int
freebsd32_sigreturn(struct thread *td, struct freebsd32_sigreturn_args *uap)
{
ucontext32_t uc;
int error;
CTR2(KTR_SIG, "sigreturn: td=%p ucp=%p", td, uap->sigcntxp);
if (copyin(uap->sigcntxp, &uc, sizeof(uc)) != 0) {
CTR1(KTR_SIG, "sigreturn: efault td=%p", td);
return (EFAULT);
}
error = set_mcontext32(td, &uc.uc_mcontext);
if (error != 0)
return (error);
kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
#if 0
CTR3(KTR_SIG, "sigreturn: return td=%p pc=%#x sp=%#x",
td, uc.uc_mcontext.mc_srr0, uc.uc_mcontext.mc_gpr[1]);
#endif
return (EJUSTRETURN);
}
int
sys_swapcontext(struct thread *td, struct swapcontext_args *uap)
{
ucontext_t uc;
int ret;
if (uap->oucp == NULL || uap->ucp == NULL)
ret = EINVAL;
else {
bzero(&uc, sizeof(ucontext_t));
get_mcontext(td, &uc.uc_mcontext, GET_MC_CLEAR_RET);
PROC_LOCK(td->td_proc);
uc.uc_sigmask = td->td_sigmask;
PROC_UNLOCK(td->td_proc);
ret = copyout(&uc, uap->oucp, UC_COPY_SIZE);
if (ret == 0) {
ret = copyin(uap->ucp, &uc, UC_COPY_SIZE);
if (ret == 0) {
ret = set_mcontext(td, &uc.uc_mcontext);
if (ret == 0) {
kern_sigprocmask(td, SIG_SETMASK,
&uc.uc_sigmask, NULL, 0);
}
}
}
}
return (ret == 0 ? EJUSTRETURN : ret);
}
int
sys_sigreturn(struct thread *td, struct sigreturn_args *uap)
{
uint64_t sstatus;
ucontext_t uc;
int error;
if (uap == NULL)
return (EFAULT);
if (copyin(uap->sigcntxp, &uc, sizeof(uc)))
return (EFAULT);
/*
* Make sure the processor mode has not been tampered with and
* interrupts have not been disabled.
*/
sstatus = uc.uc_mcontext.mc_gpregs.gp_sstatus;
if ((sstatus & SSTATUS_PS) != 0 ||
(sstatus & SSTATUS_PIE) == 0)
return (EINVAL);
error = set_mcontext(td, &uc.uc_mcontext);
if (error != 0)
return (error);
set_fpcontext(td, &uc.uc_mcontext);
/* Restore signal mask. */
kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
return (EJUSTRETURN);
}
static void
fuse_restore_sigs(sigset_t *oldset)
{
if (kern_sigprocmask(curthread, SIG_SETMASK, oldset, NULL, 0))
panic("%s: Invalid operation for kern_sigprocmask()",
__func__);
}
int
ofreebsd32_sigreturn(struct thread *td, struct ofreebsd32_sigreturn_args *uap)
{
struct ia32_sigcontext3 sc, *scp;
struct trapframe *regs;
int eflags, error;
ksiginfo_t ksi;
regs = td->td_frame;
error = copyin(uap->sigcntxp, &sc, sizeof(sc));
if (error != 0)
return (error);
scp = ≻
eflags = scp->sc_eflags;
if (!EFL_SECURE(eflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
return (EINVAL);
}
if (!CS_SECURE(scp->sc_cs)) {
ksiginfo_init_trap(&ksi);
ksi.ksi_signo = SIGBUS;
ksi.ksi_code = BUS_OBJERR;
ksi.ksi_trapno = T_PROTFLT;
ksi.ksi_addr = (void *)regs->tf_rip;
trapsignal(td, &ksi);
return (EINVAL);
}
regs->tf_ds = scp->sc_ds;
regs->tf_es = scp->sc_es;
regs->tf_fs = scp->sc_fs;
regs->tf_gs = scp->sc_gs;
regs->tf_rax = scp->sc_eax;
regs->tf_rbx = scp->sc_ebx;
regs->tf_rcx = scp->sc_ecx;
regs->tf_rdx = scp->sc_edx;
regs->tf_rsi = scp->sc_esi;
regs->tf_rdi = scp->sc_edi;
regs->tf_cs = scp->sc_cs;
regs->tf_ss = scp->sc_ss;
regs->tf_rbp = scp->sc_ebp;
regs->tf_rsp = scp->sc_esp;
regs->tf_rip = scp->sc_eip;
regs->tf_rflags = eflags;
if (scp->sc_onstack & 1)
td->td_sigstk.ss_flags |= SS_ONSTACK;
else
td->td_sigstk.ss_flags &= ~SS_ONSTACK;
kern_sigprocmask(td, SIG_SETMASK, (sigset_t *)&scp->sc_mask, NULL,
SIGPROCMASK_OLD);
set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
return (EJUSTRETURN);
}
static void
fuse_block_sigs(sigset_t *oldset)
{
sigset_t newset;
SIGFILLSET(newset);
SIGDELSET(newset, SIGKILL);
if (kern_sigprocmask(curthread, SIG_BLOCK, &newset, oldset, 0))
panic("%s: Invalid operation for kern_sigprocmask()",
__func__);
}
/*
* 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 as specified by
* context left by sendsig.
*/
int
sys_sigreturn(struct thread *td, struct sigreturn_args *uap)
{
struct trapframe *regs;
ucontext_t *ucp;
ucontext_t uc;
int error;
ucp = &uc;
error = copyin(uap->sigcntxp, &uc, sizeof(uc));
if (error != 0)
return (error);
regs = td->td_frame;
/* #ifdef DEBUG */
if (ucp->uc_mcontext.mc_regs[ZERO] != UCONTEXT_MAGIC) {
printf("sigreturn: pid %d, ucp %p\n", td->td_proc->p_pid, ucp);
printf(" old sp %p ra %p pc %p\n",
(void *)(intptr_t)regs->sp, (void *)(intptr_t)regs->ra, (void *)(intptr_t)regs->pc);
printf(" new sp %p ra %p pc %p z %p\n",
(void *)(intptr_t)ucp->uc_mcontext.mc_regs[SP],
(void *)(intptr_t)ucp->uc_mcontext.mc_regs[RA],
(void *)(intptr_t)ucp->uc_mcontext.mc_regs[PC],
(void *)(intptr_t)ucp->uc_mcontext.mc_regs[ZERO]);
return EINVAL;
}
/* #endif */
bcopy((const void *)&ucp->uc_mcontext.mc_regs[1], (void *)®s->ast,
sizeof(ucp->uc_mcontext.mc_regs) - sizeof(register_t));
if (ucp->uc_mcontext.mc_fpused)
bcopy((const void *)ucp->uc_mcontext.mc_fpregs,
(void *)&td->td_frame->f0,
sizeof(ucp->uc_mcontext.mc_fpregs));
regs->pc = ucp->uc_mcontext.mc_pc;
regs->mullo = ucp->uc_mcontext.mullo;
regs->mulhi = ucp->uc_mcontext.mulhi;
kern_sigprocmask(td, SIG_SETMASK, &ucp->uc_sigmask, NULL, 0);
return(EJUSTRETURN);
}
int
cheriabi_sigreturn(struct thread *td, struct cheriabi_sigreturn_args *uap)
{
ucontext_c_t uc;
int error;
error = copyincap(uap->sigcntxp, &uc, sizeof(uc));
if (error != 0)
return (error);
error = cheriabi_set_mcontext(td, &uc.uc_mcontext);
if (error != 0)
return (error);
kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
return (EJUSTRETURN);
}
int
sys_setcontext(struct thread *td, struct setcontext_args *uap)
{
ucontext_t uc;
int ret;
if (uap->ucp == NULL)
ret = EINVAL;
else {
ret = copyin(uap->ucp, &uc, UC_COPY_SIZE);
if (ret == 0) {
ret = set_mcontext(td, &uc.uc_mcontext);
if (ret == 0) {
kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask,
NULL, 0);
}
}
}
return (ret == 0 ? EJUSTRETURN : ret);
}
int
sys_sigreturn(struct thread *td, struct sigreturn_args *uap)
{
ucontext_t uc;
uint32_t spsr;
if (uap == NULL)
return (EFAULT);
if (copyin(uap->sigcntxp, &uc, sizeof(uc)))
return (EFAULT);
spsr = uc.uc_mcontext.mc_gpregs.gp_spsr;
if ((spsr & PSR_M_MASK) != PSR_M_EL0t ||
(spsr & (PSR_F | PSR_I | PSR_A | PSR_D)) != 0)
return (EINVAL);
set_mcontext(td, &uc.uc_mcontext);
set_fpcontext(td, &uc.uc_mcontext);
/* Restore signal mask. */
kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
return (EJUSTRETURN);
}
/*
* Process an asynchronous software trap.
* This is relatively easy.
* This function will return with preemption disabled.
*/
void
ast(struct trapframe *framep)
{
struct thread *td;
struct proc *p;
int flags;
int sig;
td = curthread;
p = td->td_proc;
CTR3(KTR_SYSC, "ast: thread %p (pid %d, %s)", td, p->p_pid,
p->p_comm);
KASSERT(TRAPF_USERMODE(framep), ("ast in kernel mode"));
WITNESS_WARN(WARN_PANIC, NULL, "Returning to user mode");
mtx_assert(&Giant, MA_NOTOWNED);
THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
td->td_frame = framep;
td->td_pticks = 0;
/*
* This updates the td_flag's for the checks below in one
* "atomic" operation with turning off the astpending flag.
* If another AST is triggered while we are handling the
* AST's saved in flags, the astpending flag will be set and
* ast() will be called again.
*/
thread_lock(td);
flags = td->td_flags;
td->td_flags &= ~(TDF_ASTPENDING | TDF_NEEDSIGCHK | TDF_NEEDSUSPCHK |
TDF_NEEDRESCHED | TDF_ALRMPEND | TDF_PROFPEND | TDF_MACPEND);
thread_unlock(td);
PCPU_INC(cnt.v_trap);
if (td->td_ucred != p->p_ucred)
cred_update_thread(td);
if (td->td_pflags & TDP_OWEUPC && p->p_flag & P_PROFIL) {
addupc_task(td, td->td_profil_addr, td->td_profil_ticks);
td->td_profil_ticks = 0;
td->td_pflags &= ~TDP_OWEUPC;
}
#ifdef HWPMC_HOOKS
/* Handle Software PMC callchain capture. */
if (PMC_IS_PENDING_CALLCHAIN(td))
PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_USER_CALLCHAIN_SOFT, (void *) framep);
#endif
if (flags & TDF_ALRMPEND) {
PROC_LOCK(p);
kern_psignal(p, SIGVTALRM);
PROC_UNLOCK(p);
}
if (flags & TDF_PROFPEND) {
PROC_LOCK(p);
kern_psignal(p, SIGPROF);
PROC_UNLOCK(p);
}
#ifdef MAC
if (flags & TDF_MACPEND)
mac_thread_userret(td);
#endif
if (flags & TDF_NEEDRESCHED) {
#ifdef KTRACE
if (KTRPOINT(td, KTR_CSW))
ktrcsw(1, 1, __func__);
#endif
thread_lock(td);
sched_prio(td, td->td_user_pri);
mi_switch(SW_INVOL | SWT_NEEDRESCHED, NULL);
thread_unlock(td);
#ifdef KTRACE
if (KTRPOINT(td, KTR_CSW))
ktrcsw(0, 1, __func__);
#endif
}
/*
* Check for signals. Unlocked reads of p_pendingcnt or
* p_siglist might cause process-directed signal to be handled
* later.
*/
if (flags & TDF_NEEDSIGCHK || p->p_pendingcnt > 0 ||
!SIGISEMPTY(p->p_siglist)) {
PROC_LOCK(p);
mtx_lock(&p->p_sigacts->ps_mtx);
while ((sig = cursig(td)) != 0)
postsig(sig);
mtx_unlock(&p->p_sigacts->ps_mtx);
PROC_UNLOCK(p);
}
/*
* We need to check to see if we have to exit or wait due to a
* single threading requirement or some other STOP condition.
*/
if (flags & TDF_NEEDSUSPCHK) {
PROC_LOCK(p);
thread_suspend_check(0);
PROC_UNLOCK(p);
}
if (td->td_pflags & TDP_OLDMASK) {
td->td_pflags &= ~TDP_OLDMASK;
kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
}
userret(td, framep);
}
/*
* System call to cleanup state after a signal
* has been taken. Reset signal mask and
* stack state from context left by rt_sendsig (above).
* Return to previous pc and psl as specified by
* context left by sendsig. Check carefully to
* make sure that the user has not modified the
* psl to gain improper privileges or to cause
* a machine fault.
*/
int
linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
{
struct l_ucontext uc;
struct l_sigcontext *context;
sigset_t bmask;
l_stack_t *lss;
stack_t ss;
struct trapframe *regs;
int eflags;
ksiginfo_t ksi;
regs = td->td_frame;
#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;
if (!EFLAGS_SECURE(eflags, regs->tf_rflags))
return(EINVAL);
/*
* Don't allow users to load a valid privileged %cs. Let the
* hardware check for invalid selectors, excess privilege in
* other selectors, invalid %eip's and invalid %esp's.
*/
#define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
if (!CS_SECURE(context->sc_cs)) {
ksiginfo_init_trap(&ksi);
ksi.ksi_signo = SIGBUS;
ksi.ksi_code = BUS_OBJERR;
ksi.ksi_trapno = T_PROTFLT;
ksi.ksi_addr = (void *)regs->tf_rip;
trapsignal(td, &ksi);
return(EINVAL);
}
linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
/*
* Restore signal context
*/
regs->tf_gs = context->sc_gs;
regs->tf_fs = context->sc_fs;
regs->tf_es = context->sc_es;
regs->tf_ds = context->sc_ds;
regs->tf_rdi = context->sc_edi;
regs->tf_rsi = context->sc_esi;
regs->tf_rbp = context->sc_ebp;
regs->tf_rbx = context->sc_ebx;
regs->tf_rdx = context->sc_edx;
regs->tf_rcx = context->sc_ecx;
regs->tf_rax = context->sc_eax;
regs->tf_rip = context->sc_eip;
regs->tf_cs = context->sc_cs;
regs->tf_rflags = eflags;
regs->tf_rsp = context->sc_esp_at_signal;
regs->tf_ss = context->sc_ss;
set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
/*
* call sigaltstack & ignore results..
*/
lss = &uc.uc_stack;
ss.ss_sp = PTRIN(lss->ss_sp);
ss.ss_size = lss->ss_size;
ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
#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);
}
/*
* System call to cleanup state after a signal
* has been taken. Reset signal mask and
* stack state from context left by sendsig (above).
* Return to previous pc and psl as specified by
* context left by sendsig. Check carefully to
* make sure that the user has not modified the
* psl to gain improper privileges or to cause
* a machine fault.
*/
int
linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
{
struct l_sigframe frame;
struct trapframe *regs;
sigset_t bmask;
l_sigset_t lmask;
int eflags, 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;
if (!EFLAGS_SECURE(eflags, regs->tf_rflags))
return(EINVAL);
/*
* Don't allow users to load a valid privileged %cs. Let the
* hardware check for invalid selectors, excess privilege in
* other selectors, invalid %eip's and invalid %esp's.
*/
#define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
if (!CS_SECURE(frame.sf_sc.sc_cs)) {
ksiginfo_init_trap(&ksi);
ksi.ksi_signo = SIGBUS;
ksi.ksi_code = BUS_OBJERR;
ksi.ksi_trapno = T_PROTFLT;
ksi.ksi_addr = (void *)regs->tf_rip;
trapsignal(td, &ksi);
return(EINVAL);
}
lmask.__bits[0] = frame.sf_sc.sc_mask;
for (i = 0; i < (LINUX_NSIG_WORDS-1); i++)
lmask.__bits[i+1] = frame.sf_extramask[i];
linux_to_bsd_sigset(&lmask, &bmask);
kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
/*
* Restore signal context.
*/
regs->tf_rdi = frame.sf_sc.sc_edi;
regs->tf_rsi = frame.sf_sc.sc_esi;
regs->tf_rbp = frame.sf_sc.sc_ebp;
regs->tf_rbx = frame.sf_sc.sc_ebx;
regs->tf_rdx = frame.sf_sc.sc_edx;
regs->tf_rcx = frame.sf_sc.sc_ecx;
regs->tf_rax = frame.sf_sc.sc_eax;
regs->tf_rip = frame.sf_sc.sc_eip;
regs->tf_cs = frame.sf_sc.sc_cs;
regs->tf_ds = frame.sf_sc.sc_ds;
regs->tf_es = frame.sf_sc.sc_es;
regs->tf_fs = frame.sf_sc.sc_fs;
regs->tf_gs = frame.sf_sc.sc_gs;
regs->tf_rflags = eflags;
regs->tf_rsp = frame.sf_sc.sc_esp_at_signal;
regs->tf_ss = frame.sf_sc.sc_ss;
set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
return (EJUSTRETURN);
}
/*
* System call to cleanup state after a signal
* has been taken. Reset signal mask and
* stack state from context left by rt_sendsig (above).
* Return to previous pc and psl as specified by
* context left by sendsig. Check carefully to
* make sure that the user has not modified the
* psl to gain improper privileges or to cause
* a machine fault.
*/
int
linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
{
struct l_ucontext uc;
struct l_sigcontext *context;
sigset_t bmask;
l_stack_t *lss;
stack_t ss;
struct trapframe *regs;
int eflags;
ksiginfo_t ksi;
regs = td->td_frame;
#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);
}
linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
/*
* Restore signal context
*/
/* %gs was restored by the trampoline. */
regs->tf_fs = context->sc_fs;
regs->tf_es = context->sc_es;
regs->tf_ds = context->sc_ds;
regs->tf_edi = context->sc_edi;
regs->tf_esi = context->sc_esi;
regs->tf_ebp = context->sc_ebp;
regs->tf_ebx = context->sc_ebx;
regs->tf_edx = context->sc_edx;
regs->tf_ecx = context->sc_ecx;
regs->tf_eax = context->sc_eax;
regs->tf_eip = context->sc_eip;
regs->tf_cs = context->sc_cs;
regs->tf_eflags = eflags;
regs->tf_esp = context->sc_esp_at_signal;
regs->tf_ss = context->sc_ss;
/*
* call sigaltstack & ignore results..
*/
lss = &uc.uc_stack;
ss.ss_sp = lss->ss_sp;
ss.ss_size = lss->ss_size;
ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
#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);
}
/*
* System call to cleanup state after a signal
* has been taken. Reset signal mask and
* stack state from context left by sendsig (above).
* Return to previous pc and psl as specified by
* context left by sendsig. Check carefully to
* make sure that the user has not modified the
* psl to gain improper privileges or to cause
* a machine fault.
*/
int
linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
{
struct l_sigframe frame;
struct trapframe *regs;
l_sigset_t lmask;
sigset_t bmask;
int eflags, 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];
linux_to_bsd_sigset(&lmask, &bmask);
kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
/*
* Restore signal context.
*/
/* %gs was restored by the trampoline. */
regs->tf_fs = frame.sf_sc.sc_fs;
regs->tf_es = frame.sf_sc.sc_es;
regs->tf_ds = frame.sf_sc.sc_ds;
regs->tf_edi = frame.sf_sc.sc_edi;
regs->tf_esi = frame.sf_sc.sc_esi;
regs->tf_ebp = frame.sf_sc.sc_ebp;
regs->tf_ebx = frame.sf_sc.sc_ebx;
regs->tf_edx = frame.sf_sc.sc_edx;
regs->tf_ecx = frame.sf_sc.sc_ecx;
regs->tf_eax = frame.sf_sc.sc_eax;
regs->tf_eip = frame.sf_sc.sc_eip;
regs->tf_cs = frame.sf_sc.sc_cs;
regs->tf_eflags = eflags;
regs->tf_esp = frame.sf_sc.sc_esp_at_signal;
regs->tf_ss = frame.sf_sc.sc_ss;
return (EJUSTRETURN);
}
static void
pmclog_loop(void *arg)
{
struct pmclog_proc_init_args *ia;
struct pmc_owner *po;
struct pmclog_buffer *lb;
struct proc *p;
struct ucred *ownercred;
struct ucred *mycred;
struct thread *td;
sigset_t unb;
struct uio auio;
struct iovec aiov;
size_t nbytes;
int error;
td = curthread;
SIGEMPTYSET(unb);
SIGADDSET(unb, SIGHUP);
(void)kern_sigprocmask(td, SIG_UNBLOCK, &unb, NULL, 0);
ia = arg;
MPASS(ia->kthr == curproc);
MPASS(!ia->acted);
mtx_lock(&pmc_kthread_mtx);
while (ia->po == NULL && !ia->exit)
msleep(ia, &pmc_kthread_mtx, PWAIT, "pmclogi", 0);
if (ia->exit) {
ia->acted = true;
wakeup(ia);
mtx_unlock(&pmc_kthread_mtx);
kproc_exit(0);
}
MPASS(ia->po != NULL);
po = ia->po;
ia->acted = true;
wakeup(ia);
mtx_unlock(&pmc_kthread_mtx);
ia = NULL;
p = po->po_owner;
mycred = td->td_ucred;
PROC_LOCK(p);
ownercred = crhold(p->p_ucred);
PROC_UNLOCK(p);
PMCDBG2(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)
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 */
PMCDBG3(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);
mtx_lock(&pmc_kthread_mtx);
po->po_error = error; /* save for flush log */
PMCDBG2(LOG,WRI,2, "po=%p error=%d", po, error);
break;
}
mtx_lock(&pmc_kthread_mtx);
/* put the used buffer back into the global pool */
PMCLOG_INIT_BUFFER_DESCRIPTOR(lb);
mtx_lock_spin(&pmc_bufferlist_mtx);
TAILQ_INSERT_HEAD(&pmc_bufferlist, lb, plb_next);
mtx_unlock_spin(&pmc_bufferlist_mtx);
lb = NULL;
}
wakeup_one(po->po_kthread);
po->po_kthread = NULL;
mtx_unlock(&pmc_kthread_mtx);
/* return the current I/O buffer to the global pool */
if (lb) {
PMCLOG_INIT_BUFFER_DESCRIPTOR(lb);
mtx_lock_spin(&pmc_bufferlist_mtx);
TAILQ_INSERT_HEAD(&pmc_bufferlist, lb, plb_next);
mtx_unlock_spin(&pmc_bufferlist_mtx);
}
/*
* Exit this thread, signalling the waiter
*/
crfree(ownercred);
kproc_exit(0);
}
