diff --git a/sys/amd64/linux/linux_sysvec.c b/sys/amd64/linux/linux_sysvec.c index dc9a77093bcd..fede294ce706 100644 --- a/sys/amd64/linux/linux_sysvec.c +++ b/sys/amd64/linux/linux_sysvec.c @@ -1,1061 +1,1062 @@ /*- * 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. */ #include __FBSDID("$FreeBSD$"); #define __ELF_WORD_SIZE 64 #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include 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]; SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); static int linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base); static int linux_fixup_elf(uintptr_t *stack_base, struct image_params *iparams); static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel); static void linux_vdso_install(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 int linux_vsyscall(struct thread *td); #define LINUX_T_UNKNOWN 255 static int _bsd_to_linux_trapcode[] = { LINUX_T_UNKNOWN, /* 0 */ 6, /* 1 T_PRIVINFLT */ LINUX_T_UNKNOWN, /* 2 */ 3, /* 3 T_BPTFLT */ LINUX_T_UNKNOWN, /* 4 */ LINUX_T_UNKNOWN, /* 5 */ 16, /* 6 T_ARITHTRAP */ 254, /* 7 T_ASTFLT */ LINUX_T_UNKNOWN, /* 8 */ 13, /* 9 T_PROTFLT */ 1, /* 10 T_TRCTRAP */ LINUX_T_UNKNOWN, /* 11 */ 14, /* 12 T_PAGEFLT */ LINUX_T_UNKNOWN, /* 13 */ 17, /* 14 T_ALIGNFLT */ LINUX_T_UNKNOWN, /* 15 */ LINUX_T_UNKNOWN, /* 16 */ LINUX_T_UNKNOWN, /* 17 */ 0, /* 18 T_DIVIDE */ 2, /* 19 T_NMI */ 4, /* 20 T_OFLOW */ 5, /* 21 T_BOUND */ 7, /* 22 T_DNA */ 8, /* 23 T_DOUBLEFLT */ 9, /* 24 T_FPOPFLT */ 10, /* 25 T_TSSFLT */ 11, /* 26 T_SEGNPFLT */ 12, /* 27 T_STKFLT */ 18, /* 28 T_MCHK */ 19, /* 29 T_XMMFLT */ 15 /* 30 T_RESERVED */ }; #define bsd_to_linux_trapcode(code) \ ((code)td_proc; frame = td->td_frame; sa = &td->td_sa; sa->args[0] = frame->tf_rdi; sa->args[1] = frame->tf_rsi; sa->args[2] = frame->tf_rdx; sa->args[3] = frame->tf_rcx; sa->args[4] = frame->tf_r8; sa->args[5] = frame->tf_r9; sa->code = frame->tf_rax; sa->original_code = sa->code; if (sa->code >= p->p_sysent->sv_size) /* nosys */ sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; else sa->callp = &p->p_sysent->sv_table[sa->code]; td->td_retval[0] = 0; return (0); } static void linux_set_syscall_retval(struct thread *td, int error) { struct trapframe *frame; frame = td->td_frame; switch (error) { case 0: frame->tf_rax = td->td_retval[0]; frame->tf_r10 = frame->tf_rcx; break; case ERESTART: /* * Reconstruct pc, we know that 'syscall' is 2 bytes, * lcall $X,y is 7 bytes, int 0x80 is 2 bytes. * We saved this in tf_err. * */ frame->tf_rip -= frame->tf_err; frame->tf_r10 = frame->tf_rcx; break; case EJUSTRETURN: break; default: frame->tf_rax = bsd_to_linux_errno(error); frame->tf_r10 = frame->tf_rcx; break; } /* * Differently from FreeBSD native ABI, on Linux only %rcx * and %r11 values are not preserved across the syscall. * Require full context restore to get all registers except * those two restored at return to usermode. * * XXX: Would be great to be able to avoid PCB_FULL_IRET * for the error == 0 case. */ set_pcb_flags(td->td_pcb, PCB_FULL_IRET); } static int linux_copyout_auxargs(struct image_params *imgp, uintptr_t base) { Elf_Auxargs *args; Elf_Auxinfo *argarray, *pos; struct proc *p; int error, issetugid; p = imgp->proc; args = (Elf64_Auxargs *)imgp->auxargs; argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, M_WAITOK | M_ZERO); issetugid = p->p_flag & P_SUGID ? 1 : 0; AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base); AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); AUXARGS_ENTRY(pos, AT_PHENT, args->phent); AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); AUXARGS_ENTRY(pos, AT_BASE, args->base); AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary); AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0); if (imgp->execpathp != 0) AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp); if (args->execfd != -1) AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); AUXARGS_ENTRY(pos, AT_NULL, 0); free(imgp->auxargs, M_TEMP); imgp->auxargs = NULL; KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); error = copyout(argarray, (void *)base, sizeof(*argarray) * LINUX_AT_COUNT); free(argarray, M_TEMP); return (error); } static int linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp) { Elf_Addr *base; base = (Elf64_Addr *)*stack_base; base--; if (suword(base, (uint64_t)imgp->args->argc) == -1) return (EFAULT); *stack_base = (uintptr_t)base; return (0); } /* * Copy strings out to the new process address space, constructing new arg * and env vector tables. Return a pointer to the base so that it can be used * as the initial stack pointer. */ static int linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) { int argc, envc, error; char **vectp; char *stringp; uintptr_t destp, ustringp; struct ps_strings *arginfo; char canary[LINUX_AT_RANDOM_LEN]; size_t execpath_len; struct proc *p; /* Calculate string base and vector table pointers. */ if (imgp->execpath != NULL && imgp->auxargs != NULL) execpath_len = strlen(imgp->execpath) + 1; else execpath_len = 0; p = imgp->proc; arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; destp = (uintptr_t)arginfo; if (execpath_len != 0) { destp -= execpath_len; destp = rounddown2(destp, sizeof(void *)); imgp->execpathp = (void *)destp; error = copyout(imgp->execpath, imgp->execpathp, execpath_len); if (error != 0) return (error); } /* Prepare the canary for SSP. */ arc4rand(canary, sizeof(canary), 0); destp -= roundup(sizeof(canary), sizeof(void *)); imgp->canary = (void *)destp; error = copyout(canary, imgp->canary, sizeof(canary)); if (error != 0) return (error); /* Allocate room for the argument and environment strings. */ destp -= ARG_MAX - imgp->args->stringspace; destp = rounddown2(destp, sizeof(void *)); ustringp = destp; if (imgp->auxargs) { /* * Allocate room on the stack for the ELF auxargs * array. It has LINUX_AT_COUNT entries. */ destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo); destp = rounddown2(destp, sizeof(void *)); } vectp = (char **)destp; /* * Allocate room for the argv[] and env vectors including the * terminating NULL pointers. */ vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; /* * Starting with 2.24, glibc depends on a 16-byte stack alignment. * One "long argc" will be prepended later. */ vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8); /* vectp also becomes our initial stack base. */ *stack_base = (uintptr_t)vectp; stringp = imgp->args->begin_argv; argc = imgp->args->argc; envc = imgp->args->envc; /* Copy out strings - arguments and environment. */ error = copyout(stringp, (void *)ustringp, ARG_MAX - imgp->args->stringspace); if (error != 0) return (error); /* Fill in "ps_strings" struct for ps, w, etc. */ if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || suword(&arginfo->ps_nargvstr, argc) != 0) return (EFAULT); /* Fill in argument portion of vector table. */ for (; argc > 0; --argc) { if (suword(vectp++, ustringp) != 0) return (EFAULT); while (*stringp++ != 0) ustringp++; ustringp++; } /* A null vector table pointer separates the argp's from the envp's. */ if (suword(vectp++, 0) != 0) return (EFAULT); if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || suword(&arginfo->ps_nenvstr, envc) != 0) return (EFAULT); /* Fill in environment portion of vector table. */ for (; envc > 0; --envc) { if (suword(vectp++, ustringp) != 0) return (EFAULT); while (*stringp++ != 0) ustringp++; ustringp++; } /* The end of the vector table is a null pointer. */ if (suword(vectp, 0) != 0) return (EFAULT); if (imgp->auxargs) { vectp++; error = imgp->sysent->sv_copyout_auxargs(imgp, (uintptr_t)vectp); if (error != 0) return (error); } return (0); } /* * Reset registers to default values on exec. */ static void linux_exec_setregs(struct thread *td, struct image_params *imgp, uintptr_t stack) { struct trapframe *regs; struct pcb *pcb; register_t saved_rflags; regs = td->td_frame; pcb = td->td_pcb; if (td->td_proc->p_md.md_ldt != NULL) user_ldt_free(td); pcb->pcb_fsbase = 0; pcb->pcb_gsbase = 0; clear_pcb_flags(pcb, PCB_32BIT); pcb->pcb_initial_fpucw = __LINUX_NPXCW__; set_pcb_flags(pcb, PCB_FULL_IRET); saved_rflags = regs->tf_rflags & PSL_T; bzero((char *)regs, sizeof(struct trapframe)); regs->tf_rip = imgp->entry_addr; regs->tf_rsp = stack; regs->tf_rflags = PSL_USER | saved_rflags; regs->tf_ss = _udatasel; regs->tf_cs = _ucodesel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; regs->tf_fs = _ufssel; regs->tf_gs = _ugssel; regs->tf_flags = TF_HASSEGS; x86_clear_dbregs(pcb); /* * Drop the FP state if we hold it, so that the process gets a * clean FP state if it uses the FPU again. */ fpstate_drop(td); } /* * Copied from amd64/amd64/machdep.c * * XXX fpu state need? don't think so */ int linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) { struct proc *p; struct l_ucontext uc; struct l_sigcontext *context; struct trapframe *regs; unsigned long rflags; int error; ksiginfo_t ksi; regs = td->td_frame; error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc)); if (error != 0) return (error); p = td->td_proc; context = &uc.uc_mcontext; rflags = context->sc_rflags; /* * Don't allow users to change privileged or reserved flags. */ /* * XXX do allow users to change the privileged flag PSL_RF. * The cpu sets PSL_RF in tf_rflags for faults. Debuggers * should sometimes set it there too. tf_rflags is kept in * the signal context during signal handling and there is no * other place to remember it, so the PSL_RF bit may be * corrupted by the signal handler without us knowing. * Corruption of the PSL_RF bit at worst causes one more or * one less debugger trap, so allowing it is fairly harmless. */ #define RFLAG_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { printf("linux_rt_sigreturn: rflags = 0x%lx\n", 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. */ #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) if (!CS_SECURE(context->sc_cs)) { printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs); ksiginfo_init_trap(&ksi); ksi.ksi_signo = SIGBUS; ksi.ksi_code = BUS_OBJERR; ksi.ksi_trapno = T_PROTFLT; ksi.ksi_addr = (void *)regs->tf_rip; trapsignal(td, &ksi); return (EINVAL); } PROC_LOCK(p); linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask); SIG_CANTMASK(td->td_sigmask); signotify(td); PROC_UNLOCK(p); regs->tf_rdi = context->sc_rdi; regs->tf_rsi = context->sc_rsi; regs->tf_rdx = context->sc_rdx; regs->tf_rbp = context->sc_rbp; regs->tf_rbx = context->sc_rbx; regs->tf_rcx = context->sc_rcx; regs->tf_rax = context->sc_rax; regs->tf_rip = context->sc_rip; regs->tf_rsp = context->sc_rsp; regs->tf_r8 = context->sc_r8; regs->tf_r9 = context->sc_r9; regs->tf_r10 = context->sc_r10; regs->tf_r11 = context->sc_r11; regs->tf_r12 = context->sc_r12; regs->tf_r13 = context->sc_r13; regs->tf_r14 = context->sc_r14; regs->tf_r15 = context->sc_r15; regs->tf_cs = context->sc_cs; regs->tf_err = context->sc_err; regs->tf_rflags = rflags; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); return (EJUSTRETURN); } /* * copied from amd64/amd64/machdep.c * * Send an interrupt to process. */ static void linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct l_rt_sigframe sf, *sfp; struct proc *p; struct thread *td; struct sigacts *psp; caddr_t sp; struct trapframe *regs; int sig, code; int oonstack; td = curthread; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); sig = ksi->ksi_signo; psp = p->p_sigacts; code = ksi->ksi_code; mtx_assert(&psp->ps_mtx, MA_OWNED); regs = td->td_frame; oonstack = sigonstack(regs->tf_rsp); LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u", catcher, sig, mask, code); /* Translate the signal. */ sig = bsd_to_linux_signal(sig); /* Save user context. */ bzero(&sf, sizeof(sf)); bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask); bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask); sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi; sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi; sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx; sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp; sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx; sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx; sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax; sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip; sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp; sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8; sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9; sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10; sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11; sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12; sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13; sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14; sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15; sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags; sf.sf_sc.uc_mcontext.sc_err = regs->tf_err; sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; /* Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe); } else sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128; /* Align to 16 bytes. */ sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); /* Build the argument list for the signal handler. */ regs->tf_rdi = sig; /* arg 1 in %rdi */ regs->tf_rax = 0; regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */ /* Fill in POSIX parts. */ siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig); sf.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) { PROC_LOCK(p); sigexit(td, SIGILL); } regs->tf_rsp = (long)sfp; regs->tf_rip = linux_rt_sigcode; regs->tf_rflags &= ~(PSL_T | PSL_D); regs->tf_cs = _ucodesel; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } #define LINUX_VSYSCALL_START (-10UL << 20) #define LINUX_VSYSCALL_SZ 1024 const unsigned long linux_vsyscall_vector[] = { LINUX_SYS_gettimeofday, LINUX_SYS_linux_time, LINUX_SYS_linux_getcpu, }; static int linux_vsyscall(struct thread *td) { struct trapframe *frame; uint64_t retqaddr; int code, traced; int error; frame = td->td_frame; /* Check %rip for vsyscall area. */ if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START)) return (EINVAL); if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0) return (EINVAL); code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ; if (code >= nitems(linux_vsyscall_vector)) return (EINVAL); /* * vsyscall called as callq *(%rax), so we must * use return address from %rsp and also fixup %rsp. */ error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr)); if (error) return (error); frame->tf_rip = retqaddr; frame->tf_rax = linux_vsyscall_vector[code]; frame->tf_rsp += 8; traced = (frame->tf_flags & PSL_T); amd64_syscall(td, traced); return (0); } struct sysentvec elf_linux_sysvec = { .sv_size = LINUX_SYS_MAXSYSCALL, .sv_table = linux_sysent, .sv_transtrap = linux_translate_traps, .sv_fixup = linux_fixup_elf, .sv_sendsig = linux_rt_sendsig, .sv_sigcode = &_binary_linux_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_imgact_try = linux_exec_imgact_try, .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_stackprot = VM_PROT_ALL, .sv_copyout_auxargs = linux_copyout_auxargs, .sv_copyout_strings = linux_copyout_strings, .sv_setregs = linux_exec_setregs, .sv_fixlimit = NULL, .sv_maxssiz = NULL, .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN | SV_SIG_WAITNDQ | SV_TIMEKEEP, .sv_set_syscall_retval = linux_set_syscall_retval, .sv_fetch_syscall_args = linux_fetch_syscall_args, .sv_syscallnames = NULL, .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_onexec = linux_on_exec_vmspace, .sv_onexit = linux_on_exit, .sv_ontdexit = linux_thread_dtor, .sv_setid_allowed = &linux_setid_allowed_query, }; 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) 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_timekeep_base; 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); } 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 char GNULINUX_ABI_VENDOR[] = "GNU"; static int GNULINUX_ABI_DESC = 0; static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel) { const Elf32_Word *desc; uintptr_t p; p = (uintptr_t)(note + 1); p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); desc = (const Elf32_Word *)p; if (desc[0] != GNULINUX_ABI_DESC) return (false); /* * For Linux we encode osrel using the Linux convention of * (version << 16) | (major << 8) | (minor) * See macro in linux_mib.h */ *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); return (true); } static Elf_Brandnote linux64_brandnote = { .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR), .hdr.n_descsz = 16, .hdr.n_type = 1, .vendor = GNULINUX_ABI_VENDOR, .flags = BN_TRANSLATE_OSREL, .trans_osrel = linux_trans_osrel }; static Elf64_Brandinfo linux_glibc2brand = { .brand = ELFOSABI_LINUX, .machine = EM_X86_64, .compat_3_brand = "Linux", .emul_path = linux_emul_path, .interp_path = "/lib64/ld-linux-x86-64.so.2", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux64_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; static Elf64_Brandinfo linux_glibc2brandshort = { .brand = ELFOSABI_LINUX, .machine = EM_X86_64, .compat_3_brand = "Linux", .emul_path = linux_emul_path, .interp_path = "/lib64/ld-linux.so.2", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux64_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; static Elf64_Brandinfo linux_muslbrand = { .brand = ELFOSABI_LINUX, .machine = EM_X86_64, .compat_3_brand = "Linux", .emul_path = linux_emul_path, .interp_path = "/lib/ld-musl-x86_64.so.1", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux64_brandnote, - .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE + .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE | + LINUX_BI_FUTEX_REQUEUE }; 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 5176029605bf..901c79fcdda8 100644 --- a/sys/amd64/linux32/linux32_sysvec.c +++ b/sys/amd64/linux32/linux32_sysvec.c @@ -1,1224 +1,1225 @@ /*- * 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. */ #include "opt_compat.h" #include __FBSDID("$FreeBSD$"); #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 #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]; SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); static int linux_fixup_elf(uintptr_t *stack_base, struct image_params *iparams); 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 bool linux32_trans_osrel(const Elf_Note *note, int32_t *osrel); 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_syscall_retval(struct thread *td, int error); #define LINUX_T_UNKNOWN 255 static int _bsd_to_linux_trapcode[] = { LINUX_T_UNKNOWN, /* 0 */ 6, /* 1 T_PRIVINFLT */ LINUX_T_UNKNOWN, /* 2 */ 3, /* 3 T_BPTFLT */ LINUX_T_UNKNOWN, /* 4 */ LINUX_T_UNKNOWN, /* 5 */ 16, /* 6 T_ARITHTRAP */ 254, /* 7 T_ASTFLT */ LINUX_T_UNKNOWN, /* 8 */ 13, /* 9 T_PROTFLT */ 1, /* 10 T_TRCTRAP */ LINUX_T_UNKNOWN, /* 11 */ 14, /* 12 T_PAGEFLT */ LINUX_T_UNKNOWN, /* 13 */ 17, /* 14 T_ALIGNFLT */ LINUX_T_UNKNOWN, /* 15 */ LINUX_T_UNKNOWN, /* 16 */ LINUX_T_UNKNOWN, /* 17 */ 0, /* 18 T_DIVIDE */ 2, /* 19 T_NMI */ 4, /* 20 T_OFLOW */ 5, /* 21 T_BOUND */ 7, /* 22 T_DNA */ 8, /* 23 T_DOUBLEFLT */ 9, /* 24 T_FPOPFLT */ 10, /* 25 T_TSSFLT */ 11, /* 26 T_SEGNPFLT */ 12, /* 27 T_STKFLT */ 18, /* 28 T_MCHK */ 19, /* 29 T_XMMFLT */ 15 /* 30 T_RESERVED */ }; #define bsd_to_linux_trapcode(code) \ ((code)auxargs; argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, M_WAITOK | M_ZERO); issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0; 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, AT_PAGESZ, args->pagesz); /* * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0, * as it has appeared in the 2.4.0-rc7 first time. * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK), * glibc falls back to the hard-coded CLK_TCK value when aux entry * is not present. * Also see linux_times() implementation. */ if (linux_kernver(curthread) >= LINUX_KERNVER_2004000) AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); AUXARGS_ENTRY(pos, AT_PHENT, args->phent); AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); AUXARGS_ENTRY(pos, AT_BASE, args->base); AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary)); AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0); if (imgp->execpathp != 0) AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp)); if (args->execfd != -1) AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); AUXARGS_ENTRY(pos, AT_NULL, 0); free(imgp->auxargs, M_TEMP); imgp->auxargs = NULL; KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); error = copyout(argarray, (void *)base, sizeof(*argarray) * LINUX_AT_COUNT); free(argarray, M_TEMP); return (error); } static int linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp) { Elf32_Addr *base; base = (Elf32_Addr *)*stack_base; base--; if (suword32(base, (uint32_t)imgp->args->argc) == -1) return (EFAULT); *stack_base = (uintptr_t)base; return (0); } 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 = ksi->ksi_signo; 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_handler = PTROUT(catcher); frame.sf_sig = sig; frame.sf_siginfo = PTROUT(&fp->sf_si); frame.sf_ucontext = PTROUT(&fp->sf_sc); /* 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_sc.uc_flags = 0; /* XXX ??? */ frame.sf_sc.uc_link = 0; /* XXX ??? */ frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; frame.sf_sc.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_sc.uc_sigmask); frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask; frame.sf_sc.uc_mcontext.sc_edi = regs->tf_rdi; frame.sf_sc.uc_mcontext.sc_esi = regs->tf_rsi; frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_rbp; frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_rbx; frame.sf_sc.uc_mcontext.sc_esp = regs->tf_rsp; frame.sf_sc.uc_mcontext.sc_edx = regs->tf_rdx; frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_rcx; frame.sf_sc.uc_mcontext.sc_eax = regs->tf_rax; frame.sf_sc.uc_mcontext.sc_eip = regs->tf_rip; frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; frame.sf_sc.uc_mcontext.sc_gs = regs->tf_gs; frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs; frame.sf_sc.uc_mcontext.sc_es = regs->tf_es; frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds; frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags; frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp; frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss; frame.sf_sc.uc_mcontext.sc_err = regs->tf_err; frame.sf_sc.uc_mcontext.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; frame.sf_sc.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 = __kernel_rt_sigreturn; 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 = ksi->ksi_signo; 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_handler = PTROUT(catcher); frame.sf_sig = sig; 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); frame.sf_extramask[0] = lmask.__mask; 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 = __kernel_sigreturn; 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; sigset_t bmask; l_sigset_t lmask; 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. */ #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) eflags = frame.sf_sc.sc_eflags; if (!EFLAGS_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. */ #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 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); } lmask.__mask = frame.sf_sc.sc_mask; lmask.__mask = frame.sf_extramask[0]; linux_to_bsd_sigset(&lmask, &bmask); kern_sigprocmask(td, SIG_SETMASK, &bmask, 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. */ #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) eflags = context->sc_eflags; if (!EFLAGS_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. */ #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 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; /* Unconfirmed */ sa->code = frame->tf_rax; sa->original_code = sa->code; if (sa->code >= p->p_sysent->sv_size) /* nosys */ sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; else sa->callp = &p->p_sysent->sv_table[sa->code]; td->td_retval[0] = 0; 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); } } /* * 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); 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; /* Calculate string base and vector table pointers. */ if (imgp->execpath != NULL && imgp->auxargs != NULL) execpath_len = strlen(imgp->execpath) + 1; else execpath_len = 0; arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS; destp = (uintptr_t)arginfo; if (execpath_len != 0) { 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, ""); 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_transtrap = linux_translate_traps, .sv_fixup = linux_fixup_elf, .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_imgact_try = linux_exec_imgact_try, .sv_minsigstksz = LINUX_MINSIGSTKSZ, .sv_minuser = VM_MIN_ADDRESS, .sv_maxuser = LINUX32_MAXUSER, .sv_usrstack = LINUX32_USRSTACK, .sv_psstrings = LINUX32_PS_STRINGS, .sv_stackprot = VM_PROT_ALL, .sv_copyout_auxargs = linux_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 = NULL, .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_onexec = linux_on_exec_vmspace, .sv_onexit = linux_on_exit, .sv_ontdexit = linux_thread_dtor, .sv_setid_allowed = &linux_setid_allowed_query, }; 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) 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_timekeep_base; 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); } 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 char GNU_ABI_VENDOR[] = "GNU"; static int GNULINUX_ABI_DESC = 0; static bool linux32_trans_osrel(const Elf_Note *note, int32_t *osrel) { const Elf32_Word *desc; uintptr_t p; p = (uintptr_t)(note + 1); p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); desc = (const Elf32_Word *)p; if (desc[0] != GNULINUX_ABI_DESC) return (false); /* * For Linux we encode osrel using the Linux convention of * (version << 16) | (major << 8) | (minor) * See macro in linux_mib.h */ *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); return (true); } static Elf_Brandnote 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 = linux32_trans_osrel }; static Elf32_Brandinfo linux_brand = { .brand = ELFOSABI_LINUX, .machine = EM_386, .compat_3_brand = "Linux", .emul_path = linux_emul_path, .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", .emul_path = linux_emul_path, .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", .emul_path = linux_emul_path, .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 + .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE | + LINUX_BI_FUTEX_REQUEUE }; 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/compat/linux/linux.h b/sys/compat/linux/linux.h index 15830b5b387e..730aece66a6a 100644 --- a/sys/compat/linux/linux.h +++ b/sys/compat/linux/linux.h @@ -1,266 +1,271 @@ /*- * Copyright (c) 2015 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. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _LINUX_MI_H_ #define _LINUX_MI_H_ #include +/* + * Private Brandinfo flags + */ +#define LINUX_BI_FUTEX_REQUEUE 0x01000000 + /* * poll() */ #define LINUX_POLLIN 0x0001 #define LINUX_POLLPRI 0x0002 #define LINUX_POLLOUT 0x0004 #define LINUX_POLLERR 0x0008 #define LINUX_POLLHUP 0x0010 #define LINUX_POLLNVAL 0x0020 #define LINUX_POLLRDNORM 0x0040 #define LINUX_POLLRDBAND 0x0080 #define LINUX_POLLWRNORM 0x0100 #define LINUX_POLLWRBAND 0x0200 #define LINUX_POLLMSG 0x0400 #define LINUX_POLLREMOVE 0x1000 #define LINUX_POLLRDHUP 0x2000 #define LINUX_IFHWADDRLEN 6 #define LINUX_IFNAMSIZ 16 /* * Criteria for interface name translation */ #define IFP_IS_ETH(ifp) (ifp->if_type == IFT_ETHER) #define IFP_IS_LOOP(ifp) (ifp->if_type == IFT_LOOP) struct l_sockaddr { unsigned short sa_family; char sa_data[14]; }; #define LINUX_ARPHRD_ETHER 1 #define LINUX_ARPHRD_LOOPBACK 772 /* * Supported address families */ #define LINUX_AF_UNSPEC 0 #define LINUX_AF_UNIX 1 #define LINUX_AF_INET 2 #define LINUX_AF_AX25 3 #define LINUX_AF_IPX 4 #define LINUX_AF_APPLETALK 5 #define LINUX_AF_INET6 10 #define LINUX_AF_NETLINK 16 #define LINUX_NETLINK_ROUTE 0 #define LINUX_NETLINK_SOCK_DIAG 4 #define LINUX_NETLINK_NFLOG 5 #define LINUX_NETLINK_SELINUX 7 #define LINUX_NETLINK_AUDIT 9 #define LINUX_NETLINK_FIB_LOOKUP 10 #define LINUX_NETLINK_NETFILTER 12 #define LINUX_NETLINK_KOBJECT_UEVENT 15 /* * net device flags */ #define LINUX_IFF_UP 0x0001 #define LINUX_IFF_BROADCAST 0x0002 #define LINUX_IFF_DEBUG 0x0004 #define LINUX_IFF_LOOPBACK 0x0008 #define LINUX_IFF_POINTOPOINT 0x0010 #define LINUX_IFF_NOTRAILERS 0x0020 #define LINUX_IFF_RUNNING 0x0040 #define LINUX_IFF_NOARP 0x0080 #define LINUX_IFF_PROMISC 0x0100 #define LINUX_IFF_ALLMULTI 0x0200 #define LINUX_IFF_MASTER 0x0400 #define LINUX_IFF_SLAVE 0x0800 #define LINUX_IFF_MULTICAST 0x1000 #define LINUX_IFF_PORTSEL 0x2000 #define LINUX_IFF_AUTOMEDIA 0x4000 #define LINUX_IFF_DYNAMIC 0x8000 /* sigaltstack */ #define LINUX_SS_ONSTACK 1 #define LINUX_SS_DISABLE 2 int linux_to_bsd_sigaltstack(int lsa); int bsd_to_linux_sigaltstack(int bsa); /* sigset */ typedef struct { uint64_t __mask; } l_sigset_t; /* primitives to manipulate sigset_t */ #define LINUX_SIGEMPTYSET(set) (set).__mask = 0 #define LINUX_SIGISMEMBER(set, sig) (1UL & ((set).__mask >> _SIG_IDX(sig))) #define LINUX_SIGADDSET(set, sig) (set).__mask |= 1UL << _SIG_IDX(sig) void linux_to_bsd_sigset(l_sigset_t *, sigset_t *); void bsd_to_linux_sigset(sigset_t *, l_sigset_t *); /* signaling */ #define LINUX_SIGHUP 1 #define LINUX_SIGINT 2 #define LINUX_SIGQUIT 3 #define LINUX_SIGILL 4 #define LINUX_SIGTRAP 5 #define LINUX_SIGABRT 6 #define LINUX_SIGIOT LINUX_SIGABRT #define LINUX_SIGBUS 7 #define LINUX_SIGFPE 8 #define LINUX_SIGKILL 9 #define LINUX_SIGUSR1 10 #define LINUX_SIGSEGV 11 #define LINUX_SIGUSR2 12 #define LINUX_SIGPIPE 13 #define LINUX_SIGALRM 14 #define LINUX_SIGTERM 15 #define LINUX_SIGSTKFLT 16 #define LINUX_SIGCHLD 17 #define LINUX_SIGCONT 18 #define LINUX_SIGSTOP 19 #define LINUX_SIGTSTP 20 #define LINUX_SIGTTIN 21 #define LINUX_SIGTTOU 22 #define LINUX_SIGURG 23 #define LINUX_SIGXCPU 24 #define LINUX_SIGXFSZ 25 #define LINUX_SIGVTALRM 26 #define LINUX_SIGPROF 27 #define LINUX_SIGWINCH 28 #define LINUX_SIGIO 29 #define LINUX_SIGPOLL LINUX_SIGIO #define LINUX_SIGPWR 30 #define LINUX_SIGSYS 31 #define LINUX_SIGTBLSZ 31 #define LINUX_SIGRTMIN 32 #define LINUX_SIGRTMAX 64 #define LINUX_SIG_VALID(sig) ((sig) <= LINUX_SIGRTMAX && (sig) > 0) int linux_to_bsd_signal(int sig); int bsd_to_linux_signal(int sig); extern LIST_HEAD(futex_list, futex) futex_list; extern struct mtx futex_mtx; void linux_dev_shm_create(void); void linux_dev_shm_destroy(void); /* * mask=0 is not sensible for this application, so it will be taken to mean * a mask equivalent to the value. Otherwise, (word & mask) == value maps to * (word & ~mask) | value in a bitfield for the platform we're converting to. */ struct bsd_to_linux_bitmap { int bsd_mask; int bsd_value; int linux_mask; int linux_value; }; int bsd_to_linux_bits_(int value, struct bsd_to_linux_bitmap *bitmap, size_t mapcnt, int no_value); int linux_to_bsd_bits_(int value, struct bsd_to_linux_bitmap *bitmap, size_t mapcnt, int no_value); /* * These functions are used for simplification of BSD <-> Linux bit conversions. * Given `value`, a bit field, these functions will walk the given bitmap table * and set the appropriate bits for the target platform. If any bits were * successfully converted, then the return value is the equivalent of value * represented with the bit values appropriate for the target platform. * Otherwise, the value supplied as `no_value` is returned. */ #define bsd_to_linux_bits(_val, _bmap, _noval) \ bsd_to_linux_bits_((_val), (_bmap), nitems((_bmap)), (_noval)) #define linux_to_bsd_bits(_val, _bmap, _noval) \ linux_to_bsd_bits_((_val), (_bmap), nitems((_bmap)), (_noval)) /* * Easy mapping helpers. BITMAP_EASY_LINUX represents a single bit to be * translated, and the FreeBSD and Linux values are supplied. BITMAP_1t1_LINUX * is the extreme version of this, where not only is it a single bit, but the * name of the macro used to represent the Linux version of a bit literally has * LINUX_ prepended to the normal name. */ #define BITMAP_EASY_LINUX(_name, _linux_name) \ { \ .bsd_value = (_name), \ .linux_value = (_linux_name), \ } #define BITMAP_1t1_LINUX(_name) BITMAP_EASY_LINUX(_name, LINUX_##_name) int bsd_to_linux_errno(int error); void linux_check_errtbl(void); #define STATX_BASIC_STATS 0x07ff #define STATX_BTIME 0x0800 #define STATX_ALL 0x0fff #define STATX_ATTR_COMPRESSED 0x0004 #define STATX_ATTR_IMMUTABLE 0x0010 #define STATX_ATTR_APPEND 0x0020 #define STATX_ATTR_NODUMP 0x0040 #define STATX_ATTR_ENCRYPTED 0x0800 #define STATX_ATTR_AUTOMOUNT 0x1000 struct l_statx_timestamp { int64_t tv_sec; int32_t tv_nsec; int32_t __spare0; }; struct l_statx { uint32_t stx_mask; uint32_t stx_blksize; uint64_t stx_attributes; uint32_t stx_nlink; uint32_t stx_uid; uint32_t stx_gid; uint16_t stx_mode; uint16_t __spare0[1]; uint64_t stx_ino; uint64_t stx_size; uint64_t stx_blocks; uint64_t stx_attributes_mask; struct l_statx_timestamp stx_atime; struct l_statx_timestamp stx_btime; struct l_statx_timestamp stx_ctime; struct l_statx_timestamp stx_mtime; uint32_t stx_rdev_major; uint32_t stx_rdev_minor; uint32_t stx_dev_major; uint32_t stx_dev_minor; uint64_t stx_mnt_id; uint64_t __spare2[13]; }; #endif /* _LINUX_MI_H_ */ diff --git a/sys/compat/linux/linux_futex.c b/sys/compat/linux/linux_futex.c index 1ba11ed3fc57..2ee18c26cf24 100644 --- a/sys/compat/linux/linux_futex.c +++ b/sys/compat/linux/linux_futex.c @@ -1,1254 +1,1269 @@ /* $NetBSD: linux_futex.c,v 1.7 2006/07/24 19:01:49 manu Exp $ */ /*- * SPDX-License-Identifier: BSD-4-Clause * * Copyright (c) 2005 Emmanuel Dreyfus * All rights reserved. * Copyright (c) 2009-2016 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. * 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 Emmanuel Dreyfus * 4. 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 THE AUTHOR AND CONTRIBUTORS ``AS IS'' * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #if 0 __KERNEL_RCSID(1, "$NetBSD: linux_futex.c,v 1.7 2006/07/24 19:01:49 manu Exp $"); #endif #include "opt_compat.h" #include #include #include +#include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef COMPAT_LINUX32 #include #include #else #include #include #endif #include #include #include #include #include /* DTrace init */ LIN_SDT_PROVIDER_DECLARE(LINUX_DTRACE); /** * Futex part for the special DTrace module "locks". */ LIN_SDT_PROBE_DEFINE1(locks, futex_mtx, locked, "struct mtx *"); LIN_SDT_PROBE_DEFINE1(locks, futex_mtx, unlock, "struct mtx *"); /** * Per futex probes. */ LIN_SDT_PROBE_DEFINE1(futex, futex, create, "struct sx *"); LIN_SDT_PROBE_DEFINE1(futex, futex, destroy, "struct sx *"); /** * DTrace probes in this module. */ LIN_SDT_PROBE_DEFINE3(futex, futex_put, destroy, "uint32_t *", "uint32_t", "int"); LIN_SDT_PROBE_DEFINE3(futex, futex_put, unlock, "uint32_t *", "uint32_t", "int"); LIN_SDT_PROBE_DEFINE1(futex, futex_get0, umtx_key_get_error, "int"); LIN_SDT_PROBE_DEFINE3(futex, futex_get0, shared, "uint32_t *", "uint32_t", "int"); LIN_SDT_PROBE_DEFINE1(futex, futex_get0, null, "uint32_t *"); LIN_SDT_PROBE_DEFINE3(futex, futex_get0, new, "uint32_t *", "uint32_t", "int"); LIN_SDT_PROBE_DEFINE0(futex, futex_get, error); LIN_SDT_PROBE_DEFINE5(futex, futex_sleep, requeue_error, "int", "uint32_t *", "struct waiting_proc *", "uint32_t *", "uint32_t"); LIN_SDT_PROBE_DEFINE3(futex, futex_sleep, sleep_error, "int", "uint32_t *", "struct waiting_proc *"); LIN_SDT_PROBE_DEFINE3(futex, futex_wake, iterate, "uint32_t", "struct waiting_proc *", "uint32_t"); LIN_SDT_PROBE_DEFINE1(futex, futex_wake, wakeup, "struct waiting_proc *"); LIN_SDT_PROBE_DEFINE1(futex, futex_requeue, wakeup, "struct waiting_proc *"); LIN_SDT_PROBE_DEFINE3(futex, futex_requeue, requeue, "uint32_t *", "struct waiting_proc *", "uint32_t"); LIN_SDT_PROBE_DEFINE1(futex, futex_wait, sleep_error, "int"); LIN_SDT_PROBE_DEFINE4(futex, futex_atomic_op, decoded_op, "int", "int", "int", "int"); LIN_SDT_PROBE_DEFINE0(futex, futex_atomic_op, missing_access_check); LIN_SDT_PROBE_DEFINE1(futex, futex_atomic_op, unimplemented_op, "int"); LIN_SDT_PROBE_DEFINE1(futex, futex_atomic_op, unimplemented_cmp, "int"); LIN_SDT_PROBE_DEFINE0(futex, linux_futex, unimplemented_clockswitch); LIN_SDT_PROBE_DEFINE1(futex, linux_futex, copyin_error, "int"); LIN_SDT_PROBE_DEFINE0(futex, linux_futex, invalid_cmp_requeue_use); LIN_SDT_PROBE_DEFINE3(futex, linux_futex, debug_wait, "uint32_t *", "uint32_t", "uint32_t"); LIN_SDT_PROBE_DEFINE4(futex, linux_futex, debug_wait_value_neq, "uint32_t *", "uint32_t", "int", "uint32_t"); LIN_SDT_PROBE_DEFINE3(futex, linux_futex, debug_wake, "uint32_t *", "uint32_t", "uint32_t"); LIN_SDT_PROBE_DEFINE5(futex, linux_futex, debug_cmp_requeue, "uint32_t *", "uint32_t", "uint32_t", "uint32_t *", "struct l_timespec *"); LIN_SDT_PROBE_DEFINE2(futex, linux_futex, debug_cmp_requeue_value_neq, "uint32_t", "int"); LIN_SDT_PROBE_DEFINE5(futex, linux_futex, debug_wake_op, "uint32_t *", "int", "uint32_t", "uint32_t *", "uint32_t"); LIN_SDT_PROBE_DEFINE0(futex, linux_futex, unhandled_efault); LIN_SDT_PROBE_DEFINE0(futex, linux_futex, unimplemented_lock_pi); LIN_SDT_PROBE_DEFINE0(futex, linux_futex, unimplemented_unlock_pi); LIN_SDT_PROBE_DEFINE0(futex, linux_futex, unimplemented_trylock_pi); LIN_SDT_PROBE_DEFINE0(futex, linux_futex, deprecated_requeue); LIN_SDT_PROBE_DEFINE0(futex, linux_futex, unimplemented_wait_requeue_pi); LIN_SDT_PROBE_DEFINE0(futex, linux_futex, unimplemented_cmp_requeue_pi); LIN_SDT_PROBE_DEFINE1(futex, linux_futex, unknown_operation, "int"); LIN_SDT_PROBE_DEFINE0(futex, linux_set_robust_list, size_error); LIN_SDT_PROBE_DEFINE1(futex, linux_get_robust_list, copyout_error, "int"); LIN_SDT_PROBE_DEFINE1(futex, handle_futex_death, copyin_error, "int"); LIN_SDT_PROBE_DEFINE1(futex, fetch_robust_entry, copyin_error, "int"); LIN_SDT_PROBE_DEFINE1(futex, release_futexes, copyin_error, "int"); struct futex; struct waiting_proc { uint32_t wp_flags; struct futex *wp_futex; TAILQ_ENTRY(waiting_proc) wp_list; }; struct futex { struct mtx f_lck; uint32_t *f_uaddr; /* user-supplied value, for debug */ struct umtx_key f_key; uint32_t f_refcount; uint32_t f_bitset; LIST_ENTRY(futex) f_list; TAILQ_HEAD(lf_waiting_proc, waiting_proc) f_waiting_proc; }; #define FUTEX_LOCK(f) mtx_lock(&(f)->f_lck) #define FUTEX_LOCKED(f) mtx_owned(&(f)->f_lck) #define FUTEX_UNLOCK(f) mtx_unlock(&(f)->f_lck) #define FUTEX_INIT(f) do { \ mtx_init(&(f)->f_lck, "ftlk", NULL, \ MTX_DUPOK); \ LIN_SDT_PROBE1(futex, futex, create, \ &(f)->f_lck); \ } while (0) #define FUTEX_DESTROY(f) do { \ LIN_SDT_PROBE1(futex, futex, destroy, \ &(f)->f_lck); \ mtx_destroy(&(f)->f_lck); \ } while (0) #define FUTEX_ASSERT_LOCKED(f) mtx_assert(&(f)->f_lck, MA_OWNED) #define FUTEX_ASSERT_UNLOCKED(f) mtx_assert(&(f)->f_lck, MA_NOTOWNED) #define FUTEXES_LOCK do { \ mtx_lock(&futex_mtx); \ LIN_SDT_PROBE1(locks, futex_mtx, \ locked, &futex_mtx); \ } while (0) #define FUTEXES_UNLOCK do { \ LIN_SDT_PROBE1(locks, futex_mtx, \ unlock, &futex_mtx); \ mtx_unlock(&futex_mtx); \ } while (0) /* flags for futex_get() */ #define FUTEX_CREATE_WP 0x1 /* create waiting_proc */ #define FUTEX_DONTCREATE 0x2 /* don't create futex if not exists */ #define FUTEX_DONTEXISTS 0x4 /* return EINVAL if futex exists */ #define FUTEX_SHARED 0x8 /* shared futex */ #define FUTEX_DONTLOCK 0x10 /* don't lock futex */ /* wp_flags */ #define FUTEX_WP_REQUEUED 0x1 /* wp requeued - wp moved from wp_list * of futex where thread sleep to wp_list * of another futex. */ #define FUTEX_WP_REMOVED 0x2 /* wp is woken up and removed from futex * wp_list to prevent double wakeup. */ static void futex_put(struct futex *, struct waiting_proc *); static int futex_get0(uint32_t *, struct futex **f, uint32_t); static int futex_get(uint32_t *, struct waiting_proc **, struct futex **, uint32_t); static int futex_sleep(struct futex *, struct waiting_proc *, struct timespec *); static int futex_wake(struct futex *, int, uint32_t); static int futex_requeue(struct futex *, int, struct futex *, int); static int futex_wait(struct futex *, struct waiting_proc *, struct timespec *, uint32_t); static void futex_lock(struct futex *); static void futex_unlock(struct futex *); static int futex_atomic_op(struct thread *, int, uint32_t *); static int handle_futex_death(struct linux_emuldata *, uint32_t *, unsigned int); static int fetch_robust_entry(struct linux_robust_list **, struct linux_robust_list **, unsigned int *); struct linux_futex_args { uint32_t *uaddr; int32_t op; uint32_t flags; bool clockrt; uint32_t val; struct timespec *ts; uint32_t *uaddr2; uint32_t val3; + bool val3_compare; struct timespec kts; }; static int linux_futex(struct thread *, struct linux_futex_args *); static int linux_futex_wait(struct thread *, struct linux_futex_args *); static int linux_futex_wake(struct thread *, struct linux_futex_args *); static int linux_futex_requeue(struct thread *, struct linux_futex_args *); static int linux_futex_wakeop(struct thread *, struct linux_futex_args *); static void futex_put(struct futex *f, struct waiting_proc *wp) { if (wp != NULL) { if ((wp->wp_flags & FUTEX_WP_REMOVED) == 0) TAILQ_REMOVE(&f->f_waiting_proc, wp, wp_list); free(wp, M_FUTEX_WP); } FUTEXES_LOCK; if (--f->f_refcount == 0) { LIST_REMOVE(f, f_list); FUTEXES_UNLOCK; if (FUTEX_LOCKED(f)) futex_unlock(f); LIN_SDT_PROBE3(futex, futex_put, destroy, f->f_uaddr, f->f_refcount, f->f_key.shared); LINUX_CTR3(sys_futex, "futex_put destroy uaddr %p ref %d " "shared %d", f->f_uaddr, f->f_refcount, f->f_key.shared); umtx_key_release(&f->f_key); FUTEX_DESTROY(f); free(f, M_FUTEX); return; } LIN_SDT_PROBE3(futex, futex_put, unlock, f->f_uaddr, f->f_refcount, f->f_key.shared); LINUX_CTR3(sys_futex, "futex_put uaddr %p ref %d shared %d", f->f_uaddr, f->f_refcount, f->f_key.shared); if (FUTEX_LOCKED(f)) futex_unlock(f); FUTEXES_UNLOCK; } static int futex_get0(uint32_t *uaddr, struct futex **newf, uint32_t flags) { struct futex *f, *tmpf; struct umtx_key key; int error; *newf = tmpf = NULL; error = umtx_key_get(uaddr, TYPE_FUTEX, (flags & FUTEX_SHARED) ? AUTO_SHARE : THREAD_SHARE, &key); if (error) { LIN_SDT_PROBE1(futex, futex_get0, umtx_key_get_error, error); return (error); } retry: FUTEXES_LOCK; LIST_FOREACH(f, &futex_list, f_list) { if (umtx_key_match(&f->f_key, &key)) { if (tmpf != NULL) { if (FUTEX_LOCKED(tmpf)) futex_unlock(tmpf); FUTEX_DESTROY(tmpf); free(tmpf, M_FUTEX); } if (flags & FUTEX_DONTEXISTS) { FUTEXES_UNLOCK; umtx_key_release(&key); return (EINVAL); } /* * Increment refcount of the found futex to * prevent it from deallocation before FUTEX_LOCK() */ ++f->f_refcount; FUTEXES_UNLOCK; umtx_key_release(&key); if ((flags & FUTEX_DONTLOCK) == 0) futex_lock(f); *newf = f; LIN_SDT_PROBE3(futex, futex_get0, shared, uaddr, f->f_refcount, f->f_key.shared); LINUX_CTR3(sys_futex, "futex_get uaddr %p ref %d shared %d", uaddr, f->f_refcount, f->f_key.shared); return (0); } } if (flags & FUTEX_DONTCREATE) { FUTEXES_UNLOCK; umtx_key_release(&key); LIN_SDT_PROBE1(futex, futex_get0, null, uaddr); LINUX_CTR1(sys_futex, "futex_get uaddr %p null", uaddr); return (0); } if (tmpf == NULL) { FUTEXES_UNLOCK; tmpf = malloc(sizeof(*tmpf), M_FUTEX, M_WAITOK | M_ZERO); tmpf->f_uaddr = uaddr; tmpf->f_key = key; tmpf->f_refcount = 1; tmpf->f_bitset = FUTEX_BITSET_MATCH_ANY; FUTEX_INIT(tmpf); TAILQ_INIT(&tmpf->f_waiting_proc); /* * Lock the new futex before an insert into the futex_list * to prevent futex usage by other. */ if ((flags & FUTEX_DONTLOCK) == 0) futex_lock(tmpf); goto retry; } LIST_INSERT_HEAD(&futex_list, tmpf, f_list); FUTEXES_UNLOCK; LIN_SDT_PROBE3(futex, futex_get0, new, uaddr, tmpf->f_refcount, tmpf->f_key.shared); LINUX_CTR3(sys_futex, "futex_get uaddr %p ref %d shared %d new", uaddr, tmpf->f_refcount, tmpf->f_key.shared); *newf = tmpf; return (0); } static int futex_get(uint32_t *uaddr, struct waiting_proc **wp, struct futex **f, uint32_t flags) { int error; if (flags & FUTEX_CREATE_WP) { *wp = malloc(sizeof(struct waiting_proc), M_FUTEX_WP, M_WAITOK); (*wp)->wp_flags = 0; } error = futex_get0(uaddr, f, flags); if (error) { LIN_SDT_PROBE0(futex, futex_get, error); if (flags & FUTEX_CREATE_WP) free(*wp, M_FUTEX_WP); return (error); } if (flags & FUTEX_CREATE_WP) { TAILQ_INSERT_HEAD(&(*f)->f_waiting_proc, *wp, wp_list); (*wp)->wp_futex = *f; } return (error); } static inline void futex_lock(struct futex *f) { LINUX_CTR3(sys_futex, "futex_lock uaddr %p ref %d shared %d", f->f_uaddr, f->f_refcount, f->f_key.shared); FUTEX_ASSERT_UNLOCKED(f); FUTEX_LOCK(f); } static inline void futex_unlock(struct futex *f) { LINUX_CTR3(sys_futex, "futex_unlock uaddr %p ref %d shared %d", f->f_uaddr, f->f_refcount, f->f_key.shared); FUTEX_ASSERT_LOCKED(f); FUTEX_UNLOCK(f); } static int futex_sleep(struct futex *f, struct waiting_proc *wp, struct timespec *ts) { sbintime_t sbt, prec, tmp; time_t over; int error; FUTEX_ASSERT_LOCKED(f); if (ts != NULL) { if (ts->tv_sec > INT32_MAX / 2) { over = ts->tv_sec - INT32_MAX / 2; ts->tv_sec -= over; } tmp = tstosbt(*ts); if (TIMESEL(&sbt, tmp)) sbt += tc_tick_sbt; sbt += tmp; prec = tmp; prec >>= tc_precexp; } else { sbt = 0; prec = 0; } LINUX_CTR4(sys_futex, "futex_sleep enter uaddr %p wp %p timo %ld ref %d", f->f_uaddr, wp, sbt, f->f_refcount); error = msleep_sbt(wp, &f->f_lck, PCATCH, "futex", sbt, prec, C_ABSOLUTE); if (wp->wp_flags & FUTEX_WP_REQUEUED) { KASSERT(f != wp->wp_futex, ("futex != wp_futex")); if (error) { LIN_SDT_PROBE5(futex, futex_sleep, requeue_error, error, f->f_uaddr, wp, wp->wp_futex->f_uaddr, wp->wp_futex->f_refcount); } LINUX_CTR5(sys_futex, "futex_sleep out error %d uaddr %p wp" " %p requeued uaddr %p ref %d", error, f->f_uaddr, wp, wp->wp_futex->f_uaddr, wp->wp_futex->f_refcount); futex_put(f, NULL); f = wp->wp_futex; futex_lock(f); } else { if (error) { LIN_SDT_PROBE3(futex, futex_sleep, sleep_error, error, f->f_uaddr, wp); } LINUX_CTR3(sys_futex, "futex_sleep out error %d uaddr %p wp %p", error, f->f_uaddr, wp); } futex_put(f, wp); return (error); } static int futex_wake(struct futex *f, int n, uint32_t bitset) { struct waiting_proc *wp, *wpt; int count = 0; if (bitset == 0) return (EINVAL); FUTEX_ASSERT_LOCKED(f); TAILQ_FOREACH_SAFE(wp, &f->f_waiting_proc, wp_list, wpt) { LIN_SDT_PROBE3(futex, futex_wake, iterate, f->f_uaddr, wp, f->f_refcount); LINUX_CTR3(sys_futex, "futex_wake uaddr %p wp %p ref %d", f->f_uaddr, wp, f->f_refcount); /* * Unless we find a matching bit in * the bitset, continue searching. */ if (!(wp->wp_futex->f_bitset & bitset)) continue; wp->wp_flags |= FUTEX_WP_REMOVED; TAILQ_REMOVE(&f->f_waiting_proc, wp, wp_list); LIN_SDT_PROBE1(futex, futex_wake, wakeup, wp); wakeup_one(wp); if (++count == n) break; } return (count); } static int futex_requeue(struct futex *f, int nrwake, struct futex *f2, int nrrequeue) { struct waiting_proc *wp, *wpt; int count = 0; FUTEX_ASSERT_LOCKED(f); FUTEX_ASSERT_LOCKED(f2); TAILQ_FOREACH_SAFE(wp, &f->f_waiting_proc, wp_list, wpt) { if (++count <= nrwake) { LINUX_CTR2(sys_futex, "futex_req_wake uaddr %p wp %p", f->f_uaddr, wp); wp->wp_flags |= FUTEX_WP_REMOVED; TAILQ_REMOVE(&f->f_waiting_proc, wp, wp_list); LIN_SDT_PROBE1(futex, futex_requeue, wakeup, wp); wakeup_one(wp); } else { LIN_SDT_PROBE3(futex, futex_requeue, requeue, f->f_uaddr, wp, f2->f_uaddr); LINUX_CTR3(sys_futex, "futex_requeue uaddr %p wp %p to %p", f->f_uaddr, wp, f2->f_uaddr); wp->wp_flags |= FUTEX_WP_REQUEUED; /* Move wp to wp_list of f2 futex */ TAILQ_REMOVE(&f->f_waiting_proc, wp, wp_list); TAILQ_INSERT_HEAD(&f2->f_waiting_proc, wp, wp_list); /* * Thread which sleeps on wp after waking should * acquire f2 lock, so increment refcount of f2 to * prevent it from premature deallocation. */ wp->wp_futex = f2; FUTEXES_LOCK; ++f2->f_refcount; FUTEXES_UNLOCK; if (count - nrwake >= nrrequeue) break; } } return (count); } static int futex_wait(struct futex *f, struct waiting_proc *wp, struct timespec *ts, uint32_t bitset) { int error; if (bitset == 0) { futex_put(f, wp); return (EINVAL); } f->f_bitset = bitset; error = futex_sleep(f, wp, ts); if (error) LIN_SDT_PROBE1(futex, futex_wait, sleep_error, error); if (error == EWOULDBLOCK) error = ETIMEDOUT; return (error); } static int futex_atomic_op(struct thread *td, int encoded_op, uint32_t *uaddr) { int op = (encoded_op >> 28) & 7; int cmp = (encoded_op >> 24) & 15; int oparg = (encoded_op << 8) >> 20; int cmparg = (encoded_op << 20) >> 20; int oldval = 0, ret; if (encoded_op & (FUTEX_OP_OPARG_SHIFT << 28)) oparg = 1 << oparg; LIN_SDT_PROBE4(futex, futex_atomic_op, decoded_op, op, cmp, oparg, cmparg); /* XXX: Linux verifies access here and returns EFAULT */ LIN_SDT_PROBE0(futex, futex_atomic_op, missing_access_check); switch (op) { case FUTEX_OP_SET: ret = futex_xchgl(oparg, uaddr, &oldval); break; case FUTEX_OP_ADD: ret = futex_addl(oparg, uaddr, &oldval); break; case FUTEX_OP_OR: ret = futex_orl(oparg, uaddr, &oldval); break; case FUTEX_OP_ANDN: ret = futex_andl(~oparg, uaddr, &oldval); break; case FUTEX_OP_XOR: ret = futex_xorl(oparg, uaddr, &oldval); break; default: LIN_SDT_PROBE1(futex, futex_atomic_op, unimplemented_op, op); ret = -ENOSYS; break; } if (ret) return (ret); switch (cmp) { case FUTEX_OP_CMP_EQ: ret = (oldval == cmparg); break; case FUTEX_OP_CMP_NE: ret = (oldval != cmparg); break; case FUTEX_OP_CMP_LT: ret = (oldval < cmparg); break; case FUTEX_OP_CMP_GE: ret = (oldval >= cmparg); break; case FUTEX_OP_CMP_LE: ret = (oldval <= cmparg); break; case FUTEX_OP_CMP_GT: ret = (oldval > cmparg); break; default: LIN_SDT_PROBE1(futex, futex_atomic_op, unimplemented_cmp, cmp); ret = -ENOSYS; } return (ret); } static int linux_futex(struct thread *td, struct linux_futex_args *args) { struct linux_pemuldata *pem; + struct proc *p; if (args->op & LINUX_FUTEX_PRIVATE_FLAG) { args->flags = 0; args->op &= ~LINUX_FUTEX_PRIVATE_FLAG; } else args->flags = FUTEX_SHARED; /* * Currently support for switching between CLOCK_MONOTONIC and * CLOCK_REALTIME is not present. However Linux forbids the use of * FUTEX_CLOCK_REALTIME with any op except FUTEX_WAIT_BITSET and * FUTEX_WAIT_REQUEUE_PI. */ args->clockrt = args->op & LINUX_FUTEX_CLOCK_REALTIME; args->op = args->op & ~LINUX_FUTEX_CLOCK_REALTIME; if (args->clockrt && args->op != LINUX_FUTEX_WAIT_BITSET && args->op != LINUX_FUTEX_WAIT_REQUEUE_PI) { LIN_SDT_PROBE0(futex, linux_futex, unimplemented_clockswitch); return (ENOSYS); } switch (args->op) { case LINUX_FUTEX_WAIT: args->val3 = FUTEX_BITSET_MATCH_ANY; /* FALLTHROUGH */ case LINUX_FUTEX_WAIT_BITSET: LIN_SDT_PROBE3(futex, linux_futex, debug_wait, args->uaddr, args->val, args->val3); LINUX_CTR3(sys_futex, "WAIT uaddr %p val 0x%x bitset 0x%x", args->uaddr, args->val, args->val3); return (linux_futex_wait(td, args)); case LINUX_FUTEX_WAKE: args->val3 = FUTEX_BITSET_MATCH_ANY; /* FALLTHROUGH */ case LINUX_FUTEX_WAKE_BITSET: LIN_SDT_PROBE3(futex, linux_futex, debug_wake, args->uaddr, args->val, args->val3); LINUX_CTR3(sys_futex, "WAKE uaddr %p nrwake 0x%x bitset 0x%x", args->uaddr, args->val, args->val3); return (linux_futex_wake(td, args)); + case LINUX_FUTEX_REQUEUE: + /* + * Glibc does not use this operation since version 2.3.3, + * as it is racy and replaced by FUTEX_CMP_REQUEUE operation. + * Glibc versions prior to 2.3.3 fall back to FUTEX_WAKE when + * FUTEX_REQUEUE returned EINVAL. + */ + pem = pem_find(td->td_proc); + if ((pem->flags & LINUX_XDEPR_REQUEUEOP) == 0) { + linux_msg(td, "unsupported FUTEX_REQUEUE"); + pem->flags |= LINUX_XDEPR_REQUEUEOP; + LIN_SDT_PROBE0(futex, linux_futex, + deprecated_requeue); + } + + /* + * The above is true, however musl libc does make use of the + * futex requeue operation, allow operation for brands which + * set LINUX_BI_FUTEX_REQUEUE bit of Brandinfo flags. + */ + p = td->td_proc; + Elf_Brandinfo *bi = p->p_elf_brandinfo; + if (bi == NULL || ((bi->flags & LINUX_BI_FUTEX_REQUEUE)) == 0) + return (EINVAL); + args->val3_compare = false; + /* FALLTHROUGH */ + case LINUX_FUTEX_CMP_REQUEUE: LIN_SDT_PROBE5(futex, linux_futex, debug_cmp_requeue, args->uaddr, args->val, args->val3, args->uaddr2, args->ts); LINUX_CTR5(sys_futex, "CMP_REQUEUE uaddr %p " "nrwake 0x%x uval 0x%x uaddr2 %p nrequeue 0x%x", args->uaddr, args->val, args->val3, args->uaddr2, args->ts); return (linux_futex_requeue(td, args)); case LINUX_FUTEX_WAKE_OP: LIN_SDT_PROBE5(futex, linux_futex, debug_wake_op, args->uaddr, args->op, args->val, args->uaddr2, args->val3); LINUX_CTR5(sys_futex, "WAKE_OP " "uaddr %p nrwake 0x%x uaddr2 %p op 0x%x nrwake2 0x%x", args->uaddr, args->val, args->uaddr2, args->val3, args->ts); return (linux_futex_wakeop(td, args)); case LINUX_FUTEX_LOCK_PI: /* not yet implemented */ pem = pem_find(td->td_proc); if ((pem->flags & LINUX_XUNSUP_FUTEXPIOP) == 0) { linux_msg(td, "unsupported FUTEX_LOCK_PI"); pem->flags |= LINUX_XUNSUP_FUTEXPIOP; LIN_SDT_PROBE0(futex, linux_futex, unimplemented_lock_pi); } return (ENOSYS); case LINUX_FUTEX_UNLOCK_PI: /* not yet implemented */ pem = pem_find(td->td_proc); if ((pem->flags & LINUX_XUNSUP_FUTEXPIOP) == 0) { linux_msg(td, "unsupported FUTEX_UNLOCK_PI"); pem->flags |= LINUX_XUNSUP_FUTEXPIOP; LIN_SDT_PROBE0(futex, linux_futex, unimplemented_unlock_pi); } return (ENOSYS); case LINUX_FUTEX_TRYLOCK_PI: /* not yet implemented */ pem = pem_find(td->td_proc); if ((pem->flags & LINUX_XUNSUP_FUTEXPIOP) == 0) { linux_msg(td, "unsupported FUTEX_TRYLOCK_PI"); pem->flags |= LINUX_XUNSUP_FUTEXPIOP; LIN_SDT_PROBE0(futex, linux_futex, unimplemented_trylock_pi); } return (ENOSYS); - case LINUX_FUTEX_REQUEUE: - /* - * Glibc does not use this operation since version 2.3.3, - * as it is racy and replaced by FUTEX_CMP_REQUEUE operation. - * Glibc versions prior to 2.3.3 fall back to FUTEX_WAKE when - * FUTEX_REQUEUE returned EINVAL. - */ - pem = pem_find(td->td_proc); - if ((pem->flags & LINUX_XDEPR_REQUEUEOP) == 0) { - linux_msg(td, "unsupported FUTEX_REQUEUE"); - pem->flags |= LINUX_XDEPR_REQUEUEOP; - LIN_SDT_PROBE0(futex, linux_futex, - deprecated_requeue); - } - return (EINVAL); - case LINUX_FUTEX_WAIT_REQUEUE_PI: /* not yet implemented */ pem = pem_find(td->td_proc); if ((pem->flags & LINUX_XUNSUP_FUTEXPIOP) == 0) { linux_msg(td, "unsupported FUTEX_WAIT_REQUEUE_PI"); pem->flags |= LINUX_XUNSUP_FUTEXPIOP; LIN_SDT_PROBE0(futex, linux_futex, unimplemented_wait_requeue_pi); } return (ENOSYS); case LINUX_FUTEX_CMP_REQUEUE_PI: /* not yet implemented */ pem = pem_find(td->td_proc); if ((pem->flags & LINUX_XUNSUP_FUTEXPIOP) == 0) { linux_msg(td, "unsupported FUTEX_CMP_REQUEUE_PI"); pem->flags |= LINUX_XUNSUP_FUTEXPIOP; LIN_SDT_PROBE0(futex, linux_futex, unimplemented_cmp_requeue_pi); } return (ENOSYS); default: linux_msg(td, "unsupported futex op %d", args->op); LIN_SDT_PROBE1(futex, linux_futex, unknown_operation, args->op); return (ENOSYS); } } static int linux_futex_wakeop(struct thread *td, struct linux_futex_args *args) { int nrwake, op_ret, ret; struct futex *f, *f2; int error, save; uint32_t val; if (args->uaddr == args->uaddr2) return (EINVAL); retry: f = f2 = NULL; error = futex_get(args->uaddr, NULL, &f, args->flags | FUTEX_DONTLOCK); if (error != 0) return (error); error = futex_get(args->uaddr2, NULL, &f2, args->flags | FUTEX_DONTLOCK); if (error != 0) { futex_put(f, NULL); return (error); } futex_lock(f); futex_lock(f2); /* * This function returns positive number as results and * negative as errors */ save = vm_fault_disable_pagefaults(); op_ret = futex_atomic_op(td, args->val3, args->uaddr2); vm_fault_enable_pagefaults(save); LINUX_CTR2(sys_futex, "WAKE_OP atomic_op uaddr %p ret 0x%x", args->uaddr, op_ret); if (op_ret < 0) { if (f2 != NULL) futex_put(f2, NULL); futex_put(f, NULL); error = copyin(args->uaddr2, &val, sizeof(val)); if (error == 0) goto retry; return (error); } ret = futex_wake(f, args->val, args->val3); if (op_ret > 0) { op_ret = 0; nrwake = (int)(unsigned long)args->ts; if (f2 != NULL) op_ret += futex_wake(f2, nrwake, args->val3); else op_ret += futex_wake(f, nrwake, args->val3); ret += op_ret; } if (f2 != NULL) futex_put(f2, NULL); futex_put(f, NULL); td->td_retval[0] = ret; return (0); } static int linux_futex_requeue(struct thread *td, struct linux_futex_args *args) { int nrwake, nrrequeue; struct futex *f, *f2; int error; uint32_t val; /* * Linux allows this, we would not, it is an incorrect * usage of declared ABI, so return EINVAL. */ if (args->uaddr == args->uaddr2) { LIN_SDT_PROBE0(futex, linux_futex, invalid_cmp_requeue_use); return (EINVAL); } nrrequeue = (int)(unsigned long)args->ts; nrwake = args->val; /* * Sanity check to prevent signed integer overflow, * see Linux CVE-2018-6927 */ if (nrwake < 0 || nrrequeue < 0) return (EINVAL); retry: f = f2 = NULL; error = futex_get(args->uaddr, NULL, &f, args->flags | FUTEX_DONTLOCK); if (error != 0) return (error); /* * To avoid deadlocks return EINVAL if second futex * exists at this time. * * Glibc fall back to FUTEX_WAKE in case of any error * returned by FUTEX_CMP_REQUEUE. */ error = futex_get(args->uaddr2, NULL, &f2, args->flags | FUTEX_DONTEXISTS | FUTEX_DONTLOCK); if (error != 0) { futex_put(f, NULL); return (error); } futex_lock(f); futex_lock(f2); error = copyin_nofault(args->uaddr, &val, sizeof(val)); if (error != 0) { futex_put(f2, NULL); futex_put(f, NULL); error = copyin(args->uaddr, &val, sizeof(val)); if (error == 0) goto retry; LIN_SDT_PROBE1(futex, linux_futex, copyin_error, error); LINUX_CTR1(sys_futex, "CMP_REQUEUE copyin failed %d", error); return (error); } - if (val != args->val3) { + if (args->val3_compare == true && val != args->val3) { LIN_SDT_PROBE2(futex, linux_futex, debug_cmp_requeue_value_neq, args->val, val); LINUX_CTR2(sys_futex, "CMP_REQUEUE val 0x%x != uval 0x%x", args->val, val); futex_put(f2, NULL); futex_put(f, NULL); return (EAGAIN); } td->td_retval[0] = futex_requeue(f, nrwake, f2, nrrequeue); futex_put(f2, NULL); futex_put(f, NULL); return (0); } static int linux_futex_wake(struct thread *td, struct linux_futex_args *args) { struct futex *f; int error; f = NULL; error = futex_get(args->uaddr, NULL, &f, args->flags | FUTEX_DONTCREATE); if (error != 0) return (error); if (f == NULL) { td->td_retval[0] = 0; return (error); } td->td_retval[0] = futex_wake(f, args->val, args->val3); futex_put(f, NULL); return (0); } static int linux_futex_wait(struct thread *td, struct linux_futex_args *args) { struct waiting_proc *wp; struct timespec kts; struct futex *f; int error; uint32_t val; if (args->ts != NULL) { if (args->clockrt) { nanotime(&kts); timespecsub(args->ts, &kts, args->ts); } else if (args->op == LINUX_FUTEX_WAIT_BITSET) { nanouptime(&kts); timespecsub(args->ts, &kts, args->ts); } } retry: f = NULL; error = futex_get(args->uaddr, &wp, &f, args->flags | FUTEX_CREATE_WP); if (error != 0) return (error); error = copyin_nofault(args->uaddr, &val, sizeof(val)); if (error != 0) { futex_put(f, wp); error = copyin(args->uaddr, &val, sizeof(val)); if (error == 0) goto retry; LIN_SDT_PROBE1(futex, linux_futex, copyin_error, error); LINUX_CTR1(sys_futex, "WAIT copyin failed %d", error); return (error); } if (val != args->val) { LIN_SDT_PROBE4(futex, linux_futex, debug_wait_value_neq, args->uaddr, args->val, val, args->val3); LINUX_CTR3(sys_futex, "WAIT uaddr %p val 0x%x != uval 0x%x", args->uaddr, args->val, val); futex_put(f, wp); return (EWOULDBLOCK); } return (futex_wait(f, wp, args->ts, args->val3)); } int linux_sys_futex(struct thread *td, struct linux_sys_futex_args *args) { struct linux_futex_args fargs = { .uaddr = args->uaddr, .op = args->op, .val = args->val, .ts = NULL, .uaddr2 = args->uaddr2, .val3 = args->val3, + .val3_compare = true, }; struct l_timespec lts; int error; switch (args->op & LINUX_FUTEX_CMD_MASK) { case LINUX_FUTEX_WAIT: case LINUX_FUTEX_WAIT_BITSET: if (args->timeout != NULL) { error = copyin(args->timeout, <s, sizeof(lts)); if (error != 0) return (error); error = linux_to_native_timespec(&fargs.kts, <s); if (error != 0) return (error); fargs.ts = &fargs.kts; } break; default: fargs.ts = PTRIN(args->timeout); } return (linux_futex(td, &fargs)); } #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32)) int linux_sys_futex_time64(struct thread *td, struct linux_sys_futex_time64_args *args) { struct linux_futex_args fargs = { .uaddr = args->uaddr, .op = args->op, .val = args->val, .ts = NULL, .uaddr2 = args->uaddr2, .val3 = args->val3, }; struct l_timespec64 lts; int error; switch (args->op & LINUX_FUTEX_CMD_MASK) { case LINUX_FUTEX_WAIT: case LINUX_FUTEX_WAIT_BITSET: if (args->timeout != NULL) { error = copyin(args->timeout, <s, sizeof(lts)); if (error != 0) return (error); error = linux_to_native_timespec64(&fargs.kts, <s); if (error != 0) return (error); fargs.ts = &fargs.kts; } break; default: fargs.ts = PTRIN(args->timeout); } return (linux_futex(td, &fargs)); } #endif int linux_set_robust_list(struct thread *td, struct linux_set_robust_list_args *args) { struct linux_emuldata *em; if (args->len != sizeof(struct linux_robust_list_head)) { LIN_SDT_PROBE0(futex, linux_set_robust_list, size_error); return (EINVAL); } em = em_find(td); em->robust_futexes = args->head; return (0); } int linux_get_robust_list(struct thread *td, struct linux_get_robust_list_args *args) { struct linux_emuldata *em; struct linux_robust_list_head *head; l_size_t len = sizeof(struct linux_robust_list_head); struct thread *td2; int error = 0; if (!args->pid) { em = em_find(td); KASSERT(em != NULL, ("get_robust_list: emuldata notfound.\n")); head = em->robust_futexes; } else { td2 = tdfind(args->pid, -1); if (td2 == NULL) return (ESRCH); if (SV_PROC_ABI(td2->td_proc) != SV_ABI_LINUX) { PROC_UNLOCK(td2->td_proc); return (EPERM); } em = em_find(td2); KASSERT(em != NULL, ("get_robust_list: emuldata notfound.\n")); /* XXX: ptrace? */ if (priv_check(td, PRIV_CRED_SETUID) || priv_check(td, PRIV_CRED_SETEUID) || p_candebug(td, td2->td_proc)) { PROC_UNLOCK(td2->td_proc); return (EPERM); } head = em->robust_futexes; PROC_UNLOCK(td2->td_proc); } error = copyout(&len, args->len, sizeof(l_size_t)); if (error) { LIN_SDT_PROBE1(futex, linux_get_robust_list, copyout_error, error); return (EFAULT); } error = copyout(&head, args->head, sizeof(head)); if (error) { LIN_SDT_PROBE1(futex, linux_get_robust_list, copyout_error, error); } return (error); } static int handle_futex_death(struct linux_emuldata *em, uint32_t *uaddr, unsigned int pi) { uint32_t uval, nval, mval; struct futex *f; int error; retry: error = copyin(uaddr, &uval, 4); if (error) { LIN_SDT_PROBE1(futex, handle_futex_death, copyin_error, error); return (EFAULT); } if ((uval & FUTEX_TID_MASK) == em->em_tid) { mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; nval = casuword32(uaddr, uval, mval); if (nval == -1) return (EFAULT); if (nval != uval) goto retry; if (!pi && (uval & FUTEX_WAITERS)) { error = futex_get(uaddr, NULL, &f, FUTEX_DONTCREATE | FUTEX_SHARED); if (error) return (error); if (f != NULL) { futex_wake(f, 1, FUTEX_BITSET_MATCH_ANY); futex_put(f, NULL); } } } return (0); } static int fetch_robust_entry(struct linux_robust_list **entry, struct linux_robust_list **head, unsigned int *pi) { l_ulong uentry; int error; error = copyin((const void *)head, &uentry, sizeof(l_ulong)); if (error) { LIN_SDT_PROBE1(futex, fetch_robust_entry, copyin_error, error); return (EFAULT); } *entry = (void *)(uentry & ~1UL); *pi = uentry & 1; return (0); } /* This walks the list of robust futexes releasing them. */ void release_futexes(struct thread *td, struct linux_emuldata *em) { struct linux_robust_list_head *head = NULL; struct linux_robust_list *entry, *next_entry, *pending; unsigned int limit = 2048, pi, next_pi, pip; l_long futex_offset; int rc, error; head = em->robust_futexes; if (head == NULL) return; if (fetch_robust_entry(&entry, PTRIN(&head->list.next), &pi)) return; error = copyin(&head->futex_offset, &futex_offset, sizeof(futex_offset)); if (error) { LIN_SDT_PROBE1(futex, release_futexes, copyin_error, error); return; } if (fetch_robust_entry(&pending, PTRIN(&head->pending_list), &pip)) return; while (entry != &head->list) { rc = fetch_robust_entry(&next_entry, PTRIN(&entry->next), &next_pi); if (entry != pending) if (handle_futex_death(em, (uint32_t *)((caddr_t)entry + futex_offset), pi)) { return; } if (rc) return; entry = next_entry; pi = next_pi; if (!--limit) break; sched_relinquish(curthread); } if (pending) handle_futex_death(em, (uint32_t *)((caddr_t)pending + futex_offset), pip); } diff --git a/sys/i386/linux/linux_sysvec.c b/sys/i386/linux/linux_sysvec.c index f3101db96d41..1fcc1b6eb5a0 100644 --- a/sys/i386/linux/linux_sysvec.c +++ b/sys/i386/linux/linux_sysvec.c @@ -1,1180 +1,1181 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * 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. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #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 LINUX_VDSOPAGE_SIZE PAGE_SIZE * 2 #define LINUX_VDSOPAGE (VM_MAXUSER_ADDRESS - LINUX_VDSOPAGE_SIZE) #define LINUX_SHAREDPAGE (LINUX_VDSOPAGE - PAGE_SIZE) /* * PAGE_SIZE - the size * of the native SHAREDPAGE */ #define LINUX_USRSTACK LINUX_SHAREDPAGE #define LINUX_PS_STRINGS (LINUX_USRSTACK - 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]; SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); static int linux_fixup(uintptr_t *stack_base, struct image_params *iparams); static int linux_fixup_elf(uintptr_t *stack_base, struct image_params *iparams); 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 int linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base); static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel); 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); #define LINUX_T_UNKNOWN 255 static int _bsd_to_linux_trapcode[] = { LINUX_T_UNKNOWN, /* 0 */ 6, /* 1 T_PRIVINFLT */ LINUX_T_UNKNOWN, /* 2 */ 3, /* 3 T_BPTFLT */ LINUX_T_UNKNOWN, /* 4 */ LINUX_T_UNKNOWN, /* 5 */ 16, /* 6 T_ARITHTRAP */ 254, /* 7 T_ASTFLT */ LINUX_T_UNKNOWN, /* 8 */ 13, /* 9 T_PROTFLT */ 1, /* 10 T_TRCTRAP */ LINUX_T_UNKNOWN, /* 11 */ 14, /* 12 T_PAGEFLT */ LINUX_T_UNKNOWN, /* 13 */ 17, /* 14 T_ALIGNFLT */ LINUX_T_UNKNOWN, /* 15 */ LINUX_T_UNKNOWN, /* 16 */ LINUX_T_UNKNOWN, /* 17 */ 0, /* 18 T_DIVIDE */ 2, /* 19 T_NMI */ 4, /* 20 T_OFLOW */ 5, /* 21 T_BOUND */ 7, /* 22 T_DNA */ 8, /* 23 T_DOUBLEFLT */ 9, /* 24 T_FPOPFLT */ 10, /* 25 T_TSSFLT */ 11, /* 26 T_SEGNPFLT */ 12, /* 27 T_STKFLT */ 18, /* 28 T_MCHK */ 19, /* 29 T_XMMFLT */ 15 /* 30 T_RESERVED */ }; #define bsd_to_linux_trapcode(code) \ ((code)args->argc + 1); base--; suword(base, (intptr_t)envp); base--; suword(base, (intptr_t)argv); base--; suword(base, imgp->args->argc); *stack_base = (uintptr_t)base; return (0); } static int linux_copyout_auxargs(struct image_params *imgp, uintptr_t base) { struct proc *p; Elf32_Auxargs *args; Elf32_Auxinfo *argarray, *pos; struct ps_strings *arginfo; int error, issetugid; p = imgp->proc; issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0; arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; args = (Elf32_Auxargs *)imgp->auxargs; argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, M_WAITOK | M_ZERO); AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base); AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, __kernel_vsyscall); AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); /* * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0, * as it has appeared in the 2.4.0-rc7 first time. * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK), * glibc falls back to the hard-coded CLK_TCK value when aux entry * is not present. * Also see linux_times() implementation. */ if (linux_kernver(curthread) >= LINUX_KERNVER_2004000) AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); AUXARGS_ENTRY(pos, AT_PHENT, args->phent); AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); AUXARGS_ENTRY(pos, AT_BASE, args->base); AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary); if (imgp->execpathp != 0) AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp); if (args->execfd != -1) AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); AUXARGS_ENTRY(pos, AT_NULL, 0); free(imgp->auxargs, M_TEMP); imgp->auxargs = NULL; KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); error = copyout(argarray, (void *)base, sizeof(*argarray) * LINUX_AT_COUNT); free(argarray, M_TEMP); return (error); } static int linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp) { register_t *base; base = (register_t *)*stack_base; base--; if (suword(base, (register_t)imgp->args->argc) == -1) return (EFAULT); *stack_base = (uintptr_t)base; return (0); } /* * Copied from kern/kern_exec.c */ static int linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) { int argc, envc, error; char **vectp; char *stringp; uintptr_t destp, ustringp; struct ps_strings *arginfo; char canary[LINUX_AT_RANDOM_LEN]; size_t execpath_len; struct proc *p; /* Calculate string base and vector table pointers. */ p = imgp->proc; if (imgp->execpath != NULL && imgp->auxargs != NULL) execpath_len = strlen(imgp->execpath) + 1; else execpath_len = 0; arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; destp = (uintptr_t)arginfo; if (execpath_len != 0) { destp -= execpath_len; destp = rounddown2(destp, sizeof(void *)); imgp->execpathp = (void *)destp; error = copyout(imgp->execpath, imgp->execpathp, execpath_len); if (error != 0) return (error); } /* Prepare the canary for SSP. */ arc4rand(canary, sizeof(canary), 0); destp -= roundup(sizeof(canary), sizeof(void *)); imgp->canary = (void *)destp; error = copyout(canary, imgp->canary, sizeof(canary)); if (error != 0) return (error); /* Allocate room for the argument and environment strings. */ destp -= ARG_MAX - imgp->args->stringspace; destp = rounddown2(destp, sizeof(void *)); ustringp = destp; if (imgp->auxargs) { /* * Allocate room on the stack for the ELF auxargs * array. It has LINUX_AT_COUNT entries. */ destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo); destp = rounddown2(destp, sizeof(void *)); } vectp = (char **)destp; /* * Allocate room for the argv[] and env vectors including the * terminating NULL pointers. */ vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; /* vectp also becomes our initial stack base. */ *stack_base = (uintptr_t)vectp; stringp = imgp->args->begin_argv; argc = imgp->args->argc; envc = imgp->args->envc; /* Copy out strings - arguments and environment. */ error = copyout(stringp, (void *)ustringp, ARG_MAX - imgp->args->stringspace); if (error != 0) return (error); /* Fill in "ps_strings" struct for ps, w, etc. */ if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || suword(&arginfo->ps_nargvstr, argc) != 0) return (EFAULT); /* Fill in argument portion of vector table. */ for (; argc > 0; --argc) { if (suword(vectp++, ustringp) != 0) return (EFAULT); while (*stringp++ != 0) ustringp++; ustringp++; } /* A null vector table pointer separates the argp's from the envp's. */ if (suword(vectp++, 0) != 0) return (EFAULT); if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || suword(&arginfo->ps_nenvstr, envc) != 0) return (EFAULT); /* Fill in environment portion of vector table. */ for (; envc > 0; --envc) { if (suword(vectp++, ustringp) != 0) return (EFAULT); while (*stringp++ != 0) ustringp++; ustringp++; } /* The end of the vector table is a null pointer. */ if (suword(vectp, 0) != 0) return (EFAULT); if (imgp->auxargs) { vectp++; error = imgp->sysent->sv_copyout_auxargs(imgp, (uintptr_t)vectp); if (error != 0) return (error); } return (0); } 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 sig, code; int oonstack; sig = ksi->ksi_signo; 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_esp); /* 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_esp - 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_handler = catcher; frame.sf_sig = sig; frame.sf_siginfo = &fp->sf_si; frame.sf_ucontext = &fp->sf_sc; /* Fill in POSIX parts. */ siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig); /* Build the signal context to be used by sigreturn. */ frame.sf_sc.uc_flags = 0; /* XXX ??? */ frame.sf_sc.uc_link = NULL; /* XXX ??? */ frame.sf_sc.uc_stack.ss_sp = td->td_sigstk.ss_sp; frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; frame.sf_sc.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_sc.uc_sigmask); frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask; frame.sf_sc.uc_mcontext.sc_gs = rgs(); frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs; frame.sf_sc.uc_mcontext.sc_es = regs->tf_es; frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds; frame.sf_sc.uc_mcontext.sc_edi = regs->tf_edi; frame.sf_sc.uc_mcontext.sc_esi = regs->tf_esi; frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_ebp; frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_ebx; frame.sf_sc.uc_mcontext.sc_esp = regs->tf_esp; frame.sf_sc.uc_mcontext.sc_edx = regs->tf_edx; frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_ecx; frame.sf_sc.uc_mcontext.sc_eax = regs->tf_eax; frame.sf_sc.uc_mcontext.sc_eip = regs->tf_eip; frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_eflags; frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_esp; frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss; frame.sf_sc.uc_mcontext.sc_err = regs->tf_err; frame.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; frame.sf_sc.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_esp = (int)fp; regs->tf_eip = __kernel_rt_sigreturn; regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D); regs->tf_cs = _ucodesel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; regs->tf_fs = _udatasel; regs->tf_ss = _udatasel; 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 sig, code; int oonstack; PROC_LOCK_ASSERT(p, MA_OWNED); psp = p->p_sigacts; sig = ksi->ksi_signo; code = ksi->ksi_code; 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_esp); /* 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_esp - 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_handler = catcher; frame.sf_sig = sig; 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 = rgs(); 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_edi; frame.sf_sc.sc_esi = regs->tf_esi; frame.sf_sc.sc_ebp = regs->tf_ebp; frame.sf_sc.sc_ebx = regs->tf_ebx; frame.sf_sc.sc_esp = regs->tf_esp; frame.sf_sc.sc_edx = regs->tf_edx; frame.sf_sc.sc_ecx = regs->tf_ecx; frame.sf_sc.sc_eax = regs->tf_eax; frame.sf_sc.sc_eip = regs->tf_eip; frame.sf_sc.sc_cs = regs->tf_cs; frame.sf_sc.sc_eflags = regs->tf_eflags; frame.sf_sc.sc_esp_at_signal = regs->tf_esp; frame.sf_sc.sc_ss = regs->tf_ss; frame.sf_sc.sc_err = regs->tf_err; frame.sf_sc.sc_cr2 = (register_t)ksi->ksi_addr; frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno); frame.sf_extramask[0] = lmask.__mask; 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_esp = (int)fp; regs->tf_eip = __kernel_sigreturn; regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D); regs->tf_cs = _ucodesel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; regs->tf_fs = _udatasel; regs->tf_ss = _udatasel; 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; l_sigset_t lmask; sigset_t bmask; 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. */ #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) eflags = frame.sf_sc.sc_eflags; if (!EFLAGS_SECURE(eflags, regs->tf_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. */ #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 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_eip; trapsignal(td, &ksi); return (EINVAL); } lmask.__mask = frame.sf_sc.sc_mask; linux_to_bsd_sigset(&lmask, &bmask); kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); /* Restore signal context. */ /* %gs was restored by the trampoline. */ regs->tf_fs = frame.sf_sc.sc_fs; regs->tf_es = frame.sf_sc.sc_es; regs->tf_ds = frame.sf_sc.sc_ds; regs->tf_edi = frame.sf_sc.sc_edi; regs->tf_esi = frame.sf_sc.sc_esi; regs->tf_ebp = frame.sf_sc.sc_ebp; regs->tf_ebx = frame.sf_sc.sc_ebx; regs->tf_edx = frame.sf_sc.sc_edx; regs->tf_ecx = frame.sf_sc.sc_ecx; regs->tf_eax = frame.sf_sc.sc_eax; regs->tf_eip = frame.sf_sc.sc_eip; regs->tf_cs = frame.sf_sc.sc_cs; regs->tf_eflags = eflags; regs->tf_esp = frame.sf_sc.sc_esp_at_signal; regs->tf_ss = frame.sf_sc.sc_ss; 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. */ #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) eflags = context->sc_eflags; if (!EFLAGS_SECURE(eflags, regs->tf_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. */ #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 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_eip; trapsignal(td, &ksi); return (EINVAL); } linux_to_bsd_sigset(&uc.uc_sigmask, &bmask); kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); /* Restore signal context. */ /* %gs was restored by the trampoline. */ regs->tf_fs = context->sc_fs; regs->tf_es = context->sc_es; regs->tf_ds = context->sc_ds; regs->tf_edi = context->sc_edi; regs->tf_esi = context->sc_esi; regs->tf_ebp = context->sc_ebp; regs->tf_ebx = context->sc_ebx; regs->tf_edx = context->sc_edx; regs->tf_ecx = context->sc_ecx; regs->tf_eax = context->sc_eax; regs->tf_eip = context->sc_eip; regs->tf_cs = context->sc_cs; regs->tf_eflags = eflags; regs->tf_esp = context->sc_esp_at_signal; regs->tf_ss = context->sc_ss; /* Call sigaltstack & ignore results. */ lss = &uc.uc_stack; ss.ss_sp = 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 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->code = frame->tf_eax; sa->original_code = sa->code; sa->args[0] = frame->tf_ebx; sa->args[1] = frame->tf_ecx; sa->args[2] = frame->tf_edx; sa->args[3] = frame->tf_esi; sa->args[4] = frame->tf_edi; sa->args[5] = frame->tf_ebp; /* Unconfirmed */ if (sa->code >= p->p_sysent->sv_size) /* nosys */ sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; else sa->callp = &p->p_sysent->sv_table[sa->code]; td->td_retval[0] = 0; td->td_retval[1] = frame->tf_edx; return (0); } static void linux_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_eax = bsd_to_linux_errno(error); } } /* * exec_setregs may initialize some registers differently than Linux * does, thus potentially confusing Linux binaries. If necessary, we * override the exec_setregs default(s) here. */ static void linux_exec_setregs(struct thread *td, struct image_params *imgp, uintptr_t stack) { struct pcb *pcb = td->td_pcb; exec_setregs(td, imgp, stack); /* Linux sets %gs to 0, we default to _udatasel. */ pcb->pcb_gs = 0; load_gs(0); pcb->pcb_initial_npxcw = __LINUX_NPXCW__; } struct sysentvec linux_sysvec = { .sv_size = LINUX_SYS_MAXSYSCALL, .sv_table = linux_sysent, .sv_transtrap = linux_translate_traps, .sv_fixup = linux_fixup, .sv_sendsig = linux_sendsig, .sv_sigcode = &_binary_linux_vdso_so_o_start, .sv_szsigcode = &linux_szsigcode, .sv_name = "Linux a.out", .sv_coredump = NULL, .sv_imgact_try = linux_exec_imgact_try, .sv_minsigstksz = LINUX_MINSIGSTKSZ, .sv_minuser = VM_MIN_ADDRESS, .sv_maxuser = VM_MAXUSER_ADDRESS, .sv_usrstack = LINUX_USRSTACK, .sv_psstrings = PS_STRINGS, .sv_stackprot = VM_PROT_ALL, .sv_copyout_strings = exec_copyout_strings, .sv_setregs = linux_exec_setregs, .sv_fixlimit = NULL, .sv_maxssiz = NULL, .sv_flags = SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32 | SV_SIG_DISCIGN | SV_SIG_WAITNDQ, .sv_set_syscall_retval = linux_set_syscall_retval, .sv_fetch_syscall_args = linux_fetch_syscall_args, .sv_syscallnames = NULL, .sv_schedtail = linux_schedtail, .sv_thread_detach = linux_thread_detach, .sv_trap = NULL, .sv_onexec = linux_on_exec_vmspace, .sv_onexit = linux_on_exit, .sv_ontdexit = linux_thread_dtor, .sv_setid_allowed = &linux_setid_allowed_query, }; INIT_SYSENTVEC(aout_sysvec, &linux_sysvec); struct sysentvec elf_linux_sysvec = { .sv_size = LINUX_SYS_MAXSYSCALL, .sv_table = linux_sysent, .sv_transtrap = linux_translate_traps, .sv_fixup = linux_fixup_elf, .sv_sendsig = linux_sendsig, .sv_sigcode = &_binary_linux_vdso_so_o_start, .sv_szsigcode = &linux_szsigcode, .sv_name = "Linux ELF32", .sv_coredump = elf32_coredump, .sv_elf_core_osabi = ELFOSABI_FREEBSD, .sv_elf_core_abi_vendor = FREEBSD_ABI_VENDOR, .sv_elf_core_prepare_notes = elf32_prepare_notes, .sv_imgact_try = linux_exec_imgact_try, .sv_minsigstksz = LINUX_MINSIGSTKSZ, .sv_minuser = VM_MIN_ADDRESS, .sv_maxuser = VM_MAXUSER_ADDRESS, .sv_usrstack = LINUX_USRSTACK, .sv_psstrings = LINUX_PS_STRINGS, .sv_stackprot = VM_PROT_ALL, .sv_copyout_auxargs = linux_copyout_auxargs, .sv_copyout_strings = linux_copyout_strings, .sv_setregs = linux_exec_setregs, .sv_fixlimit = NULL, .sv_maxssiz = NULL, .sv_flags = SV_ABI_LINUX | SV_IA32 | SV_ILP32 | 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 = NULL, .sv_shared_page_base = LINUX_SHAREDPAGE, .sv_shared_page_len = PAGE_SIZE, .sv_schedtail = linux_schedtail, .sv_thread_detach = linux_thread_detach, .sv_trap = NULL, .sv_onexec = linux_on_exec_vmspace, .sv_onexit = linux_on_exit, .sv_ontdexit = linux_thread_dtor, .sv_setid_allowed = &linux_setid_allowed_query, }; static int linux_on_exec_vmspace(struct proc *p, struct image_params *imgp) { int error = 0; if (SV_PROC_FLAG(p, SV_SHP) != 0) error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base, LINUX_VDSOPAGE_SIZE, imgp); if (error == 0) 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_timekeep_base; 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); } 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; __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_Shdr *shdr; const Elf_Rel *rel; const Elf_Ehdr *ehdr; Elf_Addr *where; Elf_Size rtype, symidx; Elf_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 = (Elf_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 = (Elf_Addr)PTROUT(offset + addend); if (*where != addr) *where = addr; break; case R_386_IRELATIVE: printf("Linux i386 vDSO: unexpected ifunc relocation, " "symbol index %d\n", symidx); break; default: printf("Linux i386 vDSO: unexpected relocation type %d, " "symbol index %d\n", rtype, symidx); } } } static char GNU_ABI_VENDOR[] = "GNU"; static int GNULINUX_ABI_DESC = 0; static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel) { const Elf32_Word *desc; uintptr_t p; p = (uintptr_t)(note + 1); p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); desc = (const Elf32_Word *)p; if (desc[0] != GNULINUX_ABI_DESC) return (false); /* * For Linux we encode osrel using the Linux convention of * (version << 16) | (major << 8) | (minor) * See macro in linux_mib.h */ *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); return (true); } static Elf_Brandnote linux_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", .emul_path = linux_emul_path, .interp_path = "/lib/ld-linux.so.1", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; static Elf32_Brandinfo linux_glibc2brand = { .brand = ELFOSABI_LINUX, .machine = EM_386, .compat_3_brand = "Linux", .emul_path = linux_emul_path, .interp_path = "/lib/ld-linux.so.2", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; static Elf32_Brandinfo linux_muslbrand = { .brand = ELFOSABI_LINUX, .machine = EM_386, .compat_3_brand = "Linux", .emul_path = linux_emul_path, .interp_path = "/lib/ld-musl-i386.so.1", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux_brandnote, - .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE + .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE | + LINUX_BI_FUTEX_REQUEUE }; 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) linux_ioctl_register_handler(*lihp); LIST_INIT(&futex_list); mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF); linux_dev_shm_create(); linux_osd_jail_register(); stclohz = (stathz ? stathz : hz); if (bootverbose) printf("Linux ELF exec handler installed\n"); } else printf("cannot insert Linux 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) linux_ioctl_unregister_handler(*lihp); mtx_destroy(&futex_mtx); linux_dev_shm_destroy(); linux_osd_jail_deregister(); if (bootverbose) printf("Linux ELF exec handler removed\n"); } else printf("Could not deinstall ELF interpreter entry\n"); break; default: return (EOPNOTSUPP); } return (error); } static moduledata_t linux_elf_mod = { "linuxelf", linux_elf_modevent, 0 }; DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); FEATURE(linux, "Linux 32bit support"); /* * linux_vdso_install() and linux_exec_sysvec_init() must be called * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY). */