/* ARGSUSED */
static int64_t
cfork(int isvfork, int isfork1, int flags)
{
proc_t *p = ttoproc(curthread);
struct as *as;
proc_t *cp, **orphpp;
klwp_t *clone;
kthread_t *t;
task_t *tk;
rval_t r;
int error;
int i;
rctl_set_t *dup_set;
rctl_alloc_gp_t *dup_gp;
rctl_entity_p_t e;
lwpdir_t *ldp;
lwpent_t *lep;
lwpent_t *clep;
/*
* Allow only these two flags.
*/
if ((flags & ~(FORK_NOSIGCHLD | FORK_WAITPID)) != 0) {
error = EINVAL;
goto forkerr;
}
/*
* fork is not supported for the /proc agent lwp.
*/
if (curthread == p->p_agenttp) {
error = ENOTSUP;
goto forkerr;
}
if ((error = secpolicy_basic_fork(CRED())) != 0)
goto forkerr;
/*
* If the calling lwp is doing a fork1() then the
* other lwps in this process are not duplicated and
* don't need to be held where their kernel stacks can be
* cloned. If doing forkall(), the process is held with
* SHOLDFORK, so that the lwps are at a point where their
* stacks can be copied which is on entry or exit from
* the kernel.
*/
if (!holdlwps(isfork1 ? SHOLDFORK1 : SHOLDFORK)) {
aston(curthread);
error = EINTR;
goto forkerr;
}
#if defined(__sparc)
/*
* Ensure that the user stack is fully constructed
* before creating the child process structure.
*/
(void) flush_user_windows_to_stack(NULL);
#endif
mutex_enter(&p->p_lock);
/*
* If this is vfork(), cancel any suspend request we might
* have gotten from some other thread via lwp_suspend().
* Otherwise we could end up with a deadlock on return
* from the vfork() in both the parent and the child.
*/
if (isvfork)
curthread->t_proc_flag &= ~TP_HOLDLWP;
/*
* Prevent our resource set associations from being changed during fork.
*/
pool_barrier_enter();
mutex_exit(&p->p_lock);
/*
* Create a child proc struct. Place a VN_HOLD on appropriate vnodes.
*/
if (getproc(&cp, 0) < 0) {
mutex_enter(&p->p_lock);
pool_barrier_exit();
continuelwps(p);
mutex_exit(&p->p_lock);
error = EAGAIN;
goto forkerr;
}
TRACE_2(TR_FAC_PROC, TR_PROC_FORK, "proc_fork:cp %p p %p", cp, p);
/*
* Assign an address space to child
*/
if (isvfork) {
/*
* Clear any watched areas and remember the
* watched pages for restoring in vfwait().
*/
as = p->p_as;
if (avl_numnodes(&as->a_wpage) != 0) {
AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER);
as_clearwatch(as);
p->p_wpage = as->a_wpage;
avl_create(&as->a_wpage, wp_compare,
sizeof (struct watched_page),
offsetof(struct watched_page, wp_link));
AS_LOCK_EXIT(as, &as->a_lock);
}
cp->p_as = as;
cp->p_flag |= SVFORK;
} else {
/*
* Common code for pr_mappage() and pr_unmappage().
*/
static int
pr_do_mappage(caddr_t addr, size_t size, int mapin, enum seg_rw rw, int kernel)
{
proc_t *p = curproc;
struct as *as = p->p_as;
char *eaddr = addr + size;
int prot_rw = rw_to_prot(rw);
int xrw = rw_to_index(rw);
int rv = 0;
struct watched_page *pwp;
struct watched_page tpw;
avl_index_t where;
uint_t prot;
ASSERT(as != &kas);
startover:
ASSERT(rv == 0);
if (avl_numnodes(&as->a_wpage) == 0)
return (0);
/*
* as->a_wpage can only be changed while the process is totally stopped.
* Don't grab p_lock here. Holding p_lock while grabbing the address
* space lock leads to deadlocks with the clock thread. Note that if an
* as_fault() is servicing a fault to a watched page on behalf of an
* XHAT provider, watchpoint will be temporarily cleared (and wp_prot
* will be set to wp_oprot). Since this is done while holding as writer
* lock, we need to grab as lock (reader lock is good enough).
*
* p_maplock prevents simultaneous execution of this function. Under
* normal circumstances, holdwatch() will stop all other threads, so the
* lock isn't really needed. But there may be multiple threads within
* stop() when SWATCHOK is set, so we need to handle multiple threads
* at once. See holdwatch() for the details of this dance.
*/
mutex_enter(&p->p_maplock);
AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
tpw.wp_vaddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK);
if ((pwp = avl_find(&as->a_wpage, &tpw, &where)) == NULL)
pwp = avl_nearest(&as->a_wpage, where, AVL_AFTER);
for (; pwp != NULL && pwp->wp_vaddr < eaddr;
pwp = AVL_NEXT(&as->a_wpage, pwp)) {
/*
* If the requested protection has not been
* removed, we need not remap this page.
*/
prot = pwp->wp_prot;
if (kernel || (prot & PROT_USER))
if (prot & prot_rw)
continue;
/*
* If the requested access does not exist in the page's
* original protections, we need not remap this page.
* If the page does not exist yet, we can't test it.
*/
if ((prot = pwp->wp_oprot) != 0) {
if (!(kernel || (prot & PROT_USER)))
continue;
if (!(prot & prot_rw))
continue;
}
if (mapin) {
/*
* Before mapping the page in, ensure that
* all other lwps are held in the kernel.
*/
if (p->p_mapcnt == 0) {
/*
* Release as lock while in holdwatch()
* in case other threads need to grab it.
*/
AS_LOCK_EXIT(as, &as->a_lock);
mutex_exit(&p->p_maplock);
if (holdwatch() != 0) {
/*
* We stopped in holdwatch().
* Start all over again because the
* watched page list may have changed.
*/
goto startover;
}
mutex_enter(&p->p_maplock);
AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
}
p->p_mapcnt++;
}
addr = pwp->wp_vaddr;
rv++;
prot = pwp->wp_prot;
if (mapin) {
if (kernel)
pwp->wp_kmap[xrw]++;
else
pwp->wp_umap[xrw]++;
pwp->wp_flags |= WP_NOWATCH;
if (pwp->wp_kmap[X] + pwp->wp_umap[X])
/* cannot have exec-only protection */
prot |= PROT_READ|PROT_EXEC;
if (pwp->wp_kmap[R] + pwp->wp_umap[R])
prot |= PROT_READ;
if (pwp->wp_kmap[W] + pwp->wp_umap[W])
/* cannot have write-only protection */
prot |= PROT_READ|PROT_WRITE;
#if 0 /* damned broken mmu feature! */
if (sum(pwp->wp_umap) == 0)
prot &= ~PROT_USER;
#endif
} else {
ASSERT(pwp->wp_flags & WP_NOWATCH);
if (kernel) {
ASSERT(pwp->wp_kmap[xrw] != 0);
--pwp->wp_kmap[xrw];
} else {
ASSERT(pwp->wp_umap[xrw] != 0);
--pwp->wp_umap[xrw];
}
if (sum(pwp->wp_kmap) + sum(pwp->wp_umap) == 0)
pwp->wp_flags &= ~WP_NOWATCH;
else {
if (pwp->wp_kmap[X] + pwp->wp_umap[X])
/* cannot have exec-only protection */
prot |= PROT_READ|PROT_EXEC;
if (pwp->wp_kmap[R] + pwp->wp_umap[R])
prot |= PROT_READ;
if (pwp->wp_kmap[W] + pwp->wp_umap[W])
/* cannot have write-only protection */
prot |= PROT_READ|PROT_WRITE;
#if 0 /* damned broken mmu feature! */
if (sum(pwp->wp_umap) == 0)
prot &= ~PROT_USER;
#endif
}
}
if (pwp->wp_oprot != 0) { /* if page exists */
struct seg *seg;
uint_t oprot;
int err, retrycnt = 0;
AS_LOCK_EXIT(as, &as->a_lock);
AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER);
retry:
seg = as_segat(as, addr);
ASSERT(seg != NULL);
SEGOP_GETPROT(seg, addr, 0, &oprot);
if (prot != oprot) {
err = SEGOP_SETPROT(seg, addr, PAGESIZE, prot);
if (err == IE_RETRY) {
ASSERT(retrycnt == 0);
retrycnt++;
goto retry;
}
}
AS_LOCK_EXIT(as, &as->a_lock);
} else
AS_LOCK_EXIT(as, &as->a_lock);
/*
* When all pages are mapped back to their normal state,
* continue the other lwps.
*/
if (!mapin) {
ASSERT(p->p_mapcnt > 0);
p->p_mapcnt--;
if (p->p_mapcnt == 0) {
mutex_exit(&p->p_maplock);
mutex_enter(&p->p_lock);
continuelwps(p);
mutex_exit(&p->p_lock);
mutex_enter(&p->p_maplock);
}
}
AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
}
AS_LOCK_EXIT(as, &as->a_lock);
mutex_exit(&p->p_maplock);
return (rv);
}
/*
* 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);
}