Print this page
5045 use atomic_{inc,dec}_* instead of atomic_add_*
Split |
Close |
Expand all |
Collapse all |
--- old/usr/src/uts/common/os/fork.c
+++ new/usr/src/uts/common/os/fork.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 */
25 25
26 26 /* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
27 27 /* All Rights Reserved */
28 28
29 29 #include <sys/types.h>
30 30 #include <sys/param.h>
31 31 #include <sys/sysmacros.h>
32 32 #include <sys/signal.h>
33 33 #include <sys/cred.h>
34 34 #include <sys/policy.h>
35 35 #include <sys/user.h>
36 36 #include <sys/systm.h>
37 37 #include <sys/cpuvar.h>
38 38 #include <sys/vfs.h>
39 39 #include <sys/vnode.h>
40 40 #include <sys/file.h>
41 41 #include <sys/errno.h>
42 42 #include <sys/time.h>
43 43 #include <sys/proc.h>
44 44 #include <sys/cmn_err.h>
45 45 #include <sys/acct.h>
46 46 #include <sys/tuneable.h>
47 47 #include <sys/class.h>
48 48 #include <sys/kmem.h>
49 49 #include <sys/session.h>
50 50 #include <sys/ucontext.h>
51 51 #include <sys/stack.h>
52 52 #include <sys/procfs.h>
53 53 #include <sys/prsystm.h>
54 54 #include <sys/vmsystm.h>
55 55 #include <sys/vtrace.h>
56 56 #include <sys/debug.h>
57 57 #include <sys/shm_impl.h>
58 58 #include <sys/door_data.h>
59 59 #include <vm/as.h>
60 60 #include <vm/rm.h>
61 61 #include <c2/audit.h>
62 62 #include <sys/var.h>
63 63 #include <sys/schedctl.h>
64 64 #include <sys/utrap.h>
65 65 #include <sys/task.h>
66 66 #include <sys/resource.h>
67 67 #include <sys/cyclic.h>
68 68 #include <sys/lgrp.h>
69 69 #include <sys/rctl.h>
70 70 #include <sys/contract_impl.h>
71 71 #include <sys/contract/process_impl.h>
72 72 #include <sys/list.h>
73 73 #include <sys/dtrace.h>
74 74 #include <sys/pool.h>
75 75 #include <sys/zone.h>
76 76 #include <sys/sdt.h>
77 77 #include <sys/class.h>
78 78 #include <sys/corectl.h>
79 79 #include <sys/brand.h>
80 80 #include <sys/fork.h>
81 81
82 82 static int64_t cfork(int, int, int);
83 83 static int getproc(proc_t **, pid_t, uint_t);
84 84 #define GETPROC_USER 0x0
85 85 #define GETPROC_KERNEL 0x1
86 86
87 87 static void fork_fail(proc_t *);
88 88 static void forklwp_fail(proc_t *);
89 89
90 90 int fork_fail_pending;
91 91
92 92 extern struct kmem_cache *process_cache;
93 93
94 94 /*
95 95 * The vfork() system call trap is no longer invoked by libc.
96 96 * It is retained only for the benefit of applications running
97 97 * within a solaris10 branded zone. It should be eliminated
98 98 * when we no longer support solaris10 branded zones.
99 99 */
100 100 int64_t
101 101 vfork(void)
102 102 {
103 103 curthread->t_post_sys = 1; /* so vfwait() will be called */
104 104 return (cfork(1, 1, 0));
105 105 }
106 106
107 107 /*
108 108 * forksys system call - forkx, forkallx, vforkx. This is the
109 109 * interface invoked by libc for fork1(), forkall(), and vfork()
110 110 */
111 111 int64_t
112 112 forksys(int subcode, int flags)
113 113 {
114 114 switch (subcode) {
115 115 case 0:
116 116 return (cfork(0, 1, flags)); /* forkx(flags) */
117 117 case 1:
118 118 return (cfork(0, 0, flags)); /* forkallx(flags) */
119 119 case 2:
120 120 curthread->t_post_sys = 1; /* so vfwait() will be called */
121 121 return (cfork(1, 1, flags)); /* vforkx(flags) */
122 122 default:
123 123 return ((int64_t)set_errno(EINVAL));
124 124 }
125 125 }
126 126
127 127 /* ARGSUSED */
128 128 static int64_t
129 129 cfork(int isvfork, int isfork1, int flags)
130 130 {
131 131 proc_t *p = ttoproc(curthread);
132 132 struct as *as;
133 133 proc_t *cp, **orphpp;
134 134 klwp_t *clone;
135 135 kthread_t *t;
136 136 task_t *tk;
137 137 rval_t r;
138 138 int error;
139 139 int i;
140 140 rctl_set_t *dup_set;
141 141 rctl_alloc_gp_t *dup_gp;
142 142 rctl_entity_p_t e;
143 143 lwpdir_t *ldp;
144 144 lwpent_t *lep;
145 145 lwpent_t *clep;
146 146
147 147 /*
148 148 * Allow only these two flags.
149 149 */
150 150 if ((flags & ~(FORK_NOSIGCHLD | FORK_WAITPID)) != 0) {
151 151 error = EINVAL;
152 152 goto forkerr;
153 153 }
154 154
155 155 /*
156 156 * fork is not supported for the /proc agent lwp.
157 157 */
158 158 if (curthread == p->p_agenttp) {
159 159 error = ENOTSUP;
160 160 goto forkerr;
161 161 }
162 162
163 163 if ((error = secpolicy_basic_fork(CRED())) != 0)
164 164 goto forkerr;
165 165
166 166 /*
167 167 * If the calling lwp is doing a fork1() then the
168 168 * other lwps in this process are not duplicated and
169 169 * don't need to be held where their kernel stacks can be
170 170 * cloned. If doing forkall(), the process is held with
171 171 * SHOLDFORK, so that the lwps are at a point where their
172 172 * stacks can be copied which is on entry or exit from
173 173 * the kernel.
174 174 */
175 175 if (!holdlwps(isfork1 ? SHOLDFORK1 : SHOLDFORK)) {
176 176 aston(curthread);
177 177 error = EINTR;
178 178 goto forkerr;
179 179 }
180 180
181 181 #if defined(__sparc)
182 182 /*
183 183 * Ensure that the user stack is fully constructed
184 184 * before creating the child process structure.
185 185 */
186 186 (void) flush_user_windows_to_stack(NULL);
187 187 #endif
188 188
189 189 mutex_enter(&p->p_lock);
190 190 /*
191 191 * If this is vfork(), cancel any suspend request we might
192 192 * have gotten from some other thread via lwp_suspend().
193 193 * Otherwise we could end up with a deadlock on return
194 194 * from the vfork() in both the parent and the child.
195 195 */
196 196 if (isvfork)
197 197 curthread->t_proc_flag &= ~TP_HOLDLWP;
198 198 /*
199 199 * Prevent our resource set associations from being changed during fork.
200 200 */
201 201 pool_barrier_enter();
202 202 mutex_exit(&p->p_lock);
203 203
204 204 /*
205 205 * Create a child proc struct. Place a VN_HOLD on appropriate vnodes.
206 206 */
207 207 if (getproc(&cp, 0, GETPROC_USER) < 0) {
208 208 mutex_enter(&p->p_lock);
209 209 pool_barrier_exit();
210 210 continuelwps(p);
211 211 mutex_exit(&p->p_lock);
212 212 error = EAGAIN;
213 213 goto forkerr;
214 214 }
215 215
216 216 TRACE_2(TR_FAC_PROC, TR_PROC_FORK, "proc_fork:cp %p p %p", cp, p);
217 217
218 218 /*
219 219 * Assign an address space to child
220 220 */
221 221 if (isvfork) {
222 222 /*
223 223 * Clear any watched areas and remember the
224 224 * watched pages for restoring in vfwait().
225 225 */
226 226 as = p->p_as;
227 227 if (avl_numnodes(&as->a_wpage) != 0) {
228 228 AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER);
229 229 as_clearwatch(as);
230 230 p->p_wpage = as->a_wpage;
231 231 avl_create(&as->a_wpage, wp_compare,
232 232 sizeof (struct watched_page),
233 233 offsetof(struct watched_page, wp_link));
234 234 AS_LOCK_EXIT(as, &as->a_lock);
235 235 }
236 236 cp->p_as = as;
237 237 cp->p_flag |= SVFORK;
238 238
239 239 /*
240 240 * Use the parent's shm segment list information for
241 241 * the child as it uses its address space till it execs.
242 242 */
243 243 cp->p_segacct = p->p_segacct;
244 244 } else {
245 245 /*
246 246 * We need to hold P_PR_LOCK until the address space has
247 247 * been duplicated and we've had a chance to remove from the
248 248 * child any DTrace probes that were in the parent. Holding
249 249 * P_PR_LOCK prevents any new probes from being added and any
250 250 * extant probes from being removed.
251 251 */
252 252 mutex_enter(&p->p_lock);
253 253 sprlock_proc(p);
254 254 p->p_flag |= SFORKING;
255 255 mutex_exit(&p->p_lock);
256 256
257 257 error = as_dup(p->p_as, cp);
258 258 if (error != 0) {
259 259 mutex_enter(&p->p_lock);
260 260 sprunlock(p);
261 261 fork_fail(cp);
262 262 mutex_enter(&pidlock);
263 263 orphpp = &p->p_orphan;
264 264 while (*orphpp != cp)
265 265 orphpp = &(*orphpp)->p_nextorph;
266 266 *orphpp = cp->p_nextorph;
↓ open down ↓ |
266 lines elided |
↑ open up ↑ |
267 267 if (p->p_child == cp)
268 268 p->p_child = cp->p_sibling;
269 269 if (cp->p_sibling)
270 270 cp->p_sibling->p_psibling = cp->p_psibling;
271 271 if (cp->p_psibling)
272 272 cp->p_psibling->p_sibling = cp->p_sibling;
273 273 mutex_enter(&cp->p_lock);
274 274 tk = cp->p_task;
275 275 task_detach(cp);
276 276 ASSERT(cp->p_pool->pool_ref > 0);
277 - atomic_add_32(&cp->p_pool->pool_ref, -1);
277 + atomic_dec_32(&cp->p_pool->pool_ref);
278 278 mutex_exit(&cp->p_lock);
279 279 pid_exit(cp, tk);
280 280 mutex_exit(&pidlock);
281 281 task_rele(tk);
282 282
283 283 mutex_enter(&p->p_lock);
284 284 p->p_flag &= ~SFORKING;
285 285 pool_barrier_exit();
286 286 continuelwps(p);
287 287 mutex_exit(&p->p_lock);
288 288 /*
289 289 * Preserve ENOMEM error condition but
290 290 * map all others to EAGAIN.
291 291 */
292 292 error = (error == ENOMEM) ? ENOMEM : EAGAIN;
293 293 goto forkerr;
294 294 }
295 295
296 296 /*
297 297 * Remove all DTrace tracepoints from the child process. We
298 298 * need to do this _before_ duplicating USDT providers since
299 299 * any associated probes may be immediately enabled.
300 300 */
301 301 if (p->p_dtrace_count > 0)
302 302 dtrace_fasttrap_fork(p, cp);
303 303
304 304 mutex_enter(&p->p_lock);
305 305 sprunlock(p);
306 306
307 307 /* Duplicate parent's shared memory */
308 308 if (p->p_segacct)
309 309 shmfork(p, cp);
310 310
311 311 /*
312 312 * Duplicate any helper actions and providers. The SFORKING
313 313 * we set above informs the code to enable USDT probes that
314 314 * sprlock() may fail because the child is being forked.
315 315 */
316 316 if (p->p_dtrace_helpers != NULL) {
317 317 ASSERT(dtrace_helpers_fork != NULL);
318 318 (*dtrace_helpers_fork)(p, cp);
319 319 }
320 320
321 321 mutex_enter(&p->p_lock);
322 322 p->p_flag &= ~SFORKING;
323 323 mutex_exit(&p->p_lock);
324 324 }
325 325
326 326 /*
327 327 * Duplicate parent's resource controls.
328 328 */
329 329 dup_set = rctl_set_create();
330 330 for (;;) {
331 331 dup_gp = rctl_set_dup_prealloc(p->p_rctls);
332 332 mutex_enter(&p->p_rctls->rcs_lock);
333 333 if (rctl_set_dup_ready(p->p_rctls, dup_gp))
334 334 break;
335 335 mutex_exit(&p->p_rctls->rcs_lock);
336 336 rctl_prealloc_destroy(dup_gp);
337 337 }
338 338 e.rcep_p.proc = cp;
339 339 e.rcep_t = RCENTITY_PROCESS;
340 340 cp->p_rctls = rctl_set_dup(p->p_rctls, p, cp, &e, dup_set, dup_gp,
341 341 RCD_DUP | RCD_CALLBACK);
342 342 mutex_exit(&p->p_rctls->rcs_lock);
343 343
344 344 rctl_prealloc_destroy(dup_gp);
345 345
346 346 /*
347 347 * Allocate the child's lwp directory and lwpid hash table.
348 348 */
349 349 if (isfork1)
350 350 cp->p_lwpdir_sz = 2;
351 351 else
352 352 cp->p_lwpdir_sz = p->p_lwpdir_sz;
353 353 cp->p_lwpdir = cp->p_lwpfree = ldp =
354 354 kmem_zalloc(cp->p_lwpdir_sz * sizeof (lwpdir_t), KM_SLEEP);
355 355 for (i = 1; i < cp->p_lwpdir_sz; i++, ldp++)
356 356 ldp->ld_next = ldp + 1;
357 357 cp->p_tidhash_sz = (cp->p_lwpdir_sz + 2) / 2;
358 358 cp->p_tidhash =
359 359 kmem_zalloc(cp->p_tidhash_sz * sizeof (tidhash_t), KM_SLEEP);
360 360
361 361 /*
362 362 * Duplicate parent's lwps.
363 363 * Mutual exclusion is not needed because the process is
364 364 * in the hold state and only the current lwp is running.
365 365 */
366 366 klgrpset_clear(cp->p_lgrpset);
367 367 if (isfork1) {
368 368 clone = forklwp(ttolwp(curthread), cp, curthread->t_tid);
369 369 if (clone == NULL)
370 370 goto forklwperr;
371 371 /*
372 372 * Inherit only the lwp_wait()able flag,
373 373 * Daemon threads should not call fork1(), but oh well...
374 374 */
375 375 lwptot(clone)->t_proc_flag |=
376 376 (curthread->t_proc_flag & TP_TWAIT);
377 377 } else {
378 378 /* this is forkall(), no one can be in lwp_wait() */
379 379 ASSERT(p->p_lwpwait == 0 && p->p_lwpdwait == 0);
380 380 /* for each entry in the parent's lwp directory... */
381 381 for (i = 0, ldp = p->p_lwpdir; i < p->p_lwpdir_sz; i++, ldp++) {
382 382 klwp_t *clwp;
383 383 kthread_t *ct;
384 384
385 385 if ((lep = ldp->ld_entry) == NULL)
386 386 continue;
387 387
388 388 if ((t = lep->le_thread) != NULL) {
389 389 clwp = forklwp(ttolwp(t), cp, t->t_tid);
390 390 if (clwp == NULL)
391 391 goto forklwperr;
392 392 ct = lwptot(clwp);
393 393 /*
394 394 * Inherit lwp_wait()able and daemon flags.
395 395 */
396 396 ct->t_proc_flag |=
397 397 (t->t_proc_flag & (TP_TWAIT|TP_DAEMON));
398 398 /*
399 399 * Keep track of the clone of curthread to
400 400 * post return values through lwp_setrval().
401 401 * Mark other threads for special treatment
402 402 * by lwp_rtt() / post_syscall().
403 403 */
404 404 if (t == curthread)
405 405 clone = clwp;
406 406 else
407 407 ct->t_flag |= T_FORKALL;
408 408 } else {
409 409 /*
410 410 * Replicate zombie lwps in the child.
411 411 */
412 412 clep = kmem_zalloc(sizeof (*clep), KM_SLEEP);
413 413 clep->le_lwpid = lep->le_lwpid;
414 414 clep->le_start = lep->le_start;
415 415 lwp_hash_in(cp, clep,
416 416 cp->p_tidhash, cp->p_tidhash_sz, 0);
417 417 }
418 418 }
419 419 }
420 420
421 421 /*
422 422 * Put new process in the parent's process contract, or put it
423 423 * in a new one if there is an active process template. Send a
424 424 * fork event (if requested) to whatever contract the child is
425 425 * a member of. Fails if the parent has been SIGKILLed.
426 426 */
427 427 if (contract_process_fork(NULL, cp, p, B_TRUE) == NULL)
428 428 goto forklwperr;
429 429
430 430 /*
431 431 * No fork failures occur beyond this point.
432 432 */
433 433
434 434 cp->p_lwpid = p->p_lwpid;
435 435 if (!isfork1) {
436 436 cp->p_lwpdaemon = p->p_lwpdaemon;
437 437 cp->p_zombcnt = p->p_zombcnt;
438 438 /*
439 439 * If the parent's lwp ids have wrapped around, so have the
440 440 * child's.
441 441 */
442 442 cp->p_flag |= p->p_flag & SLWPWRAP;
443 443 }
444 444
445 445 mutex_enter(&p->p_lock);
446 446 corectl_path_hold(cp->p_corefile = p->p_corefile);
447 447 corectl_content_hold(cp->p_content = p->p_content);
448 448 mutex_exit(&p->p_lock);
449 449
450 450 /*
451 451 * Duplicate process context ops, if any.
452 452 */
453 453 if (p->p_pctx)
454 454 forkpctx(p, cp);
455 455
456 456 #ifdef __sparc
457 457 utrap_dup(p, cp);
458 458 #endif
459 459 /*
460 460 * If the child process has been marked to stop on exit
461 461 * from this fork, arrange for all other lwps to stop in
462 462 * sympathy with the active lwp.
463 463 */
464 464 if (PTOU(cp)->u_systrap &&
465 465 prismember(&PTOU(cp)->u_exitmask, curthread->t_sysnum)) {
466 466 mutex_enter(&cp->p_lock);
467 467 t = cp->p_tlist;
468 468 do {
469 469 t->t_proc_flag |= TP_PRSTOP;
470 470 aston(t); /* so TP_PRSTOP will be seen */
471 471 } while ((t = t->t_forw) != cp->p_tlist);
472 472 mutex_exit(&cp->p_lock);
473 473 }
474 474 /*
475 475 * If the parent process has been marked to stop on exit
476 476 * from this fork, and its asynchronous-stop flag has not
477 477 * been set, arrange for all other lwps to stop before
478 478 * they return back to user level.
479 479 */
480 480 if (!(p->p_proc_flag & P_PR_ASYNC) && PTOU(p)->u_systrap &&
481 481 prismember(&PTOU(p)->u_exitmask, curthread->t_sysnum)) {
482 482 mutex_enter(&p->p_lock);
483 483 t = p->p_tlist;
484 484 do {
485 485 t->t_proc_flag |= TP_PRSTOP;
486 486 aston(t); /* so TP_PRSTOP will be seen */
487 487 } while ((t = t->t_forw) != p->p_tlist);
488 488 mutex_exit(&p->p_lock);
489 489 }
490 490
491 491 if (PROC_IS_BRANDED(p))
492 492 BROP(p)->b_lwp_setrval(clone, p->p_pid, 1);
493 493 else
494 494 lwp_setrval(clone, p->p_pid, 1);
495 495
496 496 /* set return values for parent */
497 497 r.r_val1 = (int)cp->p_pid;
498 498 r.r_val2 = 0;
499 499
500 500 /*
501 501 * pool_barrier_exit() can now be called because the child process has:
502 502 * - all identifying features cloned or set (p_pid, p_task, p_pool)
503 503 * - all resource sets associated (p_tlist->*->t_cpupart, p_as->a_mset)
504 504 * - any other fields set which are used in resource set binding.
505 505 */
506 506 mutex_enter(&p->p_lock);
507 507 pool_barrier_exit();
508 508 mutex_exit(&p->p_lock);
509 509
510 510 mutex_enter(&pidlock);
511 511 mutex_enter(&cp->p_lock);
512 512
513 513 /*
514 514 * Set flags telling the child what (not) to do on exit.
515 515 */
516 516 if (flags & FORK_NOSIGCHLD)
517 517 cp->p_pidflag |= CLDNOSIGCHLD;
518 518 if (flags & FORK_WAITPID)
519 519 cp->p_pidflag |= CLDWAITPID;
520 520
521 521 /*
522 522 * Now that there are lwps and threads attached, add the new
523 523 * process to the process group.
524 524 */
525 525 pgjoin(cp, p->p_pgidp);
526 526 cp->p_stat = SRUN;
527 527 /*
528 528 * We are now done with all the lwps in the child process.
529 529 */
530 530 t = cp->p_tlist;
531 531 do {
532 532 /*
533 533 * Set the lwp_suspend()ed lwps running.
534 534 * They will suspend properly at syscall exit.
535 535 */
536 536 if (t->t_proc_flag & TP_HOLDLWP)
537 537 lwp_create_done(t);
538 538 else {
539 539 /* set TS_CREATE to allow continuelwps() to work */
540 540 thread_lock(t);
541 541 ASSERT(t->t_state == TS_STOPPED &&
542 542 !(t->t_schedflag & (TS_CREATE|TS_CSTART)));
543 543 t->t_schedflag |= TS_CREATE;
544 544 thread_unlock(t);
545 545 }
546 546 } while ((t = t->t_forw) != cp->p_tlist);
547 547 mutex_exit(&cp->p_lock);
548 548
549 549 if (isvfork) {
550 550 CPU_STATS_ADDQ(CPU, sys, sysvfork, 1);
551 551 mutex_enter(&p->p_lock);
552 552 p->p_flag |= SVFWAIT;
553 553 curthread->t_flag |= T_VFPARENT;
554 554 DTRACE_PROC1(create, proc_t *, cp);
555 555 cv_broadcast(&pr_pid_cv[p->p_slot]); /* inform /proc */
556 556 mutex_exit(&p->p_lock);
557 557 /*
558 558 * Grab child's p_lock before dropping pidlock to ensure
559 559 * the process will not disappear before we set it running.
560 560 */
561 561 mutex_enter(&cp->p_lock);
562 562 mutex_exit(&pidlock);
563 563 sigdefault(cp);
564 564 continuelwps(cp);
565 565 mutex_exit(&cp->p_lock);
566 566 } else {
567 567 CPU_STATS_ADDQ(CPU, sys, sysfork, 1);
568 568 DTRACE_PROC1(create, proc_t *, cp);
569 569 /*
570 570 * It is CL_FORKRET's job to drop pidlock.
571 571 * If we do it here, the process could be set running
572 572 * and disappear before CL_FORKRET() is called.
573 573 */
574 574 CL_FORKRET(curthread, cp->p_tlist);
575 575 schedctl_set_cidpri(curthread);
576 576 ASSERT(MUTEX_NOT_HELD(&pidlock));
577 577 }
578 578
579 579 return (r.r_vals);
580 580
581 581 forklwperr:
582 582 if (isvfork) {
583 583 if (avl_numnodes(&p->p_wpage) != 0) {
584 584 /* restore watchpoints to parent */
585 585 as = p->p_as;
586 586 AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER);
587 587 as->a_wpage = p->p_wpage;
588 588 avl_create(&p->p_wpage, wp_compare,
589 589 sizeof (struct watched_page),
590 590 offsetof(struct watched_page, wp_link));
591 591 as_setwatch(as);
592 592 AS_LOCK_EXIT(as, &as->a_lock);
593 593 }
594 594 } else {
595 595 if (cp->p_segacct)
596 596 shmexit(cp);
597 597 as = cp->p_as;
598 598 cp->p_as = &kas;
599 599 as_free(as);
600 600 }
601 601
602 602 if (cp->p_lwpdir) {
603 603 for (i = 0, ldp = cp->p_lwpdir; i < cp->p_lwpdir_sz; i++, ldp++)
604 604 if ((lep = ldp->ld_entry) != NULL)
605 605 kmem_free(lep, sizeof (*lep));
606 606 kmem_free(cp->p_lwpdir,
607 607 cp->p_lwpdir_sz * sizeof (*cp->p_lwpdir));
608 608 }
609 609 cp->p_lwpdir = NULL;
610 610 cp->p_lwpfree = NULL;
611 611 cp->p_lwpdir_sz = 0;
612 612
613 613 if (cp->p_tidhash)
614 614 kmem_free(cp->p_tidhash,
615 615 cp->p_tidhash_sz * sizeof (*cp->p_tidhash));
616 616 cp->p_tidhash = NULL;
617 617 cp->p_tidhash_sz = 0;
618 618
619 619 forklwp_fail(cp);
620 620 fork_fail(cp);
↓ open down ↓ |
333 lines elided |
↑ open up ↑ |
621 621 rctl_set_free(cp->p_rctls);
622 622 mutex_enter(&pidlock);
623 623
624 624 /*
625 625 * Detach failed child from task.
626 626 */
627 627 mutex_enter(&cp->p_lock);
628 628 tk = cp->p_task;
629 629 task_detach(cp);
630 630 ASSERT(cp->p_pool->pool_ref > 0);
631 - atomic_add_32(&cp->p_pool->pool_ref, -1);
631 + atomic_dec_32(&cp->p_pool->pool_ref);
632 632 mutex_exit(&cp->p_lock);
633 633
634 634 orphpp = &p->p_orphan;
635 635 while (*orphpp != cp)
636 636 orphpp = &(*orphpp)->p_nextorph;
637 637 *orphpp = cp->p_nextorph;
638 638 if (p->p_child == cp)
639 639 p->p_child = cp->p_sibling;
640 640 if (cp->p_sibling)
641 641 cp->p_sibling->p_psibling = cp->p_psibling;
642 642 if (cp->p_psibling)
643 643 cp->p_psibling->p_sibling = cp->p_sibling;
644 644 pid_exit(cp, tk);
645 645 mutex_exit(&pidlock);
646 646
647 647 task_rele(tk);
648 648
649 649 mutex_enter(&p->p_lock);
650 650 pool_barrier_exit();
651 651 continuelwps(p);
652 652 mutex_exit(&p->p_lock);
653 653 error = EAGAIN;
654 654 forkerr:
655 655 return ((int64_t)set_errno(error));
656 656 }
657 657
658 658 /*
659 659 * Free allocated resources from getproc() if a fork failed.
660 660 */
661 661 static void
662 662 fork_fail(proc_t *cp)
663 663 {
664 664 uf_info_t *fip = P_FINFO(cp);
665 665
666 666 fcnt_add(fip, -1);
667 667 sigdelq(cp, NULL, 0);
668 668
669 669 mutex_enter(&pidlock);
670 670 upcount_dec(crgetruid(cp->p_cred), crgetzoneid(cp->p_cred));
671 671 mutex_exit(&pidlock);
672 672
673 673 /*
674 674 * single threaded, so no locking needed here
675 675 */
676 676 crfree(cp->p_cred);
677 677
678 678 kmem_free(fip->fi_list, fip->fi_nfiles * sizeof (uf_entry_t));
679 679
680 680 VN_RELE(PTOU(curproc)->u_cdir);
681 681 if (PTOU(curproc)->u_rdir)
682 682 VN_RELE(PTOU(curproc)->u_rdir);
683 683 if (cp->p_exec)
684 684 VN_RELE(cp->p_exec);
685 685 if (cp->p_execdir)
686 686 VN_RELE(cp->p_execdir);
687 687 if (PTOU(curproc)->u_cwd)
688 688 refstr_rele(PTOU(curproc)->u_cwd);
689 689 if (PROC_IS_BRANDED(cp)) {
690 690 brand_clearbrand(cp, B_TRUE);
691 691 }
692 692 }
693 693
694 694 /*
695 695 * Clean up the lwps already created for this child process.
696 696 * The fork failed while duplicating all the lwps of the parent
697 697 * and those lwps already created must be freed.
698 698 * This process is invisible to the rest of the system,
699 699 * so we don't need to hold p->p_lock to protect the list.
700 700 */
701 701 static void
702 702 forklwp_fail(proc_t *p)
703 703 {
704 704 kthread_t *t;
705 705 task_t *tk;
706 706 int branded = 0;
707 707
708 708 if (PROC_IS_BRANDED(p))
709 709 branded = 1;
710 710
711 711 while ((t = p->p_tlist) != NULL) {
712 712 /*
713 713 * First remove the lwp from the process's p_tlist.
714 714 */
715 715 if (t != t->t_forw)
716 716 p->p_tlist = t->t_forw;
717 717 else
718 718 p->p_tlist = NULL;
719 719 p->p_lwpcnt--;
720 720 t->t_forw->t_back = t->t_back;
721 721 t->t_back->t_forw = t->t_forw;
722 722
723 723 tk = p->p_task;
724 724 mutex_enter(&p->p_zone->zone_nlwps_lock);
725 725 tk->tk_nlwps--;
726 726 tk->tk_proj->kpj_nlwps--;
727 727 p->p_zone->zone_nlwps--;
728 728 mutex_exit(&p->p_zone->zone_nlwps_lock);
729 729
730 730 ASSERT(t->t_schedctl == NULL);
731 731
732 732 if (branded)
733 733 BROP(p)->b_freelwp(ttolwp(t));
734 734
735 735 if (t->t_door != NULL) {
736 736 kmem_free(t->t_door, sizeof (door_data_t));
737 737 t->t_door = NULL;
738 738 }
739 739 lwp_ctmpl_clear(ttolwp(t));
740 740
741 741 /*
742 742 * Remove the thread from the all threads list.
743 743 * We need to hold pidlock for this.
744 744 */
745 745 mutex_enter(&pidlock);
746 746 t->t_next->t_prev = t->t_prev;
747 747 t->t_prev->t_next = t->t_next;
748 748 CL_EXIT(t); /* tell the scheduler that we're exiting */
749 749 cv_broadcast(&t->t_joincv); /* tell anyone in thread_join */
750 750 mutex_exit(&pidlock);
751 751
752 752 /*
753 753 * Let the lgroup load averages know that this thread isn't
754 754 * going to show up (i.e. un-do what was done on behalf of
755 755 * this thread by the earlier lgrp_move_thread()).
756 756 */
757 757 kpreempt_disable();
758 758 lgrp_move_thread(t, NULL, 1);
759 759 kpreempt_enable();
760 760
761 761 /*
762 762 * The thread was created TS_STOPPED.
763 763 * We change it to TS_FREE to avoid an
764 764 * ASSERT() panic in thread_free().
765 765 */
766 766 t->t_state = TS_FREE;
767 767 thread_rele(t);
768 768 thread_free(t);
769 769 }
770 770 }
771 771
772 772 extern struct as kas;
773 773
774 774 /*
775 775 * fork a kernel process.
776 776 */
777 777 int
778 778 newproc(void (*pc)(), caddr_t arg, id_t cid, int pri, struct contract **ct,
779 779 pid_t pid)
780 780 {
781 781 proc_t *p;
782 782 struct user *up;
783 783 kthread_t *t;
784 784 cont_process_t *ctp = NULL;
785 785 rctl_entity_p_t e;
786 786
787 787 ASSERT(cid != sysdccid);
788 788 ASSERT(cid != syscid || ct == NULL);
789 789 if (CLASS_KERNEL(cid)) {
790 790 rctl_alloc_gp_t *init_gp;
791 791 rctl_set_t *init_set;
792 792
793 793 ASSERT(pid != 1);
794 794
795 795 if (getproc(&p, pid, GETPROC_KERNEL) < 0)
796 796 return (EAGAIN);
797 797
798 798 /*
799 799 * Release the hold on the p_exec and p_execdir, these
800 800 * were acquired in getproc()
801 801 */
802 802 if (p->p_execdir != NULL)
803 803 VN_RELE(p->p_execdir);
804 804 if (p->p_exec != NULL)
805 805 VN_RELE(p->p_exec);
806 806 p->p_flag |= SNOWAIT;
807 807 p->p_exec = NULL;
808 808 p->p_execdir = NULL;
809 809
810 810 init_set = rctl_set_create();
811 811 init_gp = rctl_set_init_prealloc(RCENTITY_PROCESS);
812 812
813 813 /*
814 814 * kernel processes do not inherit /proc tracing flags.
815 815 */
816 816 sigemptyset(&p->p_sigmask);
817 817 premptyset(&p->p_fltmask);
818 818 up = PTOU(p);
819 819 up->u_systrap = 0;
820 820 premptyset(&(up->u_entrymask));
821 821 premptyset(&(up->u_exitmask));
822 822 mutex_enter(&p->p_lock);
823 823 e.rcep_p.proc = p;
824 824 e.rcep_t = RCENTITY_PROCESS;
825 825 p->p_rctls = rctl_set_init(RCENTITY_PROCESS, p, &e, init_set,
826 826 init_gp);
827 827 mutex_exit(&p->p_lock);
828 828
829 829 rctl_prealloc_destroy(init_gp);
830 830
831 831 t = lwp_kernel_create(p, pc, arg, TS_STOPPED, pri);
832 832 } else {
833 833 rctl_alloc_gp_t *init_gp, *default_gp;
834 834 rctl_set_t *init_set;
835 835 task_t *tk, *tk_old;
836 836 klwp_t *lwp;
837 837
838 838 if (getproc(&p, pid, GETPROC_USER) < 0)
839 839 return (EAGAIN);
840 840 /*
841 841 * init creates a new task, distinct from the task
842 842 * containing kernel "processes".
843 843 */
844 844 tk = task_create(0, p->p_zone);
845 845 mutex_enter(&tk->tk_zone->zone_nlwps_lock);
846 846 tk->tk_proj->kpj_ntasks++;
847 847 tk->tk_nprocs++;
848 848 mutex_exit(&tk->tk_zone->zone_nlwps_lock);
849 849
850 850 default_gp = rctl_rlimit_set_prealloc(RLIM_NLIMITS);
851 851 init_gp = rctl_set_init_prealloc(RCENTITY_PROCESS);
852 852 init_set = rctl_set_create();
853 853
854 854 mutex_enter(&pidlock);
855 855 mutex_enter(&p->p_lock);
856 856 tk_old = p->p_task; /* switch to new task */
857 857
858 858 task_detach(p);
859 859 task_begin(tk, p);
860 860 mutex_exit(&pidlock);
861 861
862 862 mutex_enter(&tk_old->tk_zone->zone_nlwps_lock);
863 863 tk_old->tk_nprocs--;
864 864 mutex_exit(&tk_old->tk_zone->zone_nlwps_lock);
865 865
866 866 e.rcep_p.proc = p;
867 867 e.rcep_t = RCENTITY_PROCESS;
868 868 p->p_rctls = rctl_set_init(RCENTITY_PROCESS, p, &e, init_set,
869 869 init_gp);
870 870 rctlproc_default_init(p, default_gp);
871 871 mutex_exit(&p->p_lock);
872 872
873 873 task_rele(tk_old);
874 874 rctl_prealloc_destroy(default_gp);
875 875 rctl_prealloc_destroy(init_gp);
876 876
877 877 if ((lwp = lwp_create(pc, arg, 0, p, TS_STOPPED, pri,
878 878 &curthread->t_hold, cid, 1)) == NULL) {
879 879 task_t *tk;
880 880 fork_fail(p);
881 881 mutex_enter(&pidlock);
882 882 mutex_enter(&p->p_lock);
883 883 tk = p->p_task;
884 884 task_detach(p);
885 885 ASSERT(p->p_pool->pool_ref > 0);
886 886 atomic_add_32(&p->p_pool->pool_ref, -1);
887 887 mutex_exit(&p->p_lock);
888 888 pid_exit(p, tk);
889 889 mutex_exit(&pidlock);
890 890 task_rele(tk);
891 891
892 892 return (EAGAIN);
893 893 }
894 894 t = lwptot(lwp);
895 895
896 896 ctp = contract_process_fork(sys_process_tmpl, p, curproc,
897 897 B_FALSE);
898 898 ASSERT(ctp != NULL);
899 899 if (ct != NULL)
900 900 *ct = &ctp->conp_contract;
901 901 }
902 902
903 903 ASSERT3U(t->t_tid, ==, 1);
904 904 p->p_lwpid = 1;
905 905 mutex_enter(&pidlock);
906 906 pgjoin(p, p->p_parent->p_pgidp);
907 907 p->p_stat = SRUN;
908 908 mutex_enter(&p->p_lock);
909 909 t->t_proc_flag &= ~TP_HOLDLWP;
910 910 lwp_create_done(t);
911 911 mutex_exit(&p->p_lock);
912 912 mutex_exit(&pidlock);
913 913 return (0);
914 914 }
915 915
916 916 /*
917 917 * create a child proc struct.
918 918 */
919 919 static int
920 920 getproc(proc_t **cpp, pid_t pid, uint_t flags)
921 921 {
922 922 proc_t *pp, *cp;
923 923 pid_t newpid;
924 924 struct user *uarea;
925 925 extern uint_t nproc;
926 926 struct cred *cr;
927 927 uid_t ruid;
928 928 zoneid_t zoneid;
929 929 task_t *task;
930 930 kproject_t *proj;
931 931 zone_t *zone;
932 932 int rctlfail = 0;
933 933
934 934 if (zone_status_get(curproc->p_zone) >= ZONE_IS_SHUTTING_DOWN)
935 935 return (-1); /* no point in starting new processes */
936 936
937 937 pp = (flags & GETPROC_KERNEL) ? &p0 : curproc;
938 938 task = pp->p_task;
939 939 proj = task->tk_proj;
940 940 zone = pp->p_zone;
941 941
942 942 mutex_enter(&pp->p_lock);
943 943 mutex_enter(&zone->zone_nlwps_lock);
944 944 if (proj != proj0p) {
945 945 if (task->tk_nprocs >= task->tk_nprocs_ctl)
946 946 if (rctl_test(rc_task_nprocs, task->tk_rctls,
947 947 pp, 1, 0) & RCT_DENY)
948 948 rctlfail = 1;
949 949
950 950 if (proj->kpj_nprocs >= proj->kpj_nprocs_ctl)
951 951 if (rctl_test(rc_project_nprocs, proj->kpj_rctls,
952 952 pp, 1, 0) & RCT_DENY)
953 953 rctlfail = 1;
954 954
955 955 if (zone->zone_nprocs >= zone->zone_nprocs_ctl)
956 956 if (rctl_test(rc_zone_nprocs, zone->zone_rctls,
957 957 pp, 1, 0) & RCT_DENY)
958 958 rctlfail = 1;
959 959
960 960 if (rctlfail) {
961 961 mutex_exit(&zone->zone_nlwps_lock);
962 962 mutex_exit(&pp->p_lock);
963 963 goto punish;
964 964 }
965 965 }
966 966 task->tk_nprocs++;
967 967 proj->kpj_nprocs++;
968 968 zone->zone_nprocs++;
969 969 mutex_exit(&zone->zone_nlwps_lock);
970 970 mutex_exit(&pp->p_lock);
971 971
972 972 cp = kmem_cache_alloc(process_cache, KM_SLEEP);
973 973 bzero(cp, sizeof (proc_t));
974 974
975 975 /*
976 976 * Make proc entry for child process
977 977 */
978 978 mutex_init(&cp->p_splock, NULL, MUTEX_DEFAULT, NULL);
979 979 mutex_init(&cp->p_crlock, NULL, MUTEX_DEFAULT, NULL);
980 980 mutex_init(&cp->p_pflock, NULL, MUTEX_DEFAULT, NULL);
981 981 #if defined(__x86)
982 982 mutex_init(&cp->p_ldtlock, NULL, MUTEX_DEFAULT, NULL);
983 983 #endif
984 984 mutex_init(&cp->p_maplock, NULL, MUTEX_DEFAULT, NULL);
985 985 cp->p_stat = SIDL;
986 986 cp->p_mstart = gethrtime();
987 987 cp->p_as = &kas;
988 988 /*
989 989 * p_zone must be set before we call pid_allocate since the process
990 990 * will be visible after that and code such as prfind_zone will
991 991 * look at the p_zone field.
992 992 */
993 993 cp->p_zone = pp->p_zone;
994 994 cp->p_t1_lgrpid = LGRP_NONE;
995 995 cp->p_tr_lgrpid = LGRP_NONE;
996 996
997 997 if ((newpid = pid_allocate(cp, pid, PID_ALLOC_PROC)) == -1) {
998 998 if (nproc == v.v_proc) {
999 999 CPU_STATS_ADDQ(CPU, sys, procovf, 1);
1000 1000 cmn_err(CE_WARN, "out of processes");
1001 1001 }
1002 1002 goto bad;
1003 1003 }
1004 1004
1005 1005 mutex_enter(&pp->p_lock);
1006 1006 cp->p_exec = pp->p_exec;
1007 1007 cp->p_execdir = pp->p_execdir;
1008 1008 mutex_exit(&pp->p_lock);
1009 1009
1010 1010 if (cp->p_exec) {
1011 1011 VN_HOLD(cp->p_exec);
1012 1012 /*
1013 1013 * Each VOP_OPEN() must be paired with a corresponding
1014 1014 * VOP_CLOSE(). In this case, the executable will be
1015 1015 * closed for the child in either proc_exit() or gexec().
1016 1016 */
1017 1017 if (VOP_OPEN(&cp->p_exec, FREAD, CRED(), NULL) != 0) {
1018 1018 VN_RELE(cp->p_exec);
1019 1019 cp->p_exec = NULLVP;
1020 1020 cp->p_execdir = NULLVP;
1021 1021 goto bad;
1022 1022 }
1023 1023 }
1024 1024 if (cp->p_execdir)
1025 1025 VN_HOLD(cp->p_execdir);
1026 1026
1027 1027 /*
1028 1028 * If not privileged make sure that this user hasn't exceeded
1029 1029 * v.v_maxup processes, and that users collectively haven't
1030 1030 * exceeded v.v_maxupttl processes.
1031 1031 */
1032 1032 mutex_enter(&pidlock);
1033 1033 ASSERT(nproc < v.v_proc); /* otherwise how'd we get our pid? */
1034 1034 cr = CRED();
1035 1035 ruid = crgetruid(cr);
1036 1036 zoneid = crgetzoneid(cr);
1037 1037 if (nproc >= v.v_maxup && /* short-circuit; usually false */
1038 1038 (nproc >= v.v_maxupttl ||
1039 1039 upcount_get(ruid, zoneid) >= v.v_maxup) &&
1040 1040 secpolicy_newproc(cr) != 0) {
1041 1041 mutex_exit(&pidlock);
1042 1042 zcmn_err(zoneid, CE_NOTE,
1043 1043 "out of per-user processes for uid %d", ruid);
1044 1044 goto bad;
1045 1045 }
1046 1046
1047 1047 /*
1048 1048 * Everything is cool, put the new proc on the active process list.
1049 1049 * It is already on the pid list and in /proc.
1050 1050 * Increment the per uid process count (upcount).
1051 1051 */
1052 1052 nproc++;
1053 1053 upcount_inc(ruid, zoneid);
1054 1054
1055 1055 cp->p_next = practive;
1056 1056 practive->p_prev = cp;
1057 1057 practive = cp;
1058 1058
1059 1059 cp->p_ignore = pp->p_ignore;
1060 1060 cp->p_siginfo = pp->p_siginfo;
1061 1061 cp->p_flag = pp->p_flag & (SJCTL|SNOWAIT|SNOCD);
1062 1062 cp->p_sessp = pp->p_sessp;
1063 1063 sess_hold(pp);
1064 1064 cp->p_brand = pp->p_brand;
1065 1065 if (PROC_IS_BRANDED(pp))
1066 1066 BROP(pp)->b_copy_procdata(cp, pp);
1067 1067 cp->p_bssbase = pp->p_bssbase;
1068 1068 cp->p_brkbase = pp->p_brkbase;
1069 1069 cp->p_brksize = pp->p_brksize;
1070 1070 cp->p_brkpageszc = pp->p_brkpageszc;
1071 1071 cp->p_stksize = pp->p_stksize;
1072 1072 cp->p_stkpageszc = pp->p_stkpageszc;
1073 1073 cp->p_stkprot = pp->p_stkprot;
1074 1074 cp->p_datprot = pp->p_datprot;
1075 1075 cp->p_usrstack = pp->p_usrstack;
1076 1076 cp->p_model = pp->p_model;
1077 1077 cp->p_ppid = pp->p_pid;
1078 1078 cp->p_ancpid = pp->p_pid;
1079 1079 cp->p_portcnt = pp->p_portcnt;
1080 1080
1081 1081 /*
1082 1082 * Initialize watchpoint structures
1083 1083 */
1084 1084 avl_create(&cp->p_warea, wa_compare, sizeof (struct watched_area),
1085 1085 offsetof(struct watched_area, wa_link));
1086 1086
1087 1087 /*
1088 1088 * Initialize immediate resource control values.
1089 1089 */
1090 1090 cp->p_stk_ctl = pp->p_stk_ctl;
1091 1091 cp->p_fsz_ctl = pp->p_fsz_ctl;
1092 1092 cp->p_vmem_ctl = pp->p_vmem_ctl;
1093 1093 cp->p_fno_ctl = pp->p_fno_ctl;
1094 1094
1095 1095 /*
1096 1096 * Link up to parent-child-sibling chain. No need to lock
1097 1097 * in general since only a call to freeproc() (done by the
1098 1098 * same parent as newproc()) diddles with the child chain.
1099 1099 */
1100 1100 cp->p_sibling = pp->p_child;
1101 1101 if (pp->p_child)
1102 1102 pp->p_child->p_psibling = cp;
1103 1103
1104 1104 cp->p_parent = pp;
1105 1105 pp->p_child = cp;
1106 1106
1107 1107 cp->p_child_ns = NULL;
1108 1108 cp->p_sibling_ns = NULL;
1109 1109
1110 1110 cp->p_nextorph = pp->p_orphan;
1111 1111 cp->p_nextofkin = pp;
1112 1112 pp->p_orphan = cp;
1113 1113
1114 1114 /*
1115 1115 * Inherit profiling state; do not inherit REALPROF profiling state.
1116 1116 */
1117 1117 cp->p_prof = pp->p_prof;
1118 1118 cp->p_rprof_cyclic = CYCLIC_NONE;
1119 1119
1120 1120 /*
↓ open down ↓ |
479 lines elided |
↑ open up ↑ |
1121 1121 * Inherit pool pointer from the parent. Kernel processes are
1122 1122 * always bound to the default pool.
1123 1123 */
1124 1124 mutex_enter(&pp->p_lock);
1125 1125 if (flags & GETPROC_KERNEL) {
1126 1126 cp->p_pool = pool_default;
1127 1127 cp->p_flag |= SSYS;
1128 1128 } else {
1129 1129 cp->p_pool = pp->p_pool;
1130 1130 }
1131 - atomic_add_32(&cp->p_pool->pool_ref, 1);
1131 + atomic_inc_32(&cp->p_pool->pool_ref);
1132 1132 mutex_exit(&pp->p_lock);
1133 1133
1134 1134 /*
1135 1135 * Add the child process to the current task. Kernel processes
1136 1136 * are always attached to task0.
1137 1137 */
1138 1138 mutex_enter(&cp->p_lock);
1139 1139 if (flags & GETPROC_KERNEL)
1140 1140 task_attach(task0p, cp);
1141 1141 else
1142 1142 task_attach(pp->p_task, cp);
1143 1143 mutex_exit(&cp->p_lock);
1144 1144 mutex_exit(&pidlock);
1145 1145
1146 1146 avl_create(&cp->p_ct_held, contract_compar, sizeof (contract_t),
1147 1147 offsetof(contract_t, ct_ctlist));
1148 1148
1149 1149 /*
1150 1150 * Duplicate any audit information kept in the process table
1151 1151 */
1152 1152 if (audit_active) /* copy audit data to cp */
1153 1153 audit_newproc(cp);
1154 1154
1155 1155 crhold(cp->p_cred = cr);
1156 1156
1157 1157 /*
1158 1158 * Bump up the counts on the file structures pointed at by the
1159 1159 * parent's file table since the child will point at them too.
1160 1160 */
1161 1161 fcnt_add(P_FINFO(pp), 1);
1162 1162
1163 1163 if (PTOU(pp)->u_cdir) {
1164 1164 VN_HOLD(PTOU(pp)->u_cdir);
1165 1165 } else {
1166 1166 ASSERT(pp == &p0);
1167 1167 /*
1168 1168 * We must be at or before vfs_mountroot(); it will take care of
1169 1169 * assigning our current directory.
1170 1170 */
1171 1171 }
1172 1172 if (PTOU(pp)->u_rdir)
1173 1173 VN_HOLD(PTOU(pp)->u_rdir);
1174 1174 if (PTOU(pp)->u_cwd)
1175 1175 refstr_hold(PTOU(pp)->u_cwd);
1176 1176
1177 1177 /*
1178 1178 * copy the parent's uarea.
1179 1179 */
1180 1180 uarea = PTOU(cp);
1181 1181 bcopy(PTOU(pp), uarea, sizeof (*uarea));
1182 1182 flist_fork(P_FINFO(pp), P_FINFO(cp));
1183 1183
1184 1184 gethrestime(&uarea->u_start);
1185 1185 uarea->u_ticks = ddi_get_lbolt();
1186 1186 uarea->u_mem = rm_asrss(pp->p_as);
1187 1187 uarea->u_acflag = AFORK;
1188 1188
1189 1189 /*
1190 1190 * If inherit-on-fork, copy /proc tracing flags to child.
1191 1191 */
1192 1192 if ((pp->p_proc_flag & P_PR_FORK) != 0) {
1193 1193 cp->p_proc_flag |= pp->p_proc_flag & (P_PR_TRACE|P_PR_FORK);
1194 1194 cp->p_sigmask = pp->p_sigmask;
1195 1195 cp->p_fltmask = pp->p_fltmask;
1196 1196 } else {
1197 1197 sigemptyset(&cp->p_sigmask);
1198 1198 premptyset(&cp->p_fltmask);
1199 1199 uarea->u_systrap = 0;
1200 1200 premptyset(&uarea->u_entrymask);
1201 1201 premptyset(&uarea->u_exitmask);
1202 1202 }
1203 1203 /*
1204 1204 * If microstate accounting is being inherited, mark child
1205 1205 */
1206 1206 if ((pp->p_flag & SMSFORK) != 0)
1207 1207 cp->p_flag |= pp->p_flag & (SMSFORK|SMSACCT);
1208 1208
1209 1209 /*
1210 1210 * Inherit fixalignment flag from the parent
1211 1211 */
1212 1212 cp->p_fixalignment = pp->p_fixalignment;
1213 1213
1214 1214 *cpp = cp;
1215 1215 return (0);
1216 1216
1217 1217 bad:
1218 1218 ASSERT(MUTEX_NOT_HELD(&pidlock));
1219 1219
1220 1220 mutex_destroy(&cp->p_crlock);
1221 1221 mutex_destroy(&cp->p_pflock);
1222 1222 #if defined(__x86)
1223 1223 mutex_destroy(&cp->p_ldtlock);
1224 1224 #endif
1225 1225 if (newpid != -1) {
1226 1226 proc_entry_free(cp->p_pidp);
1227 1227 (void) pid_rele(cp->p_pidp);
1228 1228 }
1229 1229 kmem_cache_free(process_cache, cp);
1230 1230
1231 1231 mutex_enter(&zone->zone_nlwps_lock);
1232 1232 task->tk_nprocs--;
1233 1233 proj->kpj_nprocs--;
1234 1234 zone->zone_nprocs--;
1235 1235 mutex_exit(&zone->zone_nlwps_lock);
1236 1236
1237 1237 punish:
1238 1238 /*
1239 1239 * We most likely got into this situation because some process is
1240 1240 * forking out of control. As punishment, put it to sleep for a
1241 1241 * bit so it can't eat the machine alive. Sleep interval is chosen
1242 1242 * to allow no more than one fork failure per cpu per clock tick
1243 1243 * on average (yes, I just made this up). This has two desirable
1244 1244 * properties: (1) it sets a constant limit on the fork failure
1245 1245 * rate, and (2) the busier the system is, the harsher the penalty
1246 1246 * for abusing it becomes.
1247 1247 */
1248 1248 INCR_COUNT(&fork_fail_pending, &pidlock);
1249 1249 delay(fork_fail_pending / ncpus + 1);
1250 1250 DECR_COUNT(&fork_fail_pending, &pidlock);
1251 1251
1252 1252 return (-1); /* out of memory or proc slots */
1253 1253 }
1254 1254
1255 1255 /*
1256 1256 * Release virtual memory.
1257 1257 * In the case of vfork(), the child was given exclusive access to its
1258 1258 * parent's address space. The parent is waiting in vfwait() for the
1259 1259 * child to release its exclusive claim via relvm().
1260 1260 */
1261 1261 void
1262 1262 relvm()
1263 1263 {
1264 1264 proc_t *p = curproc;
1265 1265
1266 1266 ASSERT((unsigned)p->p_lwpcnt <= 1);
1267 1267
1268 1268 prrelvm(); /* inform /proc */
1269 1269
1270 1270 if (p->p_flag & SVFORK) {
1271 1271 proc_t *pp = p->p_parent;
1272 1272 /*
1273 1273 * The child process is either exec'ing or exit'ing.
1274 1274 * The child is now separated from the parent's address
1275 1275 * space. The parent process is made dispatchable.
1276 1276 *
1277 1277 * This is a delicate locking maneuver, involving
1278 1278 * both the parent's p_lock and the child's p_lock.
1279 1279 * As soon as the SVFORK flag is turned off, the
1280 1280 * parent is free to run, but it must not run until
1281 1281 * we wake it up using its p_cv because it might
1282 1282 * exit and we would be referencing invalid memory.
1283 1283 * Therefore, we hold the parent with its p_lock
1284 1284 * while protecting our p_flags with our own p_lock.
1285 1285 */
1286 1286 try_again:
1287 1287 mutex_enter(&p->p_lock); /* grab child's lock first */
1288 1288 prbarrier(p); /* make sure /proc is blocked out */
1289 1289 mutex_enter(&pp->p_lock);
1290 1290
1291 1291 /*
1292 1292 * Check if parent is locked by /proc.
1293 1293 */
1294 1294 if (pp->p_proc_flag & P_PR_LOCK) {
1295 1295 /*
1296 1296 * Delay until /proc is done with the parent.
1297 1297 * We must drop our (the child's) p->p_lock, wait
1298 1298 * via prbarrier() on the parent, then start over.
1299 1299 */
1300 1300 mutex_exit(&p->p_lock);
1301 1301 prbarrier(pp);
1302 1302 mutex_exit(&pp->p_lock);
1303 1303 goto try_again;
1304 1304 }
1305 1305 p->p_flag &= ~SVFORK;
1306 1306 kpreempt_disable();
1307 1307 p->p_as = &kas;
1308 1308
1309 1309 /*
1310 1310 * notify hat of change in thread's address space
1311 1311 */
1312 1312 hat_thread_exit(curthread);
1313 1313 kpreempt_enable();
1314 1314
1315 1315 /*
1316 1316 * child sizes are copied back to parent because
1317 1317 * child may have grown.
1318 1318 */
1319 1319 pp->p_brkbase = p->p_brkbase;
1320 1320 pp->p_brksize = p->p_brksize;
1321 1321 pp->p_stksize = p->p_stksize;
1322 1322
1323 1323 /*
1324 1324 * Copy back the shm accounting information
1325 1325 * to the parent process.
1326 1326 */
1327 1327 pp->p_segacct = p->p_segacct;
1328 1328 p->p_segacct = NULL;
1329 1329
1330 1330 /*
1331 1331 * The parent is no longer waiting for the vfork()d child.
1332 1332 * Restore the parent's watched pages, if any. This is
1333 1333 * safe because we know the parent is not locked by /proc
1334 1334 */
1335 1335 pp->p_flag &= ~SVFWAIT;
1336 1336 if (avl_numnodes(&pp->p_wpage) != 0) {
1337 1337 pp->p_as->a_wpage = pp->p_wpage;
1338 1338 avl_create(&pp->p_wpage, wp_compare,
1339 1339 sizeof (struct watched_page),
1340 1340 offsetof(struct watched_page, wp_link));
1341 1341 }
1342 1342 cv_signal(&pp->p_cv);
1343 1343 mutex_exit(&pp->p_lock);
1344 1344 mutex_exit(&p->p_lock);
1345 1345 } else {
1346 1346 if (p->p_as != &kas) {
1347 1347 struct as *as;
1348 1348
1349 1349 if (p->p_segacct)
1350 1350 shmexit(p);
1351 1351
1352 1352 /*
1353 1353 * We grab p_lock for the benefit of /proc
1354 1354 */
1355 1355 kpreempt_disable();
1356 1356 mutex_enter(&p->p_lock);
1357 1357 prbarrier(p); /* make sure /proc is blocked out */
1358 1358 as = p->p_as;
1359 1359 p->p_as = &kas;
1360 1360 mutex_exit(&p->p_lock);
1361 1361
1362 1362 /*
1363 1363 * notify hat of change in thread's address space
1364 1364 */
1365 1365 hat_thread_exit(curthread);
1366 1366 kpreempt_enable();
1367 1367
1368 1368 as_free(as);
1369 1369 p->p_tr_lgrpid = LGRP_NONE;
1370 1370 }
1371 1371 }
1372 1372 }
1373 1373
1374 1374 /*
1375 1375 * Wait for child to exec or exit.
1376 1376 * Called by parent of vfork'ed process.
1377 1377 * See important comments in relvm(), above.
1378 1378 */
1379 1379 void
1380 1380 vfwait(pid_t pid)
1381 1381 {
1382 1382 int signalled = 0;
1383 1383 proc_t *pp = ttoproc(curthread);
1384 1384 proc_t *cp;
1385 1385
1386 1386 /*
1387 1387 * Wait for child to exec or exit.
1388 1388 */
1389 1389 for (;;) {
1390 1390 mutex_enter(&pidlock);
1391 1391 cp = prfind(pid);
1392 1392 if (cp == NULL || cp->p_parent != pp) {
1393 1393 /*
1394 1394 * Child has exit()ed.
1395 1395 */
1396 1396 mutex_exit(&pidlock);
1397 1397 break;
1398 1398 }
1399 1399 /*
1400 1400 * Grab the child's p_lock before releasing pidlock.
1401 1401 * Otherwise, the child could exit and we would be
1402 1402 * referencing invalid memory.
1403 1403 */
1404 1404 mutex_enter(&cp->p_lock);
1405 1405 mutex_exit(&pidlock);
1406 1406 if (!(cp->p_flag & SVFORK)) {
1407 1407 /*
1408 1408 * Child has exec()ed or is exit()ing.
1409 1409 */
1410 1410 mutex_exit(&cp->p_lock);
1411 1411 break;
1412 1412 }
1413 1413 mutex_enter(&pp->p_lock);
1414 1414 mutex_exit(&cp->p_lock);
1415 1415 /*
1416 1416 * We might be waked up spuriously from the cv_wait().
1417 1417 * We have to do the whole operation over again to be
1418 1418 * sure the child's SVFORK flag really is turned off.
1419 1419 * We cannot make reference to the child because it can
1420 1420 * exit before we return and we would be referencing
1421 1421 * invalid memory.
1422 1422 *
1423 1423 * Because this is potentially a very long-term wait,
1424 1424 * we call cv_wait_sig() (for its jobcontrol and /proc
1425 1425 * side-effects) unless there is a current signal, in
1426 1426 * which case we use cv_wait() because we cannot return
1427 1427 * from this function until the child has released the
1428 1428 * address space. Calling cv_wait_sig() with a current
1429 1429 * signal would lead to an indefinite loop here because
1430 1430 * cv_wait_sig() returns immediately in this case.
1431 1431 */
1432 1432 if (signalled)
1433 1433 cv_wait(&pp->p_cv, &pp->p_lock);
1434 1434 else
1435 1435 signalled = !cv_wait_sig(&pp->p_cv, &pp->p_lock);
1436 1436 mutex_exit(&pp->p_lock);
1437 1437 }
1438 1438
1439 1439 /* restore watchpoints to parent */
1440 1440 if (pr_watch_active(pp)) {
1441 1441 struct as *as = pp->p_as;
1442 1442 AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER);
1443 1443 as_setwatch(as);
1444 1444 AS_LOCK_EXIT(as, &as->a_lock);
1445 1445 }
1446 1446
1447 1447 mutex_enter(&pp->p_lock);
1448 1448 prbarrier(pp); /* barrier against /proc locking */
1449 1449 continuelwps(pp);
1450 1450 mutex_exit(&pp->p_lock);
1451 1451 }
↓ open down ↓ |
310 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX