)(mapping + rel->r_offset); addend = *where; rtype = ELF_R_TYPE(rel->r_info); symidx = ELF_R_SYM(rel->r_info); switch (rtype) { case R_386_NONE: /* none */ break; case R_386_RELATIVE: /* B + A */ addr = (Elf32_Addr)PTROUT(offset + addend); if (*where != addr) *where = addr; break; case R_386_IRELATIVE: printf("Linux i386 vDSO: unexpected ifunc relocation, " "symbol index %ld\n", (intmax_t)symidx); break; default: printf("Linux i386 vDSO: unexpected relocation type %ld, " "symbol index %ld\n", (intmax_t)rtype, (intmax_t)symidx); } } } static Elf_Brandnote linux32_brandnote = { .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), .hdr.n_descsz = 16, /* XXX at least 16 */ .hdr.n_type = 1, .vendor = GNU_ABI_VENDOR, .flags = BN_TRANSLATE_OSREL, .trans_osrel = linux_trans_osrel }; static Elf32_Brandinfo linux_brand = { .brand = ELFOSABI_LINUX, .machine = EM_386, .compat_3_brand = "Linux", .interp_path = "/lib/ld-linux.so.1", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux32_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; static Elf32_Brandinfo linux_glibc2brand = { .brand = ELFOSABI_LINUX, .machine = EM_386, .compat_3_brand = "Linux", .interp_path = "/lib/ld-linux.so.2", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux32_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; static Elf32_Brandinfo linux_muslbrand = { .brand = ELFOSABI_LINUX, .machine = EM_386, .compat_3_brand = "Linux", .interp_path = "/lib/ld-musl-i386.so.1", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux32_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE | LINUX_BI_FUTEX_REQUEUE }; static Elf32_Brandinfo *linux_brandlist[] = { &linux_brand, &linux_glibc2brand, &linux_muslbrand, NULL }; static int linux_elf_modevent(module_t mod, int type, void *data) { Elf32_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 (elf32_insert_brand_entry(*brandinfo) < 0) error = EINVAL; if (error == 0) { SET_FOREACH(lihp, linux_ioctl_handler_set) linux32_ioctl_register_handler(*lihp); stclohz = (stathz ? stathz : hz); if (bootverbose) printf("Linux i386 ELF exec handler installed\n"); } else printf("cannot insert Linux i386 ELF brand handler\n"); break; case MOD_UNLOAD: for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; ++brandinfo) if (elf32_brand_inuse(*brandinfo)) error = EBUSY; if (error == 0) { for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; ++brandinfo) if (elf32_remove_brand_entry(*brandinfo) < 0) error = EINVAL; } if (error == 0) { SET_FOREACH(lihp, linux_ioctl_handler_set) linux32_ioctl_unregister_handler(*lihp); if (bootverbose) printf("Linux i386 ELF exec handler removed\n"); } else printf("Could not deinstall Linux i386 ELF interpreter entry\n"); break; default: return (EOPNOTSUPP); } return (error); } static moduledata_t linux_elf_mod = { "linuxelf", linux_elf_modevent, 0 }; DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1); FEATURE(linux, "Linux 32bit support"); diff --git a/sys/x86/include/sysarch.h b/sys/x86/include/sysarch.h index 3226f3b9d93a..61f559f4bd23 100644 --- a/sys/x86/include/sysarch.h +++ b/sys/x86/include/sysarch.h @@ -1,165 +1,167 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 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. */ /* * Architecture specific syscalls (X86) */ #ifndef _MACHINE_SYSARCH_H_ #define _MACHINE_SYSARCH_H_ #include #define I386_GET_LDT 0 #define I386_SET_LDT 1 #define LDT_AUTO_ALLOC 0xffffffff /* I386_IOPL */ #define I386_GET_IOPERM 3 #define I386_SET_IOPERM 4 /* xxxxx */ #define I386_VM86 6 /* XXX Not implementable on amd64 */ #define I386_GET_FSBASE 7 #define I386_SET_FSBASE 8 #define I386_GET_GSBASE 9 #define I386_SET_GSBASE 10 #define I386_GET_XFPUSTATE 11 #define I386_SET_PKRU 12 #define I386_CLEAR_PKRU 13 /* Leave space for 0-127 for to avoid translating syscalls */ #define AMD64_GET_FSBASE 128 #define AMD64_SET_FSBASE 129 #define AMD64_GET_GSBASE 130 #define AMD64_SET_GSBASE 131 #define AMD64_GET_XFPUSTATE 132 #define AMD64_SET_PKRU 133 #define AMD64_CLEAR_PKRU 134 +#define AMD64_GET_TLSBASE 135 +#define AMD64_SET_TLSBASE 136 /* Flags for AMD64_SET_PKRU */ #define AMD64_PKRU_EXCL 0x0001 #define AMD64_PKRU_PERSIST 0x0002 struct i386_ioperm_args { unsigned int start; unsigned int length; int enable; }; #ifdef __i386__ struct i386_ldt_args { unsigned int start; union descriptor *descs; unsigned int num; }; struct i386_vm86_args { int sub_op; /* sub-operation to perform */ char *sub_args; /* args */ }; struct i386_get_xfpustate { void *addr; int len; }; #else struct i386_ldt_args { unsigned int start; struct user_segment_descriptor *descs __packed; unsigned int num; }; struct i386_get_xfpustate { unsigned int addr; int len; }; struct i386_set_pkru { unsigned int addr; unsigned int len; unsigned int keyidx; int flags; }; struct amd64_get_xfpustate { void *addr; int len; }; #endif struct amd64_set_pkru { void *addr; unsigned long len; unsigned int keyidx; int flags; }; #ifndef _KERNEL union descriptor; struct dbreg; __BEGIN_DECLS int i386_get_ldt(int, union descriptor *, int); int i386_set_ldt(int, union descriptor *, int); int i386_get_ioperm(unsigned int, unsigned int *, int *); int i386_set_ioperm(unsigned int, unsigned int, int); int i386_vm86(int, void *); int i386_get_fsbase(void **); int i386_get_gsbase(void **); int i386_set_fsbase(void *); int i386_set_gsbase(void *); int i386_set_watch(int, unsigned int, int, int, struct dbreg *); int i386_clr_watch(int, struct dbreg *); int amd64_get_fsbase(void **); int amd64_get_gsbase(void **); int amd64_set_fsbase(void *); int amd64_set_gsbase(void *); int x86_pkru_get_perm(unsigned int keyidx, int *access, int *modify); int x86_pkru_set_perm(unsigned int keyidx, int access, int modify); int x86_pkru_protect_range(void *addr, unsigned long len, unsigned int keyidx, int flag); int x86_pkru_unprotect_range(void *addr, unsigned long len); int sysarch(int, void *); __END_DECLS #else struct thread; union descriptor; int i386_get_ldt(struct thread *, struct i386_ldt_args *); int i386_set_ldt(struct thread *, struct i386_ldt_args *, union descriptor *); int i386_get_ioperm(struct thread *, struct i386_ioperm_args *); int i386_set_ioperm(struct thread *, struct i386_ioperm_args *); int amd64_get_ldt(struct thread *, struct i386_ldt_args *); int amd64_set_ldt(struct thread *, struct i386_ldt_args *, struct user_segment_descriptor *); int amd64_get_ioperm(struct thread *, struct i386_ioperm_args *); int amd64_set_ioperm(struct thread *, struct i386_ioperm_args *); #endif #endif /* !_MACHINE_SYSARCH_H_ */