/*
* Call a system call which takes a pointer to the user args struct and
* a pointer to the return values. This is a bit slower than the standard
* C arg-passing method in some cases.
*/
int64_t
syscall_ap(void)
{
uint_t error;
struct sysent *callp;
rval_t rval;
kthread_t *t = curthread;
klwp_t *lwp = ttolwp(t);
struct regs *rp = lwptoregs(lwp);
callp = LWP_GETSYSENT(lwp) + t->t_sysnum;
#if defined(__amd64)
/*
* If the arguments don't fit in registers %rdi-%r9, make sure they
* have been copied to the lwp_arg array.
*/
if (callp->sy_narg > 6 && save_syscall_args())
return ((int64_t)set_errno(EFAULT));
#endif
rval.r_val1 = 0;
rval.r_val2 = rp->r_r1;
lwp->lwp_error = 0; /* for old drivers */
error = (*(callp->sy_call))(lwp->lwp_ap, &rval);
if (error)
return ((longlong_t)set_errno(error));
return (rval.r_vals);
}
/*
* Call a system call which takes a pointer to the user args struct and
* a pointer to the return values. This is a bit slower than the standard
* C arg-passing method in some cases.
*/
int64_t
syscall_ap()
{
uint_t error;
struct sysent *callp;
rval_t rval;
klwp_t *lwp = ttolwp(curthread);
struct regs *rp = lwptoregs(lwp);
callp = LWP_GETSYSENT(lwp) + curthread->t_sysnum;
/*
* If the arguments don't fit in registers %o0 - o5, make sure they
* have been copied to the lwp_arg array.
*/
if (callp->sy_narg > 6 && save_syscall_args())
return ((int64_t)set_errno(EFAULT));
rval.r_val1 = 0;
rval.r_val2 = (int)rp->r_o1;
lwp->lwp_error = 0; /* for old drivers */
error = (*(callp->sy_call))(lwp->lwp_ap, &rval);
if (error)
return ((int64_t)set_errno(error));
return (rval.r_vals);
}
/*
* Handle indirect system calls.
* This interface should be deprecated. The library can handle
* this more efficiently, but keep this implementation for old binaries.
*
* XX64 Needs some work.
*/
int64_t
indir(int code, long a0, long a1, long a2, long a3, long a4)
{
klwp_t *lwp = ttolwp(curthread);
struct sysent *callp;
if (code <= 0 || code >= NSYSCALL)
return (nosys());
ASSERT(lwp->lwp_ap != NULL);
curthread->t_sysnum = code;
callp = LWP_GETSYSENT(lwp) + code;
/*
* Handle argument setup, unless already done in pre_syscall().
*/
if (callp->sy_narg > 5) {
if (save_syscall_args()) /* move args to LWP array */
return ((int64_t)set_errno(EFAULT));
} else if (!lwp->lwp_argsaved) {
long *ap;
ap = lwp->lwp_ap; /* args haven't been saved */
lwp->lwp_ap = ap + 1; /* advance arg pointer */
curthread->t_post_sys = 1; /* so lwp_ap will be reset */
}
return ((*callp->sy_callc)(a0, a1, a2, a3, a4, lwp->lwp_arg[5]));
}
/*
* 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;
}
}
/*
* Perform pre-system-call processing, including stopping for tracing,
* auditing, etc.
*
* This routine is called only if the t_pre_sys flag is set. Any condition
* requiring pre-syscall handling must set the t_pre_sys flag. If the
* condition is persistent, this routine will repost t_pre_sys.
*/
int
pre_syscall()
{
kthread_t *t = curthread;
unsigned code = t->t_sysnum;
klwp_t *lwp = ttolwp(t);
proc_t *p = ttoproc(t);
int repost;
t->t_pre_sys = repost = 0; /* clear pre-syscall processing flag */
ASSERT(t->t_schedflag & TS_DONT_SWAP);
#if defined(DEBUG)
/*
* On the i386 kernel, lwp_ap points at the piece of the thread
* stack that we copy the users arguments into.
*
* On the amd64 kernel, the syscall arguments in the rdi..r9
* registers should be pointed at by lwp_ap. If the args need to
* be copied so that those registers can be changed without losing
* the ability to get the args for /proc, they can be saved by
* save_syscall_args(), and lwp_ap will be restored by post_syscall().
*/
if (lwp_getdatamodel(lwp) == DATAMODEL_NATIVE) {
#if defined(_LP64)
ASSERT(lwp->lwp_ap == (long *)&lwptoregs(lwp)->r_rdi);
} else {
#endif
ASSERT((caddr_t)lwp->lwp_ap > t->t_stkbase &&
(caddr_t)lwp->lwp_ap < t->t_stk);
}
#endif /* DEBUG */
/*
* Make sure the thread is holding the latest credentials for the
* process. The credentials in the process right now apply to this
* thread for the entire system call.
*/
if (t->t_cred != p->p_cred) {
cred_t *oldcred = t->t_cred;
/*
* DTrace accesses t_cred in probe context. t_cred must
* always be either NULL, or point to a valid, allocated cred
* structure.
*/
t->t_cred = crgetcred();
crfree(oldcred);
}
/*
* From the proc(4) manual page:
* When entry to a system call is being traced, the traced process
* stops after having begun the call to the system but before the
* system call arguments have been fetched from the process.
*/
if (PTOU(p)->u_systrap) {
if (prismember(&PTOU(p)->u_entrymask, code)) {
mutex_enter(&p->p_lock);
/*
* Recheck stop condition, now that lock is held.
*/
if (PTOU(p)->u_systrap &&
prismember(&PTOU(p)->u_entrymask, code)) {
stop(PR_SYSENTRY, code);
/*
* /proc may have modified syscall args,
* either in regs for amd64 or on ustack
* for ia32. Either way, arrange to
* copy them again, both for the syscall
* handler and for other consumers in
* post_syscall (like audit). Here, we
* only do amd64, and just set lwp_ap
* back to the kernel-entry stack copy;
* the syscall ml code redoes
* move-from-regs to set up for the
* syscall handler after we return. For
* ia32, save_syscall_args() below makes
* an lwp_ap-accessible copy.
*/
#if defined(_LP64)
if (lwp_getdatamodel(lwp) == DATAMODEL_NATIVE) {
lwp->lwp_argsaved = 0;
lwp->lwp_ap =
(long *)&lwptoregs(lwp)->r_rdi;
}
#endif
}
mutex_exit(&p->p_lock);
}
repost = 1;
}
/*
* ia32 kernel, or ia32 proc on amd64 kernel: keep args in
* lwp_arg for post-syscall processing, regardless of whether
* they might have been changed in /proc above.
*/
#if defined(_LP64)
if (lwp_getdatamodel(lwp) != DATAMODEL_NATIVE)
#endif
(void) save_syscall_args();
if (lwp->lwp_sysabort) {
/*
* lwp_sysabort may have been set via /proc while the process
* was stopped on PR_SYSENTRY. If so, abort the system call.
* Override any error from the copyin() of the arguments.
*/
lwp->lwp_sysabort = 0;
(void) set_errno(EINTR); /* forces post_sys */
t->t_pre_sys = 1; /* repost anyway */
return (1); /* don't do system call, return EINTR */
}
#ifdef C2_AUDIT
if (audit_active) { /* begin auditing for this syscall */
int error;
if (error = audit_start(T_SYSCALL, code, 0, lwp)) {
t->t_pre_sys = 1; /* repost anyway */
(void) set_errno(error);
return (1);
}
repost = 1;
}
#endif /* C2_AUDIT */
#ifndef NPROBE
/* Kernel probe */
if (tnf_tracing_active) {
TNF_PROBE_1(syscall_start, "syscall thread", /* CSTYLED */,
tnf_sysnum, sysnum, t->t_sysnum);
t->t_post_sys = 1; /* make sure post_syscall runs */
repost = 1;
}
#endif /* NPROBE */
#ifdef SYSCALLTRACE
if (syscalltrace) {
int i;
long *ap;
char *cp;
char *sysname;
struct sysent *callp;
if (code >= NSYSCALL)
callp = &nosys_ent; /* nosys has no args */
else
callp = LWP_GETSYSENT(lwp) + code;
(void) save_syscall_args();
mutex_enter(&systrace_lock);
printf("%d: ", p->p_pid);
if (code >= NSYSCALL)
printf("0x%x", code);
else {
sysname = mod_getsysname(code);
printf("%s[0x%x/0x%p]", sysname == NULL ? "NULL" :
sysname, code, callp->sy_callc);
}
cp = "(";
for (i = 0, ap = lwp->lwp_ap; i < callp->sy_narg; i++, ap++) {
printf("%s%lx", cp, *ap);
cp = ", ";
}
if (i)
printf(")");
printf(" %s id=0x%p\n", PTOU(p)->u_comm, curthread);
mutex_exit(&systrace_lock);
}
#endif /* SYSCALLTRACE */
/*
* If there was a continuing reason for pre-syscall processing,
* set the t_pre_sys flag for the next system call.
*/
if (repost)
t->t_pre_sys = 1;
lwp->lwp_error = 0; /* for old drivers */
lwp->lwp_badpriv = PRIV_NONE;
return (0);
}
/*
* System call to create an lwp.
*
* Notes on the LWP_DETACHED and LWP_DAEMON flags:
*
* A detached lwp (LWP_DETACHED) cannot be the specific target of
* lwp_wait() (it is not joinable), but lwp_wait(0, ...) is required
* to sleep until all non-daemon detached lwps have terminated before
* returning EDEADLK because a detached lwp might create a non-detached lwp
* that could then be returned by lwp_wait(0, ...). See also lwp_detach().
*
* A daemon lwp (LWP_DAEMON) is a detached lwp that has the additional
* property that it does not affect the termination condition of the
* process: The last non-daemon lwp to call lwp_exit() causes the process
* to exit and lwp_wait(0, ...) does not sleep waiting for daemon lwps
* to terminate. See the block comment before lwp_wait().
*/
int
syslwp_create(ucontext_t *ucp, int flags, id_t *new_lwp)
{
klwp_t *lwp;
proc_t *p = ttoproc(curthread);
kthread_t *t;
ucontext_t uc;
#ifdef _SYSCALL32_IMPL
ucontext32_t uc32;
#endif /* _SYSCALL32_IMPL */
k_sigset_t sigmask;
int tid;
model_t model = get_udatamodel();
uintptr_t thrptr = 0;
if (flags & ~(LWP_DAEMON|LWP_DETACHED|LWP_SUSPENDED))
return (set_errno(EINVAL));
/*
* lwp_create() is disallowed for the /proc agent lwp.
*/
if (curthread == p->p_agenttp)
return (set_errno(ENOTSUP));
if (model == DATAMODEL_NATIVE) {
if (copyin(ucp, &uc, sizeof (ucontext_t)))
return (set_errno(EFAULT));
sigutok(&uc.uc_sigmask, &sigmask);
#if defined(__i386)
/*
* libc stashed thrptr into unused kernel %sp.
* See setup_context() in libc.
*/
thrptr = (uint32_t)uc.uc_mcontext.gregs[ESP];
#endif
}
#ifdef _SYSCALL32_IMPL
else {
if (copyin(ucp, &uc32, sizeof (ucontext32_t)))
return (set_errno(EFAULT));
sigutok(&uc32.uc_sigmask, &sigmask);
#if defined(__sparc)
ucontext_32ton(&uc32, &uc, NULL, NULL);
#else /* __amd64 */
ucontext_32ton(&uc32, &uc);
/*
* libc stashed thrptr into unused kernel %sp.
* See setup_context() in libc.
*/
thrptr = (uint32_t)uc32.uc_mcontext.gregs[ESP];
#endif
}
#endif /* _SYSCALL32_IMPL */
(void) save_syscall_args(); /* save args for tracing first */
mutex_enter(&curproc->p_lock);
pool_barrier_enter();
mutex_exit(&curproc->p_lock);
lwp = lwp_create(lwp_rtt, NULL, NULL, curproc, TS_STOPPED,
curthread->t_pri, &sigmask, curthread->t_cid, 0);
mutex_enter(&curproc->p_lock);
pool_barrier_exit();
mutex_exit(&curproc->p_lock);
if (lwp == NULL)
return (set_errno(EAGAIN));
lwp_load(lwp, uc.uc_mcontext.gregs, thrptr);
t = lwptot(lwp);
/*
* Copy the new lwp's lwpid into the caller's specified buffer.
*/
if (new_lwp && copyout(&t->t_tid, new_lwp, sizeof (id_t))) {
/*
* caller's buffer is not writable, return
* EFAULT, and terminate new lwp.
*/
mutex_enter(&p->p_lock);
t->t_proc_flag |= TP_EXITLWP;
t->t_sig_check = 1;
t->t_sysnum = 0;
t->t_proc_flag &= ~TP_HOLDLWP;
lwp_create_done(t);
mutex_exit(&p->p_lock);
return (set_errno(EFAULT));
}
/*
* clone callers context, if any. must be invoked
* while -not- holding p_lock.
*/
if (curthread->t_ctx)
lwp_createctx(curthread, t);
/*
* copy current contract templates
*/
lwp_ctmpl_copy(lwp, ttolwp(curthread));
mutex_enter(&p->p_lock);
/*
* Copy the syscall arguments to the new lwp's arg area
* for the benefit of debuggers.
*/
t->t_sysnum = SYS_lwp_create;
lwp->lwp_ap = lwp->lwp_arg;
lwp->lwp_arg[0] = (long)ucp;
lwp->lwp_arg[1] = (long)flags;
lwp->lwp_arg[2] = (long)new_lwp;
lwp->lwp_argsaved = 1;
if (!(flags & (LWP_DETACHED|LWP_DAEMON)))
t->t_proc_flag |= TP_TWAIT;
if (flags & LWP_DAEMON) {
t->t_proc_flag |= TP_DAEMON;
p->p_lwpdaemon++;
}
tid = (int)t->t_tid; /* for /proc debuggers */
/*
* We now set the newly-created lwp running.
* If it is being created as LWP_SUSPENDED, we leave its
* TP_HOLDLWP flag set so it will stop in system call exit.
*/
if (!(flags & LWP_SUSPENDED))
t->t_proc_flag &= ~TP_HOLDLWP;
lwp_create_done(t);
mutex_exit(&p->p_lock);
return (tid);
}
/*
* Perform pre-system-call processing, including stopping for tracing,
* auditing, microstate-accounting, etc.
*
* This routine is called only if the t_pre_sys flag is set. Any condition
* requiring pre-syscall handling must set the t_pre_sys flag. If the
* condition is persistent, this routine will repost t_pre_sys.
*/
int
pre_syscall(int arg0)
{
unsigned int code;
kthread_t *t = curthread;
proc_t *p = ttoproc(t);
klwp_t *lwp = ttolwp(t);
struct regs *rp = lwptoregs(lwp);
int repost;
t->t_pre_sys = repost = 0; /* clear pre-syscall processing flag */
ASSERT(t->t_schedflag & TS_DONT_SWAP);
syscall_mstate(LMS_USER, LMS_SYSTEM);
/*
* The syscall arguments in the out registers should be pointed to
* by lwp_ap. If the args need to be copied so that the outs can
* be changed without losing the ability to get the args for /proc,
* they can be saved by save_syscall_args(), and lwp_ap will be
* restored by post_syscall().
*/
ASSERT(lwp->lwp_ap == (long *)&rp->r_o0);
/*
* Make sure the thread is holding the latest credentials for the
* process. The credentials in the process right now apply to this
* thread for the entire system call.
*/
if (t->t_cred != p->p_cred) {
cred_t *oldcred = t->t_cred;
/*
* DTrace accesses t_cred in probe context. t_cred must
* always be either NULL, or point to a valid, allocated cred
* structure.
*/
t->t_cred = crgetcred();
crfree(oldcred);
}
/*
* Undo special arrangements to single-step the lwp
* so that a debugger will see valid register contents.
* Also so that the pc is valid for syncfpu().
* Also so that a syscall like exec() can be stepped.
*/
if (lwp->lwp_pcb.pcb_step != STEP_NONE) {
(void) prundostep();
repost = 1;
}
/*
* Check for indirect system call in case we stop for tracing.
* Don't allow multiple indirection.
*/
code = t->t_sysnum;
if (code == 0 && arg0 != 0) { /* indirect syscall */
code = arg0;
t->t_sysnum = arg0;
}
/*
* From the proc(4) manual page:
* When entry to a system call is being traced, the traced process
* stops after having begun the call to the system but before the
* system call arguments have been fetched from the process.
* If proc changes the args we must refetch them after starting.
*/
if (PTOU(p)->u_systrap) {
if (prismember(&PTOU(p)->u_entrymask, code)) {
/*
* Recheck stop condition, now that lock is held.
*/
mutex_enter(&p->p_lock);
if (PTOU(p)->u_systrap &&
prismember(&PTOU(p)->u_entrymask, code)) {
stop(PR_SYSENTRY, code);
/*
* Must refetch args since they were
* possibly modified by /proc. Indicate
* that the valid copy is in the
* registers.
*/
lwp->lwp_argsaved = 0;
lwp->lwp_ap = (long *)&rp->r_o0;
}
mutex_exit(&p->p_lock);
}
repost = 1;
}
if (lwp->lwp_sysabort) {
/*
* lwp_sysabort may have been set via /proc while the process
* was stopped on PR_SYSENTRY. If so, abort the system call.
* Override any error from the copyin() of the arguments.
*/
lwp->lwp_sysabort = 0;
(void) set_errno(EINTR); /* sets post-sys processing */
t->t_pre_sys = 1; /* repost anyway */
return (1); /* don't do system call, return EINTR */
}
#ifdef C2_AUDIT
if (audit_active) { /* begin auditing for this syscall */
int error;
if (error = audit_start(T_SYSCALL, code, 0, lwp)) {
t->t_pre_sys = 1; /* repost anyway */
lwp->lwp_error = 0; /* for old drivers */
return (error);
}
repost = 1;
}
#endif /* C2_AUDIT */
#ifndef NPROBE
/* Kernel probe */
if (tnf_tracing_active) {
TNF_PROBE_1(syscall_start, "syscall thread", /* CSTYLED */,
tnf_sysnum, sysnum, t->t_sysnum);
t->t_post_sys = 1; /* make sure post_syscall runs */
repost = 1;
}
#endif /* NPROBE */
#ifdef SYSCALLTRACE
if (syscalltrace) {
int i;
long *ap;
char *cp;
char *sysname;
struct sysent *callp;
if (code >= NSYSCALL)
callp = &nosys_ent; /* nosys has no args */
else
callp = LWP_GETSYSENT(lwp) + code;
(void) save_syscall_args();
mutex_enter(&systrace_lock);
printf("%d: ", p->p_pid);
if (code >= NSYSCALL)
printf("0x%x", code);
else {
sysname = mod_getsysname(code);
printf("%s[0x%x]", sysname == NULL ? "NULL" :
sysname, code);
}
cp = "(";
for (i = 0, ap = lwp->lwp_ap; i < callp->sy_narg; i++, ap++) {
printf("%s%lx", cp, *ap);
cp = ", ";
}
if (i)
printf(")");
printf(" %s id=0x%p\n", PTOU(p)->u_comm, curthread);
mutex_exit(&systrace_lock);
}
#endif /* SYSCALLTRACE */
/*
* If there was a continuing reason for pre-syscall processing,
* set the t_pre_sys flag for the next system call.
*/
if (repost)
t->t_pre_sys = 1;
lwp->lwp_error = 0; /* for old drivers */
lwp->lwp_badpriv = PRIV_NONE; /* for privilege tracing */
return (0);
}
/*
* Clear registers on exec(2).
*/
void
setregs(uarg_t *args)
{
struct regs *rp;
kthread_t *t = curthread;
klwp_t *lwp = ttolwp(t);
pcb_t *pcb = &lwp->lwp_pcb;
greg_t sp;
/*
* Initialize user registers
*/
(void) save_syscall_args(); /* copy args from registers first */
rp = lwptoregs(lwp);
sp = rp->r_sp;
bzero(rp, sizeof (*rp));
rp->r_ss = UDS_SEL;
rp->r_sp = sp;
rp->r_pc = args->entry;
rp->r_ps = PSL_USER;
#if defined(__amd64)
pcb->pcb_fs = pcb->pcb_gs = 0;
pcb->pcb_fsbase = pcb->pcb_gsbase = 0;
if (ttoproc(t)->p_model == DATAMODEL_NATIVE) {
rp->r_cs = UCS_SEL;
/*
* Only allow 64-bit user code descriptor to be present.
*/
gdt_ucode_model(DATAMODEL_NATIVE);
/*
* Arrange that the virtualized %fs and %gs GDT descriptors
* have a well-defined initial state (present, ring 3
* and of type data).
*/
pcb->pcb_fsdesc = pcb->pcb_gsdesc = zero_udesc;
/*
* thrptr is either NULL or a value used by DTrace.
* 64-bit processes use %fs as their "thread" register.
*/
if (args->thrptr)
(void) lwp_setprivate(lwp, _LWP_FSBASE, args->thrptr);
} else {
rp->r_cs = U32CS_SEL;
rp->r_ds = rp->r_es = UDS_SEL;
/*
* only allow 32-bit user code selector to be present.
*/
gdt_ucode_model(DATAMODEL_ILP32);
pcb->pcb_fsdesc = pcb->pcb_gsdesc = zero_u32desc;
/*
* thrptr is either NULL or a value used by DTrace.
* 32-bit processes use %gs as their "thread" register.
*/
if (args->thrptr)
(void) lwp_setprivate(lwp, _LWP_GSBASE, args->thrptr);
}
pcb->pcb_ds = rp->r_ds;
pcb->pcb_es = rp->r_es;
pcb->pcb_rupdate = 1;
#elif defined(__i386)
rp->r_cs = UCS_SEL;
rp->r_ds = rp->r_es = UDS_SEL;
/*
* Arrange that the virtualized %fs and %gs GDT descriptors
* have a well-defined initial state (present, ring 3
* and of type data).
*/
pcb->pcb_fsdesc = pcb->pcb_gsdesc = zero_udesc;
/*
* For %gs we need to reset LWP_GSBASE in pcb and the
* per-cpu GDT descriptor. thrptr is either NULL
* or a value used by DTrace.
*/
if (args->thrptr)
(void) lwp_setprivate(lwp, _LWP_GSBASE, args->thrptr);
#endif
lwp->lwp_eosys = JUSTRETURN;
t->t_post_sys = 1;
/*
* Here we initialize minimal fpu state.
* The rest is done at the first floating
* point instruction that a process executes.
*/
pcb->pcb_fpu.fpu_flags = 0;
/*
* Add the lwp context handlers that virtualize segment registers,
* and/or system call stacks etc.
*/
lwp_installctx(lwp);
}
