/*
* スケジューラの呼び出し
*/
StatusType
Schedule(void)
{
/*
* ここでの ercd の初期化は本来は不要であるが，コンパイラの警
* 告メッセージを避けるために初期化している．
*/
StatusType ercd = E_OK;
LOG_SCHED_ENTER();
CHECK_CALLEVEL(TCL_TASK);
CHECK_RESOURCE(tcb_lastres[runtsk] == RESID_NULL);
lock_cpu();
if (tinib_inipri[runtsk] < nextpri) {
tcb_curpri[runtsk] = tinib_inipri[runtsk];
preempt();
dispatch();
tcb_curpri[runtsk] = tinib_exepri[runtsk];
}
exit:
unlock_cpu();
LOG_SCHED_LEAVE(ercd);
return(ercd);
error_exit:
lock_cpu();
call_errorhook(ercd, OSServiceId_Schedule);
goto exit;
}
static void unboost_priority(struct task_struct* t)
{
unsigned long flags;
psnedf_domain_t* pedf = task_pedf(t);
lt_t now;
raw_readyq_lock_irqsave(&pedf->slock, flags);
now = litmus_clock();
/* assumption: this only happens when the job is scheduled */
BUG_ON(pedf->scheduled != t);
TRACE_TASK(t, "priority restored at %llu\n", now);
/* priority boosted jobs must be scheduled */
BUG_ON(pedf->scheduled != t);
tsk_rt(t)->priority_boosted = 0;
tsk_rt(t)->boost_start_time = 0;
/* check if this changes anything */
if (edf_preemption_needed(&pedf->domain, pedf->scheduled))
preempt(pedf);
raw_readyq_unlock_irqrestore(&pedf->slock, flags);
}
static inline void
userret(struct proc *p, struct trapframe *frame, u_quad_t oticks)
{
int sig;
/* take pending signals */
while ((sig = CURSIG(p)) != 0)
postsig(sig);
p->p_priority = p->p_usrpri;
if (want_resched) {
/*
* We're being preempted.
*/
preempt(NULL);
while ((sig = CURSIG(p)) != 0)
postsig(sig);
}
/*
* If profiling, charge recent system time to the trapped pc.
*/
if (p->p_flag & P_PROFIL) {
extern int psratio;
addupc_task(p, frame->tf_sxip & XIP_ADDR,
(int)(p->p_sticks - oticks) * psratio);
}
curpriority = p->p_priority;
}
void
userret(struct proc *p, u_int32_t pc, quad_t oticks)
{
int sig;
/* Take pending signals. */
while ((sig = (CURSIG(p))) != 0)
postsig(sig);
p->p_priority = p->p_usrpri;
if (want_resched) {
/*
* We're being preempted.
*/
preempt(NULL);
while ((sig = CURSIG(p)) != 0)
postsig(sig);
}
/*
* If profiling, charge recent system time to the trapped pc.
*/
if (p->p_flag & P_PROFIL) {
extern int psratio;
addupc_task(p, pc, (int)(p->p_sticks - oticks) * psratio);
}
curpriority = p->p_priority;
}
static __inline void
userret (struct lwp *l, register_t pc, u_quad_t oticks)
{
struct proc *p = l->l_proc;
int sig;
/* take pending signals */
while ((sig = CURSIG(l)) != 0)
postsig(sig);
l->l_priority = l->l_usrpri;
if (want_resched) {
/*
* We're being preempted.
*/
preempt(0);
while ((sig = CURSIG(l)) != 0)
postsig(sig);
}
/*
* If profiling, charge recent system time to the trapped pc.
*/
if (l->l_flag & P_PROFIL) {
extern int psratio;
addupc_task(p, pc, (int)(p->p_sticks - oticks) * psratio);
}
curcpu()->ci_schedstate.spc_curpriority = l->l_priority;
}
void
ast(struct trapframe *tf)
{
struct proc *p = curproc;;
#ifdef acorn26
/* Enable interrupts if they were enabled before the trap. */
if ((tf->tf_r15 & R15_IRQ_DISABLE) == 0)
int_on();
#else
/* Interrupts were restored by exception_exit. */
#endif
uvmexp.traps++;
uvmexp.softs++;
#ifdef DEBUG
if (p == NULL)
panic("ast: no curproc!");
if (&p->p_addr->u_pcb == 0)
panic("ast: no pcb!");
#endif
if (p->p_flag & P_OWEUPC) {
p->p_flag &= ~P_OWEUPC;
ADDUPROF(p);
}
/* Allow a forced task switch. */
if (want_resched)
preempt(0);
userret(p, tf->tf_pc, p->p_sticks); /* XXX */
}
static void boost_priority(struct task_struct* t)
{
unsigned long flags;
psnedf_domain_t* pedf = task_pedf(t);
lt_t now;
raw_readyq_lock_irqsave(&pedf->slock, flags);
now = litmus_clock();
TRACE_TASK(t, "priority boosted at %llu\n", now);
tsk_rt(t)->priority_boosted = 1;
tsk_rt(t)->boost_start_time = now;
if (pedf->scheduled != t) {
/* holder may be queued: first stop queue changes */
raw_spin_lock(&pedf->domain.release_lock);
if (is_queued(t) &&
/* If it is queued, then we need to re-order. */
bheap_decrease(edf_ready_order, tsk_rt(t)->heap_node) &&
/* If we bubbled to the top, then we need to check for preemptions. */
edf_preemption_needed(&pedf->domain, pedf->scheduled))
preempt(pedf);
raw_spin_unlock(&pedf->domain.release_lock);
} /* else: nothing to do since the job is not queued while scheduled */
raw_readyq_unlock_irqrestore(&pedf->slock, flags);
}
/*
* Attempt to release a writer lock. Return true on success.
*/
static int
write_unlock_try(rwlock_t *rwlp)
{
volatile uint32_t *rwstate = (volatile uint32_t *)&rwlp->rwlock_readers;
uint32_t readers;
ulwp_t *self = curthread;
no_preempt(self);
while (((readers = *rwstate) & URW_HAS_WAITERS) == 0) {
if (atomic_cas_32(rwstate, readers, 0) == readers) {
preempt(self);
return (1);
}
}
preempt(self);
return (0);
}
static int psnedf_preempt_check(psnedf_domain_t *pedf)
{
if (edf_preemption_needed(&pedf->domain, pedf->scheduled)) {
preempt(pedf);
return 1;
} else
return 0;
}
void irq_dispatch(regs r, uint32 number)
{
mask_irq(number);
if(irq_task_map[number]){
if(irq_task_map[number]->flags == tSLEEP_IRQ){
preempt(irq_task_map[number],ERR_NONE);
}
}
}
/*
* Attempt to acquire a readers lock. Return true on success.
*/
static int
read_lock_try(rwlock_t *rwlp, int ignore_waiters_flag)
{
volatile uint32_t *rwstate = (volatile uint32_t *)&rwlp->rwlock_readers;
uint32_t mask = ignore_waiters_flag?
URW_WRITE_LOCKED : (URW_HAS_WAITERS | URW_WRITE_LOCKED);
uint32_t readers;
ulwp_t *self = curthread;
no_preempt(self);
while (((readers = *rwstate) & mask) == 0) {
if (atomic_cas_32(rwstate, readers, readers + 1) == readers) {
preempt(self);
return (1);
}
}
preempt(self);
return (0);
}
void
amap_wipeout(struct vm_amap *amap)
{
int lcv, slot;
struct vm_anon *anon;
UVMHIST_FUNC("amap_wipeout"); UVMHIST_CALLED(maphist);
UVMHIST_LOG(maphist,"(amap=0x%x)", amap, 0,0,0);
KASSERT(amap->am_ref == 0);
if (__predict_false((amap->am_flags & AMAP_SWAPOFF) != 0)) {
/*
* amap_swap_off will call us again.
*/
amap_unlock(amap);
return;
}
amap_list_remove(amap);
amap_unlock(amap);
for (lcv = 0 ; lcv < amap->am_nused ; lcv++) {
int refs;
slot = amap->am_slots[lcv];
anon = amap->am_anon[slot];
if (anon == NULL || anon->an_ref == 0)
panic("amap_wipeout: corrupt amap");
mutex_enter(&anon->an_lock);
UVMHIST_LOG(maphist," processing anon 0x%x, ref=%d", anon,
anon->an_ref, 0, 0);
refs = --anon->an_ref;
mutex_exit(&anon->an_lock);
if (refs == 0) {
/*
* we had the last reference to a vm_anon. free it.
*/
uvm_anfree(anon);
}
if (curlwp->l_cpu->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
preempt();
}
/*
* now we free the map
*/
amap->am_nused = 0;
amap_free(amap); /* will unlock and free amap */
UVMHIST_LOG(maphist,"<- done!", 0,0,0,0);
}
/*
* This is the loop for the CPU threads. The general idea:
*
* 1) Each CPU thread has a state variable. While the library is using the
* CPU thread to simulate a process, this variable is set to CPU_RUNNING.
*
* 2) To "simulate" a process, we simply block on a condition variable.
* Each CPU thread has a dedicated condition variable.
*
* 3) For simplicity, the supervisor thread actually does all of the work.
* This makes synchronization in the simulator much easier, since all
* the real work is done by a single thread. So, when the supervisor
* wants to dispatch an event to a CPU thread, it needs to unblock the
* CPU thread. It does this by setting the CPU thread's state variable
* to inform the CPU thread of the event, then it signals the condition
* variable.
*
* 4) Once the CPU thread unblocks, it calls the students event handler,
* then goes back to step 1.
*
* There is one special case: idle. Idle is simulated by the student's code,
* not the library's. So we simply set the state variable to CPU_IDLE, and
* call the student's code.
*/
static void simulator_cpu_thread(unsigned int cpu_id) {
simulator_cpu_state_t state;
while (1) {
pthread_mutex_lock(&simulator_mutex);
if (simulator_cpu_data[cpu_id].current == NULL) {
/* the idle process was selected */
simulator_cpu_data[cpu_id].state = CPU_IDLE;
} else {
/* a process was scheduled */
simulator_cpu_data[cpu_id].state = CPU_RUNNING;
while (simulator_cpu_data[cpu_id].state == CPU_RUNNING)
pthread_cond_wait(&simulator_cpu_data[cpu_id].wakeup, &simulator_mutex);
}
state = simulator_cpu_data[cpu_id].state;
pthread_mutex_unlock(&simulator_mutex);
/* Call student's code */
switch (state) {
case CPU_IDLE:
/*
* We can't lock the student_lock for idle(); otherwise we can't
* print statistics while any CPU is idling.
*/
idle(cpu_id);
break;
case CPU_PREEMPT:
IRWL_WRITER_LOCK(student_lock)
preempt(cpu_id);
IRWL_WRITER_UNLOCK(student_lock)
break;
case CPU_YIELD:
IRWL_WRITER_LOCK(student_lock)
yield(cpu_id);
IRWL_WRITER_UNLOCK(student_lock)
break;
case CPU_TERMINATE:
processes_terminated++;
IRWL_WRITER_LOCK(student_lock)
terminate(cpu_id);
IRWL_WRITER_UNLOCK(student_lock)
break;
case CPU_RUNNING:
/* This should never happen!!! */
break;
}
}
}
int
main(int argc, char *argv[])
{
printf(1, "usertests starting\n");
if(open("usertests.ran", 0) >= 0){
printf(1, "already ran user tests -- rebuild fs.img\n");
exit();
}
close(open("usertests.ran", O_CREATE));
createdelete();
linkunlink();
concreate();
fourfiles();
sharedfd();
bigargtest();
bigwrite();
bigargtest();
bsstest();
sbrktest();
validatetest();
opentest();
writetest();
writetest1();
createtest();
openiputtest();
exitiputtest();
iputtest();
mem();
pipe1();
preempt();
exitwait();
rmdot();
fourteen();
bigfile();
subdir();
linktest();
unlinkread();
dirfile();
iref();
forktest();
bigdir(); // slow
exectest();
exit();
}
/*
* リソースの返却
*/
StatusType
ReleaseResource(ResourceType resid)
{
StatusType ercd = E_OK;
LOG_RELRES_ENTER(resid);
CHECK_CALLEVEL(TCL_TASK | TCL_ISR2);
CHECK_RESID(resid);
if (callevel == TCL_TASK) {
CHECK_ACCESS(tinib_inipri[runtsk] <= resinib_ceilpri[resid]);
CHECK_NOFUNC(tcb_lastres[runtsk] == resid);
lock_cpu();
if (rescb_prevpri[resid] >= TPRI_MINISR) {
set_ipl(rescb_prevpri[resid] - TPRI_MINISR);
}
else{
if (tcb_curpri[runtsk] >= TPRI_MINISR) {
set_ipl(IPL_ENA_ALL);
}
}
tcb_curpri[runtsk] = rescb_prevpri[resid];
tcb_lastres[runtsk] = rescb_prevres[resid];
rescb_prevpri[resid] = TPRI_NULL;
if (tcb_curpri[runtsk] < nextpri) {
preempt();
dispatch();
}
}
else {
CHECK_ACCESS(isrinib_intpri[runisr] <= resinib_ceilpri[resid]);
CHECK_NOFUNC(isrcb_lastres[runisr] == resid);
lock_cpu();
set_ipl(rescb_prevpri[resid] - TPRI_MINISR);
isrcb_lastres[runisr] = rescb_prevres[resid];
rescb_prevpri[resid] = TPRI_NULL;
}
exit:
unlock_cpu();
LOG_RELRES_LEAVE(ercd);
return(ercd);
error_exit:
lock_cpu();
_errorhook_par1.resid = resid;
call_errorhook(ercd, OSServiceId_ReleaseResource);
goto exit;
}
/*
* Handle asynchronous software traps.
*/
void
ast(struct trapframe *frame)
{
struct cpu_info *ci = curcpu();
struct proc *p = ci->ci_curproc;
uvmexp.softs++;
p->p_md.md_astpending = 0;
if (p->p_flag & P_OWEUPC) {
KERNEL_LOCK();
ADDUPROF(p);
KERNEL_UNLOCK();
}
if (ci->ci_want_resched)
preempt(NULL);
userret(p);
}
/*
* Handle an AST for the current process.
*/
void
ast()
{
struct cpu_info *ci = curcpu();
struct proc *p = ci->ci_curproc;
uvmexp.softs++;
if (p->p_md.md_astpending == 0)
panic("unexpected ast p %p astpending %p\n", p, &p->p_md.md_astpending);
p->p_md.md_astpending = 0;
if (p->p_flag & P_OWEUPC) {
ADDUPROF(p);
}
if (ci->ci_want_resched)
preempt(NULL);
userret(p);
}
int
rw_write_held(rwlock_t *rwlp)
{
volatile uint32_t *rwstate = (volatile uint32_t *)&rwlp->rwlock_readers;
uint32_t readers;
ulwp_t *self = curthread;
int rval;
no_preempt(self);
readers = *rwstate;
ASSERT_CONSISTENT_STATE(readers);
rval = ((readers & URW_WRITE_LOCKED) &&
rwlp->rwlock_owner == (uintptr_t)self &&
(rwlp->rwlock_type == USYNC_THREAD ||
rwlp->rwlock_ownerpid == self->ul_uberdata->pid));
preempt(self);
return (rval);
}
/**
Called when a new job arrives.
If multiple cores are idle, the job should be assigned to the core with the
lowest id.
If the job arriving should be scheduled to run during the next
time cycle, return the zero-based index of the core the job should be
scheduled on. If another job is already running on the core specified,
this will preempt the currently running job.
Assumptions:
- You may assume that every job wil have a unique arrival time.
@param job_number a globally unique identification number of the job arriving.
@param time the current time of the simulator.
@param running_time the total number of time units this job will run before it will be finished.
@param priority the priority of the job. (The lower the value, the higher the priority.)
@return index of core job should be scheduled on
@return -1 if no scheduling changes should be made.
*/
int scheduler_new_job(int job_number, int time, int running_time, int priority)
{
int i;
inc_time(time);
job_t* job = create_job(job_number, time, running_time, priority);
if ( (i = get_core()) != -1 )
{
insert_job(i,job);
return i;
}//if
else if ( is_prempt() )
{
i = preempt(job);
if ( i == -1 )
priqueue_offer(jobs,job);
return i;
}//else if
priqueue_offer(jobs,job);
return -1;
}//scheduler_new_job
/**
* trace preemption when the given task moves in a different CPU.
* the active queue must be locked.
*/
void preempt_in(resch_task_t *rt)
{
unsigned long flags;
int cpu = rt->cpu_id;
resch_task_t *curr = lo[cpu].current_task;
if (curr && curr->prio >= rt->prio) {
preempt_lock(cpu, &flags);
/* the previous task in the active queue must be valid. */
rt->preempter = active_prev_prio_task(rt);
rt->preemptee = rt->preempter->preemptee;
rt->preempter->preemptee = rt;
if (rt->preemptee) {
rt->preemptee->preempter = rt;
}
preempt_unlock(cpu, &flags);
}
else {
preempt(rt);
}
}
void
ast(struct trapframe *tf)
{
struct lwp * const l = curlwp;
#ifdef acorn26
/* Enable interrupts if they were enabled before the trap. */
if ((tf->tf_r15 & R15_IRQ_DISABLE) == 0)
int_on();
#else
/* Interrupts were restored by exception_exit. */
#endif
#ifdef __PROG32
KASSERT(VALID_R15_PSR(tf->tf_pc, tf->tf_spsr));
#endif
#ifdef __HAVE_PREEMPTION
kpreempt_disable();
#endif
struct cpu_info * const ci = curcpu();
ci->ci_data.cpu_ntrap++;
KDASSERT(ci->ci_cpl == IPL_NONE);
const int want_resched = ci->ci_want_resched;
#ifdef __HAVE_PREEMPTION
kpreempt_enable();
#endif
if (l->l_pflag & LP_OWEUPC) {
l->l_pflag &= ~LP_OWEUPC;
ADDUPROF(l);
}
/* Allow a forced task switch. */
if (want_resched)
preempt();
userret(l);
}
void
dme_phy_update_media(struct dme_softc *sc)
{
u_int ifm_media = sc->sc_media.ifm_media;
uint32_t reg;
if (IFM_SUBTYPE(ifm_media) == IFM_AUTO) {
/* If auto-negotiation is used, ensures that it is completed
before trying to extract any media information. */
reg = dme_phy_read(sc, DM9000_PHY_BMSR);
if ((reg & DM9000_PHY_BMSR_AUTO_NEG_AB) == 0) {
/* Auto-negotation not possible, therefore there is no
reason to try obtain any media information. */
return;
}
/* Then loop until the negotiation is completed. */
while ((reg & DM9000_PHY_BMSR_AUTO_NEG_COM) == 0) {
/* TODO: Bail out after a finite number of attempts
in case something goes wrong. */
preempt();
reg = dme_phy_read(sc, DM9000_PHY_BMSR);
}
}
sc->sc_media_active = IFM_ETHER;
reg = dme_phy_read(sc, DM9000_PHY_BMCR);
if (reg & DM9000_PHY_BMCR_SPEED_SELECT) {
sc->sc_media_active |= IFM_100_TX;
} else {
sc->sc_media_active |= IFM_10_T;
}
if (reg & DM9000_PHY_BMCR_DUPLEX_MODE) {
sc->sc_media_active |= IFM_FDX;
}
}
int
rw_read_held(rwlock_t *rwlp)
{
volatile uint32_t *rwstate = (volatile uint32_t *)&rwlp->rwlock_readers;
uint32_t readers;
ulwp_t *self = curthread;
readlock_t *readlockp;
uint_t nlocks;
int rval = 0;
no_preempt(self);
readers = *rwstate;
ASSERT_CONSISTENT_STATE(readers);
if (!(readers & URW_WRITE_LOCKED) &&
(readers & URW_READERS_MASK) != 0) {
/*
* The lock is held for reading by some thread.
* Search our array of rwlocks held for reading for a match.
*/
if ((nlocks = self->ul_rdlockcnt) != 0)
readlockp = self->ul_readlock.array;
else {
nlocks = 1;
readlockp = &self->ul_readlock.single;
}
for (; nlocks; nlocks--, readlockp++) {
if (readlockp->rd_rwlock == rwlp) {
if (readlockp->rd_count)
rval = 1;
break;
}
}
}
preempt(self);
return (rval);
}
/*
* This function does the same as uiomove, but takes an explicit
* direction, and does not update the uio structure.
*/
static int
_bus_dma_uiomove(void *buf, struct uio *uio, size_t n, int direction)
{
struct iovec *iov;
int error;
struct vmspace *vm;
char *cp;
size_t resid, cnt;
int i;
iov = uio->uio_iov;
vm = uio->uio_vmspace;
cp = buf;
resid = n;
for (i = 0; i < uio->uio_iovcnt && resid > 0; i++) {
iov = &uio->uio_iov[i];
if (iov->iov_len == 0)
continue;
cnt = MIN(resid, iov->iov_len);
if (!VMSPACE_IS_KERNEL_P(vm) &&
(curlwp->l_cpu->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
!= 0) {
preempt();
}
if (direction == UIO_READ) {
error = copyout_vmspace(vm, cp, iov->iov_base, cnt);
} else {
error = copyin_vmspace(vm, iov->iov_base, cp, cnt);
}
if (error)
return (error);
cp += cnt;
resid -= cnt;
}
return (0);
}
void
userret(struct proc *p, register_t pc, u_quad_t oticks)
{
int sig;
/* take pending signals */
while ((sig = CURSIG(p)) != 0)
postsig(sig);
p->p_priority = p->p_usrpri;
if (astpending) {
astpending = 0;
if (p->p_flag & P_OWEUPC) {
ADDUPROF(p);
}
}
if (want_resched) {
/*
* We're being preempted.
*/
preempt(NULL);
while ((sig = CURSIG(p)) != 0)
postsig(sig);
}
/*
* If profiling, charge recent system time to the trapped pc.
*/
if (p->p_flag & P_PROFIL) {
extern int psratio;
addupc_task(p, pc, (int)(p->p_sticks - oticks) * psratio);
}
p->p_cpu->ci_schedstate.spc_curpriority = p->p_priority;
}
static void psnedf_task_exit(struct task_struct * t)
{
unsigned long flags;
psnedf_domain_t* pedf = task_pedf(t);
rt_domain_t* edf;
raw_readyq_lock_irqsave(&pedf->slock, flags);
/* disable budget enforcement */
budget_state_machine(t,on_exit);
if (is_queued(t)) {
/* dequeue */
edf = task_edf(t);
remove(edf, t);
}
if (pedf->scheduled == t)
pedf->scheduled = NULL;
TRACE_TASK(t, "RIP, now reschedule\n");
preempt(pedf);
raw_readyq_unlock_irqrestore(&pedf->slock, flags);
}
/*
* Post-syscall processing. Perform abnormal system call completion
* actions such as /proc tracing, profiling, signals, preemption, etc.
*
* This routine is called only if t_post_sys, t_sig_check, or t_astflag is set.
* Any condition requiring pre-syscall handling must set one of these.
* If the condition is persistent, this routine will repost t_post_sys.
*/
void
post_syscall(long rval1, long rval2)
{
kthread_t *t = curthread;
klwp_t *lwp = ttolwp(t);
proc_t *p = ttoproc(t);
struct regs *rp = lwptoregs(lwp);
uint_t error;
uint_t code = t->t_sysnum;
int repost = 0;
int proc_stop = 0; /* non-zero if stopping */
int sigprof = 0; /* non-zero if sending SIGPROF */
t->t_post_sys = 0;
error = lwp->lwp_errno;
/*
* Code can be zero if this is a new LWP returning after a forkall(),
* other than the one which matches the one in the parent which called
* forkall(). In these LWPs, skip most of post-syscall activity.
*/
if (code == 0)
goto sig_check;
/*
* If the trace flag is set, mark the lwp to take a single-step trap
* on return to user level (below). The x86 lcall interface and
* sysenter has already done this, and turned off the flag, but
* amd64 syscall interface has not.
*/
if (rp->r_ps & PS_T) {
lwp->lwp_pcb.pcb_flags |= DEBUG_PENDING;
rp->r_ps &= ~PS_T;
aston(curthread);
}
#ifdef C2_AUDIT
if (audit_active) { /* put out audit record for this syscall */
rval_t rval;
/* XX64 -- truncation of 64-bit return values? */
rval.r_val1 = (int)rval1;
rval.r_val2 = (int)rval2;
audit_finish(T_SYSCALL, code, error, &rval);
repost = 1;
}
#endif /* C2_AUDIT */
if (curthread->t_pdmsg != NULL) {
char *m = curthread->t_pdmsg;
uprintf("%s", m);
kmem_free(m, strlen(m) + 1);
curthread->t_pdmsg = NULL;
}
/*
* If we're going to stop for /proc tracing, set the flag and
* save the arguments so that the return values don't smash them.
*/
if (PTOU(p)->u_systrap) {
if (prismember(&PTOU(p)->u_exitmask, code)) {
if (lwp_getdatamodel(lwp) == DATAMODEL_LP64)
(void) save_syscall_args();
proc_stop = 1;
}
repost = 1;
}
/*
* Similarly check to see if SIGPROF might be sent.
*/
if (curthread->t_rprof != NULL &&
curthread->t_rprof->rp_anystate != 0) {
if (lwp_getdatamodel(lwp) == DATAMODEL_LP64)
(void) save_syscall_args();
sigprof = 1;
}
if (lwp->lwp_eosys == NORMALRETURN) {
if (error == 0) {
#ifdef SYSCALLTRACE
if (syscalltrace) {
mutex_enter(&systrace_lock);
printf(
"%d: r_val1=0x%lx, r_val2=0x%lx, id 0x%p\n",
p->p_pid, rval1, rval2, curthread);
mutex_exit(&systrace_lock);
}
#endif /* SYSCALLTRACE */
rp->r_ps &= ~PS_C;
rp->r_r0 = rval1;
rp->r_r1 = rval2;
} else {
int sig;
#ifdef SYSCALLTRACE
if (syscalltrace) {
mutex_enter(&systrace_lock);
printf("%d: error=%d, id 0x%p\n",
p->p_pid, error, curthread);
mutex_exit(&systrace_lock);
}
#endif /* SYSCALLTRACE */
if (error == EINTR && t->t_activefd.a_stale)
error = EBADF;
if (error == EINTR &&
(sig = lwp->lwp_cursig) != 0 &&
sigismember(&PTOU(p)->u_sigrestart, sig) &&
PTOU(p)->u_signal[sig - 1] != SIG_DFL &&
PTOU(p)->u_signal[sig - 1] != SIG_IGN)
error = ERESTART;
rp->r_r0 = error;
rp->r_ps |= PS_C;
}
}
/*
* From the proc(4) manual page:
* When exit from a system call is being traced, the traced process
* stops on completion of the system call just prior to checking for
* signals and returning to user level. At this point all return
* values have been stored into the traced process's saved registers.
*/
if (proc_stop) {
mutex_enter(&p->p_lock);
if (PTOU(p)->u_systrap &&
prismember(&PTOU(p)->u_exitmask, code))
stop(PR_SYSEXIT, code);
mutex_exit(&p->p_lock);
}
/*
* If we are the parent returning from a successful
* vfork, wait for the child to exec or exit.
* This code must be here and not in the bowels of the system
* so that /proc can intercept exit from vfork in a timely way.
*/
if (code == SYS_vfork && rp->r_r1 == 0 && error == 0)
vfwait((pid_t)rval1);
/*
* If profiling is active, bill the current PC in user-land
* and keep reposting until profiling is disabled.
*/
if (p->p_prof.pr_scale) {
if (lwp->lwp_oweupc)
profil_tick(rp->r_pc);
repost = 1;
}
sig_check:
/*
* Reset flag for next time.
* We must do this after stopping on PR_SYSEXIT
* because /proc uses the information in lwp_eosys.
*/
lwp->lwp_eosys = NORMALRETURN;
clear_stale_fd();
t->t_flag &= ~T_FORKALL;
if (t->t_astflag | t->t_sig_check) {
/*
* Turn off the AST flag before checking all the conditions that
* may have caused an AST. This flag is on whenever a signal or
* unusual condition should be handled after the next trap or
* syscall.
*/
astoff(t);
/*
* If a single-step trap occurred on a syscall (see trap())
* recognize it now. Do this before checking for signals
* because deferred_singlestep_trap() may generate a SIGTRAP to
* the LWP or may otherwise mark the LWP to call issig(FORREAL).
*/
if (lwp->lwp_pcb.pcb_flags & DEBUG_PENDING)
deferred_singlestep_trap((caddr_t)rp->r_pc);
t->t_sig_check = 0;
/*
* The following check is legal for the following reasons:
* 1) The thread we are checking, is ourselves, so there is
* no way the proc can go away.
* 2) The only time we need to be protected by the
* lock is if the binding is changed.
*
* Note we will still take the lock and check the binding
* if the condition was true without the lock held. This
* prevents lock contention among threads owned by the
* same proc.
*/
if (curthread->t_proc_flag & TP_CHANGEBIND) {
mutex_enter(&p->p_lock);
if (curthread->t_proc_flag & TP_CHANGEBIND) {
timer_lwpbind();
curthread->t_proc_flag &= ~TP_CHANGEBIND;
}
mutex_exit(&p->p_lock);
}
/*
* for kaio requests on the special kaio poll queue,
* copyout their results to user memory.
*/
if (p->p_aio)
aio_cleanup(0);
/*
* If this LWP was asked to hold, call holdlwp(), which will
* stop. holdlwps() sets this up and calls pokelwps() which
* sets the AST flag.
*
* Also check TP_EXITLWP, since this is used by fresh new LWPs
* through lwp_rtt(). That flag is set if the lwp_create(2)
* syscall failed after creating the LWP.
*/
if (ISHOLD(p) || (t->t_proc_flag & TP_EXITLWP))
holdlwp();
/*
* All code that sets signals and makes ISSIG_PENDING
* evaluate true must set t_sig_check afterwards.
*/
if (ISSIG_PENDING(t, lwp, p)) {
if (issig(FORREAL))
psig();
t->t_sig_check = 1; /* recheck next time */
}
if (sigprof) {
realsigprof(code, error);
t->t_sig_check = 1; /* recheck next time */
}
/*
* If a performance counter overflow interrupt was
* delivered *during* the syscall, then re-enable the
* AST so that we take a trip through trap() to cause
* the SIGEMT to be delivered.
*/
if (lwp->lwp_pcb.pcb_flags & CPC_OVERFLOW)
aston(t);
/*
* /proc can't enable/disable the trace bit itself
* because that could race with the call gate used by
* system calls via "lcall". If that happened, an
* invalid EFLAGS would result. prstep()/prnostep()
* therefore schedule an AST for the purpose.
*/
if (lwp->lwp_pcb.pcb_flags & REQUEST_STEP) {
lwp->lwp_pcb.pcb_flags &= ~REQUEST_STEP;
rp->r_ps |= PS_T;
}
if (lwp->lwp_pcb.pcb_flags & REQUEST_NOSTEP) {
lwp->lwp_pcb.pcb_flags &= ~REQUEST_NOSTEP;
rp->r_ps &= ~PS_T;
}
}
lwp->lwp_errno = 0; /* clear error for next time */
#ifndef NPROBE
/* Kernel probe */
if (tnf_tracing_active) {
TNF_PROBE_3(syscall_end, "syscall thread", /* CSTYLED */,
tnf_long, rval1, rval1,
tnf_long, rval2, rval2,
tnf_long, errno, (long)error);
repost = 1;
}
#endif /* NPROBE */
/*
* Set state to LWP_USER here so preempt won't give us a kernel
* priority if it occurs after this point. Call CL_TRAPRET() to
* restore the user-level priority.
*
* It is important that no locks (other than spinlocks) be entered
* after this point before returning to user mode (unless lwp_state
* is set back to LWP_SYS).
*
* XXX Sampled times past this point are charged to the user.
*/
lwp->lwp_state = LWP_USER;
if (t->t_trapret) {
t->t_trapret = 0;
thread_lock(t);
CL_TRAPRET(t);
thread_unlock(t);
}
if (CPU->cpu_runrun)
preempt();
lwp->lwp_errno = 0; /* clear error for next time */
/*
* The thread lock must be held in order to clear sysnum and reset
* lwp_ap atomically with respect to other threads in the system that
* may be looking at the args via lwp_ap from get_syscall_args().
*/
thread_lock(t);
t->t_sysnum = 0; /* no longer in a system call */
if (lwp_getdatamodel(lwp) == DATAMODEL_NATIVE) {
#if defined(_LP64)
/*
* In case the args were copied to the lwp, reset the
* pointer so the next syscall will have the right
* lwp_ap pointer.
*/
lwp->lwp_ap = (long *)&rp->r_rdi;
} else {
#endif
lwp->lwp_ap = NULL; /* reset on every syscall entry */
}
thread_unlock(t);
lwp->lwp_argsaved = 0;
/*
* If there was a continuing reason for post-syscall processing,
* set the t_post_sys flag for the next system call.
*/
if (repost)
t->t_post_sys = 1;
/*
* If there is a ustack registered for this lwp, and the stack rlimit
* has been altered, read in the ustack. If the saved stack rlimit
* matches the bounds of the ustack, update the ustack to reflect
* the new rlimit. If the new stack rlimit is RLIM_INFINITY, disable
* stack checking by setting the size to 0.
*/
if (lwp->lwp_ustack != 0 && lwp->lwp_old_stk_ctl != 0) {
rlim64_t new_size;
caddr_t top;
stack_t stk;
struct rlimit64 rl;
mutex_enter(&p->p_lock);
new_size = p->p_stk_ctl;
top = p->p_usrstack;
(void) rctl_rlimit_get(rctlproc_legacy[RLIMIT_STACK], p, &rl);
mutex_exit(&p->p_lock);
if (rl.rlim_cur == RLIM64_INFINITY)
new_size = 0;
if (copyin((stack_t *)lwp->lwp_ustack, &stk,
sizeof (stack_t)) == 0 &&
(stk.ss_size == lwp->lwp_old_stk_ctl ||
stk.ss_size == 0) &&
stk.ss_sp == top - stk.ss_size) {
stk.ss_sp = (void *)((uintptr_t)stk.ss_sp +
stk.ss_size - (uintptr_t)new_size);
stk.ss_size = new_size;
(void) copyout(&stk, (stack_t *)lwp->lwp_ustack,
sizeof (stack_t));
}
lwp->lwp_old_stk_ctl = 0;
}
}
/*ARGSUSED*/
void
trap(struct frame *fp, int type, unsigned code, unsigned v)
{
extern char fubail[], subail[];
struct lwp *l;
struct proc *p;
struct pcb *pcb;
void *onfault;
ksiginfo_t ksi;
int s;
int rv;
u_quad_t sticks = 0 /* XXX initialiser works around compiler bug */;
static int panicking __diagused;
curcpu()->ci_data.cpu_ntrap++;
l = curlwp;
p = l->l_proc;
pcb = lwp_getpcb(l);
KSI_INIT_TRAP(&ksi);
ksi.ksi_trap = type & ~T_USER;
if (USERMODE(fp->f_sr)) {
type |= T_USER;
sticks = p->p_sticks;
l->l_md.md_regs = fp->f_regs;
LWP_CACHE_CREDS(l, p);
}
switch (type) {
default:
dopanic:
/*
* Let the kernel debugger see the trap frame that
* caused us to panic. This is a convenience so
* one can see registers at the point of failure.
*/
s = splhigh();
panicking = 1;
printf("trap type %d, code = 0x%x, v = 0x%x\n", type, code, v);
printf("%s program counter = 0x%x\n",
(type & T_USER) ? "user" : "kernel", fp->f_pc);
#ifdef KGDB
/* If connected, step or cont returns 1 */
if (kgdb_trap(type, (db_regs_t *)fp))
goto kgdb_cont;
#endif
#ifdef DDB
(void)kdb_trap(type, (db_regs_t *)fp);
#endif
#ifdef KGDB
kgdb_cont:
#endif
splx(s);
if (panicstr) {
printf("trap during panic!\n");
#ifdef DEBUG
/* XXX should be a machine-dependent hook */
printf("(press a key)\n"); (void)cngetc();
#endif
}
regdump((struct trapframe *)fp, 128);
type &= ~T_USER;
if ((u_int)type < trap_types)
panic(trap_type[type]);
panic("trap");
case T_BUSERR: /* kernel bus error */
onfault = pcb->pcb_onfault;
if (onfault == NULL)
goto dopanic;
rv = EFAULT;
/* FALLTHROUGH */
copyfault:
/*
* If we have arranged to catch this fault in any of the
* copy to/from user space routines, set PC to return to
* indicated location and set flag informing buserror code
* that it may need to clean up stack frame.
*/
fp->f_stackadj = exframesize[fp->f_format];
fp->f_format = fp->f_vector = 0;
fp->f_pc = (int)onfault;
fp->f_regs[D0] = rv;
return;
case T_BUSERR|T_USER: /* bus error */
case T_ADDRERR|T_USER: /* address error */
ksi.ksi_addr = (void *)v;
ksi.ksi_signo = SIGBUS;
ksi.ksi_code = (type == (T_BUSERR|T_USER)) ?
BUS_OBJERR : BUS_ADRERR;
break;
case T_COPERR: /* kernel coprocessor violation */
case T_FMTERR|T_USER: /* do all RTE errors come in as T_USER? */
case T_FMTERR: /* ...just in case... */
/*
* The user has most likely trashed the RTE or FP state info
* in the stack frame of a signal handler.
*/
printf("pid %d: kernel %s exception\n", p->p_pid,
type==T_COPERR ? "coprocessor" : "format");
type |= T_USER;
mutex_enter(p->p_lock);
SIGACTION(p, SIGILL).sa_handler = SIG_DFL;
sigdelset(&p->p_sigctx.ps_sigignore, SIGILL);
sigdelset(&p->p_sigctx.ps_sigcatch, SIGILL);
sigdelset(&l->l_sigmask, SIGILL);
mutex_exit(p->p_lock);
ksi.ksi_signo = SIGILL;
ksi.ksi_addr = (void *)(int)fp->f_format;
/* XXX was ILL_RESAD_FAULT */
ksi.ksi_code = (type == T_COPERR) ?
ILL_COPROC : ILL_ILLOPC;
break;
case T_COPERR|T_USER: /* user coprocessor violation */
/* What is a proper response here? */
ksi.ksi_signo = SIGFPE;
ksi.ksi_code = FPE_FLTINV;
break;
case T_FPERR|T_USER: /* 68881 exceptions */
/*
* We pass along the 68881 status register which locore stashed
* in code for us.
*/
ksi.ksi_signo = SIGFPE;
ksi.ksi_code = fpsr2siginfocode(code);
break;
#ifdef M68040
case T_FPEMULI|T_USER: /* unimplemented FP instruction */
case T_FPEMULD|T_USER: /* unimplemented FP data type */
/* XXX need to FSAVE */
printf("pid %d(%s): unimplemented FP %s at %x (EA %x)\n",
p->p_pid, p->p_comm,
fp->f_format == 2 ? "instruction" : "data type",
fp->f_pc, fp->f_fmt2.f_iaddr);
/* XXX need to FRESTORE */
ksi.ksi_signo = SIGFPE;
ksi.ksi_code = FPE_FLTINV;
break;
#endif
case T_ILLINST|T_USER: /* illegal instruction fault */
case T_PRIVINST|T_USER: /* privileged instruction fault */
ksi.ksi_addr = (void *)(int)fp->f_format;
/* XXX was ILL_PRIVIN_FAULT */
ksi.ksi_signo = SIGILL;
ksi.ksi_code = (type == (T_PRIVINST|T_USER)) ?
ILL_PRVOPC : ILL_ILLOPC;
break;
case T_ZERODIV|T_USER: /* Divide by zero */
ksi.ksi_addr = (void *)(int)fp->f_format;
/* XXX was FPE_INTDIV_TRAP */
ksi.ksi_signo = SIGFPE;
ksi.ksi_code = FPE_FLTDIV;
break;
case T_CHKINST|T_USER: /* CHK instruction trap */
ksi.ksi_addr = (void *)(int)fp->f_format;
/* XXX was FPE_SUBRNG_TRAP */
ksi.ksi_signo = SIGFPE;
break;
case T_TRAPVINST|T_USER: /* TRAPV instruction trap */
ksi.ksi_addr = (void *)(int)fp->f_format;
/* XXX was FPE_INTOVF_TRAP */
ksi.ksi_signo = SIGFPE;
break;
/*
* XXX: Trace traps are a nightmare.
*
* HP-UX uses trap #1 for breakpoints,
* NetBSD/m68k uses trap #2,
* SUN 3.x uses trap #15,
* DDB and KGDB uses trap #15 (for kernel breakpoints;
* handled elsewhere).
*
* NetBSD and HP-UX traps both get mapped by locore.s into T_TRACE.
* SUN 3.x traps get passed through as T_TRAP15 and are not really
* supported yet.
*
* XXX: We should never get kernel-mode T_TRAP15
* XXX: because locore.s now gives them special treatment.
*/
case T_TRAP15: /* kernel breakpoint */
#ifdef DEBUG
printf("unexpected kernel trace trap, type = %d\n", type);
printf("program counter = 0x%x\n", fp->f_pc);
#endif
fp->f_sr &= ~PSL_T;
return;
case T_TRACE|T_USER: /* user trace trap */
#ifdef COMPAT_SUNOS
/*
* SunOS uses Trap #2 for a "CPU cache flush".
* Just flush the on-chip caches and return.
*/
if (p->p_emul == &emul_sunos) {
ICIA();
DCIU();
return;
}
#endif
/* FALLTHROUGH */
case T_TRACE: /* tracing a trap instruction */
case T_TRAP15|T_USER: /* SUN user trace trap */
fp->f_sr &= ~PSL_T;
ksi.ksi_signo = SIGTRAP;
break;
case T_ASTFLT: /* system async trap, cannot happen */
goto dopanic;
case T_ASTFLT|T_USER: /* user async trap */
astpending = 0;
/*
* We check for software interrupts first. This is because
* they are at a higher level than ASTs, and on a VAX would
* interrupt the AST. We assume that if we are processing
* an AST that we must be at IPL0 so we don't bother to
* check. Note that we ensure that we are at least at SIR
* IPL while processing the SIR.
*/
spl1();
/* fall into... */
case T_SSIR: /* software interrupt */
case T_SSIR|T_USER:
/*
* If this was not an AST trap, we are all done.
*/
if (type != (T_ASTFLT|T_USER)) {
curcpu()->ci_data.cpu_ntrap--;
return;
}
spl0();
if (l->l_pflag & LP_OWEUPC) {
l->l_pflag &= ~LP_OWEUPC;
ADDUPROF(l);
}
if (curcpu()->ci_want_resched)
preempt();
goto out;
case T_MMUFLT: /* kernel mode page fault */
/*
* If we were doing profiling ticks or other user mode
* stuff from interrupt code, Just Say No.
*/
onfault = pcb->pcb_onfault;
if (onfault == fubail || onfault == subail) {
rv = EFAULT;
goto copyfault;
}
/* fall into ... */
case T_MMUFLT|T_USER: /* page fault */
{
vaddr_t va;
struct vmspace *vm = p->p_vmspace;
struct vm_map *map;
vm_prot_t ftype;
extern struct vm_map *kernel_map;
onfault = pcb->pcb_onfault;
#ifdef DEBUG
if ((mmudebug & MDB_WBFOLLOW) || MDB_ISPID(p->p_pid))
printf("trap: T_MMUFLT pid=%d, code=%x, v=%x, pc=%x, sr=%x\n",
p->p_pid, code, v, fp->f_pc, fp->f_sr);
#endif
/*
* It is only a kernel address space fault iff:
* 1. (type & T_USER) == 0 and
* 2. pcb_onfault not set or
* 3. pcb_onfault set but supervisor space data fault
* The last can occur during an exec() copyin where the
* argument space is lazy-allocated.
*/
if ((type & T_USER) == 0 && (onfault == NULL || KDFAULT(code)))
map = kernel_map;
else {
map = vm ? &vm->vm_map : kernel_map;
}
if (WRFAULT(code))
ftype = VM_PROT_WRITE;
else
ftype = VM_PROT_READ;
va = trunc_page((vaddr_t)v);
if (map == kernel_map && va == 0) {
printf("trap: bad kernel %s access at 0x%x\n",
(ftype & VM_PROT_WRITE) ? "read/write" :
"read", v);
goto dopanic;
}
#ifdef DIAGNOSTIC
if (interrupt_depth && !panicking) {
printf("trap: calling uvm_fault() from interrupt!\n");
goto dopanic;
}
#endif
pcb->pcb_onfault = NULL;
rv = uvm_fault(map, va, ftype);
pcb->pcb_onfault = onfault;
#ifdef DEBUG
if (rv && MDB_ISPID(p->p_pid))
printf("uvm_fault(%p, 0x%lx, 0x%x) -> 0x%x\n",
map, va, ftype, rv);
#endif
/*
* If this was a stack access we keep track of the maximum
* accessed stack size. Also, if vm_fault gets a protection
* failure it is due to accessing the stack region outside
* the current limit and we need to reflect that as an access
* error.
*/
if (rv == 0) {
if (map != kernel_map && (void *)va >= vm->vm_maxsaddr)
uvm_grow(p, va);
if (type == T_MMUFLT) {
if (ucas_ras_check(&fp->F_t)) {
return;
}
#ifdef M68040
if (cputype == CPU_68040)
(void) writeback(fp, 1);
#endif
return;
}
goto out;
}
if (rv == EACCES) {
ksi.ksi_code = SEGV_ACCERR;
rv = EFAULT;
} else
ksi.ksi_code = SEGV_MAPERR;
if (type == T_MMUFLT) {
if (onfault)
goto copyfault;
printf("uvm_fault(%p, 0x%lx, 0x%x) -> 0x%x\n",
map, va, ftype, rv);
printf(" type %x, code [mmu,,ssw]: %x\n",
type, code);
goto dopanic;
}
ksi.ksi_addr = (void *)v;
switch (rv) {
case ENOMEM:
printf("UVM: pid %d (%s), uid %d killed: out of swap\n",
p->p_pid, p->p_comm,
l->l_cred ?
kauth_cred_geteuid(l->l_cred) : -1);
ksi.ksi_signo = SIGKILL;
break;
case EINVAL:
ksi.ksi_signo = SIGBUS;
ksi.ksi_code = BUS_ADRERR;
break;
case EACCES:
ksi.ksi_signo = SIGSEGV;
ksi.ksi_code = SEGV_ACCERR;
break;
default:
ksi.ksi_signo = SIGSEGV;
ksi.ksi_code = SEGV_MAPERR;
break;
}
break;
}
}
trapsignal(l, &ksi);
if ((type & T_USER) == 0)
return;
out:
userret(l, fp, sticks, v, 1);
}
/*
* Convert a component of a pathname into a pointer to a locked inode.
* This is a very central and rather complicated routine.
* If the file system is not maintained in a strict tree hierarchy,
* this can result in a deadlock situation (see comments in code below).
*
* The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
* on whether the name is to be looked up, created, renamed, or deleted.
* When CREATE, RENAME, or DELETE is specified, information usable in
* creating, renaming, or deleting a directory entry may be calculated.
* If flag has LOCKPARENT or'ed into it and the target of the pathname
* exists, lookup returns both the target and its parent directory locked.
* When creating or renaming and LOCKPARENT is specified, the target may
* not be ".". When deleting and LOCKPARENT is specified, the target may
* be "."., but the caller must check to ensure it does an vrele and vput
* instead of two vputs.
*
* Overall outline of ext2fs_lookup:
*
* check accessibility of directory
* look for name in cache, if found, then if at end of path
* and deleting or creating, drop it, else return name
* search for name in directory, to found or notfound
* notfound:
* if creating, return locked directory, leaving info on available slots
* else return error
* found:
* if at end of path and deleting, return information to allow delete
* if at end of path and rewriting (RENAME and LOCKPARENT), lock target
* inode and return info to allow rewrite
* if not at end, add name to cache; if at end and neither creating
* nor deleting, add name to cache
*/
int
ext2fs_lookup(void *v)
{
struct vop_lookup_v2_args /* {
struct vnode *a_dvp;
struct vnode **a_vpp;
struct componentname *a_cnp;
} */ *ap = v;
struct vnode *vdp = ap->a_dvp; /* vnode for directory being searched */
struct inode *dp = VTOI(vdp); /* inode for directory being searched */
struct buf *bp; /* a buffer of directory entries */
struct ext2fs_direct *ep; /* the current directory entry */
int entryoffsetinblock; /* offset of ep in bp's buffer */
enum {NONE, COMPACT, FOUND} slotstatus;
doff_t slotoffset; /* offset of area with free space */
int slotsize; /* size of area at slotoffset */
int slotfreespace; /* amount of space free in slot */
int slotneeded; /* size of the entry we're seeking */
int numdirpasses; /* strategy for directory search */
doff_t endsearch; /* offset to end directory search */
doff_t prevoff; /* prev entry dp->i_offset */
struct vnode *tdp; /* returned by vcache_get */
doff_t enduseful; /* pointer past last used dir slot */
u_long bmask; /* block offset mask */
int namlen, error;
struct vnode **vpp = ap->a_vpp;
struct componentname *cnp = ap->a_cnp;
kauth_cred_t cred = cnp->cn_cred;
int flags;
int nameiop = cnp->cn_nameiop;
struct ufsmount *ump = dp->i_ump;
int dirblksiz = ump->um_dirblksiz;
ino_t foundino;
struct ufs_lookup_results *results;
flags = cnp->cn_flags;
bp = NULL;
slotoffset = -1;
*vpp = NULL;
/*
* Produce the auxiliary lookup results into i_crap. Increment
* its serial number so elsewhere we can tell if we're using
* stale results. This should not be done this way. XXX.
*/
results = &dp->i_crap;
dp->i_crapcounter++;
/*
* Check accessiblity of directory.
*/
if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
return (error);
if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
(cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
return (EROFS);
/*
* We now have a segment name to search for, and a directory to search.
*
* Before tediously performing a linear scan of the directory,
* check the name cache to see if the directory/name pair
* we are looking for is known already.
*/
if (cache_lookup(vdp, cnp->cn_nameptr, cnp->cn_namelen,
cnp->cn_nameiop, cnp->cn_flags, NULL, vpp)) {
return *vpp == NULLVP ? ENOENT : 0;
}
/*
* Suppress search for slots unless creating
* file and at end of pathname, in which case
* we watch for a place to put the new file in
* case it doesn't already exist.
*/
slotstatus = FOUND;
slotfreespace = slotsize = slotneeded = 0;
if ((nameiop == CREATE || nameiop == RENAME) &&
(flags & ISLASTCN)) {
slotstatus = NONE;
slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
}
/*
* If there is cached information on a previous search of
* this directory, pick up where we last left off.
* We cache only lookups as these are the most common
* and have the greatest payoff. Caching CREATE has little
* benefit as it usually must search the entire directory
* to determine that the entry does not exist. Caching the
* location of the last DELETE or RENAME has not reduced
* profiling time and hence has been removed in the interest
* of simplicity.
*/
bmask = vdp->v_mount->mnt_stat.f_iosize - 1;
if (nameiop != LOOKUP || results->ulr_diroff == 0 ||
results->ulr_diroff >= ext2fs_size(dp)) {
entryoffsetinblock = 0;
results->ulr_offset = 0;
numdirpasses = 1;
} else {
results->ulr_offset = results->ulr_diroff;
if ((entryoffsetinblock = results->ulr_offset & bmask) &&
(error = ext2fs_blkatoff(vdp, (off_t)results->ulr_offset, NULL, &bp)))
return (error);
numdirpasses = 2;
namecache_count_2passes();
}
prevoff = results->ulr_offset;
endsearch = roundup(ext2fs_size(dp), dirblksiz);
enduseful = 0;
searchloop:
while (results->ulr_offset < endsearch) {
if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
preempt();
/*
* If necessary, get the next directory block.
*/
if ((results->ulr_offset & bmask) == 0) {
if (bp != NULL)
brelse(bp, 0);
error = ext2fs_blkatoff(vdp, (off_t)results->ulr_offset, NULL,
&bp);
if (error != 0)
return (error);
entryoffsetinblock = 0;
}
/*
* If still looking for a slot, and at a dirblksize
* boundary, have to start looking for free space again.
*/
if (slotstatus == NONE &&
(entryoffsetinblock & (dirblksiz - 1)) == 0) {
slotoffset = -1;
slotfreespace = 0;
}
/*
* Get pointer to next entry.
* Full validation checks are slow, so we only check
* enough to insure forward progress through the
* directory. Complete checks can be run by patching
* "dirchk" to be true.
*/
KASSERT(bp != NULL);
ep = (struct ext2fs_direct *)
((char *)bp->b_data + entryoffsetinblock);
if (ep->e2d_reclen == 0 ||
(dirchk &&
ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) {
int i;
ufs_dirbad(dp, results->ulr_offset, "mangled entry");
i = dirblksiz - (entryoffsetinblock & (dirblksiz - 1));
results->ulr_offset += i;
entryoffsetinblock += i;
continue;
}
/*
* If an appropriate sized slot has not yet been found,
* check to see if one is available. Also accumulate space
* in the current block so that we can determine if
* compaction is viable.
*/
if (slotstatus != FOUND) {
int size = fs2h16(ep->e2d_reclen);
if (ep->e2d_ino != 0)
size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
if (size > 0) {
if (size >= slotneeded) {
slotstatus = FOUND;
slotoffset = results->ulr_offset;
slotsize = fs2h16(ep->e2d_reclen);
} else if (slotstatus == NONE) {
slotfreespace += size;
if (slotoffset == -1)
slotoffset = results->ulr_offset;
if (slotfreespace >= slotneeded) {
slotstatus = COMPACT;
slotsize = results->ulr_offset +
fs2h16(ep->e2d_reclen) -
slotoffset;
}
}
}
}
/*
* Check for a name match.
*/
if (ep->e2d_ino) {
namlen = ep->e2d_namlen;
if (namlen == cnp->cn_namelen &&
!memcmp(cnp->cn_nameptr, ep->e2d_name,
(unsigned)namlen)) {
/*
* Save directory entry's inode number and
* reclen in ndp->ni_ufs area, and release
* directory buffer.
*/
foundino = fs2h32(ep->e2d_ino);
results->ulr_reclen = fs2h16(ep->e2d_reclen);
goto found;
}
}
prevoff = results->ulr_offset;
results->ulr_offset += fs2h16(ep->e2d_reclen);
entryoffsetinblock += fs2h16(ep->e2d_reclen);
if (ep->e2d_ino)
enduseful = results->ulr_offset;
}
/* notfound: */
/*
* If we started in the middle of the directory and failed
* to find our target, we must check the beginning as well.
*/
if (numdirpasses == 2) {
numdirpasses--;
results->ulr_offset = 0;
endsearch = results->ulr_diroff;
goto searchloop;
}
if (bp != NULL)
brelse(bp, 0);
/*
* If creating, and at end of pathname and current
* directory has not been removed, then can consider
* allowing file to be created.
*/
if ((nameiop == CREATE || nameiop == RENAME) &&
(flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
/*
* Access for write is interpreted as allowing
* creation of files in the directory.
*/
error = VOP_ACCESS(vdp, VWRITE, cred);
if (error)
return (error);
/*
* Return an indication of where the new directory
* entry should be put. If we didn't find a slot,
* then set results->ulr_count to 0 indicating
* that the new slot belongs at the end of the
* directory. If we found a slot, then the new entry
* can be put in the range from results->ulr_offset to
* results->ulr_offset + results->ulr_count.
*/
if (slotstatus == NONE) {
results->ulr_offset = roundup(ext2fs_size(dp), dirblksiz);
results->ulr_count = 0;
enduseful = results->ulr_offset;
} else {
results->ulr_offset = slotoffset;
results->ulr_count = slotsize;
if (enduseful < slotoffset + slotsize)
enduseful = slotoffset + slotsize;
}
results->ulr_endoff = roundup(enduseful, dirblksiz);
#if 0
dp->i_flag |= IN_CHANGE | IN_UPDATE;
#endif
/*
* We return with the directory locked, so that
* the parameters we set up above will still be
* valid if we actually decide to do a direnter().
* We return ni_vp == NULL to indicate that the entry
* does not currently exist; we leave a pointer to
* the (locked) directory inode in ndp->ni_dvp.
*
* NB - if the directory is unlocked, then this
* information cannot be used.
*/
return (EJUSTRETURN);
}
/*
* Insert name into cache (as non-existent) if appropriate.
*/
if (nameiop != CREATE) {
cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen,
cnp->cn_flags);
}
return ENOENT;
found:
if (numdirpasses == 2)
namecache_count_pass2();
/*
* Check that directory length properly reflects presence
* of this entry.
*/
if (results->ulr_offset + EXT2FS_DIRSIZ(ep->e2d_namlen) > ext2fs_size(dp)) {
ufs_dirbad(dp, results->ulr_offset, "i_size too small");
error = ext2fs_setsize(dp,
results->ulr_offset + EXT2FS_DIRSIZ(ep->e2d_namlen));
if (error) {
brelse(bp, 0);
return (error);
}
dp->i_flag |= IN_CHANGE | IN_UPDATE;
uvm_vnp_setsize(vdp, ext2fs_size(dp));
}
brelse(bp, 0);
/*
* Found component in pathname.
* If the final component of path name, save information
* in the cache as to where the entry was found.
*/
if ((flags & ISLASTCN) && nameiop == LOOKUP)
results->ulr_diroff = results->ulr_offset &~ (dirblksiz - 1);
/*
* If deleting, and at end of pathname, return
* parameters which can be used to remove file.
* Lock the inode, being careful with ".".
*/
if (nameiop == DELETE && (flags & ISLASTCN)) {
/*
* Return pointer to current entry in results->ulr_offset,
* and distance past previous entry (if there
* is a previous entry in this block) in results->ulr_count.
* Save directory inode pointer in ndp->ni_dvp for dirremove().
*/
if ((results->ulr_offset & (dirblksiz - 1)) == 0)
results->ulr_count = 0;
else
results->ulr_count = results->ulr_offset - prevoff;
if (dp->i_number == foundino) {
vref(vdp);
tdp = vdp;
} else {
error = vcache_get(vdp->v_mount,
&foundino, sizeof(foundino), &tdp);
if (error)
return (error);
}
/*
* Write access to directory required to delete files.
*/
if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0) {
vrele(tdp);
return (error);
}
/*
* If directory is "sticky", then user must own
* the directory, or the file in it, else she
* may not delete it (unless she's root). This
* implements append-only directories.
*/
if (dp->i_e2fs_mode & ISVTX) {
error = kauth_authorize_vnode(cred, KAUTH_VNODE_DELETE,
tdp, vdp, genfs_can_sticky(cred, dp->i_uid,
VTOI(tdp)->i_uid));
if (error) {
vrele(tdp);
return (EPERM);
}
}
*vpp = tdp;
return (0);
}
/*
* If rewriting (RENAME), return the inode and the
* information required to rewrite the present directory
* Must get inode of directory entry to verify it's a
* regular file, or empty directory.
*/
if (nameiop == RENAME && (flags & ISLASTCN)) {
error = VOP_ACCESS(vdp, VWRITE, cred);
if (error)
return (error);
/*
* Careful about locking second inode.
* This can only occur if the target is ".".
*/
if (dp->i_number == foundino)
return (EISDIR);
error = vcache_get(vdp->v_mount,
&foundino, sizeof(foundino), &tdp);
if (error)
return (error);
*vpp = tdp;
return (0);
}
if (dp->i_number == foundino) {
vref(vdp); /* we want ourself, ie "." */
*vpp = vdp;
} else {
error = vcache_get(vdp->v_mount,
&foundino, sizeof(foundino), &tdp);
if (error)
return (error);
*vpp = tdp;
}
/*
* Insert name into cache if appropriate.
*/
cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags);
return 0;
}
/*ARGSUSED*/
void
trap(struct trapframe *frame)
{
struct proc *p = curproc;
int type = (int)frame->tf_trapno;
struct pcb *pcb;
extern char doreti_iret[], resume_iret[];
caddr_t onfault;
int error;
uint64_t cr2;
union sigval sv;
uvmexp.traps++;
pcb = (p != NULL && p->p_addr != NULL) ? &p->p_addr->u_pcb : NULL;
#ifdef DEBUG
if (trapdebug) {
printf("trap %d code %lx rip %lx cs %lx rflags %lx cr2 %lx "
"cpl %x\n",
type, frame->tf_err, frame->tf_rip, frame->tf_cs,
frame->tf_rflags, rcr2(), curcpu()->ci_ilevel);
printf("curproc %p\n", curproc);
if (curproc)
printf("pid %d\n", p->p_pid);
}
#endif
if (!KERNELMODE(frame->tf_cs, frame->tf_rflags)) {
type |= T_USER;
p->p_md.md_regs = frame;
}
switch (type) {
default:
we_re_toast:
#ifdef KGDB
if (kgdb_trap(type, frame))
return;
else {
/*
* If this is a breakpoint, don't panic
* if we're not connected.
*/
if (type == T_BPTFLT) {
printf("kgdb: ignored %s\n", trap_type[type]);
return;
}
}
#endif
#ifdef DDB
if (kdb_trap(type, 0, frame))
return;
#endif
if (frame->tf_trapno < trap_types)
printf("fatal %s", trap_type[frame->tf_trapno]);
else
printf("unknown trap %ld", (u_long)frame->tf_trapno);
printf(" in %s mode\n", (type & T_USER) ? "user" : "supervisor");
printf("trap type %d code %lx rip %lx cs %lx rflags %lx cr2 "
" %lx cpl %x rsp %lx\n",
type, frame->tf_err, (u_long)frame->tf_rip, frame->tf_cs,
frame->tf_rflags, rcr2(), curcpu()->ci_ilevel, frame->tf_rsp);
panic("trap type %d, code=%lx, pc=%lx",
type, frame->tf_err, frame->tf_rip);
/*NOTREACHED*/
case T_PROTFLT:
case T_SEGNPFLT:
case T_ALIGNFLT:
case T_TSSFLT:
if (p == NULL)
goto we_re_toast;
/* Check for copyin/copyout fault. */
if (pcb->pcb_onfault != 0) {
error = EFAULT;
copyfault:
frame->tf_rip = (u_int64_t)pcb->pcb_onfault;
frame->tf_rax = error;
return;
}
/*
* Check for failure during return to user mode.
* We do this by looking at the address of the
* instruction that faulted.
*/
if (frame->tf_rip == (u_int64_t)doreti_iret) {
frame->tf_rip = (u_int64_t)resume_iret;
return;
}
goto we_re_toast;
case T_PROTFLT|T_USER: /* protection fault */
case T_TSSFLT|T_USER:
case T_SEGNPFLT|T_USER:
case T_STKFLT|T_USER:
case T_NMI|T_USER:
#ifdef TRAP_SIGDEBUG
printf("pid %d (%s): BUS at rip %lx addr %lx\n",
p->p_pid, p->p_comm, frame->tf_rip, rcr2());
frame_dump(frame);
#endif
sv.sival_ptr = (void *)frame->tf_rip;
KERNEL_LOCK();
trapsignal(p, SIGBUS, type & ~T_USER, BUS_OBJERR, sv);
KERNEL_UNLOCK();
goto out;
case T_ALIGNFLT|T_USER:
sv.sival_ptr = (void *)frame->tf_rip;
KERNEL_LOCK();
trapsignal(p, SIGBUS, type & ~T_USER, BUS_ADRALN, sv);
KERNEL_UNLOCK();
goto out;
case T_PRIVINFLT|T_USER: /* privileged instruction fault */
sv.sival_ptr = (void *)frame->tf_rip;
KERNEL_LOCK();
trapsignal(p, SIGILL, type & ~T_USER, ILL_PRVOPC, sv);
KERNEL_UNLOCK();
goto out;
case T_FPOPFLT|T_USER: /* coprocessor operand fault */
#ifdef TRAP_SIGDEBUG
printf("pid %d (%s): ILL at rip %lx addr %lx\n",
p->p_pid, p->p_comm, frame->tf_rip, rcr2());
frame_dump(frame);
#endif
sv.sival_ptr = (void *)frame->tf_rip;
KERNEL_LOCK();
trapsignal(p, SIGILL, type & ~T_USER, ILL_COPROC, sv);
KERNEL_UNLOCK();
goto out;
case T_ASTFLT|T_USER: /* Allow process switch */
uvmexp.softs++;
if (p->p_flag & P_OWEUPC) {
KERNEL_LOCK();
ADDUPROF(p);
KERNEL_UNLOCK();
}
/* Allow a forced task switch. */
if (curcpu()->ci_want_resched)
preempt(NULL);
goto out;
case T_BOUND|T_USER:
sv.sival_ptr = (void *)frame->tf_rip;
KERNEL_LOCK();
trapsignal(p, SIGFPE, type &~ T_USER, FPE_FLTSUB, sv);
KERNEL_UNLOCK();
goto out;
case T_OFLOW|T_USER:
sv.sival_ptr = (void *)frame->tf_rip;
KERNEL_LOCK();
trapsignal(p, SIGFPE, type &~ T_USER, FPE_INTOVF, sv);
KERNEL_UNLOCK();
goto out;
case T_DIVIDE|T_USER:
sv.sival_ptr = (void *)frame->tf_rip;
KERNEL_LOCK();
trapsignal(p, SIGFPE, type &~ T_USER, FPE_INTDIV, sv);
KERNEL_UNLOCK();
goto out;
case T_ARITHTRAP|T_USER:
case T_XMM|T_USER:
fputrap(frame);
goto out;
case T_PAGEFLT: /* allow page faults in kernel mode */
if (p == NULL)
goto we_re_toast;
cr2 = rcr2();
KERNEL_LOCK();
goto faultcommon;
case T_PAGEFLT|T_USER: { /* page fault */
vaddr_t va, fa;
struct vmspace *vm;
struct vm_map *map;
vm_prot_t ftype;
extern struct vm_map *kernel_map;
cr2 = rcr2();
KERNEL_LOCK();
faultcommon:
vm = p->p_vmspace;
if (vm == NULL)
goto we_re_toast;
fa = cr2;
va = trunc_page((vaddr_t)cr2);
/*
* It is only a kernel address space fault iff:
* 1. (type & T_USER) == 0 and
* 2. pcb_onfault not set or
* 3. pcb_onfault set but supervisor space fault
* The last can occur during an exec() copyin where the
* argument space is lazy-allocated.
*/
if (type == T_PAGEFLT && va >= VM_MIN_KERNEL_ADDRESS)
map = kernel_map;
else
map = &vm->vm_map;
if (frame->tf_err & PGEX_W)
ftype = VM_PROT_WRITE;
else if (frame->tf_err & PGEX_I)
ftype = VM_PROT_EXECUTE;
else
ftype = VM_PROT_READ;
#ifdef DIAGNOSTIC
if (map == kernel_map && va == 0) {
printf("trap: bad kernel access at %lx\n", va);
goto we_re_toast;
}
#endif
/* Fault the original page in. */
onfault = pcb->pcb_onfault;
pcb->pcb_onfault = NULL;
error = uvm_fault(map, va, frame->tf_err & PGEX_P?
VM_FAULT_PROTECT : VM_FAULT_INVALID, ftype);
pcb->pcb_onfault = onfault;
if (error == 0) {
if (map != kernel_map)
uvm_grow(p, va);
if (type == T_PAGEFLT) {
KERNEL_UNLOCK();
return;
}
KERNEL_UNLOCK();
goto out;
}
if (error == EACCES) {
error = EFAULT;
}
if (type == T_PAGEFLT) {
if (pcb->pcb_onfault != 0) {
KERNEL_UNLOCK();
goto copyfault;
}
printf("uvm_fault(%p, 0x%lx, 0, %d) -> %x\n",
map, va, ftype, error);
goto we_re_toast;
}
if (error == ENOMEM) {
printf("UVM: pid %d (%s), uid %d killed: out of swap\n",
p->p_pid, p->p_comm,
p->p_cred && p->p_ucred ?
(int)p->p_ucred->cr_uid : -1);
sv.sival_ptr = (void *)fa;
trapsignal(p, SIGKILL, T_PAGEFLT, SEGV_MAPERR, sv);
} else {
#ifdef TRAP_SIGDEBUG
printf("pid %d (%s): SEGV at rip %lx addr %lx\n",
p->p_pid, p->p_comm, frame->tf_rip, va);
frame_dump(frame);
#endif
sv.sival_ptr = (void *)fa;
trapsignal(p, SIGSEGV, T_PAGEFLT, SEGV_MAPERR, sv);
}
KERNEL_UNLOCK();
break;
}
case T_TRCTRAP:
goto we_re_toast;
case T_BPTFLT|T_USER: /* bpt instruction fault */
case T_TRCTRAP|T_USER: /* trace trap */
#ifdef MATH_EMULATE
trace:
#endif
KERNEL_LOCK();
trapsignal(p, SIGTRAP, type &~ T_USER, TRAP_BRKPT, sv);
KERNEL_UNLOCK();
break;
#if NISA > 0
case T_NMI:
#if defined(KGDB) || defined(DDB)
/* NMI can be hooked up to a pushbutton for debugging */
printf ("NMI ... going to debugger\n");
#ifdef KGDB
if (kgdb_trap(type, frame))
return;
#endif
#ifdef DDB
if (kdb_trap(type, 0, frame))
return;
#endif
#endif /* KGDB || DDB */
/* machine/parity/power fail/"kitchen sink" faults */
if (x86_nmi() != 0)
goto we_re_toast;
else
return;
#endif /* NISA > 0 */
}
if ((type & T_USER) == 0)
return;
out:
userret(p);
}
