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_kdtrace.h" 31#include "opt_hwpmc_hooks.h" 32 33#include <sys/cdefs.h> 34__FBSDID("$FreeBSD: releng/10.2/sys/kern/kern_thread.c 283765 2015-05-30 08:54:42Z kib $"); 35 36#include <sys/param.h> 37#include <sys/systm.h> 38#include <sys/kernel.h> 39#include <sys/lock.h> 40#include <sys/mutex.h> 41#include <sys/proc.h> 42#include <sys/rangelock.h> 43#include <sys/resourcevar.h> 44#include <sys/sdt.h> 45#include <sys/smp.h> 46#include <sys/sched.h> 47#include <sys/sleepqueue.h> 48#include <sys/selinfo.h> 49#include <sys/turnstile.h> 50#include <sys/ktr.h> 51#include <sys/rwlock.h> 52#include <sys/umtx.h> 53#include <sys/cpuset.h> 54#ifdef HWPMC_HOOKS 55#include <sys/pmckern.h> 56#endif 57 58#include <security/audit/audit.h> 59 60#include <vm/vm.h> 61#include <vm/vm_extern.h> 62#include <vm/uma.h> 63#include <sys/eventhandler.h> 64 65SDT_PROVIDER_DECLARE(proc); 66SDT_PROBE_DEFINE(proc, , , lwp__exit); 67 68/* 69 * thread related storage. 70 */ 71static uma_zone_t thread_zone; 72 73TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads); 74static struct mtx zombie_lock; 75MTX_SYSINIT(zombie_lock, &zombie_lock, "zombie lock", MTX_SPIN); 76 77static void thread_zombie(struct thread *); 78static int thread_unsuspend_one(struct thread *td, struct proc *p, 79 bool boundary); 80 81#define TID_BUFFER_SIZE 1024 82 83struct mtx tid_lock; 84static struct unrhdr *tid_unrhdr; 85static lwpid_t tid_buffer[TID_BUFFER_SIZE]; 86static int tid_head, tid_tail; 87static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash"); 88 89struct tidhashhead *tidhashtbl; 90u_long tidhash; 91struct rwlock tidhash_lock; 92 93static lwpid_t 94tid_alloc(void) 95{ 96 lwpid_t tid; 97 98 tid = alloc_unr(tid_unrhdr); 99 if (tid != -1) 100 return (tid); 101 mtx_lock(&tid_lock); 102 if (tid_head == tid_tail) { 103 mtx_unlock(&tid_lock); 104 return (-1); 105 } 106 tid = tid_buffer[tid_head]; 107 tid_head = (tid_head + 1) % TID_BUFFER_SIZE; 108 mtx_unlock(&tid_lock); 109 return (tid); 110} 111 112static void 113tid_free(lwpid_t tid) 114{ 115 lwpid_t tmp_tid = -1; 116 117 mtx_lock(&tid_lock); 118 if ((tid_tail + 1) % TID_BUFFER_SIZE == tid_head) { 119 tmp_tid = tid_buffer[tid_head]; 120 tid_head = (tid_head + 1) % TID_BUFFER_SIZE; 121 } 122 tid_buffer[tid_tail] = tid; 123 tid_tail = (tid_tail + 1) % TID_BUFFER_SIZE; 124 mtx_unlock(&tid_lock); 125 if (tmp_tid != -1) 126 free_unr(tid_unrhdr, tmp_tid); 127} 128 129/* 130 * Prepare a thread for use. 131 */ 132static int 133thread_ctor(void *mem, int size, void *arg, int flags) 134{ 135 struct thread *td; 136 137 td = (struct thread *)mem; 138 td->td_state = TDS_INACTIVE; 139 td->td_oncpu = NOCPU; 140 141 td->td_tid = tid_alloc(); 142 143 /* 144 * Note that td_critnest begins life as 1 because the thread is not 145 * running and is thereby implicitly waiting to be on the receiving 146 * end of a context switch. 147 */ 148 td->td_critnest = 1; 149 td->td_lend_user_pri = PRI_MAX; 150 EVENTHANDLER_INVOKE(thread_ctor, td); 151#ifdef AUDIT 152 audit_thread_alloc(td); 153#endif 154 umtx_thread_alloc(td); 155 return (0); 156} 157 158/* 159 * Reclaim a thread after use. 160 */ 161static void 162thread_dtor(void *mem, int size, void *arg) 163{ 164 struct thread *td; 165 166 td = (struct thread *)mem; 167 168#ifdef INVARIANTS 169 /* Verify that this thread is in a safe state to free. */ 170 switch (td->td_state) { 171 case TDS_INHIBITED: 172 case TDS_RUNNING: 173 case TDS_CAN_RUN: 174 case TDS_RUNQ: 175 /* 176 * We must never unlink a thread that is in one of 177 * these states, because it is currently active. 178 */ 179 panic("bad state for thread unlinking"); 180 /* NOTREACHED */ 181 case TDS_INACTIVE: 182 break; 183 default: 184 panic("bad thread state"); 185 /* NOTREACHED */ 186 } 187#endif 188#ifdef AUDIT 189 audit_thread_free(td); 190#endif 191 /* Free all OSD associated to this thread. */ 192 osd_thread_exit(td); 193 194 EVENTHANDLER_INVOKE(thread_dtor, td); 195 tid_free(td->td_tid); 196} 197 198/* 199 * Initialize type-stable parts of a thread (when newly created). 200 */ 201static int 202thread_init(void *mem, int size, int flags) 203{ 204 struct thread *td; 205 206 td = (struct thread *)mem; 207 208 td->td_sleepqueue = sleepq_alloc(); 209 td->td_turnstile = turnstile_alloc(); 210 td->td_rlqe = NULL; 211 EVENTHANDLER_INVOKE(thread_init, td); 212 td->td_sched = (struct td_sched *)&td[1]; 213 umtx_thread_init(td); 214 td->td_kstack = 0; 215 td->td_sel = NULL; 216 return (0); 217} 218 219/* 220 * Tear down type-stable parts of a thread (just before being discarded). 221 */ 222static void 223thread_fini(void *mem, int size) 224{ 225 struct thread *td; 226 227 td = (struct thread *)mem; 228 EVENTHANDLER_INVOKE(thread_fini, td); 229 rlqentry_free(td->td_rlqe); 230 turnstile_free(td->td_turnstile); 231 sleepq_free(td->td_sleepqueue); 232 umtx_thread_fini(td); 233 seltdfini(td); 234} 235 236/* 237 * For a newly created process, 238 * link up all the structures and its initial threads etc. 239 * called from: 240 * {arch}/{arch}/machdep.c ia64_init(), init386() etc. 241 * proc_dtor() (should go away) 242 * proc_init() 243 */ 244void 245proc_linkup0(struct proc *p, struct thread *td) 246{ 247 TAILQ_INIT(&p->p_threads); /* all threads in proc */ 248 proc_linkup(p, td); 249} 250 251void 252proc_linkup(struct proc *p, struct thread *td) 253{ 254 255 sigqueue_init(&p->p_sigqueue, p); 256 p->p_ksi = ksiginfo_alloc(1); 257 if (p->p_ksi != NULL) { 258 /* XXX p_ksi may be null if ksiginfo zone is not ready */ 259 p->p_ksi->ksi_flags = KSI_EXT | KSI_INS; 260 } 261 LIST_INIT(&p->p_mqnotifier); 262 p->p_numthreads = 0; 263 thread_link(td, p); 264} 265 266/* 267 * Initialize global thread allocation resources. 268 */ 269void 270threadinit(void) 271{ 272 273 mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF); 274 275 /* 276 * pid_max cannot be greater than PID_MAX. 277 * leave one number for thread0. 278 */ 279 tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock); 280 281 thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(), 282 thread_ctor, thread_dtor, thread_init, thread_fini, 283 16 - 1, 0); 284 tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash); 285 rw_init(&tidhash_lock, "tidhash"); 286} 287 288/* 289 * Place an unused thread on the zombie list. 290 * Use the slpq as that must be unused by now. 291 */ 292void 293thread_zombie(struct thread *td) 294{ 295 mtx_lock_spin(&zombie_lock); 296 TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq); 297 mtx_unlock_spin(&zombie_lock); 298} 299 300/* 301 * Release a thread that has exited after cpu_throw(). 302 */ 303void 304thread_stash(struct thread *td) 305{ 306 atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1); 307 thread_zombie(td); 308} 309 310/* 311 * Reap zombie resources. 312 */ 313void 314thread_reap(void) 315{ 316 struct thread *td_first, *td_next; 317 318 /* 319 * Don't even bother to lock if none at this instant, 320 * we really don't care about the next instant.. 321 */ 322 if (!TAILQ_EMPTY(&zombie_threads)) { 323 mtx_lock_spin(&zombie_lock); 324 td_first = TAILQ_FIRST(&zombie_threads); 325 if (td_first) 326 TAILQ_INIT(&zombie_threads); 327 mtx_unlock_spin(&zombie_lock); 328 while (td_first) { 329 td_next = TAILQ_NEXT(td_first, td_slpq); 330 if (td_first->td_ucred) 331 crfree(td_first->td_ucred); 332 thread_free(td_first); 333 td_first = td_next; 334 } 335 } 336} 337 338/* 339 * Allocate a thread. 340 */ 341struct thread * 342thread_alloc(int pages) 343{ 344 struct thread *td; 345 346 thread_reap(); /* check if any zombies to get */ 347 348 td = (struct thread *)uma_zalloc(thread_zone, M_WAITOK); 349 KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack")); 350 if (!vm_thread_new(td, pages)) { 351 uma_zfree(thread_zone, td); 352 return (NULL); 353 } 354 cpu_thread_alloc(td); 355 return (td); 356} 357 358int 359thread_alloc_stack(struct thread *td, int pages) 360{ 361 362 KASSERT(td->td_kstack == 0, 363 ("thread_alloc_stack called on a thread with kstack")); 364 if (!vm_thread_new(td, pages)) 365 return (0); 366 cpu_thread_alloc(td); 367 return (1); 368} 369 370/* 371 * Deallocate a thread. 372 */ 373void 374thread_free(struct thread *td) 375{ 376 377 lock_profile_thread_exit(td); 378 if (td->td_cpuset) 379 cpuset_rel(td->td_cpuset); 380 td->td_cpuset = NULL; 381 cpu_thread_free(td); 382 if (td->td_kstack != 0) 383 vm_thread_dispose(td); 384 uma_zfree(thread_zone, td); 385} 386 387/* 388 * Discard the current thread and exit from its context. 389 * Always called with scheduler locked. 390 * 391 * Because we can't free a thread while we're operating under its context, 392 * push the current thread into our CPU's deadthread holder. This means 393 * we needn't worry about someone else grabbing our context before we 394 * do a cpu_throw(). 395 */ 396void 397thread_exit(void) 398{ 399 uint64_t runtime, new_switchtime; 400 struct thread *td; 401 struct thread *td2; 402 struct proc *p; 403 int wakeup_swapper; 404 405 td = curthread; 406 p = td->td_proc; 407 408 PROC_SLOCK_ASSERT(p, MA_OWNED); 409 mtx_assert(&Giant, MA_NOTOWNED); 410 411 PROC_LOCK_ASSERT(p, MA_OWNED); 412 KASSERT(p != NULL, ("thread exiting without a process")); 413 CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td, 414 (long)p->p_pid, td->td_name); 415 KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending")); 416 417#ifdef AUDIT 418 AUDIT_SYSCALL_EXIT(0, td); 419#endif 420 /* 421 * drop FPU & debug register state storage, or any other 422 * architecture specific resources that 423 * would not be on a new untouched process. 424 */ 425 cpu_thread_exit(td); /* XXXSMP */ 426 427 /* 428 * The last thread is left attached to the process 429 * So that the whole bundle gets recycled. Skip 430 * all this stuff if we never had threads. 431 * EXIT clears all sign of other threads when 432 * it goes to single threading, so the last thread always 433 * takes the short path. 434 */ 435 if (p->p_flag & P_HADTHREADS) { 436 if (p->p_numthreads > 1) { 437 atomic_add_int(&td->td_proc->p_exitthreads, 1); 438 thread_unlink(td); 439 td2 = FIRST_THREAD_IN_PROC(p); 440 sched_exit_thread(td2, td); 441 442 /* 443 * The test below is NOT true if we are the 444 * sole exiting thread. P_STOPPED_SINGLE is unset 445 * in exit1() after it is the only survivor. 446 */ 447 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 448 if (p->p_numthreads == p->p_suspcount) { 449 thread_lock(p->p_singlethread); 450 wakeup_swapper = thread_unsuspend_one( 451 p->p_singlethread, p, false); 452 thread_unlock(p->p_singlethread); 453 if (wakeup_swapper) 454 kick_proc0(); 455 } 456 } 457 458 PCPU_SET(deadthread, td); 459 } else { 460 /* 461 * The last thread is exiting.. but not through exit() 462 */ 463 panic ("thread_exit: Last thread exiting on its own"); 464 } 465 } 466#ifdef HWPMC_HOOKS 467 /* 468 * If this thread is part of a process that is being tracked by hwpmc(4), 469 * inform the module of the thread's impending exit. 470 */ 471 if (PMC_PROC_IS_USING_PMCS(td->td_proc)) 472 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT); 473#endif 474 PROC_UNLOCK(p); 475 476 /* Do the same timestamp bookkeeping that mi_switch() would do. */ 477 new_switchtime = cpu_ticks(); 478 runtime = new_switchtime - PCPU_GET(switchtime); 479 td->td_runtime += runtime; 480 td->td_incruntime += runtime; 481 PCPU_SET(switchtime, new_switchtime); 482 PCPU_SET(switchticks, ticks); 483 PCPU_INC(cnt.v_swtch); 484 485 /* Save our resource usage in our process. */ 486 td->td_ru.ru_nvcsw++; 487 ruxagg(p, td); 488 rucollect(&p->p_ru, &td->td_ru); 489 490 thread_lock(td); 491 PROC_SUNLOCK(p); 492 td->td_state = TDS_INACTIVE; 493#ifdef WITNESS 494 witness_thread_exit(td); 495#endif 496 CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td); 497 sched_throw(td); 498 panic("I'm a teapot!"); 499 /* NOTREACHED */ 500} 501 502/* 503 * Do any thread specific cleanups that may be needed in wait() 504 * called with Giant, proc and schedlock not held. 505 */ 506void 507thread_wait(struct proc *p) 508{ 509 struct thread *td; 510 511 mtx_assert(&Giant, MA_NOTOWNED); 512 KASSERT(p->p_numthreads == 1, ("multiple threads in thread_wait()")); 513 KASSERT(p->p_exitthreads == 0, ("p_exitthreads leaking")); 514 td = FIRST_THREAD_IN_PROC(p); 515 /* Lock the last thread so we spin until it exits cpu_throw(). */ 516 thread_lock(td); 517 thread_unlock(td); 518 lock_profile_thread_exit(td); 519 cpuset_rel(td->td_cpuset); 520 td->td_cpuset = NULL; 521 cpu_thread_clean(td); 522 crfree(td->td_ucred); 523 thread_reap(); /* check for zombie threads etc. */ 524} 525 526/* 527 * Link a thread to a process. 528 * set up anything that needs to be initialized for it to 529 * be used by the process. 530 */ 531void 532thread_link(struct thread *td, struct proc *p) 533{ 534 535 /* 536 * XXX This can't be enabled because it's called for proc0 before 537 * its lock has been created. 538 * PROC_LOCK_ASSERT(p, MA_OWNED); 539 */ 540 td->td_state = TDS_INACTIVE; 541 td->td_proc = p; 542 td->td_flags = TDF_INMEM; 543 544 LIST_INIT(&td->td_contested); 545 LIST_INIT(&td->td_lprof[0]); 546 LIST_INIT(&td->td_lprof[1]); 547 sigqueue_init(&td->td_sigqueue, p); 548 callout_init(&td->td_slpcallout, CALLOUT_MPSAFE); 549 TAILQ_INSERT_HEAD(&p->p_threads, td, td_plist); 550 p->p_numthreads++; 551} 552 553/* 554 * Called from: 555 * thread_exit() 556 */ 557void 558thread_unlink(struct thread *td) 559{ 560 struct proc *p = td->td_proc; 561 562 PROC_LOCK_ASSERT(p, MA_OWNED); 563 TAILQ_REMOVE(&p->p_threads, td, td_plist); 564 p->p_numthreads--; 565 /* could clear a few other things here */ 566 /* Must NOT clear links to proc! */ 567} 568 569static int 570calc_remaining(struct proc *p, int mode) 571{ 572 int remaining; 573 574 PROC_LOCK_ASSERT(p, MA_OWNED); 575 PROC_SLOCK_ASSERT(p, MA_OWNED); 576 if (mode == SINGLE_EXIT) 577 remaining = p->p_numthreads; 578 else if (mode == SINGLE_BOUNDARY) 579 remaining = p->p_numthreads - p->p_boundary_count; 580 else if (mode == SINGLE_NO_EXIT || mode == SINGLE_ALLPROC) 581 remaining = p->p_numthreads - p->p_suspcount; 582 else 583 panic("calc_remaining: wrong mode %d", mode); 584 return (remaining); 585} 586 587static int 588remain_for_mode(int mode) 589{ 590 591 return (mode == SINGLE_ALLPROC ? 0 : 1); 592} 593 594static int 595weed_inhib(int mode, struct thread *td2, struct proc *p) 596{ 597 int wakeup_swapper; 598 599 PROC_LOCK_ASSERT(p, MA_OWNED); 600 PROC_SLOCK_ASSERT(p, MA_OWNED); 601 THREAD_LOCK_ASSERT(td2, MA_OWNED); 602 603 wakeup_swapper = 0; 604 switch (mode) { 605 case SINGLE_EXIT: 606 if (TD_IS_SUSPENDED(td2)) 607 wakeup_swapper |= thread_unsuspend_one(td2, p, true); 608 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) 609 wakeup_swapper |= sleepq_abort(td2, EINTR); 610 break; 611 case SINGLE_BOUNDARY: 612 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & TDF_BOUNDARY) == 0) 613 wakeup_swapper |= thread_unsuspend_one(td2, p, false); 614 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) 615 wakeup_swapper |= sleepq_abort(td2, ERESTART); 616 break; 617 case SINGLE_NO_EXIT: 618 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & TDF_BOUNDARY) == 0) 619 wakeup_swapper |= thread_unsuspend_one(td2, p, false); 620 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) 621 wakeup_swapper |= sleepq_abort(td2, ERESTART); 622 break; 623 case SINGLE_ALLPROC: 624 /* 625 * ALLPROC suspend tries to avoid spurious EINTR for 626 * threads sleeping interruptable, by suspending the 627 * thread directly, similarly to sig_suspend_threads(). 628 * Since such sleep is not performed at the user 629 * boundary, TDF_BOUNDARY flag is not set, and TDF_ALLPROCSUSP 630 * is used to avoid immediate un-suspend. 631 */ 632 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & (TDF_BOUNDARY | 633 TDF_ALLPROCSUSP)) == 0) 634 wakeup_swapper |= thread_unsuspend_one(td2, p, false); 635 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) { 636 if ((td2->td_flags & TDF_SBDRY) == 0) { 637 thread_suspend_one(td2); 638 td2->td_flags |= TDF_ALLPROCSUSP; 639 } else { 640 wakeup_swapper |= sleepq_abort(td2, ERESTART); 641 } 642 } 643 break; 644 } 645 return (wakeup_swapper); 646} 647 648/* 649 * Enforce single-threading. 650 * 651 * Returns 1 if the caller must abort (another thread is waiting to 652 * exit the process or similar). Process is locked! 653 * Returns 0 when you are successfully the only thread running. 654 * A process has successfully single threaded in the suspend mode when 655 * There are no threads in user mode. Threads in the kernel must be 656 * allowed to continue until they get to the user boundary. They may even 657 * copy out their return values and data before suspending. They may however be 658 * accelerated in reaching the user boundary as we will wake up 659 * any sleeping threads that are interruptable. (PCATCH). 660 */ 661int 662thread_single(struct proc *p, int mode) 663{ 664 struct thread *td; 665 struct thread *td2; 666 int remaining, wakeup_swapper; 667 668 td = curthread; 669 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY || 670 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT, 671 ("invalid mode %d", mode)); 672 /* 673 * If allowing non-ALLPROC singlethreading for non-curproc 674 * callers, calc_remaining() and remain_for_mode() should be 675 * adjusted to also account for td->td_proc != p. For now 676 * this is not implemented because it is not used. 677 */ 678 KASSERT((mode == SINGLE_ALLPROC && td->td_proc != p) || 679 (mode != SINGLE_ALLPROC && td->td_proc == p), 680 ("mode %d proc %p curproc %p", mode, p, td->td_proc)); 681 mtx_assert(&Giant, MA_NOTOWNED); 682 PROC_LOCK_ASSERT(p, MA_OWNED); 683 684 if ((p->p_flag & P_HADTHREADS) == 0 && mode != SINGLE_ALLPROC) 685 return (0); 686 687 /* Is someone already single threading? */ 688 if (p->p_singlethread != NULL && p->p_singlethread != td) 689 return (1); 690 691 if (mode == SINGLE_EXIT) { 692 p->p_flag |= P_SINGLE_EXIT; 693 p->p_flag &= ~P_SINGLE_BOUNDARY; 694 } else { 695 p->p_flag &= ~P_SINGLE_EXIT; 696 if (mode == SINGLE_BOUNDARY) 697 p->p_flag |= P_SINGLE_BOUNDARY; 698 else 699 p->p_flag &= ~P_SINGLE_BOUNDARY; 700 } 701 if (mode == SINGLE_ALLPROC) 702 p->p_flag |= P_TOTAL_STOP; 703 p->p_flag |= P_STOPPED_SINGLE; 704 PROC_SLOCK(p); 705 p->p_singlethread = td; 706 remaining = calc_remaining(p, mode); 707 while (remaining != remain_for_mode(mode)) { 708 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE) 709 goto stopme; 710 wakeup_swapper = 0; 711 FOREACH_THREAD_IN_PROC(p, td2) { 712 if (td2 == td) 713 continue; 714 thread_lock(td2); 715 td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK; 716 if (TD_IS_INHIBITED(td2)) { 717 wakeup_swapper |= weed_inhib(mode, td2, p); 718#ifdef SMP 719 } else if (TD_IS_RUNNING(td2) && td != td2) { 720 forward_signal(td2); 721#endif 722 } 723 thread_unlock(td2); 724 } 725 if (wakeup_swapper) 726 kick_proc0(); 727 remaining = calc_remaining(p, mode); 728 729 /* 730 * Maybe we suspended some threads.. was it enough? 731 */ 732 if (remaining == remain_for_mode(mode)) 733 break; 734 735stopme: 736 /* 737 * Wake us up when everyone else has suspended. 738 * In the mean time we suspend as well. 739 */ 740 thread_suspend_switch(td, p); 741 remaining = calc_remaining(p, mode); 742 } 743 if (mode == SINGLE_EXIT) { 744 /* 745 * Convert the process to an unthreaded process. The 746 * SINGLE_EXIT is called by exit1() or execve(), in 747 * both cases other threads must be retired. 748 */ 749 KASSERT(p->p_numthreads == 1, ("Unthreading with >1 threads")); 750 p->p_singlethread = NULL; 751 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_HADTHREADS); 752 753 /* 754 * Wait for any remaining threads to exit cpu_throw(). 755 */ 756 while (p->p_exitthreads != 0) { 757 PROC_SUNLOCK(p); 758 PROC_UNLOCK(p); 759 sched_relinquish(td); 760 PROC_LOCK(p); 761 PROC_SLOCK(p); 762 } 763 } else if (mode == SINGLE_BOUNDARY) { 764 /* 765 * Wait until all suspended threads are removed from 766 * the processors. The thread_suspend_check() 767 * increments p_boundary_count while it is still 768 * running, which makes it possible for the execve() 769 * to destroy vmspace while our other threads are 770 * still using the address space. 771 * 772 * We lock the thread, which is only allowed to 773 * succeed after context switch code finished using 774 * the address space. 775 */ 776 FOREACH_THREAD_IN_PROC(p, td2) { 777 if (td2 == td) 778 continue; 779 thread_lock(td2); 780 KASSERT((td2->td_flags & TDF_BOUNDARY) != 0, 781 ("td %p not on boundary", td2)); 782 KASSERT(TD_IS_SUSPENDED(td2), 783 ("td %p is not suspended", td2)); 784 thread_unlock(td2); 785 } 786 } 787 PROC_SUNLOCK(p); 788 return (0); 789} 790 791bool 792thread_suspend_check_needed(void) 793{ 794 struct proc *p; 795 struct thread *td; 796 797 td = curthread; 798 p = td->td_proc; 799 PROC_LOCK_ASSERT(p, MA_OWNED); 800 return (P_SHOULDSTOP(p) || ((p->p_flag & P_TRACED) != 0 && 801 (td->td_dbgflags & TDB_SUSPEND) != 0)); 802} 803 804/* 805 * Called in from locations that can safely check to see 806 * whether we have to suspend or at least throttle for a 807 * single-thread event (e.g. fork). 808 * 809 * Such locations include userret(). 810 * If the "return_instead" argument is non zero, the thread must be able to 811 * accept 0 (caller may continue), or 1 (caller must abort) as a result. 812 * 813 * The 'return_instead' argument tells the function if it may do a 814 * thread_exit() or suspend, or whether the caller must abort and back 815 * out instead. 816 * 817 * If the thread that set the single_threading request has set the 818 * P_SINGLE_EXIT bit in the process flags then this call will never return 819 * if 'return_instead' is false, but will exit. 820 * 821 * P_SINGLE_EXIT | return_instead == 0| return_instead != 0 822 *---------------+--------------------+--------------------- 823 * 0 | returns 0 | returns 0 or 1 824 * | when ST ends | immediately 825 *---------------+--------------------+--------------------- 826 * 1 | thread exits | returns 1 827 * | | immediately 828 * 0 = thread_exit() or suspension ok, 829 * other = return error instead of stopping the thread. 830 * 831 * While a full suspension is under effect, even a single threading 832 * thread would be suspended if it made this call (but it shouldn't). 833 * This call should only be made from places where 834 * thread_exit() would be safe as that may be the outcome unless 835 * return_instead is set. 836 */ 837int 838thread_suspend_check(int return_instead) 839{ 840 struct thread *td; 841 struct proc *p; 842 int wakeup_swapper; 843 844 td = curthread; 845 p = td->td_proc; 846 mtx_assert(&Giant, MA_NOTOWNED); 847 PROC_LOCK_ASSERT(p, MA_OWNED); 848 while (thread_suspend_check_needed()) { 849 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 850 KASSERT(p->p_singlethread != NULL, 851 ("singlethread not set")); 852 /* 853 * The only suspension in action is a 854 * single-threading. Single threader need not stop. 855 * XXX Should be safe to access unlocked 856 * as it can only be set to be true by us. 857 */ 858 if (p->p_singlethread == td) 859 return (0); /* Exempt from stopping. */ 860 } 861 if ((p->p_flag & P_SINGLE_EXIT) && return_instead) 862 return (EINTR); 863 864 /* Should we goto user boundary if we didn't come from there? */ 865 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE && 866 (p->p_flag & P_SINGLE_BOUNDARY) && return_instead) 867 return (ERESTART); 868 869 /* 870 * Ignore suspend requests if they are deferred. 871 */ 872 if ((td->td_flags & TDF_SBDRY) != 0) { 873 KASSERT(return_instead, 874 ("TDF_SBDRY set for unsafe thread_suspend_check")); 875 return (0); 876 } 877 878 /* 879 * If the process is waiting for us to exit, 880 * this thread should just suicide. 881 * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE. 882 */ 883 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) { 884 PROC_UNLOCK(p); 885 tidhash_remove(td); 886 PROC_LOCK(p); 887 tdsigcleanup(td); 888 umtx_thread_exit(td); 889 PROC_SLOCK(p); 890 thread_stopped(p); 891 thread_exit(); 892 } 893 894 PROC_SLOCK(p); 895 thread_stopped(p); 896 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 897 if (p->p_numthreads == p->p_suspcount + 1) { 898 thread_lock(p->p_singlethread); 899 wakeup_swapper = thread_unsuspend_one( 900 p->p_singlethread, p, false); 901 thread_unlock(p->p_singlethread); 902 if (wakeup_swapper) 903 kick_proc0(); 904 } 905 } 906 PROC_UNLOCK(p); 907 thread_lock(td); 908 /* 909 * When a thread suspends, it just 910 * gets taken off all queues. 911 */ 912 thread_suspend_one(td); 913 if (return_instead == 0) { 914 p->p_boundary_count++; 915 td->td_flags |= TDF_BOUNDARY; 916 } 917 PROC_SUNLOCK(p); 918 mi_switch(SW_INVOL | SWT_SUSPEND, NULL); 919 thread_unlock(td); 920 PROC_LOCK(p); 921 } 922 return (0); 923} 924 925void 926thread_suspend_switch(struct thread *td, struct proc *p) 927{ 928 929 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended")); 930 PROC_LOCK_ASSERT(p, MA_OWNED); 931 PROC_SLOCK_ASSERT(p, MA_OWNED); 932 /* 933 * We implement thread_suspend_one in stages here to avoid 934 * dropping the proc lock while the thread lock is owned. 935 */ 936 if (p == td->td_proc) { 937 thread_stopped(p); 938 p->p_suspcount++; 939 } 940 PROC_UNLOCK(p); 941 thread_lock(td); 942 td->td_flags &= ~TDF_NEEDSUSPCHK; 943 TD_SET_SUSPENDED(td); 944 sched_sleep(td, 0); 945 PROC_SUNLOCK(p); 946 DROP_GIANT(); 947 mi_switch(SW_VOL | SWT_SUSPEND, NULL); 948 thread_unlock(td); 949 PICKUP_GIANT(); 950 PROC_LOCK(p); 951 PROC_SLOCK(p); 952} 953 954void 955thread_suspend_one(struct thread *td) 956{ 957 struct proc *p; 958 959 p = td->td_proc; 960 PROC_SLOCK_ASSERT(p, MA_OWNED); 961 THREAD_LOCK_ASSERT(td, MA_OWNED); 962 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended")); 963 p->p_suspcount++; 964 td->td_flags &= ~TDF_NEEDSUSPCHK; 965 TD_SET_SUSPENDED(td); 966 sched_sleep(td, 0); 967} 968 969static int 970thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary) 971{ 972 973 THREAD_LOCK_ASSERT(td, MA_OWNED); 974 KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended")); 975 TD_CLR_SUSPENDED(td); 976 td->td_flags &= ~TDF_ALLPROCSUSP; 977 if (td->td_proc == p) { 978 PROC_SLOCK_ASSERT(p, MA_OWNED); 979 p->p_suspcount--; 980 if (boundary && (td->td_flags & TDF_BOUNDARY) != 0) { 981 td->td_flags &= ~TDF_BOUNDARY; 982 p->p_boundary_count--; 983 } 984 } 985 return (setrunnable(td)); 986} 987 988/* 989 * Allow all threads blocked by single threading to continue running. 990 */ 991void 992thread_unsuspend(struct proc *p) 993{ 994 struct thread *td; 995 int wakeup_swapper; 996 997 PROC_LOCK_ASSERT(p, MA_OWNED); 998 PROC_SLOCK_ASSERT(p, MA_OWNED); 999 wakeup_swapper = 0; 1000 if (!P_SHOULDSTOP(p)) { 1001 FOREACH_THREAD_IN_PROC(p, td) { 1002 thread_lock(td); 1003 if (TD_IS_SUSPENDED(td)) { 1004 wakeup_swapper |= thread_unsuspend_one(td, p, 1005 true); 1006 } 1007 thread_unlock(td); 1008 } 1009 } else if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE && 1010 p->p_numthreads == p->p_suspcount) { 1011 /* 1012 * Stopping everything also did the job for the single 1013 * threading request. Now we've downgraded to single-threaded, 1014 * let it continue. 1015 */ 1016 if (p->p_singlethread->td_proc == p) { 1017 thread_lock(p->p_singlethread); 1018 wakeup_swapper = thread_unsuspend_one( 1019 p->p_singlethread, p, false); 1020 thread_unlock(p->p_singlethread); 1021 } 1022 } 1023 if (wakeup_swapper) 1024 kick_proc0(); 1025} 1026 1027/* 1028 * End the single threading mode.. 1029 */ 1030void 1031thread_single_end(struct proc *p, int mode) 1032{ 1033 struct thread *td; 1034 int wakeup_swapper; 1035 1036 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY || 1037 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT, 1038 ("invalid mode %d", mode)); 1039 PROC_LOCK_ASSERT(p, MA_OWNED); 1040 KASSERT((mode == SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) != 0) || 1041 (mode != SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) == 0), 1042 ("mode %d does not match P_TOTAL_STOP", mode)); 1043 KASSERT(mode == SINGLE_ALLPROC || p->p_singlethread == curthread, 1044 ("thread_single_end from other thread %p %p", 1045 curthread, p->p_singlethread)); 1046 KASSERT(mode != SINGLE_BOUNDARY || 1047 (p->p_flag & P_SINGLE_BOUNDARY) != 0, 1048 ("mis-matched SINGLE_BOUNDARY flags %x", p->p_flag)); 1049 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY | 1050 P_TOTAL_STOP); 1051 PROC_SLOCK(p); 1052 p->p_singlethread = NULL; 1053 wakeup_swapper = 0; 1054 /* 1055 * If there are other threads they may now run, 1056 * unless of course there is a blanket 'stop order' 1057 * on the process. The single threader must be allowed 1058 * to continue however as this is a bad place to stop. 1059 */ 1060 if (p->p_numthreads != remain_for_mode(mode) && !P_SHOULDSTOP(p)) { 1061 FOREACH_THREAD_IN_PROC(p, td) { 1062 thread_lock(td); 1063 if (TD_IS_SUSPENDED(td)) { 1064 wakeup_swapper |= thread_unsuspend_one(td, p, 1065 mode == SINGLE_BOUNDARY); 1066 } 1067 thread_unlock(td); 1068 } 1069 } 1070 KASSERT(mode != SINGLE_BOUNDARY || p->p_boundary_count == 0, 1071 ("inconsistent boundary count %d", p->p_boundary_count)); 1072 PROC_SUNLOCK(p); 1073 if (wakeup_swapper) 1074 kick_proc0(); 1075} 1076 1077struct thread * 1078thread_find(struct proc *p, lwpid_t tid) 1079{ 1080 struct thread *td; 1081 1082 PROC_LOCK_ASSERT(p, MA_OWNED); 1083 FOREACH_THREAD_IN_PROC(p, td) { 1084 if (td->td_tid == tid) 1085 break; 1086 } 1087 return (td); 1088} 1089 1090/* Locate a thread by number; return with proc lock held. */ 1091struct thread * 1092tdfind(lwpid_t tid, pid_t pid) 1093{ 1094#define RUN_THRESH 16 1095 struct thread *td; 1096 int run = 0; 1097 1098 rw_rlock(&tidhash_lock); 1099 LIST_FOREACH(td, TIDHASH(tid), td_hash) { 1100 if (td->td_tid == tid) { 1101 if (pid != -1 && td->td_proc->p_pid != pid) { 1102 td = NULL; 1103 break; 1104 } 1105 PROC_LOCK(td->td_proc); 1106 if (td->td_proc->p_state == PRS_NEW) { 1107 PROC_UNLOCK(td->td_proc); 1108 td = NULL; 1109 break; 1110 } 1111 if (run > RUN_THRESH) { 1112 if (rw_try_upgrade(&tidhash_lock)) { 1113 LIST_REMOVE(td, td_hash); 1114 LIST_INSERT_HEAD(TIDHASH(td->td_tid), 1115 td, td_hash); 1116 rw_wunlock(&tidhash_lock); 1117 return (td); 1118 } 1119 } 1120 break; 1121 } 1122 run++; 1123 } 1124 rw_runlock(&tidhash_lock); 1125 return (td); 1126} 1127 1128void 1129tidhash_add(struct thread *td) 1130{ 1131 rw_wlock(&tidhash_lock); 1132 LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash); 1133 rw_wunlock(&tidhash_lock); 1134} 1135 1136void 1137tidhash_remove(struct thread *td) 1138{ 1139 rw_wlock(&tidhash_lock); 1140 LIST_REMOVE(td, td_hash); 1141 rw_wunlock(&tidhash_lock); 1142} 1143