static void
ia32_osendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
{
	struct ia32_sigframe3 sf, *fp;
	struct proc *p;
	struct thread *td;
	struct sigacts *psp;
	struct trapframe *regs;
	int sig;
	int oonstack;

	td = curthread;
	p = td->td_proc;
	PROC_LOCK_ASSERT(p, MA_OWNED);
	sig = ksi->ksi_signo;
	psp = p->p_sigacts;
	mtx_assert(&psp->ps_mtx, MA_OWNED);
	regs = td->td_frame;
	oonstack = sigonstack(regs->tf_rsp);

	/* Allocate space for the signal handler context. */
	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
		fp = (struct ia32_sigframe3 *)(td->td_sigstk.ss_sp +
		    td->td_sigstk.ss_size - sizeof(sf));
		td->td_sigstk.ss_flags |= SS_ONSTACK;
	} else
		fp = (struct ia32_sigframe3 *)regs->tf_rsp - 1;

	/* Translate the signal if appropriate. */
	if (p->p_sysent->sv_sigtbl && sig <= p->p_sysent->sv_sigsize)
		sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];

	/* Build the argument list for the signal handler. */
	sf.sf_signum = sig;
	sf.sf_scp = (register_t)&fp->sf_siginfo.si_sc;
	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
		/* Signal handler installed with SA_SIGINFO. */
		sf.sf_arg2 = (register_t)&fp->sf_siginfo;
		sf.sf_siginfo.si_signo = sig;
		sf.sf_siginfo.si_code = ksi->ksi_code;
		sf.sf_ah = (uintptr_t)catcher;
	} else {
		/* Old FreeBSD-style arguments. */
		sf.sf_arg2 = ksi->ksi_code;
		sf.sf_addr = (register_t)ksi->ksi_addr;
		sf.sf_ah = (uintptr_t)catcher;
	}
	mtx_unlock(&psp->ps_mtx);
	PROC_UNLOCK(p);

	/* Save most if not all of trap frame. */
	sf.sf_siginfo.si_sc.sc_eax = regs->tf_rax;
	sf.sf_siginfo.si_sc.sc_ebx = regs->tf_rbx;
	sf.sf_siginfo.si_sc.sc_ecx = regs->tf_rcx;
	sf.sf_siginfo.si_sc.sc_edx = regs->tf_rdx;
	sf.sf_siginfo.si_sc.sc_esi = regs->tf_rsi;
	sf.sf_siginfo.si_sc.sc_edi = regs->tf_rdi;
	sf.sf_siginfo.si_sc.sc_cs = regs->tf_cs;
	sf.sf_siginfo.si_sc.sc_ds = regs->tf_ds;
	sf.sf_siginfo.si_sc.sc_ss = regs->tf_ss;
	sf.sf_siginfo.si_sc.sc_es = regs->tf_es;
	sf.sf_siginfo.si_sc.sc_fs = regs->tf_fs;
	sf.sf_siginfo.si_sc.sc_gs = regs->tf_gs;
	sf.sf_siginfo.si_sc.sc_isp = regs->tf_rsp;

	/* Build the signal context to be used by osigreturn(). */
	sf.sf_siginfo.si_sc.sc_onstack = (oonstack) ? 1 : 0;
	SIG2OSIG(*mask, sf.sf_siginfo.si_sc.sc_mask);
	sf.sf_siginfo.si_sc.sc_esp = regs->tf_rsp;
	sf.sf_siginfo.si_sc.sc_ebp = regs->tf_rbp;
	sf.sf_siginfo.si_sc.sc_eip = regs->tf_rip;
	sf.sf_siginfo.si_sc.sc_eflags = regs->tf_rflags;
	sf.sf_siginfo.si_sc.sc_trapno = regs->tf_trapno;
	sf.sf_siginfo.si_sc.sc_err = regs->tf_err;

	/*
	 * Copy the sigframe out to the user's stack.
	 */
	if (copyout(&sf, fp, sizeof(*fp)) != 0) {
#ifdef DEBUG
		printf("process %ld has trashed its stack\n", (long)p->p_pid);
#endif
		PROC_LOCK(p);
		sigexit(td, SIGILL);
	}

	regs->tf_rsp = (uintptr_t)fp;
	regs->tf_rip = p->p_sysent->sv_psstrings - sz_ia32_osigcode;
	regs->tf_rflags &= ~(PSL_T | PSL_D);
	regs->tf_cs = _ucode32sel;
	regs->tf_ds = _udatasel;
	regs->tf_es = _udatasel;
	regs->tf_fs = _udatasel;
	regs->tf_ss = _udatasel;
	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
	PROC_LOCK(p);
	mtx_lock(&psp->ps_mtx);
}

static int
linux_common_open(struct thread *td, int dirfd, char *path, int l_flags, int mode)
{
	cap_rights_t rights;
	struct proc *p = td->td_proc;
	struct file *fp;
	int fd;
	int bsd_flags, error;

	bsd_flags = 0;
	switch (l_flags & LINUX_O_ACCMODE) {
	case LINUX_O_WRONLY:
		bsd_flags |= O_WRONLY;
		break;
	case LINUX_O_RDWR:
		bsd_flags |= O_RDWR;
		break;
	default:
		bsd_flags |= O_RDONLY;
	}
	if (l_flags & LINUX_O_NDELAY)
		bsd_flags |= O_NONBLOCK;
	if (l_flags & LINUX_O_APPEND)
		bsd_flags |= O_APPEND;
	if (l_flags & LINUX_O_SYNC)
		bsd_flags |= O_FSYNC;
	if (l_flags & LINUX_O_NONBLOCK)
		bsd_flags |= O_NONBLOCK;
	if (l_flags & LINUX_FASYNC)
		bsd_flags |= O_ASYNC;
	if (l_flags & LINUX_O_CREAT)
		bsd_flags |= O_CREAT;
	if (l_flags & LINUX_O_TRUNC)
		bsd_flags |= O_TRUNC;
	if (l_flags & LINUX_O_EXCL)
		bsd_flags |= O_EXCL;
	if (l_flags & LINUX_O_NOCTTY)
		bsd_flags |= O_NOCTTY;
	if (l_flags & LINUX_O_DIRECT)
		bsd_flags |= O_DIRECT;
	if (l_flags & LINUX_O_NOFOLLOW)
		bsd_flags |= O_NOFOLLOW;
	if (l_flags & LINUX_O_DIRECTORY)
		bsd_flags |= O_DIRECTORY;
	/* XXX LINUX_O_NOATIME: unable to be easily implemented. */

	error = kern_openat(td, dirfd, path, UIO_SYSSPACE, bsd_flags, mode);
	if (error != 0)
		goto done;
	if (bsd_flags & O_NOCTTY)
		goto done;

	/*
	 * XXX In between kern_open() and fget(), another process
	 * having the same filedesc could use that fd without
	 * checking below.
	*/
	fd = td->td_retval[0];
	if (fget(td, fd, cap_rights_init(&rights, CAP_IOCTL), &fp) == 0) {
		if (fp->f_type != DTYPE_VNODE) {
			fdrop(fp, td);
			goto done;
		}
		sx_slock(&proctree_lock);
		PROC_LOCK(p);
		if (SESS_LEADER(p) && !(p->p_flag & P_CONTROLT)) {
			PROC_UNLOCK(p);
			sx_sunlock(&proctree_lock);
			/* XXXPJD: Verify if TIOCSCTTY is allowed. */
			(void) fo_ioctl(fp, TIOCSCTTY, (caddr_t) 0,
			    td->td_ucred, td);
		} else {
			PROC_UNLOCK(p);
			sx_sunlock(&proctree_lock);
		}
		fdrop(fp, td);
	}

done:
#ifdef DEBUG
	if (ldebug(open))
		printf(LMSG("open returns error %d"), error);
#endif
	LFREEPATH(path);
	return (error);
}

//.........这里部分代码省略.........
			 * userland that actually runs in a priveledged mode
			 * but uses USR mode permissions for its accesses.
			 */
			user = 1;
			ksig.signb = SIGSEGV;
			ksig.code = 0;
			goto do_trapsignal;
		}
	} else {
		map = &td->td_proc->p_vmspace->vm_map;
	}

	/*
	 * We need to know whether the page should be mapped
	 * as R or R/W. The MMU does not give us the info as
	 * to whether the fault was caused by a read or a write.
	 *
	 * However, we know that a permission fault can only be
	 * the result of a write to a read-only location, so
	 * we can deal with those quickly.
	 *
	 * Otherwise we need to disassemble the instruction
	 * responsible to determine if it was a write.
	 */
	if (IS_PERMISSION_FAULT(fsr))
		ftype = VM_PROT_WRITE;
	else {
		u_int insn = ReadWord(tf->tf_pc);

		if (((insn & 0x0c100000) == 0x04000000) ||	/* STR/STRB */
		    ((insn & 0x0e1000b0) == 0x000000b0) ||	/* STRH/STRD */
		    ((insn & 0x0a100000) == 0x08000000)) {	/* STM/CDT */
			ftype = VM_PROT_WRITE;
		} else {
			if ((insn & 0x0fb00ff0) == 0x01000090)	/* SWP */
				ftype = VM_PROT_READ | VM_PROT_WRITE;
			else
				ftype = VM_PROT_READ;
		}
	}

	/*
	 * See if the fault is as a result of ref/mod emulation,
	 * or domain mismatch.
	 */
#ifdef DEBUG
	last_fault_code = fsr;
#endif
	if (pmap_fault_fixup(vmspace_pmap(td->td_proc->p_vmspace), va, ftype,
	    user)) {
		goto out;
	}

	onfault = pcb->pcb_onfault;
	pcb->pcb_onfault = NULL;
	if (map != kernel_map) {
		PROC_LOCK(p);
		p->p_lock++;
		PROC_UNLOCK(p);
	}
	error = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
	pcb->pcb_onfault = onfault;

	if (map != kernel_map) {
		PROC_LOCK(p);
		p->p_lock--;
		PROC_UNLOCK(p);
	}
	if (__predict_true(error == 0))
		goto out;
	if (user == 0) {
		if (pcb->pcb_onfault) {
			tf->tf_r0 = error;
			tf->tf_pc = (register_t)(intptr_t) pcb->pcb_onfault;
			return;
		}

		printf("\nvm_fault(%p, %x, %x, 0) -> %x\n", map, va, ftype,
		    error);
		dab_fatal(tf, fsr, far, td, &ksig);
	}


	if (error == ENOMEM) {
		printf("VM: pid %d (%s), uid %d killed: "
		    "out of swap\n", td->td_proc->p_pid, td->td_name,
		    (td->td_proc->p_ucred) ?
		     td->td_proc->p_ucred->cr_uid : -1);
		ksig.signb = SIGKILL;
	} else {
		ksig.signb = SIGSEGV;
	}
	ksig.code = 0;
do_trapsignal:
	call_trapsignal(td, ksig.signb, ksig.code);
out:
	/* If returning to user mode, make sure to invoke userret() */
	if (user)
		userret(td, tf);
}

static void
pmclog_loop(void *arg)
{
	int error;
	struct pmc_owner *po;
	struct pmclog_buffer *lb;
	struct proc *p;
	struct ucred *ownercred;
	struct ucred *mycred;
	struct thread *td;
	struct uio auio;
	struct iovec aiov;
	size_t nbytes;

	po = (struct pmc_owner *) arg;
	p = po->po_owner;
	td = curthread;
	mycred = td->td_ucred;

	PROC_LOCK(p);
	ownercred = crhold(p->p_ucred);
	PROC_UNLOCK(p);

	PMCDBG(LOG,INI,1, "po=%p kt=%p", po, po->po_kthread);
	KASSERT(po->po_kthread == curthread->td_proc,
	    ("[pmclog,%d] proc mismatch po=%p po/kt=%p curproc=%p", __LINE__,
		po, po->po_kthread, curthread->td_proc));

	lb = NULL;


	/*
	 * Loop waiting for I/O requests to be added to the owner
	 * struct's queue.  The loop is exited when the log file
	 * is deconfigured.
	 */

	mtx_lock(&pmc_kthread_mtx);

	for (;;) {

		/* check if we've been asked to exit */
		if ((po->po_flags & PMC_PO_OWNS_LOGFILE) == 0)
			break;

		if (lb == NULL) { /* look for a fresh buffer to write */
			mtx_lock_spin(&po->po_mtx);
			if ((lb = TAILQ_FIRST(&po->po_logbuffers)) == NULL) {
				mtx_unlock_spin(&po->po_mtx);

				/* No more buffers and shutdown required. */
				if (po->po_flags & PMC_PO_SHUTDOWN) {
					mtx_unlock(&pmc_kthread_mtx);
					/*
			 		 * Close the file to get PMCLOG_EOF
					 * error in pmclog(3).
					 */
					fo_close(po->po_file, curthread);
					mtx_lock(&pmc_kthread_mtx);
					break;
				}

				(void) msleep(po, &pmc_kthread_mtx, PWAIT,
				    "pmcloop", 0);
				continue;
			}

			TAILQ_REMOVE(&po->po_logbuffers, lb, plb_next);
			mtx_unlock_spin(&po->po_mtx);
		}

		mtx_unlock(&pmc_kthread_mtx);

		/* process the request */
		PMCDBG(LOG,WRI,2, "po=%p base=%p ptr=%p", po,
		    lb->plb_base, lb->plb_ptr);
		/* change our thread's credentials before issuing the I/O */

		aiov.iov_base = lb->plb_base;
		aiov.iov_len  = nbytes = lb->plb_ptr - lb->plb_base;

		auio.uio_iov    = &aiov;
		auio.uio_iovcnt = 1;
		auio.uio_offset = -1;
		auio.uio_resid  = nbytes;
		auio.uio_rw     = UIO_WRITE;
		auio.uio_segflg = UIO_SYSSPACE;
		auio.uio_td     = td;

		/* switch thread credentials -- see kern_ktrace.c */
		td->td_ucred = ownercred;
		error = fo_write(po->po_file, &auio, ownercred, 0, td);
		td->td_ucred = mycred;

		if (error) {
			/* XXX some errors are recoverable */
			/* send a SIGIO to the owner and exit */
			PROC_LOCK(p);
			kern_psignal(p, SIGIO);
			PROC_UNLOCK(p);
//.........这里部分代码省略.........

/*
 * Exit: deallocate address space and other resources, change proc state to
 * zombie, and unlink proc from allproc and parent's lists.  Save exit status
 * and rusage for wait().  Check for child processes and orphan them.
 */
void
exit1(struct thread *td, int rv)
{
	struct proc *p, *nq, *q;
	struct vnode *vtmp;
	struct vnode *ttyvp = NULL;
	struct plimit *plim;

	mtx_assert(&Giant, MA_NOTOWNED);

	p = td->td_proc;
	/*
	 * XXX in case we're rebooting we just let init die in order to
	 * work around an unsolved stack overflow seen very late during
	 * shutdown on sparc64 when the gmirror worker process exists.
	 */
	if (p == initproc && rebooting == 0) {
		printf("init died (signal %d, exit %d)\n",
		    WTERMSIG(rv), WEXITSTATUS(rv));
		panic("Going nowhere without my init!");
	}

	/*
	 * MUST abort all other threads before proceeding past here.
	 */
	PROC_LOCK(p);
	while (p->p_flag & P_HADTHREADS) {
		/*
		 * First check if some other thread got here before us.
		 * If so, act appropriately: exit or suspend.
		 */
		thread_suspend_check(0);

		/*
		 * Kill off the other threads. This requires
		 * some co-operation from other parts of the kernel
		 * so it may not be instantaneous.  With this state set
		 * any thread entering the kernel from userspace will
		 * thread_exit() in trap().  Any thread attempting to
		 * sleep will return immediately with EINTR or EWOULDBLOCK
		 * which will hopefully force them to back out to userland
		 * freeing resources as they go.  Any thread attempting
		 * to return to userland will thread_exit() from userret().
		 * thread_exit() will unsuspend us when the last of the
		 * other threads exits.
		 * If there is already a thread singler after resumption,
		 * calling thread_single will fail; in that case, we just
		 * re-check all suspension request, the thread should
		 * either be suspended there or exit.
		 */
		if (!thread_single(SINGLE_EXIT))
			break;

		/*
		 * All other activity in this process is now stopped.
		 * Threading support has been turned off.
		 */
	}
	KASSERT(p->p_numthreads == 1,
	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
	racct_sub(p, RACCT_NTHR, 1);
	/*
	 * Wakeup anyone in procfs' PIOCWAIT.  They should have a hold
	 * on our vmspace, so we should block below until they have
	 * released their reference to us.  Note that if they have
	 * requested S_EXIT stops we will block here until they ack
	 * via PIOCCONT.
	 */
	_STOPEVENT(p, S_EXIT, rv);

	/*
	 * Ignore any pending request to stop due to a stop signal.
	 * Once P_WEXIT is set, future requests will be ignored as
	 * well.
	 */
	p->p_flag &= ~P_STOPPED_SIG;
	KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));

	/*
	 * Note that we are exiting and do another wakeup of anyone in
	 * PIOCWAIT in case they aren't listening for S_EXIT stops or
	 * decided to wait again after we told them we are exiting.
	 */
	p->p_flag |= P_WEXIT;
	wakeup(&p->p_stype);

	/*
	 * Wait for any processes that have a hold on our vmspace to
	 * release their reference.
	 */
	while (p->p_lock > 0)
		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);

	p->p_xstat = rv;	/* Let event handler change exit status */
	PROC_UNLOCK(p);
//.........这里部分代码省略.........

/*
 * 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);
}

/*
 * The CheriABI version of sendsig(9) largely borrows from the MIPS version,
 * and it is important to keep them in sync.  It differs primarily in that it
 * must also be aware of user stack-handling ABIs, so is also sensitive to our
 * (fluctuating) design choices in how $stc and $sp interact.  The current
 * design uses ($stc + $sp) for stack-relative references, so early on we have
 * to calculate a 'relocated' version of $sp that we can then use for
 * MIPS-style access.
 *
 * This code, as with the CHERI-aware MIPS code, makes a privilege
 * determination in order to decide whether to trust the stack exposed by the
 * user code for the purposes of signal handling.  We must use the alternative
 * stack if there is any indication that using the user thread's stack state
 * might violate the userspace compartmentalisation model.
 */
static void
cheriabi_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
{
	struct proc *p;
	struct thread *td;
	struct trapframe *regs;
	struct sigacts *psp;
	struct sigframe_c sf, *sfp;
	uintptr_t stackbase;
	vm_offset_t sp;
	int cheri_is_sandboxed;
	int sig;
	int oonstack;

	td = curthread;
	p = td->td_proc;
	PROC_LOCK_ASSERT(p, MA_OWNED);
	sig = ksi->ksi_signo;
	psp = p->p_sigacts;
	mtx_assert(&psp->ps_mtx, MA_OWNED);

	regs = td->td_frame;

	/*
	 * In CheriABI, $sp is $stc relative, so calculate a relocation base
	 * that must be combined with regs->sp from this point onwards.
	 * Unfortunately, we won't retain bounds and permissions information
	 * (as is the case elsewhere in CheriABI).  While 'stackbase'
	 * suggests that $stc's offset isn't included, in practice it will be,
	 * although we may reasonably assume that it will be zero.
	 *
	 * If it turns out we will be delivering to the alternative signal
	 * stack, we'll recalculate stackbase later.
	 */
	CHERI_CLC(CHERI_CR_CTEMP0, CHERI_CR_KDC, &td->td_pcb->pcb_regs.stc,
	    0);
	CHERI_CTOPTR(stackbase, CHERI_CR_CTEMP0, CHERI_CR_KDC);
	oonstack = sigonstack(stackbase + regs->sp);

	/*
	 * CHERI affects signal delivery in the following ways:
	 *
	 * (1) Additional capability-coprocessor state is exposed via
	 *     extensions to the context frame placed on the stack.
	 *
	 * (2) If the user $pcc doesn't include CHERI_PERM_SYSCALL, then we
	 *     consider user state to be 'sandboxed' and therefore to require
	 *     special delivery handling which includes a domain-switch to the
	 *     thread's context-switch domain.  (This is done by
	 *     cheri_sendsig()).
	 *
	 * (3) If an alternative signal stack is not defined, and we are in a
	 *     'sandboxed' state, then we have two choices: (a) if the signal
	 *     is of type SA_SANDBOX_UNWIND, we will automatically unwind the
	 *     trusted stack by one frame; (b) otherwise, we will terminate
	 *     the process unconditionally.
	 */
	cheri_is_sandboxed = cheri_signal_sandboxed(td);

	/*
	 * We provide the ability to drop into the debugger in two different
	 * circumstances: (1) if the code running is sandboxed; and (2) if the
	 * fault is a CHERI protection fault.  Handle both here for the
	 * non-unwind case.  Do this before we rewrite any general-purpose or
	 * capability register state for the thread.
	 */
#if DDB
	if (cheri_is_sandboxed && security_cheri_debugger_on_sandbox_signal)
		kdb_enter(KDB_WHY_CHERI, "Signal delivery to CHERI sandbox");
	else if (sig == SIGPROT && security_cheri_debugger_on_sigprot)
		kdb_enter(KDB_WHY_CHERI,
		    "SIGPROT delivered outside sandbox");
#endif

	/*
	 * If a thread is running sandboxed, we can't rely on $sp which may
	 * not point at a valid stack in the ambient context, or even be
	 * maliciously manipulated.  We must therefore always use the
	 * alternative stack.  We are also therefore unable to tell whether we
	 * are on the alternative stack, so must clear 'oonstack' here.
	 *
	 * XXXRW: This requires significant further thinking; however, the net
	 * upshot is that it is not a good idea to do an object-capability
	 * invoke() from a signal handler, as with so many other things in
	 * life.
//.........这里部分代码省略.........

int
ext2_bmaparray(struct vnode *vp, daddr_t bn, daddr_t *bnp, int *runp, int *runb)
{
	struct inode *ip;
	struct buf *bp;
	struct ext2mount *ump;
	struct mount *mp;
	struct indir a[NIADDR+1], *ap;
	daddr_t daddr;
	e2fs_lbn_t metalbn;
	int error, num, maxrun = 0, bsize;
	int *nump;

	ap = NULL;
	ip = VTOI(vp);
	mp = vp->v_mount;
	ump = VFSTOEXT2(mp);

	bsize = EXT2_BLOCK_SIZE(ump->um_e2fs);

	if (runp) {
		maxrun = mp->mnt_iosize_max / bsize - 1;
		*runp = 0;
	}

	if (runb) {
		*runb = 0;
	}


	ap = a;
	nump = #
	error = ext2_getlbns(vp, bn, ap, nump);
	if (error)
		return (error);

	num = *nump;
	if (num == 0) {
		*bnp = blkptrtodb(ump, ip->i_db[bn]);
		if (*bnp == 0) {
			*bnp = -1;
		} else if (runp) {
			daddr_t bnb = bn;
			for (++bn; bn < NDADDR && *runp < maxrun &&
			    is_sequential(ump, ip->i_db[bn - 1], ip->i_db[bn]);
			    ++bn, ++*runp);
			bn = bnb;
			if (runb && (bn > 0)) {
				for (--bn; (bn >= 0) && (*runb < maxrun) &&
					is_sequential(ump, ip->i_db[bn],
						ip->i_db[bn + 1]);
						--bn, ++*runb);
			}
		}
		return (0);
	}


	/* Get disk address out of indirect block array */
	daddr = ip->i_ib[ap->in_off];

	for (bp = NULL, ++ap; --num; ++ap) {
		/*
		 * Exit the loop if there is no disk address assigned yet and
		 * the indirect block isn't in the cache, or if we were
		 * looking for an indirect block and we've found it.
		 */

		metalbn = ap->in_lbn;
		if ((daddr == 0 && !incore(&vp->v_bufobj, metalbn)) || metalbn == bn)
			break;
		/*
		 * If we get here, we've either got the block in the cache
		 * or we have a disk address for it, go fetch it.
		 */
		if (bp)
			bqrelse(bp);

		bp = getblk(vp, metalbn, bsize, 0, 0, 0);
		if ((bp->b_flags & B_CACHE) == 0) {
#ifdef INVARIANTS
			if (!daddr)
				panic("ext2_bmaparray: indirect block not in cache");
#endif
			bp->b_blkno = blkptrtodb(ump, daddr);
			bp->b_iocmd = BIO_READ;
			bp->b_flags &= ~B_INVAL;
			bp->b_ioflags &= ~BIO_ERROR;
			vfs_busy_pages(bp, 0);
			bp->b_iooffset = dbtob(bp->b_blkno);
			bstrategy(bp);
#ifdef RACCT
			if (racct_enable) {
				PROC_LOCK(curproc);
				racct_add_buf(curproc, bp, 0);
				PROC_UNLOCK(curproc);
			}
#endif
			curthread->td_ru.ru_inblock++;
			error = bufwait(bp);
//.........这里部分代码省略.........

int
linux_clone(struct thread *td, struct linux_clone_args *args)
{
	int error, ff = RFPROC | RFSTOPPED;
	struct proc *p2;
	struct thread *td2;
	int exit_signal;
	struct linux_emuldata *em;

#ifdef DEBUG
	if (ldebug(clone)) {
		printf(ARGS(clone, "flags %x, stack %p, parent tid: %p, "
		    "child tid: %p"), (unsigned)args->flags,
		    args->stack, args->parent_tidptr, args->child_tidptr);
	}
#endif

	exit_signal = args->flags & 0x000000ff;
	if (LINUX_SIG_VALID(exit_signal)) {
		if (exit_signal <= LINUX_SIGTBLSZ)
			exit_signal =
			    linux_to_bsd_signal[_SIG_IDX(exit_signal)];
	} else if (exit_signal != 0)
		return (EINVAL);

	if (args->flags & LINUX_CLONE_VM)
		ff |= RFMEM;
	if (args->flags & LINUX_CLONE_SIGHAND)
		ff |= RFSIGSHARE;
	/*
	 * XXX: In Linux, sharing of fs info (chroot/cwd/umask)
	 * and open files is independant.  In FreeBSD, its in one
	 * structure but in reality it does not cause any problems
	 * because both of these flags are usually set together.
	 */
	if (!(args->flags & (LINUX_CLONE_FILES | LINUX_CLONE_FS)))
		ff |= RFFDG;

	/*
	 * Attempt to detect when linux_clone(2) is used for creating
	 * kernel threads. Unfortunately despite the existence of the
	 * CLONE_THREAD flag, version of linuxthreads package used in
	 * most popular distros as of beginning of 2005 doesn't make
	 * any use of it. Therefore, this detection relies on
	 * empirical observation that linuxthreads sets certain
	 * combination of flags, so that we can make more or less
	 * precise detection and notify the FreeBSD kernel that several
	 * processes are in fact part of the same threading group, so
	 * that special treatment is necessary for signal delivery
	 * between those processes and fd locking.
	 */
	if ((args->flags & 0xffffff00) == LINUX_THREADING_FLAGS)
		ff |= RFTHREAD;

	if (args->flags & LINUX_CLONE_PARENT_SETTID)
		if (args->parent_tidptr == NULL)
			return (EINVAL);

	error = fork1(td, ff, 0, &p2, NULL, 0);
	if (error)
		return (error);

	if (args->flags & (LINUX_CLONE_PARENT | LINUX_CLONE_THREAD)) {
	   	sx_xlock(&proctree_lock);
		PROC_LOCK(p2);
		proc_reparent(p2, td->td_proc->p_pptr);
		PROC_UNLOCK(p2);
		sx_xunlock(&proctree_lock);
	}

	/* create the emuldata */
	error = linux_proc_init(td, p2->p_pid, args->flags);
	/* reference it - no need to check this */
	em = em_find(p2, EMUL_DOLOCK);
	KASSERT(em != NULL, ("clone: emuldata not found."));
	/* and adjust it */

	if (args->flags & LINUX_CLONE_THREAD) {
#ifdef notyet
	   	PROC_LOCK(p2);
	   	p2->p_pgrp = td->td_proc->p_pgrp;
	   	PROC_UNLOCK(p2);
#endif
		exit_signal = 0;
	}

	if (args->flags & LINUX_CLONE_CHILD_SETTID)
		em->child_set_tid = args->child_tidptr;
	else
	   	em->child_set_tid = NULL;

	if (args->flags & LINUX_CLONE_CHILD_CLEARTID)
		em->child_clear_tid = args->child_tidptr;
	else
	   	em->child_clear_tid = NULL;

	EMUL_UNLOCK(&emul_lock);

	if (args->flags & LINUX_CLONE_PARENT_SETTID) {
		error = copyout(&p2->p_pid, args->parent_tidptr,
//.........这里部分代码省略.........

void
ia32_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
{
	struct ia32_sigframe sf, *sfp;
	struct siginfo32 siginfo;
	struct proc *p;
	struct thread *td;
	struct sigacts *psp;
	char *sp;
	struct trapframe *regs;
	char *xfpusave;
	size_t xfpusave_len;
	int oonstack;
	int sig;

	siginfo_to_siginfo32(&ksi->ksi_info, &siginfo);
	td = curthread;
	p = td->td_proc;
	PROC_LOCK_ASSERT(p, MA_OWNED);
	sig = siginfo.si_signo;
	psp = p->p_sigacts;
#ifdef COMPAT_FREEBSD4
	if (SIGISMEMBER(psp->ps_freebsd4, sig)) {
		freebsd4_ia32_sendsig(catcher, ksi, mask);
		return;
	}
#endif
#ifdef COMPAT_43
	if (SIGISMEMBER(psp->ps_osigset, sig)) {
		ia32_osendsig(catcher, ksi, mask);
		return;
	}
#endif
	mtx_assert(&psp->ps_mtx, MA_OWNED);
	regs = td->td_frame;
	oonstack = sigonstack(regs->tf_rsp);

	if (cpu_max_ext_state_size > sizeof(struct savefpu) && use_xsave) {
		xfpusave_len = cpu_max_ext_state_size - sizeof(struct savefpu);
		xfpusave = __builtin_alloca(xfpusave_len);
	} else {
		xfpusave_len = 0;
		xfpusave = NULL;
	}

	/* Save user context. */
	bzero(&sf, sizeof(sf));
	sf.sf_uc.uc_sigmask = *mask;
	sf.sf_uc.uc_stack.ss_sp = (uintptr_t)td->td_sigstk.ss_sp;
	sf.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size;
	sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
	    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
	sf.sf_uc.uc_mcontext.mc_onstack = (oonstack) ? 1 : 0;
	sf.sf_uc.uc_mcontext.mc_edi = regs->tf_rdi;
	sf.sf_uc.uc_mcontext.mc_esi = regs->tf_rsi;
	sf.sf_uc.uc_mcontext.mc_ebp = regs->tf_rbp;
	sf.sf_uc.uc_mcontext.mc_isp = regs->tf_rsp; /* XXX */
	sf.sf_uc.uc_mcontext.mc_ebx = regs->tf_rbx;
	sf.sf_uc.uc_mcontext.mc_edx = regs->tf_rdx;
	sf.sf_uc.uc_mcontext.mc_ecx = regs->tf_rcx;
	sf.sf_uc.uc_mcontext.mc_eax = regs->tf_rax;
	sf.sf_uc.uc_mcontext.mc_trapno = regs->tf_trapno;
	sf.sf_uc.uc_mcontext.mc_err = regs->tf_err;
	sf.sf_uc.uc_mcontext.mc_eip = regs->tf_rip;
	sf.sf_uc.uc_mcontext.mc_cs = regs->tf_cs;
	sf.sf_uc.uc_mcontext.mc_eflags = regs->tf_rflags;
	sf.sf_uc.uc_mcontext.mc_esp = regs->tf_rsp;
	sf.sf_uc.uc_mcontext.mc_ss = regs->tf_ss;
	sf.sf_uc.uc_mcontext.mc_ds = regs->tf_ds;
	sf.sf_uc.uc_mcontext.mc_es = regs->tf_es;
	sf.sf_uc.uc_mcontext.mc_fs = regs->tf_fs;
	sf.sf_uc.uc_mcontext.mc_gs = regs->tf_gs;
	sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext); /* magic */
	ia32_get_fpcontext(td, &sf.sf_uc.uc_mcontext, xfpusave, xfpusave_len);
	fpstate_drop(td);
	sf.sf_uc.uc_mcontext.mc_fsbase = td->td_pcb->pcb_fsbase;
	sf.sf_uc.uc_mcontext.mc_gsbase = td->td_pcb->pcb_gsbase;
	bzero(sf.sf_uc.__spare__, sizeof(sf.sf_uc.__spare__));

	/* Allocate space for the signal handler context. */
	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
	    SIGISMEMBER(psp->ps_sigonstack, sig))
		sp = td->td_sigstk.ss_sp + td->td_sigstk.ss_size;
	else
		sp = (char *)regs->tf_rsp;
	if (xfpusave != NULL) {
		sp -= xfpusave_len;
		sp = (char *)((unsigned long)sp & ~0x3Ful);
		sf.sf_uc.uc_mcontext.mc_xfpustate = (register_t)sp;
	}
	sp -= sizeof(sf);
	/* Align to 16 bytes. */
	sfp = (struct ia32_sigframe *)((uintptr_t)sp & ~0xF);
	PROC_UNLOCK(p);

	/* Translate the signal if appropriate. */
	if (p->p_sysent->sv_sigtbl && sig <= p->p_sysent->sv_sigsize)
		sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];

	/* Build the argument list for the signal handler. */
//.........这里部分代码省略.........

static void
freebsd4_ia32_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
{
	struct ia32_sigframe4 sf, *sfp;
	struct siginfo32 siginfo;
	struct proc *p;
	struct thread *td;
	struct sigacts *psp;
	struct trapframe *regs;
	int oonstack;
	int sig;

	td = curthread;
	p = td->td_proc;
	siginfo_to_siginfo32(&ksi->ksi_info, &siginfo);

	PROC_LOCK_ASSERT(p, MA_OWNED);
	sig = siginfo.si_signo;
	psp = p->p_sigacts;
	mtx_assert(&psp->ps_mtx, MA_OWNED);
	regs = td->td_frame;
	oonstack = sigonstack(regs->tf_rsp);

	/* Save user context. */
	bzero(&sf, sizeof(sf));
	sf.sf_uc.uc_sigmask = *mask;
	sf.sf_uc.uc_stack.ss_sp = (uintptr_t)td->td_sigstk.ss_sp;
	sf.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size;
	sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
	    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
	sf.sf_uc.uc_mcontext.mc_onstack = (oonstack) ? 1 : 0;
	sf.sf_uc.uc_mcontext.mc_edi = regs->tf_rdi;
	sf.sf_uc.uc_mcontext.mc_esi = regs->tf_rsi;
	sf.sf_uc.uc_mcontext.mc_ebp = regs->tf_rbp;
	sf.sf_uc.uc_mcontext.mc_isp = regs->tf_rsp; /* XXX */
	sf.sf_uc.uc_mcontext.mc_ebx = regs->tf_rbx;
	sf.sf_uc.uc_mcontext.mc_edx = regs->tf_rdx;
	sf.sf_uc.uc_mcontext.mc_ecx = regs->tf_rcx;
	sf.sf_uc.uc_mcontext.mc_eax = regs->tf_rax;
	sf.sf_uc.uc_mcontext.mc_trapno = regs->tf_trapno;
	sf.sf_uc.uc_mcontext.mc_err = regs->tf_err;
	sf.sf_uc.uc_mcontext.mc_eip = regs->tf_rip;
	sf.sf_uc.uc_mcontext.mc_cs = regs->tf_cs;
	sf.sf_uc.uc_mcontext.mc_eflags = regs->tf_rflags;
	sf.sf_uc.uc_mcontext.mc_esp = regs->tf_rsp;
	sf.sf_uc.uc_mcontext.mc_ss = regs->tf_ss;
	sf.sf_uc.uc_mcontext.mc_ds = regs->tf_ds;
	sf.sf_uc.uc_mcontext.mc_es = regs->tf_es;
	sf.sf_uc.uc_mcontext.mc_fs = regs->tf_fs;
	sf.sf_uc.uc_mcontext.mc_gs = regs->tf_gs;
	bzero(sf.sf_uc.uc_mcontext.mc_fpregs,
	    sizeof(sf.sf_uc.uc_mcontext.mc_fpregs));
	bzero(sf.sf_uc.uc_mcontext.__spare__,
	    sizeof(sf.sf_uc.uc_mcontext.__spare__));
	bzero(sf.sf_uc.__spare__, sizeof(sf.sf_uc.__spare__));

	/* Allocate space for the signal handler context. */
	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
		sfp = (struct ia32_sigframe4 *)(td->td_sigstk.ss_sp +
		    td->td_sigstk.ss_size - sizeof(sf));
	} else
		sfp = (struct ia32_sigframe4 *)regs->tf_rsp - 1;
	PROC_UNLOCK(p);

	/* Translate the signal if appropriate. */
	if (p->p_sysent->sv_sigtbl && sig <= p->p_sysent->sv_sigsize)
		sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];

	/* Build the argument list for the signal handler. */
	sf.sf_signum = sig;
	sf.sf_ucontext = (register_t)&sfp->sf_uc;
	bzero(&sf.sf_si, sizeof(sf.sf_si));
	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
		/* Signal handler installed with SA_SIGINFO. */
		sf.sf_siginfo = (u_int32_t)(uintptr_t)&sfp->sf_si;
		sf.sf_ah = (u_int32_t)(uintptr_t)catcher;

		/* Fill in POSIX parts */
		sf.sf_si = siginfo;
		sf.sf_si.si_signo = sig;
	} else {
		/* Old FreeBSD-style arguments. */
		sf.sf_siginfo = siginfo.si_code;
		sf.sf_addr = (u_int32_t)siginfo.si_addr;
		sf.sf_ah = (u_int32_t)(uintptr_t)catcher;
	}
	mtx_unlock(&psp->ps_mtx);

	/*
	 * Copy the sigframe out to the user's stack.
	 */
	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
#ifdef DEBUG
		printf("process %ld has trashed its stack\n", (long)p->p_pid);
#endif
		PROC_LOCK(p);
		sigexit(td, SIGILL);
	}

//.........这里部分代码省略.........

int
thread_create(struct thread *td, struct rtprio *rtp,
    int (*initialize_thread)(struct thread *, void *), void *thunk)
{
	struct thread *newtd;
	struct proc *p;
	int error;

	p = td->td_proc;

	if (rtp != NULL) {
		switch(rtp->type) {
		case RTP_PRIO_REALTIME:
		case RTP_PRIO_FIFO:
			/* Only root can set scheduler policy */
			if (priv_check(td, PRIV_SCHED_SETPOLICY) != 0)
				return (EPERM);
			if (rtp->prio > RTP_PRIO_MAX)
				return (EINVAL);
			break;
		case RTP_PRIO_NORMAL:
			rtp->prio = 0;
			break;
		default:
			return (EINVAL);
		}
	}

#ifdef RACCT
	if (racct_enable) {
		PROC_LOCK(p);
		error = racct_add(p, RACCT_NTHR, 1);
		PROC_UNLOCK(p);
		if (error != 0)
			return (EPROCLIM);
	}
#endif

	/* Initialize our td */
	error = kern_thr_alloc(p, 0, &newtd);
	if (error)
		goto fail;

	cpu_set_upcall(newtd, td);

	bzero(&newtd->td_startzero,
	    __rangeof(struct thread, td_startzero, td_endzero));
	bcopy(&td->td_startcopy, &newtd->td_startcopy,
	    __rangeof(struct thread, td_startcopy, td_endcopy));
	newtd->td_proc = td->td_proc;
	thread_cow_get(newtd, td);

	error = initialize_thread(newtd, thunk);
	if (error != 0) {
		thread_cow_free(newtd);
		thread_free(newtd);
		goto fail;
	}

	PROC_LOCK(p);
	p->p_flag |= P_HADTHREADS;
	thread_link(newtd, p);
	bcopy(p->p_comm, newtd->td_name, sizeof(newtd->td_name));
	newtd->td_pax = p->p_pax;
	thread_lock(td);
	/* let the scheduler know about these things. */
	sched_fork_thread(td, newtd);
	thread_unlock(td);
	if (P_SHOULDSTOP(p))
		newtd->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
	if (p->p_flag2 & P2_LWP_EVENTS)
		newtd->td_dbgflags |= TDB_BORN;

	/*
	 * Copy the existing thread VM policy into the new thread.
	 */
	vm_domain_policy_localcopy(&newtd->td_vm_dom_policy,
	    &td->td_vm_dom_policy);

	PROC_UNLOCK(p);

	tidhash_add(newtd);

	thread_lock(newtd);
	if (rtp != NULL) {
		if (!(td->td_pri_class == PRI_TIMESHARE &&
		      rtp->type == RTP_PRIO_NORMAL)) {
			rtp_to_pri(rtp, newtd);
			sched_prio(newtd, newtd->td_user_pri);
		} /* ignore timesharing class */
	}
	TD_SET_CAN_RUN(newtd);
	sched_add(newtd, SRQ_BORING);
	thread_unlock(newtd);

	return (0);

fail:
#ifdef RACCT
	if (racct_enable) {
//.........这里部分代码省略.........

int
cheriabi_sysarch(struct thread *td, struct cheriabi_sysarch_args *uap)
{
	struct trapframe *regs = &td->td_pcb->pcb_regs;
	int error;
	int parms_from_cap = 1;
	size_t reqsize;
	register_t reqperms;

	/*
	 * The sysarch() fill_uap function is machine-independent so can not
	 * check the validity of the capabilty which becomes uap->parms.  As
	 * such, it makes no attempt to convert the result.  We need to
	 * perform those checks here.
	 */
	switch (uap->op) {
	case MIPS_SET_TLS:
		reqsize = 0;
		reqperms = 0;
		break;

	case MIPS_GET_TLS:
	case CHERI_GET_STACK:
	case CHERI_GET_TYPECAP:
		reqsize = sizeof(struct chericap);
		reqperms = CHERI_PERM_STORE|CHERI_PERM_STORE_CAP;
		break;

	case CHERI_SET_STACK:
		reqsize = sizeof(struct chericap);
		reqperms = CHERI_PERM_LOAD|CHERI_PERM_LOAD_CAP;
		break;

	case CHERI_MMAP_GETBASE:
	case CHERI_MMAP_GETLEN:
	case CHERI_MMAP_GETOFFSET:
	case CHERI_MMAP_GETPERM:
	case CHERI_MMAP_SETOFFSET:
	case CHERI_MMAP_SETBOUNDS:
		reqsize = sizeof(uint64_t);
		reqperms = CHERI_PERM_STORE;
		break;

	case CHERI_MMAP_ANDPERM:
		reqsize = sizeof(uint64_t);
		reqperms = CHERI_PERM_LOAD|CHERI_PERM_STORE;
		break;

	case MIPS_GET_COUNT:
		parms_from_cap = 0;
		break;

#ifdef CPU_QEMU_MALTA
	case QEMU_GET_QTRACE:
		reqsize = sizeof(int);
		reqperms = CHERI_PERM_STORE;
		break;

	case QEMU_SET_QTRACE:
		reqsize = sizeof(int);
		reqperms = CHERI_PERM_LOAD;
		break;
#endif

	default:
		return (EINVAL);
	}
	if (parms_from_cap) {
		error = cheriabi_cap_to_ptr(&uap->parms, &regs->c3,
		    reqsize, reqperms, 0);
		if (error != 0)
			return (error);
	}

	switch (uap->op) {
	case MIPS_SET_TLS:
		return (cheriabi_set_user_tls(td, &regs->c3));

	case MIPS_GET_TLS:
		error = copyoutcap(&td->td_md.md_tls_cap, uap->parms,
		    sizeof(struct chericap));
		return (error);

	case CHERI_MMAP_GETBASE: {
		size_t base;

		PROC_LOCK(td->td_proc);
		CHERI_CLC(CHERI_CR_CTEMP0, CHERI_CR_KDC,
		    &td->td_proc->p_md.md_cheri_mmap_cap, 0);
		CHERI_CGETBASE(base, CHERI_CR_CTEMP0);
		PROC_UNLOCK(td->td_proc);
		if (suword64(uap->parms, base) != 0)
			return (EFAULT);
		return (0);
	}

	case CHERI_MMAP_GETLEN: {
		size_t len;

		PROC_LOCK(td->td_proc);
//.........这里部分代码省略.........

void
sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
{
	struct thread *td;
	struct proc *p;
	struct trapframe *tf;
	struct sigframe *fp, frame;
	struct sigacts *psp;
	int code, onstack, sig;

	td = curthread;
	p = td->td_proc;
	PROC_LOCK_ASSERT(p, MA_OWNED);

	sig = ksi->ksi_signo;
	code = ksi->ksi_code;
	psp = p->p_sigacts;
	mtx_assert(&psp->ps_mtx, MA_OWNED);

	tf = td->td_frame;
	onstack = sigonstack(tf->tf_sp);

	CTR4(KTR_SIG, "sendsig: td=%p (%s) catcher=%p sig=%d", td, p->p_comm,
	    catcher, sig);

	/* Allocate and validate space for the signal handler context. */
	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !onstack &&
	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
		fp = (struct sigframe *)(td->td_sigstk.ss_sp +
		    td->td_sigstk.ss_size);
#if defined(COMPAT_43)
		td->td_sigstk.ss_flags |= SS_ONSTACK;
#endif
	} else {
		fp = (struct sigframe *)td->td_frame->tf_sp;
	}

	/* Make room, keeping the stack aligned */
	fp--;
	fp = (struct sigframe *)STACKALIGN(fp);

	/* Fill in the frame to copy out */
	get_mcontext(td, &frame.sf_uc.uc_mcontext, 0);
	get_fpcontext(td, &frame.sf_uc.uc_mcontext);
	frame.sf_si = ksi->ksi_info;
	frame.sf_uc.uc_sigmask = *mask;
	frame.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) ?
	    ((onstack) ? SS_ONSTACK : 0) : SS_DISABLE;
	frame.sf_uc.uc_stack = td->td_sigstk;
	mtx_unlock(&psp->ps_mtx);
	PROC_UNLOCK(td->td_proc);

	/* Copy the sigframe out to the user's stack. */
	if (copyout(&frame, fp, sizeof(*fp)) != 0) {
		/* Process has trashed its stack. Kill it. */
		CTR2(KTR_SIG, "sendsig: sigexit td=%p fp=%p", td, fp);
		PROC_LOCK(p);
		sigexit(td, SIGILL);
	}

	tf->tf_x[0]= sig;
	tf->tf_x[1] = (register_t)&fp->sf_si;
	tf->tf_x[2] = (register_t)&fp->sf_uc;

	tf->tf_elr = (register_t)catcher;
	tf->tf_sp = (register_t)fp;
	tf->tf_lr = (register_t)(PS_STRINGS - *(p->p_sysent->sv_szsigcode));

	CTR3(KTR_SIG, "sendsig: return td=%p pc=%#x sp=%#x", td, tf->tf_elr,
	    tf->tf_sp);

	PROC_LOCK(p);
	mtx_lock(&psp->ps_mtx);
}

static int
linux_clone_proc(struct thread *td, struct linux_clone_args *args)
{
	struct fork_req fr;
	int error, ff = RFPROC | RFSTOPPED;
	struct proc *p2;
	struct thread *td2;
	int exit_signal;
	struct linux_emuldata *em;

#ifdef DEBUG
	if (ldebug(clone)) {
		printf(ARGS(clone, "flags %x, stack %p, parent tid: %p, "
		    "child tid: %p"), (unsigned)args->flags,
		    args->stack, args->parent_tidptr, args->child_tidptr);
	}
#endif

	exit_signal = args->flags & 0x000000ff;
	if (LINUX_SIG_VALID(exit_signal)) {
		exit_signal = linux_to_bsd_signal(exit_signal);
	} else if (exit_signal != 0)
		return (EINVAL);

	if (args->flags & LINUX_CLONE_VM)
		ff |= RFMEM;
	if (args->flags & LINUX_CLONE_SIGHAND)
		ff |= RFSIGSHARE;
	/*
	 * XXX: In Linux, sharing of fs info (chroot/cwd/umask)
	 * and open files is independent.  In FreeBSD, its in one
	 * structure but in reality it does not cause any problems
	 * because both of these flags are usually set together.
	 */
	if (!(args->flags & (LINUX_CLONE_FILES | LINUX_CLONE_FS)))
		ff |= RFFDG;

	if (args->flags & LINUX_CLONE_PARENT_SETTID)
		if (args->parent_tidptr == NULL)
			return (EINVAL);

	if (args->flags & LINUX_CLONE_VFORK)
		ff |= RFPPWAIT;

	bzero(&fr, sizeof(fr));
	fr.fr_flags = ff;
	fr.fr_procp = &p2;
	error = fork1(td, &fr);
	if (error)
		return (error);

	td2 = FIRST_THREAD_IN_PROC(p2);

	/* create the emuldata */
	linux_proc_init(td, td2, args->flags);

	em = em_find(td2);
	KASSERT(em != NULL, ("clone_proc: emuldata not found.\n"));

	if (args->flags & LINUX_CLONE_CHILD_SETTID)
		em->child_set_tid = args->child_tidptr;
	else
	   	em->child_set_tid = NULL;

	if (args->flags & LINUX_CLONE_CHILD_CLEARTID)
		em->child_clear_tid = args->child_tidptr;
	else
	   	em->child_clear_tid = NULL;

	if (args->flags & LINUX_CLONE_PARENT_SETTID) {
		error = copyout(&p2->p_pid, args->parent_tidptr,
		    sizeof(p2->p_pid));
		if (error)
			printf(LMSG("copyout failed!"));
	}

	PROC_LOCK(p2);
	p2->p_sigparent = exit_signal;
	PROC_UNLOCK(p2);
	/*
	 * In a case of stack = NULL, we are supposed to COW calling process
	 * stack. This is what normal fork() does, so we just keep tf_rsp arg
	 * intact.
	 */
	linux_set_upcall_kse(td2, PTROUT(args->stack));

	if (args->flags & LINUX_CLONE_SETTLS)
		linux_set_cloned_tls(td2, args->tls);

	/*
	 * If CLONE_PARENT is set, then the parent of the new process will be 
	 * the same as that of the calling process.
	 */
	if (args->flags & LINUX_CLONE_PARENT) {
		sx_xlock(&