static int
linux_proc_read_fpxregs(struct thread *td, struct linux_pt_fpxreg *fpxregs)
{

	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
	if (cpu_fxsr == 0 || (td->td_proc->p_flag & P_INMEM) == 0)
		return (EIO);
	bcopy(&get_pcb_user_save_td(td)->sv_xmm, fpxregs, sizeof(*fpxregs));
	return (0);
}

static void
clear_orphan(struct proc *p)
{

	PROC_LOCK_ASSERT(p, MA_OWNED);

	if (p->p_flag & P_ORPHAN) {
		LIST_REMOVE(p, p_orphan);
		p->p_flag &= ~P_ORPHAN;
	}
}

/* Attach the filemon to the process. */
static int
filemon_attach_proc(struct filemon *filemon, struct proc *p)
{
	struct filemon *filemon2;

	sx_assert(&filemon->lock, SA_XLOCKED);
	PROC_LOCK_ASSERT(p, MA_OWNED);
	KASSERT((p->p_flag & P_WEXIT) == 0,
	    ("%s: filemon %p attaching to exiting process %p",
	    __func__, filemon, p));
	KASSERT((p->p_flag & P_INEXEC) == 0,
	    ("%s: filemon %p attaching to execing process %p",
	    __func__, filemon, p));

	if (p->p_filemon == filemon)
		return (0);
	/*
	 * Don't allow truncating other process traces.  It is
	 * not really intended to trace procs other than curproc
	 * anyhow.
	 */
	if (p->p_filemon != NULL && p != curproc)
		return (EBUSY);
	/*
	 * Historic behavior of filemon has been to let a child initiate
	 * tracing on itself and cease existing tracing.  Bmake
	 * .META + .MAKE relies on this.  It is only relevant for attaching to
	 * curproc.
	 */
	while (p->p_filemon != NULL) {
		PROC_UNLOCK(p);
		sx_xunlock(&filemon->lock);
		while ((filemon2 = filemon_proc_get(p)) != NULL) {
			/* It may have changed. */
			if (p->p_filemon == filemon2)
				filemon_proc_drop(p);
			filemon_drop(filemon2);
		}
		sx_xlock(&filemon->lock);
		PROC_LOCK(p);
		/*
		 * It may have been attached to, though unlikely.
		 * Try again if needed.
		 */
	}

	KASSERT(p->p_filemon == NULL,
	    ("%s: proc %p didn't detach filemon %p", __func__, p,
	    p->p_filemon));
	p->p_filemon = filemon_acquire(filemon);
	++filemon->proccnt;

	return (0);
}

/*
 * Clean up on last close
 */
int
procfs_close(PFS_CLOSE_ARGS)
{
	if (p != NULL && (p->p_pfsflags & PF_LINGER) == 0) {
		PROC_LOCK_ASSERT(p, MA_OWNED);
		p->p_pfsflags = 0;
		p->p_stops = 0;
		p->p_step = 0;
		wakeup(&p->p_step);
	}
	return (0);
}

int
mac_proc_check_wait(struct ucred *cred, struct proc *p)
{
	int error;

	PROC_LOCK_ASSERT(p, MA_OWNED);

	MAC_POLICY_CHECK_NOSLEEP(proc_check_wait, cred, p);
	MAC_CHECK_PROBE2(proc_check_wait, error, cred, p);

	return (error);
}

int
mac_proc_check_signal(struct ucred *cred, struct proc *p, int signum)
{
	int error;

	PROC_LOCK_ASSERT(p, MA_OWNED);

	MAC_POLICY_CHECK_NOSLEEP(proc_check_signal, cred, p, signum);
	MAC_CHECK_PROBE3(proc_check_signal, error, cred, p, signum);

	return (error);
}

/*
 * Stop a process because of a debugging event;
 * stay stopped until p->p_step is cleared
 * (cleared by PIOCCONT in procfs).
 */
void
stopevent(struct proc *p, unsigned int event, unsigned int val)
{

	PROC_LOCK_ASSERT(p, MA_OWNED | MA_NOTRECURSED);
	p->p_step = 1;

	do {
		p->p_xstat = val;
		p->p_stype = event;	/* Which event caused the stop? */
		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
	} while (p->p_step);
}

/*
 * Increase allocation of 'resource' by 'amount' for process 'p'.
 * Doesn't check for limits and never fails.
 */
void
racct_add_force(struct proc *p, int resource, uint64_t amount)
{

	SDT_PROBE(racct, kernel, rusage, add_force, p, resource, amount, 0, 0);

	/*
	 * We need proc lock to dereference p->p_ucred.
	 */
	PROC_LOCK_ASSERT(p, MA_OWNED);

	mtx_lock(&racct_lock);
	racct_alloc_resource(p->p_racct, resource, amount);
	mtx_unlock(&racct_lock);
	racct_add_cred(p->p_ucred, resource, amount);
}

void
racct_set_force(struct proc *p, int resource, uint64_t amount)
{
	int64_t diff;

	SDT_PROBE(racct, kernel, rusage, set, p, resource, amount, 0, 0);

	/*
	 * We need proc lock to dereference p->p_ucred.
	 */
	PROC_LOCK_ASSERT(p, MA_OWNED);

	mtx_lock(&racct_lock);
	diff = amount - p->p_racct->r_resources[resource];
	racct_alloc_resource(p->p_racct, resource, diff);
	if (diff > 0)
		racct_add_cred_locked(p->p_ucred, resource, diff);
	else if (diff < 0)
		racct_sub_cred_locked(p->p_ucred, resource, -diff);
	mtx_unlock(&racct_lock);
}

/*
 * Decrease allocation of 'resource' by 'amount' for process 'p'.
 */
void
racct_sub(struct proc *p, int resource, uint64_t amount)
{

	SDT_PROBE(racct, kernel, rusage, sub, p, resource, amount, 0, 0);

	/*
	 * We need proc lock to dereference p->p_ucred.
	 */
	PROC_LOCK_ASSERT(p, MA_OWNED);
	KASSERT(RACCT_IS_RECLAIMABLE(resource),
	    ("racct_sub: called for non-reclaimable resource %d", resource));

	mtx_lock(&racct_lock);
	KASSERT(amount <= p->p_racct->r_resources[resource],
	    ("racct_sub: freeing %ju of resource %d, which is more "
	     "than allocated %jd for %s (pid %d)", amount, resource,
	    (intmax_t)p->p_racct->r_resources[resource], p->p_comm, p->p_pid));

	racct_alloc_resource(p->p_racct, resource, -amount);
	racct_sub_cred_locked(p->p_ucred, resource, amount);
	mtx_unlock(&racct_lock);
}

/*
 * Adjust mode for some nodes that need it
 */
int
procfs_attr(PFS_ATTR_ARGS)
{

	/* XXX inefficient, split into separate functions */
	if (strcmp(pn->pn_name, "note") == 0 ||
	    strcmp(pn->pn_name, "notepg") == 0)
		vap->va_mode = 0200;
	else if (strcmp(pn->pn_name, "mem") == 0 ||
	    strcmp(pn->pn_name, "regs") == 0 ||
	    strcmp(pn->pn_name, "dbregs") == 0 ||
	    strcmp(pn->pn_name, "fpregs") == 0 ||
	    strcmp(pn->pn_name, "osrel") == 0)
		vap->va_mode = 0600;

	if (p != NULL) {
		PROC_LOCK_ASSERT(p, MA_OWNED);

		if ((p->p_flag & P_SUGID) && pn->pn_type != pfstype_procdir)
			vap->va_mode = 0;
	}

	return (0);
}

/*
 * 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.........

/*
 * copied from amd64/amd64/machdep.c
 *
 * Send an interrupt to process.
 */
static void
linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
{
	struct l_rt_sigframe sf, *sfp;
	struct proc *p;
	struct thread *td;
	struct sigacts *psp;
	caddr_t sp;
	struct trapframe *regs;
	int sig, code;
	int oonstack;

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

	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
	    catcher, sig, mask, code);

	/* Allocate space for the signal handler context. */
	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
		sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
		    sizeof(struct l_rt_sigframe);
	} else
		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
	/* Align to 16 bytes. */
	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
	mtx_unlock(&psp->ps_mtx);

	/* Translate the signal. */
	sig = bsd_to_linux_signal(sig);

	/* Save user context. */
	bzero(&sf, sizeof(sf));
	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);

	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
	PROC_UNLOCK(p);

	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;

	/* Build the argument list for the signal handler. */
	regs->tf_rdi = sig;			/* arg 1 in %rdi */
	regs->tf_rax = 0;
	regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
	regs->tf_rdx = (register_t)&sfp->sf_sc;	/* arg 3 in %rdx */

	sf.sf_handler = catcher;
	/* Fill in POSIX parts */
	ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);

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

	regs->tf_rsp = (long)sfp;
//.........这里部分代码省略.........

static void
cheriabi_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
{
	struct proc *p;
	struct thread *td;
	struct trapframe *regs;
	struct cheri_frame *capreg;
	struct sigacts *psp;
	struct sigframe_c sf, *sfp;
	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;
	capreg = &td->td_pcb->pcb_cheriframe;
	oonstack = sigonstack(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 sandbox 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.
	 */
	if (cheri_is_sandboxed != 0)
		oonstack = 0;

	/* save user context */
	bzero(&sf, sizeof(struct sigframe));
	sf.sf_uc.uc_sigmask = *mask;
#if 0
	/*
	 * XXX-BD: stack_t type differs and we can't just fake a capabilty.
	 * We don't restore the value so what purpose does it serve?
	 */
	sf.sf_uc.uc_stack = td->td_sigstk;
#endif
	sf.sf_uc.uc_mcontext.mc_onstack = (oonstack) ? 1 : 0;
	sf.sf_uc.uc_mcontext.mc_pc = regs->pc;
	sf.sf_uc.uc_mcontext.mullo = regs->mullo;
	sf.sf_uc.uc_mcontext.mulhi = regs->mulhi;
#if 0
	/* XXX-BD: what actually makes sense here? */
	sf.sf_uc.uc_mcontext.mc_tls = td->td_md.md_tls;
#endif
	sf.sf_uc.uc_mcontext.mc_regs[0] = UCONTEXT_MAGIC;  /* magic number */
	bcopy((void *)&regs->ast, (void *)&sf.sf_uc.uc_mcontext.mc_regs[1],
	    sizeof(sf.sf_uc.uc_mcontext.mc_regs) - sizeof(register_t));
	sf.sf_uc.uc_mcontext.mc_fpused = td->td_md.md_flags & MDTD_FPUSED;
	if (sf.sf_uc.uc_mcontext.mc_fpused) {
		/* if FPU has current state, save it first */
		if (td == PCPU_GET(fpcurthread))
//.........这里部分代码省略.........

/*
 * Send an interrupt to process.
 *
 * Stack is set up to allow sigcode stored
 * at top to call routine, followed by kcall
 * to sigreturn routine below.	After sigreturn
 * resets the signal mask, the stack, and the
 * frame pointer, it returns to the user
 * specified pc, psl.
 */
static void
freebsd32_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
{
	struct proc *p;
	struct thread *td;
	struct fpreg32 fpregs;
	struct reg32 regs;
	struct sigacts *psp;
	struct sigframe32 sf, *sfp;
	int sig;
	int oonstack;
	unsigned i;

	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);

	fill_regs32(td, &regs);
	oonstack = sigonstack(td->td_frame->sp);

	/* save user context */
	bzero(&sf, sizeof sf);
	sf.sf_uc.uc_sigmask = *mask;
	sf.sf_uc.uc_stack.ss_sp = (int32_t)(intptr_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_sigstk.ss_flags;
	sf.sf_uc.uc_mcontext.mc_onstack = (oonstack) ? 1 : 0;
	sf.sf_uc.uc_mcontext.mc_pc = regs.r_regs[PC];
	sf.sf_uc.uc_mcontext.mullo = regs.r_regs[MULLO];
	sf.sf_uc.uc_mcontext.mulhi = regs.r_regs[MULHI];
	sf.sf_uc.uc_mcontext.mc_tls = (int32_t)(intptr_t)td->td_md.md_tls;
	sf.sf_uc.uc_mcontext.mc_regs[0] = UCONTEXT_MAGIC;  /* magic number */
	for (i = 1; i < 32; i++)
		sf.sf_uc.uc_mcontext.mc_regs[i] = regs.r_regs[i];
	sf.sf_uc.uc_mcontext.mc_fpused = td->td_md.md_flags & MDTD_FPUSED;
	if (sf.sf_uc.uc_mcontext.mc_fpused) {
		/* if FPU has current state, save it first */
		if (td == PCPU_GET(fpcurthread))
			MipsSaveCurFPState(td);
		fill_fpregs32(td, &fpregs);
		for (i = 0; i < 33; i++)
			sf.sf_uc.uc_mcontext.mc_fpregs[i] = fpregs.r_regs[i];
	}

	/* Allocate and validate space for the signal handler context. */
	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
		sfp = (struct sigframe32 *)(((uintptr_t)td->td_sigstk.ss_sp +
		    td->td_sigstk.ss_size - sizeof(struct sigframe32))
		    & ~(sizeof(__int64_t) - 1));
	} else
		sfp = (struct sigframe32 *)((vm_offset_t)(td->td_frame->sp - 
		    sizeof(struct sigframe32)) & ~(sizeof(__int64_t) - 1));

	/* Build the argument list for the signal handler. */
	td->td_frame->a0 = sig;
	td->td_frame->a2 = (register_t)(intptr_t)&sfp->sf_uc;
	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
		/* Signal handler installed with SA_SIGINFO. */
		td->td_frame->a1 = (register_t)(intptr_t)&sfp->sf_si;
		/* sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher; */

		/* fill siginfo structure */
		sf.sf_si.si_signo = sig;
		sf.sf_si.si_code = ksi->ksi_code;
		sf.sf_si.si_addr = td->td_frame->badvaddr;
	} else {
		/* Old FreeBSD-style arguments. */
		td->td_frame->a1 = ksi->ksi_code;
		td->td_frame->a3 = td->td_frame->badvaddr;
		/* sf.sf_ahu.sf_handler = catcher; */
	}

	mtx_unlock(&psp->ps_mtx);
	PROC_UNLOCK(p);

	/*
	 * Copy the sigframe out to the user's stack.
	 */
	if (copyout(&sf, sfp, sizeof(struct sigframe32)) != 0) {
		/*
		 * Something is wrong with the stack pointer.
		 * ...Kill the process.
		 */
		PROC_LOCK(p);
		sigexit(td, SIGILL);
	}
//.........这里部分代码省略.........

/*
 * Process ioctls
 */
int
procfs_ioctl(PFS_IOCTL_ARGS)
{
	struct procfs_status *ps;
#ifdef COMPAT_FREEBSD32
	struct procfs_status32 *ps32;
#endif
	int error, flags, sig;
#ifdef COMPAT_FREEBSD6
	int ival;
#endif

	KASSERT(p != NULL,
	    ("%s() called without a process", __func__));
	PROC_LOCK_ASSERT(p, MA_OWNED);

	error = 0;
	switch (cmd) {
#if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
	case _IOC(IOC_IN, 'p', 1, 0):
#endif
#ifdef COMPAT_FREEBSD6
	case _IO('p', 1):
		ival = IOCPARM_IVAL(data);
		data = &ival;
#endif
	case PIOCBIS:
		p->p_stops |= *(unsigned int *)data;
		break;
#if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
	case _IOC(IOC_IN, 'p', 2, 0):
#endif
#ifdef COMPAT_FREEBSD6
	case _IO('p', 2):
		ival = IOCPARM_IVAL(data);
		data = &ival;
#endif
	case PIOCBIC:
		p->p_stops &= ~*(unsigned int *)data;
		break;
#if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
	case _IOC(IOC_IN, 'p', 3, 0):
#endif
#ifdef COMPAT_FREEBSD6
	case _IO('p', 3):
		ival = IOCPARM_IVAL(data);
		data = &ival;
#endif
	case PIOCSFL:
		flags = *(unsigned int *)data;
		if (flags & PF_ISUGID) {
			/*
			 * XXXRW: Is this specific check required here, as
			 * p_candebug() should implement it, or other checks
			 * are missing.
			 */
			error = priv_check(td, PRIV_DEBUG_SUGID);
			if (error)
				break;
		}
		p->p_pfsflags = flags;
		break;
	case PIOCGFL:
		*(unsigned int *)data = p->p_pfsflags;
		break;
	case PIOCWAIT:
		while (p->p_step == 0 && (p->p_flag & P_WEXIT) == 0) {
			/* sleep until p stops */
			_PHOLD(p);
			error = msleep(&p->p_stype, &p->p_mtx,
			    PWAIT|PCATCH, "pioctl", 0);
			_PRELE(p);
			if (error != 0)
				break;
		}
		/* fall through to PIOCSTATUS */
	case PIOCSTATUS:
		ps = (struct procfs_status *)data;
		ps->state = (p->p_step == 0);
		ps->flags = 0; /* nope */
		ps->events = p->p_stops;
		ps->why = p->p_step ? p->p_stype : 0;
		ps->val = p->p_step ? p->p_xstat : 0;
		break;
#ifdef COMPAT_FREEBSD32
	case PIOCWAIT32:
		while (p->p_step == 0 && (p->p_flag & P_WEXIT) == 0) {
			/* sleep until p stops */
			_PHOLD(p);
			error = msleep(&p->p_stype, &p->p_mtx,
			    PWAIT|PCATCH, "pioctl", 0);
			_PRELE(p);
			if (error != 0)
				break;
		}
		/* fall through to PIOCSTATUS32 */
	case PIOCSTATUS32:
//.........这里部分代码省略.........

/*
 * 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     = regs->tf_gs;
	frame.sf_sc.sc_fs     = regs->tf_fs;
	frame.sf_sc.sc_es     = regs->tf_es;
	frame.sf_sc.sc_ds     = regs->tf_ds;
	frame.sf_sc.sc_edi    = regs->tf_rdi;
	frame.sf_sc.sc_esi    = regs->tf_rsi;
	frame.sf_sc.sc_ebp    = regs->tf_rbp;
	frame.sf_sc.sc_ebx    = regs->tf_rbx;
	frame.sf_sc.sc_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.
		 */
//.........这里部分代码省略.........

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 */
	ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);

	/*
	 * Build the signal context to be used by sigreturn.
	 */
	frame.sf_sc.uc_flags = 0;		/* XXX ??? */
	frame.sf_sc.uc_link = 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_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_gs     = regs->tf_gs;
	frame.sf_sc.uc_mcontext.sc_fs     = regs->tf_fs;
	frame.sf_sc.uc_mcontext.sc_es     = regs->tf_es;
	frame.sf_sc.uc_mcontext.sc_ds     = regs->tf_ds;
	frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags;
	frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp;
	frame.sf_sc.uc_mcontext.sc_ss     = regs->tf_ss;
	frame.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
	frame.sf_sc.uc_mcontext.sc_cr2    = (u_int32_t)(uintptr_t)ksi->ksi_addr;
	frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);

#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.
		 */
//.........这里部分代码省略.........

/*
 * Visibility: some files are only visible to process that can debug
 * the target process.
 */
int
procfs_candebug(PFS_VIS_ARGS)
{
	PROC_LOCK_ASSERT(p, MA_OWNED);
	return ((p->p_flag & P_SYSTEM) == 0 && p_candebug(td, p) == 0);
}

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);
}