1 /*- 2 * Copyright (C) 2001 Julian Elischer <julian@freebsd.org>. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice(s), this list of conditions and the following disclaimer as 10 * the first lines of this file unmodified other than the possible 11 * addition of one or more copyright notices. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice(s), this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY 17 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 18 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 19 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY 20 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 21 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 22 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 23 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH 26 * DAMAGE. 27 */ 28 29 #include "opt_witness.h" 30 #include "opt_hwpmc_hooks.h" 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/kernel.h> 38 #include <sys/lock.h> 39 #include <sys/mutex.h> 40 #include <sys/proc.h> 41 #include <sys/rangelock.h> 42 #include <sys/resourcevar.h> 43 #include <sys/sdt.h> 44 #include <sys/smp.h> 45 #include <sys/sched.h> 46 #include <sys/sleepqueue.h> 47 #include <sys/selinfo.h> 48 #include <sys/turnstile.h> 49 #include <sys/ktr.h> 50 #include <sys/rwlock.h> 51 #include <sys/umtx.h> 52 #include <sys/cpuset.h> 53 #ifdef HWPMC_HOOKS 54 #include <sys/pmckern.h> 55 #endif 56 57 #include <security/audit/audit.h> 58 59 #include <vm/vm.h> 60 #include <vm/vm_extern.h> 61 #include <vm/uma.h> 62 #include <sys/eventhandler.h> 63 64 SDT_PROVIDER_DECLARE(proc); 65 SDT_PROBE_DEFINE(proc, , , lwp__exit); 66 67 /* 68 * thread related storage. 69 */ 70 static uma_zone_t thread_zone; 71 72 TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads); 73 static struct mtx zombie_lock; 74 MTX_SYSINIT(zombie_lock, &zombie_lock, "zombie lock", MTX_SPIN); 75 76 static void thread_zombie(struct thread *); 77 static int thread_unsuspend_one(struct thread *td, struct proc *p, 78 bool boundary); 79 80 #define TID_BUFFER_SIZE 1024 81 82 struct mtx tid_lock; 83 static struct unrhdr *tid_unrhdr; 84 static lwpid_t tid_buffer[TID_BUFFER_SIZE]; 85 static int tid_head, tid_tail; 86 static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash"); 87 88 struct tidhashhead *tidhashtbl; 89 u_long tidhash; 90 struct rwlock tidhash_lock; 91 92 static lwpid_t 93 tid_alloc(void) 94 { 95 lwpid_t tid; 96 97 tid = alloc_unr(tid_unrhdr); 98 if (tid != -1) 99 return (tid); 100 mtx_lock(&tid_lock); 101 if (tid_head == tid_tail) { 102 mtx_unlock(&tid_lock); 103 return (-1); 104 } 105 tid = tid_buffer[tid_head]; 106 tid_head = (tid_head + 1) % TID_BUFFER_SIZE; 107 mtx_unlock(&tid_lock); 108 return (tid); 109 } 110 111 static void 112 tid_free(lwpid_t tid) 113 { 114 lwpid_t tmp_tid = -1; 115 116 mtx_lock(&tid_lock); 117 if ((tid_tail + 1) % TID_BUFFER_SIZE == tid_head) { 118 tmp_tid = tid_buffer[tid_head]; 119 tid_head = (tid_head + 1) % TID_BUFFER_SIZE; 120 } 121 tid_buffer[tid_tail] = tid; 122 tid_tail = (tid_tail + 1) % TID_BUFFER_SIZE; 123 mtx_unlock(&tid_lock); 124 if (tmp_tid != -1) 125 free_unr(tid_unrhdr, tmp_tid); 126 } 127 128 /* 129 * Prepare a thread for use. 130 */ 131 static int 132 thread_ctor(void *mem, int size, void *arg, int flags) 133 { 134 struct thread *td; 135 136 td = (struct thread *)mem; 137 td->td_state = TDS_INACTIVE; 138 td->td_oncpu = NOCPU; 139 140 td->td_tid = tid_alloc(); 141 142 /* 143 * Note that td_critnest begins life as 1 because the thread is not 144 * running and is thereby implicitly waiting to be on the receiving 145 * end of a context switch. 146 */ 147 td->td_critnest = 1; 148 td->td_lend_user_pri = PRI_MAX; 149 EVENTHANDLER_INVOKE(thread_ctor, td); 150 #ifdef AUDIT 151 audit_thread_alloc(td); 152 #endif 153 umtx_thread_alloc(td); 154 return (0); 155 } 156 157 /* 158 * Reclaim a thread after use. 159 */ 160 static void 161 thread_dtor(void *mem, int size, void *arg) 162 { 163 struct thread *td; 164 165 td = (struct thread *)mem; 166 167 #ifdef INVARIANTS 168 /* Verify that this thread is in a safe state to free. */ 169 switch (td->td_state) { 170 case TDS_INHIBITED: 171 case TDS_RUNNING: 172 case TDS_CAN_RUN: 173 case TDS_RUNQ: 174 /* 175 * We must never unlink a thread that is in one of 176 * these states, because it is currently active. 177 */ 178 panic("bad state for thread unlinking"); 179 /* NOTREACHED */ 180 case TDS_INACTIVE: 181 break; 182 default: 183 panic("bad thread state"); 184 /* NOTREACHED */ 185 } 186 #endif 187 #ifdef AUDIT 188 audit_thread_free(td); 189 #endif 190 /* Free all OSD associated to this thread. */ 191 osd_thread_exit(td); 192 193 EVENTHANDLER_INVOKE(thread_dtor, td); 194 tid_free(td->td_tid); 195 } 196 197 /* 198 * Initialize type-stable parts of a thread (when newly created). 199 */ 200 static int 201 thread_init(void *mem, int size, int flags) 202 { 203 struct thread *td; 204 205 td = (struct thread *)mem; 206 207 td->td_sleepqueue = sleepq_alloc(); 208 td->td_turnstile = turnstile_alloc(); 209 td->td_rlqe = NULL; 210 EVENTHANDLER_INVOKE(thread_init, td); 211 td->td_sched = (struct td_sched *)&td[1]; 212 umtx_thread_init(td); 213 td->td_kstack = 0; 214 td->td_sel = NULL; 215 return (0); 216 } 217 218 /* 219 * Tear down type-stable parts of a thread (just before being discarded). 220 */ 221 static void 222 thread_fini(void *mem, int size) 223 { 224 struct thread *td; 225 226 td = (struct thread *)mem; 227 EVENTHANDLER_INVOKE(thread_fini, td); 228 rlqentry_free(td->td_rlqe); 229 turnstile_free(td->td_turnstile); 230 sleepq_free(td->td_sleepqueue); 231 umtx_thread_fini(td); 232 seltdfini(td); 233 } 234 235 /* 236 * For a newly created process, 237 * link up all the structures and its initial threads etc. 238 * called from: 239 * {arch}/{arch}/machdep.c {arch}_init(), init386() etc. 240 * proc_dtor() (should go away) 241 * proc_init() 242 */ 243 void 244 proc_linkup0(struct proc *p, struct thread *td) 245 { 246 TAILQ_INIT(&p->p_threads); /* all threads in proc */ 247 proc_linkup(p, td); 248 } 249 250 void 251 proc_linkup(struct proc *p, struct thread *td) 252 { 253 254 sigqueue_init(&p->p_sigqueue, p); 255 p->p_ksi = ksiginfo_alloc(1); 256 if (p->p_ksi != NULL) { 257 /* XXX p_ksi may be null if ksiginfo zone is not ready */ 258 p->p_ksi->ksi_flags = KSI_EXT | KSI_INS; 259 } 260 LIST_INIT(&p->p_mqnotifier); 261 p->p_numthreads = 0; 262 thread_link(td, p); 263 } 264 265 /* 266 * Initialize global thread allocation resources. 267 */ 268 void 269 threadinit(void) 270 { 271 272 mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF); 273 274 /* 275 * pid_max cannot be greater than PID_MAX. 276 * leave one number for thread0. 277 */ 278 tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock); 279 280 thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(), 281 thread_ctor, thread_dtor, thread_init, thread_fini, 282 16 - 1, 0); 283 tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash); 284 rw_init(&tidhash_lock, "tidhash"); 285 } 286 287 /* 288 * Place an unused thread on the zombie list. 289 * Use the slpq as that must be unused by now. 290 */ 291 void 292 thread_zombie(struct thread *td) 293 { 294 mtx_lock_spin(&zombie_lock); 295 TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq); 296 mtx_unlock_spin(&zombie_lock); 297 } 298 299 /* 300 * Release a thread that has exited after cpu_throw(). 301 */ 302 void 303 thread_stash(struct thread *td) 304 { 305 atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1); 306 thread_zombie(td); 307 } 308 309 /* 310 * Reap zombie resources. 311 */ 312 void 313 thread_reap(void) 314 { 315 struct thread *td_first, *td_next; 316 317 /* 318 * Don't even bother to lock if none at this instant, 319 * we really don't care about the next instant.. 320 */ 321 if (!TAILQ_EMPTY(&zombie_threads)) { 322 mtx_lock_spin(&zombie_lock); 323 td_first = TAILQ_FIRST(&zombie_threads); 324 if (td_first) 325 TAILQ_INIT(&zombie_threads); 326 mtx_unlock_spin(&zombie_lock); 327 while (td_first) { 328 td_next = TAILQ_NEXT(td_first, td_slpq); 329 if (td_first->td_ucred) 330 crfree(td_first->td_ucred); 331 thread_free(td_first); 332 td_first = td_next; 333 } 334 } 335 } 336 337 /* 338 * Allocate a thread. 339 */ 340 struct thread * 341 thread_alloc(int pages) 342 { 343 struct thread *td; 344 345 thread_reap(); /* check if any zombies to get */ 346 347 td = (struct thread *)uma_zalloc(thread_zone, M_WAITOK); 348 KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack")); 349 if (!vm_thread_new(td, pages)) { 350 uma_zfree(thread_zone, td); 351 return (NULL); 352 } 353 cpu_thread_alloc(td); 354 return (td); 355 } 356 357 int 358 thread_alloc_stack(struct thread *td, int pages) 359 { 360 361 KASSERT(td->td_kstack == 0, 362 ("thread_alloc_stack called on a thread with kstack")); 363 if (!vm_thread_new(td, pages)) 364 return (0); 365 cpu_thread_alloc(td); 366 return (1); 367 } 368 369 /* 370 * Deallocate a thread. 371 */ 372 void 373 thread_free(struct thread *td) 374 { 375 376 lock_profile_thread_exit(td); 377 if (td->td_cpuset) 378 cpuset_rel(td->td_cpuset); 379 td->td_cpuset = NULL; 380 cpu_thread_free(td); 381 if (td->td_kstack != 0) 382 vm_thread_dispose(td); 383 uma_zfree(thread_zone, td); 384 } 385 386 /* 387 * Discard the current thread and exit from its context. 388 * Always called with scheduler locked. 389 * 390 * Because we can't free a thread while we're operating under its context, 391 * push the current thread into our CPU's deadthread holder. This means 392 * we needn't worry about someone else grabbing our context before we 393 * do a cpu_throw(). 394 */ 395 void 396 thread_exit(void) 397 { 398 uint64_t runtime, new_switchtime; 399 struct thread *td; 400 struct thread *td2; 401 struct proc *p; 402 int wakeup_swapper; 403 404 td = curthread; 405 p = td->td_proc; 406 407 PROC_SLOCK_ASSERT(p, MA_OWNED); 408 mtx_assert(&Giant, MA_NOTOWNED); 409 410 PROC_LOCK_ASSERT(p, MA_OWNED); 411 KASSERT(p != NULL, ("thread exiting without a process")); 412 CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td, 413 (long)p->p_pid, td->td_name); 414 KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending")); 415 416 #ifdef AUDIT 417 AUDIT_SYSCALL_EXIT(0, td); 418 #endif 419 /* 420 * drop FPU & debug register state storage, or any other 421 * architecture specific resources that 422 * would not be on a new untouched process. 423 */ 424 cpu_thread_exit(td); /* XXXSMP */ 425 426 /* 427 * The last thread is left attached to the process 428 * So that the whole bundle gets recycled. Skip 429 * all this stuff if we never had threads. 430 * EXIT clears all sign of other threads when 431 * it goes to single threading, so the last thread always 432 * takes the short path. 433 */ 434 if (p->p_flag & P_HADTHREADS) { 435 if (p->p_numthreads > 1) { 436 atomic_add_int(&td->td_proc->p_exitthreads, 1); 437 thread_unlink(td); 438 td2 = FIRST_THREAD_IN_PROC(p); 439 sched_exit_thread(td2, td); 440 441 /* 442 * The test below is NOT true if we are the 443 * sole exiting thread. P_STOPPED_SINGLE is unset 444 * in exit1() after it is the only survivor. 445 */ 446 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 447 if (p->p_numthreads == p->p_suspcount) { 448 thread_lock(p->p_singlethread); 449 wakeup_swapper = thread_unsuspend_one( 450 p->p_singlethread, p, false); 451 thread_unlock(p->p_singlethread); 452 if (wakeup_swapper) 453 kick_proc0(); 454 } 455 } 456 457 PCPU_SET(deadthread, td); 458 } else { 459 /* 460 * The last thread is exiting.. but not through exit() 461 */ 462 panic ("thread_exit: Last thread exiting on its own"); 463 } 464 } 465 #ifdef HWPMC_HOOKS 466 /* 467 * If this thread is part of a process that is being tracked by hwpmc(4), 468 * inform the module of the thread's impending exit. 469 */ 470 if (PMC_PROC_IS_USING_PMCS(td->td_proc)) 471 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT); 472 #endif 473 PROC_UNLOCK(p); 474 PROC_STATLOCK(p); 475 thread_lock(td); 476 PROC_SUNLOCK(p); 477 478 /* Do the same timestamp bookkeeping that mi_switch() would do. */ 479 new_switchtime = cpu_ticks(); 480 runtime = new_switchtime - PCPU_GET(switchtime); 481 td->td_runtime += runtime; 482 td->td_incruntime += runtime; 483 PCPU_SET(switchtime, new_switchtime); 484 PCPU_SET(switchticks, ticks); 485 PCPU_INC(cnt.v_swtch); 486 487 /* Save our resource usage in our process. */ 488 td->td_ru.ru_nvcsw++; 489 ruxagg(p, td); 490 rucollect(&p->p_ru, &td->td_ru); 491 PROC_STATUNLOCK(p); 492 493 td->td_state = TDS_INACTIVE; 494 #ifdef WITNESS 495 witness_thread_exit(td); 496 #endif 497 CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td); 498 sched_throw(td); 499 panic("I'm a teapot!"); 500 /* NOTREACHED */ 501 } 502 503 /* 504 * Do any thread specific cleanups that may be needed in wait() 505 * called with Giant, proc and schedlock not held. 506 */ 507 void 508 thread_wait(struct proc *p) 509 { 510 struct thread *td; 511 512 mtx_assert(&Giant, MA_NOTOWNED); 513 KASSERT(p->p_numthreads == 1, ("multiple threads in thread_wait()")); 514 KASSERT(p->p_exitthreads == 0, ("p_exitthreads leaking")); 515 td = FIRST_THREAD_IN_PROC(p); 516 /* Lock the last thread so we spin until it exits cpu_throw(). */ 517 thread_lock(td); 518 thread_unlock(td); 519 lock_profile_thread_exit(td); 520 cpuset_rel(td->td_cpuset); 521 td->td_cpuset = NULL; 522 cpu_thread_clean(td); 523 crfree(td->td_ucred); 524 thread_reap(); /* check for zombie threads etc. */ 525 } 526 527 /* 528 * Link a thread to a process. 529 * set up anything that needs to be initialized for it to 530 * be used by the process. 531 */ 532 void 533 thread_link(struct thread *td, struct proc *p) 534 { 535 536 /* 537 * XXX This can't be enabled because it's called for proc0 before 538 * its lock has been created. 539 * PROC_LOCK_ASSERT(p, MA_OWNED); 540 */ 541 td->td_state = TDS_INACTIVE; 542 td->td_proc = p; 543 td->td_flags = TDF_INMEM; 544 545 LIST_INIT(&td->td_contested); 546 LIST_INIT(&td->td_lprof[0]); 547 LIST_INIT(&td->td_lprof[1]); 548 sigqueue_init(&td->td_sigqueue, p); 549 callout_init(&td->td_slpcallout, 1); 550 TAILQ_INSERT_TAIL(&p->p_threads, td, td_plist); 551 p->p_numthreads++; 552 } 553 554 /* 555 * Called from: 556 * thread_exit() 557 */ 558 void 559 thread_unlink(struct thread *td) 560 { 561 struct proc *p = td->td_proc; 562 563 PROC_LOCK_ASSERT(p, MA_OWNED); 564 TAILQ_REMOVE(&p->p_threads, td, td_plist); 565 p->p_numthreads--; 566 /* could clear a few other things here */ 567 /* Must NOT clear links to proc! */ 568 } 569 570 static int 571 calc_remaining(struct proc *p, int mode) 572 { 573 int remaining; 574 575 PROC_LOCK_ASSERT(p, MA_OWNED); 576 PROC_SLOCK_ASSERT(p, MA_OWNED); 577 if (mode == SINGLE_EXIT) 578 remaining = p->p_numthreads; 579 else if (mode == SINGLE_BOUNDARY) 580 remaining = p->p_numthreads - p->p_boundary_count; 581 else if (mode == SINGLE_NO_EXIT || mode == SINGLE_ALLPROC) 582 remaining = p->p_numthreads - p->p_suspcount; 583 else 584 panic("calc_remaining: wrong mode %d", mode); 585 return (remaining); 586 } 587 588 static int 589 remain_for_mode(int mode) 590 { 591 592 return (mode == SINGLE_ALLPROC ? 0 : 1); 593 } 594 595 static int 596 weed_inhib(int mode, struct thread *td2, struct proc *p) 597 { 598 int wakeup_swapper; 599 600 PROC_LOCK_ASSERT(p, MA_OWNED); 601 PROC_SLOCK_ASSERT(p, MA_OWNED); 602 THREAD_LOCK_ASSERT(td2, MA_OWNED); 603 604 wakeup_swapper = 0; 605 switch (mode) { 606 case SINGLE_EXIT: 607 if (TD_IS_SUSPENDED(td2)) 608 wakeup_swapper |= thread_unsuspend_one(td2, p, true); 609 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) 610 wakeup_swapper |= sleepq_abort(td2, EINTR); 611 break; 612 case SINGLE_BOUNDARY: 613 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & TDF_BOUNDARY) == 0) 614 wakeup_swapper |= thread_unsuspend_one(td2, p, false); 615 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) 616 wakeup_swapper |= sleepq_abort(td2, ERESTART); 617 break; 618 case SINGLE_NO_EXIT: 619 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & TDF_BOUNDARY) == 0) 620 wakeup_swapper |= thread_unsuspend_one(td2, p, false); 621 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) 622 wakeup_swapper |= sleepq_abort(td2, ERESTART); 623 break; 624 case SINGLE_ALLPROC: 625 /* 626 * ALLPROC suspend tries to avoid spurious EINTR for 627 * threads sleeping interruptable, by suspending the 628 * thread directly, similarly to sig_suspend_threads(). 629 * Since such sleep is not performed at the user 630 * boundary, TDF_BOUNDARY flag is not set, and TDF_ALLPROCSUSP 631 * is used to avoid immediate un-suspend. 632 */ 633 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & (TDF_BOUNDARY | 634 TDF_ALLPROCSUSP)) == 0) 635 wakeup_swapper |= thread_unsuspend_one(td2, p, false); 636 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) { 637 if ((td2->td_flags & TDF_SBDRY) == 0) { 638 thread_suspend_one(td2); 639 td2->td_flags |= TDF_ALLPROCSUSP; 640 } else { 641 wakeup_swapper |= sleepq_abort(td2, ERESTART); 642 } 643 } 644 break; 645 } 646 return (wakeup_swapper); 647 } 648 649 /* 650 * Enforce single-threading. 651 * 652 * Returns 1 if the caller must abort (another thread is waiting to 653 * exit the process or similar). Process is locked! 654 * Returns 0 when you are successfully the only thread running. 655 * A process has successfully single threaded in the suspend mode when 656 * There are no threads in user mode. Threads in the kernel must be 657 * allowed to continue until they get to the user boundary. They may even 658 * copy out their return values and data before suspending. They may however be 659 * accelerated in reaching the user boundary as we will wake up 660 * any sleeping threads that are interruptable. (PCATCH). 661 */ 662 int 663 thread_single(struct proc *p, int mode) 664 { 665 struct thread *td; 666 struct thread *td2; 667 int remaining, wakeup_swapper; 668 669 td = curthread; 670 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY || 671 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT, 672 ("invalid mode %d", mode)); 673 /* 674 * If allowing non-ALLPROC singlethreading for non-curproc 675 * callers, calc_remaining() and remain_for_mode() should be 676 * adjusted to also account for td->td_proc != p. For now 677 * this is not implemented because it is not used. 678 */ 679 KASSERT((mode == SINGLE_ALLPROC && td->td_proc != p) || 680 (mode != SINGLE_ALLPROC && td->td_proc == p), 681 ("mode %d proc %p curproc %p", mode, p, td->td_proc)); 682 mtx_assert(&Giant, MA_NOTOWNED); 683 PROC_LOCK_ASSERT(p, MA_OWNED); 684 685 if ((p->p_flag & P_HADTHREADS) == 0 && mode != SINGLE_ALLPROC) 686 return (0); 687 688 /* Is someone already single threading? */ 689 if (p->p_singlethread != NULL && p->p_singlethread != td) 690 return (1); 691 692 if (mode == SINGLE_EXIT) { 693 p->p_flag |= P_SINGLE_EXIT; 694 p->p_flag &= ~P_SINGLE_BOUNDARY; 695 } else { 696 p->p_flag &= ~P_SINGLE_EXIT; 697 if (mode == SINGLE_BOUNDARY) 698 p->p_flag |= P_SINGLE_BOUNDARY; 699 else 700 p->p_flag &= ~P_SINGLE_BOUNDARY; 701 } 702 if (mode == SINGLE_ALLPROC) 703 p->p_flag |= P_TOTAL_STOP; 704 p->p_flag |= P_STOPPED_SINGLE; 705 PROC_SLOCK(p); 706 p->p_singlethread = td; 707 remaining = calc_remaining(p, mode); 708 while (remaining != remain_for_mode(mode)) { 709 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE) 710 goto stopme; 711 wakeup_swapper = 0; 712 FOREACH_THREAD_IN_PROC(p, td2) { 713 if (td2 == td) 714 continue; 715 thread_lock(td2); 716 td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK; 717 if (TD_IS_INHIBITED(td2)) { 718 wakeup_swapper |= weed_inhib(mode, td2, p); 719 #ifdef SMP 720 } else if (TD_IS_RUNNING(td2) && td != td2) { 721 forward_signal(td2); 722 #endif 723 } 724 thread_unlock(td2); 725 } 726 if (wakeup_swapper) 727 kick_proc0(); 728 remaining = calc_remaining(p, mode); 729 730 /* 731 * Maybe we suspended some threads.. was it enough? 732 */ 733 if (remaining == remain_for_mode(mode)) 734 break; 735 736 stopme: 737 /* 738 * Wake us up when everyone else has suspended. 739 * In the mean time we suspend as well. 740 */ 741 thread_suspend_switch(td, p); 742 remaining = calc_remaining(p, mode); 743 } 744 if (mode == SINGLE_EXIT) { 745 /* 746 * Convert the process to an unthreaded process. The 747 * SINGLE_EXIT is called by exit1() or execve(), in 748 * both cases other threads must be retired. 749 */ 750 KASSERT(p->p_numthreads == 1, ("Unthreading with >1 threads")); 751 p->p_singlethread = NULL; 752 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_HADTHREADS); 753 754 /* 755 * Wait for any remaining threads to exit cpu_throw(). 756 */ 757 while (p->p_exitthreads != 0) { 758 PROC_SUNLOCK(p); 759 PROC_UNLOCK(p); 760 sched_relinquish(td); 761 PROC_LOCK(p); 762 PROC_SLOCK(p); 763 } 764 } else if (mode == SINGLE_BOUNDARY) { 765 /* 766 * Wait until all suspended threads are removed from 767 * the processors. The thread_suspend_check() 768 * increments p_boundary_count while it is still 769 * running, which makes it possible for the execve() 770 * to destroy vmspace while our other threads are 771 * still using the address space. 772 * 773 * We lock the thread, which is only allowed to 774 * succeed after context switch code finished using 775 * the address space. 776 */ 777 FOREACH_THREAD_IN_PROC(p, td2) { 778 if (td2 == td) 779 continue; 780 thread_lock(td2); 781 KASSERT((td2->td_flags & TDF_BOUNDARY) != 0, 782 ("td %p not on boundary", td2)); 783 KASSERT(TD_IS_SUSPENDED(td2), 784 ("td %p is not suspended", td2)); 785 thread_unlock(td2); 786 } 787 } 788 PROC_SUNLOCK(p); 789 return (0); 790 } 791 792 bool 793 thread_suspend_check_needed(void) 794 { 795 struct proc *p; 796 struct thread *td; 797 798 td = curthread; 799 p = td->td_proc; 800 PROC_LOCK_ASSERT(p, MA_OWNED); 801 return (P_SHOULDSTOP(p) || ((p->p_flag & P_TRACED) != 0 && 802 (td->td_dbgflags & TDB_SUSPEND) != 0)); 803 } 804 805 /* 806 * Called in from locations that can safely check to see 807 * whether we have to suspend or at least throttle for a 808 * single-thread event (e.g. fork). 809 * 810 * Such locations include userret(). 811 * If the "return_instead" argument is non zero, the thread must be able to 812 * accept 0 (caller may continue), or 1 (caller must abort) as a result. 813 * 814 * The 'return_instead' argument tells the function if it may do a 815 * thread_exit() or suspend, or whether the caller must abort and back 816 * out instead. 817 * 818 * If the thread that set the single_threading request has set the 819 * P_SINGLE_EXIT bit in the process flags then this call will never return 820 * if 'return_instead' is false, but will exit. 821 * 822 * P_SINGLE_EXIT | return_instead == 0| return_instead != 0 823 *---------------+--------------------+--------------------- 824 * 0 | returns 0 | returns 0 or 1 825 * | when ST ends | immediately 826 *---------------+--------------------+--------------------- 827 * 1 | thread exits | returns 1 828 * | | immediately 829 * 0 = thread_exit() or suspension ok, 830 * other = return error instead of stopping the thread. 831 * 832 * While a full suspension is under effect, even a single threading 833 * thread would be suspended if it made this call (but it shouldn't). 834 * This call should only be made from places where 835 * thread_exit() would be safe as that may be the outcome unless 836 * return_instead is set. 837 */ 838 int 839 thread_suspend_check(int return_instead) 840 { 841 struct thread *td; 842 struct proc *p; 843 int wakeup_swapper; 844 845 td = curthread; 846 p = td->td_proc; 847 mtx_assert(&Giant, MA_NOTOWNED); 848 PROC_LOCK_ASSERT(p, MA_OWNED); 849 while (thread_suspend_check_needed()) { 850 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 851 KASSERT(p->p_singlethread != NULL, 852 ("singlethread not set")); 853 /* 854 * The only suspension in action is a 855 * single-threading. Single threader need not stop. 856 * XXX Should be safe to access unlocked 857 * as it can only be set to be true by us. 858 */ 859 if (p->p_singlethread == td) 860 return (0); /* Exempt from stopping. */ 861 } 862 if ((p->p_flag & P_SINGLE_EXIT) && return_instead) 863 return (EINTR); 864 865 /* Should we goto user boundary if we didn't come from there? */ 866 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE && 867 (p->p_flag & P_SINGLE_BOUNDARY) && return_instead) 868 return (ERESTART); 869 870 /* 871 * Ignore suspend requests if they are deferred. 872 */ 873 if ((td->td_flags & TDF_SBDRY) != 0) { 874 KASSERT(return_instead, 875 ("TDF_SBDRY set for unsafe thread_suspend_check")); 876 return (0); 877 } 878 879 /* 880 * If the process is waiting for us to exit, 881 * this thread should just suicide. 882 * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE. 883 */ 884 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) { 885 PROC_UNLOCK(p); 886 tidhash_remove(td); 887 PROC_LOCK(p); 888 tdsigcleanup(td); 889 umtx_thread_exit(td); 890 PROC_SLOCK(p); 891 thread_stopped(p); 892 thread_exit(); 893 } 894 895 PROC_SLOCK(p); 896 thread_stopped(p); 897 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 898 if (p->p_numthreads == p->p_suspcount + 1) { 899 thread_lock(p->p_singlethread); 900 wakeup_swapper = thread_unsuspend_one( 901 p->p_singlethread, p, false); 902 thread_unlock(p->p_singlethread); 903 if (wakeup_swapper) 904 kick_proc0(); 905 } 906 } 907 PROC_UNLOCK(p); 908 thread_lock(td); 909 /* 910 * When a thread suspends, it just 911 * gets taken off all queues. 912 */ 913 thread_suspend_one(td); 914 if (return_instead == 0) { 915 p->p_boundary_count++; 916 td->td_flags |= TDF_BOUNDARY; 917 } 918 PROC_SUNLOCK(p); 919 mi_switch(SW_INVOL | SWT_SUSPEND, NULL); 920 thread_unlock(td); 921 PROC_LOCK(p); 922 } 923 return (0); 924 } 925 926 void 927 thread_suspend_switch(struct thread *td, struct proc *p) 928 { 929 930 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended")); 931 PROC_LOCK_ASSERT(p, MA_OWNED); 932 PROC_SLOCK_ASSERT(p, MA_OWNED); 933 /* 934 * We implement thread_suspend_one in stages here to avoid 935 * dropping the proc lock while the thread lock is owned. 936 */ 937 if (p == td->td_proc) { 938 thread_stopped(p); 939 p->p_suspcount++; 940 } 941 PROC_UNLOCK(p); 942 thread_lock(td); 943 td->td_flags &= ~TDF_NEEDSUSPCHK; 944 TD_SET_SUSPENDED(td); 945 sched_sleep(td, 0); 946 PROC_SUNLOCK(p); 947 DROP_GIANT(); 948 mi_switch(SW_VOL | SWT_SUSPEND, NULL); 949 thread_unlock(td); 950 PICKUP_GIANT(); 951 PROC_LOCK(p); 952 PROC_SLOCK(p); 953 } 954 955 void 956 thread_suspend_one(struct thread *td) 957 { 958 struct proc *p; 959 960 p = td->td_proc; 961 PROC_SLOCK_ASSERT(p, MA_OWNED); 962 THREAD_LOCK_ASSERT(td, MA_OWNED); 963 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended")); 964 p->p_suspcount++; 965 td->td_flags &= ~TDF_NEEDSUSPCHK; 966 TD_SET_SUSPENDED(td); 967 sched_sleep(td, 0); 968 } 969 970 static int 971 thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary) 972 { 973 974 THREAD_LOCK_ASSERT(td, MA_OWNED); 975 KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended")); 976 TD_CLR_SUSPENDED(td); 977 td->td_flags &= ~TDF_ALLPROCSUSP; 978 if (td->td_proc == p) { 979 PROC_SLOCK_ASSERT(p, MA_OWNED); 980 p->p_suspcount--; 981 if (boundary && (td->td_flags & TDF_BOUNDARY) != 0) { 982 td->td_flags &= ~TDF_BOUNDARY; 983 p->p_boundary_count--; 984 } 985 } 986 return (setrunnable(td)); 987 } 988 989 /* 990 * Allow all threads blocked by single threading to continue running. 991 */ 992 void 993 thread_unsuspend(struct proc *p) 994 { 995 struct thread *td; 996 int wakeup_swapper; 997 998 PROC_LOCK_ASSERT(p, MA_OWNED); 999 PROC_SLOCK_ASSERT(p, MA_OWNED); 1000 wakeup_swapper = 0; 1001 if (!P_SHOULDSTOP(p)) { 1002 FOREACH_THREAD_IN_PROC(p, td) { 1003 thread_lock(td); 1004 if (TD_IS_SUSPENDED(td)) { 1005 wakeup_swapper |= thread_unsuspend_one(td, p, 1006 true); 1007 } 1008 thread_unlock(td); 1009 } 1010 } else if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE && 1011 p->p_numthreads == p->p_suspcount) { 1012 /* 1013 * Stopping everything also did the job for the single 1014 * threading request. Now we've downgraded to single-threaded, 1015 * let it continue. 1016 */ 1017 if (p->p_singlethread->td_proc == p) { 1018 thread_lock(p->p_singlethread); 1019 wakeup_swapper = thread_unsuspend_one( 1020 p->p_singlethread, p, false); 1021 thread_unlock(p->p_singlethread); 1022 } 1023 } 1024 if (wakeup_swapper) 1025 kick_proc0(); 1026 } 1027 1028 /* 1029 * End the single threading mode.. 1030 */ 1031 void 1032 thread_single_end(struct proc *p, int mode) 1033 { 1034 struct thread *td; 1035 int wakeup_swapper; 1036 1037 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY || 1038 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT, 1039 ("invalid mode %d", mode)); 1040 PROC_LOCK_ASSERT(p, MA_OWNED); 1041 KASSERT((mode == SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) != 0) || 1042 (mode != SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) == 0), 1043 ("mode %d does not match P_TOTAL_STOP", mode)); 1044 KASSERT(mode == SINGLE_ALLPROC || p->p_singlethread == curthread, 1045 ("thread_single_end from other thread %p %p", 1046 curthread, p->p_singlethread)); 1047 KASSERT(mode != SINGLE_BOUNDARY || 1048 (p->p_flag & P_SINGLE_BOUNDARY) != 0, 1049 ("mis-matched SINGLE_BOUNDARY flags %x", p->p_flag)); 1050 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY | 1051 P_TOTAL_STOP); 1052 PROC_SLOCK(p); 1053 p->p_singlethread = NULL; 1054 wakeup_swapper = 0; 1055 /* 1056 * If there are other threads they may now run, 1057 * unless of course there is a blanket 'stop order' 1058 * on the process. The single threader must be allowed 1059 * to continue however as this is a bad place to stop. 1060 */ 1061 if (p->p_numthreads != remain_for_mode(mode) && !P_SHOULDSTOP(p)) { 1062 FOREACH_THREAD_IN_PROC(p, td) { 1063 thread_lock(td); 1064 if (TD_IS_SUSPENDED(td)) { 1065 wakeup_swapper |= thread_unsuspend_one(td, p, 1066 mode == SINGLE_BOUNDARY); 1067 } 1068 thread_unlock(td); 1069 } 1070 } 1071 KASSERT(mode != SINGLE_BOUNDARY || p->p_boundary_count == 0, 1072 ("inconsistent boundary count %d", p->p_boundary_count)); 1073 PROC_SUNLOCK(p); 1074 if (wakeup_swapper) 1075 kick_proc0(); 1076 } 1077 1078 struct thread * 1079 thread_find(struct proc *p, lwpid_t tid) 1080 { 1081 struct thread *td; 1082 1083 PROC_LOCK_ASSERT(p, MA_OWNED); 1084 FOREACH_THREAD_IN_PROC(p, td) { 1085 if (td->td_tid == tid) 1086 break; 1087 } 1088 return (td); 1089 } 1090 1091 /* Locate a thread by number; return with proc lock held. */ 1092 struct thread * 1093 tdfind(lwpid_t tid, pid_t pid) 1094 { 1095 #define RUN_THRESH 16 1096 struct thread *td; 1097 int run = 0; 1098 1099 rw_rlock(&tidhash_lock); 1100 LIST_FOREACH(td, TIDHASH(tid), td_hash) { 1101 if (td->td_tid == tid) { 1102 if (pid != -1 && td->td_proc->p_pid != pid) { 1103 td = NULL; 1104 break; 1105 } 1106 PROC_LOCK(td->td_proc); 1107 if (td->td_proc->p_state == PRS_NEW) { 1108 PROC_UNLOCK(td->td_proc); 1109 td = NULL; 1110 break; 1111 } 1112 if (run > RUN_THRESH) { 1113 if (rw_try_upgrade(&tidhash_lock)) { 1114 LIST_REMOVE(td, td_hash); 1115 LIST_INSERT_HEAD(TIDHASH(td->td_tid), 1116 td, td_hash); 1117 rw_wunlock(&tidhash_lock); 1118 return (td); 1119 } 1120 } 1121 break; 1122 } 1123 run++; 1124 } 1125 rw_runlock(&tidhash_lock); 1126 return (td); 1127 } 1128 1129 void 1130 tidhash_add(struct thread *td) 1131 { 1132 rw_wlock(&tidhash_lock); 1133 LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash); 1134 rw_wunlock(&tidhash_lock); 1135 } 1136 1137 void 1138 tidhash_remove(struct thread *td) 1139 { 1140 rw_wlock(&tidhash_lock); 1141 LIST_REMOVE(td, td_hash); 1142 rw_wunlock(&tidhash_lock); 1143 } 1144