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));
}
static int
wracct_task(ac_info_t *ac_task, taskid_t tkid, int flag, size_t *sizep)
{
task_t *tk;
int error;
mutex_enter(&ac_task->ac_lock);
if (ac_task->ac_state == AC_OFF || ac_task->ac_vnode == NULL) {
mutex_exit(&ac_task->ac_lock);
return (ENOTACTIVE);
}
mutex_exit(&ac_task->ac_lock);
if ((tk = task_hold_by_id(tkid)) == NULL)
return (ESRCH);
error = exacct_assemble_task_usage(ac_task, tk, exacct_commit_callback,
NULL, 0, sizep, flag);
task_rele(tk);
return (error);
}
static int
getacct_task(ac_info_t *ac_task, taskid_t tkid, void *buf, size_t bufsize,
size_t *sizep)
{
task_t *tk;
int error;
mutex_enter(&ac_task->ac_lock);
if (ac_task->ac_state == AC_OFF) {
mutex_exit(&ac_task->ac_lock);
return (ENOTACTIVE);
}
mutex_exit(&ac_task->ac_lock);
if ((tk = task_hold_by_id(tkid)) == NULL)
return (ESRCH);
error = exacct_assemble_task_usage(ac_task, tk,
getacct_callback, buf, bufsize, sizep, EW_PARTIAL);
task_rele(tk);
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);
}
/*
* task_t *task_create(projid_t, zone *)
*
* Overview
* A process constructing a new task calls task_create() to construct and
* preinitialize the task for the appropriate destination project. Only one
* task, the primordial task0, is not created with task_create().
*
* Return values
* None.
*
* Caller's context
* Caller's context should be safe for KM_SLEEP allocations.
* The caller should appropriately bump the kpj_ntasks counter on the
* project that contains this task.
*/
task_t *
task_create(projid_t projid, zone_t *zone)
{
task_t *tk = kmem_cache_alloc(task_cache, KM_SLEEP);
task_t *ancestor_tk;
taskid_t tkid;
task_usage_t *tu = kmem_zalloc(sizeof (task_usage_t), KM_SLEEP);
mod_hash_hndl_t hndl;
rctl_set_t *set = rctl_set_create();
rctl_alloc_gp_t *gp;
rctl_entity_p_t e;
bzero(tk, sizeof (task_t));
tk->tk_tkid = tkid = id_alloc(taskid_space);
tk->tk_nlwps = 0;
tk->tk_nlwps_ctl = INT_MAX;
tk->tk_nprocs = 0;
tk->tk_nprocs_ctl = INT_MAX;
tk->tk_usage = tu;
tk->tk_inherited = kmem_zalloc(sizeof (task_usage_t), KM_SLEEP);
tk->tk_proj = project_hold_by_id(projid, zone, PROJECT_HOLD_INSERT);
tk->tk_flags = TASK_NORMAL;
tk->tk_commit_next = NULL;
/*
* Copy ancestor task's resource controls.
*/
zone_task_hold(zone);
mutex_enter(&curproc->p_lock);
ancestor_tk = curproc->p_task;
task_hold(ancestor_tk);
tk->tk_zone = zone;
mutex_exit(&curproc->p_lock);
for (;;) {
gp = rctl_set_dup_prealloc(ancestor_tk->tk_rctls);
mutex_enter(&ancestor_tk->tk_rctls->rcs_lock);
if (rctl_set_dup_ready(ancestor_tk->tk_rctls, gp))
break;
mutex_exit(&ancestor_tk->tk_rctls->rcs_lock);
rctl_prealloc_destroy(gp);
}
/*
* At this point, curproc does not have the appropriate linkage
* through the task to the project. So, rctl_set_dup should only
* copy the rctls, and leave the callbacks for later.
*/
e.rcep_p.task = tk;
e.rcep_t = RCENTITY_TASK;
tk->tk_rctls = rctl_set_dup(ancestor_tk->tk_rctls, curproc, curproc, &e,
set, gp, RCD_DUP);
mutex_exit(&ancestor_tk->tk_rctls->rcs_lock);
rctl_prealloc_destroy(gp);
/*
* Record the ancestor task's ID for use by extended accounting.
*/
tu->tu_anctaskid = ancestor_tk->tk_tkid;
task_rele(ancestor_tk);
/*
* Put new task structure in the hash table.
*/
(void) mod_hash_reserve(task_hash, &hndl);
mutex_enter(&task_hash_lock);
ASSERT(task_find(tkid, zone->zone_id) == NULL);
if (mod_hash_insert_reserve(task_hash, (mod_hash_key_t)(uintptr_t)tkid,
(mod_hash_val_t *)tk, hndl) != 0) {
mod_hash_cancel(task_hash, &hndl);
panic("unable to insert task %d(%p)", tkid, (void *)tk);
}
mutex_exit(&task_hash_lock);
tk->tk_nprocs_kstat = task_kstat_create(tk, zone);
return (tk);
}
/*
* Remove zombie children from the process table.
*/
void
freeproc(proc_t *p)
{
proc_t *q;
task_t *tk;
ASSERT(p->p_stat == SZOMB);
ASSERT(p->p_tlist == NULL);
ASSERT(MUTEX_HELD(&pidlock));
sigdelq(p, NULL, 0);
if (p->p_killsqp) {
siginfofree(p->p_killsqp);
p->p_killsqp = NULL;
}
prfree(p); /* inform /proc */
/*
* Don't free the init processes.
* Other dying processes will access it.
*/
if (p == proc_init)
return;
/*
* We wait until now to free the cred structure because a
* zombie process's credentials may be examined by /proc.
* No cred locking needed because there are no threads at this point.
*/
upcount_dec(crgetruid(p->p_cred), crgetzoneid(p->p_cred));
crfree(p->p_cred);
if (p->p_corefile != NULL) {
corectl_path_rele(p->p_corefile);
p->p_corefile = NULL;
}
if (p->p_content != NULL) {
corectl_content_rele(p->p_content);
p->p_content = NULL;
}
if (p->p_nextofkin && !((p->p_nextofkin->p_flag & SNOWAIT) ||
(PTOU(p->p_nextofkin)->u_signal[SIGCLD - 1] == SIG_IGN))) {
/*
* This should still do the right thing since p_utime/stime
* get set to the correct value on process exit, so it
* should get properly updated
*/
p->p_nextofkin->p_cutime += p->p_utime;
p->p_nextofkin->p_cstime += p->p_stime;
p->p_nextofkin->p_cacct[LMS_USER] += p->p_acct[LMS_USER];
p->p_nextofkin->p_cacct[LMS_SYSTEM] += p->p_acct[LMS_SYSTEM];
p->p_nextofkin->p_cacct[LMS_TRAP] += p->p_acct[LMS_TRAP];
p->p_nextofkin->p_cacct[LMS_TFAULT] += p->p_acct[LMS_TFAULT];
p->p_nextofkin->p_cacct[LMS_DFAULT] += p->p_acct[LMS_DFAULT];
p->p_nextofkin->p_cacct[LMS_KFAULT] += p->p_acct[LMS_KFAULT];
p->p_nextofkin->p_cacct[LMS_USER_LOCK]
+= p->p_acct[LMS_USER_LOCK];
p->p_nextofkin->p_cacct[LMS_SLEEP] += p->p_acct[LMS_SLEEP];
p->p_nextofkin->p_cacct[LMS_WAIT_CPU]
+= p->p_acct[LMS_WAIT_CPU];
p->p_nextofkin->p_cacct[LMS_STOPPED] += p->p_acct[LMS_STOPPED];
p->p_nextofkin->p_cru.minflt += p->p_ru.minflt;
p->p_nextofkin->p_cru.majflt += p->p_ru.majflt;
p->p_nextofkin->p_cru.nswap += p->p_ru.nswap;
p->p_nextofkin->p_cru.inblock += p->p_ru.inblock;
p->p_nextofkin->p_cru.oublock += p->p_ru.oublock;
p->p_nextofkin->p_cru.msgsnd += p->p_ru.msgsnd;
p->p_nextofkin->p_cru.msgrcv += p->p_ru.msgrcv;
p->p_nextofkin->p_cru.nsignals += p->p_ru.nsignals;
p->p_nextofkin->p_cru.nvcsw += p->p_ru.nvcsw;
p->p_nextofkin->p_cru.nivcsw += p->p_ru.nivcsw;
p->p_nextofkin->p_cru.sysc += p->p_ru.sysc;
p->p_nextofkin->p_cru.ioch += p->p_ru.ioch;
}
q = p->p_nextofkin;
if (q && q->p_orphan == p)
q->p_orphan = p->p_nextorph;
else if (q) {
for (q = q->p_orphan; q; q = q->p_nextorph)
if (q->p_nextorph == p)
break;
ASSERT(q && q->p_nextorph == p);
q->p_nextorph = p->p_nextorph;
}
/*
* The process table slot is being freed, so it is now safe to give up
* task and project membership.
*/
mutex_enter(&p->p_lock);
tk = p->p_task;
task_detach(p);
mutex_exit(&p->p_lock);
proc_detach(p);
pid_exit(p, tk); /* frees pid and proc structure */
task_rele(tk);
}
/*
* taskid_t tasksys_settaskid(projid_t projid, uint_t flags);
*
* Overview
* Place the calling process in a new task if sufficiently privileged. If the
* present task is finalized, the process may not create a new task.
*
* Return values
* 0 on success, errno on failure.
*/
static long
tasksys_settaskid(projid_t projid, uint_t flags)
{
proc_t *p = ttoproc(curthread);
kproject_t *oldpj;
kproject_t *kpj;
task_t *tk, *oldtk;
rctl_entity_p_t e;
zone_t *zone;
int rctlfail = 0;
if (secpolicy_tasksys(CRED()) != 0)
return (set_errno(EPERM));
if (projid < 0 || projid > MAXPROJID)
return (set_errno(EINVAL));
if (flags & ~TASK_FINAL)
return (set_errno(EINVAL));
mutex_enter(&pidlock);
if (p->p_task->tk_flags & TASK_FINAL) {
mutex_exit(&pidlock);
return (set_errno(EACCES));
}
mutex_exit(&pidlock);
/*
* Try to stop all other lwps in the process while we're changing
* our project. This way, curthread doesn't need to grab its own
* thread_lock to find its project ID (see curprojid()). If this
* is the /proc agent lwp, we know that the other lwps are already
* held. If we failed to hold all lwps, bail out and return EINTR.
*/
if (curthread != p->p_agenttp && !holdlwps(SHOLDFORK1))
return (set_errno(EINTR));
/*
* Put a hold on our new project and make sure that nobody is
* trying to bind it to a pool while we're joining.
*/
kpj = project_hold_by_id(projid, p->p_zone, PROJECT_HOLD_INSERT);
e.rcep_p.proj = kpj;
e.rcep_t = RCENTITY_PROJECT;
mutex_enter(&p->p_lock);
oldpj = p->p_task->tk_proj;
zone = p->p_zone;
mutex_enter(&zone->zone_nlwps_lock);
mutex_enter(&zone->zone_mem_lock);
if (kpj->kpj_nlwps + p->p_lwpcnt > kpj->kpj_nlwps_ctl)
if (rctl_test_entity(rc_project_nlwps, kpj->kpj_rctls, p, &e,
p->p_lwpcnt, 0) & RCT_DENY)
rctlfail = 1;
if (kpj->kpj_ntasks + 1 > kpj->kpj_ntasks_ctl)
if (rctl_test_entity(rc_project_ntasks, kpj->kpj_rctls, p, &e,
1, 0) & RCT_DENY)
rctlfail = 1;
if (kpj->kpj_data.kpd_locked_mem + p->p_locked_mem >
kpj->kpj_data.kpd_locked_mem_ctl)
if (rctl_test_entity(rc_project_locked_mem, kpj->kpj_rctls, p,
&e, p->p_locked_mem, 0) & RCT_DENY)
rctlfail = 1;
mutex_enter(&(kpj->kpj_data.kpd_crypto_lock));
if (kpj->kpj_data.kpd_crypto_mem + p->p_crypto_mem >
kpj->kpj_data.kpd_crypto_mem_ctl)
if (rctl_test_entity(rc_project_crypto_mem, kpj->kpj_rctls, p,
&e, p->p_crypto_mem, 0) & RCT_DENY)
rctlfail = 1;
if (rctlfail) {
mutex_exit(&(kpj->kpj_data.kpd_crypto_lock));
mutex_exit(&zone->zone_mem_lock);
mutex_exit(&zone->zone_nlwps_lock);
if (curthread != p->p_agenttp)
continuelwps(p);
mutex_exit(&p->p_lock);
return (set_errno(EAGAIN));
}
kpj->kpj_data.kpd_crypto_mem += p->p_crypto_mem;
mutex_exit(&(kpj->kpj_data.kpd_crypto_lock));
kpj->kpj_data.kpd_locked_mem += p->p_locked_mem;
kpj->kpj_nlwps += p->p_lwpcnt;
kpj->kpj_ntasks++;
oldpj->kpj_data.kpd_locked_mem -= p->p_locked_mem;
mutex_enter(&(oldpj->kpj_data.kpd_crypto_lock));
oldpj->kpj_data.kpd_crypto_mem -= p->p_crypto_mem;
mutex_exit(&(oldpj->kpj_data.kpd_crypto_lock));
oldpj->kpj_nlwps -= p->p_lwpcnt;
mutex_exit(&zone->zone_mem_lock);
mutex_exit(&zone->zone_nlwps_lock);
mutex_exit(&p->p_lock);
mutex_enter(&kpj->kpj_poolbind);
tk = task_create(projid, curproc->p_zone);
mutex_enter(&cpu_lock);
/*
* Returns with p_lock held.
*/
oldtk = task_join(tk, flags);
if (curthread != p->p_agenttp)
continuelwps(p);
mutex_exit(&p->p_lock);
mutex_exit(&cpu_lock);
mutex_exit(&kpj->kpj_poolbind);
task_rele(oldtk);
project_rele(kpj);
return (tk->tk_tkid);
}