1/*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (C) 2001 Julian Elischer <julian@freebsd.org>. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice(s), this list of conditions and the following disclaimer as 12 * the first lines of this file unmodified other than the possible 13 * addition of one or more copyright notices. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice(s), this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY 19 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 20 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 21 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY 22 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 23 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 24 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 25 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH 28 * DAMAGE. 29 */ 30 31#include "opt_witness.h" 32#include "opt_hwpmc_hooks.h" 33 34#include <sys/cdefs.h> 35__FBSDID("$FreeBSD$"); 36 37#include <sys/param.h> 38#include <sys/systm.h> 39#include <sys/kernel.h> 40#include <sys/lock.h> 41#include <sys/mutex.h> 42#include <sys/proc.h> 43#include <sys/rangelock.h> 44#include <sys/resourcevar.h> 45#include <sys/sdt.h> 46#include <sys/smp.h> 47#include <sys/sched.h> 48#include <sys/sleepqueue.h> 49#include <sys/selinfo.h> 50#include <sys/syscallsubr.h> 51#include <sys/sysent.h> 52#include <sys/turnstile.h> 53#include <sys/ktr.h> 54#include <sys/rwlock.h> 55#include <sys/umtx.h> 56#include <sys/vmmeter.h> 57#include <sys/cpuset.h> 58#ifdef HWPMC_HOOKS 59#include <sys/pmckern.h> 60#endif 61 62#include <security/audit/audit.h> 63 64#include <vm/vm.h> 65#include <vm/vm_extern.h> 66#include <vm/uma.h> 67#include <sys/eventhandler.h> 68 69/* 70 * Asserts below verify the stability of struct thread and struct proc 71 * layout, as exposed by KBI to modules. On head, the KBI is allowed 72 * to drift, change to the structures must be accompanied by the 73 * assert update. 74 * 75 * On the stable branches after KBI freeze, conditions must not be 76 * violated. Typically new fields are moved to the end of the 77 * structures. 78 */ 79#ifdef __amd64__ 80_Static_assert(offsetof(struct thread, td_flags) == 0xfc, 81 "struct thread KBI td_flags"); 82_Static_assert(offsetof(struct thread, td_pflags) == 0x104, 83 "struct thread KBI td_pflags"); 84_Static_assert(offsetof(struct thread, td_frame) == 0x470, 85 "struct thread KBI td_frame"); 86_Static_assert(offsetof(struct thread, td_emuldata) == 0x528, 87 "struct thread KBI td_emuldata"); 88_Static_assert(offsetof(struct proc, p_flag) == 0xb0, 89 "struct proc KBI p_flag"); 90_Static_assert(offsetof(struct proc, p_pid) == 0xbc, 91 "struct proc KBI p_pid"); 92_Static_assert(offsetof(struct proc, p_filemon) == 0x3d0, 93 "struct proc KBI p_filemon"); 94_Static_assert(offsetof(struct proc, p_comm) == 0x3e4, 95 "struct proc KBI p_comm"); 96_Static_assert(offsetof(struct proc, p_emuldata) == 0x4b8, 97 "struct proc KBI p_emuldata"); 98#endif 99#ifdef __i386__ 100_Static_assert(offsetof(struct thread, td_flags) == 0x98, 101 "struct thread KBI td_flags"); 102_Static_assert(offsetof(struct thread, td_pflags) == 0xa0, 103 "struct thread KBI td_pflags"); 104_Static_assert(offsetof(struct thread, td_frame) == 0x2e8, 105 "struct thread KBI td_frame"); 106_Static_assert(offsetof(struct thread, td_emuldata) == 0x334, 107 "struct thread KBI td_emuldata"); 108_Static_assert(offsetof(struct proc, p_flag) == 0x68, 109 "struct proc KBI p_flag"); 110_Static_assert(offsetof(struct proc, p_pid) == 0x74, 111 "struct proc KBI p_pid"); 112_Static_assert(offsetof(struct proc, p_filemon) == 0x27c, 113 "struct proc KBI p_filemon"); 114_Static_assert(offsetof(struct proc, p_comm) == 0x28c, 115 "struct proc KBI p_comm"); 116_Static_assert(offsetof(struct proc, p_emuldata) == 0x318, 117 "struct proc KBI p_emuldata"); 118#endif 119 120SDT_PROVIDER_DECLARE(proc); 121SDT_PROBE_DEFINE(proc, , , lwp__exit); 122 123/* 124 * thread related storage. 125 */ 126static uma_zone_t thread_zone; 127 128TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads); 129static struct mtx zombie_lock; 130MTX_SYSINIT(zombie_lock, &zombie_lock, "zombie lock", MTX_SPIN); 131 132static void thread_zombie(struct thread *); 133static int thread_unsuspend_one(struct thread *td, struct proc *p, 134 bool boundary); 135 136#define TID_BUFFER_SIZE 1024 137 138struct mtx tid_lock; 139static struct unrhdr *tid_unrhdr; 140static lwpid_t tid_buffer[TID_BUFFER_SIZE]; 141static int tid_head, tid_tail; 142static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash"); 143 144struct tidhashhead *tidhashtbl; 145u_long tidhash; 146struct rwlock tidhash_lock; 147 148EVENTHANDLER_LIST_DEFINE(thread_ctor); 149EVENTHANDLER_LIST_DEFINE(thread_dtor); 150EVENTHANDLER_LIST_DEFINE(thread_init); 151EVENTHANDLER_LIST_DEFINE(thread_fini); 152 153static lwpid_t 154tid_alloc(void) 155{ 156 lwpid_t tid; 157 158 tid = alloc_unr(tid_unrhdr); 159 if (tid != -1) 160 return (tid); 161 mtx_lock(&tid_lock); 162 if (tid_head == tid_tail) { 163 mtx_unlock(&tid_lock); 164 return (-1); 165 } 166 tid = tid_buffer[tid_head]; 167 tid_head = (tid_head + 1) % TID_BUFFER_SIZE; 168 mtx_unlock(&tid_lock); 169 return (tid); 170} 171 172static void 173tid_free(lwpid_t tid) 174{ 175 lwpid_t tmp_tid = -1; 176 177 mtx_lock(&tid_lock); 178 if ((tid_tail + 1) % TID_BUFFER_SIZE == tid_head) { 179 tmp_tid = tid_buffer[tid_head]; 180 tid_head = (tid_head + 1) % TID_BUFFER_SIZE; 181 } 182 tid_buffer[tid_tail] = tid; 183 tid_tail = (tid_tail + 1) % TID_BUFFER_SIZE; 184 mtx_unlock(&tid_lock); 185 if (tmp_tid != -1) 186 free_unr(tid_unrhdr, tmp_tid); 187} 188 189/* 190 * Prepare a thread for use. 191 */ 192static int 193thread_ctor(void *mem, int size, void *arg, int flags) 194{ 195 struct thread *td; 196 197 td = (struct thread *)mem; 198 td->td_state = TDS_INACTIVE; 199 td->td_lastcpu = td->td_oncpu = NOCPU; 200 201 td->td_tid = tid_alloc(); 202 203 /* 204 * Note that td_critnest begins life as 1 because the thread is not 205 * running and is thereby implicitly waiting to be on the receiving 206 * end of a context switch. 207 */ 208 td->td_critnest = 1; 209 td->td_lend_user_pri = PRI_MAX; 210 EVENTHANDLER_DIRECT_INVOKE(thread_ctor, td); 211#ifdef AUDIT 212 audit_thread_alloc(td); 213#endif 214 umtx_thread_alloc(td); 215 return (0); 216} 217 218/* 219 * Reclaim a thread after use. 220 */ 221static void 222thread_dtor(void *mem, int size, void *arg) 223{ 224 struct thread *td; 225 226 td = (struct thread *)mem; 227 228#ifdef INVARIANTS 229 /* Verify that this thread is in a safe state to free. */ 230 switch (td->td_state) { 231 case TDS_INHIBITED: 232 case TDS_RUNNING: 233 case TDS_CAN_RUN: 234 case TDS_RUNQ: 235 /* 236 * We must never unlink a thread that is in one of 237 * these states, because it is currently active. 238 */ 239 panic("bad state for thread unlinking"); 240 /* NOTREACHED */ 241 case TDS_INACTIVE: 242 break; 243 default: 244 panic("bad thread state"); 245 /* NOTREACHED */ 246 } 247#endif 248#ifdef AUDIT 249 audit_thread_free(td); 250#endif 251 /* Free all OSD associated to this thread. */ 252 osd_thread_exit(td); 253 td_softdep_cleanup(td); 254 MPASS(td->td_su == NULL); 255 256 EVENTHANDLER_DIRECT_INVOKE(thread_dtor, td); 257 tid_free(td->td_tid); 258} 259 260/* 261 * Initialize type-stable parts of a thread (when newly created). 262 */ 263static int 264thread_init(void *mem, int size, int flags) 265{ 266 struct thread *td; 267 268 td = (struct thread *)mem; 269 270 td->td_sleepqueue = sleepq_alloc(); 271 td->td_turnstile = turnstile_alloc(); 272 td->td_rlqe = NULL; 273 EVENTHANDLER_DIRECT_INVOKE(thread_init, td); 274 umtx_thread_init(td); 275 td->td_kstack = 0; 276 td->td_sel = NULL; 277 return (0); 278} 279 280/* 281 * Tear down type-stable parts of a thread (just before being discarded). 282 */ 283static void 284thread_fini(void *mem, int size) 285{ 286 struct thread *td; 287 288 td = (struct thread *)mem; 289 EVENTHANDLER_DIRECT_INVOKE(thread_fini, td); 290 rlqentry_free(td->td_rlqe); 291 turnstile_free(td->td_turnstile); 292 sleepq_free(td->td_sleepqueue); 293 umtx_thread_fini(td); 294 seltdfini(td); 295} 296 297/* 298 * For a newly created process, 299 * link up all the structures and its initial threads etc. 300 * called from: 301 * {arch}/{arch}/machdep.c {arch}_init(), init386() etc. 302 * proc_dtor() (should go away) 303 * proc_init() 304 */ 305void 306proc_linkup0(struct proc *p, struct thread *td) 307{ 308 TAILQ_INIT(&p->p_threads); /* all threads in proc */ 309 proc_linkup(p, td); 310} 311 312void 313proc_linkup(struct proc *p, struct thread *td) 314{ 315 316 sigqueue_init(&p->p_sigqueue, p); 317 p->p_ksi = ksiginfo_alloc(1); 318 if (p->p_ksi != NULL) { 319 /* XXX p_ksi may be null if ksiginfo zone is not ready */ 320 p->p_ksi->ksi_flags = KSI_EXT | KSI_INS; 321 } 322 LIST_INIT(&p->p_mqnotifier); 323 p->p_numthreads = 0; 324 thread_link(td, p); 325} 326 327/* 328 * Initialize global thread allocation resources. 329 */ 330void 331threadinit(void) 332{ 333 334 mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF); 335 336 /* 337 * pid_max cannot be greater than PID_MAX. 338 * leave one number for thread0. 339 */ 340 tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock); 341 342 thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(), 343 thread_ctor, thread_dtor, thread_init, thread_fini, 344 32 - 1, UMA_ZONE_NOFREE); 345 tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash); 346 rw_init(&tidhash_lock, "tidhash"); 347} 348 349/* 350 * Place an unused thread on the zombie list. 351 * Use the slpq as that must be unused by now. 352 */ 353void 354thread_zombie(struct thread *td) 355{ 356 mtx_lock_spin(&zombie_lock); 357 TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq); 358 mtx_unlock_spin(&zombie_lock); 359} 360 361/* 362 * Release a thread that has exited after cpu_throw(). 363 */ 364void 365thread_stash(struct thread *td) 366{ 367 atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1); 368 thread_zombie(td); 369} 370 371/* 372 * Reap zombie resources. 373 */ 374void 375thread_reap(void) 376{ 377 struct thread *td_first, *td_next; 378 379 /* 380 * Don't even bother to lock if none at this instant, 381 * we really don't care about the next instant. 382 */ 383 if (!TAILQ_EMPTY(&zombie_threads)) { 384 mtx_lock_spin(&zombie_lock); 385 td_first = TAILQ_FIRST(&zombie_threads); 386 if (td_first) 387 TAILQ_INIT(&zombie_threads); 388 mtx_unlock_spin(&zombie_lock); 389 while (td_first) { 390 td_next = TAILQ_NEXT(td_first, td_slpq); 391 thread_cow_free(td_first); 392 thread_free(td_first); 393 td_first = td_next; 394 } 395 } 396} 397 398/* 399 * Allocate a thread. 400 */ 401struct thread * 402thread_alloc(int pages) 403{ 404 struct thread *td; 405 406 thread_reap(); /* check if any zombies to get */ 407 408 td = (struct thread *)uma_zalloc(thread_zone, M_WAITOK); 409 KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack")); 410 if (!vm_thread_new(td, pages)) { 411 uma_zfree(thread_zone, td); 412 return (NULL); 413 } 414 cpu_thread_alloc(td); 415 return (td); 416} 417 418int 419thread_alloc_stack(struct thread *td, int pages) 420{ 421 422 KASSERT(td->td_kstack == 0, 423 ("thread_alloc_stack called on a thread with kstack")); 424 if (!vm_thread_new(td, pages)) 425 return (0); 426 cpu_thread_alloc(td); 427 return (1); 428} 429 430/* 431 * Deallocate a thread. 432 */ 433void 434thread_free(struct thread *td) 435{ 436 437 lock_profile_thread_exit(td); 438 if (td->td_cpuset) 439 cpuset_rel(td->td_cpuset); 440 td->td_cpuset = NULL; 441 cpu_thread_free(td); 442 if (td->td_kstack != 0) 443 vm_thread_dispose(td); 444 callout_drain(&td->td_slpcallout); 445 uma_zfree(thread_zone, td); 446} 447 448void 449thread_cow_get_proc(struct thread *newtd, struct proc *p) 450{ 451 452 PROC_LOCK_ASSERT(p, MA_OWNED); 453 newtd->td_ucred = crhold(p->p_ucred); 454 newtd->td_limit = lim_hold(p->p_limit); 455 newtd->td_cowgen = p->p_cowgen; 456} 457 458void 459thread_cow_get(struct thread *newtd, struct thread *td) 460{ 461 462 newtd->td_ucred = crhold(td->td_ucred); 463 newtd->td_limit = lim_hold(td->td_limit); 464 newtd->td_cowgen = td->td_cowgen; 465} 466 467void 468thread_cow_free(struct thread *td) 469{ 470 471 if (td->td_ucred != NULL) 472 crfree(td->td_ucred); 473 if (td->td_limit != NULL) 474 lim_free(td->td_limit); 475} 476 477void 478thread_cow_update(struct thread *td) 479{ 480 struct proc *p; 481 struct ucred *oldcred; 482 struct plimit *oldlimit; 483 484 p = td->td_proc; 485 oldcred = NULL; 486 oldlimit = NULL; 487 PROC_LOCK(p); 488 if (td->td_ucred != p->p_ucred) { 489 oldcred = td->td_ucred; 490 td->td_ucred = crhold(p->p_ucred); 491 } 492 if (td->td_limit != p->p_limit) { 493 oldlimit = td->td_limit; 494 td->td_limit = lim_hold(p->p_limit); 495 } 496 td->td_cowgen = p->p_cowgen; 497 PROC_UNLOCK(p); 498 if (oldcred != NULL) 499 crfree(oldcred); 500 if (oldlimit != NULL) 501 lim_free(oldlimit); 502} 503 504/* 505 * Discard the current thread and exit from its context. 506 * Always called with scheduler locked. 507 * 508 * Because we can't free a thread while we're operating under its context, 509 * push the current thread into our CPU's deadthread holder. This means 510 * we needn't worry about someone else grabbing our context before we 511 * do a cpu_throw(). 512 */ 513void 514thread_exit(void) 515{ 516 uint64_t runtime, new_switchtime; 517 struct thread *td; 518 struct thread *td2; 519 struct proc *p; 520 int wakeup_swapper; 521 522 td = curthread; 523 p = td->td_proc; 524 525 PROC_SLOCK_ASSERT(p, MA_OWNED); 526 mtx_assert(&Giant, MA_NOTOWNED); 527 528 PROC_LOCK_ASSERT(p, MA_OWNED); 529 KASSERT(p != NULL, ("thread exiting without a process")); 530 CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td, 531 (long)p->p_pid, td->td_name); 532 SDT_PROBE0(proc, , , lwp__exit); 533 KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending")); 534 535 /* 536 * drop FPU & debug register state storage, or any other 537 * architecture specific resources that 538 * would not be on a new untouched process. 539 */ 540 cpu_thread_exit(td); 541 542 /* 543 * The last thread is left attached to the process 544 * So that the whole bundle gets recycled. Skip 545 * all this stuff if we never had threads. 546 * EXIT clears all sign of other threads when 547 * it goes to single threading, so the last thread always 548 * takes the short path. 549 */ 550 if (p->p_flag & P_HADTHREADS) { 551 if (p->p_numthreads > 1) { 552 atomic_add_int(&td->td_proc->p_exitthreads, 1); 553 thread_unlink(td); 554 td2 = FIRST_THREAD_IN_PROC(p); 555 sched_exit_thread(td2, td); 556 557 /* 558 * The test below is NOT true if we are the 559 * sole exiting thread. P_STOPPED_SINGLE is unset 560 * in exit1() after it is the only survivor. 561 */ 562 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 563 if (p->p_numthreads == p->p_suspcount) { 564 thread_lock(p->p_singlethread); 565 wakeup_swapper = thread_unsuspend_one( 566 p->p_singlethread, p, false); 567 thread_unlock(p->p_singlethread); 568 if (wakeup_swapper) 569 kick_proc0(); 570 } 571 } 572 573 PCPU_SET(deadthread, td); 574 } else { 575 /* 576 * The last thread is exiting.. but not through exit() 577 */ 578 panic ("thread_exit: Last thread exiting on its own"); 579 } 580 } 581#ifdef HWPMC_HOOKS 582 /* 583 * If this thread is part of a process that is being tracked by hwpmc(4), 584 * inform the module of the thread's impending exit. 585 */ 586 if (PMC_PROC_IS_USING_PMCS(td->td_proc)) { 587 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT); 588 PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_EXIT, NULL); 589 } else if (PMC_SYSTEM_SAMPLING_ACTIVE()) 590 PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_EXIT_LOG, NULL); 591#endif 592 PROC_UNLOCK(p); 593 PROC_STATLOCK(p); 594 thread_lock(td); 595 PROC_SUNLOCK(p); 596 597 /* Do the same timestamp bookkeeping that mi_switch() would do. */ 598 new_switchtime = cpu_ticks(); 599 runtime = new_switchtime - PCPU_GET(switchtime); 600 td->td_runtime += runtime; 601 td->td_incruntime += runtime; 602 PCPU_SET(switchtime, new_switchtime); 603 PCPU_SET(switchticks, ticks); 604 VM_CNT_INC(v_swtch); 605 606 /* Save our resource usage in our process. */ 607 td->td_ru.ru_nvcsw++; 608 ruxagg(p, td); 609 rucollect(&p->p_ru, &td->td_ru); 610 PROC_STATUNLOCK(p); 611 612 td->td_state = TDS_INACTIVE; 613#ifdef WITNESS 614 witness_thread_exit(td); 615#endif 616 CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td); 617 sched_throw(td); 618 panic("I'm a teapot!"); 619 /* NOTREACHED */ 620} 621 622/* 623 * Do any thread specific cleanups that may be needed in wait() 624 * called with Giant, proc and schedlock not held. 625 */ 626void 627thread_wait(struct proc *p) 628{ 629 struct thread *td; 630 631 mtx_assert(&Giant, MA_NOTOWNED); 632 KASSERT(p->p_numthreads == 1, ("multiple threads in thread_wait()")); 633 KASSERT(p->p_exitthreads == 0, ("p_exitthreads leaking")); 634 td = FIRST_THREAD_IN_PROC(p); 635 /* Lock the last thread so we spin until it exits cpu_throw(). */ 636 thread_lock(td); 637 thread_unlock(td); 638 lock_profile_thread_exit(td); 639 cpuset_rel(td->td_cpuset); 640 td->td_cpuset = NULL; 641 cpu_thread_clean(td); 642 thread_cow_free(td); 643 callout_drain(&td->td_slpcallout); 644 thread_reap(); /* check for zombie threads etc. */ 645} 646 647/* 648 * Link a thread to a process. 649 * set up anything that needs to be initialized for it to 650 * be used by the process. 651 */ 652void 653thread_link(struct thread *td, struct proc *p) 654{ 655 656 /* 657 * XXX This can't be enabled because it's called for proc0 before 658 * its lock has been created. 659 * PROC_LOCK_ASSERT(p, MA_OWNED); 660 */ 661 td->td_state = TDS_INACTIVE; 662 td->td_proc = p; 663 td->td_flags = TDF_INMEM; 664 665 LIST_INIT(&td->td_contested); 666 LIST_INIT(&td->td_lprof[0]); 667 LIST_INIT(&td->td_lprof[1]); 668 sigqueue_init(&td->td_sigqueue, p); 669 callout_init(&td->td_slpcallout, 1); 670 TAILQ_INSERT_TAIL(&p->p_threads, td, td_plist); 671 p->p_numthreads++; 672} 673 674/* 675 * Called from: 676 * thread_exit() 677 */ 678void 679thread_unlink(struct thread *td) 680{ 681 struct proc *p = td->td_proc; 682 683 PROC_LOCK_ASSERT(p, MA_OWNED); 684 TAILQ_REMOVE(&p->p_threads, td, td_plist); 685 p->p_numthreads--; 686 /* could clear a few other things here */ 687 /* Must NOT clear links to proc! */ 688} 689 690static int 691calc_remaining(struct proc *p, int mode) 692{ 693 int remaining; 694 695 PROC_LOCK_ASSERT(p, MA_OWNED); 696 PROC_SLOCK_ASSERT(p, MA_OWNED); 697 if (mode == SINGLE_EXIT) 698 remaining = p->p_numthreads; 699 else if (mode == SINGLE_BOUNDARY) 700 remaining = p->p_numthreads - p->p_boundary_count; 701 else if (mode == SINGLE_NO_EXIT || mode == SINGLE_ALLPROC) 702 remaining = p->p_numthreads - p->p_suspcount; 703 else 704 panic("calc_remaining: wrong mode %d", mode); 705 return (remaining); 706} 707 708static int 709remain_for_mode(int mode) 710{ 711 712 return (mode == SINGLE_ALLPROC ? 0 : 1); 713} 714 715static int 716weed_inhib(int mode, struct thread *td2, struct proc *p) 717{ 718 int wakeup_swapper; 719 720 PROC_LOCK_ASSERT(p, MA_OWNED); 721 PROC_SLOCK_ASSERT(p, MA_OWNED); 722 THREAD_LOCK_ASSERT(td2, MA_OWNED); 723 724 wakeup_swapper = 0; 725 switch (mode) { 726 case SINGLE_EXIT: 727 if (TD_IS_SUSPENDED(td2)) 728 wakeup_swapper |= thread_unsuspend_one(td2, p, true); 729 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) 730 wakeup_swapper |= sleepq_abort(td2, EINTR); 731 break; 732 case SINGLE_BOUNDARY: 733 case SINGLE_NO_EXIT: 734 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & TDF_BOUNDARY) == 0) 735 wakeup_swapper |= thread_unsuspend_one(td2, p, false); 736 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) 737 wakeup_swapper |= sleepq_abort(td2, ERESTART); 738 break; 739 case SINGLE_ALLPROC: 740 /* 741 * ALLPROC suspend tries to avoid spurious EINTR for 742 * threads sleeping interruptable, by suspending the 743 * thread directly, similarly to sig_suspend_threads(). 744 * Since such sleep is not performed at the user 745 * boundary, TDF_BOUNDARY flag is not set, and TDF_ALLPROCSUSP 746 * is used to avoid immediate un-suspend. 747 */ 748 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & (TDF_BOUNDARY | 749 TDF_ALLPROCSUSP)) == 0) 750 wakeup_swapper |= thread_unsuspend_one(td2, p, false); 751 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) { 752 if ((td2->td_flags & TDF_SBDRY) == 0) { 753 thread_suspend_one(td2); 754 td2->td_flags |= TDF_ALLPROCSUSP; 755 } else { 756 wakeup_swapper |= sleepq_abort(td2, ERESTART); 757 } 758 } 759 break; 760 } 761 return (wakeup_swapper); 762} 763 764/* 765 * Enforce single-threading. 766 * 767 * Returns 1 if the caller must abort (another thread is waiting to 768 * exit the process or similar). Process is locked! 769 * Returns 0 when you are successfully the only thread running. 770 * A process has successfully single threaded in the suspend mode when 771 * There are no threads in user mode. Threads in the kernel must be 772 * allowed to continue until they get to the user boundary. They may even 773 * copy out their return values and data before suspending. They may however be 774 * accelerated in reaching the user boundary as we will wake up 775 * any sleeping threads that are interruptable. (PCATCH). 776 */ 777int 778thread_single(struct proc *p, int mode) 779{ 780 struct thread *td; 781 struct thread *td2; 782 int remaining, wakeup_swapper; 783 784 td = curthread; 785 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY || 786 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT, 787 ("invalid mode %d", mode)); 788 /* 789 * If allowing non-ALLPROC singlethreading for non-curproc 790 * callers, calc_remaining() and remain_for_mode() should be 791 * adjusted to also account for td->td_proc != p. For now 792 * this is not implemented because it is not used. 793 */ 794 KASSERT((mode == SINGLE_ALLPROC && td->td_proc != p) || 795 (mode != SINGLE_ALLPROC && td->td_proc == p), 796 ("mode %d proc %p curproc %p", mode, p, td->td_proc)); 797 mtx_assert(&Giant, MA_NOTOWNED); 798 PROC_LOCK_ASSERT(p, MA_OWNED); 799 800 if ((p->p_flag & P_HADTHREADS) == 0 && mode != SINGLE_ALLPROC) 801 return (0); 802 803 /* Is someone already single threading? */ 804 if (p->p_singlethread != NULL && p->p_singlethread != td) 805 return (1); 806 807 if (mode == SINGLE_EXIT) { 808 p->p_flag |= P_SINGLE_EXIT; 809 p->p_flag &= ~P_SINGLE_BOUNDARY; 810 } else { 811 p->p_flag &= ~P_SINGLE_EXIT; 812 if (mode == SINGLE_BOUNDARY) 813 p->p_flag |= P_SINGLE_BOUNDARY; 814 else 815 p->p_flag &= ~P_SINGLE_BOUNDARY; 816 } 817 if (mode == SINGLE_ALLPROC) 818 p->p_flag |= P_TOTAL_STOP; 819 p->p_flag |= P_STOPPED_SINGLE; 820 PROC_SLOCK(p); 821 p->p_singlethread = td; 822 remaining = calc_remaining(p, mode); 823 while (remaining != remain_for_mode(mode)) { 824 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE) 825 goto stopme; 826 wakeup_swapper = 0; 827 FOREACH_THREAD_IN_PROC(p, td2) { 828 if (td2 == td) 829 continue; 830 thread_lock(td2); 831 td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK; 832 if (TD_IS_INHIBITED(td2)) { 833 wakeup_swapper |= weed_inhib(mode, td2, p); 834#ifdef SMP 835 } else if (TD_IS_RUNNING(td2) && td != td2) { 836 forward_signal(td2); 837#endif 838 } 839 thread_unlock(td2); 840 } 841 if (wakeup_swapper) 842 kick_proc0(); 843 remaining = calc_remaining(p, mode); 844 845 /* 846 * Maybe we suspended some threads.. was it enough? 847 */ 848 if (remaining == remain_for_mode(mode)) 849 break; 850 851stopme: 852 /* 853 * Wake us up when everyone else has suspended. 854 * In the mean time we suspend as well. 855 */ 856 thread_suspend_switch(td, p); 857 remaining = calc_remaining(p, mode); 858 } 859 if (mode == SINGLE_EXIT) { 860 /* 861 * Convert the process to an unthreaded process. The 862 * SINGLE_EXIT is called by exit1() or execve(), in 863 * both cases other threads must be retired. 864 */ 865 KASSERT(p->p_numthreads == 1, ("Unthreading with >1 threads")); 866 p->p_singlethread = NULL; 867 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_HADTHREADS); 868 869 /* 870 * Wait for any remaining threads to exit cpu_throw(). 871 */ 872 while (p->p_exitthreads != 0) { 873 PROC_SUNLOCK(p); 874 PROC_UNLOCK(p); 875 sched_relinquish(td); 876 PROC_LOCK(p); 877 PROC_SLOCK(p); 878 } 879 } else if (mode == SINGLE_BOUNDARY) { 880 /* 881 * Wait until all suspended threads are removed from 882 * the processors. The thread_suspend_check() 883 * increments p_boundary_count while it is still 884 * running, which makes it possible for the execve() 885 * to destroy vmspace while our other threads are 886 * still using the address space. 887 * 888 * We lock the thread, which is only allowed to 889 * succeed after context switch code finished using 890 * the address space. 891 */ 892 FOREACH_THREAD_IN_PROC(p, td2) { 893 if (td2 == td) 894 continue; 895 thread_lock(td2); 896 KASSERT((td2->td_flags & TDF_BOUNDARY) != 0, 897 ("td %p not on boundary", td2)); 898 KASSERT(TD_IS_SUSPENDED(td2), 899 ("td %p is not suspended", td2)); 900 thread_unlock(td2); 901 } 902 } 903 PROC_SUNLOCK(p); 904 return (0); 905} 906 907bool 908thread_suspend_check_needed(void) 909{ 910 struct proc *p; 911 struct thread *td; 912 913 td = curthread; 914 p = td->td_proc; 915 PROC_LOCK_ASSERT(p, MA_OWNED); 916 return (P_SHOULDSTOP(p) || ((p->p_flag & P_TRACED) != 0 && 917 (td->td_dbgflags & TDB_SUSPEND) != 0)); 918} 919 920/* 921 * Called in from locations that can safely check to see 922 * whether we have to suspend or at least throttle for a 923 * single-thread event (e.g. fork). 924 * 925 * Such locations include userret(). 926 * If the "return_instead" argument is non zero, the thread must be able to 927 * accept 0 (caller may continue), or 1 (caller must abort) as a result. 928 * 929 * The 'return_instead' argument tells the function if it may do a 930 * thread_exit() or suspend, or whether the caller must abort and back 931 * out instead. 932 * 933 * If the thread that set the single_threading request has set the 934 * P_SINGLE_EXIT bit in the process flags then this call will never return 935 * if 'return_instead' is false, but will exit. 936 * 937 * P_SINGLE_EXIT | return_instead == 0| return_instead != 0 938 *---------------+--------------------+--------------------- 939 * 0 | returns 0 | returns 0 or 1 940 * | when ST ends | immediately 941 *---------------+--------------------+--------------------- 942 * 1 | thread exits | returns 1 943 * | | immediately 944 * 0 = thread_exit() or suspension ok, 945 * other = return error instead of stopping the thread. 946 * 947 * While a full suspension is under effect, even a single threading 948 * thread would be suspended if it made this call (but it shouldn't). 949 * This call should only be made from places where 950 * thread_exit() would be safe as that may be the outcome unless 951 * return_instead is set. 952 */ 953int 954thread_suspend_check(int return_instead) 955{ 956 struct thread *td; 957 struct proc *p; 958 int wakeup_swapper; 959 960 td = curthread; 961 p = td->td_proc; 962 mtx_assert(&Giant, MA_NOTOWNED); 963 PROC_LOCK_ASSERT(p, MA_OWNED); 964 while (thread_suspend_check_needed()) { 965 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 966 KASSERT(p->p_singlethread != NULL, 967 ("singlethread not set")); 968 /* 969 * The only suspension in action is a 970 * single-threading. Single threader need not stop. 971 * It is safe to access p->p_singlethread unlocked 972 * because it can only be set to our address by us. 973 */ 974 if (p->p_singlethread == td) 975 return (0); /* Exempt from stopping. */ 976 } 977 if ((p->p_flag & P_SINGLE_EXIT) && return_instead) 978 return (EINTR); 979 980 /* Should we goto user boundary if we didn't come from there? */ 981 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE && 982 (p->p_flag & P_SINGLE_BOUNDARY) && return_instead) 983 return (ERESTART); 984 985 /* 986 * Ignore suspend requests if they are deferred. 987 */ 988 if ((td->td_flags & TDF_SBDRY) != 0) { 989 KASSERT(return_instead, 990 ("TDF_SBDRY set for unsafe thread_suspend_check")); 991 KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) != 992 (TDF_SEINTR | TDF_SERESTART), 993 ("both TDF_SEINTR and TDF_SERESTART")); 994 return (TD_SBDRY_INTR(td) ? TD_SBDRY_ERRNO(td) : 0); 995 } 996 997 /* 998 * If the process is waiting for us to exit, 999 * this thread should just suicide. 1000 * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE. 1001 */ 1002 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) { 1003 PROC_UNLOCK(p); 1004 1005 /* 1006 * Allow Linux emulation layer to do some work 1007 * before thread suicide. 1008 */ 1009 if (__predict_false(p->p_sysent->sv_thread_detach != NULL)) 1010 (p->p_sysent->sv_thread_detach)(td); 1011 umtx_thread_exit(td); 1012 kern_thr_exit(td); 1013 panic("stopped thread did not exit"); 1014 } 1015 1016 PROC_SLOCK(p); 1017 thread_stopped(p); 1018 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 1019 if (p->p_numthreads == p->p_suspcount + 1) { 1020 thread_lock(p->p_singlethread); 1021 wakeup_swapper = thread_unsuspend_one( 1022 p->p_singlethread, p, false); 1023 thread_unlock(p->p_singlethread); 1024 if (wakeup_swapper) 1025 kick_proc0(); 1026 } 1027 } 1028 PROC_UNLOCK(p); 1029 thread_lock(td); 1030 /* 1031 * When a thread suspends, it just 1032 * gets taken off all queues. 1033 */ 1034 thread_suspend_one(td); 1035 if (return_instead == 0) { 1036 p->p_boundary_count++; 1037 td->td_flags |= TDF_BOUNDARY; 1038 } 1039 PROC_SUNLOCK(p); 1040 mi_switch(SW_INVOL | SWT_SUSPEND, NULL); 1041 thread_unlock(td); 1042 PROC_LOCK(p); 1043 } 1044 return (0); 1045} 1046 1047/* 1048 * Check for possible stops and suspensions while executing a 1049 * casueword or similar transiently failing operation. 1050 * 1051 * The sleep argument controls whether the function can handle a stop 1052 * request itself or it should return ERESTART and the request is 1053 * proceed at the kernel/user boundary in ast. 1054 * 1055 * Typically, when retrying due to casueword(9) failure (rv == 1), we 1056 * should handle the stop requests there, with exception of cases when 1057 * the thread owns a kernel resource, for instance busied the umtx 1058 * key, or when functions return immediately if thread_check_susp() 1059 * returned non-zero. On the other hand, retrying the whole lock 1060 * operation, we better not stop there but delegate the handling to 1061 * ast. 1062 * 1063 * If the request is for thread termination P_SINGLE_EXIT, we cannot 1064 * handle it at all, and simply return EINTR. 1065 */ 1066int 1067thread_check_susp(struct thread *td, bool sleep) 1068{ 1069 struct proc *p; 1070 int error; 1071 1072 /* 1073 * The check for TDF_NEEDSUSPCHK is racy, but it is enough to 1074 * eventually break the lockstep loop. 1075 */ 1076 if ((td->td_flags & TDF_NEEDSUSPCHK) == 0) 1077 return (0); 1078 error = 0; 1079 p = td->td_proc; 1080 PROC_LOCK(p); 1081 if (p->p_flag & P_SINGLE_EXIT) 1082 error = EINTR; 1083 else if (P_SHOULDSTOP(p) || 1084 ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_SUSPEND))) 1085 error = sleep ? thread_suspend_check(0) : ERESTART; 1086 PROC_UNLOCK(p); 1087 return (error); 1088} 1089 1090void 1091thread_suspend_switch(struct thread *td, struct proc *p) 1092{ 1093 1094 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended")); 1095 PROC_LOCK_ASSERT(p, MA_OWNED); 1096 PROC_SLOCK_ASSERT(p, MA_OWNED); 1097 /* 1098 * We implement thread_suspend_one in stages here to avoid 1099 * dropping the proc lock while the thread lock is owned. 1100 */ 1101 if (p == td->td_proc) { 1102 thread_stopped(p); 1103 p->p_suspcount++; 1104 } 1105 PROC_UNLOCK(p); 1106 thread_lock(td); 1107 td->td_flags &= ~TDF_NEEDSUSPCHK; 1108 TD_SET_SUSPENDED(td); 1109 sched_sleep(td, 0); 1110 PROC_SUNLOCK(p); 1111 DROP_GIANT(); 1112 mi_switch(SW_VOL | SWT_SUSPEND, NULL); 1113 thread_unlock(td); 1114 PICKUP_GIANT(); 1115 PROC_LOCK(p); 1116 PROC_SLOCK(p); 1117} 1118 1119void 1120thread_suspend_one(struct thread *td) 1121{ 1122 struct proc *p; 1123 1124 p = td->td_proc; 1125 PROC_SLOCK_ASSERT(p, MA_OWNED); 1126 THREAD_LOCK_ASSERT(td, MA_OWNED); 1127 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended")); 1128 p->p_suspcount++; 1129 td->td_flags &= ~TDF_NEEDSUSPCHK; 1130 TD_SET_SUSPENDED(td); 1131 sched_sleep(td, 0); 1132} 1133 1134static int 1135thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary) 1136{ 1137 1138 THREAD_LOCK_ASSERT(td, MA_OWNED); 1139 KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended")); 1140 TD_CLR_SUSPENDED(td); 1141 td->td_flags &= ~TDF_ALLPROCSUSP; 1142 if (td->td_proc == p) { 1143 PROC_SLOCK_ASSERT(p, MA_OWNED); 1144 p->p_suspcount--; 1145 if (boundary && (td->td_flags & TDF_BOUNDARY) != 0) { 1146 td->td_flags &= ~TDF_BOUNDARY; 1147 p->p_boundary_count--; 1148 } 1149 } 1150 return (setrunnable(td)); 1151} 1152 1153/* 1154 * Allow all threads blocked by single threading to continue running. 1155 */ 1156void 1157thread_unsuspend(struct proc *p) 1158{ 1159 struct thread *td; 1160 int wakeup_swapper; 1161 1162 PROC_LOCK_ASSERT(p, MA_OWNED); 1163 PROC_SLOCK_ASSERT(p, MA_OWNED); 1164 wakeup_swapper = 0; 1165 if (!P_SHOULDSTOP(p)) { 1166 FOREACH_THREAD_IN_PROC(p, td) { 1167 thread_lock(td); 1168 if (TD_IS_SUSPENDED(td)) { 1169 wakeup_swapper |= thread_unsuspend_one(td, p, 1170 true); 1171 } 1172 thread_unlock(td); 1173 } 1174 } else if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE && 1175 p->p_numthreads == p->p_suspcount) { 1176 /* 1177 * Stopping everything also did the job for the single 1178 * threading request. Now we've downgraded to single-threaded, 1179 * let it continue. 1180 */ 1181 if (p->p_singlethread->td_proc == p) { 1182 thread_lock(p->p_singlethread); 1183 wakeup_swapper = thread_unsuspend_one( 1184 p->p_singlethread, p, false); 1185 thread_unlock(p->p_singlethread); 1186 } 1187 } 1188 if (wakeup_swapper) 1189 kick_proc0(); 1190} 1191 1192/* 1193 * End the single threading mode.. 1194 */ 1195void 1196thread_single_end(struct proc *p, int mode) 1197{ 1198 struct thread *td; 1199 int wakeup_swapper; 1200 1201 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY || 1202 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT, 1203 ("invalid mode %d", mode)); 1204 PROC_LOCK_ASSERT(p, MA_OWNED); 1205 KASSERT((mode == SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) != 0) || 1206 (mode != SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) == 0), 1207 ("mode %d does not match P_TOTAL_STOP", mode)); 1208 KASSERT(mode == SINGLE_ALLPROC || p->p_singlethread == curthread, 1209 ("thread_single_end from other thread %p %p", 1210 curthread, p->p_singlethread)); 1211 KASSERT(mode != SINGLE_BOUNDARY || 1212 (p->p_flag & P_SINGLE_BOUNDARY) != 0, 1213 ("mis-matched SINGLE_BOUNDARY flags %x", p->p_flag)); 1214 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY | 1215 P_TOTAL_STOP); 1216 PROC_SLOCK(p); 1217 p->p_singlethread = NULL; 1218 wakeup_swapper = 0; 1219 /* 1220 * If there are other threads they may now run, 1221 * unless of course there is a blanket 'stop order' 1222 * on the process. The single threader must be allowed 1223 * to continue however as this is a bad place to stop. 1224 */ 1225 if (p->p_numthreads != remain_for_mode(mode) && !P_SHOULDSTOP(p)) { 1226 FOREACH_THREAD_IN_PROC(p, td) { 1227 thread_lock(td); 1228 if (TD_IS_SUSPENDED(td)) { 1229 wakeup_swapper |= thread_unsuspend_one(td, p, 1230 mode == SINGLE_BOUNDARY); 1231 } 1232 thread_unlock(td); 1233 } 1234 } 1235 KASSERT(mode != SINGLE_BOUNDARY || p->p_boundary_count == 0, 1236 ("inconsistent boundary count %d", p->p_boundary_count)); 1237 PROC_SUNLOCK(p); 1238 if (wakeup_swapper) 1239 kick_proc0(); 1240} 1241 1242struct thread * 1243thread_find(struct proc *p, lwpid_t tid) 1244{ 1245 struct thread *td; 1246 1247 PROC_LOCK_ASSERT(p, MA_OWNED); 1248 FOREACH_THREAD_IN_PROC(p, td) { 1249 if (td->td_tid == tid) 1250 break; 1251 } 1252 return (td); 1253} 1254 1255/* Locate a thread by number; return with proc lock held. */ 1256struct thread * 1257tdfind(lwpid_t tid, pid_t pid) 1258{ 1259#define RUN_THRESH 16 1260 struct thread *td; 1261 int run = 0; 1262 1263 td = curthread; 1264 if (td->td_tid == tid) { 1265 if (pid != -1 && td->td_proc->p_pid != pid) 1266 return (NULL); 1267 PROC_LOCK(td->td_proc); 1268 return (td); 1269 } 1270 1271 rw_rlock(&tidhash_lock); 1272 LIST_FOREACH(td, TIDHASH(tid), td_hash) { 1273 if (td->td_tid == tid) { 1274 if (pid != -1 && td->td_proc->p_pid != pid) { 1275 td = NULL; 1276 break; 1277 } 1278 PROC_LOCK(td->td_proc); 1279 if (td->td_proc->p_state == PRS_NEW) { 1280 PROC_UNLOCK(td->td_proc); 1281 td = NULL; 1282 break; 1283 } 1284 if (run > RUN_THRESH) { 1285 if (rw_try_upgrade(&tidhash_lock)) { 1286 LIST_REMOVE(td, td_hash); 1287 LIST_INSERT_HEAD(TIDHASH(td->td_tid), 1288 td, td_hash); 1289 rw_wunlock(&tidhash_lock); 1290 return (td); 1291 } 1292 } 1293 break; 1294 } 1295 run++; 1296 } 1297 rw_runlock(&tidhash_lock); 1298 return (td); 1299} 1300 1301void 1302tidhash_add(struct thread *td) 1303{ 1304 rw_wlock(&tidhash_lock); 1305 LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash); 1306 rw_wunlock(&tidhash_lock); 1307} 1308 1309void 1310tidhash_remove(struct thread *td) 1311{ 1312 rw_wlock(&tidhash_lock); 1313 LIST_REMOVE(td, td_hash); 1314 rw_wunlock(&tidhash_lock); 1315} 1316