/*
* 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);
}
/*
* 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);
}
/*
* See the comment above pool_do_bind() for the semantics of the pset_bind_*()
* functions. These must be kept in sync with cpupart_move_thread, and
* anything else that could fail a pool_pset_bind.
*
* Returns non-zero errno on failure and zero on success.
* Iff successful, cpu_lock is held on return.
*/
int
pset_bind_start(proc_t **procs, pool_t *pool)
{
cred_t *pcred;
proc_t *p, **pp;
kthread_t *t;
cpupart_t *newpp;
int ret;
extern int cpupart_movable_thread(kthread_id_t, cpupart_t *, int);
ASSERT(pool_lock_held());
ASSERT(INGLOBALZONE(curproc));
mutex_enter(&cpu_lock);
weakbinding_stop();
newpp = cpupart_find(pool->pool_pset->pset_id);
ASSERT(newpp != NULL);
if (newpp->cp_cpulist == NULL) {
weakbinding_start();
mutex_exit(&cpu_lock);
return (ENOTSUP);
}
pcred = crgetcred();
/*
* Check for the PRIV_PROC_PRIOCNTL privilege that is required
* to enter and exit scheduling classes. If other privileges
* are required by CL_ENTERCLASS/CL_CANEXIT types of routines
* in the future, this code will have to be updated.
*/
if (secpolicy_setpriority(pcred) != 0) {
weakbinding_start();
mutex_exit(&cpu_lock);
crfree(pcred);
return (EPERM);
}
for (pp = procs; (p = *pp) != NULL; pp++) {
mutex_enter(&p->p_lock);
if ((t = p->p_tlist) == NULL) {
mutex_exit(&p->p_lock);
continue;
}
/*
* Check our basic permissions to control this process.
*/
if (!prochasprocperm(p, curproc, pcred)) {
mutex_exit(&p->p_lock);
weakbinding_start();
mutex_exit(&cpu_lock);
crfree(pcred);
return (EPERM);
}
do {
/*
* Check that all threads can be moved to
* a new processor set.
*/
thread_lock(t);
ret = cpupart_movable_thread(t, newpp, 0);
thread_unlock(t);
if (ret != 0) {
mutex_exit(&p->p_lock);
weakbinding_start();
mutex_exit(&cpu_lock);
crfree(pcred);
return (ret);
}
} while ((t = t->t_forw) != p->p_tlist);
mutex_exit(&p->p_lock);
}
crfree(pcred);
return (0); /* with cpu_lock held and weakbinding stopped */
}
