0
|
1 |
/*
|
|
2 |
* CDDL HEADER START
|
|
3 |
*
|
|
4 |
* The contents of this file are subject to the terms of the
|
|
5 |
* Common Development and Distribution License, Version 1.0 only
|
|
6 |
* (the "License"). You may not use this file except in compliance
|
|
7 |
* with the License.
|
|
8 |
*
|
|
9 |
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
|
|
10 |
* or http://www.opensolaris.org/os/licensing.
|
|
11 |
* See the License for the specific language governing permissions
|
|
12 |
* and limitations under the License.
|
|
13 |
*
|
|
14 |
* When distributing Covered Code, include this CDDL HEADER in each
|
|
15 |
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
|
|
16 |
* If applicable, add the following below this CDDL HEADER, with the
|
|
17 |
* fields enclosed by brackets "[]" replaced with your own identifying
|
|
18 |
* information: Portions Copyright [yyyy] [name of copyright owner]
|
|
19 |
*
|
|
20 |
* CDDL HEADER END
|
|
21 |
*/
|
875
|
22 |
|
0
|
23 |
/*
|
1217
|
24 |
* Copyright 2006 Sun Microsystems, Inc. All rights reserved.
|
0
|
25 |
* Use is subject to license terms.
|
|
26 |
*/
|
|
27 |
|
|
28 |
/* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
|
|
29 |
/* All Rights Reserved */
|
|
30 |
|
|
31 |
|
|
32 |
#pragma ident "%Z%%M% %I% %E% SMI"
|
|
33 |
|
|
34 |
#include <sys/types.h>
|
|
35 |
#include <sys/param.h>
|
|
36 |
#include <sys/sysmacros.h>
|
|
37 |
#include <sys/signal.h>
|
|
38 |
#include <sys/cred.h>
|
|
39 |
#include <sys/policy.h>
|
|
40 |
#include <sys/user.h>
|
|
41 |
#include <sys/systm.h>
|
|
42 |
#include <sys/cpuvar.h>
|
|
43 |
#include <sys/vfs.h>
|
|
44 |
#include <sys/vnode.h>
|
|
45 |
#include <sys/file.h>
|
|
46 |
#include <sys/errno.h>
|
|
47 |
#include <sys/time.h>
|
|
48 |
#include <sys/proc.h>
|
|
49 |
#include <sys/cmn_err.h>
|
|
50 |
#include <sys/acct.h>
|
|
51 |
#include <sys/tuneable.h>
|
|
52 |
#include <sys/class.h>
|
|
53 |
#include <sys/kmem.h>
|
|
54 |
#include <sys/session.h>
|
|
55 |
#include <sys/ucontext.h>
|
|
56 |
#include <sys/stack.h>
|
|
57 |
#include <sys/procfs.h>
|
|
58 |
#include <sys/prsystm.h>
|
|
59 |
#include <sys/vmsystm.h>
|
|
60 |
#include <sys/vtrace.h>
|
|
61 |
#include <sys/debug.h>
|
|
62 |
#include <sys/shm_impl.h>
|
|
63 |
#include <sys/door_data.h>
|
|
64 |
#include <vm/as.h>
|
|
65 |
#include <vm/rm.h>
|
|
66 |
#include <c2/audit.h>
|
|
67 |
#include <sys/var.h>
|
|
68 |
#include <sys/schedctl.h>
|
|
69 |
#include <sys/utrap.h>
|
|
70 |
#include <sys/task.h>
|
|
71 |
#include <sys/resource.h>
|
|
72 |
#include <sys/cyclic.h>
|
|
73 |
#include <sys/lgrp.h>
|
|
74 |
#include <sys/rctl.h>
|
|
75 |
#include <sys/contract_impl.h>
|
|
76 |
#include <sys/contract/process_impl.h>
|
|
77 |
#include <sys/list.h>
|
|
78 |
#include <sys/dtrace.h>
|
|
79 |
#include <sys/pool.h>
|
|
80 |
#include <sys/zone.h>
|
|
81 |
#include <sys/sdt.h>
|
|
82 |
#include <sys/class.h>
|
|
83 |
#include <sys/corectl.h>
|
|
84 |
|
|
85 |
static int64_t cfork(int, int);
|
|
86 |
static int getproc(proc_t **, int);
|
|
87 |
static void fork_fail(proc_t *);
|
|
88 |
static void forklwp_fail(proc_t *);
|
|
89 |
|
|
90 |
int fork_fail_pending;
|
|
91 |
|
|
92 |
extern struct kmem_cache *process_cache;
|
|
93 |
|
|
94 |
/*
|
|
95 |
* forkall system call.
|
|
96 |
*/
|
|
97 |
int64_t
|
|
98 |
forkall(void)
|
|
99 |
{
|
|
100 |
return (cfork(0, 0));
|
|
101 |
}
|
|
102 |
|
|
103 |
/*
|
|
104 |
* The parent is stopped until the child invokes relvm().
|
|
105 |
*/
|
|
106 |
int64_t
|
|
107 |
vfork(void)
|
|
108 |
{
|
|
109 |
curthread->t_post_sys = 1; /* so vfwait() will be called */
|
|
110 |
return (cfork(1, 1));
|
|
111 |
}
|
|
112 |
|
|
113 |
/*
|
|
114 |
* fork1 system call
|
|
115 |
*/
|
|
116 |
int64_t
|
|
117 |
fork1(void)
|
|
118 |
{
|
|
119 |
return (cfork(0, 1));
|
|
120 |
}
|
|
121 |
|
|
122 |
/* ARGSUSED */
|
|
123 |
static int64_t
|
|
124 |
cfork(int isvfork, int isfork1)
|
|
125 |
{
|
|
126 |
proc_t *p = ttoproc(curthread);
|
|
127 |
struct as *as;
|
|
128 |
proc_t *cp, **orphpp;
|
|
129 |
klwp_t *clone;
|
|
130 |
kthread_t *t;
|
|
131 |
task_t *tk;
|
|
132 |
rval_t r;
|
|
133 |
int error;
|
|
134 |
int i;
|
|
135 |
rctl_set_t *dup_set;
|
|
136 |
rctl_alloc_gp_t *dup_gp;
|
|
137 |
rctl_entity_p_t e;
|
|
138 |
lwpdir_t *ldp;
|
|
139 |
lwpent_t *lep;
|
|
140 |
lwpent_t *clep;
|
|
141 |
|
|
142 |
/*
|
|
143 |
* fork is not supported for the /proc agent lwp.
|
|
144 |
*/
|
|
145 |
if (curthread == p->p_agenttp) {
|
|
146 |
error = ENOTSUP;
|
|
147 |
goto forkerr;
|
|
148 |
}
|
|
149 |
|
|
150 |
if ((error = secpolicy_basic_fork(CRED())) != 0)
|
|
151 |
goto forkerr;
|
|
152 |
|
|
153 |
/*
|
|
154 |
* If the calling lwp is doing a fork1() then the
|
|
155 |
* other lwps in this process are not duplicated and
|
|
156 |
* don't need to be held where their kernel stacks can be
|
|
157 |
* cloned. If doing forkall(), the process is held with
|
|
158 |
* SHOLDFORK, so that the lwps are at a point where their
|
|
159 |
* stacks can be copied which is on entry or exit from
|
|
160 |
* the kernel.
|
|
161 |
*/
|
|
162 |
if (!holdlwps(isfork1 ? SHOLDFORK1 : SHOLDFORK)) {
|
|
163 |
aston(curthread);
|
|
164 |
error = EINTR;
|
|
165 |
goto forkerr;
|
|
166 |
}
|
|
167 |
|
1091
|
168 |
#if defined(__sparc)
|
|
169 |
/*
|
|
170 |
* Ensure that the user stack is fully constructed
|
|
171 |
* before creating the child process structure.
|
|
172 |
*/
|
|
173 |
(void) flush_user_windows_to_stack(NULL);
|
|
174 |
#endif
|
|
175 |
|
|
176 |
mutex_enter(&p->p_lock);
|
875
|
177 |
/*
|
|
178 |
* If this is vfork(), cancel any suspend request we might
|
|
179 |
* have gotten from some other thread via lwp_suspend().
|
|
180 |
* Otherwise we could end up with a deadlock on return
|
|
181 |
* from the vfork() in both the parent and the child.
|
|
182 |
*/
|
|
183 |
if (isvfork)
|
|
184 |
curthread->t_proc_flag &= ~TP_HOLDLWP;
|
0
|
185 |
/*
|
|
186 |
* Prevent our resource set associations from being changed during fork.
|
|
187 |
*/
|
|
188 |
pool_barrier_enter();
|
|
189 |
mutex_exit(&p->p_lock);
|
|
190 |
|
|
191 |
/*
|
|
192 |
* Create a child proc struct. Place a VN_HOLD on appropriate vnodes.
|
|
193 |
*/
|
|
194 |
if (getproc(&cp, 0) < 0) {
|
|
195 |
mutex_enter(&p->p_lock);
|
|
196 |
pool_barrier_exit();
|
|
197 |
continuelwps(p);
|
|
198 |
mutex_exit(&p->p_lock);
|
|
199 |
error = EAGAIN;
|
|
200 |
goto forkerr;
|
|
201 |
}
|
|
202 |
|
|
203 |
TRACE_2(TR_FAC_PROC, TR_PROC_FORK, "proc_fork:cp %p p %p", cp, p);
|
|
204 |
|
|
205 |
/*
|
|
206 |
* Assign an address space to child
|
|
207 |
*/
|
|
208 |
if (isvfork) {
|
|
209 |
/*
|
|
210 |
* Clear any watched areas and remember the
|
|
211 |
* watched pages for restoring in vfwait().
|
|
212 |
*/
|
|
213 |
as = p->p_as;
|
|
214 |
if (avl_numnodes(&as->a_wpage) != 0) {
|
|
215 |
AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER);
|
|
216 |
as_clearwatch(as);
|
|
217 |
p->p_wpage = as->a_wpage;
|
|
218 |
avl_create(&as->a_wpage, wp_compare,
|
|
219 |
sizeof (struct watched_page),
|
|
220 |
offsetof(struct watched_page, wp_link));
|
|
221 |
AS_LOCK_EXIT(as, &as->a_lock);
|
|
222 |
}
|
|
223 |
cp->p_as = as;
|
|
224 |
cp->p_flag |= SVFORK;
|
|
225 |
} else {
|
|
226 |
/*
|
|
227 |
* We need to hold P_PR_LOCK until the address space has
|
|
228 |
* been duplicated and we've had a chance to remove from the
|
|
229 |
* child any DTrace probes that were in the parent. Holding
|
|
230 |
* P_PR_LOCK prevents any new probes from being added and any
|
|
231 |
* extant probes from being removed.
|
|
232 |
*/
|
|
233 |
mutex_enter(&p->p_lock);
|
|
234 |
sprlock_proc(p);
|
|
235 |
mutex_exit(&p->p_lock);
|
|
236 |
|
|
237 |
error = as_dup(p->p_as, &cp->p_as);
|
|
238 |
if (error != 0) {
|
|
239 |
fork_fail(cp);
|
|
240 |
mutex_enter(&pidlock);
|
|
241 |
orphpp = &p->p_orphan;
|
|
242 |
while (*orphpp != cp)
|
|
243 |
orphpp = &(*orphpp)->p_nextorph;
|
|
244 |
*orphpp = cp->p_nextorph;
|
|
245 |
ASSERT(p->p_child == cp);
|
|
246 |
p->p_child = cp->p_sibling;
|
|
247 |
if (p->p_child) {
|
|
248 |
p->p_child->p_psibling = NULL;
|
|
249 |
}
|
|
250 |
mutex_enter(&cp->p_lock);
|
|
251 |
tk = cp->p_task;
|
|
252 |
task_detach(cp);
|
|
253 |
ASSERT(cp->p_pool->pool_ref > 0);
|
|
254 |
atomic_add_32(&cp->p_pool->pool_ref, -1);
|
|
255 |
mutex_exit(&cp->p_lock);
|
|
256 |
pid_exit(cp);
|
|
257 |
mutex_exit(&pidlock);
|
|
258 |
task_rele(tk);
|
|
259 |
|
|
260 |
mutex_enter(&p->p_lock);
|
|
261 |
pool_barrier_exit();
|
|
262 |
continuelwps(p);
|
|
263 |
sprunlock(p);
|
|
264 |
/*
|
|
265 |
* Preserve ENOMEM error condition but
|
|
266 |
* map all others to EAGAIN.
|
|
267 |
*/
|
|
268 |
error = (error == ENOMEM) ? ENOMEM : EAGAIN;
|
|
269 |
goto forkerr;
|
|
270 |
}
|
|
271 |
/* Duplicate parent's shared memory */
|
|
272 |
if (p->p_segacct)
|
|
273 |
shmfork(p, cp);
|
|
274 |
|
|
275 |
if (p->p_dtrace_helpers != NULL) {
|
|
276 |
ASSERT(dtrace_helpers_fork != NULL);
|
|
277 |
(*dtrace_helpers_fork)(p, cp);
|
|
278 |
}
|
|
279 |
|
|
280 |
/*
|
|
281 |
* Remove all DTrace tracepoints from the child process.
|
|
282 |
*/
|
|
283 |
mutex_enter(&p->p_lock);
|
|
284 |
if (p->p_dtrace_count > 0)
|
|
285 |
dtrace_fasttrap_fork(p, cp);
|
|
286 |
sprunlock(p);
|
|
287 |
}
|
|
288 |
|
|
289 |
/*
|
|
290 |
* Duplicate parent's resource controls.
|
|
291 |
*/
|
|
292 |
dup_set = rctl_set_create();
|
|
293 |
for (;;) {
|
|
294 |
dup_gp = rctl_set_dup_prealloc(p->p_rctls);
|
|
295 |
mutex_enter(&p->p_rctls->rcs_lock);
|
|
296 |
if (rctl_set_dup_ready(p->p_rctls, dup_gp))
|
|
297 |
break;
|
|
298 |
mutex_exit(&p->p_rctls->rcs_lock);
|
|
299 |
rctl_prealloc_destroy(dup_gp);
|
|
300 |
}
|
|
301 |
e.rcep_p.proc = cp;
|
|
302 |
e.rcep_t = RCENTITY_PROCESS;
|
|
303 |
cp->p_rctls = rctl_set_dup(p->p_rctls, p, cp, &e, dup_set, dup_gp,
|
|
304 |
RCD_DUP | RCD_CALLBACK);
|
|
305 |
mutex_exit(&p->p_rctls->rcs_lock);
|
|
306 |
|
|
307 |
rctl_prealloc_destroy(dup_gp);
|
|
308 |
|
|
309 |
/*
|
|
310 |
* Allocate the child's lwp directory and lwpid hash table.
|
|
311 |
*/
|
|
312 |
if (isfork1)
|
|
313 |
cp->p_lwpdir_sz = 2;
|
|
314 |
else
|
|
315 |
cp->p_lwpdir_sz = p->p_lwpdir_sz;
|
|
316 |
cp->p_lwpdir = cp->p_lwpfree = ldp =
|
|
317 |
kmem_zalloc(cp->p_lwpdir_sz * sizeof (lwpdir_t), KM_SLEEP);
|
|
318 |
for (i = 1; i < cp->p_lwpdir_sz; i++, ldp++)
|
|
319 |
ldp->ld_next = ldp + 1;
|
|
320 |
cp->p_tidhash_sz = (cp->p_lwpdir_sz + 2) / 2;
|
|
321 |
cp->p_tidhash =
|
|
322 |
kmem_zalloc(cp->p_tidhash_sz * sizeof (lwpdir_t *), KM_SLEEP);
|
|
323 |
|
|
324 |
/*
|
|
325 |
* Duplicate parent's lwps.
|
|
326 |
* Mutual exclusion is not needed because the process is
|
|
327 |
* in the hold state and only the current lwp is running.
|
|
328 |
*/
|
|
329 |
klgrpset_clear(cp->p_lgrpset);
|
|
330 |
if (isfork1) {
|
|
331 |
clone = forklwp(ttolwp(curthread), cp, curthread->t_tid);
|
|
332 |
if (clone == NULL)
|
|
333 |
goto forklwperr;
|
|
334 |
/*
|
|
335 |
* Inherit only the lwp_wait()able flag,
|
|
336 |
* Daemon threads should not call fork1(), but oh well...
|
|
337 |
*/
|
|
338 |
lwptot(clone)->t_proc_flag |=
|
|
339 |
(curthread->t_proc_flag & TP_TWAIT);
|
|
340 |
} else {
|
|
341 |
/* this is forkall(), no one can be in lwp_wait() */
|
|
342 |
ASSERT(p->p_lwpwait == 0 && p->p_lwpdwait == 0);
|
|
343 |
/* for each entry in the parent's lwp directory... */
|
|
344 |
for (i = 0, ldp = p->p_lwpdir; i < p->p_lwpdir_sz; i++, ldp++) {
|
|
345 |
klwp_t *clwp;
|
769
|
346 |
kthread_t *ct;
|
0
|
347 |
|
|
348 |
if ((lep = ldp->ld_entry) == NULL)
|
|
349 |
continue;
|
|
350 |
|
|
351 |
if ((t = lep->le_thread) != NULL) {
|
|
352 |
clwp = forklwp(ttolwp(t), cp, t->t_tid);
|
|
353 |
if (clwp == NULL)
|
|
354 |
goto forklwperr;
|
769
|
355 |
ct = lwptot(clwp);
|
0
|
356 |
/*
|
|
357 |
* Inherit lwp_wait()able and daemon flags.
|
|
358 |
*/
|
769
|
359 |
ct->t_proc_flag |=
|
0
|
360 |
(t->t_proc_flag & (TP_TWAIT|TP_DAEMON));
|
|
361 |
/*
|
|
362 |
* Keep track of the clone of curthread to
|
|
363 |
* post return values through lwp_setrval().
|
769
|
364 |
* Mark other threads for special treatment
|
|
365 |
* by lwp_rtt() / post_syscall().
|
0
|
366 |
*/
|
|
367 |
if (t == curthread)
|
|
368 |
clone = clwp;
|
769
|
369 |
else
|
|
370 |
ct->t_flag |= T_FORKALL;
|
0
|
371 |
} else {
|
|
372 |
/*
|
|
373 |
* Replicate zombie lwps in the child.
|
|
374 |
*/
|
|
375 |
clep = kmem_zalloc(sizeof (*clep), KM_SLEEP);
|
|
376 |
clep->le_lwpid = lep->le_lwpid;
|
|
377 |
clep->le_start = lep->le_start;
|
|
378 |
lwp_hash_in(cp, clep);
|
|
379 |
}
|
|
380 |
}
|
|
381 |
}
|
|
382 |
|
|
383 |
/*
|
|
384 |
* Put new process in the parent's process contract, or put it
|
|
385 |
* in a new one if there is an active process template. Send a
|
|
386 |
* fork event (if requested) to whatever contract the child is
|
|
387 |
* a member of. Fails if the parent has been SIGKILLed.
|
|
388 |
*/
|
|
389 |
if (contract_process_fork(NULL, cp, p, B_TRUE) == NULL)
|
|
390 |
goto forklwperr;
|
|
391 |
|
|
392 |
/*
|
|
393 |
* No fork failures occur beyond this point.
|
|
394 |
*/
|
|
395 |
|
|
396 |
cp->p_lwpid = p->p_lwpid;
|
|
397 |
if (!isfork1) {
|
|
398 |
cp->p_lwpdaemon = p->p_lwpdaemon;
|
|
399 |
cp->p_zombcnt = p->p_zombcnt;
|
|
400 |
/*
|
|
401 |
* If the parent's lwp ids have wrapped around, so have the
|
|
402 |
* child's.
|
|
403 |
*/
|
|
404 |
cp->p_flag |= p->p_flag & SLWPWRAP;
|
|
405 |
}
|
|
406 |
|
|
407 |
corectl_path_hold(cp->p_corefile = p->p_corefile);
|
|
408 |
corectl_content_hold(cp->p_content = p->p_content);
|
|
409 |
|
|
410 |
/*
|
1217
|
411 |
* Duplicate process context ops, if any.
|
0
|
412 |
*/
|
1217
|
413 |
if (p->p_pctx)
|
|
414 |
forkpctx(p, cp);
|
0
|
415 |
|
|
416 |
#ifdef __sparc
|
|
417 |
utrap_dup(p, cp);
|
|
418 |
#endif
|
|
419 |
/*
|
|
420 |
* If the child process has been marked to stop on exit
|
|
421 |
* from this fork, arrange for all other lwps to stop in
|
|
422 |
* sympathy with the active lwp.
|
|
423 |
*/
|
|
424 |
if (PTOU(cp)->u_systrap &&
|
|
425 |
prismember(&PTOU(cp)->u_exitmask, curthread->t_sysnum)) {
|
|
426 |
mutex_enter(&cp->p_lock);
|
|
427 |
t = cp->p_tlist;
|
|
428 |
do {
|
|
429 |
t->t_proc_flag |= TP_PRSTOP;
|
|
430 |
aston(t); /* so TP_PRSTOP will be seen */
|
|
431 |
} while ((t = t->t_forw) != cp->p_tlist);
|
|
432 |
mutex_exit(&cp->p_lock);
|
|
433 |
}
|
|
434 |
/*
|
|
435 |
* If the parent process has been marked to stop on exit
|
|
436 |
* from this fork, and its asynchronous-stop flag has not
|
|
437 |
* been set, arrange for all other lwps to stop before
|
|
438 |
* they return back to user level.
|
|
439 |
*/
|
|
440 |
if (!(p->p_proc_flag & P_PR_ASYNC) && PTOU(p)->u_systrap &&
|
|
441 |
prismember(&PTOU(p)->u_exitmask, curthread->t_sysnum)) {
|
|
442 |
mutex_enter(&p->p_lock);
|
|
443 |
t = p->p_tlist;
|
|
444 |
do {
|
|
445 |
t->t_proc_flag |= TP_PRSTOP;
|
|
446 |
aston(t); /* so TP_PRSTOP will be seen */
|
|
447 |
} while ((t = t->t_forw) != p->p_tlist);
|
|
448 |
mutex_exit(&p->p_lock);
|
|
449 |
}
|
|
450 |
|
|
451 |
/* set return values for child */
|
|
452 |
lwp_setrval(clone, p->p_pid, 1);
|
|
453 |
|
|
454 |
/* set return values for parent */
|
|
455 |
r.r_val1 = (int)cp->p_pid;
|
|
456 |
r.r_val2 = 0;
|
|
457 |
|
|
458 |
/*
|
|
459 |
* pool_barrier_exit() can now be called because the child process has:
|
|
460 |
* - all identifying features cloned or set (p_pid, p_task, p_pool)
|
|
461 |
* - all resource sets associated (p_tlist->*->t_cpupart, p_as->a_mset)
|
|
462 |
* - any other fields set which are used in resource set binding.
|
|
463 |
*/
|
|
464 |
mutex_enter(&p->p_lock);
|
|
465 |
pool_barrier_exit();
|
|
466 |
mutex_exit(&p->p_lock);
|
|
467 |
|
|
468 |
mutex_enter(&pidlock);
|
|
469 |
mutex_enter(&cp->p_lock);
|
|
470 |
|
|
471 |
/*
|
|
472 |
* Now that there are lwps and threads attached, add the new
|
|
473 |
* process to the process group.
|
|
474 |
*/
|
|
475 |
pgjoin(cp, p->p_pgidp);
|
|
476 |
cp->p_stat = SRUN;
|
|
477 |
/*
|
|
478 |
* We are now done with all the lwps in the child process.
|
|
479 |
*/
|
|
480 |
t = cp->p_tlist;
|
|
481 |
do {
|
|
482 |
/*
|
|
483 |
* Set the lwp_suspend()ed lwps running.
|
|
484 |
* They will suspend properly at syscall exit.
|
|
485 |
*/
|
|
486 |
if (t->t_proc_flag & TP_HOLDLWP)
|
|
487 |
lwp_create_done(t);
|
|
488 |
else {
|
|
489 |
/* set TS_CREATE to allow continuelwps() to work */
|
|
490 |
thread_lock(t);
|
|
491 |
ASSERT(t->t_state == TS_STOPPED &&
|
|
492 |
!(t->t_schedflag & (TS_CREATE|TS_CSTART)));
|
|
493 |
t->t_schedflag |= TS_CREATE;
|
|
494 |
thread_unlock(t);
|
|
495 |
}
|
|
496 |
} while ((t = t->t_forw) != cp->p_tlist);
|
|
497 |
mutex_exit(&cp->p_lock);
|
|
498 |
|
|
499 |
if (isvfork) {
|
|
500 |
CPU_STATS_ADDQ(CPU, sys, sysvfork, 1);
|
|
501 |
mutex_enter(&p->p_lock);
|
|
502 |
p->p_flag |= SVFWAIT;
|
|
503 |
DTRACE_PROC1(create, proc_t *, cp);
|
|
504 |
cv_broadcast(&pr_pid_cv[p->p_slot]); /* inform /proc */
|
|
505 |
mutex_exit(&p->p_lock);
|
|
506 |
/*
|
|
507 |
* Grab child's p_lock before dropping pidlock to ensure
|
|
508 |
* the process will not disappear before we set it running.
|
|
509 |
*/
|
|
510 |
mutex_enter(&cp->p_lock);
|
|
511 |
mutex_exit(&pidlock);
|
|
512 |
sigdefault(cp);
|
|
513 |
continuelwps(cp);
|
|
514 |
mutex_exit(&cp->p_lock);
|
|
515 |
} else {
|
|
516 |
CPU_STATS_ADDQ(CPU, sys, sysfork, 1);
|
|
517 |
DTRACE_PROC1(create, proc_t *, cp);
|
|
518 |
/*
|
|
519 |
* It is CL_FORKRET's job to drop pidlock.
|
|
520 |
* If we do it here, the process could be set running
|
|
521 |
* and disappear before CL_FORKRET() is called.
|
|
522 |
*/
|
|
523 |
CL_FORKRET(curthread, cp->p_tlist);
|
|
524 |
ASSERT(MUTEX_NOT_HELD(&pidlock));
|
|
525 |
}
|
|
526 |
|
|
527 |
return (r.r_vals);
|
|
528 |
|
|
529 |
forklwperr:
|
|
530 |
if (isvfork) {
|
|
531 |
if (avl_numnodes(&p->p_wpage) != 0) {
|
|
532 |
/* restore watchpoints to parent */
|
|
533 |
as = p->p_as;
|
|
534 |
AS_LOCK_ENTER(as, &as->a_lock,
|
|
535 |
RW_WRITER);
|
|
536 |
as->a_wpage = p->p_wpage;
|
|
537 |
avl_create(&p->p_wpage, wp_compare,
|
|
538 |
sizeof (struct watched_page),
|
|
539 |
offsetof(struct watched_page, wp_link));
|
|
540 |
as_setwatch(as);
|
|
541 |
AS_LOCK_EXIT(as, &as->a_lock);
|
|
542 |
}
|
|
543 |
} else {
|
|
544 |
if (cp->p_segacct)
|
|
545 |
shmexit(cp);
|
|
546 |
as = cp->p_as;
|
|
547 |
cp->p_as = &kas;
|
|
548 |
as_free(as);
|
|
549 |
}
|
|
550 |
|
|
551 |
if (cp->p_lwpdir) {
|
|
552 |
for (i = 0, ldp = cp->p_lwpdir; i < cp->p_lwpdir_sz; i++, ldp++)
|
|
553 |
if ((lep = ldp->ld_entry) != NULL)
|
|
554 |
kmem_free(lep, sizeof (*lep));
|
|
555 |
kmem_free(cp->p_lwpdir,
|
|
556 |
cp->p_lwpdir_sz * sizeof (*cp->p_lwpdir));
|
|
557 |
}
|
|
558 |
cp->p_lwpdir = NULL;
|
|
559 |
cp->p_lwpfree = NULL;
|
|
560 |
cp->p_lwpdir_sz = 0;
|
|
561 |
|
|
562 |
if (cp->p_tidhash)
|
|
563 |
kmem_free(cp->p_tidhash,
|
|
564 |
cp->p_tidhash_sz * sizeof (*cp->p_tidhash));
|
|
565 |
cp->p_tidhash = NULL;
|
|
566 |
cp->p_tidhash_sz = 0;
|
|
567 |
|
|
568 |
forklwp_fail(cp);
|
|
569 |
fork_fail(cp);
|
|
570 |
rctl_set_free(cp->p_rctls);
|
|
571 |
mutex_enter(&pidlock);
|
|
572 |
|
|
573 |
/*
|
|
574 |
* Detach failed child from task.
|
|
575 |
*/
|
|
576 |
mutex_enter(&cp->p_lock);
|
|
577 |
tk = cp->p_task;
|
|
578 |
task_detach(cp);
|
|
579 |
ASSERT(cp->p_pool->pool_ref > 0);
|
|
580 |
atomic_add_32(&cp->p_pool->pool_ref, -1);
|
|
581 |
mutex_exit(&cp->p_lock);
|
|
582 |
|
|
583 |
orphpp = &p->p_orphan;
|
|
584 |
while (*orphpp != cp)
|
|
585 |
orphpp = &(*orphpp)->p_nextorph;
|
|
586 |
*orphpp = cp->p_nextorph;
|
|
587 |
ASSERT(p->p_child == cp);
|
|
588 |
p->p_child = cp->p_sibling;
|
|
589 |
if (p->p_child) {
|
|
590 |
p->p_child->p_psibling = NULL;
|
|
591 |
}
|
|
592 |
pid_exit(cp);
|
|
593 |
mutex_exit(&pidlock);
|
|
594 |
|
|
595 |
task_rele(tk);
|
|
596 |
|
|
597 |
mutex_enter(&p->p_lock);
|
|
598 |
pool_barrier_exit();
|
|
599 |
continuelwps(p);
|
|
600 |
mutex_exit(&p->p_lock);
|
|
601 |
error = EAGAIN;
|
|
602 |
forkerr:
|
|
603 |
return ((int64_t)set_errno(error));
|
|
604 |
}
|
|
605 |
|
|
606 |
/*
|
|
607 |
* Free allocated resources from getproc() if a fork failed.
|
|
608 |
*/
|
|
609 |
static void
|
|
610 |
fork_fail(proc_t *cp)
|
|
611 |
{
|
|
612 |
uf_info_t *fip = P_FINFO(cp);
|
|
613 |
|
|
614 |
fcnt_add(fip, -1);
|
|
615 |
sigdelq(cp, NULL, 0);
|
|
616 |
|
|
617 |
mutex_enter(&pidlock);
|
|
618 |
upcount_dec(crgetruid(cp->p_cred), crgetzoneid(cp->p_cred));
|
|
619 |
mutex_exit(&pidlock);
|
|
620 |
|
|
621 |
/*
|
|
622 |
* single threaded, so no locking needed here
|
|
623 |
*/
|
|
624 |
crfree(cp->p_cred);
|
|
625 |
|
|
626 |
kmem_free(fip->fi_list, fip->fi_nfiles * sizeof (uf_entry_t));
|
|
627 |
|
|
628 |
VN_RELE(u.u_cdir);
|
|
629 |
if (u.u_rdir)
|
|
630 |
VN_RELE(u.u_rdir);
|
|
631 |
if (cp->p_exec)
|
|
632 |
VN_RELE(cp->p_exec);
|
|
633 |
if (cp->p_execdir)
|
|
634 |
VN_RELE(cp->p_execdir);
|
|
635 |
if (u.u_cwd)
|
|
636 |
refstr_rele(u.u_cwd);
|
|
637 |
}
|
|
638 |
|
|
639 |
/*
|
|
640 |
* Clean up the lwps already created for this child process.
|
|
641 |
* The fork failed while duplicating all the lwps of the parent
|
|
642 |
* and those lwps already created must be freed.
|
|
643 |
* This process is invisible to the rest of the system,
|
|
644 |
* so we don't need to hold p->p_lock to protect the list.
|
|
645 |
*/
|
|
646 |
static void
|
|
647 |
forklwp_fail(proc_t *p)
|
|
648 |
{
|
|
649 |
kthread_t *t;
|
|
650 |
task_t *tk;
|
|
651 |
|
|
652 |
while ((t = p->p_tlist) != NULL) {
|
|
653 |
/*
|
|
654 |
* First remove the lwp from the process's p_tlist.
|
|
655 |
*/
|
|
656 |
if (t != t->t_forw)
|
|
657 |
p->p_tlist = t->t_forw;
|
|
658 |
else
|
|
659 |
p->p_tlist = NULL;
|
|
660 |
p->p_lwpcnt--;
|
|
661 |
t->t_forw->t_back = t->t_back;
|
|
662 |
t->t_back->t_forw = t->t_forw;
|
|
663 |
|
|
664 |
tk = p->p_task;
|
|
665 |
mutex_enter(&p->p_zone->zone_nlwps_lock);
|
|
666 |
tk->tk_nlwps--;
|
|
667 |
tk->tk_proj->kpj_nlwps--;
|
|
668 |
p->p_zone->zone_nlwps--;
|
|
669 |
mutex_exit(&p->p_zone->zone_nlwps_lock);
|
|
670 |
|
|
671 |
ASSERT(t->t_schedctl == NULL);
|
|
672 |
|
|
673 |
if (t->t_door != NULL) {
|
|
674 |
kmem_free(t->t_door, sizeof (door_data_t));
|
|
675 |
t->t_door = NULL;
|
|
676 |
}
|
|
677 |
lwp_ctmpl_clear(ttolwp(t));
|
|
678 |
|
|
679 |
/*
|
|
680 |
* Remove the thread from the all threads list.
|
|
681 |
* We need to hold pidlock for this.
|
|
682 |
*/
|
|
683 |
mutex_enter(&pidlock);
|
|
684 |
t->t_next->t_prev = t->t_prev;
|
|
685 |
t->t_prev->t_next = t->t_next;
|
|
686 |
CL_EXIT(t); /* tell the scheduler that we're exiting */
|
|
687 |
cv_broadcast(&t->t_joincv); /* tell anyone in thread_join */
|
|
688 |
mutex_exit(&pidlock);
|
|
689 |
|
|
690 |
/*
|
|
691 |
* Let the lgroup load averages know that this thread isn't
|
|
692 |
* going to show up (i.e. un-do what was done on behalf of
|
|
693 |
* this thread by the earlier lgrp_move_thread()).
|
|
694 |
*/
|
|
695 |
kpreempt_disable();
|
|
696 |
lgrp_move_thread(t, NULL, 1);
|
|
697 |
kpreempt_enable();
|
|
698 |
|
|
699 |
/*
|
|
700 |
* The thread was created TS_STOPPED.
|
|
701 |
* We change it to TS_FREE to avoid an
|
|
702 |
* ASSERT() panic in thread_free().
|
|
703 |
*/
|
|
704 |
t->t_state = TS_FREE;
|
|
705 |
thread_rele(t);
|
|
706 |
thread_free(t);
|
|
707 |
}
|
|
708 |
}
|
|
709 |
|
|
710 |
extern struct as kas;
|
|
711 |
|
|
712 |
/*
|
|
713 |
* fork a kernel process.
|
|
714 |
*/
|
|
715 |
int
|
|
716 |
newproc(void (*pc)(), caddr_t arg, id_t cid, int pri, struct contract **ct)
|
|
717 |
{
|
|
718 |
proc_t *p;
|
|
719 |
struct user *up;
|
|
720 |
klwp_t *lwp;
|
|
721 |
cont_process_t *ctp = NULL;
|
|
722 |
rctl_entity_p_t e;
|
|
723 |
|
|
724 |
ASSERT(!(cid == syscid && ct != NULL));
|
|
725 |
if (cid == syscid) {
|
|
726 |
rctl_alloc_gp_t *init_gp;
|
|
727 |
rctl_set_t *init_set;
|
|
728 |
|
|
729 |
if (getproc(&p, 1) < 0)
|
|
730 |
return (EAGAIN);
|
|
731 |
|
|
732 |
p->p_flag |= SNOWAIT;
|
|
733 |
p->p_exec = NULL;
|
|
734 |
p->p_execdir = NULL;
|
|
735 |
|
|
736 |
init_set = rctl_set_create();
|
|
737 |
init_gp = rctl_set_init_prealloc(RCENTITY_PROCESS);
|
|
738 |
|
|
739 |
/*
|
|
740 |
* kernel processes do not inherit /proc tracing flags.
|
|
741 |
*/
|
|
742 |
sigemptyset(&p->p_sigmask);
|
|
743 |
premptyset(&p->p_fltmask);
|
|
744 |
up = PTOU(p);
|
|
745 |
up->u_systrap = 0;
|
|
746 |
premptyset(&(up->u_entrymask));
|
|
747 |
premptyset(&(up->u_exitmask));
|
|
748 |
mutex_enter(&p->p_lock);
|
|
749 |
e.rcep_p.proc = p;
|
|
750 |
e.rcep_t = RCENTITY_PROCESS;
|
|
751 |
p->p_rctls = rctl_set_init(RCENTITY_PROCESS, p, &e, init_set,
|
|
752 |
init_gp);
|
|
753 |
mutex_exit(&p->p_lock);
|
|
754 |
|
|
755 |
rctl_prealloc_destroy(init_gp);
|
|
756 |
} else {
|
|
757 |
rctl_alloc_gp_t *init_gp, *default_gp;
|
|
758 |
rctl_set_t *init_set;
|
|
759 |
task_t *tk, *tk_old;
|
|
760 |
|
|
761 |
if (getproc(&p, 0) < 0)
|
|
762 |
return (EAGAIN);
|
|
763 |
/*
|
|
764 |
* init creates a new task, distinct from the task
|
|
765 |
* containing kernel "processes".
|
|
766 |
*/
|
|
767 |
tk = task_create(0, p->p_zone);
|
|
768 |
mutex_enter(&tk->tk_zone->zone_nlwps_lock);
|
|
769 |
tk->tk_proj->kpj_ntasks++;
|
|
770 |
mutex_exit(&tk->tk_zone->zone_nlwps_lock);
|
|
771 |
|
|
772 |
default_gp = rctl_rlimit_set_prealloc(RLIM_NLIMITS);
|
|
773 |
init_gp = rctl_set_init_prealloc(RCENTITY_PROCESS);
|
|
774 |
init_set = rctl_set_create();
|
|
775 |
|
|
776 |
mutex_enter(&pidlock);
|
|
777 |
mutex_enter(&p->p_lock);
|
|
778 |
tk_old = p->p_task; /* switch to new task */
|
|
779 |
|
|
780 |
task_detach(p);
|
|
781 |
task_begin(tk, p);
|
|
782 |
mutex_exit(&pidlock);
|
|
783 |
|
|
784 |
e.rcep_p.proc = p;
|
|
785 |
e.rcep_t = RCENTITY_PROCESS;
|
|
786 |
p->p_rctls = rctl_set_init(RCENTITY_PROCESS, p, &e, init_set,
|
|
787 |
init_gp);
|
|
788 |
rctlproc_default_init(p, default_gp);
|
|
789 |
mutex_exit(&p->p_lock);
|
|
790 |
|
|
791 |
task_rele(tk_old);
|
|
792 |
rctl_prealloc_destroy(default_gp);
|
|
793 |
rctl_prealloc_destroy(init_gp);
|
|
794 |
}
|
|
795 |
|
|
796 |
p->p_as = &kas;
|
|
797 |
|
|
798 |
if ((lwp = lwp_create(pc, arg, 0, p, TS_STOPPED, pri,
|
|
799 |
&curthread->t_hold, cid, 1)) == NULL) {
|
|
800 |
task_t *tk;
|
|
801 |
fork_fail(p);
|
|
802 |
mutex_enter(&pidlock);
|
|
803 |
mutex_enter(&p->p_lock);
|
|
804 |
tk = p->p_task;
|
|
805 |
task_detach(p);
|
|
806 |
ASSERT(p->p_pool->pool_ref > 0);
|
|
807 |
atomic_add_32(&p->p_pool->pool_ref, -1);
|
|
808 |
mutex_exit(&p->p_lock);
|
|
809 |
pid_exit(p);
|
|
810 |
mutex_exit(&pidlock);
|
|
811 |
task_rele(tk);
|
|
812 |
|
|
813 |
return (EAGAIN);
|
|
814 |
}
|
|
815 |
|
|
816 |
if (cid != syscid) {
|
|
817 |
ctp = contract_process_fork(sys_process_tmpl, p, curproc,
|
|
818 |
B_FALSE);
|
|
819 |
ASSERT(ctp != NULL);
|
|
820 |
if (ct != NULL)
|
|
821 |
*ct = &ctp->conp_contract;
|
|
822 |
}
|
|
823 |
|
|
824 |
p->p_lwpid = 1;
|
|
825 |
mutex_enter(&pidlock);
|
|
826 |
pgjoin(p, curproc->p_pgidp);
|
|
827 |
p->p_stat = SRUN;
|
|
828 |
mutex_enter(&p->p_lock);
|
|
829 |
lwptot(lwp)->t_proc_flag &= ~TP_HOLDLWP;
|
|
830 |
lwp_create_done(lwptot(lwp));
|
|
831 |
mutex_exit(&p->p_lock);
|
|
832 |
mutex_exit(&pidlock);
|
|
833 |
return (0);
|
|
834 |
}
|
|
835 |
|
|
836 |
/*
|
|
837 |
* create a child proc struct.
|
|
838 |
*/
|
|
839 |
static int
|
|
840 |
getproc(proc_t **cpp, int kernel)
|
|
841 |
{
|
|
842 |
proc_t *pp, *cp;
|
|
843 |
pid_t newpid;
|
|
844 |
struct user *uarea;
|
|
845 |
extern uint_t nproc;
|
|
846 |
struct cred *cr;
|
|
847 |
uid_t ruid;
|
|
848 |
zoneid_t zoneid;
|
|
849 |
|
|
850 |
if (!page_mem_avail(tune.t_minarmem))
|
|
851 |
return (-1);
|
|
852 |
if (zone_status_get(curproc->p_zone) >= ZONE_IS_SHUTTING_DOWN)
|
|
853 |
return (-1); /* no point in starting new processes */
|
|
854 |
|
|
855 |
pp = curproc;
|
|
856 |
cp = kmem_cache_alloc(process_cache, KM_SLEEP);
|
|
857 |
bzero(cp, sizeof (proc_t));
|
|
858 |
|
|
859 |
/*
|
|
860 |
* Make proc entry for child process
|
|
861 |
*/
|
|
862 |
mutex_init(&cp->p_crlock, NULL, MUTEX_DEFAULT, NULL);
|
|
863 |
mutex_init(&cp->p_pflock, NULL, MUTEX_DEFAULT, NULL);
|
|
864 |
#if defined(__x86)
|
|
865 |
mutex_init(&cp->p_ldtlock, NULL, MUTEX_DEFAULT, NULL);
|
|
866 |
#endif
|
|
867 |
mutex_init(&cp->p_maplock, NULL, MUTEX_DEFAULT, NULL);
|
|
868 |
cp->p_stat = SIDL;
|
|
869 |
cp->p_mstart = gethrtime();
|
|
870 |
|
|
871 |
if ((newpid = pid_assign(cp)) == -1) {
|
|
872 |
if (nproc == v.v_proc) {
|
|
873 |
CPU_STATS_ADDQ(CPU, sys, procovf, 1);
|
|
874 |
cmn_err(CE_WARN, "out of processes");
|
|
875 |
}
|
|
876 |
goto bad;
|
|
877 |
}
|
|
878 |
|
|
879 |
/*
|
|
880 |
* If not privileged make sure that this user hasn't exceeded
|
|
881 |
* v.v_maxup processes, and that users collectively haven't
|
|
882 |
* exceeded v.v_maxupttl processes.
|
|
883 |
*/
|
|
884 |
mutex_enter(&pidlock);
|
|
885 |
ASSERT(nproc < v.v_proc); /* otherwise how'd we get our pid? */
|
|
886 |
cr = CRED();
|
|
887 |
ruid = crgetruid(cr);
|
|
888 |
zoneid = crgetzoneid(cr);
|
|
889 |
if (nproc >= v.v_maxup && /* short-circuit; usually false */
|
|
890 |
(nproc >= v.v_maxupttl ||
|
|
891 |
upcount_get(ruid, zoneid) >= v.v_maxup) &&
|
|
892 |
secpolicy_newproc(cr) != 0) {
|
|
893 |
mutex_exit(&pidlock);
|
|
894 |
zcmn_err(zoneid, CE_NOTE,
|
|
895 |
"out of per-user processes for uid %d", ruid);
|
|
896 |
goto bad;
|
|
897 |
}
|
|
898 |
|
|
899 |
/*
|
|
900 |
* Everything is cool, put the new proc on the active process list.
|
|
901 |
* It is already on the pid list and in /proc.
|
|
902 |
* Increment the per uid process count (upcount).
|
|
903 |
*/
|
|
904 |
nproc++;
|
|
905 |
upcount_inc(ruid, zoneid);
|
|
906 |
|
|
907 |
cp->p_next = practive;
|
|
908 |
practive->p_prev = cp;
|
|
909 |
practive = cp;
|
|
910 |
|
|
911 |
cp->p_ignore = pp->p_ignore;
|
|
912 |
cp->p_siginfo = pp->p_siginfo;
|
|
913 |
cp->p_flag = pp->p_flag & (SJCTL|SNOWAIT|SNOCD);
|
|
914 |
cp->p_sessp = pp->p_sessp;
|
|
915 |
SESS_HOLD(pp->p_sessp);
|
|
916 |
cp->p_exec = pp->p_exec;
|
|
917 |
cp->p_execdir = pp->p_execdir;
|
|
918 |
cp->p_zone = pp->p_zone;
|
|
919 |
|
|
920 |
cp->p_bssbase = pp->p_bssbase;
|
|
921 |
cp->p_brkbase = pp->p_brkbase;
|
|
922 |
cp->p_brksize = pp->p_brksize;
|
|
923 |
cp->p_brkpageszc = pp->p_brkpageszc;
|
|
924 |
cp->p_stksize = pp->p_stksize;
|
|
925 |
cp->p_stkpageszc = pp->p_stkpageszc;
|
|
926 |
cp->p_stkprot = pp->p_stkprot;
|
|
927 |
cp->p_datprot = pp->p_datprot;
|
|
928 |
cp->p_usrstack = pp->p_usrstack;
|
|
929 |
cp->p_model = pp->p_model;
|
|
930 |
cp->p_ppid = pp->p_pid;
|
|
931 |
cp->p_ancpid = pp->p_pid;
|
|
932 |
cp->p_portcnt = pp->p_portcnt;
|
|
933 |
|
|
934 |
/*
|
|
935 |
* Initialize watchpoint structures
|
|
936 |
*/
|
|
937 |
avl_create(&cp->p_warea, wa_compare, sizeof (struct watched_area),
|
|
938 |
offsetof(struct watched_area, wa_link));
|
|
939 |
|
|
940 |
/*
|
|
941 |
* Initialize immediate resource control values.
|
|
942 |
*/
|
|
943 |
cp->p_stk_ctl = pp->p_stk_ctl;
|
|
944 |
cp->p_fsz_ctl = pp->p_fsz_ctl;
|
|
945 |
cp->p_vmem_ctl = pp->p_vmem_ctl;
|
|
946 |
cp->p_fno_ctl = pp->p_fno_ctl;
|
|
947 |
|
|
948 |
/*
|
|
949 |
* Link up to parent-child-sibling chain. No need to lock
|
|
950 |
* in general since only a call to freeproc() (done by the
|
|
951 |
* same parent as newproc()) diddles with the child chain.
|
|
952 |
*/
|
|
953 |
cp->p_sibling = pp->p_child;
|
|
954 |
if (pp->p_child)
|
|
955 |
pp->p_child->p_psibling = cp;
|
|
956 |
|
|
957 |
cp->p_parent = pp;
|
|
958 |
pp->p_child = cp;
|
|
959 |
|
|
960 |
cp->p_child_ns = NULL;
|
|
961 |
cp->p_sibling_ns = NULL;
|
|
962 |
|
|
963 |
cp->p_nextorph = pp->p_orphan;
|
|
964 |
cp->p_nextofkin = pp;
|
|
965 |
pp->p_orphan = cp;
|
|
966 |
|
|
967 |
/*
|
|
968 |
* Inherit profiling state; do not inherit REALPROF profiling state.
|
|
969 |
*/
|
|
970 |
cp->p_prof = pp->p_prof;
|
|
971 |
cp->p_rprof_cyclic = CYCLIC_NONE;
|
|
972 |
|
|
973 |
/*
|
|
974 |
* Inherit pool pointer from the parent. Kernel processes are
|
|
975 |
* always bound to the default pool.
|
|
976 |
*/
|
|
977 |
mutex_enter(&pp->p_lock);
|
|
978 |
if (kernel) {
|
|
979 |
cp->p_pool = pool_default;
|
|
980 |
cp->p_flag |= SSYS;
|
|
981 |
} else {
|
|
982 |
cp->p_pool = pp->p_pool;
|
|
983 |
}
|
|
984 |
atomic_add_32(&cp->p_pool->pool_ref, 1);
|
|
985 |
mutex_exit(&pp->p_lock);
|
|
986 |
|
|
987 |
/*
|
|
988 |
* Add the child process to the current task. Kernel processes
|
|
989 |
* are always attached to task0.
|
|
990 |
*/
|
|
991 |
mutex_enter(&cp->p_lock);
|
|
992 |
if (kernel)
|
|
993 |
task_attach(task0p, cp);
|
|
994 |
else
|
|
995 |
task_attach(pp->p_task, cp);
|
|
996 |
mutex_exit(&cp->p_lock);
|
|
997 |
mutex_exit(&pidlock);
|
|
998 |
|
|
999 |
avl_create(&cp->p_ct_held, contract_compar, sizeof (contract_t),
|
|
1000 |
offsetof(contract_t, ct_ctlist));
|
|
1001 |
|
|
1002 |
/*
|
|
1003 |
* Duplicate any audit information kept in the process table
|
|
1004 |
*/
|
|
1005 |
#ifdef C2_AUDIT
|
|
1006 |
if (audit_active) /* copy audit data to cp */
|
|
1007 |
audit_newproc(cp);
|
|
1008 |
#endif
|
|
1009 |
|
|
1010 |
crhold(cp->p_cred = cr);
|
|
1011 |
|
|
1012 |
/*
|
|
1013 |
* Bump up the counts on the file structures pointed at by the
|
|
1014 |
* parent's file table since the child will point at them too.
|
|
1015 |
*/
|
|
1016 |
fcnt_add(P_FINFO(pp), 1);
|
|
1017 |
|
|
1018 |
VN_HOLD(u.u_cdir);
|
|
1019 |
if (u.u_rdir)
|
|
1020 |
VN_HOLD(u.u_rdir);
|
|
1021 |
if (u.u_cwd)
|
|
1022 |
refstr_hold(u.u_cwd);
|
|
1023 |
|
|
1024 |
/*
|
|
1025 |
* copy the parent's uarea.
|
|
1026 |
*/
|
|
1027 |
uarea = PTOU(cp);
|
|
1028 |
bcopy(PTOU(pp), uarea, sizeof (user_t));
|
|
1029 |
flist_fork(P_FINFO(pp), P_FINFO(cp));
|
|
1030 |
|
|
1031 |
gethrestime(&uarea->u_start);
|
|
1032 |
uarea->u_ticks = lbolt;
|
|
1033 |
uarea->u_mem = rm_asrss(pp->p_as);
|
|
1034 |
uarea->u_acflag = AFORK;
|
|
1035 |
|
|
1036 |
/*
|
|
1037 |
* If inherit-on-fork, copy /proc tracing flags to child.
|
|
1038 |
*/
|
|
1039 |
if ((pp->p_proc_flag & P_PR_FORK) != 0) {
|
|
1040 |
cp->p_proc_flag |= pp->p_proc_flag & (P_PR_TRACE|P_PR_FORK);
|
|
1041 |
cp->p_sigmask = pp->p_sigmask;
|
|
1042 |
cp->p_fltmask = pp->p_fltmask;
|
|
1043 |
} else {
|
|
1044 |
sigemptyset(&cp->p_sigmask);
|
|
1045 |
premptyset(&cp->p_fltmask);
|
|
1046 |
uarea->u_systrap = 0;
|
|
1047 |
premptyset(&uarea->u_entrymask);
|
|
1048 |
premptyset(&uarea->u_exitmask);
|
|
1049 |
}
|
|
1050 |
/*
|
|
1051 |
* If microstate accounting is being inherited, mark child
|
|
1052 |
*/
|
|
1053 |
if ((pp->p_flag & SMSFORK) != 0)
|
|
1054 |
cp->p_flag |= pp->p_flag & (SMSFORK|SMSACCT);
|
|
1055 |
|
|
1056 |
/*
|
|
1057 |
* Inherit fixalignment flag from the parent
|
|
1058 |
*/
|
|
1059 |
cp->p_fixalignment = pp->p_fixalignment;
|
|
1060 |
|
|
1061 |
if (cp->p_exec)
|
|
1062 |
VN_HOLD(cp->p_exec);
|
|
1063 |
if (cp->p_execdir)
|
|
1064 |
VN_HOLD(cp->p_execdir);
|
|
1065 |
*cpp = cp;
|
|
1066 |
return (0);
|
|
1067 |
|
|
1068 |
bad:
|
|
1069 |
ASSERT(MUTEX_NOT_HELD(&pidlock));
|
|
1070 |
|
|
1071 |
mutex_destroy(&cp->p_crlock);
|
|
1072 |
mutex_destroy(&cp->p_pflock);
|
|
1073 |
#if defined(__x86)
|
|
1074 |
mutex_destroy(&cp->p_ldtlock);
|
|
1075 |
#endif
|
|
1076 |
if (newpid != -1) {
|
|
1077 |
proc_entry_free(cp->p_pidp);
|
|
1078 |
(void) pid_rele(cp->p_pidp);
|
|
1079 |
}
|
|
1080 |
kmem_cache_free(process_cache, cp);
|
|
1081 |
|
|
1082 |
/*
|
|
1083 |
* We most likely got into this situation because some process is
|
|
1084 |
* forking out of control. As punishment, put it to sleep for a
|
|
1085 |
* bit so it can't eat the machine alive. Sleep interval is chosen
|
|
1086 |
* to allow no more than one fork failure per cpu per clock tick
|
|
1087 |
* on average (yes, I just made this up). This has two desirable
|
|
1088 |
* properties: (1) it sets a constant limit on the fork failure
|
|
1089 |
* rate, and (2) the busier the system is, the harsher the penalty
|
|
1090 |
* for abusing it becomes.
|
|
1091 |
*/
|
|
1092 |
INCR_COUNT(&fork_fail_pending, &pidlock);
|
|
1093 |
delay(fork_fail_pending / ncpus + 1);
|
|
1094 |
DECR_COUNT(&fork_fail_pending, &pidlock);
|
|
1095 |
|
|
1096 |
return (-1); /* out of memory or proc slots */
|
|
1097 |
}
|
|
1098 |
|
|
1099 |
/*
|
|
1100 |
* Release virtual memory.
|
|
1101 |
* In the case of vfork(), the child was given exclusive access to its
|
|
1102 |
* parent's address space. The parent is waiting in vfwait() for the
|
|
1103 |
* child to release its exclusive claim via relvm().
|
|
1104 |
*/
|
|
1105 |
void
|
|
1106 |
relvm()
|
|
1107 |
{
|
|
1108 |
proc_t *p = curproc;
|
|
1109 |
|
|
1110 |
ASSERT((unsigned)p->p_lwpcnt <= 1);
|
|
1111 |
|
|
1112 |
prrelvm(); /* inform /proc */
|
|
1113 |
|
|
1114 |
if (p->p_flag & SVFORK) {
|
|
1115 |
proc_t *pp = p->p_parent;
|
|
1116 |
/*
|
|
1117 |
* The child process is either exec'ing or exit'ing.
|
|
1118 |
* The child is now separated from the parent's address
|
|
1119 |
* space. The parent process is made dispatchable.
|
|
1120 |
*
|
|
1121 |
* This is a delicate locking maneuver, involving
|
|
1122 |
* both the parent's p_lock and the child's p_lock.
|
|
1123 |
* As soon as the SVFORK flag is turned off, the
|
|
1124 |
* parent is free to run, but it must not run until
|
|
1125 |
* we wake it up using its p_cv because it might
|
|
1126 |
* exit and we would be referencing invalid memory.
|
|
1127 |
* Therefore, we hold the parent with its p_lock
|
|
1128 |
* while protecting our p_flags with our own p_lock.
|
|
1129 |
*/
|
|
1130 |
try_again:
|
|
1131 |
mutex_enter(&p->p_lock); /* grab child's lock first */
|
|
1132 |
prbarrier(p); /* make sure /proc is blocked out */
|
|
1133 |
mutex_enter(&pp->p_lock);
|
|
1134 |
|
|
1135 |
/*
|
|
1136 |
* Check if parent is locked by /proc.
|
|
1137 |
*/
|
|
1138 |
if (pp->p_proc_flag & P_PR_LOCK) {
|
|
1139 |
/*
|
|
1140 |
* Delay until /proc is done with the parent.
|
|
1141 |
* We must drop our (the child's) p->p_lock, wait
|
|
1142 |
* via prbarrier() on the parent, then start over.
|
|
1143 |
*/
|
|
1144 |
mutex_exit(&p->p_lock);
|
|
1145 |
prbarrier(pp);
|
|
1146 |
mutex_exit(&pp->p_lock);
|
|
1147 |
goto try_again;
|
|
1148 |
}
|
|
1149 |
p->p_flag &= ~SVFORK;
|
|
1150 |
kpreempt_disable();
|
|
1151 |
p->p_as = &kas;
|
|
1152 |
|
|
1153 |
/*
|
|
1154 |
* notify hat of change in thread's address space
|
|
1155 |
*/
|
|
1156 |
hat_thread_exit(curthread);
|
|
1157 |
kpreempt_enable();
|
|
1158 |
|
|
1159 |
/*
|
|
1160 |
* child sizes are copied back to parent because
|
|
1161 |
* child may have grown.
|
|
1162 |
*/
|
|
1163 |
pp->p_brkbase = p->p_brkbase;
|
|
1164 |
pp->p_brksize = p->p_brksize;
|
|
1165 |
pp->p_stksize = p->p_stksize;
|
|
1166 |
/*
|
|
1167 |
* The parent is no longer waiting for the vfork()d child.
|
|
1168 |
* Restore the parent's watched pages, if any. This is
|
|
1169 |
* safe because we know the parent is not locked by /proc
|
|
1170 |
*/
|
|
1171 |
pp->p_flag &= ~SVFWAIT;
|
|
1172 |
if (avl_numnodes(&pp->p_wpage) != 0) {
|
|
1173 |
pp->p_as->a_wpage = pp->p_wpage;
|
|
1174 |
avl_create(&pp->p_wpage, wp_compare,
|
|
1175 |
sizeof (struct watched_page),
|
|
1176 |
offsetof(struct watched_page, wp_link));
|
|
1177 |
}
|
|
1178 |
cv_signal(&pp->p_cv);
|
|
1179 |
mutex_exit(&pp->p_lock);
|
|
1180 |
mutex_exit(&p->p_lock);
|
|
1181 |
} else {
|
|
1182 |
if (p->p_as != &kas) {
|
|
1183 |
struct as *as;
|
|
1184 |
|
|
1185 |
if (p->p_segacct)
|
|
1186 |
shmexit(p);
|
|
1187 |
/*
|
|
1188 |
* We grab p_lock for the benefit of /proc
|
|
1189 |
*/
|
|
1190 |
kpreempt_disable();
|
|
1191 |
mutex_enter(&p->p_lock);
|
|
1192 |
prbarrier(p); /* make sure /proc is blocked out */
|
|
1193 |
as = p->p_as;
|
|
1194 |
p->p_as = &kas;
|
|
1195 |
mutex_exit(&p->p_lock);
|
|
1196 |
|
|
1197 |
/*
|
|
1198 |
* notify hat of change in thread's address space
|
|
1199 |
*/
|
|
1200 |
hat_thread_exit(curthread);
|
|
1201 |
kpreempt_enable();
|
|
1202 |
|
|
1203 |
as_free(as);
|
|
1204 |
}
|
|
1205 |
}
|
|
1206 |
}
|
|
1207 |
|
|
1208 |
/*
|
|
1209 |
* Wait for child to exec or exit.
|
|
1210 |
* Called by parent of vfork'ed process.
|
|
1211 |
* See important comments in relvm(), above.
|
|
1212 |
*/
|
|
1213 |
void
|
|
1214 |
vfwait(pid_t pid)
|
|
1215 |
{
|
|
1216 |
int signalled = 0;
|
|
1217 |
proc_t *pp = ttoproc(curthread);
|
|
1218 |
proc_t *cp;
|
|
1219 |
|
|
1220 |
/*
|
|
1221 |
* Wait for child to exec or exit.
|
|
1222 |
*/
|
|
1223 |
for (;;) {
|
|
1224 |
mutex_enter(&pidlock);
|
|
1225 |
cp = prfind(pid);
|
|
1226 |
if (cp == NULL || cp->p_parent != pp) {
|
|
1227 |
/*
|
|
1228 |
* Child has exit()ed.
|
|
1229 |
*/
|
|
1230 |
mutex_exit(&pidlock);
|
|
1231 |
break;
|
|
1232 |
}
|
|
1233 |
/*
|
|
1234 |
* Grab the child's p_lock before releasing pidlock.
|
|
1235 |
* Otherwise, the child could exit and we would be
|
|
1236 |
* referencing invalid memory.
|
|
1237 |
*/
|
|
1238 |
mutex_enter(&cp->p_lock);
|
|
1239 |
mutex_exit(&pidlock);
|
|
1240 |
if (!(cp->p_flag & SVFORK)) {
|
|
1241 |
/*
|
|
1242 |
* Child has exec()ed or is exit()ing.
|
|
1243 |
*/
|
|
1244 |
mutex_exit(&cp->p_lock);
|
|
1245 |
break;
|
|
1246 |
}
|
|
1247 |
mutex_enter(&pp->p_lock);
|
|
1248 |
mutex_exit(&cp->p_lock);
|
|
1249 |
/*
|
|
1250 |
* We might be waked up spuriously from the cv_wait().
|
|
1251 |
* We have to do the whole operation over again to be
|
|
1252 |
* sure the child's SVFORK flag really is turned off.
|
|
1253 |
* We cannot make reference to the child because it can
|
|
1254 |
* exit before we return and we would be referencing
|
|
1255 |
* invalid memory.
|
|
1256 |
*
|
|
1257 |
* Because this is potentially a very long-term wait,
|
|
1258 |
* we call cv_wait_sig() (for its jobcontrol and /proc
|
|
1259 |
* side-effects) unless there is a current signal, in
|
|
1260 |
* which case we use cv_wait() because we cannot return
|
|
1261 |
* from this function until the child has released the
|
|
1262 |
* address space. Calling cv_wait_sig() with a current
|
|
1263 |
* signal would lead to an indefinite loop here because
|
|
1264 |
* cv_wait_sig() returns immediately in this case.
|
|
1265 |
*/
|
|
1266 |
if (signalled)
|
|
1267 |
cv_wait(&pp->p_cv, &pp->p_lock);
|
|
1268 |
else
|
|
1269 |
signalled = !cv_wait_sig(&pp->p_cv, &pp->p_lock);
|
|
1270 |
mutex_exit(&pp->p_lock);
|
|
1271 |
}
|
|
1272 |
|
|
1273 |
/* restore watchpoints to parent */
|
|
1274 |
if (pr_watch_active(pp)) {
|
|
1275 |
struct as *as = pp->p_as;
|
|
1276 |
AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER);
|
|
1277 |
as_setwatch(as);
|
|
1278 |
AS_LOCK_EXIT(as, &as->a_lock);
|
|
1279 |
}
|
|
1280 |
|
|
1281 |
mutex_enter(&pp->p_lock);
|
|
1282 |
prbarrier(pp); /* barrier against /proc locking */
|
|
1283 |
continuelwps(pp);
|
|
1284 |
mutex_exit(&pp->p_lock);
|
|
1285 |
}
|