static int
setpmask(caddr_t data)
{
STRUCT_DECL(auditpinfo, apinfo);
proc_t *proc;
cred_t *newcred;
auditinfo_addr_t *ainfo;
struct p_audit_data *pad;
model_t model;
model = get_udatamodel();
STRUCT_INIT(apinfo, model);
if (copyin(data, STRUCT_BUF(apinfo), STRUCT_SIZE(apinfo)))
return (EFAULT);
mutex_enter(&pidlock);
if ((proc = prfind(STRUCT_FGET(apinfo, ap_pid))) == NULL) {
mutex_exit(&pidlock);
return (ESRCH);
}
mutex_enter(&proc->p_lock); /* so process doesn't go away */
mutex_exit(&pidlock);
newcred = cralloc();
if ((ainfo = crgetauinfo_modifiable(newcred)) == NULL) {
mutex_exit(&proc->p_lock);
crfree(newcred);
return (EINVAL);
}
mutex_enter(&proc->p_crlock);
crcopy_to(proc->p_cred, newcred);
proc->p_cred = newcred;
ainfo->ai_mask = STRUCT_FGET(apinfo, ap_mask);
/*
* Unlock. No need to broadcast changes via set_proc_pre_sys(),
* since t_pre_sys is ALWAYS on when audit is enabled... due to
* syscall auditing.
*/
crfree(newcred);
mutex_exit(&proc->p_crlock);
/* Reset flag for any previous pending mask change; this supercedes */
pad = P2A(proc);
ASSERT(pad != NULL);
mutex_enter(&(pad->pad_lock));
pad->pad_flags &= ~PAD_SETMASK;
mutex_exit(&(pad->pad_lock));
mutex_exit(&proc->p_lock);
return (0);
}
/*
* Bind the lwp:id of process:pid to processor set: pset
*/
static int
pset_bind_lwp(psetid_t pset, id_t id, pid_t pid, psetid_t *opset)
{
kthread_t *tp;
proc_t *pp;
psetid_t oldpset;
void *projbuf, *zonebuf;
int error = 0;
pool_lock();
mutex_enter(&cpu_lock);
projbuf = fss_allocbuf(FSS_NPROJ_BUF, FSS_ALLOC_PROJ);
zonebuf = fss_allocbuf(FSS_NPROJ_BUF, FSS_ALLOC_ZONE);
mutex_enter(&pidlock);
if ((pid == P_MYID && id == P_MYID) ||
(pid == curproc->p_pid && id == P_MYID)) {
pp = curproc;
tp = curthread;
mutex_enter(&pp->p_lock);
} else {
if (pid == P_MYID) {
pp = curproc;
} else if ((pp = prfind(pid)) == NULL) {
error = ESRCH;
goto err;
}
if (pp != curproc && id == P_MYID) {
error = EINVAL;
goto err;
}
mutex_enter(&pp->p_lock);
if ((tp = idtot(pp, id)) == NULL) {
mutex_exit(&pp->p_lock);
error = ESRCH;
goto err;
}
}
error = pset_bind_thread(tp, pset, &oldpset, projbuf, zonebuf);
mutex_exit(&pp->p_lock);
err:
mutex_exit(&pidlock);
fss_freebuf(projbuf, FSS_ALLOC_PROJ);
fss_freebuf(zonebuf, FSS_ALLOC_ZONE);
mutex_exit(&cpu_lock);
pool_unlock();
if (opset != NULL) {
if (copyout(&oldpset, opset, sizeof (psetid_t)) != 0)
return (set_errno(EFAULT));
}
if (error != 0)
return (set_errno(error));
return (0);
}
static int
putacct(idtype_t idtype, id_t id, void *buf, size_t bufsize, int flags)
{
int error;
taskid_t tkid;
proc_t *p;
task_t *tk;
void *kbuf;
struct exacct_globals *acg;
if (bufsize == 0 || bufsize > EXACCT_MAX_BUFSIZE)
return (set_errno(EINVAL));
kbuf = kmem_alloc(bufsize, KM_SLEEP);
if (copyin(buf, kbuf, bufsize) != 0) {
error = EFAULT;
goto out;
}
acg = zone_getspecific(exacct_zone_key, curproc->p_zone);
switch (idtype) {
case P_PID:
mutex_enter(&pidlock);
if ((p = prfind(id)) == NULL) {
mutex_exit(&pidlock);
error = ESRCH;
} else {
zone_t *zone = p->p_zone;
tkid = p->p_task->tk_tkid;
zone_hold(zone);
mutex_exit(&pidlock);
error = exacct_tag_proc(&acg->ac_proc, id, tkid, kbuf,
bufsize, flags, zone->zone_nodename);
zone_rele(zone);
}
break;
case P_TASKID:
if ((tk = task_hold_by_id(id)) != NULL) {
error = exacct_tag_task(&acg->ac_task, tk, kbuf,
bufsize, flags);
task_rele(tk);
} else {
error = ESRCH;
}
break;
default:
error = EINVAL;
break;
}
out:
kmem_free(kbuf, bufsize);
return (error == 0 ? error : set_errno(error));
}
/* ARGSUSED */
int
sys_setpgid(struct proc *curp, void *v, register_t *retval)
{
struct sys_setpgid_args /* {
syscallarg(pid_t) pid;
syscallarg(int) pgid;
} */ *uap = v;
struct process *curpr = curp->p_p;
struct process *targpr; /* target process */
struct pgrp *pgrp, *newpgrp; /* target pgrp */
pid_t pid;
int pgid, error;
pid = SCARG(uap, pid);
pgid = SCARG(uap, pgid);
if (pgid < 0)
return (EINVAL);
newpgrp = pool_get(&pgrp_pool, PR_WAITOK);
if (pid != 0 && pid != curpr->ps_pid) {
if ((targpr = prfind(pid)) == 0 || !inferior(targpr, curpr)) {
error = ESRCH;
goto out;
}
if (targpr->ps_session != curpr->ps_session) {
error = EPERM;
goto out;
}
if (targpr->ps_flags & PS_EXEC) {
error = EACCES;
goto out;
}
} else
targpr = curpr;
if (SESS_LEADER(targpr)) {
error = EPERM;
goto out;
}
if (pgid == 0)
pgid = targpr->ps_pid;
else if (pgid != targpr->ps_pid)
if ((pgrp = pgfind(pgid)) == 0 ||
pgrp->pg_session != curpr->ps_session) {
error = EPERM;
goto out;
}
return (enterpgrp(targpr, pgid, newpgrp, NULL));
out:
pool_put(&pgrp_pool, newpgrp);
return (error);
}
/*
* lxpr_lock():
*
* Lookup process from pid and return with p_plock and P_PR_LOCK held.
*/
proc_t *
lxpr_lock(pid_t pid)
{
proc_t *p;
kmutex_t *mp;
ASSERT(!MUTEX_HELD(&pidlock));
for (;;) {
mutex_enter(&pidlock);
/*
* If the pid is 1, we really want the zone's init process
*/
p = prfind((pid == 1) ?
curproc->p_zone->zone_proc_initpid : pid);
if (p == NULL || p->p_stat == SIDL) {
mutex_exit(&pidlock);
return (NULL);
}
/*
* p_lock is persistent, but p itself is not -- it could
* vanish during cv_wait(). Load p->p_lock now so we can
* drop it after cv_wait() without referencing p.
*/
mp = &p->p_lock;
mutex_enter(mp);
mutex_exit(&pidlock);
if (p->p_flag & SEXITING) {
/*
* This process is exiting -- let it go.
*/
mutex_exit(mp);
return (NULL);
}
if (!(p->p_proc_flag & P_PR_LOCK))
break;
cv_wait(&pr_pid_cv[p->p_slot], mp);
mutex_exit(mp);
}
p->p_proc_flag |= P_PR_LOCK;
THREAD_KPRI_REQUEST();
return (p);
}
static int
sigqkill(pid_t pid, sigsend_t *sigsend)
{
proc_t *p;
int error;
if ((uint_t)sigsend->sig >= NSIG)
return (EINVAL);
if (pid == -1) {
procset_t set;
setprocset(&set, POP_AND, P_ALL, P_MYID, P_ALL, P_MYID);
error = sigsendset(&set, sigsend);
} else if (pid > 0) {
mutex_enter(&pidlock);
if ((p = prfind(pid)) == NULL || p->p_stat == SIDL)
error = ESRCH;
else {
error = sigsendproc(p, sigsend);
if (error == 0 && sigsend->perm == 0)
error = EPERM;
}
mutex_exit(&pidlock);
} else {
int nfound = 0;
pid_t pgid;
if (pid == 0)
pgid = ttoproc(curthread)->p_pgrp;
else
pgid = -pid;
error = 0;
mutex_enter(&pidlock);
for (p = pgfind(pgid); p && !error; p = p->p_pglink) {
if (p->p_stat != SIDL) {
nfound++;
error = sigsendproc(p, sigsend);
}
}
mutex_exit(&pidlock);
if (nfound == 0)
error = ESRCH;
else if (error == 0 && sigsend->perm == 0)
error = EPERM;
}
return (error);
}
static void
xen_shutdown(void *arg)
{
int cmd = (uintptr_t)arg;
proc_t *initpp;
ASSERT(cmd > SHUTDOWN_INVALID && cmd < SHUTDOWN_MAX);
if (cmd == SHUTDOWN_SUSPEND) {
xen_suspend_domain();
return;
}
switch (cmd) {
case SHUTDOWN_POWEROFF:
force_shutdown_method = AD_POWEROFF;
break;
case SHUTDOWN_HALT:
force_shutdown_method = AD_HALT;
break;
case SHUTDOWN_REBOOT:
force_shutdown_method = AD_BOOT;
break;
}
/*
* If we're still booting and init(1) isn't set up yet, simply halt.
*/
mutex_enter(&pidlock);
initpp = prfind(P_INITPID);
mutex_exit(&pidlock);
if (initpp == NULL) {
extern void halt(char *);
halt("Power off the System"); /* just in case */
}
/*
* else, graceful shutdown with inittab and all getting involved
*/
psignal(initpp, SIGPWR);
(void) timeout(xen_dirty_shutdown, arg,
SHUTDOWN_TIMEOUT_SECS * drv_usectohz(MICROSEC));
}
/*
* SysVR.4 compatible getpgid()
*/
pid_t
sys_getpgid(struct proc *curp, void *v, register_t *retval)
{
struct sys_getpgid_args /* {
syscallarg(pid_t) pid;
} */ *uap = v;
struct process *targpr = curp->p_p;
if (SCARG(uap, pid) == 0 || SCARG(uap, pid) == targpr->ps_pid)
goto found;
if ((targpr = prfind(SCARG(uap, pid))) == NULL)
return (ESRCH);
if (targpr->ps_session != curp->p_p->ps_session)
return (EPERM);
found:
*retval = targpr->ps_pgid;
return (0);
}
static int
getacct_proc(ac_info_t *ac_proc, pid_t pid, void *buf, size_t bufsize,
size_t *sizep)
{
proc_t *p;
proc_usage_t *pu;
ulong_t mask[AC_MASK_SZ];
ulong_t *ac_mask = &mask[0];
int error;
mutex_enter(&ac_proc->ac_lock);
if (ac_proc->ac_state == AC_OFF) {
mutex_exit(&ac_proc->ac_lock);
return (ENOTACTIVE);
}
bt_copy(&ac_proc->ac_mask[0], ac_mask, AC_MASK_SZ);
mutex_exit(&ac_proc->ac_lock);
pu = kmem_zalloc(sizeof (proc_usage_t), KM_SLEEP);
pu->pu_command = kmem_zalloc(MAXCOMLEN + 1, KM_SLEEP);
mutex_enter(&pidlock);
if ((p = prfind(pid)) == NULL) {
mutex_exit(&pidlock);
kmem_free(pu->pu_command, MAXCOMLEN + 1);
kmem_free(pu, sizeof (proc_usage_t));
return (ESRCH);
}
mutex_enter(&p->p_lock);
mutex_exit(&pidlock);
exacct_calculate_proc_usage(p, pu, ac_mask, EW_PARTIAL, 0);
mutex_exit(&p->p_lock);
error = exacct_assemble_proc_usage(ac_proc, pu,
getacct_callback, buf, bufsize, sizep, EW_PARTIAL);
kmem_free(pu->pu_command, MAXCOMLEN + 1);
kmem_free(pu, sizeof (proc_usage_t));
return (error);
}
static int
pset_bind(psetid_t pset, idtype_t idtype, id_t id, psetid_t *opset)
{
kthread_t *tp;
proc_t *pp;
task_t *tk;
kproject_t *kpj;
contract_t *ct;
zone_t *zptr;
psetid_t oldpset;
int error = 0;
void *projbuf, *zonebuf;
pool_lock();
if ((pset != PS_QUERY) && (pset != PS_SOFT) &&
(pset != PS_HARD) && (pset != PS_QUERY_TYPE)) {
/*
* Check if the set actually exists before checking
* permissions. This is the historical error
* precedence. Note that if pset was PS_MYID, the
* cpupart_get_cpus call will change it to the
* processor set id of the caller (or PS_NONE if the
* caller is not bound to a processor set).
*/
if (pool_state == POOL_ENABLED) {
pool_unlock();
return (set_errno(ENOTSUP));
}
if (cpupart_get_cpus(&pset, NULL, NULL) != 0) {
pool_unlock();
return (set_errno(EINVAL));
} else if (pset != PS_NONE && secpolicy_pset(CRED()) != 0) {
pool_unlock();
return (set_errno(EPERM));
}
}
/*
* Pre-allocate enough buffers for FSS for all active projects
* and for all active zones on the system. Unused buffers will
* be freed later by fss_freebuf().
*/
mutex_enter(&cpu_lock);
projbuf = fss_allocbuf(FSS_NPROJ_BUF, FSS_ALLOC_PROJ);
zonebuf = fss_allocbuf(FSS_NPROJ_BUF, FSS_ALLOC_ZONE);
switch (idtype) {
case P_LWPID:
pp = curproc;
mutex_enter(&pidlock);
mutex_enter(&pp->p_lock);
if (id == P_MYID) {
tp = curthread;
} else {
if ((tp = idtot(pp, id)) == NULL) {
mutex_exit(&pp->p_lock);
mutex_exit(&pidlock);
error = ESRCH;
break;
}
}
error = pset_bind_thread(tp, pset, &oldpset, projbuf, zonebuf);
mutex_exit(&pp->p_lock);
mutex_exit(&pidlock);
break;
case P_PID:
mutex_enter(&pidlock);
if (id == P_MYID) {
pp = curproc;
} else if ((pp = prfind(id)) == NULL) {
mutex_exit(&pidlock);
error = ESRCH;
break;
}
error = pset_bind_process(pp, pset, &oldpset, projbuf, zonebuf);
mutex_exit(&pidlock);
break;
case P_TASKID:
mutex_enter(&pidlock);
if (id == P_MYID)
id = curproc->p_task->tk_tkid;
if ((tk = task_hold_by_id(id)) == NULL) {
mutex_exit(&pidlock);
error = ESRCH;
break;
}
error = pset_bind_task(tk, pset, &oldpset, projbuf, zonebuf);
mutex_exit(&pidlock);
task_rele(tk);
break;
case P_PROJID:
pp = curproc;
if (id == P_MYID)
id = curprojid();
if ((kpj = project_hold_by_id(id, pp->p_zone,
PROJECT_HOLD_FIND)) == NULL) {
error = ESRCH;
break;
}
mutex_enter(&pidlock);
error = pset_bind_project(kpj, pset, &oldpset, projbuf,
zonebuf);
mutex_exit(&pidlock);
project_rele(kpj);
break;
case P_ZONEID:
if (id == P_MYID)
id = getzoneid();
if ((zptr = zone_find_by_id(id)) == NULL) {
error = ESRCH;
break;
}
mutex_enter(&pidlock);
error = pset_bind_zone(zptr, pset, &oldpset, projbuf, zonebuf);
mutex_exit(&pidlock);
zone_rele(zptr);
break;
case P_CTID:
if (id == P_MYID)
id = PRCTID(curproc);
if ((ct = contract_type_ptr(process_type, id,
curproc->p_zone->zone_uniqid)) == NULL) {
error = ESRCH;
break;
}
mutex_enter(&pidlock);
error = pset_bind_contract(ct->ct_data, pset, &oldpset, projbuf,
zonebuf);
mutex_exit(&pidlock);
contract_rele(ct);
break;
case P_PSETID:
if (id == P_MYID || pset != PS_NONE || !INGLOBALZONE(curproc)) {
error = EINVAL;
break;
}
error = pset_unbind(id, projbuf, zonebuf, idtype);
break;
case P_ALL:
if (id == P_MYID || pset != PS_NONE || !INGLOBALZONE(curproc)) {
error = EINVAL;
break;
}
error = pset_unbind(PS_NONE, projbuf, zonebuf, idtype);
break;
default:
error = EINVAL;
break;
}
fss_freebuf(projbuf, FSS_ALLOC_PROJ);
fss_freebuf(zonebuf, FSS_ALLOC_ZONE);
mutex_exit(&cpu_lock);
pool_unlock();
if (error != 0)
return (set_errno(error));
if (opset != NULL) {
if (copyout(&oldpset, opset, sizeof (psetid_t)) != 0)
return (set_errno(EFAULT));
}
return (0);
}
int
corectl(int subcode, uintptr_t arg1, uintptr_t arg2, uintptr_t arg3)
{
int error = 0;
proc_t *p;
refstr_t *rp;
size_t size;
char *path;
core_content_t content = CC_CONTENT_INVALID;
struct core_globals *cg;
zone_t *zone = curproc->p_zone;
cg = zone_getspecific(core_zone_key, zone);
ASSERT(cg != NULL);
switch (subcode) {
case CC_SET_OPTIONS:
if ((error = secpolicy_coreadm(CRED())) == 0) {
if (arg1 & ~CC_OPTIONS)
error = EINVAL;
else
cg->core_options = (uint32_t)arg1;
}
break;
case CC_GET_OPTIONS:
return (cg->core_options);
case CC_GET_GLOBAL_PATH:
case CC_GET_DEFAULT_PATH:
case CC_GET_PROCESS_PATH:
if (subcode == CC_GET_GLOBAL_PATH) {
mutex_enter(&cg->core_lock);
if ((rp = cg->core_file) != NULL)
refstr_hold(rp);
mutex_exit(&cg->core_lock);
} else if (subcode == CC_GET_DEFAULT_PATH) {
rp = corectl_path_value(cg->core_default_path);
} else {
rp = NULL;
mutex_enter(&pidlock);
if ((p = prfind((pid_t)arg3)) == NULL ||
p->p_stat == SIDL) {
mutex_exit(&pidlock);
error = ESRCH;
} else {
mutex_enter(&p->p_lock);
mutex_exit(&pidlock);
mutex_enter(&p->p_crlock);
if (!hasprocperm(p->p_cred, CRED()))
error = EPERM;
else if (p->p_corefile != NULL)
rp = corectl_path_value(p->p_corefile);
mutex_exit(&p->p_crlock);
mutex_exit(&p->p_lock);
}
}
if (rp == NULL) {
if (error == 0 && suword8((void *)arg1, 0))
error = EFAULT;
} else {
error = copyoutstr(refstr_value(rp), (char *)arg1,
(size_t)arg2, NULL);
refstr_rele(rp);
}
break;
case CC_SET_GLOBAL_PATH:
case CC_SET_DEFAULT_PATH:
if ((error = secpolicy_coreadm(CRED())) != 0)
break;
/* FALLTHROUGH */
case CC_SET_PROCESS_PATH:
if ((size = MIN((size_t)arg2, MAXPATHLEN)) == 0) {
error = EINVAL;
break;
}
path = kmem_alloc(size, KM_SLEEP);
error = copyinstr((char *)arg1, path, size, NULL);
if (error == 0) {
if (subcode == CC_SET_PROCESS_PATH) {
error = set_proc_info((pid_t)arg3, path, 0);
} else if (subcode == CC_SET_DEFAULT_PATH) {
corectl_path_set(cg->core_default_path, path);
} else if (*path != '\0' && *path != '/') {
error = EINVAL;
} else {
refstr_t *nrp = refstr_alloc(path);
mutex_enter(&cg->core_lock);
rp = cg->core_file;
if (*path == '\0')
cg->core_file = NULL;
else
refstr_hold(cg->core_file = nrp);
mutex_exit(&cg->core_lock);
if (rp != NULL)
refstr_rele(rp);
refstr_rele(nrp);
}
}
kmem_free(path, size);
break;
case CC_SET_GLOBAL_CONTENT:
case CC_SET_DEFAULT_CONTENT:
if ((error = secpolicy_coreadm(CRED())) != 0)
break;
/* FALLTHROUGH */
case CC_SET_PROCESS_CONTENT:
error = copyin((void *)arg1, &content, sizeof (content));
if (error != 0)
break;
/*
* If any unknown bits are set, don't let this charade
* continue.
*/
if (content & ~CC_CONTENT_ALL) {
error = EINVAL;
break;
}
if (subcode == CC_SET_PROCESS_CONTENT) {
error = set_proc_info((pid_t)arg2, NULL, content);
} else if (subcode == CC_SET_DEFAULT_CONTENT) {
corectl_content_set(cg->core_default_content, content);
} else {
mutex_enter(&cg->core_lock);
cg->core_content = content;
mutex_exit(&cg->core_lock);
}
break;
case CC_GET_GLOBAL_CONTENT:
content = cg->core_content;
error = copyout(&content, (void *)arg1, sizeof (content));
break;
case CC_GET_DEFAULT_CONTENT:
content = corectl_content_value(cg->core_default_content);
error = copyout(&content, (void *)arg1, sizeof (content));
break;
case CC_GET_PROCESS_CONTENT:
mutex_enter(&pidlock);
if ((p = prfind((pid_t)arg2)) == NULL || p->p_stat == SIDL) {
mutex_exit(&pidlock);
error = ESRCH;
break;
}
mutex_enter(&p->p_lock);
mutex_exit(&pidlock);
mutex_enter(&p->p_crlock);
if (!hasprocperm(p->p_cred, CRED()))
error = EPERM;
else if (p->p_content == NULL)
content = CC_CONTENT_NONE;
else
content = corectl_content_value(p->p_content);
mutex_exit(&p->p_crlock);
mutex_exit(&p->p_lock);
if (error == 0)
error = copyout(&content, (void *)arg1,
sizeof (content));
break;
default:
error = EINVAL;
break;
}
if (error)
return (set_errno(error));
return (0);
}
/*
* The dotoprocs function locates the process(es) specified
* by the procset structure pointed to by psp. funcp points to a
* function which dotoprocs will call for each process in the
* specified set. The arguments to this function will be a pointer
* to the current process from the set and arg.
* If the called function returns -1, it means that processing of the
* procset should stop and a normal (non-error) return should be made
* to the caller of dotoprocs.
* If the called function returns any other non-zero value the search
* is terminated and the function's return value is returned to
* the caller of dotoprocs. This will normally be an error code.
* Otherwise, dotoprocs will return zero after processing the entire
* process set unless no processes were found in which case ESRCH will
* be returned.
*/
int
dotoprocs(procset_t *psp, int (*funcp)(), char *arg)
{
proc_t *prp; /* A process from the set */
int error;
int nfound; /* Nbr of processes found. */
proc_t *lastprp; /* Last proc found. */
ASSERT(funcp != NULL);
/*
* Check that the procset_t is valid.
*/
error = checkprocset(psp);
if (error) {
return (error);
}
/*
* Check for the special value P_MYID in either operand
* and replace it with the correct value. We don't check
* for an error return from getmyid() because the idtypes
* have been validated by the checkprocset() call above.
*/
mutex_enter(&pidlock);
if (psp->p_lid == P_MYID) {
psp->p_lid = getmyid(psp->p_lidtype);
}
if (psp->p_rid == P_MYID) {
psp->p_rid = getmyid(psp->p_ridtype);
}
/*
* If psp only acts on a single proc, we can reduce pidlock hold time
* by avoiding a needless scan of the entire proc list. Although
* there are many procset_t combinations which might boil down to a
* single proc, the most common case is an AND operation where one
* side is a specific pid, and the other side is P_ALL, so that is
* the case for which we will provide a fast path. Other cases could
* be added in a similar fashion if they were to become significant
* pidlock bottlenecks.
*
* Perform the check symmetrically: either the left or right side may
* specify a pid, with the opposite side being 'all'.
*/
if (psp->p_op == POP_AND) {
if (((psp->p_lidtype == P_PID) && (psp->p_ridtype == P_ALL)) ||
((psp->p_ridtype == P_PID) && (psp->p_lidtype == P_ALL))) {
id_t pid;
pid = (psp->p_lidtype == P_PID) ?
psp->p_lid : psp->p_rid;
if (((prp = prfind((pid_t)pid)) == NULL) ||
(prp->p_stat == SIDL || prp->p_stat == SZOMB ||
prp->p_tlist == NULL || prp->p_flag & SSYS)) {
/*
* Specified proc doesn't exist or should
* not be operated on.
* Don't need to make HASZONEACCESS check
* here since prfind() takes care of that.
*/
mutex_exit(&pidlock);
return (ESRCH);
}
/*
* Operate only on the specified proc. It's okay
* if it's init.
*/
error = (*funcp)(prp, arg);
mutex_exit(&pidlock);
if (error == -1)
error = 0;
return (error);
}
}
nfound = 0;
error = 0;
for (prp = practive; prp != NULL; prp = prp->p_next) {
/*
* If caller is in a non-global zone, skip processes
* in other zones.
*/
if (!HASZONEACCESS(curproc, prp->p_zone->zone_id))
continue;
/*
* Ignore this process if it's coming or going,
* if it's a system process or if it's not in
* the given procset_t.
*/
if (prp->p_stat == SIDL || prp->p_stat == SZOMB)
continue;
mutex_enter(&prp->p_lock);
if (prp->p_flag & SSYS || prp->p_tlist == NULL ||
procinset(prp, psp) == 0) {
mutex_exit(&prp->p_lock);
} else {
mutex_exit(&prp->p_lock);
nfound++;
lastprp = prp;
if (prp != proc_init) {
error = (*funcp)(prp, arg);
if (error == -1) {
mutex_exit(&pidlock);
return (0);
} else if (error) {
mutex_exit(&pidlock);
return (error);
}
}
}
}
if (nfound == 0) {
mutex_exit(&pidlock);
return (ESRCH);
}
if (nfound == 1 && lastprp == proc_init)
error = (*funcp)(lastprp, arg);
if (error == -1)
error = 0;
mutex_exit(&pidlock);
return (error);
}
int
klpd_unreg(int did, idtype_t type, id_t id)
{
door_handle_t dh;
int res = 0;
proc_t *p;
pid_t pid;
projid_t proj;
kproject_t *kpp = NULL;
credklpd_t *ckp;
switch (type) {
case P_PID:
pid = (pid_t)id;
break;
case P_PROJID:
proj = (projid_t)id;
kpp = project_hold_by_id(proj, crgetzone(CRED()),
PROJECT_HOLD_FIND);
if (kpp == NULL)
return (set_errno(ESRCH));
break;
default:
return (set_errno(ENOTSUP));
}
dh = door_ki_lookup(did);
if (dh == NULL) {
if (kpp != NULL)
project_rele(kpp);
return (set_errno(EINVAL));
}
if (kpp != NULL) {
mutex_enter(&klpd_mutex);
if (kpp->kpj_klpd == NULL)
res = ESRCH;
else
klpd_freelist(&kpp->kpj_klpd);
mutex_exit(&klpd_mutex);
project_rele(kpp);
goto out;
} else if ((int)pid > 0) {
mutex_enter(&pidlock);
p = prfind(pid);
if (p == NULL) {
mutex_exit(&pidlock);
door_ki_rele(dh);
return (set_errno(ESRCH));
}
mutex_enter(&p->p_crlock);
mutex_exit(&pidlock);
} else if (pid == 0) {
p = curproc;
mutex_enter(&p->p_crlock);
} else {
res = klpd_unreg_dh(dh);
goto out;
}
ckp = crgetcrklpd(p->p_cred);
if (ckp != NULL) {
crklpd_setreg(ckp, NULL);
} else {
res = ESRCH;
}
mutex_exit(&p->p_crlock);
out:
door_ki_rele(dh);
if (res != 0)
return (set_errno(res));
return (0);
}
/*
* Register the klpd.
* If the pid_t passed in is positive, update the registration for
* the specific process; that is only possible if the process already
* has a registration on it. This change of registration will affect
* all processes which share common ancestry.
*
* MY_PID (pid 0) can be used to create or change the context for
* the current process, typically done after fork().
*
* A negative value can be used to register a klpd globally.
*
* The per-credential klpd needs to be cleaned up when entering
* a zone or unsetting the flag.
*/
int
klpd_reg(int did, idtype_t type, id_t id, priv_set_t *psetbuf)
{
cred_t *cr = CRED();
door_handle_t dh;
klpd_reg_t *kpd;
priv_set_t pset;
door_info_t di;
credklpd_t *ckp = NULL;
pid_t pid = -1;
projid_t proj = -1;
kproject_t *kpp = NULL;
if (CR_FLAGS(cr) & PRIV_XPOLICY)
return (set_errno(EINVAL));
if (copyin(psetbuf, &pset, sizeof (priv_set_t)))
return (set_errno(EFAULT));
if (!priv_issubset(&pset, &CR_OEPRIV(cr)))
return (set_errno(EPERM));
switch (type) {
case P_PID:
pid = (pid_t)id;
if (pid == P_MYPID)
pid = curproc->p_pid;
if (pid == curproc->p_pid)
ckp = crklpd_alloc();
break;
case P_PROJID:
proj = (projid_t)id;
kpp = project_hold_by_id(proj, crgetzone(cr),
PROJECT_HOLD_FIND);
if (kpp == NULL)
return (set_errno(ESRCH));
break;
default:
return (set_errno(ENOTSUP));
}
/*
* Verify the door passed in; it must be a door and we won't
* allow processes to be called on their own behalf.
*/
dh = door_ki_lookup(did);
if (dh == NULL || door_ki_info(dh, &di) != 0) {
if (ckp != NULL)
crklpd_rele(ckp);
if (kpp != NULL)
project_rele(kpp);
return (set_errno(EBADF));
}
if (type == P_PID && pid == di.di_target) {
if (ckp != NULL)
crklpd_rele(ckp);
ASSERT(kpp == NULL);
return (set_errno(EINVAL));
}
kpd = kmem_zalloc(sizeof (*kpd), KM_SLEEP);
crhold(kpd->klpd_cred = cr);
kpd->klpd_door = dh;
kpd->klpd_door_pid = di.di_target;
kpd->klpd_ref = 1;
kpd->klpd_pset = pset;
if (kpp != NULL) {
mutex_enter(&klpd_mutex);
kpd = klpd_link(kpd, &kpp->kpj_klpd, B_TRUE);
mutex_exit(&klpd_mutex);
if (kpd != NULL)
klpd_rele(kpd);
project_rele(kpp);
} else if ((int)pid < 0) {
/* Global daemon */
mutex_enter(&klpd_mutex);
(void) klpd_link(kpd, &klpd_list, B_FALSE);
mutex_exit(&klpd_mutex);
} else if (pid == curproc->p_pid) {
proc_t *p = curproc;
cred_t *newcr = cralloc();
/* No need to lock, sole reference to ckp */
kpd = klpd_link(kpd, &ckp->crkl_reg, B_TRUE);
if (kpd != NULL)
klpd_rele(kpd);
mutex_enter(&p->p_crlock);
cr = p->p_cred;
crdup_to(cr, newcr);
crsetcrklpd(newcr, ckp);
p->p_cred = newcr; /* Already held for p_cred */
crhold(newcr); /* Hold once for the current thread */
mutex_exit(&p->p_crlock);
crfree(cr); /* One for the p_cred */
crset(p, newcr);
} else {
proc_t *p;
cred_t *pcr;
mutex_enter(&pidlock);
p = prfind(pid);
if (p == NULL || !prochasprocperm(p, curproc, CRED())) {
mutex_exit(&pidlock);
klpd_rele(kpd);
return (set_errno(p == NULL ? ESRCH : EPERM));
}
mutex_enter(&p->p_crlock);
crhold(pcr = p->p_cred);
mutex_exit(&pidlock);
mutex_exit(&p->p_crlock);
/*
* We're going to update the credential's ckp in place;
* this requires that it exists.
*/
ckp = crgetcrklpd(pcr);
if (ckp == NULL) {
crfree(pcr);
klpd_rele(kpd);
return (set_errno(EINVAL));
}
crklpd_setreg(ckp, kpd);
crfree(pcr);
}
return (0);
}
static int
set_proc_info(pid_t pid, const char *path, core_content_t content)
{
proc_t *p;
counter_t counter;
int error = 0;
counter.cc_count = 0;
/*
* Only one of the core file path or content can be set at a time.
*/
if (path != NULL) {
counter.cc_path = corectl_path_alloc(path);
counter.cc_content = NULL;
} else {
counter.cc_path = NULL;
counter.cc_content = corectl_content_alloc(content);
}
if (pid == -1) {
procset_t set;
setprocset(&set, POP_AND, P_ALL, P_MYID, P_ALL, P_MYID);
error = dotoprocs(&set, set_one_proc_info, (char *)&counter);
if (error == 0 && counter.cc_count == 0)
error = EPERM;
} else if (pid > 0) {
mutex_enter(&pidlock);
if ((p = prfind(pid)) == NULL || p->p_stat == SIDL) {
error = ESRCH;
} else {
(void) set_one_proc_info(p, &counter);
if (counter.cc_count == 0)
error = EPERM;
}
mutex_exit(&pidlock);
} else {
int nfound = 0;
pid_t pgid;
if (pid == 0)
pgid = curproc->p_pgrp;
else
pgid = -pid;
mutex_enter(&pidlock);
for (p = pgfind(pgid); p != NULL; p = p->p_pglink) {
if (p->p_stat != SIDL) {
nfound++;
(void) set_one_proc_info(p, &counter);
}
}
mutex_exit(&pidlock);
if (nfound == 0)
error = ESRCH;
else if (counter.cc_count == 0)
error = EPERM;
}
if (path != NULL)
corectl_path_rele(counter.cc_path);
else
corectl_content_rele(counter.cc_content);
if (error)
return (set_errno(error));
return (0);
}
static int
getpinfo_addr(caddr_t data, int len)
{
STRUCT_DECL(auditpinfo_addr, apinfo);
proc_t *proc;
const auditinfo_addr_t *ainfo;
model_t model;
cred_t *cr, *newcred;
model = get_udatamodel();
STRUCT_INIT(apinfo, model);
if (len < STRUCT_SIZE(apinfo))
return (EOVERFLOW);
if (copyin(data, STRUCT_BUF(apinfo), STRUCT_SIZE(apinfo)))
return (EFAULT);
newcred = cralloc();
mutex_enter(&pidlock);
if ((proc = prfind(STRUCT_FGET(apinfo, ap_pid))) == NULL) {
mutex_exit(&pidlock);
crfree(newcred);
return (ESRCH);
}
mutex_enter(&proc->p_lock); /* so process doesn't go away */
mutex_exit(&pidlock);
audit_update_context(proc, newcred); /* make sure it's up-to-date */
mutex_enter(&proc->p_crlock);
crhold(cr = proc->p_cred);
mutex_exit(&proc->p_crlock);
mutex_exit(&proc->p_lock);
ainfo = crgetauinfo(cr);
if (ainfo == NULL) {
crfree(cr);
return (EINVAL);
}
STRUCT_FSET(apinfo, ap_auid, ainfo->ai_auid);
STRUCT_FSET(apinfo, ap_asid, ainfo->ai_asid);
#ifdef _LP64
if (model == DATAMODEL_ILP32) {
dev32_t dev;
/* convert internal 64 bit form to 32 bit version */
if (cmpldev(&dev, ainfo->ai_termid.at_port) == 0) {
crfree(cr);
return (EOVERFLOW);
}
STRUCT_FSET(apinfo, ap_termid.at_port, dev);
} else
STRUCT_FSET(apinfo, ap_termid.at_port,
ainfo->ai_termid.at_port);
#else
STRUCT_FSET(apinfo, ap_termid.at_port, ainfo->ai_termid.at_port);
#endif
STRUCT_FSET(apinfo, ap_termid.at_type, ainfo->ai_termid.at_type);
STRUCT_FSET(apinfo, ap_termid.at_addr[0], ainfo->ai_termid.at_addr[0]);
STRUCT_FSET(apinfo, ap_termid.at_addr[1], ainfo->ai_termid.at_addr[1]);
STRUCT_FSET(apinfo, ap_termid.at_addr[2], ainfo->ai_termid.at_addr[2]);
STRUCT_FSET(apinfo, ap_termid.at_addr[3], ainfo->ai_termid.at_addr[3]);
STRUCT_FSET(apinfo, ap_mask, ainfo->ai_mask);
crfree(cr);
if (copyout(STRUCT_BUF(apinfo), data, STRUCT_SIZE(apinfo)))
return (EFAULT);
return (0);
}
/* ARGSUSED */
int
setpgrp(int flag, int pid, int pgid)
{
register proc_t *p = ttoproc(curthread);
register int retval = 0;
switch (flag) {
case 1: /* setpgrp() */
mutex_enter(&pidlock);
if (p->p_sessp->s_sidp != p->p_pidp && !pgmembers(p->p_pid)) {
mutex_exit(&pidlock);
sess_create();
} else
mutex_exit(&pidlock);
return (p->p_sessp->s_sid);
case 3: /* setsid() */
mutex_enter(&pidlock);
if (p->p_pgidp == p->p_pidp || pgmembers(p->p_pid)) {
mutex_exit(&pidlock);
return (set_errno(EPERM));
}
mutex_exit(&pidlock);
sess_create();
return (p->p_sessp->s_sid);
case 5: /* setpgid() */
{
mutex_enter(&pidlock);
if (pid == 0)
pid = p->p_pid;
else if (pid < 0 || pid >= maxpid) {
mutex_exit(&pidlock);
return (set_errno(EINVAL));
} else if (pid != p->p_pid) {
for (p = p->p_child; /* empty */; p = p->p_sibling) {
if (p == NULL) {
mutex_exit(&pidlock);
return (set_errno(ESRCH));
}
if (p->p_pid == pid)
break;
}
if (p->p_flag & SEXECED) {
mutex_exit(&pidlock);
return (set_errno(EACCES));
}
if (p->p_sessp != ttoproc(curthread)->p_sessp) {
mutex_exit(&pidlock);
return (set_errno(EPERM));
}
}
if (p->p_sessp->s_sid == pid) {
mutex_exit(&pidlock);
return (set_errno(EPERM));
}
if (pgid == 0)
pgid = p->p_pid;
else if (pgid < 0 || pgid >= maxpid) {
mutex_exit(&pidlock);
return (set_errno(EINVAL));
}
if (p->p_pgrp == pgid) {
mutex_exit(&pidlock);
break;
} else if (p->p_pid == pgid) {
/*
* We need to protect p_pgidp with p_lock because
* /proc looks at it while holding only p_lock.
*/
mutex_enter(&p->p_lock);
pgexit(p);
pgjoin(p, p->p_pidp);
mutex_exit(&p->p_lock);
} else {
register proc_t *q;
if ((q = pgfind(pgid)) == NULL ||
q->p_sessp != p->p_sessp) {
mutex_exit(&pidlock);
return (set_errno(EPERM));
}
/*
* See comment above about p_lock and /proc
*/
mutex_enter(&p->p_lock);
pgexit(p);
pgjoin(p, q->p_pgidp);
mutex_exit(&p->p_lock);
}
mutex_exit(&pidlock);
break;
}
case 0: /* getpgrp() */
mutex_enter(&pidlock);
retval = p->p_pgrp;
mutex_exit(&pidlock);
break;
case 2: /* getsid() */
case 4: /* getpgid() */
if (pid < 0 || pid >= maxpid) {
return (set_errno(EINVAL));
}
mutex_enter(&pidlock);
if (pid != 0 && p->p_pid != pid &&
((p = prfind(pid)) == NULL || p->p_stat == SIDL)) {
mutex_exit(&pidlock);
return (set_errno(ESRCH));
}
if (flag == 2)
retval = p->p_sessp->s_sid;
else
retval = p->p_pgrp;
mutex_exit(&pidlock);
break;
}
return (retval);
}
/*
* Process debugging system call.
*/
int
sys_ptrace(struct proc *p, void *v, register_t *retval)
{
struct sys_ptrace_args /* {
syscallarg(int) req;
syscallarg(pid_t) pid;
syscallarg(caddr_t) addr;
syscallarg(int) data;
} */ *uap = v;
struct proc *t; /* target thread */
struct process *tr; /* target process */
struct uio uio;
struct iovec iov;
struct ptrace_io_desc piod;
struct ptrace_event pe;
struct ptrace_thread_state pts;
struct reg *regs;
#if defined (PT_SETFPREGS) || defined (PT_GETFPREGS)
struct fpreg *fpregs;
#endif
#if defined (PT_SETXMMREGS) || defined (PT_GETXMMREGS)
struct xmmregs *xmmregs;
#endif
#ifdef PT_WCOOKIE
register_t wcookie;
#endif
int error, write;
int temp;
int req = SCARG(uap, req);
int s;
/* "A foolish consistency..." XXX */
switch (req) {
case PT_TRACE_ME:
t = p;
break;
/* calls that only operate on the PID */
case PT_READ_I:
case PT_READ_D:
case PT_WRITE_I:
case PT_WRITE_D:
case PT_KILL:
case PT_ATTACH:
case PT_IO:
case PT_SET_EVENT_MASK:
case PT_GET_EVENT_MASK:
case PT_GET_PROCESS_STATE:
case PT_GET_THREAD_FIRST:
case PT_GET_THREAD_NEXT:
default:
/* Find the process we're supposed to be operating on. */
if ((t = pfind(SCARG(uap, pid))) == NULL)
return (ESRCH);
if (t->p_flag & P_THREAD)
return (ESRCH);
break;
/* calls that accept a PID or a thread ID */
case PT_CONTINUE:
case PT_DETACH:
#ifdef PT_STEP
case PT_STEP:
#endif
case PT_GETREGS:
case PT_SETREGS:
#ifdef PT_GETFPREGS
case PT_GETFPREGS:
#endif
#ifdef PT_SETFPREGS
case PT_SETFPREGS:
#endif
#ifdef PT_GETXMMREGS
case PT_GETXMMREGS:
#endif
#ifdef PT_SETXMMREGS
case PT_SETXMMREGS:
#endif
if (SCARG(uap, pid) > THREAD_PID_OFFSET) {
t = pfind(SCARG(uap, pid) - THREAD_PID_OFFSET);
if (t == NULL)
return (ESRCH);
} else {
if ((t = pfind(SCARG(uap, pid))) == NULL)
return (ESRCH);
if (t->p_flag & P_THREAD)
return (ESRCH);
}
break;
}
tr = t->p_p;
if ((tr->ps_flags & PS_INEXEC) != 0)
return (EAGAIN);
/* Make sure we can operate on it. */
switch (req) {
case PT_TRACE_ME:
/* Saying that you're being traced is always legal. */
break;
case PT_ATTACH:
/*
* You can't attach to a process if:
* (1) it's the process that's doing the attaching,
*/
if (tr == p->p_p)
return (EINVAL);
/*
* (2) it's a system process
*/
if (ISSET(tr->ps_flags, PS_SYSTEM))
return (EPERM);
/*
* (3) it's already being traced, or
*/
if (ISSET(tr->ps_flags, PS_TRACED))
return (EBUSY);
/*
* (4) it's not owned by you, or the last exec
* gave us setuid/setgid privs (unless
* you're root), or...
*
* [Note: once PS_SUGID or PS_SUGIDEXEC gets set in
* execve(), they stay set until the process does
* another execve(). Hence this prevents a setuid
* process which revokes its special privileges using
* setuid() from being traced. This is good security.]
*/
if ((tr->ps_ucred->cr_ruid != p->p_ucred->cr_ruid ||
ISSET(tr->ps_flags, PS_SUGIDEXEC | PS_SUGID)) &&
(error = suser(p, 0)) != 0)
return (error);
/*
* (4.5) it's not a child of the tracing process.
*/
if (global_ptrace == 0 && !inferior(tr, p->p_p) &&
(error = suser(p, 0)) != 0)
return (error);
/*
* (5) ...it's init, which controls the security level
* of the entire system, and the system was not
* compiled with permanently insecure mode turned
* on.
*/
if ((tr->ps_pid == 1) && (securelevel > -1))
return (EPERM);
/*
* (6) it's an ancestor of the current process and
* not init (because that would create a loop in
* the process graph).
*/
if (tr->ps_pid != 1 && inferior(p->p_p, tr))
return (EINVAL);
break;
case PT_READ_I:
case PT_READ_D:
case PT_WRITE_I:
case PT_WRITE_D:
case PT_IO:
case PT_CONTINUE:
case PT_KILL:
case PT_DETACH:
#ifdef PT_STEP
case PT_STEP:
#endif
case PT_SET_EVENT_MASK:
case PT_GET_EVENT_MASK:
case PT_GET_PROCESS_STATE:
case PT_GETREGS:
case PT_SETREGS:
#ifdef PT_GETFPREGS
case PT_GETFPREGS:
#endif
#ifdef PT_SETFPREGS
case PT_SETFPREGS:
#endif
#ifdef PT_GETXMMREGS
case PT_GETXMMREGS:
#endif
#ifdef PT_SETXMMREGS
case PT_SETXMMREGS:
#endif
#ifdef PT_WCOOKIE
case PT_WCOOKIE:
#endif
/*
* You can't do what you want to the process if:
* (1) It's not being traced at all,
*/
if (!ISSET(tr->ps_flags, PS_TRACED))
return (EPERM);
/*
* (2) it's not being traced by _you_, or
*/
if (tr->ps_pptr != p->p_p)
return (EBUSY);
/*
* (3) it's not currently stopped.
*/
if (t->p_stat != SSTOP || !ISSET(tr->ps_flags, PS_WAITED))
return (EBUSY);
break;
case PT_GET_THREAD_FIRST:
case PT_GET_THREAD_NEXT:
/*
* You can't do what you want to the process if:
* (1) It's not being traced at all,
*/
if (!ISSET(tr->ps_flags, PS_TRACED))
return (EPERM);
/*
* (2) it's not being traced by _you_, or
*/
if (tr->ps_pptr != p->p_p)
return (EBUSY);
/*
* Do the work here because the request isn't actually
* associated with 't'
*/
if (SCARG(uap, data) != sizeof(pts))
return (EINVAL);
if (req == PT_GET_THREAD_NEXT) {
error = copyin(SCARG(uap, addr), &pts, sizeof(pts));
if (error)
return (error);
t = pfind(pts.pts_tid - THREAD_PID_OFFSET);
if (t == NULL || ISSET(t->p_flag, P_WEXIT))
return (ESRCH);
if (t->p_p != tr)
return (EINVAL);
t = TAILQ_NEXT(t, p_thr_link);
} else {
t = TAILQ_FIRST(&tr->ps_threads);
}
if (t == NULL)
pts.pts_tid = -1;
else
pts.pts_tid = t->p_pid + THREAD_PID_OFFSET;
return (copyout(&pts, SCARG(uap, addr), sizeof(pts)));
default: /* It was not a legal request. */
return (EINVAL);
}
/* Do single-step fixup if needed. */
FIX_SSTEP(t);
/* Now do the operation. */
write = 0;
*retval = 0;
switch (req) {
case PT_TRACE_ME:
/* Just set the trace flag. */
atomic_setbits_int(&tr->ps_flags, PS_TRACED);
tr->ps_oppid = tr->ps_pptr->ps_pid;
if (tr->ps_ptstat == NULL)
tr->ps_ptstat = malloc(sizeof(*tr->ps_ptstat),
M_SUBPROC, M_WAITOK);
memset(tr->ps_ptstat, 0, sizeof(*tr->ps_ptstat));
return (0);
case PT_WRITE_I: /* XXX no separate I and D spaces */
case PT_WRITE_D:
write = 1;
temp = SCARG(uap, data);
case PT_READ_I: /* XXX no separate I and D spaces */
case PT_READ_D:
/* write = 0 done above. */
iov.iov_base = (caddr_t)&temp;
iov.iov_len = sizeof(int);
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = (off_t)(vaddr_t)SCARG(uap, addr);
uio.uio_resid = sizeof(int);
uio.uio_segflg = UIO_SYSSPACE;
uio.uio_rw = write ? UIO_WRITE : UIO_READ;
uio.uio_procp = p;
error = process_domem(p, t, &uio, write ? PT_WRITE_I :
PT_READ_I);
if (write == 0)
*retval = temp;
return (error);
case PT_IO:
error = copyin(SCARG(uap, addr), &piod, sizeof(piod));
if (error)
return (error);
iov.iov_base = piod.piod_addr;
iov.iov_len = piod.piod_len;
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = (off_t)(vaddr_t)piod.piod_offs;
uio.uio_resid = piod.piod_len;
uio.uio_segflg = UIO_USERSPACE;
uio.uio_procp = p;
switch (piod.piod_op) {
case PIOD_READ_I:
req = PT_READ_I;
uio.uio_rw = UIO_READ;
break;
case PIOD_READ_D:
req = PT_READ_D;
uio.uio_rw = UIO_READ;
break;
case PIOD_WRITE_I:
req = PT_WRITE_I;
uio.uio_rw = UIO_WRITE;
break;
case PIOD_WRITE_D:
req = PT_WRITE_D;
uio.uio_rw = UIO_WRITE;
break;
case PIOD_READ_AUXV:
req = PT_READ_D;
uio.uio_rw = UIO_READ;
temp = tr->ps_emul->e_arglen * sizeof(char *);
if (uio.uio_offset > temp)
return (EIO);
if (uio.uio_resid > temp - uio.uio_offset)
uio.uio_resid = temp - uio.uio_offset;
piod.piod_len = iov.iov_len = uio.uio_resid;
error = process_auxv_offset(p, t, &uio);
if (error)
return (error);
break;
default:
return (EINVAL);
}
error = process_domem(p, t, &uio, req);
piod.piod_len -= uio.uio_resid;
(void) copyout(&piod, SCARG(uap, addr), sizeof(piod));
return (error);
#ifdef PT_STEP
case PT_STEP:
/*
* From the 4.4BSD PRM:
* "Execution continues as in request PT_CONTINUE; however
* as soon as possible after execution of at least one
* instruction, execution stops again. [ ... ]"
*/
#endif
case PT_CONTINUE:
/*
* From the 4.4BSD PRM:
* "The data argument is taken as a signal number and the
* child's execution continues at location addr as if it
* incurred that signal. Normally the signal number will
* be either 0 to indicate that the signal that caused the
* stop should be ignored, or that value fetched out of
* the process's image indicating which signal caused
* the stop. If addr is (int *)1 then execution continues
* from where it stopped."
*/
if (SCARG(uap, pid) < THREAD_PID_OFFSET && tr->ps_single)
t = tr->ps_single;
/* Check that the data is a valid signal number or zero. */
if (SCARG(uap, data) < 0 || SCARG(uap, data) >= NSIG)
return (EINVAL);
/* If the address parameter is not (int *)1, set the pc. */
if ((int *)SCARG(uap, addr) != (int *)1)
if ((error = process_set_pc(t, SCARG(uap, addr))) != 0)
goto relebad;
#ifdef PT_STEP
/*
* Arrange for a single-step, if that's requested and possible.
*/
error = process_sstep(t, req == PT_STEP);
if (error)
goto relebad;
#endif
goto sendsig;
case PT_DETACH:
/*
* From the 4.4BSD PRM:
* "The data argument is taken as a signal number and the
* child's execution continues at location addr as if it
* incurred that signal. Normally the signal number will
* be either 0 to indicate that the signal that caused the
* stop should be ignored, or that value fetched out of
* the process's image indicating which signal caused
* the stop. If addr is (int *)1 then execution continues
* from where it stopped."
*/
if (SCARG(uap, pid) < THREAD_PID_OFFSET && tr->ps_single)
t = tr->ps_single;
/* Check that the data is a valid signal number or zero. */
if (SCARG(uap, data) < 0 || SCARG(uap, data) >= NSIG)
return (EINVAL);
#ifdef PT_STEP
/*
* Arrange for a single-step, if that's requested and possible.
*/
error = process_sstep(t, req == PT_STEP);
if (error)
goto relebad;
#endif
/* give process back to original parent or init */
if (tr->ps_oppid != tr->ps_pptr->ps_pid) {
struct process *ppr;
ppr = prfind(tr->ps_oppid);
proc_reparent(tr, ppr ? ppr : initprocess);
}
/* not being traced any more */
tr->ps_oppid = 0;
atomic_clearbits_int(&tr->ps_flags, PS_TRACED|PS_WAITED);
sendsig:
memset(tr->ps_ptstat, 0, sizeof(*tr->ps_ptstat));
/* Finally, deliver the requested signal (or none). */
if (t->p_stat == SSTOP) {
t->p_xstat = SCARG(uap, data);
SCHED_LOCK(s);
setrunnable(t);
SCHED_UNLOCK(s);
} else {
if (SCARG(uap, data) != 0)
psignal(t, SCARG(uap, data));
}
return (0);
relebad:
return (error);
case PT_KILL:
if (SCARG(uap, pid) < THREAD_PID_OFFSET && tr->ps_single)
t = tr->ps_single;
/* just send the process a KILL signal. */
SCARG(uap, data) = SIGKILL;
goto sendsig; /* in PT_CONTINUE, above. */
case PT_ATTACH:
/*
* As was done in procfs:
* Go ahead and set the trace flag.
* Save the old parent (it's reset in
* _DETACH, and also in kern_exit.c:wait4()
* Reparent the process so that the tracing
* proc gets to see all the action.
* Stop the target.
*/
atomic_setbits_int(&tr->ps_flags, PS_TRACED);
tr->ps_oppid = tr->ps_pptr->ps_pid;
if (tr->ps_pptr != p->p_p)
proc_reparent(tr, p->p_p);
if (tr->ps_ptstat == NULL)
tr->ps_ptstat = malloc(sizeof(*tr->ps_ptstat),
M_SUBPROC, M_WAITOK);
SCARG(uap, data) = SIGSTOP;
goto sendsig;
case PT_GET_EVENT_MASK:
if (SCARG(uap, data) != sizeof(pe))
return (EINVAL);
memset(&pe, 0, sizeof(pe));
pe.pe_set_event = tr->ps_ptmask;
return (copyout(&pe, SCARG(uap, addr), sizeof(pe)));
case PT_SET_EVENT_MASK:
if (SCARG(uap, data) != sizeof(pe))
return (EINVAL);
if ((error = copyin(SCARG(uap, addr), &pe, sizeof(pe))))
return (error);
tr->ps_ptmask = pe.pe_set_event;
return (0);
case PT_GET_PROCESS_STATE:
if (SCARG(uap, data) != sizeof(*tr->ps_ptstat))
return (EINVAL);
if (tr->ps_single)
tr->ps_ptstat->pe_tid =
tr->ps_single->p_pid + THREAD_PID_OFFSET;
return (copyout(tr->ps_ptstat, SCARG(uap, addr),
sizeof(*tr->ps_ptstat)));
case PT_SETREGS:
KASSERT((p->p_flag & P_SYSTEM) == 0);
if ((error = process_checkioperm(p, tr)) != 0)
return (error);
regs = malloc(sizeof(*regs), M_TEMP, M_WAITOK);
error = copyin(SCARG(uap, addr), regs, sizeof(*regs));
if (error == 0) {
error = process_write_regs(t, regs);
}
free(regs, M_TEMP, sizeof(*regs));
return (error);
case PT_GETREGS:
KASSERT((p->p_flag & P_SYSTEM) == 0);
if ((error = process_checkioperm(p, tr)) != 0)
return (error);
regs = malloc(sizeof(*regs), M_TEMP, M_WAITOK);
error = process_read_regs(t, regs);
if (error == 0)
error = copyout(regs,
SCARG(uap, addr), sizeof (*regs));
free(regs, M_TEMP, sizeof(*regs));
return (error);
#ifdef PT_SETFPREGS
case PT_SETFPREGS:
KASSERT((p->p_flag & P_SYSTEM) == 0);
if ((error = process_checkioperm(p, tr)) != 0)
return (error);
fpregs = malloc(sizeof(*fpregs), M_TEMP, M_WAITOK);
error = copyin(SCARG(uap, addr), fpregs, sizeof(*fpregs));
if (error == 0) {
error = process_write_fpregs(t, fpregs);
}
free(fpregs, M_TEMP, sizeof(*fpregs));
return (error);
#endif
#ifdef PT_GETFPREGS
case PT_GETFPREGS:
KASSERT((p->p_flag & P_SYSTEM) == 0);
if ((error = process_checkioperm(p, tr)) != 0)
return (error);
fpregs = malloc(sizeof(*fpregs), M_TEMP, M_WAITOK);
error = process_read_fpregs(t, fpregs);
if (error == 0)
error = copyout(fpregs,
SCARG(uap, addr), sizeof(*fpregs));
free(fpregs, M_TEMP, sizeof(*fpregs));
return (error);
#endif
#ifdef PT_SETXMMREGS
case PT_SETXMMREGS:
KASSERT((p->p_flag & P_SYSTEM) == 0);
if ((error = process_checkioperm(p, tr)) != 0)
return (error);
xmmregs = malloc(sizeof(*xmmregs), M_TEMP, M_WAITOK);
error = copyin(SCARG(uap, addr), xmmregs, sizeof(*xmmregs));
if (error == 0) {
error = process_write_xmmregs(t, xmmregs);
}
free(xmmregs, M_TEMP, sizeof(*xmmregs));
return (error);
#endif
#ifdef PT_GETXMMREGS
case PT_GETXMMREGS:
KASSERT((p->p_flag & P_SYSTEM) == 0);
if ((error = process_checkioperm(p, tr)) != 0)
return (error);
xmmregs = malloc(sizeof(*xmmregs), M_TEMP, M_WAITOK);
error = process_read_xmmregs(t, xmmregs);
if (error == 0)
error = copyout(xmmregs,
SCARG(uap, addr), sizeof(*xmmregs));
free(xmmregs, M_TEMP, sizeof(*xmmregs));
return (error);
#endif
#ifdef PT_WCOOKIE
case PT_WCOOKIE:
wcookie = process_get_wcookie (t);
return (copyout(&wcookie, SCARG(uap, addr),
sizeof (register_t)));
#endif
}
#ifdef DIAGNOSTIC
panic("ptrace: impossible");
#endif
return 0;
}
int
signotify(int cmd, siginfo_t *siginfo, signotify_id_t *sn_id)
{
k_siginfo_t info;
signotify_id_t id;
proc_t *p;
proc_t *cp = curproc;
signotifyq_t *snqp;
struct cred *cr;
sigqueue_t *sqp;
sigqhdr_t *sqh;
u_longlong_t sid;
model_t datamodel = get_udatamodel();
if (copyin(sn_id, &id, sizeof (signotify_id_t)))
return (set_errno(EFAULT));
if (id.sn_index >= _SIGNOTIFY_MAX || id.sn_index < 0)
return (set_errno(EINVAL));
switch (cmd) {
case SN_PROC:
/* get snid for the given user address of signotifyid_t */
sid = get_sigid(cp, (caddr_t)sn_id);
if (id.sn_pid > 0) {
mutex_enter(&pidlock);
if ((p = prfind(id.sn_pid)) != NULL) {
mutex_enter(&p->p_lock);
if (p->p_signhdr != NULL) {
snqp = SIGN_PTR(p, id.sn_index);
if (snqp->sn_snid == sid) {
mutex_exit(&p->p_lock);
mutex_exit(&pidlock);
return (set_errno(EBUSY));
}
}
mutex_exit(&p->p_lock);
}
mutex_exit(&pidlock);
}
if (copyin_siginfo(datamodel, siginfo, &info))
return (set_errno(EFAULT));
/* The si_code value must indicate the signal will be queued */
if (!sigwillqueue(info.si_signo, info.si_code))
return (set_errno(EINVAL));
if (cp->p_signhdr == NULL) {
/* Allocate signotify pool first time */
sqh = sigqhdralloc(sizeof (signotifyq_t),
_SIGNOTIFY_MAX);
mutex_enter(&cp->p_lock);
if (cp->p_signhdr == NULL) {
/* hang the pool head on proc */
cp->p_signhdr = sqh;
} else {
/* another lwp allocated the pool, free ours */
sigqhdrfree(sqh);
}
} else {
mutex_enter(&cp->p_lock);
}
sqp = sigqalloc(cp->p_signhdr);
if (sqp == NULL) {
mutex_exit(&cp->p_lock);
return (set_errno(EAGAIN));
}
cr = CRED();
sqp->sq_info = info;
sqp->sq_info.si_pid = cp->p_pid;
sqp->sq_info.si_ctid = PRCTID(cp);
sqp->sq_info.si_zoneid = getzoneid();
sqp->sq_info.si_uid = crgetruid(cr);
/* fill the signotifyq_t fields */
((signotifyq_t *)sqp)->sn_snid = sid;
mutex_exit(&cp->p_lock);
/* complete the signotify_id_t fields */
id.sn_index = (signotifyq_t *)sqp - SIGN_PTR(cp, 0);
id.sn_pid = cp->p_pid;
break;
case SN_CANCEL:
case SN_SEND:
sid = get_sigid(cp, (caddr_t)sn_id);
mutex_enter(&pidlock);
if ((id.sn_pid <= 0) || ((p = prfind(id.sn_pid)) == NULL)) {
mutex_exit(&pidlock);
return (set_errno(EINVAL));
}
mutex_enter(&p->p_lock);
mutex_exit(&pidlock);
if (p->p_signhdr == NULL) {
mutex_exit(&p->p_lock);
return (set_errno(EINVAL));
}
snqp = SIGN_PTR(p, id.sn_index);
if (snqp->sn_snid == 0) {
mutex_exit(&p->p_lock);
return (set_errno(EINVAL));
}
if (snqp->sn_snid != sid) {
mutex_exit(&p->p_lock);
return (set_errno(EINVAL));
}
snqp->sn_snid = 0;
/* cmd == SN_CANCEL or signo == 0 (SIGEV_NONE) */
if (((sigqueue_t *)snqp)->sq_info.si_signo <= 0)
cmd = SN_CANCEL;
sigqsend(cmd, p, 0, (sigqueue_t *)snqp);
mutex_exit(&p->p_lock);
id.sn_pid = 0;
id.sn_index = 0;
break;
default :
return (set_errno(EINVAL));
}
if (copyout(&id, sn_id, sizeof (signotify_id_t)))
return (set_errno(EFAULT));
return (0);
}
/*
* static long rctlsys_set(char *name, rctl_opaque_t *old_rblk,
* rctl_opaque_t *new_rblk, int flags)
*
* Overview
* rctlsys_set() is the implementation of the core login of setrctl(2), which
* allows the establishment of resource control values. Flags may take on any
* of three exclusive values: RCTL_INSERT, RCTL_DELETE, and RCTL_REPLACE.
* RCTL_INSERT ignores old_rblk and inserts the value in the appropriate
* position in the ordered sequence of resource control values. RCTL_DELETE
* ignores old_rblk and deletes the first resource control value matching
* (value, priority) in the given resource block. If no matching value is
* found, -1 is returned and errno is set to ENOENT. Finally, in the case of
* RCTL_REPLACE, old_rblk is used to match (value, priority); the matching
* resource control value in the sequence is replaced with the contents of
* new_rblk. Again, if no match is found, -1 is returned and errno is set to
* ENOENT.
*
* rctlsys_set() causes a cursor test, which can reactivate resource controls
* that have previously fired.
*/
static long
rctlsys_set(char *name, rctl_opaque_t *old_rblk, rctl_opaque_t *new_rblk,
int flags)
{
rctl_val_t *nval;
rctl_dict_entry_t *rde;
rctl_opaque_t *nblk;
rctl_hndl_t hndl;
char *kname;
size_t klen;
long ret = 0;
proc_t *pp = NULL;
pid_t pid;
int action = flags & (~RCTLSYS_ACTION_MASK);
rctl_val_t *oval;
rctl_val_t *rval1;
rctl_val_t *rval2;
rctl_val_t *tval;
rctl_opaque_t *oblk;
if (flags & (~RCTLSYS_MASK))
return (set_errno(EINVAL));
if (action != RCTL_INSERT &&
action != RCTL_DELETE &&
action != RCTL_REPLACE)
return (set_errno(EINVAL));
if (new_rblk == NULL || name == NULL)
return (set_errno(EFAULT));
kname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
if (copyinstr(name, kname, MAXPATHLEN, &klen) != 0) {
kmem_free(kname, MAXPATHLEN);
return (set_errno(EFAULT));
}
if ((hndl = rctl_hndl_lookup(kname)) == -1) {
kmem_free(kname, MAXPATHLEN);
return (set_errno(EINVAL));
}
kmem_free(kname, MAXPATHLEN);
rde = rctl_dict_lookup_hndl(hndl);
nblk = kmem_alloc(sizeof (rctl_opaque_t), KM_SLEEP);
if (copyin(new_rblk, nblk, sizeof (rctl_opaque_t)) == -1) {
kmem_free(nblk, sizeof (rctl_opaque_t));
return (set_errno(EFAULT));
}
nval = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
rctlsys_rblk_xfrm(nblk, NULL, nval, RBX_FROM_BLK | RBX_VAL);
if (rctl_invalid_value(rde, nval)) {
kmem_free(nblk, sizeof (rctl_opaque_t));
kmem_cache_free(rctl_val_cache, nval);
return (set_errno(EINVAL));
}
/* allocate what we might need before potentially grabbing p_lock */
oblk = kmem_alloc(sizeof (rctl_opaque_t), KM_SLEEP);
oval = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
rval1 = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
rval2 = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
if (nval->rcv_privilege == RCPRIV_BASIC) {
if (flags & RCTL_USE_RECIPIENT_PID) {
pid = nval->rcv_action_recip_pid;
/* case for manipulating rctl values on other procs */
if (pid != curproc->p_pid) {
/* cannot be other pid on process rctls */
if (rde->rcd_entity == RCENTITY_PROCESS) {
ret = set_errno(EINVAL);
goto rctlsys_out;
}
/*
* must have privilege to manipulate controls
* on other processes
*/
if (secpolicy_rctlsys(CRED(), B_FALSE) != 0) {
ret = set_errno(EACCES);
goto rctlsys_out;
}
pid = nval->rcv_action_recip_pid;
mutex_enter(&pidlock);
pp = prfind(pid);
if (!pp) {
mutex_exit(&pidlock);
ret = set_errno(ESRCH);
goto rctlsys_out;
}
/*
* idle or zombie procs have either not yet
* set up their rctls or have already done
* their rctl_set_tearoff's.
*/
if (pp->p_stat == SZOMB ||
pp->p_stat == SIDL) {
mutex_exit(&pidlock);
ret = set_errno(ESRCH);
goto rctlsys_out;
}
/*
* hold this pp's p_lock to ensure that
* it does not do it's rctl_set_tearoff
* If we did not do this, we could
* potentially add rctls to the entity
* with a recipient that is a process
* that has exited.
*/
mutex_enter(&pp->p_lock);
mutex_exit(&pidlock);
/*
* We know that curproc's task, project,
* and zone pointers will not change
* because functions that change them
* call holdlwps(SHOLDFORK1) first.
*/
/*
* verify that the found pp is in the
* current task. If it is, then it
* is also within the current project
* and zone.
*/
if (rde->rcd_entity == RCENTITY_TASK &&
pp->p_task != curproc->p_task) {
ret = set_errno(ESRCH);
goto rctlsys_out;
}
ASSERT(pp->p_task->tk_proj ==
curproc->p_task->tk_proj);
ASSERT(pp->p_zone == curproc->p_zone);
nval->rcv_action_recipient = pp;
nval->rcv_action_recip_pid = pid;
} else {
/* for manipulating rctl values on this proc */
mutex_enter(&curproc->p_lock);
pp = curproc;
nval->rcv_action_recipient = curproc;
nval->rcv_action_recip_pid = curproc->p_pid;
}
} else {
/* RCTL_USE_RECIPIENT_PID not set, use this proc */
mutex_enter(&curproc->p_lock);
pp = curproc;
nval->rcv_action_recipient = curproc;
nval->rcv_action_recip_pid = curproc->p_pid;
}
} else {
/* privileged controls have no recipient pid */
mutex_enter(&curproc->p_lock);
pp = curproc;
nval->rcv_action_recipient = NULL;
nval->rcv_action_recip_pid = -1;
}
nval->rcv_firing_time = 0;
if (action == RCTL_REPLACE) {
if (copyin(old_rblk, oblk, sizeof (rctl_opaque_t)) == -1) {
ret = set_errno(EFAULT);
goto rctlsys_out;
}
rctlsys_rblk_xfrm(oblk, NULL, oval, RBX_FROM_BLK | RBX_VAL);
if (rctl_invalid_value(rde, oval)) {
ret = set_errno(EINVAL);
goto rctlsys_out;
}
if (oval->rcv_privilege == RCPRIV_BASIC) {
if (!(flags & RCTL_USE_RECIPIENT_PID)) {
oval->rcv_action_recipient = curproc;
oval->rcv_action_recip_pid = curproc->p_pid;
}
} else {
oval->rcv_action_recipient = NULL;
oval->rcv_action_recip_pid = -1;
}
/*
* Find the real value we're attempting to replace on the
* sequence, rather than trusting the one delivered from
* userland.
*/
if (ret = rctl_local_get(hndl, NULL, rval1, pp)) {
(void) set_errno(ret);
goto rctlsys_out;
}
do {
if (rval1->rcv_privilege == RCPRIV_SYSTEM ||
rctl_val_cmp(oval, rval1, 0) == 0)
break;
tval = rval1;
rval1 = rval2;
rval2 = tval;
} while (rctl_local_get(hndl, rval2, rval1, pp) == 0);
if (rval1->rcv_privilege == RCPRIV_SYSTEM) {
if (rctl_val_cmp(oval, rval1, 1) == 0)
ret = set_errno(EPERM);
else
ret = set_errno(ESRCH);
goto rctlsys_out;
}
bcopy(rval1, oval, sizeof (rctl_val_t));
/*
* System controls are immutable.
*/
if (nval->rcv_privilege == RCPRIV_SYSTEM) {
ret = set_errno(EPERM);
goto rctlsys_out;
}
/*
* Only privileged processes in the global zone can modify
* privileged rctls of type RCENTITY_ZONE; replacing privileged
* controls with basic ones are not allowed either. Lowering a
* lowerable one might be OK for privileged processes in a
* non-global zone, but lowerable rctls probably don't make
* sense for zones (hence, not modifiable from within a zone).
*/
if (rde->rcd_entity == RCENTITY_ZONE &&
(nval->rcv_privilege == RCPRIV_PRIVILEGED ||
oval->rcv_privilege == RCPRIV_PRIVILEGED) &&
secpolicy_rctlsys(CRED(), B_TRUE) != 0) {
ret = set_errno(EACCES);
goto rctlsys_out;
}
/*
* Must be privileged to replace a privileged control with
* a basic one.
*/
if (oval->rcv_privilege == RCPRIV_PRIVILEGED &&
nval->rcv_privilege != RCPRIV_PRIVILEGED &&
secpolicy_rctlsys(CRED(), B_FALSE) != 0) {
ret = set_errno(EACCES);
goto rctlsys_out;
}
/*
* Must have lowerable global property for non-privileged
* to lower the value of a privileged control; otherwise must
* have sufficient privileges to modify privileged controls
* at all.
*/
if (oval->rcv_privilege == RCPRIV_PRIVILEGED &&
nval->rcv_privilege == RCPRIV_PRIVILEGED &&
((((rde->rcd_flagaction & RCTL_GLOBAL_LOWERABLE) == 0) ||
oval->rcv_flagaction != nval->rcv_flagaction ||
oval->rcv_action_signal != nval->rcv_action_signal ||
oval->rcv_value < nval->rcv_value)) &&
secpolicy_rctlsys(CRED(), B_FALSE) != 0) {
ret = set_errno(EACCES);
goto rctlsys_out;
}
if (ret = rctl_local_replace(hndl, oval, nval, pp)) {
(void) set_errno(ret);
goto rctlsys_out;
}
/* ensure that nval is not freed */
nval = NULL;
} else if (action == RCTL_INSERT) {
/*
* System controls are immutable.
*/
if (nval->rcv_privilege == RCPRIV_SYSTEM) {
ret = set_errno(EPERM);
goto rctlsys_out;
}
/*
* Only privileged processes in the global zone may add
* privileged zone.* rctls. Only privileged processes
* may add other privileged rctls.
*/
if (nval->rcv_privilege == RCPRIV_PRIVILEGED) {
if ((rde->rcd_entity == RCENTITY_ZONE &&
secpolicy_rctlsys(CRED(), B_TRUE) != 0) ||
(rde->rcd_entity != RCENTITY_ZONE &&
secpolicy_rctlsys(CRED(), B_FALSE) != 0)) {
ret = set_errno(EACCES);
goto rctlsys_out;
}
}
/*
* Only one basic control is allowed per rctl.
* If a basic control is being inserted, delete
* any other basic control.
*/
if ((nval->rcv_privilege == RCPRIV_BASIC) &&
(rctl_local_get(hndl, NULL, rval1, pp) == 0)) {
do {
if (rval1->rcv_privilege == RCPRIV_BASIC &&
rval1->rcv_action_recipient == curproc) {
(void) rctl_local_delete(hndl, rval1,
pp);
if (rctl_local_get(hndl, NULL, rval1,
pp) != 0)
break;
}
tval = rval1;
rval1 = rval2;
rval2 = tval;
} while (rctl_local_get(hndl, rval2, rval1, pp)
== 0);
}
if (ret = rctl_local_insert(hndl, nval, pp)) {
(void) set_errno(ret);
goto rctlsys_out;
}
/* ensure that nval is not freed */
nval = NULL;
} else {
/*
* RCTL_DELETE
*/
if (nval->rcv_privilege == RCPRIV_SYSTEM) {
ret = set_errno(EPERM);
goto rctlsys_out;
}
if (nval->rcv_privilege == RCPRIV_PRIVILEGED) {
if ((rde->rcd_entity == RCENTITY_ZONE &&
secpolicy_rctlsys(CRED(), B_TRUE) != 0) ||
(rde->rcd_entity != RCENTITY_ZONE &&
secpolicy_rctlsys(CRED(), B_FALSE) != 0)) {
ret = set_errno(EACCES);
goto rctlsys_out;
}
}
if (ret = rctl_local_delete(hndl, nval, pp)) {
(void) set_errno(ret);
goto rctlsys_out;
}
}
rctlsys_out:
if (pp)
mutex_exit(&pp->p_lock);
kmem_free(nblk, sizeof (rctl_opaque_t));
kmem_free(oblk, sizeof (rctl_opaque_t));
/* only free nval if we did not rctl_local_insert it */
if (nval)
kmem_cache_free(rctl_val_cache, nval);
kmem_cache_free(rctl_val_cache, oval);
kmem_cache_free(rctl_val_cache, rval1);
kmem_cache_free(rctl_val_cache, rval2);
return (ret);
}
/*
* Return value:
* 1 - exitlwps() failed, call (or continue) lwp_exit()
* 0 - restarting init. Return through system call path
*/
int
proc_exit(int why, int what)
{
kthread_t *t = curthread;
klwp_t *lwp = ttolwp(t);
proc_t *p = ttoproc(t);
zone_t *z = p->p_zone;
timeout_id_t tmp_id;
int rv;
proc_t *q;
task_t *tk;
vnode_t *exec_vp, *execdir_vp, *cdir, *rdir;
sigqueue_t *sqp;
lwpdir_t *lwpdir;
uint_t lwpdir_sz;
tidhash_t *tidhash;
uint_t tidhash_sz;
ret_tidhash_t *ret_tidhash;
refstr_t *cwd;
hrtime_t hrutime, hrstime;
int evaporate;
brand_t *orig_brand = NULL;
void *brand_data = NULL;
/*
* Stop and discard the process's lwps except for the current one,
* unless some other lwp beat us to it. If exitlwps() fails then
* return and the calling lwp will call (or continue in) lwp_exit().
*/
proc_is_exiting(p);
if (exitlwps(0) != 0)
return (1);
mutex_enter(&p->p_lock);
if (p->p_ttime > 0) {
/*
* Account any remaining ticks charged to this process
* on its way out.
*/
(void) task_cpu_time_incr(p->p_task, p->p_ttime);
p->p_ttime = 0;
}
mutex_exit(&p->p_lock);
DTRACE_PROC(lwp__exit);
DTRACE_PROC1(exit, int, why);
/*
* Will perform any brand specific proc exit processing. Since this
* is always the last lwp, will also perform lwp_exit and free
* brand_data, except in the case that the brand has a b_exit_with_sig
* handler. In this case we free the brand_data later within this
* function.
*/
mutex_enter(&p->p_lock);
if (PROC_IS_BRANDED(p)) {
orig_brand = p->p_brand;
if (p->p_brand_data != NULL && orig_brand->b_data_size > 0) {
brand_data = p->p_brand_data;
}
lwp_detach_brand_hdlrs(lwp);
brand_clearbrand(p, B_FALSE);
}
mutex_exit(&p->p_lock);
/*
* Don't let init exit unless zone_start_init() failed its exec, or
* we are shutting down the zone or the machine.
*
* Since we are single threaded, we don't need to lock the
* following accesses to zone_proc_initpid.
*/
if (p->p_pid == z->zone_proc_initpid) {
if (z->zone_boot_err == 0 &&
zone_status_get(z) < ZONE_IS_SHUTTING_DOWN &&
zone_status_get(global_zone) < ZONE_IS_SHUTTING_DOWN) {
if (z->zone_restart_init == B_TRUE) {
if (restart_init(what, why) == 0)
return (0);
}
z->zone_init_status = wstat(why, what);
(void) zone_kadmin(A_SHUTDOWN, AD_HALT, NULL, CRED());
}
/*
* Since we didn't or couldn't restart init, we clear
* the zone's init state and proceed with exit
* processing.
*/
z->zone_proc_initpid = -1;
}
lwp_pcb_exit();
/*
* Allocate a sigqueue now, before we grab locks.
* It will be given to sigcld(), below.
* Special case: If we will be making the process disappear
* without a trace because it is either:
* * an exiting SSYS process, or
* * a posix_spawn() vfork child who requests it,
* we don't bother to allocate a useless sigqueue.
*/
evaporate = (p->p_flag & SSYS) || ((p->p_flag & SVFORK) &&
why == CLD_EXITED && what == _EVAPORATE);
if (!evaporate)
sqp = kmem_zalloc(sizeof (sigqueue_t), KM_SLEEP);
/*
* revoke any doors created by the process.
*/
if (p->p_door_list)
door_exit();
/*
* Release schedctl data structures.
*/
if (p->p_pagep)
schedctl_proc_cleanup();
/*
* make sure all pending kaio has completed.
*/
if (p->p_aio)
aio_cleanup_exit();
/*
* discard the lwpchan cache.
*/
if (p->p_lcp != NULL)
lwpchan_destroy_cache(0);
/*
* Clean up any DTrace helper actions or probes for the process.
*/
if (p->p_dtrace_helpers != NULL) {
ASSERT(dtrace_helpers_cleanup != NULL);
(*dtrace_helpers_cleanup)();
}
/* untimeout the realtime timers */
if (p->p_itimer != NULL)
timer_exit();
if ((tmp_id = p->p_alarmid) != 0) {
p->p_alarmid = 0;
(void) untimeout(tmp_id);
}
/*
* Remove any fpollinfo_t's for this (last) thread from our file
* descriptors so closeall() can ASSERT() that they're all gone.
*/
pollcleanup();
if (p->p_rprof_cyclic != CYCLIC_NONE) {
mutex_enter(&cpu_lock);
cyclic_remove(p->p_rprof_cyclic);
mutex_exit(&cpu_lock);
}
mutex_enter(&p->p_lock);
/*
* Clean up any DTrace probes associated with this process.
*/
if (p->p_dtrace_probes) {
ASSERT(dtrace_fasttrap_exit_ptr != NULL);
dtrace_fasttrap_exit_ptr(p);
}
while ((tmp_id = p->p_itimerid) != 0) {
p->p_itimerid = 0;
mutex_exit(&p->p_lock);
(void) untimeout(tmp_id);
mutex_enter(&p->p_lock);
}
lwp_cleanup();
/*
* We are about to exit; prevent our resource associations from
* being changed.
*/
pool_barrier_enter();
/*
* Block the process against /proc now that we have really
* acquired p->p_lock (to manipulate p_tlist at least).
*/
prbarrier(p);
sigfillset(&p->p_ignore);
sigemptyset(&p->p_siginfo);
sigemptyset(&p->p_sig);
sigemptyset(&p->p_extsig);
sigemptyset(&t->t_sig);
sigemptyset(&t->t_extsig);
sigemptyset(&p->p_sigmask);
sigdelq(p, t, 0);
lwp->lwp_cursig = 0;
lwp->lwp_extsig = 0;
p->p_flag &= ~(SKILLED | SEXTKILLED);
if (lwp->lwp_curinfo) {
siginfofree(lwp->lwp_curinfo);
lwp->lwp_curinfo = NULL;
}
t->t_proc_flag |= TP_LWPEXIT;
ASSERT(p->p_lwpcnt == 1 && p->p_zombcnt == 0);
prlwpexit(t); /* notify /proc */
lwp_hash_out(p, t->t_tid);
prexit(p);
p->p_lwpcnt = 0;
p->p_tlist = NULL;
sigqfree(p);
term_mstate(t);
p->p_mterm = gethrtime();
exec_vp = p->p_exec;
execdir_vp = p->p_execdir;
p->p_exec = NULLVP;
p->p_execdir = NULLVP;
mutex_exit(&p->p_lock);
pr_free_watched_pages(p);
closeall(P_FINFO(p));
/* Free the controlling tty. (freectty() always assumes curproc.) */
ASSERT(p == curproc);
(void) freectty(B_TRUE);
#if defined(__sparc)
if (p->p_utraps != NULL)
utrap_free(p);
#endif
if (p->p_semacct) /* IPC semaphore exit */
semexit(p);
rv = wstat(why, what);
acct(rv & 0xff);
exacct_commit_proc(p, rv);
/*
* Release any resources associated with C2 auditing
*/
if (AU_AUDITING()) {
/*
* audit exit system call
*/
audit_exit(why, what);
}
/*
* Free address space.
*/
relvm();
if (exec_vp) {
/*
* Close this executable which has been opened when the process
* was created by getproc().
*/
(void) VOP_CLOSE(exec_vp, FREAD, 1, (offset_t)0, CRED(), NULL);
VN_RELE(exec_vp);
}
if (execdir_vp)
VN_RELE(execdir_vp);
/*
* Release held contracts.
*/
contract_exit(p);
/*
* Depart our encapsulating process contract.
*/
if ((p->p_flag & SSYS) == 0) {
ASSERT(p->p_ct_process);
contract_process_exit(p->p_ct_process, p, rv);
}
/*
* Remove pool association, and block if requested by pool_do_bind.
*/
mutex_enter(&p->p_lock);
ASSERT(p->p_pool->pool_ref > 0);
atomic_dec_32(&p->p_pool->pool_ref);
p->p_pool = pool_default;
/*
* Now that our address space has been freed and all other threads
* in this process have exited, set the PEXITED pool flag. This
* tells the pools subsystems to ignore this process if it was
* requested to rebind this process to a new pool.
*/
p->p_poolflag |= PEXITED;
pool_barrier_exit();
mutex_exit(&p->p_lock);
mutex_enter(&pidlock);
/*
* Delete this process from the newstate list of its parent. We
* will put it in the right place in the sigcld in the end.
*/
delete_ns(p->p_parent, p);
/*
* Reassign the orphans to the next of kin.
* Don't rearrange init's orphanage.
*/
if ((q = p->p_orphan) != NULL && p != proc_init) {
proc_t *nokp = p->p_nextofkin;
for (;;) {
q->p_nextofkin = nokp;
if (q->p_nextorph == NULL)
break;
q = q->p_nextorph;
}
q->p_nextorph = nokp->p_orphan;
nokp->p_orphan = p->p_orphan;
p->p_orphan = NULL;
}
/*
* Reassign the children to init.
* Don't try to assign init's children to init.
*/
if ((q = p->p_child) != NULL && p != proc_init) {
struct proc *np;
struct proc *initp = proc_init;
pid_t zone_initpid = 1;
struct proc *zoneinitp = NULL;
boolean_t setzonetop = B_FALSE;
if (!INGLOBALZONE(curproc)) {
zone_initpid = curproc->p_zone->zone_proc_initpid;
ASSERT(MUTEX_HELD(&pidlock));
zoneinitp = prfind(zone_initpid);
if (zoneinitp != NULL) {
initp = zoneinitp;
} else {
zone_initpid = 1;
setzonetop = B_TRUE;
}
}
pgdetach(p);
do {
np = q->p_sibling;
/*
* Delete it from its current parent new state
* list and add it to init new state list
*/
delete_ns(q->p_parent, q);
q->p_ppid = zone_initpid;
q->p_pidflag &= ~(CLDNOSIGCHLD | CLDWAITPID);
if (setzonetop) {
mutex_enter(&q->p_lock);
q->p_flag |= SZONETOP;
mutex_exit(&q->p_lock);
}
q->p_parent = initp;
/*
* Since q will be the first child,
* it will not have a previous sibling.
*/
q->p_psibling = NULL;
if (initp->p_child) {
initp->p_child->p_psibling = q;
}
q->p_sibling = initp->p_child;
initp->p_child = q;
if (q->p_proc_flag & P_PR_PTRACE) {
mutex_enter(&q->p_lock);
sigtoproc(q, NULL, SIGKILL);
mutex_exit(&q->p_lock);
}
/*
* sigcld() will add the child to parents
* newstate list.
*/
if (q->p_stat == SZOMB)
sigcld(q, NULL);
} while ((q = np) != NULL);
p->p_child = NULL;
ASSERT(p->p_child_ns == NULL);
}
TRACE_1(TR_FAC_PROC, TR_PROC_EXIT, "proc_exit: %p", p);
mutex_enter(&p->p_lock);
CL_EXIT(curthread); /* tell the scheduler that curthread is exiting */
/*
* Have our task accummulate our resource usage data before they
* become contaminated by p_cacct etc., and before we renounce
* membership of the task.
*
* We do this regardless of whether or not task accounting is active.
* This is to avoid having nonsense data reported for this task if
* task accounting is subsequently enabled. The overhead is minimal;
* by this point, this process has accounted for the usage of all its
* LWPs. We nonetheless do the work here, and under the protection of
* pidlock, so that the movement of the process's usage to the task
* happens at the same time as the removal of the process from the
* task, from the point of view of exacct_snapshot_task_usage().
*/
exacct_update_task_mstate(p);
hrutime = mstate_aggr_state(p, LMS_USER);
hrstime = mstate_aggr_state(p, LMS_SYSTEM);
p->p_utime = (clock_t)NSEC_TO_TICK(hrutime) + p->p_cutime;
p->p_stime = (clock_t)NSEC_TO_TICK(hrstime) + p->p_cstime;
p->p_acct[LMS_USER] += p->p_cacct[LMS_USER];
p->p_acct[LMS_SYSTEM] += p->p_cacct[LMS_SYSTEM];
p->p_acct[LMS_TRAP] += p->p_cacct[LMS_TRAP];
p->p_acct[LMS_TFAULT] += p->p_cacct[LMS_TFAULT];
p->p_acct[LMS_DFAULT] += p->p_cacct[LMS_DFAULT];
p->p_acct[LMS_KFAULT] += p->p_cacct[LMS_KFAULT];
p->p_acct[LMS_USER_LOCK] += p->p_cacct[LMS_USER_LOCK];
p->p_acct[LMS_SLEEP] += p->p_cacct[LMS_SLEEP];
p->p_acct[LMS_WAIT_CPU] += p->p_cacct[LMS_WAIT_CPU];
p->p_acct[LMS_STOPPED] += p->p_cacct[LMS_STOPPED];
p->p_ru.minflt += p->p_cru.minflt;
p->p_ru.majflt += p->p_cru.majflt;
p->p_ru.nswap += p->p_cru.nswap;
p->p_ru.inblock += p->p_cru.inblock;
p->p_ru.oublock += p->p_cru.oublock;
p->p_ru.msgsnd += p->p_cru.msgsnd;
p->p_ru.msgrcv += p->p_cru.msgrcv;
p->p_ru.nsignals += p->p_cru.nsignals;
p->p_ru.nvcsw += p->p_cru.nvcsw;
p->p_ru.nivcsw += p->p_cru.nivcsw;
p->p_ru.sysc += p->p_cru.sysc;
p->p_ru.ioch += p->p_cru.ioch;
p->p_stat = SZOMB;
p->p_proc_flag &= ~P_PR_PTRACE;
p->p_wdata = what;
p->p_wcode = (char)why;
cdir = PTOU(p)->u_cdir;
rdir = PTOU(p)->u_rdir;
cwd = PTOU(p)->u_cwd;
ASSERT(cdir != NULL || p->p_parent == &p0);
/*
* Release resource controls, as they are no longer enforceable.
*/
rctl_set_free(p->p_rctls);
/*
* Decrement tk_nlwps counter for our task.max-lwps resource control.
* An extended accounting record, if that facility is active, is
* scheduled to be written. We cannot give up task and project
* membership at this point because that would allow zombies to escape
* from the max-processes resource controls. Zombies stay in their
* current task and project until the process table slot is released
* in freeproc().
*/
tk = p->p_task;
mutex_enter(&p->p_zone->zone_nlwps_lock);
tk->tk_nlwps--;
tk->tk_proj->kpj_nlwps--;
p->p_zone->zone_nlwps--;
mutex_exit(&p->p_zone->zone_nlwps_lock);
/*
* Clear the lwp directory and the lwpid hash table
* now that /proc can't bother us any more.
* We free the memory below, after dropping p->p_lock.
*/
lwpdir = p->p_lwpdir;
lwpdir_sz = p->p_lwpdir_sz;
tidhash = p->p_tidhash;
tidhash_sz = p->p_tidhash_sz;
ret_tidhash = p->p_ret_tidhash;
p->p_lwpdir = NULL;
p->p_lwpfree = NULL;
p->p_lwpdir_sz = 0;
p->p_tidhash = NULL;
p->p_tidhash_sz = 0;
p->p_ret_tidhash = NULL;
/*
* If the process has context ops installed, call the exit routine
* on behalf of this last remaining thread. Normally exitpctx() is
* called during thread_exit() or lwp_exit(), but because this is the
* last thread in the process, we must call it here. By the time
* thread_exit() is called (below), the association with the relevant
* process has been lost.
*
* We also free the context here.
*/
if (p->p_pctx) {
kpreempt_disable();
exitpctx(p);
kpreempt_enable();
freepctx(p, 0);
}
/*
* curthread's proc pointer is changed to point to the 'sched'
* process for the corresponding zone, except in the case when
* the exiting process is in fact a zsched instance, in which
* case the proc pointer is set to p0. We do so, so that the
* process still points at the right zone when we call the VN_RELE()
* below.
*
* This is because curthread's original proc pointer can be freed as
* soon as the child sends a SIGCLD to its parent. We use zsched so
* that for user processes, even in the final moments of death, the
* process is still associated with its zone.
*/
if (p != t->t_procp->p_zone->zone_zsched)
t->t_procp = t->t_procp->p_zone->zone_zsched;
else
t->t_procp = &p0;
mutex_exit(&p->p_lock);
if (!evaporate) {
/*
* The brand specific code only happens when the brand has a
* function to call in place of sigcld, the data itself still
* existed, and the parent of the exiting process is not the
* global zone init. If the parent is the global zone init,
* then the process was reparented, and we don't want brand
* code delivering possibly strange signals to init. Also, init
* is not branded, so any brand specific exit data will not be
* picked up by init anyway.
* It is assumed by this code that any brand where
* b_exit_with_sig == NULL, will free its own brand_data rather
* than letting this piece of code free it.
*/
if (orig_brand != NULL &&
orig_brand->b_ops->b_exit_with_sig != NULL &&
brand_data != NULL && p->p_ppid != 1) {
/*
* The code for _fini that could unload the brand_t
* blocks until the count of zones using the module
* reaches zero. Zones decrement the refcount on their
* brands only after all user tasks in that zone have
* exited and been waited on. The decrement on the
* brand's refcount happen in zone_destroy(). That
* depends on zone_shutdown() having been completed.
* zone_shutdown() includes a call to zone_empty(),
* where the zone waits for itself to reach the state
* ZONE_IS_EMPTY. This state is only set in either
* zone_shutdown(), when there are no user processes as
* the zone enters this function, or in
* zone_task_rele(). zone_task_rele() is called from
* code triggered by waiting on processes, not by the
* processes exiting through proc_exit(). This means
* all the branded processes that could exist for a
* specific brand_t must exit and get reaped before the
* refcount on the brand_t can reach 0. _fini will
* never unload the corresponding brand module before
* proc_exit finishes execution for all processes
* branded with a particular brand_t, which makes the
* operation below safe to do. Brands that wish to use
* this mechanism must wait in _fini as described
* above.
*/
orig_brand->b_ops->b_exit_with_sig(p,
sqp, brand_data);
} else {
p->p_pidflag &= ~CLDPEND;
sigcld(p, sqp);
}
if (brand_data != NULL) {
kmem_free(brand_data, orig_brand->b_data_size);
brand_data = NULL;
orig_brand = NULL;
}
} else {
/*
* Do what sigcld() would do if the disposition
* of the SIGCHLD signal were set to be ignored.
*/
cv_broadcast(&p->p_srwchan_cv);
freeproc(p);
}
mutex_exit(&pidlock);
/*
* We don't release u_cdir and u_rdir until SZOMB is set.
* This protects us against dofusers().
*/
if (cdir)
VN_RELE(cdir);
if (rdir)
VN_RELE(rdir);
if (cwd)
refstr_rele(cwd);
/*
* task_rele() may ultimately cause the zone to go away (or
* may cause the last user process in a zone to go away, which
* signals zsched to go away). So prior to this call, we must
* no longer point at zsched.
*/
t->t_procp = &p0;
kmem_free(lwpdir, lwpdir_sz * sizeof (lwpdir_t));
kmem_free(tidhash, tidhash_sz * sizeof (tidhash_t));
while (ret_tidhash != NULL) {
ret_tidhash_t *next = ret_tidhash->rth_next;
kmem_free(ret_tidhash->rth_tidhash,
ret_tidhash->rth_tidhash_sz * sizeof (tidhash_t));
kmem_free(ret_tidhash, sizeof (*ret_tidhash));
ret_tidhash = next;
}
thread_exit();
/* NOTREACHED */
}