Index: stable/7/sys/amd64/linux32/linux32_sysvec.c =================================================================== --- stable/7/sys/amd64/linux32/linux32_sysvec.c (revision 193151) +++ stable/7/sys/amd64/linux32/linux32_sysvec.c (revision 193152) @@ -1,1178 +1,1178 @@ /*- * Copyright (c) 2004 Tim J. Robbins * Copyright (c) 2003 Peter Wemm * Copyright (c) 2002 Doug Rabson * Copyright (c) 1998-1999 Andrew Gallatin * Copyright (c) 1994-1996 Søren Schmidt * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer * in this position and unchanged. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #ifndef COMPAT_IA32 #error "Unable to compile Linux-emulator due to missing COMPAT_IA32 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 MODULE_VERSION(linux, 1); MALLOC_DEFINE(M_LINUX, "linux", "Linux mode structures"); #define AUXARGS_ENTRY_32(pos, id, val) \ do { \ suword32(pos++, id); \ suword32(pos++, val); \ } while (0) #if BYTE_ORDER == LITTLE_ENDIAN #define SHELLMAGIC 0x2123 /* #! */ #else #define SHELLMAGIC 0x2321 #endif /* * Allow the sendsig functions to use the ldebug() facility * even though they are not syscalls themselves. Map them * to syscall 0. This is slightly less bogus than using * ldebug(sigreturn). */ #define LINUX_SYS_linux_rt_sendsig 0 #define LINUX_SYS_linux_sendsig 0 const char *linux_platform = "i686"; static int linux_szplatform; extern char linux_sigcode[]; extern int linux_szsigcode; extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); SET_DECLARE(linux_device_handler_set, struct linux_device_handler); static int elf_linux_fixup(register_t **stack_base, struct image_params *iparams); static register_t *linux_copyout_strings(struct image_params *imgp); static void linux_prepsyscall(struct trapframe *tf, int *args, u_int *code, caddr_t *params); static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask); static void exec_linux_setregs(struct thread *td, u_long entry, u_long stack, u_long ps_strings); static void linux32_fixlimit(struct rlimit *rl, int which); extern LIST_HEAD(futex_list, futex) futex_list; -extern struct sx futex_sx; +extern struct mtx futex_mtx; static eventhandler_tag linux_exit_tag; static eventhandler_tag linux_schedtail_tag; static eventhandler_tag linux_exec_tag; /* * Linux syscalls return negative errno's, we do positive and map them * Reference: * FreeBSD: src/sys/sys/errno.h * Linux: linux-2.6.17.8/include/asm-generic/errno-base.h * linux-2.6.17.8/include/asm-generic/errno.h */ static int bsd_to_linux_errno[ELAST + 1] = { -0, -1, -2, -3, -4, -5, -6, -7, -8, -9, -10, -35, -12, -13, -14, -15, -16, -17, -18, -19, -20, -21, -22, -23, -24, -25, -26, -27, -28, -29, -30, -31, -32, -33, -34, -11,-115,-114, -88, -89, -90, -91, -92, -93, -94, -95, -96, -97, -98, -99, -100,-101,-102,-103,-104,-105,-106,-107,-108,-109, -110,-111, -40, -36,-112,-113, -39, -11, -87,-122, -116, -66, -6, -6, -6, -6, -6, -37, -38, -9, -6, -6, -43, -42, -75,-125, -84, -95, -16, -74, -72, -67, -71 }; int bsd_to_linux_signal[LINUX_SIGTBLSZ] = { LINUX_SIGHUP, LINUX_SIGINT, LINUX_SIGQUIT, LINUX_SIGILL, LINUX_SIGTRAP, LINUX_SIGABRT, 0, LINUX_SIGFPE, LINUX_SIGKILL, LINUX_SIGBUS, LINUX_SIGSEGV, LINUX_SIGSYS, LINUX_SIGPIPE, LINUX_SIGALRM, LINUX_SIGTERM, LINUX_SIGURG, LINUX_SIGSTOP, LINUX_SIGTSTP, LINUX_SIGCONT, LINUX_SIGCHLD, LINUX_SIGTTIN, LINUX_SIGTTOU, LINUX_SIGIO, LINUX_SIGXCPU, LINUX_SIGXFSZ, LINUX_SIGVTALRM, LINUX_SIGPROF, LINUX_SIGWINCH, 0, LINUX_SIGUSR1, LINUX_SIGUSR2 }; int linux_to_bsd_signal[LINUX_SIGTBLSZ] = { SIGHUP, SIGINT, SIGQUIT, SIGILL, SIGTRAP, SIGABRT, SIGBUS, SIGFPE, SIGKILL, SIGUSR1, SIGSEGV, SIGUSR2, SIGPIPE, SIGALRM, SIGTERM, SIGBUS, SIGCHLD, SIGCONT, SIGSTOP, SIGTSTP, SIGTTIN, SIGTTOU, SIGURG, SIGXCPU, SIGXFSZ, SIGVTALRM, SIGPROF, SIGWINCH, SIGIO, SIGURG, SIGSYS }; #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 == imgp->proc && (curthread->td_proc->p_flag & P_SA) == 0, ("unsafe elf_linux_fixup(), should be curproc")); base = (Elf32_Addr *)*stack_base; args = (Elf32_Auxargs *)imgp->auxargs; pos = base + (imgp->args->argc + imgp->args->envc + 2); AUXARGS_ENTRY_32(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_32(pos, LINUX_AT_CLKTCK, stclohz); AUXARGS_ENTRY_32(pos, AT_PHDR, args->phdr); AUXARGS_ENTRY_32(pos, AT_PHENT, args->phent); AUXARGS_ENTRY_32(pos, AT_PHNUM, args->phnum); AUXARGS_ENTRY_32(pos, AT_PAGESZ, args->pagesz); AUXARGS_ENTRY_32(pos, AT_FLAGS, args->flags); AUXARGS_ENTRY_32(pos, AT_ENTRY, args->entry); AUXARGS_ENTRY_32(pos, AT_BASE, args->base); AUXARGS_ENTRY_32(pos, LINUX_AT_SECURE, 0); AUXARGS_ENTRY_32(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); AUXARGS_ENTRY_32(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); AUXARGS_ENTRY_32(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); AUXARGS_ENTRY_32(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); AUXARGS_ENTRY_32(pos, LINUX_AT_PLATFORM, PTROUT(uplatform)); if (args->execfd != -1) AUXARGS_ENTRY_32(pos, AT_EXECFD, args->execfd); AUXARGS_ENTRY_32(pos, AT_NULL, 0); free(imgp->auxargs, M_TEMP); imgp->auxargs = NULL; base--; suword32(base, (uint32_t)imgp->args->argc); *stack_base = (register_t *)base; return 0; } extern int _ucodesel, _ucode32sel, _udatasel; extern unsigned long linux_sznonrtsigcode; 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); #ifdef DEBUG if (ldebug(rt_sendsig)) printf(ARGS(rt_sendsig, "%p, %d, %p, %u"), catcher, sig, (void*)mask, code); #endif /* * Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { fp = (struct l_rt_sigframe *)(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. */ if (p->p_sysent->sv_sigtbl) if (sig <= p->p_sysent->sv_sigsize) sig = p->p_sysent->sv_sigtbl[_SIG_IDX(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 */ frame.sf_si.lsi_signo = sig; frame.sf_si.lsi_code = code; frame.sf_si.lsi_addr = PTROUT(ksi->ksi_addr); /* * Build the signal context to be used by sigreturn. */ 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.__bits[0]; frame.sf_sc.uc_mcontext.sc_gs = rgs(); frame.sf_sc.uc_mcontext.sc_fs = rfs(); __asm __volatile("movl %%es,%0" : "=rm" (frame.sf_sc.uc_mcontext.sc_es)); __asm __volatile("movl %%ds,%0" : "=rm" (frame.sf_sc.uc_mcontext.sc_ds)); 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_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_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); #ifdef DEBUG if (ldebug(rt_sendsig)) printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"), frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp, td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask); #endif if (copyout(&frame, fp, sizeof(frame)) != 0) { /* * Process has trashed its stack; give it an illegal * instruction to halt it in its tracks. */ #ifdef DEBUG if (ldebug(rt_sendsig)) printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"), fp, oonstack); #endif PROC_LOCK(p); sigexit(td, SIGILL); } /* * Build context to run handler in. */ regs->tf_rsp = PTROUT(fp); regs->tf_rip = LINUX32_PS_STRINGS - *(p->p_sysent->sv_szsigcode) + linux_sznonrtsigcode; regs->tf_rflags &= ~(PSL_T | PSL_D); regs->tf_cs = _ucode32sel; regs->tf_ss = _udatasel; load_ds(_udatasel); td->td_pcb->pcb_ds = _udatasel; load_es(_udatasel); td->td_pcb->pcb_es = _udatasel; /* leave user %fs and %gs untouched */ 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, i; 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); #ifdef DEBUG if (ldebug(sendsig)) printf(ARGS(sendsig, "%p, %d, %p, %u"), catcher, sig, (void*)mask, code); #endif /* * Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { fp = (struct l_sigframe *)(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. */ if (p->p_sysent->sv_sigtbl) if (sig <= p->p_sysent->sv_sigsize) sig = p->p_sysent->sv_sigtbl[_SIG_IDX(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.__bits[0]; frame.sf_sc.sc_gs = rgs(); frame.sf_sc.sc_fs = rfs(); __asm __volatile("movl %%es,%0" : "=rm" (frame.sf_sc.sc_es)); __asm __volatile("movl %%ds,%0" : "=rm" (frame.sf_sc.sc_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_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); for (i = 0; i < (LINUX_NSIG_WORDS-1); i++) frame.sf_extramask[i] = lmask.__bits[i+1]; if (copyout(&frame, fp, sizeof(frame)) != 0) { /* * Process has trashed its stack; give it an illegal * instruction to halt it in its tracks. */ PROC_LOCK(p); sigexit(td, SIGILL); } /* * Build context to run handler in. */ regs->tf_rsp = PTROUT(fp); regs->tf_rip = LINUX32_PS_STRINGS - *(p->p_sysent->sv_szsigcode); regs->tf_rflags &= ~(PSL_T | PSL_D); regs->tf_cs = _ucode32sel; regs->tf_ss = _udatasel; load_ds(_udatasel); td->td_pcb->pcb_ds = _udatasel; load_es(_udatasel); td->td_pcb->pcb_es = _udatasel; /* leave user %fs and %gs untouched */ PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } /* * System call to cleanup state after a signal * has been taken. Reset signal mask and * stack state from context left by sendsig (above). * Return to previous pc and psl as specified by * context left by sendsig. Check carefully to * make sure that the user has not modified the * psl to gain improper privileges or to cause * a machine fault. */ int linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args) { struct proc *p = td->td_proc; struct l_sigframe frame; struct trapframe *regs; l_sigset_t lmask; int eflags, i; ksiginfo_t ksi; regs = td->td_frame; #ifdef DEBUG if (ldebug(sigreturn)) printf(ARGS(sigreturn, "%p"), (void *)args->sfp); #endif /* * 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; /* * XXX do allow users to change the privileged flag PSL_RF. The * cpu sets PSL_RF in tf_eflags for faults. Debuggers should * sometimes set it there too. tf_eflags is kept in the signal * context during signal handling and there is no other place * to remember it, so the PSL_RF bit may be corrupted by the * signal handler without us knowing. Corruption of the PSL_RF * bit at worst causes one more or one less debugger trap, so * allowing it is fairly harmless. */ if (!EFLAGS_SECURE(eflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) 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.__bits[0] = frame.sf_sc.sc_mask; for (i = 0; i < (LINUX_NSIG_WORDS-1); i++) lmask.__bits[i+1] = frame.sf_extramask[i]; PROC_LOCK(p); linux_to_bsd_sigset(&lmask, &td->td_sigmask); SIG_CANTMASK(td->td_sigmask); signotify(td); PROC_UNLOCK(p); /* * Restore signal context. */ /* Selectors were restored by the trampoline. */ 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_rflags = eflags; regs->tf_rsp = 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 proc *p = td->td_proc; struct l_ucontext uc; struct l_sigcontext *context; l_stack_t *lss; stack_t ss; struct trapframe *regs; int eflags; ksiginfo_t ksi; regs = td->td_frame; #ifdef DEBUG if (ldebug(rt_sigreturn)) printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp); #endif /* * 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; /* * XXX do allow users to change the privileged flag PSL_RF. The * cpu sets PSL_RF in tf_eflags for faults. Debuggers should * sometimes set it there too. tf_eflags is kept in the signal * context during signal handling and there is no other place * to remember it, so the PSL_RF bit may be corrupted by the * signal handler without us knowing. Corruption of the PSL_RF * bit at worst causes one more or one less debugger trap, so * allowing it is fairly harmless. */ if (!EFLAGS_SECURE(eflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) 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); } PROC_LOCK(p); linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask); SIG_CANTMASK(td->td_sigmask); signotify(td); PROC_UNLOCK(p); /* * Restore signal context */ /* Selectors were restored by the trampoline. */ 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; /* * 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); #ifdef DEBUG if (ldebug(rt_sigreturn)) printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"), ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask); #endif (void)kern_sigaltstack(td, &ss, NULL); return (EJUSTRETURN); } /* * MPSAFE */ static void linux_prepsyscall(struct trapframe *tf, int *args, u_int *code, caddr_t *params) { args[0] = tf->tf_rbx; args[1] = tf->tf_rcx; args[2] = tf->tf_rdx; args[3] = tf->tf_rsi; args[4] = tf->tf_rdi; args[5] = tf->tf_rbp; /* Unconfirmed */ *params = NULL; /* no copyin */ } /* * If a linux binary is exec'ing something, try this image activator * first. We override standard shell script execution in order to * be able to modify the interpreter path. We only do this if a linux * binary is doing the exec, so we do not create an EXEC module for it. */ static int exec_linux_imgact_try(struct image_params *iparams); static int exec_linux_imgact_try(struct image_params *imgp) { const char *head = (const char *)imgp->image_header; char *rpath; int error = -1, len; /* * The interpreter for shell scripts run from a linux binary needs * to be located in /compat/linux if possible in order to recursively * maintain linux path emulation. */ if (((const short *)head)[0] == SHELLMAGIC) { /* * Run our normal shell image activator. If it succeeds * attempt to use the alternate path for the interpreter. If * an alternate path is found, use our stringspace to store it. */ if ((error = exec_shell_imgact(imgp)) == 0) { linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc), imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0); if (rpath != NULL) { len = strlen(rpath) + 1; if (len <= MAXSHELLCMDLEN) { memcpy(imgp->interpreter_name, rpath, len); } free(rpath, M_TEMP); } } } return(error); } /* * Clear registers on exec * XXX copied from ia32_signal.c. */ static void exec_linux_setregs(td, entry, stack, ps_strings) struct thread *td; u_long entry; u_long stack; u_long ps_strings; { struct trapframe *regs = td->td_frame; struct pcb *pcb = td->td_pcb; 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(); load_ds(_udatasel); load_es(_udatasel); load_fs(_udatasel); load_gs(_udatasel); pcb->pcb_ds = _udatasel; pcb->pcb_es = _udatasel; pcb->pcb_fs = _udatasel; pcb->pcb_gs = _udatasel; pcb->pcb_initial_fpucw = __LINUX_NPXCW__; bzero((char *)regs, sizeof(struct trapframe)); regs->tf_rip = entry; regs->tf_rsp = stack; regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T); regs->tf_ss = _udatasel; regs->tf_cs = _ucode32sel; regs->tf_rbx = ps_strings; load_cr0(rcr0() | CR0_MP | CR0_TS); fpstate_drop(td); /* Return via doreti so that we can change to a different %cs */ pcb->pcb_flags |= PCB_FULLCTX | PCB_32BIT; pcb->pcb_flags &= ~PCB_GS32BIT; td->td_retval[1] = 0; } /* * XXX copied from ia32_sysvec.c. */ static register_t * linux_copyout_strings(struct image_params *imgp) { int argc, envc; u_int32_t *vectp; char *stringp, *destp; u_int32_t *stack_base; struct linux32_ps_strings *arginfo; /* * Calculate string base and vector table pointers. * Also deal with signal trampoline code for this exec type. */ arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS; destp = (caddr_t)arginfo - linux_szsigcode - SPARE_USRSPACE - linux_szplatform - roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *)); /* * install sigcode */ copyout(imgp->proc->p_sysent->sv_sigcode, ((caddr_t)arginfo - linux_szsigcode), linux_szsigcode); /* * Install LINUX_PLATFORM */ copyout(linux_platform, ((caddr_t)arginfo - linux_szsigcode - linux_szplatform), linux_szplatform); /* * If we have a valid auxargs ptr, prepare some room * on the stack. */ if (imgp->auxargs) { /* * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for * lower compatibility. */ imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : (LINUX_AT_COUNT * 2); /* * The '+ 2' is for the null pointers at the end of each of * the arg and env vector sets,and imgp->auxarg_size is room * for argument of Runtime loader. */ vectp = (u_int32_t *)(destp - (imgp->args->argc + imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(u_int32_t)); } else /* * The '+ 2' is for the null pointers at the end of each of * the arg and env vector sets */ vectp = (u_int32_t *)(destp - (imgp->args->argc + imgp->args->envc + 2) * sizeof(u_int32_t)); /* * vectp also becomes our initial stack base */ stack_base = vectp; stringp = imgp->args->begin_argv; argc = imgp->args->argc; envc = imgp->args->envc; /* * Copy out strings - arguments and environment. */ copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); /* * Fill in "ps_strings" struct for ps, w, etc. */ suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp); suword32(&arginfo->ps_nargvstr, argc); /* * Fill in argument portion of vector table. */ for (; argc > 0; --argc) { suword32(vectp++, (uint32_t)(intptr_t)destp); while (*stringp++ != 0) destp++; destp++; } /* a null vector table pointer separates the argp's from the envp's */ suword32(vectp++, 0); suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp); suword32(&arginfo->ps_nenvstr, envc); /* * Fill in environment portion of vector table. */ for (; envc > 0; --envc) { suword32(vectp++, (uint32_t)(intptr_t)destp); while (*stringp++ != 0) destp++; destp++; } /* end of vector table is a null pointer */ suword32(vectp, 0); return ((register_t *)stack_base); } SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW, 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 = LINUX_SYS_MAXSYSCALL, .sv_table = linux_sysent, .sv_mask = 0, .sv_sigsize = LINUX_SIGTBLSZ, .sv_sigtbl = bsd_to_linux_signal, .sv_errsize = ELAST + 1, .sv_errtbl = bsd_to_linux_errno, .sv_transtrap = translate_traps, .sv_fixup = elf_linux_fixup, .sv_sendsig = linux_sendsig, .sv_sigcode = linux_sigcode, .sv_szsigcode = &linux_szsigcode, .sv_prepsyscall = linux_prepsyscall, .sv_name = "Linux ELF32", .sv_coredump = elf32_coredump, .sv_imgact_try = exec_linux_imgact_try, .sv_minsigstksz = LINUX_MINSIGSTKSZ, .sv_pagesize = PAGE_SIZE, .sv_minuser = VM_MIN_ADDRESS, .sv_maxuser = LINUX32_USRSTACK, .sv_usrstack = LINUX32_USRSTACK, .sv_psstrings = LINUX32_PS_STRINGS, .sv_stackprot = VM_PROT_ALL, .sv_copyout_strings = linux_copyout_strings, .sv_setregs = exec_linux_setregs, .sv_fixlimit = linux32_fixlimit, .sv_maxssiz = &linux32_maxssiz, }; static char GNULINUX_ABI_VENDOR[] = "GNU"; static Elf_Brandnote linux32_brandnote = { .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR), .hdr.n_descsz = 16, .hdr.n_type = 1, .vendor = GNULINUX_ABI_VENDOR, .flags = 0 }; static Elf32_Brandinfo linux_brand = { .brand = ELFOSABI_LINUX, .machine = EM_386, .compat_3_brand = "Linux", .emul_path = "/compat/linux", .interp_path = "/lib/ld-linux.so.1", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux32_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; static Elf32_Brandinfo linux_glibc2brand = { .brand = ELFOSABI_LINUX, .machine = EM_386, .compat_3_brand = "Linux", .emul_path = "/compat/linux", .interp_path = "/lib/ld-linux.so.2", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux32_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; Elf32_Brandinfo *linux_brandlist[] = { &linux_brand, &linux_glibc2brand, NULL }; static int linux_elf_modevent(module_t mod, int type, void *data) { Elf32_Brandinfo **brandinfo; int error; struct linux_ioctl_handler **lihp; struct linux_device_handler **ldhp; 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); SET_FOREACH(ldhp, linux_device_handler_set) linux_device_register_handler(*ldhp); mtx_init(&emul_lock, "emuldata lock", NULL, MTX_DEF); sx_init(&emul_shared_lock, "emuldata->shared lock"); LIST_INIT(&futex_list); - sx_init(&futex_sx, "futex protection lock"); + mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF); linux_exit_tag = EVENTHANDLER_REGISTER(process_exit, linux_proc_exit, NULL, 1000); linux_schedtail_tag = EVENTHANDLER_REGISTER(schedtail, linux_schedtail, NULL, 1000); linux_exec_tag = EVENTHANDLER_REGISTER(process_exec, linux_proc_exec, NULL, 1000); linux_szplatform = roundup(strlen(linux_platform) + 1, sizeof(char *)); 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); SET_FOREACH(ldhp, linux_device_handler_set) linux_device_unregister_handler(*ldhp); mtx_destroy(&emul_lock); sx_destroy(&emul_shared_lock); - sx_destroy(&futex_sx); + mtx_destroy(&futex_mtx); EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag); EVENTHANDLER_DEREGISTER(schedtail, linux_schedtail_tag); EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag); 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(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); Index: stable/7/sys/compat/linux/linux_futex.c =================================================================== --- stable/7/sys/compat/linux/linux_futex.c (revision 193151) +++ stable/7/sys/compat/linux/linux_futex.c (revision 193152) @@ -1,693 +1,784 @@ /* $NetBSD: linux_futex.c,v 1.7 2006/07/24 19:01:49 manu Exp $ */ /*- * Copyright (c) 2005 Emmanuel Dreyfus, all rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. 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 #ifdef COMPAT_LINUX32 #include #include #else #include #include #endif #include #include +MALLOC_DEFINE(M_FUTEX, "futex", "Linux futexes"); +MALLOC_DEFINE(M_FUTEX_WP, "futex wp", "Linux futexes wp"); + struct futex; struct waiting_proc { - struct thread *wp_t; - struct futex *wp_new_futex; + uint32_t wp_flags; + struct futex *wp_futex; TAILQ_ENTRY(waiting_proc) wp_list; }; + struct futex { - void *f_uaddr; - int f_refcount; + struct sx f_lck; + uint32_t *f_uaddr; + uint32_t f_refcount; LIST_ENTRY(futex) f_list; TAILQ_HEAD(lf_waiting_proc, waiting_proc) f_waiting_proc; }; LIST_HEAD(futex_list, futex) futex_list; -struct sx futex_sx; /* this protects the LIST of futexes */ -#define FUTEX_LOCK sx_xlock(&futex_sx) -#define FUTEX_UNLOCK sx_xunlock(&futex_sx) +#define FUTEX_LOCK(f) sx_xlock(&(f)->f_lck) +#define FUTEX_UNLOCK(f) sx_xunlock(&(f)->f_lck) +#define FUTEX_INIT(f) sx_init_flags(&(f)->f_lck, "ftlk", 0) +#define FUTEX_DESTROY(f) sx_destroy(&(f)->f_lck) +#define FUTEX_ASSERT_LOCKED(f) sx_assert(&(f)->f_lck, SA_XLOCKED) -#define FUTEX_LOCKED 1 -#define FUTEX_UNLOCKED 0 +struct mtx futex_mtx; /* protects the futex list */ +#define FUTEXES_LOCK mtx_lock(&futex_mtx) +#define FUTEXES_UNLOCK mtx_unlock(&futex_mtx) -#define FUTEX_SYSTEM_LOCK mtx_lock(&Giant) -#define FUTEX_SYSTEM_UNLOCK mtx_unlock(&Giant) +/* 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 */ -static struct futex *futex_get(void *, int); -static void futex_put(struct futex *); -static int futex_sleep(struct futex *, struct thread *, unsigned long); -static int futex_wake(struct futex *, int, struct futex *, int); -static int futex_atomic_op(struct thread *td, int encoded_op, caddr_t uaddr); +/* 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. + */ /* support.s */ -int futex_xchgl(int oparg, caddr_t uaddr, int *oldval); -int futex_addl(int oparg, caddr_t uaddr, int *oldval); -int futex_orl(int oparg, caddr_t uaddr, int *oldval); -int futex_andl(int oparg, caddr_t uaddr, int *oldval); -int futex_xorl(int oparg, caddr_t uaddr, int *oldval); +int futex_xchgl(int oparg, uint32_t *uaddr, int *oldval); +int futex_addl(int oparg, uint32_t *uaddr, int *oldval); +int futex_orl(int oparg, uint32_t *uaddr, int *oldval); +int futex_andl(int oparg, uint32_t *uaddr, int *oldval); +int futex_xorl(int oparg, uint32_t *uaddr, int *oldval); -int -linux_sys_futex(struct thread *td, struct linux_sys_futex_args *args) +static void +futex_put(struct futex *f, struct waiting_proc *wp) { - int val; - int ret; - struct l_timespec timeout = {0, 0}; - int error = 0; - struct futex *f; - struct futex *newf; - int timeout_hz; - struct timeval tv = {0, 0}; - struct futex *f2; - int op_ret; - struct linux_emuldata *em; -#ifdef DEBUG - if (ldebug(sys_futex)) - printf(ARGS(futex, "%p, %i, %i, *, %p, %i"), args->uaddr, args->op, - args->val, args->uaddr2, args->val3); -#endif + FUTEX_ASSERT_LOCKED(f); + 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); + } - /* - * Our implementation provides only privates futexes. Most of the apps - * should use private futexes but don't claim so. Therefore we treat - * all futexes as private by clearing the FUTEX_PRIVATE_FLAG. It works - * in most cases (ie. when futexes are not shared on file descriptor - * or between different processes.). - */ - args->op = (args->op & ~LINUX_FUTEX_PRIVATE_FLAG); + FUTEXES_LOCK; + if (--f->f_refcount == 0) { + LIST_REMOVE(f, f_list); + FUTEXES_UNLOCK; + FUTEX_UNLOCK(f); - switch (args->op) { - case LINUX_FUTEX_WAIT: - FUTEX_SYSTEM_LOCK; + FUTEX_DESTROY(f); + free(f, M_FUTEX); + return; + } - if ((error = copyin(args->uaddr, - &val, sizeof(val))) != 0) { - FUTEX_SYSTEM_UNLOCK; - return error; - } + FUTEXES_UNLOCK; + FUTEX_UNLOCK(f); +} - if (val != args->val) { - FUTEX_SYSTEM_UNLOCK; - return EWOULDBLOCK; - } +static int +futex_get0(uint32_t *uaddr, struct futex **newf, uint32_t flags) +{ + struct futex *f, *tmpf; - if (args->timeout != NULL) { - if ((error = copyin(args->timeout, - &timeout, sizeof(timeout))) != 0) { - FUTEX_SYSTEM_UNLOCK; - return error; - } - } + *newf = tmpf = NULL; -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX_WAIT %d: val = %d, uaddr = %p, " - "*uaddr = %d, timeout = %d.%09lu\n", - td->td_proc->p_pid, args->val, - args->uaddr, val, timeout.tv_sec, - (unsigned long)timeout.tv_nsec); -#endif - tv.tv_usec = timeout.tv_sec * 1000000 + timeout.tv_nsec / 1000; - timeout_hz = tvtohz(&tv); - - if (timeout.tv_sec == 0 && timeout.tv_nsec == 0) - timeout_hz = 0; - /* - * If the user process requests a non null timeout, - * make sure we do not turn it into an infinite - * timeout because timeout_hz gets null. - * - * We use a minimal timeout of 1/hz. Maybe it would - * make sense to just return ETIMEDOUT without sleeping. - */ - if (((timeout.tv_sec != 0) || (timeout.tv_nsec != 0)) && - (timeout_hz == 0)) - timeout_hz = 1; - - - f = futex_get(args->uaddr, FUTEX_UNLOCKED); - ret = futex_sleep(f, td, timeout_hz); - futex_put(f); - -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX_WAIT %d: uaddr = %p, " - "ret = %d\n", td->td_proc->p_pid, args->uaddr, ret); -#endif - - FUTEX_SYSTEM_UNLOCK; - switch (ret) { - case EWOULDBLOCK: /* timeout */ - return ETIMEDOUT; - break; - case EINTR: /* signal */ - return EINTR; - break; - case 0: /* FUTEX_WAKE received */ -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX_WAIT %d: uaddr = %p, " - "got FUTEX_WAKE\n", - td->td_proc->p_pid, args->uaddr); -#endif - return 0; - break; - default: -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX_WAIT: unexpected ret = %d\n", - ret); -#endif - break; - } - - /* NOTREACHED */ - break; - - case LINUX_FUTEX_WAKE: - FUTEX_SYSTEM_LOCK; - - /* - * XXX: Linux is able to cope with different addresses - * corresponding to the same mapped memory in the sleeping - * and waker process(es). - */ -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX_WAKE %d: uaddr = %p, val = %d\n", - td->td_proc->p_pid, args->uaddr, args->val); -#endif - f = futex_get(args->uaddr, FUTEX_UNLOCKED); - td->td_retval[0] = futex_wake(f, args->val, NULL, 0); - futex_put(f); - - FUTEX_SYSTEM_UNLOCK; - break; - - case LINUX_FUTEX_CMP_REQUEUE: - FUTEX_SYSTEM_LOCK; - - if ((error = copyin(args->uaddr, - &val, sizeof(val))) != 0) { - FUTEX_SYSTEM_UNLOCK; - return error; - } - - if (val != args->val3) { - FUTEX_SYSTEM_UNLOCK; - return EAGAIN; - } - - f = futex_get(args->uaddr, FUTEX_UNLOCKED); - newf = futex_get(args->uaddr2, FUTEX_UNLOCKED); - td->td_retval[0] = futex_wake(f, args->val, newf, - (int)(unsigned long)args->timeout); - futex_put(f); - futex_put(newf); - - FUTEX_SYSTEM_UNLOCK; - break; - - case LINUX_FUTEX_WAKE_OP: - FUTEX_SYSTEM_LOCK; -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX_WAKE_OP: %d: uaddr = %p, op = %d, " - "val = %x, uaddr2 = %p, val3 = %x\n", - td->td_proc->p_pid, args->uaddr, args->op, - args->val, args->uaddr2, args->val3); -#endif - f = futex_get(args->uaddr, FUTEX_UNLOCKED); - f2 = futex_get(args->uaddr2, FUTEX_UNLOCKED); - - /* - * This function returns positive number as results and - * negative as errors - */ - op_ret = futex_atomic_op(td, args->val3, args->uaddr2); -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("futex_atomic_op ret %d\n", op_ret); -#endif - if (op_ret < 0) { - /* XXX: We don't handle the EFAULT yet. */ - if (op_ret != -EFAULT) { - futex_put(f); - futex_put(f2); - FUTEX_SYSTEM_UNLOCK; - return (-op_ret); +retry: + FUTEXES_LOCK; + LIST_FOREACH(f, &futex_list, f_list) { + if (f->f_uaddr == uaddr) { + if (tmpf != NULL) { + FUTEX_UNLOCK(tmpf); + FUTEX_DESTROY(tmpf); + free(tmpf, M_FUTEX); } + if (flags & FUTEX_DONTEXISTS) { + FUTEXES_UNLOCK; + return (EINVAL); + } - futex_put(f); - futex_put(f2); - - FUTEX_SYSTEM_UNLOCK; - return (EFAULT); - } - - ret = futex_wake(f, args->val, NULL, 0); - futex_put(f); - if (op_ret > 0) { - op_ret = 0; /* - * Linux abuses the address of the timespec parameter - * as the number of retries. + * Increment refcount of the found futex to + * prevent it from deallocation before FUTEX_LOCK() */ - op_ret += futex_wake(f2, - (int)(unsigned long)args->timeout, NULL, 0); - ret += op_ret; + ++f->f_refcount; + FUTEXES_UNLOCK; + + FUTEX_LOCK(f); + *newf = f; + return (0); } - futex_put(f2); - td->td_retval[0] = ret; + } - FUTEX_SYSTEM_UNLOCK; - break; + if (flags & FUTEX_DONTCREATE) { + FUTEXES_UNLOCK; + return (0); + } - case LINUX_FUTEX_LOCK_PI: - /* not yet implemented */ - return (ENOSYS); + if (tmpf == NULL) { + FUTEXES_UNLOCK; + tmpf = malloc(sizeof(*tmpf), M_FUTEX, M_WAITOK | M_ZERO); + tmpf->f_uaddr = uaddr; + tmpf->f_refcount = 1; + FUTEX_INIT(tmpf); + TAILQ_INIT(&tmpf->f_waiting_proc); - case LINUX_FUTEX_UNLOCK_PI: - /* not yet implemented */ - return (ENOSYS); - - case LINUX_FUTEX_TRYLOCK_PI: - /* not yet implemented */ - return (ENOSYS); - - case LINUX_FUTEX_REQUEUE: - /* - * Glibc does not use this operation since Jun 2004 (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. + * Lock the new futex before an insert into the futex_list + * to prevent futex usage by other. */ - em = em_find(td->td_proc, EMUL_DONTLOCK); - if (em->used_requeue == 0) { - printf("linux(%s (%d)) sys_futex: " - "unsupported futex_requeue op\n", - td->td_proc->p_comm, td->td_proc->p_pid); - em->used_requeue = 1; - } - return (EINVAL); - - default: - printf("linux_sys_futex: unknown op %d\n", - args->op); - return (ENOSYS); + FUTEX_LOCK(tmpf); + goto retry; } + + LIST_INSERT_HEAD(&futex_list, tmpf, f_list); + FUTEXES_UNLOCK; + + *newf = tmpf; return (0); } -static struct futex * -futex_get(void *uaddr, int locked) +static int +futex_get(uint32_t *uaddr, struct waiting_proc **wp, struct futex **f, + uint32_t flags) { - struct futex *f; + int error; - if (locked == FUTEX_UNLOCKED) - FUTEX_LOCK; - LIST_FOREACH(f, &futex_list, f_list) { - if (f->f_uaddr == uaddr) { - f->f_refcount++; - if (locked == FUTEX_UNLOCKED) - FUTEX_UNLOCK; - return f; - } + 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) { + 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; + } - f = malloc(sizeof(*f), M_LINUX, M_WAITOK); - f->f_uaddr = uaddr; - f->f_refcount = 1; - TAILQ_INIT(&f->f_waiting_proc); - LIST_INSERT_HEAD(&futex_list, f, f_list); - if (locked == FUTEX_UNLOCKED) - FUTEX_UNLOCK; - - return f; + return (error); } -static void -futex_put(f) - struct futex *f; +static int +futex_sleep(struct futex *f, struct waiting_proc *wp, unsigned long timeout) { - FUTEX_LOCK; - f->f_refcount--; - if (f->f_refcount == 0) { - LIST_REMOVE(f, f_list); - free(f, M_LINUX); + int error; + + FUTEX_ASSERT_LOCKED(f); + error = sx_sleep(wp, &f->f_lck, PCATCH, "futex", timeout); + if (wp->wp_flags & FUTEX_WP_REQUEUED) { + KASSERT(f != wp->wp_futex, ("futex != wp_futex")); + futex_put(f, NULL); + f = wp->wp_futex; + FUTEX_LOCK(f); } - FUTEX_UNLOCK; - return; + futex_put(f, wp); + return (error); } static int -futex_sleep(struct futex *f, struct thread *td, unsigned long timeout) +futex_wake(struct futex *f, int n) { - struct waiting_proc *wp; - int ret; + struct waiting_proc *wp, *wpt; + int count = 0; - wp = malloc(sizeof(*wp), M_LINUX, M_WAITOK); - wp->wp_t = td; - wp->wp_new_futex = NULL; - FUTEX_LOCK; - TAILQ_INSERT_TAIL(&f->f_waiting_proc, wp, wp_list); - FUTEX_UNLOCK; + FUTEX_ASSERT_LOCKED(f); + TAILQ_FOREACH_SAFE(wp, &f->f_waiting_proc, wp_list, wpt) { + wp->wp_flags |= FUTEX_WP_REMOVED; + TAILQ_REMOVE(&f->f_waiting_proc, wp, wp_list); + wakeup_one(wp); + if (++count == n) + break; + } -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX --> %d tlseep timeout = %ld\n", - td->td_proc->p_pid, timeout); -#endif - ret = tsleep(wp, PCATCH | PZERO, "linuxfutex", timeout); -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX -> %d tsleep returns %d\n", - td->td_proc->p_pid, ret); -#endif + return (count); +} - FUTEX_LOCK; - TAILQ_REMOVE(&f->f_waiting_proc, wp, wp_list); - FUTEX_UNLOCK; +static int +futex_requeue(struct futex *f, int n, struct futex *f2, int n2) +{ + struct waiting_proc *wp, *wpt; + int count = 0; - /* if we got woken up in futex_wake */ - if ((ret == 0) && (wp->wp_new_futex != NULL)) { - /* suspend us on the new futex */ - ret = futex_sleep(wp->wp_new_futex, td, timeout); - /* and release the old one */ - futex_put(wp->wp_new_futex); - } + FUTEX_ASSERT_LOCKED(f); + FUTEX_ASSERT_LOCKED(f2); - free(wp, M_LINUX); + TAILQ_FOREACH_SAFE(wp, &f->f_waiting_proc, wp_list, wpt) { + if (++count <= n) { + wp->wp_flags |= FUTEX_WP_REMOVED; + TAILQ_REMOVE(&f->f_waiting_proc, wp, wp_list); + wakeup_one(wp); + } else { + 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); - return ret; + /* + * 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 - n >= n2) + break; + } + } + + return (count); } static int -futex_wake(struct futex *f, int n, struct futex *newf, int n2) +futex_wait(struct futex *f, struct waiting_proc *wp, struct l_timespec *ts) { - struct waiting_proc *wp; - int count; + struct l_timespec timeout = {0, 0}; + struct timeval tv = {0, 0}; + int timeout_hz; + int error; + if (ts != NULL) { + error = copyin(ts, &timeout, sizeof(timeout)); + if (error) + return (error); + } + + tv.tv_usec = timeout.tv_sec * 1000000 + timeout.tv_nsec / 1000; + timeout_hz = tvtohz(&tv); + + if (timeout.tv_sec == 0 && timeout.tv_nsec == 0) + timeout_hz = 0; + /* - * Linux is very strange it wakes up N threads for - * all operations BUT requeue ones where its N+1 - * mimic this. + * If the user process requests a non null timeout, + * make sure we do not turn it into an infinite + * timeout because timeout_hz gets null. + * + * We use a minimal timeout of 1/hz. Maybe it would + * make sense to just return ETIMEDOUT without sleeping. */ - count = newf ? 0 : 1; + if (((timeout.tv_sec != 0) || (timeout.tv_nsec != 0)) && + (timeout_hz == 0)) + timeout_hz = 1; - FUTEX_LOCK; - TAILQ_FOREACH(wp, &f->f_waiting_proc, wp_list) { - if (count <= n) { - wakeup_one(wp); - count++; - } else { - if (newf != NULL) { - /* futex_put called after tsleep */ - wp->wp_new_futex = futex_get(newf->f_uaddr, - FUTEX_LOCKED); - wakeup_one(wp); - if (count - n >= n2) - break; - } - } - } - FUTEX_UNLOCK; + error = futex_sleep(f, wp, timeout_hz); + if (error == EWOULDBLOCK) + error = ETIMEDOUT; - return count; + return (error); } static int -futex_atomic_op(struct thread *td, int encoded_op, caddr_t uaddr) +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; #ifdef DEBUG if (ldebug(sys_futex)) printf("futex_atomic_op: op = %d, cmp = %d, oparg = %x, " "cmparg = %x, uaddr = %p\n", op, cmp, oparg, cmparg, uaddr); #endif /* XXX: linux verifies access here and returns EFAULT */ 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: ret = -ENOSYS; break; } if (ret) return (ret); switch (cmp) { case FUTEX_OP_CMP_EQ: return (oldval == cmparg); case FUTEX_OP_CMP_NE: return (oldval != cmparg); case FUTEX_OP_CMP_LT: return (oldval < cmparg); case FUTEX_OP_CMP_GE: return (oldval >= cmparg); case FUTEX_OP_CMP_LE: return (oldval <= cmparg); case FUTEX_OP_CMP_GT: return (oldval > cmparg); default: return (-ENOSYS); } } int +linux_sys_futex(struct thread *td, struct linux_sys_futex_args *args) +{ + int op_ret, val, ret, nrwake; + struct linux_emuldata *em; + struct waiting_proc *wp; + struct futex *f, *f2; + int error = 0; + + /* + * Our implementation provides only privates futexes. Most of the apps + * should use private futexes but don't claim so. Therefore we treat + * all futexes as private by clearing the FUTEX_PRIVATE_FLAG. It works + * in most cases (ie. when futexes are not shared on file descriptor + * or between different processes.). + */ + args->op = (args->op & ~LINUX_FUTEX_PRIVATE_FLAG); + + switch (args->op) { + case LINUX_FUTEX_WAIT: + +#ifdef DEBUG + if (ldebug(sys_futex)) + printf(ARGS(sys_futex, "futex_wait val %d uaddr %p"), + args->val, args->uaddr); +#endif + error = futex_get(args->uaddr, &wp, &f, FUTEX_CREATE_WP); + if (error) + return (error); + error = copyin(args->uaddr, &val, sizeof(val)); + if (error) { + futex_put(f, wp); + return (error); + } + if (val != args->val) { +#ifdef DEBUG + if (ldebug(sys_futex)) + printf(ARGS(sys_futex, "futex_wait uaddr %p WHOOPS %d != %d"), + args->uaddr, args->val, val); +#endif + futex_put(f, wp); + return (EWOULDBLOCK); + } + + error = futex_wait(f, wp, args->timeout); + break; + + case LINUX_FUTEX_WAKE: + + /* + * XXX: Linux is able to cope with different addresses + * corresponding to the same mapped memory in the sleeping + * and waker process(es). + */ +#ifdef DEBUG + if (ldebug(sys_futex)) + printf(ARGS(sys_futex, "futex_wake val %d uaddr %p"), + args->val, args->uaddr); +#endif + error = futex_get(args->uaddr, NULL, &f, FUTEX_DONTCREATE); + if (error) + return (error); + if (f == NULL) { + td->td_retval[0] = 0; + return (error);; + } + td->td_retval[0] = futex_wake(f, args->val); + futex_put(f, NULL); + break; + + case LINUX_FUTEX_CMP_REQUEUE: + +#ifdef DEBUG + if (ldebug(sys_futex)) + printf(ARGS(sys_futex, "futex_cmp_requeue uaddr %p " + "val %d val3 %d uaddr2 %p val2 %d"), + args->uaddr, args->val, args->val3, args->uaddr2, + (int)(unsigned long)args->timeout); +#endif + + /* + * Linux allows this, we would not, it is an incorrect + * usage of declared ABI, so return EINVAL. + */ + if (args->uaddr == args->uaddr2) + return (EINVAL); + error = futex_get0(args->uaddr, &f, 0); + if (error) + return (error); + + /* + * To avoid deadlocks return EINVAL if second futex + * exists at this time. Otherwise create the new futex + * and ignore false positive LOR which thus happens. + * + * Glibc fall back to FUTEX_WAKE in case of any error + * returned by FUTEX_CMP_REQUEUE. + */ + error = futex_get0(args->uaddr2, &f2, FUTEX_DONTEXISTS); + if (error) { + futex_put(f, NULL); + return (error); + } + error = copyin(args->uaddr, &val, sizeof(val)); + if (error) { + futex_put(f2, NULL); + futex_put(f, NULL); + return (error); + } + if (val != args->val3) { +#ifdef DEBUG + if (ldebug(sys_futex)) + printf(ARGS(sys_futex, "futex_cmp_requeue WHOOPS" + " VAL %d != UVAL %d"), args->val, val); +#endif + futex_put(f2, NULL); + futex_put(f, NULL); + return (EAGAIN); + } + + nrwake = (int)(unsigned long)args->timeout; + td->td_retval[0] = futex_requeue(f, args->val, f2, nrwake); + futex_put(f2, NULL); + futex_put(f, NULL); + break; + + case LINUX_FUTEX_WAKE_OP: + +#ifdef DEBUG + if (ldebug(sys_futex)) + printf(ARGS(sys_futex, "futex_wake_op " + "uaddr %p op %d val %x uaddr2 %p val3 %x"), + args->uaddr, args->op, args->val, + args->uaddr2, args->val3); +#endif + error = futex_get0(args->uaddr, &f, 0); + if (error) + return (error); + if (args->uaddr != args->uaddr2) + error = futex_get0(args->uaddr2, &f2, 0); + if (error) { + futex_put(f, NULL); + return (error); + } + + /* + * This function returns positive number as results and + * negative as errors + */ + op_ret = futex_atomic_op(td, args->val3, args->uaddr2); + + if (op_ret < 0) { + /* XXX: We don't handle the EFAULT yet. */ + if (op_ret != -EFAULT) { + if (f2 != NULL) + futex_put(f2, NULL); + futex_put(f, NULL); + return (-op_ret); + } + if (f2 != NULL) + futex_put(f2, NULL); + futex_put(f, NULL); + return (EFAULT); + } + + ret = futex_wake(f, args->val); + + if (op_ret > 0) { + op_ret = 0; + nrwake = (int)(unsigned long)args->timeout; + + if (f2 != NULL) + op_ret += futex_wake(f2, nrwake); + else + op_ret += futex_wake(f, nrwake); + ret += op_ret; + + } + if (f2 != NULL) + futex_put(f2, NULL); + futex_put(f, NULL); + td->td_retval[0] = ret; + break; + + case LINUX_FUTEX_LOCK_PI: + /* not yet implemented */ + return (ENOSYS); + + case LINUX_FUTEX_UNLOCK_PI: + /* not yet implemented */ + return (ENOSYS); + + case LINUX_FUTEX_TRYLOCK_PI: + /* not yet implemented */ + 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. + */ + em = em_find(td->td_proc, EMUL_DONTLOCK); + if (em->used_requeue == 0) { + printf("linux(%s (%d)) sys_futex: " + "unsupported futex_requeue op\n", + td->td_proc->p_comm, td->td_proc->p_pid); + em->used_requeue = 1; + } + return (EINVAL); + + default: + printf("linux_sys_futex: unknown op %d\n", args->op); + return (ENOSYS); + } + + return (error); +} + +int linux_set_robust_list(struct thread *td, struct linux_set_robust_list_args *args) { struct linux_emuldata *em; -#ifdef DEBUG +#ifdef DEBUG if (ldebug(set_robust_list)) - printf(ARGS(set_robust_list, "")); + printf(ARGS(set_robust_list, "head %p len %d"), + args->head, args->len); #endif + if (args->len != sizeof(struct linux_robust_list_head)) return (EINVAL); em = em_find(td->td_proc, EMUL_DOLOCK); em->robust_futexes = args->head; EMUL_UNLOCK(&emul_lock); 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); int error = 0; #ifdef DEBUG if (ldebug(get_robust_list)) printf(ARGS(get_robust_list, "")); #endif if (!args->pid) { em = em_find(td->td_proc, EMUL_DONTLOCK); head = em->robust_futexes; } else { struct proc *p; p = pfind(args->pid); if (p == NULL) return (ESRCH); em = em_find(p, EMUL_DONTLOCK); /* XXX: ptrace? */ if (priv_check(td, PRIV_CRED_SETUID) || priv_check(td, PRIV_CRED_SETEUID) || p_candebug(td, p)) return (EPERM); head = em->robust_futexes; PROC_UNLOCK(p); } error = copyout(&len, args->len, sizeof(l_size_t)); if (error) return (EFAULT); error = copyout(head, args->head, sizeof(struct linux_robust_list_head)); return (error); } static int -handle_futex_death(void *uaddr, pid_t pid, int pi) +handle_futex_death(struct proc *p, uint32_t *uaddr, int pi) { - int uval, nval, mval; + uint32_t uval, nval, mval; struct futex *f; + int error; retry: if (copyin(uaddr, &uval, 4)) return (EFAULT); - - if ((uval & FUTEX_TID_MASK) == pid) { + if ((uval & FUTEX_TID_MASK) == p->p_pid) { 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)) { - f = futex_get(uaddr, FUTEX_UNLOCKED); - futex_wake(f, 1, NULL, 0); + error = futex_get(uaddr, NULL, &f, + FUTEX_DONTCREATE); + if (error) + return (error); + if (f != NULL) { + futex_wake(f, 1); + futex_put(f, NULL); + } } } return (0); } static int fetch_robust_entry(struct linux_robust_list **entry, struct linux_robust_list **head, int *pi) { l_ulong uentry; if (copyin((const void *)head, &uentry, sizeof(l_ulong))) return (EFAULT); *entry = (void *)(uentry & ~1UL); *pi = uentry & 1; return (0); } /* This walks the list of robust futexes releasing them. */ void release_futexes(struct proc *p) { struct linux_robust_list_head *head = NULL; struct linux_robust_list *entry, *next_entry, *pending; unsigned int limit = 2048, pi, next_pi, pip; struct linux_emuldata *em; l_long futex_offset; int rc; em = em_find(p, EMUL_DONTLOCK); head = em->robust_futexes; if (head == NULL) return; if (fetch_robust_entry(&entry, PTRIN(&head->list.next), &pi)) return; if (copyin(&head->futex_offset, &futex_offset, sizeof(futex_offset))) 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((char *)entry + futex_offset, - p->p_pid, pi)) + if (handle_futex_death(p, (uint32_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((char *) pending + futex_offset, - p->p_pid, pip); + handle_futex_death(p, (uint32_t *)pending + futex_offset, pip); } Index: stable/7/sys/contrib/pf =================================================================== --- stable/7/sys/contrib/pf (revision 193151) +++ stable/7/sys/contrib/pf (revision 193152) Property changes on: stable/7/sys/contrib/pf ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head/sys/contrib/pf:r191719 Index: stable/7/sys/dev/ath/ath_hal =================================================================== --- stable/7/sys/dev/ath/ath_hal (revision 193151) +++ stable/7/sys/dev/ath/ath_hal (revision 193152) Property changes on: stable/7/sys/dev/ath/ath_hal ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head/sys/dev/ath/ath_hal:r191719 Index: stable/7/sys/dev/cxgb =================================================================== --- stable/7/sys/dev/cxgb (revision 193151) +++ stable/7/sys/dev/cxgb (revision 193152) Property changes on: stable/7/sys/dev/cxgb ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head/sys/dev/cxgb:r191719 Index: stable/7/sys/i386/linux/linux_sysvec.c =================================================================== --- stable/7/sys/i386/linux/linux_sysvec.c (revision 193151) +++ stable/7/sys/i386/linux/linux_sysvec.c (revision 193152) @@ -1,1149 +1,1149 @@ /*- * Copyright (c) 1994-1996 Søren Schmidt * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer * in this position and unchanged. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #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); MALLOC_DEFINE(M_LINUX, "linux", "Linux mode structures"); #if BYTE_ORDER == LITTLE_ENDIAN #define SHELLMAGIC 0x2123 /* #! */ #else #define SHELLMAGIC 0x2321 #endif /* * Allow the sendsig functions to use the ldebug() facility * even though they are not syscalls themselves. Map them * to syscall 0. This is slightly less bogus than using * ldebug(sigreturn). */ #define LINUX_SYS_linux_rt_sendsig 0 #define LINUX_SYS_linux_sendsig 0 extern char linux_sigcode[]; extern int linux_szsigcode; extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); SET_DECLARE(linux_device_handler_set, struct linux_device_handler); static int linux_fixup(register_t **stack_base, struct image_params *iparams); static int elf_linux_fixup(register_t **stack_base, struct image_params *iparams); static void linux_prepsyscall(struct trapframe *tf, int *args, u_int *code, caddr_t *params); static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask); static void exec_linux_setregs(struct thread *td, u_long entry, u_long stack, u_long ps_strings); static register_t *linux_copyout_strings(struct image_params *imgp); static int linux_szplatform; const char *linux_platform; extern LIST_HEAD(futex_list, futex) futex_list; -extern struct sx futex_sx; +extern struct mtx futex_mtx; static eventhandler_tag linux_exit_tag; static eventhandler_tag linux_schedtail_tag; static eventhandler_tag linux_exec_tag; /* * Linux syscalls return negative errno's, we do positive and map them * Reference: * FreeBSD: src/sys/sys/errno.h * Linux: linux-2.6.17.8/include/asm-generic/errno-base.h * linux-2.6.17.8/include/asm-generic/errno.h */ static int bsd_to_linux_errno[ELAST + 1] = { -0, -1, -2, -3, -4, -5, -6, -7, -8, -9, -10, -35, -12, -13, -14, -15, -16, -17, -18, -19, -20, -21, -22, -23, -24, -25, -26, -27, -28, -29, -30, -31, -32, -33, -34, -11,-115,-114, -88, -89, -90, -91, -92, -93, -94, -95, -96, -97, -98, -99, -100,-101,-102,-103,-104,-105,-106,-107,-108,-109, -110,-111, -40, -36,-112,-113, -39, -11, -87,-122, -116, -66, -6, -6, -6, -6, -6, -37, -38, -9, -6, -6, -43, -42, -75,-125, -84, -95, -16, -74, -72, -67, -71 }; int bsd_to_linux_signal[LINUX_SIGTBLSZ] = { LINUX_SIGHUP, LINUX_SIGINT, LINUX_SIGQUIT, LINUX_SIGILL, LINUX_SIGTRAP, LINUX_SIGABRT, 0, LINUX_SIGFPE, LINUX_SIGKILL, LINUX_SIGBUS, LINUX_SIGSEGV, LINUX_SIGSYS, LINUX_SIGPIPE, LINUX_SIGALRM, LINUX_SIGTERM, LINUX_SIGURG, LINUX_SIGSTOP, LINUX_SIGTSTP, LINUX_SIGCONT, LINUX_SIGCHLD, LINUX_SIGTTIN, LINUX_SIGTTOU, LINUX_SIGIO, LINUX_SIGXCPU, LINUX_SIGXFSZ, LINUX_SIGVTALRM, LINUX_SIGPROF, LINUX_SIGWINCH, 0, LINUX_SIGUSR1, LINUX_SIGUSR2 }; int linux_to_bsd_signal[LINUX_SIGTBLSZ] = { SIGHUP, SIGINT, SIGQUIT, SIGILL, SIGTRAP, SIGABRT, SIGBUS, SIGFPE, SIGKILL, SIGUSR1, SIGSEGV, SIGUSR2, SIGPIPE, SIGALRM, SIGTERM, SIGBUS, SIGCHLD, SIGCONT, SIGSTOP, SIGTSTP, SIGTTIN, SIGTTOU, SIGURG, SIGXCPU, SIGXFSZ, SIGVTALRM, SIGPROF, SIGWINCH, SIGIO, SIGURG, SIGSYS }; #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); (*stack_base)--; **stack_base = (intptr_t)(void *)envp; (*stack_base)--; **stack_base = (intptr_t)(void *)argv; (*stack_base)--; **stack_base = imgp->args->argc; return (0); } static int elf_linux_fixup(register_t **stack_base, struct image_params *imgp) { struct proc *p; Elf32_Auxargs *args; Elf32_Addr *uplatform; struct ps_strings *arginfo; register_t *pos; KASSERT(curthread->td_proc == imgp->proc && (curthread->td_proc->p_flag & P_SA) == 0, ("unsafe elf_linux_fixup(), should be curproc")); p = imgp->proc; arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; uplatform = (Elf32_Addr *)((caddr_t)arginfo - linux_szsigcode - linux_szplatform); args = (Elf32_Auxargs *)imgp->auxargs; pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2); 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, 0); 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(uplatform)); 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; (*stack_base)--; **stack_base = (register_t)imgp->args->argc; return (0); } /* * Copied from kern/kern_exec.c */ static register_t * linux_copyout_strings(struct image_params *imgp) { int argc, envc; char **vectp; char *stringp, *destp; register_t *stack_base; struct ps_strings *arginfo; struct proc *p; /* * Calculate string base and vector table pointers. * Also deal with signal trampoline code for this exec type. */ p = imgp->proc; arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; destp = (caddr_t)arginfo - linux_szsigcode - SPARE_USRSPACE - linux_szplatform - roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *)); /* * install sigcode */ copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo - linux_szsigcode), linux_szsigcode); /* * install LINUX_PLATFORM */ copyout(linux_platform, ((caddr_t)arginfo - linux_szsigcode - linux_szplatform), linux_szplatform); /* * If we have a valid auxargs ptr, prepare some room * on the stack. */ if (imgp->auxargs) { /* * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for * lower compatibility. */ imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : (LINUX_AT_COUNT * 2); /* * The '+ 2' is for the null pointers at the end of each of * the arg and env vector sets,and imgp->auxarg_size is room * for argument of Runtime loader. */ vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *)); } else { /* * The '+ 2' is for the null pointers at the end of each of * the arg and env vector sets */ vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) * sizeof(char *)); } /* * vectp also becomes our initial stack base */ stack_base = (register_t *)vectp; stringp = imgp->args->begin_argv; argc = imgp->args->argc; envc = imgp->args->envc; /* * Copy out strings - arguments and environment. */ copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); /* * Fill in "ps_strings" struct for ps, w, etc. */ suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); suword(&arginfo->ps_nargvstr, argc); /* * Fill in argument portion of vector table. */ for (; argc > 0; --argc) { suword(vectp++, (long)(intptr_t)destp); while (*stringp++ != 0) destp++; destp++; } /* a null vector table pointer separates the argp's from the envp's */ suword(vectp++, 0); suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); suword(&arginfo->ps_nenvstr, envc); /* * Fill in environment portion of vector table. */ for (; envc > 0; --envc) { suword(vectp++, (long)(intptr_t)destp); while (*stringp++ != 0) destp++; destp++; } /* end of vector table is a null pointer */ suword(vectp, 0); return (stack_base); } extern int _ucodesel, _udatasel; extern unsigned long linux_sznonrtsigcode; 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); #ifdef DEBUG if (ldebug(rt_sendsig)) printf(ARGS(rt_sendsig, "%p, %d, %p, %u"), catcher, sig, (void*)mask, code); #endif /* * Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { fp = (struct l_rt_sigframe *)(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. */ if (p->p_sysent->sv_sigtbl) if (sig <= p->p_sysent->sv_sigsize) sig = p->p_sysent->sv_sigtbl[_SIG_IDX(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 */ frame.sf_si.lsi_signo = sig; frame.sf_si.lsi_code = code; frame.sf_si.lsi_addr = ksi->ksi_addr; /* * 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.__bits[0]; 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_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); #ifdef DEBUG if (ldebug(rt_sendsig)) printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"), frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp, td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask); #endif if (copyout(&frame, fp, sizeof(frame)) != 0) { /* * Process has trashed its stack; give it an illegal * instruction to halt it in its tracks. */ #ifdef DEBUG if (ldebug(rt_sendsig)) printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"), fp, oonstack); #endif PROC_LOCK(p); sigexit(td, SIGILL); } /* * Build context to run handler in. */ regs->tf_esp = (int)fp; regs->tf_eip = PS_STRINGS - *(p->p_sysent->sv_szsigcode) + linux_sznonrtsigcode; 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, i; 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); #ifdef DEBUG if (ldebug(sendsig)) printf(ARGS(sendsig, "%p, %d, %p, %u"), catcher, sig, (void*)mask, code); #endif /* * Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { fp = (struct l_sigframe *)(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. */ if (p->p_sysent->sv_sigtbl) if (sig <= p->p_sysent->sv_sigsize) sig = p->p_sysent->sv_sigtbl[_SIG_IDX(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.__bits[0]; 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_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); for (i = 0; i < (LINUX_NSIG_WORDS-1); i++) frame.sf_extramask[i] = lmask.__bits[i+1]; 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 = PS_STRINGS - *(p->p_sysent->sv_szsigcode); 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 proc *p = td->td_proc; struct l_sigframe frame; struct trapframe *regs; l_sigset_t lmask; int eflags, i; ksiginfo_t ksi; regs = td->td_frame; #ifdef DEBUG if (ldebug(sigreturn)) printf(ARGS(sigreturn, "%p"), (void *)args->sfp); #endif /* * 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; /* * XXX do allow users to change the privileged flag PSL_RF. The * cpu sets PSL_RF in tf_eflags for faults. Debuggers should * sometimes set it there too. tf_eflags is kept in the signal * context during signal handling and there is no other place * to remember it, so the PSL_RF bit may be corrupted by the * signal handler without us knowing. Corruption of the PSL_RF * bit at worst causes one more or one less debugger trap, so * allowing it is fairly harmless. */ if (!EFLAGS_SECURE(eflags & ~PSL_RF, regs->tf_eflags & ~PSL_RF)) 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.__bits[0] = frame.sf_sc.sc_mask; for (i = 0; i < (LINUX_NSIG_WORDS-1); i++) lmask.__bits[i+1] = frame.sf_extramask[i]; PROC_LOCK(p); linux_to_bsd_sigset(&lmask, &td->td_sigmask); SIG_CANTMASK(td->td_sigmask); signotify(td); PROC_UNLOCK(p); /* * 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 proc *p = td->td_proc; struct l_ucontext uc; struct l_sigcontext *context; l_stack_t *lss; stack_t ss; struct trapframe *regs; int eflags; ksiginfo_t ksi; regs = td->td_frame; #ifdef DEBUG if (ldebug(rt_sigreturn)) printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp); #endif /* * 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; /* * XXX do allow users to change the privileged flag PSL_RF. The * cpu sets PSL_RF in tf_eflags for faults. Debuggers should * sometimes set it there too. tf_eflags is kept in the signal * context during signal handling and there is no other place * to remember it, so the PSL_RF bit may be corrupted by the * signal handler without us knowing. Corruption of the PSL_RF * bit at worst causes one more or one less debugger trap, so * allowing it is fairly harmless. */ if (!EFLAGS_SECURE(eflags & ~PSL_RF, regs->tf_eflags & ~PSL_RF)) 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); } PROC_LOCK(p); linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask); SIG_CANTMASK(td->td_sigmask); signotify(td); PROC_UNLOCK(p); /* * 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); #ifdef DEBUG if (ldebug(rt_sigreturn)) printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"), ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask); #endif (void)kern_sigaltstack(td, &ss, NULL); return (EJUSTRETURN); } /* * MPSAFE */ static void linux_prepsyscall(struct trapframe *tf, int *args, u_int *code, caddr_t *params) { args[0] = tf->tf_ebx; args[1] = tf->tf_ecx; args[2] = tf->tf_edx; args[3] = tf->tf_esi; args[4] = tf->tf_edi; args[5] = tf->tf_ebp; /* Unconfirmed */ *params = NULL; /* no copyin */ } /* * If a linux binary is exec'ing something, try this image activator * first. We override standard shell script execution in order to * be able to modify the interpreter path. We only do this if a linux * binary is doing the exec, so we do not create an EXEC module for it. */ static int exec_linux_imgact_try(struct image_params *iparams); static int exec_linux_imgact_try(struct image_params *imgp) { const char *head = (const char *)imgp->image_header; char *rpath; int error = -1, len; /* * The interpreter for shell scripts run from a linux binary needs * to be located in /compat/linux if possible in order to recursively * maintain linux path emulation. */ if (((const short *)head)[0] == SHELLMAGIC) { /* * Run our normal shell image activator. If it succeeds attempt * to use the alternate path for the interpreter. If an alternate * path is found, use our stringspace to store it. */ if ((error = exec_shell_imgact(imgp)) == 0) { linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc), imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0); if (rpath != NULL) { len = strlen(rpath) + 1; if (len <= MAXSHELLCMDLEN) { memcpy(imgp->interpreter_name, rpath, len); } free(rpath, M_TEMP); } } } return(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 exec_linux_setregs(struct thread *td, u_long entry, u_long stack, u_long ps_strings) { struct pcb *pcb = td->td_pcb; exec_setregs(td, entry, stack, ps_strings); /* Linux sets %gs to 0, we default to _udatasel */ pcb->pcb_gs = 0; load_gs(0); pcb->pcb_initial_npxcw = __LINUX_NPXCW__; } static void linux_get_machine(const char **dst) { switch (cpu_class) { case CPUCLASS_686: *dst = "i686"; break; case CPUCLASS_586: *dst = "i586"; break; case CPUCLASS_486: *dst = "i486"; break; default: *dst = "i386"; } } struct sysentvec linux_sysvec = { .sv_size = LINUX_SYS_MAXSYSCALL, .sv_table = linux_sysent, .sv_mask = 0, .sv_sigsize = LINUX_SIGTBLSZ, .sv_sigtbl = bsd_to_linux_signal, .sv_errsize = ELAST + 1, .sv_errtbl = bsd_to_linux_errno, .sv_transtrap = translate_traps, .sv_fixup = linux_fixup, .sv_sendsig = linux_sendsig, .sv_sigcode = linux_sigcode, .sv_szsigcode = &linux_szsigcode, .sv_prepsyscall = linux_prepsyscall, .sv_name = "Linux a.out", .sv_coredump = NULL, .sv_imgact_try = exec_linux_imgact_try, .sv_minsigstksz = LINUX_MINSIGSTKSZ, .sv_pagesize = PAGE_SIZE, .sv_minuser = VM_MIN_ADDRESS, .sv_maxuser = VM_MAXUSER_ADDRESS, .sv_usrstack = USRSTACK, .sv_psstrings = PS_STRINGS, .sv_stackprot = VM_PROT_ALL, .sv_copyout_strings = exec_copyout_strings, .sv_setregs = exec_linux_setregs, .sv_fixlimit = NULL, .sv_maxssiz = NULL }; struct sysentvec elf_linux_sysvec = { .sv_size = LINUX_SYS_MAXSYSCALL, .sv_table = linux_sysent, .sv_mask = 0, .sv_sigsize = LINUX_SIGTBLSZ, .sv_sigtbl = bsd_to_linux_signal, .sv_errsize = ELAST + 1, .sv_errtbl = bsd_to_linux_errno, .sv_transtrap = translate_traps, .sv_fixup = elf_linux_fixup, .sv_sendsig = linux_sendsig, .sv_sigcode = linux_sigcode, .sv_szsigcode = &linux_szsigcode, .sv_prepsyscall = linux_prepsyscall, .sv_name = "Linux ELF", .sv_coredump = elf32_coredump, .sv_imgact_try = exec_linux_imgact_try, .sv_minsigstksz = LINUX_MINSIGSTKSZ, .sv_pagesize = PAGE_SIZE, .sv_minuser = VM_MIN_ADDRESS, .sv_maxuser = VM_MAXUSER_ADDRESS, .sv_usrstack = USRSTACK, .sv_psstrings = PS_STRINGS, .sv_stackprot = VM_PROT_ALL, .sv_copyout_strings = linux_copyout_strings, .sv_setregs = exec_linux_setregs, .sv_fixlimit = NULL, .sv_maxssiz = NULL }; static char GNULINUX_ABI_VENDOR[] = "GNU"; static Elf_Brandnote linux_brandnote = { .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR), .hdr.n_descsz = 16, .hdr.n_type = 1, .vendor = GNULINUX_ABI_VENDOR, .flags = 0 }; static Elf32_Brandinfo linux_brand = { .brand = ELFOSABI_LINUX, .machine = EM_386, .compat_3_brand = "Linux", .emul_path = "/compat/linux", .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 = "/compat/linux", .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 }; Elf32_Brandinfo *linux_brandlist[] = { &linux_brand, &linux_glibc2brand, NULL }; static int linux_elf_modevent(module_t mod, int type, void *data) { Elf32_Brandinfo **brandinfo; int error; struct linux_ioctl_handler **lihp; struct linux_device_handler **ldhp; 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); SET_FOREACH(ldhp, linux_device_handler_set) linux_device_register_handler(*ldhp); mtx_init(&emul_lock, "emuldata lock", NULL, MTX_DEF); sx_init(&emul_shared_lock, "emuldata->shared lock"); LIST_INIT(&futex_list); - sx_init(&futex_sx, "futex protection lock"); + mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF); linux_exit_tag = EVENTHANDLER_REGISTER(process_exit, linux_proc_exit, NULL, 1000); linux_schedtail_tag = EVENTHANDLER_REGISTER(schedtail, linux_schedtail, NULL, 1000); linux_exec_tag = EVENTHANDLER_REGISTER(process_exec, linux_proc_exec, NULL, 1000); linux_get_machine(&linux_platform); linux_szplatform = roundup(strlen(linux_platform) + 1, sizeof(char *)); 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); SET_FOREACH(ldhp, linux_device_handler_set) linux_device_unregister_handler(*ldhp); mtx_destroy(&emul_lock); sx_destroy(&emul_shared_lock); - sx_destroy(&futex_sx); + mtx_destroy(&futex_mtx); EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag); EVENTHANDLER_DEREGISTER(schedtail, linux_schedtail_tag); EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag); 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(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); Index: stable/7/sys =================================================================== --- stable/7/sys (revision 193151) +++ stable/7/sys (revision 193152) Property changes on: stable/7/sys ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head/sys:r191719